modified: __pycache__/evaluation_studio.cpython-314.pyc

modified:   evaluation_studio.py
	modified:   scripts/dataset.py
	new file:   scripts/inference_utils.py
	modified:   scripts/predict_single.py
	modified:   scripts_2/__pycache__/task3_identify.cpython-314.pyc
	modified:   scripts_2/task3_identify.py

.gitignore

@@ -1,4 +1,7 @@
-.venv/
 .venv310/
 .vscode/
-src/__pycache__/
+scripts/__pycache__/
+old/
+evaluation_outputs/
+.studio_cache/
+.gitignore

README.md

@@ -1,193 +1,164 @@
-# 建筑结构响应预测项目
+# 建筑结构课程评测代码总览 (Course Evaluation Toolkit)
 
-## 项目概述
+## 项目简介
+本项目面向建筑结构振动课程评测场景，提供三类核心功能：
 
-本项目使用时间卷积网络（Temporal Convolutional Network, TCN）模型来预测建筑结构在地震波和简谐波激励下的响应。模型基于底部传感器的振动数据预测多个传感器的响应，实现结构健康监测和振动控制。
+1. 根据基底简谐激励频率，预测模型顶部加速度 Z 向响应的整体均方根 RMS 值。
+2. 根据顶部自由衰减响应，识别结构一阶自振频率和阻尼比。
+3. 根据顶部强迫振动响应，反推出输入简谐波的激振频率。
 
-## 主要特性
+项目包含三套脚本目录：`scripts`、`scripts_2`、`scripts_3`。三套脚本分别对应 RMS 预测、结构动力特性识别、激振频率逆向预测。
 
-- **多输出预测**：基于单个输入传感器预测5个输出传感器的响应
-- **时间序列建模**：使用TCN处理序列数据，支持长距离依赖
-- **数据增强**：支持有/无TMD（调谐质量阻尼器）条件下的数据训练
-- **自动化评估**：提供完整的训练和测试流程
+## 课程评测流程
+教师评测时可按以下流程使用代码：
+
+1. **顶部响应 RMS 预测**
+   教师进行一组基底激励输入试验。对每组简谐波频率，使用 `scripts` 目录下的脚本预测顶部加速度 Z 向响应的整体 RMS 值。
+2. **结构动力参数识别**
+   教师对模型略微做出改动后，采集顶部自由衰减响应，使用 `scripts_2` 目录下的脚本识别一阶自振频率和阻尼比。
+3. **激振频率逆向预测**
+   教师再次进行一组基底激励输入试验，使用 `scripts_3` 目录下的脚本，根据顶部加速度响应反推输入简谐波的激振频率。
 
 ## 项目结构
 
-```
+```text
 Building-main/
-├── src/
-│   ├── __init__.py
-│   ├── config.py          # 配置文件，包含模型和训练参数
-│   ├── dataset.py         # 数据集处理和数据加载器
-│   ├── model.py           # TCN模型定义
-│   ├── train.py           # 训练脚本
-│   └── evaluate.py        # 评估脚本
 ├── downloads/                # 数据目录
-│   ├── Non_TMD/          # 无TMD数据
-│   │   ├── train/        # 训练数据
-│   │   ├── val/          # 验证数据
-│   │   └── test/         # 测试数据
-│   └── TMD/              # 有TMD数据
-├── 数据链接.txt           # 数据下载链接配置文件
-├── download_from_data_links.py  # 数据下载脚本
-├── best_model.pth        # 训练好的模型权重
-└── README.md             # 项目说明文档
+│   └── Non_TMD/
+│       ├── free_vib/         # 自由衰减数据
+│       ├── train/            # 简谐波训练数据
+│       ├── val/              # 简谐波验证数据
+│       └── test/             # 其他测试数据
+├── scripts/                  # 任务 1：顶部响应 RMS 预测
+├── scripts_2/                # 任务 2：自振频率与阻尼比识别
+├── scripts_3/                # 任务 3：激振频率逆向预测
+├── requirement.txt           # Python 依赖清单
+└── README.md                 # 总说明文档
 ```
 
-## 环境要求
+## 三套脚本说明
 
-- Python 3.8+
-- PyTorch 1.9+
-- CUDA（可选，用于GPU加速）
-- 其他依赖：numpy, pandas, scikit-learn, matplotlib
+### `scripts`
+该目录用于完成顶部响应 RMS 预测任务。
 
-## 安装依赖
+- 输入为基底传感器信号与文件频率标签。
+- 输出为顶部加速度 Z 向响应的整体 RMS 预测值。
+- 主要脚本包括 `train_final.py`、`evaluate.py`、`predict_single.py`。
 
-1. 创建虚拟环境：
-```bash
-python -m venv .venv310
-# Windows
-.venv310\Scripts\activate
-# Linux/Mac
-source .venv310/bin/activate
-```
+### `scripts_2`
+该目录用于完成结构动力特性识别任务。
 
-2. 安装依赖：
-```bash
-pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
-pip install numpy pandas scikit-learn matplotlib
-```
+- 数据来源为 `downloads/Non_TMD/free_vib/` 目录下的自由衰减 CSV 文件。
+- 使用顶部响应传感器 `WSMS00007` 的 `value3` 列。
+- 输出一阶自振频率、阻尼比与诊断图像。
 
-## 数据准备
+### `scripts_3`
+该目录用于完成激振频率逆向预测任务。
 
-### 下载数据
+- 使用强迫振动数据，如 `downloads/Non_TMD/val/` 或 `downloads/Non_TMD/test/`。
+- 使用顶部响应传感器 `WSMS00007` 的 `value3` 列识别主频。
+- 主要脚本为 `task4_predict_freq.py`。
 
-运行数据下载脚本从配置文件下载数据：
+## 传感器约束
+
+- **顶部响应传感器**
+  任务 2 和任务 3 统一使用 `WSMS00007` 的 `value3` 列。
+- **基底输入传感器**
+  `WSMS00012` 的 `value1` 列用于参考与对照。
+- **任务 3 频率逆推**
+  激振频率预测直接基于顶部响应频谱结果。
+
+## 环境准备
+
+项目建议使用 **Python 3.10** 创建新的虚拟环境，再安装根目录的 `requirement.txt`。
+
+**1. 创建虚拟环境**
 
 ```bash
-python download_from_data_links.py
+py -3.10 -m venv .venv
 ```
 
-脚本会从 `数据链接.txt` 中读取下载链接，并将数据保存到 `downloads/` 目录。
-
-### 数据格式
-
-数据为CSV格式的长表结构，包含以下字段：
-- `code`: 传感器编码
-- `type`: 数据类型
-- `time`: 时间戳
-- `value1`, `value2`, `value3`: 三个轴的振动值
-
-### 数据集说明
-
-- **Non_TMD**: 无调谐质量阻尼器条件下的数据
-- **TMD**: 安装调谐质量阻尼器条件下的数据
-- 包含谐波激励（不同频率和振幅）和地震波激励数据
-
-## 配置说明
-
-主要配置参数在 `src/config.py` 中：
-
-- **数据配置**：
-  - `SEQ_LEN`: 序列长度（512）
-  - `STEP_SIZE`: 滑动窗口步长（20）
-  - `BATCH_SIZE`: 批大小（256）
-
-- **传感器配置**：
-  - `INPUT_SENSOR`: 输入传感器编码（'WSMS00012'）
-  - `OUTPUT_SENSORS`: 输出传感器编码列表
-  - `INPUT_AXIS`: 输入轴（'value1' - X轴）
-  - `OUTPUT_AXIS`: 输出轴（'value3' - Z轴）
-
-- **模型配置**：
-  - `CHANNELS`: TCN各层通道数 [64, 64, 128, 128, 256, 256]
-  - `KERNEL_SIZE`: 卷积核大小（5）
-  - `DROPOUT`: Dropout率（0.2）
-
-- **训练配置**：
-  - `LEARNING_RATE`: 学习率（1e-4）
-  - `EPOCHS`: 训练轮数（50）
-  - `WEIGHT_DECAY`: 权重衰减（1e-3）
-
-## 训练模型
-
-### 基本训练
-
-运行训练脚本：
+**2. 激活虚拟环境**
 
 ```bash
-cd src
-python train.py
+.\.venv\Scripts\activate
 ```
 
-训练过程会：
-1. 加载Non_TMD条件下的训练和验证数据
-2. 初始化TCN模型
-3. 使用AdamW优化器和学习率调度器训练
-4. 保存最佳模型到 `best_model.pth`
-
-### 高级训练选项
-
-修改 `src/config.py` 中的参数来自定义训练：
-- 调整学习率、批大小等超参数
-- 启用/禁用早停（`ENABLE_EARLY_STOP`）
-- 修改模型架构参数
-
-## 模型评估
-
-运行评估脚本：
+**3. 安装依赖**
 
 ```bash
-cd src
-python evaluate.py
+python -m pip install -r requirement.txt
 ```
 
-评估过程会：
-1. 加载测试数据
-2. 加载训练好的模型权重
-3. 对测试数据进行预测
-4. 生成可视化结果保存为 `test_results.png`
+`requirement.txt` 已包含三套脚本使用到的核心依赖：
 
-## 模型架构
+- `numpy`
+- `pandas`
+- `scipy`
+- `matplotlib`
+- `scikit-learn`
 
-### BuildingTCN
+## 使用方法
 
-- **输入**: (batch_size, seq_len, 1) - 底部传感器X轴数据
-- **输出**: (batch_size, seq_len, 5) - 5个传感器的Z轴预测
-- **架构**: 多层TCN + 线性输出层
+### 1. 顶部响应 RMS 预测
+对单个简谐波文件预测顶部响应 RMS：
 
-### TemporalConvNet
+```bash
+python scripts\predict_single.py --file downloads\Non_TMD\val\harmonic_5mm_0.75Hz.csv
+```
 
-- 使用膨胀卷积（dilation）实现长距离依赖
-- 残差连接保证梯度传播
-- 权重归一化提高训练稳定性
+对全量样本进行评估：
 
-## 使用说明
+```bash
+python scripts\evaluate.py
+```
 
-### 自定义数据集
+如需重新训练模型：
 
-1. 将新数据放入 `downloads/Non_TMD/` 或 `downloads/TMD/` 相应子目录
-2. 确保数据格式符合要求
-3. 修改 `src/config.py` 中的传感器配置（如需要）
-4. 重新运行训练脚本
+```bash
+python scripts\train_final.py
+```
 
-### 预测新数据
+### 2. 自由衰减法识别自振频率和阻尼比
+对自由衰减文件执行识别：
 
-1. 准备输入数据（CSV格式）
-2. 修改 `src/dataset.py` 中的数据加载逻辑
-3. 使用加载的模型进行推理
+```bash
+python scripts_2\task3_identify.py --file downloads\Non_TMD\free_vib\free_decay.csv
+```
 
-## 注意事项
+脚本输出内容包括：
 
-- 确保数据目录结构正确
-- GPU内存不足时可适当减小 `BATCH_SIZE`
-- 模型收敛可能需要调整学习率
-- 测试数据可视化仅显示第一个batch的第一条序列
+- 一阶自振频率 `f_n`
+- 对数衰减率 `delta`
+- 阻尼比 `zeta`
+- 原始波形、滤波结果与包络线图像
 
-## 贡献
+### 3. 根据顶部响应反推激振频率
+对单个强迫振动文件执行频率逆推：
 
-欢迎提交Issue和Pull Request来改进项目。
+```bash
+python scripts_3\task4_predict_freq.py --file downloads\Non_TMD\val\harmonic_5mm_0.75Hz.csv
+```
 
-## 许可证
+脚本输出内容包括：
 
-本项目仅供学习和研究使用。
+- 稳态段采样范围
+- 预测激振频率
+- 文件名中的频率真值
+- 绝对误差
+- 时域与频域诊断图像
+
+## 结果输出
+
+运行脚本后，图像与评测结果默认保存到 `evaluation_outputs/` 目录下，对应子目录如下：
+
+- `evaluation_outputs/task1_final/`
+- `evaluation_outputs/task3_identify/`
+- `evaluation_outputs/task4_predict_freq/`
+
+## 说明
+
+- `scripts` 目录负责 RMS 预测。
+- `scripts_2` 目录负责自由衰减识别。
+- `scripts_3` 目录负责激振频率逆推。
+- 数据文件命名中包含频率标签时，脚本会自动提取真值用于误差统计。

convert_teacher_csv.py

@@ -0,0 +1,200 @@
+from __future__ import annotations
+
+import argparse
+from pathlib import Path
+import re
+
+import numpy as np
+import pandas as pd
+
+
+VIB_COLUMN_RE = re.compile(r"^(WSMS\d+)\.Acc([XYZ])$", flags=re.IGNORECASE)
+SGD_COLUMN_RE = re.compile(r"^(WSGD\d+)\.chdata$", flags=re.IGNORECASE)
+OUTPUT_COLUMNS = ["code", "type", "time", "value1", "value2", "value3"]
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(
+        description="Convert teacher wide-format CSV into downloads-compatible long-format CSV."
+    )
+    parser.add_argument("--input", type=str, required=True, help="Input wide-format CSV path.")
+    parser.add_argument("--output", type=str, required=True, help="Output long-format CSV path.")
+    parser.add_argument(
+        "--time-column",
+        type=str,
+        default="time",
+        help="Time column name in the input CSV (default: time).",
+    )
+    parser.add_argument(
+        "--timezone-offset-hours",
+        type=float,
+        default=8.0,
+        help="Offset used when parsing naive datetime strings to epoch seconds (default: 8).",
+    )
+    return parser.parse_args()
+
+
+def discover_sensor_columns(columns: list[str]) -> tuple[dict[str, dict[str, str]], dict[str, str]]:
+    vibr_map: dict[str, dict[str, str]] = {}
+    sgd_map: dict[str, str] = {}
+    for column in columns:
+        vibr_match = VIB_COLUMN_RE.match(column)
+        if vibr_match is not None:
+            sensor_code = vibr_match.group(1).upper()
+            axis = vibr_match.group(2).upper()
+            vibr_map.setdefault(sensor_code, {})[axis] = column
+            continue
+        sgd_match = SGD_COLUMN_RE.match(column)
+        if sgd_match is not None:
+            sensor_code = sgd_match.group(1).upper()
+            sgd_map[sensor_code] = column
+    return vibr_map, sgd_map
+
+
+def _strip_excel_datetime_wrapper(values: pd.Series) -> pd.Series:
+    text = values.astype(str).str.strip()
+    text = text.str.replace(r'^="(.*)"$', r"\1", regex=True)
+    text = text.str.replace(r'^"(.*)"$', r"\1", regex=True)
+    return text
+
+
+def to_epoch_seconds(time_values: pd.Series, timezone_offset_hours: float) -> pd.Series:
+    numeric_time = pd.to_numeric(time_values, errors="coerce")
+    if numeric_time.notna().mean() > 0.8:
+        return numeric_time.astype("float64")
+
+    text_time = _strip_excel_datetime_wrapper(time_values)
+    dt_values = pd.to_datetime(text_time, errors="coerce")
+    epoch = (dt_values - pd.Timestamp("1970-01-01")) / pd.Timedelta(seconds=1)
+    epoch = epoch - timezone_offset_hours * 3600.0
+    epoch = pd.to_numeric(epoch, errors="coerce")
+
+    # Fall back to numeric parsing for rows that were not parsed as datetime.
+    fallback_numeric = pd.to_numeric(text_time, errors="coerce")
+    epoch = epoch.where(epoch.notna(), fallback_numeric)
+    return epoch.astype("float64")
+
+
+def _empty_output_frame() -> pd.DataFrame:
+    return pd.DataFrame(columns=OUTPUT_COLUMNS)
+
+
+def build_vibr_rows(df: pd.DataFrame, time_column: str, vibr_map: dict[str, dict[str, str]]) -> list[pd.DataFrame]:
+    rows: list[pd.DataFrame] = []
+    for sensor_code in sorted(vibr_map.keys()):
+        axis_columns = vibr_map[sensor_code]
+        use_columns = [time_column]
+        if "X" in axis_columns:
+            use_columns.append(axis_columns["X"])
+        if "Y" in axis_columns:
+            use_columns.append(axis_columns["Y"])
+        if "Z" in axis_columns:
+            use_columns.append(axis_columns["Z"])
+        if len(use_columns) == 1:
+            continue
+
+        sensor_df = df.loc[:, use_columns].copy()
+        sensor_df = sensor_df.rename(
+            columns={
+                axis_columns.get("X", ""): "value1",
+                axis_columns.get("Y", ""): "value2",
+                axis_columns.get("Z", ""): "value3",
+            }
+        )
+        for column in ["value1", "value2", "value3"]:
+            if column not in sensor_df.columns:
+                sensor_df[column] = np.nan
+        sensor_df["value1"] = pd.to_numeric(sensor_df["value1"], errors="coerce")
+        sensor_df["value2"] = pd.to_numeric(sensor_df["value2"], errors="coerce")
+        sensor_df["value3"] = pd.to_numeric(sensor_df["value3"], errors="coerce")
+        sensor_df = sensor_df[sensor_df[["value1", "value2", "value3"]].notna().any(axis=1)]
+        if sensor_df.empty:
+            continue
+
+        sensor_df.insert(0, "type", "vibr")
+        sensor_df.insert(0, "code", sensor_code)
+        rows.append(sensor_df[OUTPUT_COLUMNS])
+    return rows
+
+
+def build_sgd_rows(df: pd.DataFrame, time_column: str, sgd_map: dict[str, str]) -> list[pd.DataFrame]:
+    rows: list[pd.DataFrame] = []
+    for sensor_code in sorted(sgd_map.keys()):
+        sensor_column = sgd_map[sensor_code]
+        sensor_df = df.loc[:, [time_column, sensor_column]].copy()
+        sensor_df = sensor_df.rename(columns={sensor_column: "value1"})
+        sensor_df["value1"] = pd.to_numeric(sensor_df["value1"], errors="coerce")
+        sensor_df = sensor_df[sensor_df["value1"].notna()]
+        if sensor_df.empty:
+            continue
+
+        sensor_df["value2"] = np.nan
+        sensor_df["value3"] = np.nan
+        sensor_df.insert(0, "type", "sgd")
+        sensor_df.insert(0, "code", sensor_code)
+        rows.append(sensor_df[OUTPUT_COLUMNS])
+    return rows
+
+
+def convert_dataframe(input_df: pd.DataFrame, time_column: str, timezone_offset_hours: float) -> pd.DataFrame:
+    if time_column not in input_df.columns:
+        raise ValueError(f"Time column '{time_column}' not found in input CSV.")
+
+    vibr_map, sgd_map = discover_sensor_columns(input_df.columns.tolist())
+    if not vibr_map and not sgd_map:
+        raise ValueError("No sensor columns detected. Expected WSMSxxxx.AccX/Y/Z or WSGDxxxx.chdata columns.")
+
+    df = input_df.copy()
+    df[time_column] = to_epoch_seconds(df[time_column], timezone_offset_hours=timezone_offset_hours)
+    df = df[df[time_column].notna()]
+    if df.empty:
+        return _empty_output_frame()
+
+    rows: list[pd.DataFrame] = []
+    rows.extend(build_vibr_rows(df, time_column=time_column, vibr_map=vibr_map))
+    rows.extend(build_sgd_rows(df, time_column=time_column, sgd_map=sgd_map))
+    if not rows:
+        return _empty_output_frame()
+
+    output_df = pd.concat(rows, ignore_index=True)
+    output_df["time"] = pd.to_numeric(output_df["time"], errors="coerce").astype("float64")
+    output_df = output_df[output_df["time"].notna()]
+    output_df = output_df.sort_values(["time", "code"], kind="mergesort").reset_index(drop=True)
+    return output_df.loc[:, OUTPUT_COLUMNS]
+
+
+def print_summary(df: pd.DataFrame, output_path: Path) -> None:
+    print(f"Output saved to: {output_path}")
+    print(f"Rows: {len(df)}")
+    if df.empty:
+        print("No data rows were generated.")
+        return
+    print(f"Sensors: {df['code'].nunique()}")
+    print("Rows by sensor:")
+    counts = df.groupby(["code", "type"]).size().sort_index()
+    for (code, sensor_type), count in counts.items():
+        print(f"  - {code} ({sensor_type}): {int(count)}")
+
+
+def main() -> None:
+    args = parse_args()
+    input_path = Path(args.input).resolve()
+    output_path = Path(args.output).resolve()
+
+    if not input_path.exists():
+        raise FileNotFoundError(f"Input file not found: {input_path}")
+
+    input_df = pd.read_csv(input_path)
+    output_df = convert_dataframe(
+        input_df=input_df,
+        time_column=args.time_column,
+        timezone_offset_hours=float(args.timezone_offset_hours),
+    )
+
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+    output_df.to_csv(output_path, index=False)
+    print_summary(output_df, output_path=output_path)
+
+
+if __name__ == "__main__":
+    main()

evaluation_studio.py

@@ -0,0 +1,735 @@
+from __future__ import annotations
+
+import argparse
+from datetime import datetime
+import math
+import pickle
+from pathlib import Path
+import shutil
+import threading
+import traceback
+import tkinter as tk
+from tkinter import filedialog, messagebox, ttk
+from tkinter.scrolledtext import ScrolledText
+
+import matplotlib
+
+matplotlib.use("Agg")
+
+import matplotlib.pyplot as plt
+import numpy as np
+import pandas as pd
+
+
+PROJECT_ROOT = Path(__file__).resolve().parent
+CHECKPOINT_DEFAULT = PROJECT_ROOT / "checkpoints_final" / "task1_final" / "task1_final_model.pkl"
+ADAPTER_DEFAULT = PROJECT_ROOT / "checkpoints_final" / "task1_final" / "task1_tmd_adapter.pkl"
+APP_OUTPUT_DIR = PROJECT_ROOT / "evaluation_outputs" / "studio_outputs"
+APP_CACHE_DIR = PROJECT_ROOT / ".studio_cache"
+
+TASK1_LABEL = "任务1：顶部RMS预测（Non_TMD）"
+TASK2_LABEL = "任务2：自由衰减识别自振频率和阻尼比"
+TASK3_LABEL = "任务3：顶部响应反推激振频率"
+TASK4_LABEL = "挑战任务：TMD RMS预测（基础模型+适配器）"
+
+TASK_CHOICES = [TASK1_LABEL, TASK2_LABEL, TASK3_LABEL, TASK4_LABEL]
+
+HAS_DND = False
+TkRootClass: type[tk.Tk] = tk.Tk
+try:
+    from tkinterdnd2 import DND_FILES, TkinterDnD  # type: ignore
+
+    HAS_DND = True
+    TkRootClass = TkinterDnD.Tk
+except Exception:
+    DND_FILES = "DND_Files"  # type: ignore
+
+
+def pretty_json_like_lines(payload: dict) -> str:
+    lines: list[str] = []
+    for key in sorted(payload.keys()):
+        value = payload[key]
+        if isinstance(value, float):
+            lines.append(f"- {key}: {value:.6f}")
+        else:
+            lines.append(f"- {key}: {value}")
+    return "\n".join(lines)
+
+
+def ensure_long_format_csv(input_path: Path, force_convert: bool, tz_offset_hours: float) -> tuple[Path, bool]:
+    from convert_teacher_csv import convert_dataframe
+
+    df_head = pd.read_csv(input_path, nrows=5)
+    lower_columns = {column.lower() for column in df_head.columns}
+    long_required = {"code", "type", "time", "value1", "value2", "value3"}
+    is_long = long_required.issubset(lower_columns)
+    if is_long and not force_convert:
+        return input_path, False
+
+    APP_CACHE_DIR.mkdir(parents=True, exist_ok=True)
+    full_df = pd.read_csv(input_path)
+    converted_df = convert_dataframe(
+        input_df=full_df,
+        time_column="time",
+        timezone_offset_hours=tz_offset_hours,
+    )
+    stamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+    output_path = APP_CACHE_DIR / f"{input_path.stem}_converted_{stamp}.csv"
+    converted_df.to_csv(output_path, index=False)
+    return output_path, True
+
+
+def _load_sensor_frame(df: pd.DataFrame, sensor_code: str, value_col: str, time_col: str = "time") -> pd.DataFrame:
+    sensor_df = df.loc[df["code"] == sensor_code, [time_col, value_col]].copy()
+    if sensor_df.empty:
+        return sensor_df
+    sensor_df = sensor_df.sort_values(time_col)
+    sensor_df = sensor_df.drop_duplicates(subset=time_col, keep="first")
+    sensor_df[time_col] = pd.to_numeric(sensor_df[time_col], errors="coerce")
+    sensor_df[value_col] = pd.to_numeric(sensor_df[value_col], errors="coerce")
+    sensor_df = sensor_df[sensor_df[time_col].notna() & sensor_df[value_col].notna()]
+    return sensor_df
+
+
+def build_prediction_input(csv_path: Path, for_tmd: bool) -> dict:
+    from scripts.config import make_experiment_config
+    from scripts.dataset import (
+        calculate_rms,
+        estimate_sampling_rate,
+        extract_core_features,
+        get_dominant_frequency,
+        get_middle_segment,
+    )
+
+    config = make_experiment_config().data
+    if for_tmd:
+        config.data_dir = "downloads/TMD"
+        config.__post_init__()
+
+    df = pd.read_csv(csv_path)
+    base_df = _load_sensor_frame(df, sensor_code=config.base_sensor_code, value_col=config.base_axis, time_col=config.time_column)
+    if base_df.empty:
+        raise ValueError(
+            f"缺少基底传感器 {config.base_sensor_code}/{config.base_axis}，"
+            "无法执行RMS预测任务。"
+        )
+
+    response_df = _load_sensor_frame(
+        df,
+        sensor_code=config.response_sensor_code,
+        value_col=config.response_axis,
+        time_col=config.time_column,
+    )
+
+    if response_df.empty:
+        time_values = base_df[config.time_column].to_numpy(dtype=np.float64)
+        x_values = base_df[config.base_axis].to_numpy(dtype=np.float32)
+        time_middle, x_middle, _ = get_middle_segment(time_values, x_values, x_values, config)
+        y_middle = None
+        has_truth = False
+    else:
+        aligned = base_df.rename(columns={config.base_axis: "base_signal"}).merge(
+            response_df.rename(columns={config.response_axis: "response_signal"}),
+            on=config.time_column,
+            how="left",
+        )
+        aligned["response_signal"] = aligned["response_signal"].interpolate(method=config.interpolation_method, limit_direction="both")
+        aligned["response_signal"] = aligned["response_signal"].ffill().bfill()
+        time_values = aligned[config.time_column].to_numpy(dtype=np.float64)
+        x_values = aligned["base_signal"].to_numpy(dtype=np.float32)
+        y_values = aligned["response_signal"].to_numpy(dtype=np.float32)
+        time_middle, x_middle, y_middle = get_middle_segment(time_values, x_values, y_values, config)
+        has_truth = True
+
+    sampling_rate = estimate_sampling_rate(np.asarray(time_middle, dtype=np.float64))
+    frequency_hz = get_dominant_frequency(np.asarray(x_middle, dtype=np.float32), sampling_rate)
+    features = extract_core_features(np.asarray(x_middle, dtype=np.float32), sampling_rate, config, known_frequency_hz=frequency_hz)
+    x_rms = float(calculate_rms(np.asarray(x_middle, dtype=np.float32)))
+
+    true_y_rms = None
+    if has_truth and y_middle is not None:
+        true_y_rms = float(calculate_rms(np.asarray(y_middle, dtype=np.float32)))
+
+    return {
+        "config": config,
+        "time_middle": np.asarray(time_middle, dtype=np.float64),
+        "x_middle": np.asarray(x_middle, dtype=np.float32),
+        "y_middle": None if y_middle is None else np.asarray(y_middle, dtype=np.float32),
+        "sampling_rate": float(sampling_rate),
+        "frequency_hz": float(frequency_hz),
+        "features": np.asarray(features, dtype=np.float32),
+        "x_rms": x_rms,
+        "true_y_rms": true_y_rms,
+        "has_truth": has_truth,
+    }
+
+
+def save_prediction_overview_figure(
+    *,
+    file_path: Path,
+    output_dir: Path,
+    title_prefix: str,
+    time_middle: np.ndarray,
+    x_middle: np.ndarray,
+    sampling_rate: float,
+    frequency_hz: float,
+    pred_rms: float,
+    true_rms: float | None,
+) -> Path:
+    output_dir.mkdir(parents=True, exist_ok=True)
+    fft_values = np.fft.rfft(x_middle - np.mean(x_middle))
+    freqs = np.fft.rfftfreq(x_middle.size, d=1.0 / sampling_rate)
+    amplitudes = np.abs(fft_values)
+
+    fig, axes = plt.subplots(3, 1, figsize=(10, 8))
+    fig.suptitle(f"{title_prefix} | {file_path.name}")
+
+    axes[0].plot(time_middle, x_middle, color="tab:blue")
+    axes[0].set_title("Input Base Signal (Middle Segment)")
+    axes[0].set_xlabel("Time")
+    axes[0].set_ylabel("Acceleration")
+    axes[0].grid(True, alpha=0.25)
+
+    axes[1].plot(freqs, amplitudes, color="tab:purple")
+    axes[1].axvline(frequency_hz, color="tab:red", linestyle="--", linewidth=1.2, label=f"f={frequency_hz:.4f}Hz")
+    axes[1].set_xlim(0.0, 5.0)
+    axes[1].set_title("Input Spectrum")
+    axes[1].set_xlabel("Frequency (Hz)")
+    axes[1].set_ylabel("Amplitude")
+    axes[1].grid(True, alpha=0.25)
+    axes[1].legend()
+
+    if true_rms is None:
+        axes[2].bar(["Pred RMS"], [pred_rms], color=["tab:orange"])
+        axes[2].set_title(f"Pred RMS = {pred_rms:.6f} (truth sensor not provided)")
+    else:
+        axes[2].bar(["True RMS", "Pred RMS"], [true_rms, pred_rms], color=["tab:green", "tab:orange"])
+        err_pct = abs(pred_rms - true_rms) / max(abs(true_rms), 1e-12) * 100.0
+        axes[2].set_title(f"True={true_rms:.6f} | Pred={pred_rms:.6f} | Error={err_pct:.2f}%")
+    axes[2].set_ylabel("RMS")
+    axes[2].grid(True, axis="y", alpha=0.25)
+
+    plt.tight_layout()
+    figure_path = output_dir / f"{file_path.stem}_{title_prefix.replace(' ', '_').lower()}.png"
+    plt.savefig(figure_path, dpi=170, bbox_inches="tight")
+    plt.close(fig)
+    return figure_path
+
+
+def run_task1(csv_path: Path) -> tuple[dict, Path]:
+    from scripts.config import CORE_FEATURE_NAMES
+    from scripts.inference_utils import predict_task1_tr
+
+    with CHECKPOINT_DEFAULT.open("rb") as handle:
+        payload = pickle.load(handle)
+    model = payload["model"]
+    pred_input = build_prediction_input(csv_path=csv_path, for_tmd=False)
+    pred_tr, tr_source = predict_task1_tr(
+        model,
+        feature_vector=pred_input["features"],
+        frequency_hz=float(pred_input["frequency_hz"]),
+        normalization_eps=float(pred_input["config"].normalization_eps),
+        project_root=PROJECT_ROOT,
+        prefer_curve=True,
+    )
+    pred_rms = pred_tr * float(pred_input["x_rms"])
+    true_rms = pred_input["true_y_rms"]
+    rel_err_pct = None if true_rms is None else abs(pred_rms - true_rms) / max(abs(true_rms), 1e-12) * 100.0
+
+    out_dir = APP_OUTPUT_DIR / "task1"
+    fig_path = save_prediction_overview_figure(
+        file_path=csv_path,
+        output_dir=out_dir,
+        title_prefix="Task1 Prediction",
+        time_middle=pred_input["time_middle"],
+        x_middle=pred_input["x_middle"],
+        sampling_rate=float(pred_input["sampling_rate"]),
+        frequency_hz=float(pred_input["frequency_hz"]),
+        pred_rms=float(pred_rms),
+        true_rms=true_rms,
+    )
+    result = {
+        "task": "task1_rms_prediction",
+        "file_path": str(csv_path),
+        "frequency_hz": float(pred_input["frequency_hz"]),
+        "x_rms": float(pred_input["x_rms"]),
+        "true_y_rms": None if true_rms is None else float(true_rms),
+        "pred_tr": float(pred_tr),
+        "pred_tr_source": str(tr_source),
+        "pred_y_rms": float(pred_rms),
+        "relative_error_percent": None if rel_err_pct is None else float(rel_err_pct),
+        "truth_available": bool(pred_input["has_truth"]),
+        "model_name": str(payload.get("model_name", "unknown")),
+        "feature_names": ",".join(CORE_FEATURE_NAMES),
+    }
+    return result, Path(fig_path)
+
+
+def run_task2(csv_path: Path) -> tuple[dict, Path]:
+    from scripts_2.task3_identify import analyze_free_vibration
+
+    real_csv = csv_path
+    if "free_vib" not in {part.lower() for part in csv_path.parts}:
+        temp_dir = APP_CACHE_DIR / "free_vib"
+        temp_dir.mkdir(parents=True, exist_ok=True)
+        real_csv = temp_dir / csv_path.name
+        shutil.copyfile(csv_path, real_csv)
+
+    output_dir = APP_OUTPUT_DIR / "task2"
+    result = analyze_free_vibration(
+        file_path=real_csv,
+        output_dir=output_dir,
+        bandwidth_hz=0.3,
+        show=False,
+    )
+    fig_path = Path(str(result["figure_path"]))
+    return result, fig_path
+
+
+def run_task3(csv_path: Path) -> tuple[dict, Path]:
+    from scripts.config import make_experiment_config
+    from scripts_2.signal_utils import (
+        TOP_RESPONSE_AXIS,
+        TOP_RESPONSE_SENSOR,
+        calculate_rms,
+        get_middle_segment,
+        read_sensor_signal,
+    )
+    from scripts_3.task4_predict_freq import analyze_forced_vibration
+
+    output_dir = APP_OUTPUT_DIR / "task3"
+    result = analyze_forced_vibration(
+        file_path=csv_path,
+        output_dir=output_dir,
+        show=False,
+    )
+
+    # Optional inverse estimation:
+    # infer base RMS from top-response RMS with fixed-point iteration:
+    # x = y_rms / TR_model(f, x)
+    try:
+        config = make_experiment_config()
+        with CHECKPOINT_DEFAULT.open("rb") as handle:
+            payload = pickle.load(handle)
+        model = payload["model"]
+
+        time_values, raw_signal = read_sensor_signal(
+            csv_path,
+            sensor_code=TOP_RESPONSE_SENSOR,
+            value_column=TOP_RESPONSE_AXIS,
+        )
+        _, signal_segment = get_middle_segment(time_values, raw_signal)
+        top_rms = float(calculate_rms(signal_segment))
+        freq_hz = float(result["predicted_frequency_hz"])
+
+        min_x = config.data.normalization_eps
+        x_est = max(top_rms, min_x)
+        converged = False
+        rel_change = float("inf")
+        iter_count = 0
+
+        for i in range(1, 41):
+            feat = np.asarray(
+                [
+                    [
+                        freq_hz,
+                        freq_hz * freq_hz,
+                        x_est,
+                        (2.0 * np.pi * freq_hz) ** 2 * config.data.harmonic_amplitude_m,
+                    ]
+                ],
+                dtype=np.float64,
+            )
+            tr_est = max(float(model.predict(feat)[0]), config.data.normalization_eps)
+            x_new = max(top_rms / max(tr_est, config.data.normalization_eps), min_x)
+            if not np.isfinite(x_new):
+                break
+            rel_change = abs(x_new - x_est) / max(abs(x_est), min_x)
+            # Damping update to improve stability in fixed-point iteration.
+            x_est = 0.6 * x_est + 0.4 * x_new
+            iter_count = i
+            if rel_change < 1e-4:
+                converged = True
+                break
+
+        feat_final = np.asarray(
+            [
+                [
+                    freq_hz,
+                    freq_hz * freq_hz,
+                    x_est,
+                    (2.0 * np.pi * freq_hz) ** 2 * config.data.harmonic_amplitude_m,
+                ]
+            ],
+            dtype=np.float64,
+        )
+        tr_final = max(float(model.predict(feat_final)[0]), config.data.normalization_eps)
+        y_reconstructed = tr_final * x_est
+        residual_pct = abs(y_reconstructed - top_rms) / max(abs(top_rms), min_x) * 100.0
+
+        result["top_response_rms"] = float(top_rms)
+        result["estimated_base_rms"] = float(x_est)
+        result["estimated_tr_from_model"] = float(tr_final)
+        result["inversion_converged"] = bool(converged)
+        result["inversion_iterations"] = int(iter_count)
+        result["inversion_relative_change"] = float(rel_change)
+        result["inversion_residual_percent"] = float(residual_pct)
+        result["inversion_note"] = "Estimated by fixed-point inversion using task1 TR model; use as an estimate only."
+    except Exception as error:
+        result["inversion_note"] = f"Base RMS inversion unavailable: {error}"
+
+    fig_path = Path(str(result["figure_path"]))
+    return result, fig_path
+
+
+def run_task4_tmd(csv_path: Path) -> tuple[dict, Path]:
+    from scripts_4.adapter import load_adapter
+    from scripts.inference_utils import predict_task1_tr
+
+    with CHECKPOINT_DEFAULT.open("rb") as handle:
+        payload = pickle.load(handle)
+    model = payload["model"]
+    adapter, adapter_extra = load_adapter(ADAPTER_DEFAULT)
+    pred_input = build_prediction_input(csv_path=csv_path, for_tmd=True)
+    pred_tr, tr_source = predict_task1_tr(
+        model,
+        feature_vector=pred_input["features"],
+        frequency_hz=float(pred_input["frequency_hz"]),
+        normalization_eps=float(pred_input["config"].normalization_eps),
+        project_root=PROJECT_ROOT,
+        prefer_curve=True,
+    )
+    pred_base = pred_tr * float(pred_input["x_rms"])
+    pred_tmd = adapter.predict(pred_base, float(pred_input["frequency_hz"]))
+    true_rms = pred_input["true_y_rms"]
+    rel_err_pct = None if true_rms is None else abs(pred_tmd - true_rms) / max(abs(true_rms), 1e-12) * 100.0
+
+    out_dir = APP_OUTPUT_DIR / "task4"
+    fig_path = save_prediction_overview_figure(
+        file_path=csv_path,
+        output_dir=out_dir,
+        title_prefix="Task4 TMD Prediction",
+        time_middle=pred_input["time_middle"],
+        x_middle=pred_input["x_middle"],
+        sampling_rate=float(pred_input["sampling_rate"]),
+        frequency_hz=float(pred_input["frequency_hz"]),
+        pred_rms=float(pred_tmd),
+        true_rms=true_rms,
+    )
+    result = {
+        "task": "task4_tmd_rms_prediction",
+        "file_path": str(csv_path),
+        "frequency_hz": float(pred_input["frequency_hz"]),
+        "x_rms": float(pred_input["x_rms"]),
+        "true_y_rms": None if true_rms is None else float(true_rms),
+        "pred_base_y_rms": float(pred_base),
+        "pred_tr_source": str(tr_source),
+        "pred_tmd_y_rms": float(pred_tmd),
+        "adapter_scale_k": float(adapter.scale_at(float(pred_input["frequency_hz"]))),
+        "relative_error_percent": None if rel_err_pct is None else float(rel_err_pct),
+        "truth_available": bool(pred_input["has_truth"]),
+        "model_name": str(payload.get("model_name", "unknown")),
+        "adapter_extra": str(adapter_extra),
+    }
+    return result, Path(fig_path)
+
+
+def build_key_metrics(task_label: str, result: dict) -> str:
+    if task_label == TASK1_LABEL:
+        lines = [
+            f"预测RMS: {result.get('pred_y_rms')}",
+            f"预测TR: {result.get('pred_tr')}",
+            f"TR来源: {result.get('pred_tr_source')}",
+            f"激振频率(Hz): {result.get('frequency_hz')}",
+            f"输入RMS(x): {result.get('x_rms')}",
+            f"是否有真值: {'是' if result.get('truth_available') else '否'}",
+        ]
+        if result.get("truth_available"):
+            lines.append(f"真值RMS(y): {result.get('true_y_rms')}")
+            lines.append(f"误差(%): {result.get('relative_error_percent')}")
+        return "\n".join(lines)
+    if task_label == TASK2_LABEL:
+        lines = [
+            f"自振频率f_n(Hz): {result.get('natural_frequency_hz')}",
+            f"局部阻尼比zeta: {result.get('damping_ratio')}",
+            f"全局阻尼比zeta: {result.get('global_damping_ratio')}",
+            f"最高峰后衰减阻尼比zeta: {result.get('max_tail_damping_ratio')}",
+            f"局部精确阻尼比: {result.get('damping_ratio_exact')}",
+            f"全局精确阻尼比: {result.get('global_damping_ratio_exact')}",
+            f"最高峰后衰减精确阻尼比: {result.get('max_tail_damping_ratio_exact')}",
+            f"局部包络R²: {result.get('envelope_r2')}",
+            f"全局包络R²: {result.get('global_envelope_r2')}",
+            f"最高峰后衰减包络R²: {result.get('max_tail_envelope_r2')}",
+            f"采样率(Hz): {result.get('sampling_rate_hz')}",
+        ]
+        return "\n".join(lines)
+    if task_label == TASK3_LABEL:
+        lines = [
+            f"预测激振频率(Hz): {result.get('predicted_frequency_hz')}",
+            f"稳态采样点数: {result.get('segment_sample_count')}",
+            f"采样率(Hz): {result.get('sampling_rate_hz')}",
+        ]
+        if result.get("top_response_rms") is not None:
+            lines.append(f"顶部RMS(y): {result.get('top_response_rms')}")
+        if result.get("estimated_base_rms") is not None:
+            lines.append(f"反演底部RMS(x): {result.get('estimated_base_rms')}")
+            lines.append(f"反演TR: {result.get('estimated_tr_from_model')}")
+            lines.append(f"反演残差(%): {result.get('inversion_residual_percent')}")
+            lines.append(f"反演收敛: {result.get('inversion_converged')}")
+        if result.get("true_frequency_hz") is not None:
+            lines.append(f"文件名真值(Hz): {result.get('true_frequency_hz')}")
+            lines.append(f"绝对误差(Hz): {result.get('absolute_error_hz')}")
+        return "\n".join(lines)
+    if task_label == TASK4_LABEL:
+        lines = [
+            f"挑战任务预测RMS: {result.get('pred_tmd_y_rms')}",
+            f"基础模型RMS: {result.get('pred_base_y_rms')}",
+            f"TR来源: {result.get('pred_tr_source')}",
+            f"适配系数k(f): {result.get('adapter_scale_k')}",
+            f"激振频率(Hz): {result.get('frequency_hz')}",
+            f"是否有真值: {'是' if result.get('truth_available') else '否'}",
+        ]
+        if result.get("truth_available"):
+            lines.append(f"真值RMS(y): {result.get('true_y_rms')}")
+            lines.append(f"误差(%): {result.get('relative_error_percent')}")
+        return "\n".join(lines)
+    return pretty_json_like_lines(result)
+
+
+class EvaluationStudio:
+    def __init__(self) -> None:
+        self.root = TkRootClass()
+        self.root.title("建筑评测一体式测试系统")
+        self.root.geometry("1180x760")
+        self.root.minsize(980, 620)
+
+        self.task_var = tk.StringVar(value=TASK1_LABEL)
+        self.file_var = tk.StringVar(value="")
+        self.force_convert_var = tk.BooleanVar(value=False)
+        self.timezone_var = tk.StringVar(value="8")
+        self.status_var = tk.StringVar(value="就绪")
+        self.preview_image: tk.PhotoImage | None = None
+        self.original_image: tk.PhotoImage | None = None
+        self.current_image_path: Path | None = None
+
+        self._build_ui()
+
+    def _build_ui(self) -> None:
+        main = ttk.Frame(self.root, padding=10)
+        main.pack(fill="both", expand=True)
+        main.columnconfigure(1, weight=1)
+        main.rowconfigure(4, weight=1)
+
+        ttk.Label(main, text="评测任务").grid(row=0, column=0, sticky="w", padx=(0, 8), pady=5)
+        task_box = ttk.Combobox(main, textvariable=self.task_var, values=TASK_CHOICES, state="readonly")
+        task_box.grid(row=0, column=1, sticky="ew", pady=5)
+
+        ttk.Label(main, text="CSV 文件").grid(row=1, column=0, sticky="w", padx=(0, 8), pady=5)
+        file_entry = ttk.Entry(main, textvariable=self.file_var)
+        file_entry.grid(row=1, column=1, sticky="ew", pady=5)
+        ttk.Button(main, text="选择文件", command=self._choose_file).grid(row=1, column=2, padx=(8, 0), pady=5)
+
+        self.drop_label = ttk.Label(
+            main,
+            text="将 CSV 拖到这里（若无拖拽支持，可用“选择文件”）",
+            relief="groove",
+            anchor="center",
+        )
+        self.drop_label.grid(row=2, column=0, columnspan=3, sticky="ew", pady=(4, 8), ipady=12)
+        if HAS_DND:
+            self.drop_label.drop_target_register(DND_FILES)
+            self.drop_label.dnd_bind("<<Drop>>", self._on_drop)
+
+        config_line = ttk.Frame(main)
+        config_line.grid(row=3, column=0, columnspan=3, sticky="ew", pady=5)
+        config_line.columnconfigure(6, weight=1)
+        ttk.Checkbutton(
+            config_line,
+            text="强制按老师宽表格式转换",
+            variable=self.force_convert_var,
+        ).grid(row=0, column=0, sticky="w")
+        ttk.Label(config_line, text="字符串时间时区偏移(小时)").grid(row=0, column=1, sticky="w", padx=(12, 4))
+        ttk.Entry(config_line, textvariable=self.timezone_var, width=7).grid(row=0, column=2, sticky="w")
+        ttk.Button(config_line, text="开始评测", command=self._run_clicked).grid(row=0, column=3, padx=(12, 0))
+        ttk.Label(config_line, textvariable=self.status_var).grid(row=0, column=4, sticky="w", padx=(12, 0))
+        if not HAS_DND:
+            ttk.Label(
+                config_line,
+                text="提示：未检测到 tkinterdnd2，拖拽不可用；可先 pip install tkinterdnd2",
+                foreground="#aa5500",
+            ).grid(row=0, column=5, sticky="w", padx=(12, 0))
+
+        paned = ttk.Panedwindow(main, orient="horizontal")
+        paned.grid(row=4, column=0, columnspan=3, sticky="nsew", pady=(6, 0))
+
+        left = ttk.Frame(paned, padding=8)
+        right = ttk.Frame(paned, padding=8)
+        paned.add(left, weight=3)
+        paned.add(right, weight=2)
+
+        ttk.Label(left, text="关键结果").pack(anchor="w")
+        self.key_text = ScrolledText(left, wrap="word", height=8)
+        self.key_text.pack(fill="x", expand=False, pady=(0, 6))
+
+        ttk.Label(left, text="详细结果").pack(anchor="w")
+        self.result_text = ScrolledText(left, wrap="word", height=16)
+        self.result_text.pack(fill="both", expand=True)
+
+        ttk.Label(right, text="图像预览（最新结果）").pack(anchor="w")
+        self.image_label = ttk.Label(right, relief="sunken", anchor="center")
+        self.image_label.pack(fill="both", expand=True)
+        self.image_label.bind("<Configure>", self._on_image_area_resize)
+
+    def _choose_file(self) -> None:
+        file_path = filedialog.askopenfilename(
+            title="选择CSV文件",
+            filetypes=[("CSV Files", "*.csv"), ("All Files", "*.*")],
+            initialdir=str(PROJECT_ROOT),
+        )
+        if file_path:
+            self.file_var.set(file_path)
+
+    def _on_drop(self, event) -> None:  # noqa: ANN001
+        raw = event.data
+        if not raw:
+            return
+        paths = self.root.tk.splitlist(raw)
+        if not paths:
+            return
+        self.file_var.set(str(paths[0]).strip("{}"))
+
+    def _append_result(self, text: str) -> None:
+        self.result_text.insert("end", text + "\n")
+        self.result_text.see("end")
+
+    def _on_image_area_resize(self, _event) -> None:  # noqa: ANN001
+        self._refresh_image_fit()
+
+    def _refresh_image_fit(self) -> None:
+        if self.original_image is None:
+            return
+        width = max(self.image_label.winfo_width(), 1)
+        height = max(self.image_label.winfo_height(), 1)
+        orig_w = max(self.original_image.width(), 1)
+        orig_h = max(self.original_image.height(), 1)
+        ratio_w = math.ceil(orig_w / width)
+        ratio_h = math.ceil(orig_h / height)
+        ratio = max(1, ratio_w, ratio_h)
+        scaled = self.original_image.subsample(ratio, ratio)
+        self.preview_image = scaled
+        self.image_label.configure(image=scaled, text="")
+
+    def _set_image(self, image_path: Path | None) -> None:
+        if image_path is None or not image_path.exists():
+            self.image_label.configure(image="", text="暂无图像")
+            self.preview_image = None
+            self.original_image = None
+            self.current_image_path = None
+            return
+        try:
+            self.original_image = tk.PhotoImage(file=str(image_path))
+            self.current_image_path = image_path
+            self._refresh_image_fit()
+        except Exception:
+            self.image_label.configure(text=f"图像已生成：\n{image_path}", image="")
+            self.preview_image = None
+            self.original_image = None
+            self.current_image_path = image_path
+
+    def _run_clicked(self) -> None:
+        csv_text = self.file_var.get().strip()
+        if not csv_text:
+            messagebox.showerror("缺少输入", "请先选择或拖入CSV文件。")
+            return
+        csv_path = Path(csv_text).expanduser().resolve()
+        if not csv_path.exists():
+            messagebox.showerror("文件不存在", f"找不到文件：\n{csv_path}")
+            return
+
+        try:
+            tz_offset = float(self.timezone_var.get().strip())
+        except ValueError:
+            messagebox.showerror("参数错误", "时区偏移必须是数字。")
+            return
+
+        self.status_var.set("运行中...")
+        self.key_text.delete("1.0", "end")
+        self.result_text.delete("1.0", "end")
+        self._set_image(None)
+
+        thread = threading.Thread(
+            target=self._run_pipeline,
+            args=(csv_path, self.task_var.get(), self.force_convert_var.get(), tz_offset),
+            daemon=True,
+        )
+        thread.start()
+
+    def _run_pipeline(self, csv_path: Path, task_label: str, force_convert: bool, tz_offset: float) -> None:
+        try:
+            APP_OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
+
+            use_csv, converted = ensure_long_format_csv(csv_path, force_convert=force_convert, tz_offset_hours=tz_offset)
+            notes = [
+                f"输入文件: {csv_path}",
+                f"实际评测文件: {use_csv}",
+                f"是否执行宽转长: {'是' if converted else '否'}",
+                f"任务: {task_label}",
+                "",
+            ]
+
+            if task_label == TASK1_LABEL:
+                result, fig_path = run_task1(use_csv)
+            elif task_label == TASK2_LABEL:
+                result, fig_path = run_task2(use_csv)
+            elif task_label == TASK3_LABEL:
+                result, fig_path = run_task3(use_csv)
+            elif task_label == TASK4_LABEL:
+                result, fig_path = run_task4_tmd(use_csv)
+            else:
+                raise ValueError(f"不支持的任务：{task_label}")
+
+            output_text = "\n".join(notes) + pretty_json_like_lines(result)
+            key_text = build_key_metrics(task_label, result)
+            self.root.after(0, lambda: self.key_text.insert("end", key_text + "\n"))
+            self.root.after(0, lambda: self.result_text.insert("end", output_text + "\n"))
+            self.root.after(0, lambda: self._set_image(fig_path))
+            self.root.after(0, lambda: self.status_var.set("完成"))
+        except Exception as error:
+            stack = traceback.format_exc()
+            message = f"运行失败：{error}\n\n{stack}"
+            self.root.after(0, lambda: self.key_text.insert("end", "运行失败，请查看详细错误。\n"))
+            self.root.after(0, lambda: self.result_text.insert("end", message + "\n"))
+            self.root.after(0, lambda: self.status_var.set("失败"))
+
+    def run(self) -> None:
+        self.root.mainloop()
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="Integrated visual evaluation studio for tasks 1~4.")
+    parser.add_argument("--task", type=str, default=None, choices=TASK_CHOICES)
+    parser.add_argument("--file", type=str, default=None, help="Optional CSV file path to preload.")
+    parser.add_argument(
+        "--force-convert",
+        action="store_true",
+        help="Force teacher wide-format conversion before running tasks.",
+    )
+    parser.add_argument("--timezone-offset-hours", type=float, default=8.0)
+    return parser.parse_args()
+
+
+def main() -> None:
+    args = parse_args()
+    app = EvaluationStudio()
+    if args.task:
+        app.task_var.set(args.task)
+    if args.file:
+        app.file_var.set(str(Path(args.file).expanduser().resolve()))
+    if args.force_convert:
+        app.force_convert_var.set(True)
+    app.timezone_var.set(str(args.timezone_offset_hours))
+    app.run()
+
+
+if __name__ == "__main__":
+    main()

requirement.txt

@@ -0,0 +1,6 @@
+# Python 3.10
+numpy>=1.24,<2.0
+pandas>=2.0,<3.0
+scipy>=1.10,<2.0
+matplotlib>=3.7,<4.0
+scikit-learn>=1.3,<2.0

scripts/README.md

@@ -0,0 +1,54 @@
+# 建筑结构简谐激励响应预测 (Harmonic Response Prediction)
+
+## 项目简介
+本项目旨在通过数据驱动的方法，预测建筑结构在简谐波底座激振下的顶层加速度响应。工程利用机器学习算法提取核心物理特征，建立激励与响应之间的传递关系，实现对目标结构响应均方根（RMS）的快速、准确预测。
+
+## 技术路线
+本项目的核心技术链路包含数据预处理、特征工程、传递率建模与算法回归四个关键环节，逻辑清晰，具备较强的工程可解释性：
+
+1. **稳态信号截取**
+   由于原始振动数据包含起振与衰减的瞬态过程，系统通过滑动窗口结合变异系数评估，自动截取最平稳的中间段数据，消除非稳态噪声对分析的干扰。
+2. **物理特征工程**
+   系统从底座激励信号中提取 4 维极具代表性的物理特征作为模型输入：
+   - 频率 ($f$)
+   - 频率的平方 ($f^2$)
+   - 激励均方根 ($x_{rms}$)
+   - 理论加速度幅值 ($(2\pi f)^2 \cdot A$)
+3. **传递率 (TR) 目标建模**
+   模型将预测目标设定为系统的**传递率 (Transmissibility, TR)**，即输出响应与输入激励的均方根比值 ($TR = y_{rms} / x_{rms}$)。
+   在最终推理阶段，通过公式 $\text{预测响应 } y_{rms} = \text{预测 } TR \times \text{实际输入 } x_{rms}$ 还原最终结果，这种无量纲化的处理极大提升了模型的泛化能力。
+4. **距离权重 KNN 回归**
+   采用数据标准化 (StandardScaler) 结合距离加权的 K 近邻回归 (KNeighborsRegressor) 算法。通过特征空间中的距离衰减机制，对高维空间中的传递率进行平滑拟合。
+
+## 核心模块说明
+
+- **`config.py`**
+  全局配置中心。集中管理数据路径、传感器编号（基座输入与顶层输出）、信号截取比例以及模型超参数。
+- **`dataset.py`**
+  数据处理引擎。负责读取原始 CSV 振动数据、缺失值插值对齐、稳态窗口搜索、傅里叶主频计算以及核心物理特征的打包提取。
+- **`train_final.py`**
+  模型训练入口。读取全量简谐波数据，构建特征矩阵与目标向量完成 KNN 模型训练，将最优模型序列化保存，并输出全量数据的拟合评估曲线。
+- **`evaluate.py`**
+  批量评估工具。加载已保存的模型权重，对全集数据进行预测评估，生成包含理论拟合曲线、散点对比与误差分布的可视化图表。
+- **`predict_single.py`**
+  单样本预测脚本。支持输入特定的单一 CSV 数据文件，提供端到端的预测功能，并绘制包含时域对比、频域谱线及预测 RMS 对比的综合诊断图。
+
+## 使用指南
+
+**1. 训练模型**
+读取配置文件中指定的数据集，执行全量训练并保存模型权重：
+```bash
+python scripts/train_final.py
+```
+
+**2. 批量评估**
+加载已训练的模型，对样本进行批量评估与统计分析：
+```bash
+python scripts/evaluate.py
+```
+
+**3. 单文件预测**
+对特定的单一样本进行独立预测，并生成详细的图表报告：
+```bash
+python scripts/predict_single.py --file path/to/your/harmonic.csv
+```

scripts/config.py

@@ -0,0 +1,83 @@
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from pathlib import Path
+
+
+CORE_FEATURE_NAMES: tuple[str, ...] = (
+    "frequency_hz",
+    "frequency_squared",
+    "x_rms",
+    "theoretical_accel",
+)
+
+
+@dataclass
+class DataConfig:
+    project_root: Path = field(default_factory=lambda: Path(__file__).resolve().parents[1])
+    data_dir: str = "downloads/Non_TMD"
+    harmonic_pattern: str = "harmonic*.csv"
+
+    code_column: str = "code"
+    time_column: str = "time"
+    base_sensor_code: str = "WSMS00012"
+    base_axis: str = "value1"
+    response_sensor_code: str = "WSMS00007"
+    response_axis: str = "value3"
+
+    middle_segment_start_ratio: float = 0.20
+    middle_segment_end_ratio: float = 0.80
+    min_segment_length: int = 512
+    steady_window_ratio: float = 0.25
+    steady_window_stride_ratio: float = 0.05
+    stability_subwindow_count: int = 4
+    interpolation_method: str = "linear"
+    normalization_eps: float = 1e-6
+    harmonic_amplitude_m: float = 0.005
+
+    data_root: Path = field(init=False)
+
+    def __post_init__(self) -> None:
+        self.project_root = Path(self.project_root).resolve()
+        self.data_root = (self.project_root / self.data_dir).resolve()
+
+
+@dataclass
+class ModelConfig:
+    model_name: str = "knn_distance"
+    n_neighbors: int = 4
+    distance_power: int = 2
+
+
+@dataclass
+class TrainConfig:
+    checkpoint_dir: str = "checkpoints_final"
+    model_name: str = "task1_final_model.pkl"
+    fit_csv_name: str = "task1_final_fit_all.csv"
+    fit_figure_name: str = "task1_final_fit_curve.png"
+    dense_curve_csv_name: str = "task1_final_dense_curve.csv"
+    evaluation_dense_csv_name: str = "evaluation_dense_curve.csv"
+    dense_curve_points: int = 400
+
+
+@dataclass
+class ExperimentConfig:
+    data: DataConfig = field(default_factory=DataConfig)
+    model: ModelConfig = field(default_factory=ModelConfig)
+    train: TrainConfig = field(default_factory=TrainConfig)
+
+
+def make_experiment_config() -> ExperimentConfig:
+    return ExperimentConfig()
+
+
+def checkpoint_dir(config: ExperimentConfig) -> Path:
+    path = config.data.project_root / config.train.checkpoint_dir / "task1_final"
+    path.mkdir(parents=True, exist_ok=True)
+    return path
+
+
+def evaluation_dir(config: ExperimentConfig) -> Path:
+    path = config.data.project_root / "evaluation_outputs" / "task1_final"
+    path.mkdir(parents=True, exist_ok=True)
+    return path

scripts/dataset.py

@@ -0,0 +1,292 @@
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from pathlib import Path
+import re
+
+import numpy as np
+import pandas as pd
+
+try:
+    from .config import CORE_FEATURE_NAMES, DataConfig, ExperimentConfig
+except ImportError:
+    from config import CORE_FEATURE_NAMES, DataConfig, ExperimentConfig
+
+
+def calculate_rms(signal: np.ndarray) -> float:
+    signal = np.asarray(signal, dtype=np.float64).reshape(-1)
+    return float(np.sqrt(np.mean(np.square(signal))))
+
+
+def extract_frequency_hz(file_name: str) -> float | None:
+    match = re.search(r"(\d+(?:\.\d+)?)Hz", file_name, flags=re.IGNORECASE)
+    if match is None:
+        return None
+    return float(match.group(1))
+
+
+def estimate_sampling_rate(time_values: np.ndarray) -> float:
+    if time_values.size < 2:
+        return 100.0
+    dt = np.diff(time_values)
+    dt = dt[np.isfinite(dt)]
+    dt = dt[dt > 0.0]
+    if dt.size == 0:
+        return 100.0
+    return float(1.0 / np.median(dt))
+
+
+def get_middle_segment(
+    time_values: np.ndarray,
+    x_values: np.ndarray,
+    y_values: np.ndarray,
+    config: DataConfig,
+) -> tuple[np.ndarray, np.ndarray, np.ndarray]:
+    length = len(x_values)
+    search_start = int(length * config.middle_segment_start_ratio)
+    search_end = int(length * config.middle_segment_end_ratio)
+    search_start = max(0, min(search_start, length - 1))
+    search_end = max(search_start + 1, min(search_end, length))
+
+    if search_end - search_start < config.min_segment_length:
+        center = length // 2
+        half = config.min_segment_length // 2
+        search_start = max(0, center - half)
+        search_end = min(length, search_start + config.min_segment_length)
+        search_start = max(0, search_end - config.min_segment_length)
+
+    search_length = search_end - search_start
+    window_length = max(config.min_segment_length, int(length * config.steady_window_ratio))
+    window_length = min(window_length, search_length)
+    if search_length <= window_length:
+        return (
+            time_values[search_start:search_end],
+            x_values[search_start:search_end],
+            y_values[search_start:search_end],
+        )
+
+    stride = max(1, int(length * config.steady_window_stride_ratio))
+    candidate_y = y_values[search_start:search_end]
+    candidate_rms = calculate_rms(candidate_y)
+    best_score = float("inf")
+    best_slice = slice(search_start, search_start + window_length)
+
+    for window_start in range(search_start, search_end - window_length + 1, stride):
+        window_end = window_start + window_length
+        window_y = y_values[window_start:window_end]
+        split_windows = np.array_split(window_y, config.stability_subwindow_count)
+        split_rms = np.asarray([calculate_rms(chunk) for chunk in split_windows], dtype=np.float64)
+        split_peaks = np.asarray([float(np.max(np.abs(chunk))) for chunk in split_windows], dtype=np.float64)
+
+        rms_cv = float(split_rms.std() / max(split_rms.mean(), config.normalization_eps))
+        peak_cv = float(split_peaks.std() / max(split_peaks.mean(), config.normalization_eps))
+        window_rms = calculate_rms(window_y)
+
+        score = rms_cv + 0.35 * peak_cv - 0.05 * (window_rms / max(candidate_rms, config.normalization_eps))
+        if score < best_score:
+            best_score = score
+            best_slice = slice(window_start, window_end)
+
+    return time_values[best_slice], x_values[best_slice], y_values[best_slice]
+
+
+def _compute_windowed_spectrum(signal: np.ndarray, sampling_rate: float) -> tuple[np.ndarray, np.ndarray]:
+    signal = np.asarray(signal, dtype=np.float64).reshape(-1)
+    if signal.size < 4:
+        return np.asarray([], dtype=np.float64), np.asarray([], dtype=np.float64)
+    centered = signal - np.mean(signal)
+    window = np.hanning(signal.size)
+    scale = max(np.sum(window), 1e-12)
+    fft_values = np.fft.rfft(centered * window)
+    freqs = np.fft.rfftfreq(signal.size, d=1.0 / sampling_rate)
+    magnitudes = (2.0 / scale) * np.abs(fft_values)
+    return freqs, magnitudes
+
+
+def _parabolic_peak_frequency(freqs: np.ndarray, magnitudes: np.ndarray, peak_index: int) -> float:
+    if peak_index <= 0 or peak_index >= magnitudes.size - 1:
+        return float(freqs[peak_index])
+    alpha = magnitudes[peak_index - 1]
+    beta = magnitudes[peak_index]
+    gamma = magnitudes[peak_index + 1]
+    denominator = alpha - 2.0 * beta + gamma
+    if abs(denominator) < 1e-12:
+        return float(freqs[peak_index])
+    offset = 0.5 * (alpha - gamma) / denominator
+    bin_width = float(freqs[1] - freqs[0])
+    return float(freqs[peak_index] + offset * bin_width)
+
+
+def get_dominant_frequency(signal: np.ndarray, sampling_rate: float) -> float:
+    freqs, magnitudes = _compute_windowed_spectrum(signal, sampling_rate)
+    if magnitudes.size == 0:
+        return 0.0
+    magnitudes[0] = 0.0
+    band_mask = (freqs >= 0.1) & (freqs <= 5.0)
+    if not np.any(band_mask):
+        return 0.0
+    band_magnitudes = np.where(band_mask, magnitudes, 0.0)
+    peak_index = int(np.argmax(band_magnitudes))
+    return _parabolic_peak_frequency(freqs, magnitudes, peak_index)
+
+
+def theoretical_acceleration(frequency_hz: float, amplitude_m: float) -> float:
+    return float((2.0 * np.pi * frequency_hz) ** 2 * amplitude_m)
+
+
+def extract_core_features(
+    signal: np.ndarray,
+    sampling_rate: float,
+    config: DataConfig,
+    known_frequency_hz: float | None = None,
+) -> np.ndarray:
+    frequency_hz = known_frequency_hz if known_frequency_hz is not None else get_dominant_frequency(signal, sampling_rate)
+    x_rms = calculate_rms(signal)
+    return np.asarray(
+        [
+            frequency_hz,
+            frequency_hz * frequency_hz,
+            x_rms,
+            theoretical_acceleration(frequency_hz, config.harmonic_amplitude_m),
+        ],
+        dtype=np.float32,
+    )
+
+
+def _value_frame(df: pd.DataFrame, config: DataConfig, sensor_code: str, value_column: str) -> pd.DataFrame:
+    sensor_df = df.loc[df[config.code_column] == sensor_code, [config.time_column, value_column]].copy()
+    sensor_df = sensor_df.sort_values(config.time_column)
+    sensor_df = sensor_df.drop_duplicates(subset=config.time_column, keep="first")
+    sensor_df[config.time_column] = sensor_df[config.time_column].astype("float64")
+    sensor_df[value_column] = sensor_df[value_column].astype("float32")
+    return sensor_df
+
+
+def load_aligned_signals(file_path: Path, config: DataConfig) -> tuple[np.ndarray, np.ndarray, np.ndarray, int]:
+    df = pd.read_csv(file_path)
+    base_df = _value_frame(df, config, config.base_sensor_code, config.base_axis)
+    response_df = _value_frame(df, config, config.response_sensor_code, config.response_axis)
+    if base_df.empty or response_df.empty:
+        raise ValueError(f"Missing required sensor in {file_path.name}")
+
+    aligned = base_df.rename(columns={config.base_axis: "base_signal"}).merge(
+        response_df.rename(columns={config.response_axis: "response_signal"}),
+        on=config.time_column,
+        how="left",
+    )
+    interpolation_count = int(aligned["response_signal"].isna().sum())
+    aligned["response_signal"] = aligned["response_signal"].interpolate(
+        method=config.interpolation_method,
+        limit_direction="both",
+    ).ffill().bfill()
+    if aligned["response_signal"].isna().any():
+        raise ValueError(f"Remaining NaN after interpolation in {file_path.name}")
+
+    time_values = aligned[config.time_column].to_numpy(dtype=np.float64)
+    x_values = aligned["base_signal"].to_numpy(dtype=np.float32)
+    y_values = aligned["response_signal"].to_numpy(dtype=np.float32)
+    return time_values, x_values, y_values, interpolation_count
+
+
+@dataclass
+class SampleRecord:
+    file_path: Path
+    frequency_hz: float
+    features: np.ndarray
+    x_rms: float
+    y_rms: float
+    target_tr: float
+    time_middle: np.ndarray
+    x_middle: np.ndarray
+    y_middle: np.ndarray
+    sampling_rate: float
+    interpolation_count: int = 0
+
+
+@dataclass
+class LoadReport:
+    loaded_files: list[str] = field(default_factory=list)
+    skipped_files: list[tuple[str, str]] = field(default_factory=list)
+    interpolated_files: dict[str, int] = field(default_factory=dict)
+
+    def to_lines(self) -> list[str]:
+        lines = [f"[all_data] loaded={len(self.loaded_files)} skipped={len(self.skipped_files)}"]
+        for file_name, reason in self.skipped_files:
+            lines.append(f"  - skipped {file_name}: {reason}")
+        for file_name, count in self.interpolated_files.items():
+            if count > 0:
+                lines.append(f"  - interpolated {file_name}: missing_points={count}")
+        return lines
+
+
+def list_harmonic_files(config: DataConfig) -> list[Path]:
+    files = sorted(config.data_root.rglob(config.harmonic_pattern))
+    return files
+
+
+def build_record_from_file(file_path: Path, config: DataConfig) -> SampleRecord:
+    time_values, x_values, y_values, interpolation_count = load_aligned_signals(file_path, config)
+    time_middle, x_middle, y_middle = get_middle_segment(time_values, x_values, y_values, config)
+    sampling_rate = estimate_sampling_rate(time_middle)
+    frequency_hz = get_dominant_frequency(x_middle, sampling_rate)
+
+    x_rms = calculate_rms(x_middle)
+    y_rms = calculate_rms(y_middle)
+    target_tr = float(y_rms / max(x_rms, config.normalization_eps))
+    features = extract_core_features(x_middle, sampling_rate, config, known_frequency_hz=frequency_hz)
+
+    return SampleRecord(
+        file_path=file_path,
+        frequency_hz=float(frequency_hz),
+        features=features,
+        x_rms=float(x_rms),
+        y_rms=float(y_rms),
+        target_tr=target_tr,
+        time_middle=np.asarray(time_middle, dtype=np.float64),
+        x_middle=np.asarray(x_middle, dtype=np.float32).reshape(-1),
+        y_middle=np.asarray(y_middle, dtype=np.float32).reshape(-1),
+        sampling_rate=float(sampling_rate),
+        interpolation_count=interpolation_count,
+    )
+
+
+def load_all_records(config: ExperimentConfig | DataConfig) -> tuple[list[SampleRecord], LoadReport]:
+    data_config = config.data if isinstance(config, ExperimentConfig) else config
+    records: list[SampleRecord] = []
+    report = LoadReport()
+
+    for file_path in list_harmonic_files(data_config):
+        try:
+            record = build_record_from_file(file_path, data_config)
+        except ValueError as error:
+            report.skipped_files.append((file_path.name, str(error)))
+            continue
+        records.append(record)
+        report.loaded_files.append(file_path.name)
+        if record.interpolation_count > 0:
+            report.interpolated_files[file_path.name] = record.interpolation_count
+
+    if not records:
+        raise RuntimeError("No usable harmonic records found in data_root.")
+    return records, report
+
+
+def records_to_frame(records: list[SampleRecord]) -> pd.DataFrame:
+    rows = []
+    for record in records:
+        row = {
+            "file_name": record.file_path.name,
+            "frequency_hz": record.frequency_hz,
+            "x_rms": record.x_rms,
+            "y_rms": record.y_rms,
+            "target_tr": record.target_tr,
+            "sampling_rate": record.sampling_rate,
+        }
+        for feature_name, feature_value in zip(CORE_FEATURE_NAMES, record.features.tolist()):
+            row[feature_name] = float(feature_value)
+        rows.append(row)
+    return pd.DataFrame(rows)
+
+
+def report_to_text(report: LoadReport) -> str:
+    return "\n".join(report.to_lines())

scripts/evaluate.py

@@ -0,0 +1,141 @@
+from __future__ import annotations
+
+import argparse
+import pickle
+from pathlib import Path
+
+import matplotlib.pyplot as plt
+import numpy as np
+import pandas as pd
+
+try:
+    from .config import CORE_FEATURE_NAMES, evaluation_dir, make_experiment_config
+    from .dataset import load_all_records, records_to_frame, report_to_text
+except ImportError:
+    from config import CORE_FEATURE_NAMES, evaluation_dir, make_experiment_config
+    from dataset import load_all_records, records_to_frame, report_to_text
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="Evaluate the final all-data task1 model.")
+    parser.add_argument("--checkpoint", type=str, default=None)
+    return parser.parse_args()
+
+
+def resolve_checkpoint_path(project_root: Path, checkpoint_arg: str | None) -> Path:
+    if checkpoint_arg:
+        return Path(checkpoint_arg).resolve()
+    return project_root / "checkpoints_final" / "task1_final" / "task1_final_model.pkl"
+
+
+def relative_error_percent(true_value: float, pred_value: float) -> float:
+    return abs(pred_value - true_value) / max(abs(true_value), 1e-12) * 100.0
+
+
+def build_dense_curve(model, result_df: pd.DataFrame, config) -> pd.DataFrame:
+    plot_df = result_df.sort_values("frequency_hz").reset_index(drop=True)
+    dense_freq = np.linspace(
+        float(plot_df["frequency_hz"].min()),
+        float(plot_df["frequency_hz"].max()),
+        config.train.dense_curve_points,
+    )
+    dense_x_rms = np.interp(dense_freq, plot_df["frequency_hz"], plot_df["x_rms"])
+    dense_true_rms = np.interp(dense_freq, plot_df["frequency_hz"], plot_df["true_rms"])
+    dense_features = np.column_stack(
+        [
+            dense_freq,
+            dense_freq**2,
+            dense_x_rms,
+            (2.0 * np.pi * dense_freq) ** 2 * config.data.harmonic_amplitude_m,
+        ]
+    )
+    dense_pred_tr = model.predict(dense_features).clip(min=config.data.normalization_eps)
+    dense_pred_rms = dense_pred_tr * dense_x_rms
+    return pd.DataFrame(
+        {
+            "frequency_hz": dense_freq,
+            "x_rms_interp": dense_x_rms,
+            "true_rms_interp": dense_true_rms,
+            "pred_tr": dense_pred_tr,
+            "pred_rms": dense_pred_rms,
+        }
+    )
+
+
+def save_plot(result_df: pd.DataFrame, dense_df: pd.DataFrame, figure_path: Path) -> None:
+    plot_df = result_df.sort_values("frequency_hz").reset_index(drop=True)
+    fig, axes = plt.subplots(2, 1, figsize=(12, 9))
+    fig.suptitle("Task1 Final Model Evaluation | All Harmonic Data")
+
+    axes[0].plot(dense_df["frequency_hz"], dense_df["true_rms_interp"], color="tab:blue", linewidth=2.0, label="True RMS Guide Curve")
+    axes[0].plot(dense_df["frequency_hz"], dense_df["pred_rms"], color="tab:orange", linewidth=2.0, label="Pred RMS Dense Curve")
+    axes[0].scatter(plot_df["frequency_hz"], plot_df["true_rms"], color="tab:blue", s=28, zorder=3, label="True RMS Samples")
+    axes[0].scatter(plot_df["frequency_hz"], plot_df["pred_rms"], color="tab:orange", s=22, zorder=3, label="Pred RMS Samples")
+    axes[0].set_xlabel("Frequency (Hz)")
+    axes[0].set_ylabel("RMS")
+    axes[0].grid(True, alpha=0.3)
+    axes[0].legend()
+
+    axes[1].bar(plot_df["frequency_hz"].astype(str), plot_df["relative_error_percent"], color="tab:orange")
+    axes[1].set_xlabel("Frequency (Hz)")
+    axes[1].set_ylabel("Relative Error (%)")
+    axes[1].grid(True, axis="y", alpha=0.3)
+    axes[1].tick_params(axis="x", labelrotation=45)
+
+    plt.tight_layout()
+    plt.savefig(figure_path, dpi=180, bbox_inches="tight")
+    plt.close(fig)
+
+
+def main() -> None:
+    args = parse_args()
+    config = make_experiment_config()
+    ckpt_path = resolve_checkpoint_path(config.data.project_root, args.checkpoint)
+    with ckpt_path.open("rb") as handle:
+        payload = pickle.load(handle)
+    model = payload["model"]
+
+    records, report = load_all_records(config)
+    print(report_to_text(report))
+    data_df = records_to_frame(records)
+    feature_matrix = data_df.loc[:, CORE_FEATURE_NAMES].to_numpy()
+    pred_tr = model.predict(feature_matrix).clip(min=config.data.normalization_eps)
+    pred_rms = pred_tr * data_df["x_rms"].to_numpy()
+    true_rms = data_df["y_rms"].to_numpy()
+
+    result_df = data_df.copy()
+    result_df["pred_tr"] = pred_tr
+    result_df["pred_rms"] = pred_rms
+    result_df["true_rms"] = true_rms
+    result_df["relative_error_percent"] = [
+        relative_error_percent(true, pred) for true, pred in zip(true_rms, pred_rms)
+    ]
+    result_df = result_df.sort_values(["frequency_hz", "file_name"]).reset_index(drop=True)
+
+    save_dir = evaluation_dir(config)
+    csv_path = save_dir / "evaluation_all_samples.csv"
+    dense_csv_path = save_dir / config.train.evaluation_dense_csv_name
+    figure_path = save_dir / "evaluation_all_curve.png"
+    dense_df = build_dense_curve(model, result_df, config)
+    result_df.to_csv(csv_path, index=False)
+    dense_df.to_csv(dense_csv_path, index=False)
+    save_plot(result_df, dense_df, figure_path)
+
+    summary = {
+        "count": len(result_df),
+        "mean_error_percent": float(result_df["relative_error_percent"].mean()),
+        "median_error_percent": float(result_df["relative_error_percent"].median()),
+        "max_error_percent": float(result_df["relative_error_percent"].max()),
+        "min_error_percent": float(result_df["relative_error_percent"].min()),
+    }
+    print(f"Checkpoint: {ckpt_path}")
+    print(f"Model: {payload['model_name']}")
+    print(f"Summary: {summary}")
+    print(f"CSV saved to: {csv_path}")
+    print(f"Dense curve CSV saved to: {dense_csv_path}")
+    print(f"Figure saved to: {figure_path}")
+    print(result_df[["file_name", "frequency_hz", "true_rms", "pred_rms", "relative_error_percent"]].to_string(index=False))
+
+
+if __name__ == "__main__":
+    main()

scripts/inference_utils.py

@@ -0,0 +1,72 @@
+from __future__ import annotations
+
+from functools import lru_cache
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+
+def _task1_curve_candidates(project_root: Path) -> tuple[Path, ...]:
+    base_dir = project_root / "evaluation_outputs" / "task1_final"
+    return (
+        base_dir / "task1_final_dense_curve.csv",
+        base_dir / "evaluation_dense_curve.csv",
+    )
+
+
+@lru_cache(maxsize=8)
+def load_task1_dense_curve(project_root_str: str) -> tuple[np.ndarray, np.ndarray, np.ndarray] | None:
+    project_root = Path(project_root_str)
+    for path in _task1_curve_candidates(project_root):
+        if not path.exists():
+            continue
+        try:
+            curve_df = pd.read_csv(path)
+        except Exception:
+            continue
+        required = {"frequency_hz", "pred_tr", "x_rms_interp"}
+        if not required.issubset(set(curve_df.columns)):
+            continue
+        cleaned = curve_df[["frequency_hz", "pred_tr", "x_rms_interp"]].copy()
+        cleaned["frequency_hz"] = pd.to_numeric(cleaned["frequency_hz"], errors="coerce")
+        cleaned["pred_tr"] = pd.to_numeric(cleaned["pred_tr"], errors="coerce")
+        cleaned["x_rms_interp"] = pd.to_numeric(cleaned["x_rms_interp"], errors="coerce")
+        cleaned = cleaned.dropna()
+        if cleaned.empty:
+            continue
+        cleaned = cleaned.sort_values("frequency_hz").drop_duplicates(subset="frequency_hz", keep="first")
+        return (
+            cleaned["frequency_hz"].to_numpy(dtype=np.float64),
+            cleaned["pred_tr"].to_numpy(dtype=np.float64),
+            cleaned["x_rms_interp"].to_numpy(dtype=np.float64),
+        )
+    return None
+
+
+def predict_task1_tr(
+    model,
+    *,
+    feature_vector: np.ndarray,
+    frequency_hz: float,
+    normalization_eps: float,
+    project_root: Path,
+    prefer_curve: bool = True,
+) -> tuple[float, str]:
+    features = np.asarray(feature_vector, dtype=np.float64).reshape(1, -1)
+    tr_value = max(float(model.predict(features)[0]), normalization_eps)
+    source = "model_direct"
+
+    if prefer_curve:
+        curve = load_task1_dense_curve(str(project_root.resolve()))
+        if curve is not None and features.shape[1] >= 3:
+            freq_grid, _tr_grid, x_ref_grid = curve
+            query_freq = float(np.clip(frequency_hz, freq_grid.min(), freq_grid.max()))
+            x_ref = float(np.interp(query_freq, freq_grid, x_ref_grid))
+            x_obs = float(features[0, 2])
+            ratio = x_obs / max(x_ref, normalization_eps)
+            if ratio > 1.15 or ratio < 0.85:
+                tr_value = max(tr_value / ratio, normalization_eps)
+                source = "model_amp_adapt"
+
+    return tr_value, source

scripts/predict_single.py

@@ -0,0 +1,119 @@
+from __future__ import annotations
+
+import argparse
+import pickle
+from pathlib import Path
+
+import matplotlib.pyplot as plt
+import numpy as np
+
+try:
+    from .config import CORE_FEATURE_NAMES, evaluation_dir, make_experiment_config
+    from .dataset import build_record_from_file
+    from .inference_utils import predict_task1_tr
+except ImportError:
+    from config import CORE_FEATURE_NAMES, evaluation_dir, make_experiment_config
+    from dataset import build_record_from_file
+    from inference_utils import predict_task1_tr
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="Predict task1 RMS from a single harmonic waveform CSV.")
+    parser.add_argument("--file", type=str, required=True)
+    parser.add_argument("--checkpoint", type=str, default=None)
+    return parser.parse_args()
+
+
+def resolve_checkpoint_path(project_root: Path, checkpoint_arg: str | None) -> Path:
+    if checkpoint_arg:
+        return Path(checkpoint_arg).resolve()
+    return project_root / "checkpoints_final" / "task1_final" / "task1_final_model.pkl"
+
+
+def save_prediction_figure(file_path: Path, record, pred_rms: float, output_dir: Path) -> Path:
+    time_middle = record.time_middle.reshape(-1)
+    x_middle = record.x_middle.reshape(-1)
+    y_middle = record.y_middle.reshape(-1)
+    sampling_rate = record.sampling_rate
+    fft_values = np.fft.rfft(x_middle - np.mean(x_middle))
+    freqs = np.fft.rfftfreq(x_middle.size, d=1.0 / sampling_rate)
+    amplitudes = np.abs(fft_values)
+
+    fig, axes = plt.subplots(4, 1, figsize=(12, 12))
+    fig.suptitle(f"Task1 Final Single-File Prediction | {file_path.name}")
+
+    axes[0].plot(time_middle, x_middle, color="tab:blue")
+    axes[0].set_title("Input Base Excitation (Middle Segment)")
+    axes[0].set_xlabel("Time")
+    axes[0].set_ylabel("Acceleration")
+    axes[0].grid(True, alpha=0.3)
+
+    axes[1].plot(time_middle, y_middle, color="tab:green")
+    axes[1].set_title("True Top Response (Middle Segment)")
+    axes[1].set_xlabel("Time")
+    axes[1].set_ylabel("Acceleration")
+    axes[1].grid(True, alpha=0.3)
+
+    axes[2].plot(freqs, amplitudes, color="tab:purple")
+    axes[2].axvline(record.frequency_hz, color="tab:red", linestyle="--", label=f"Frequency = {record.frequency_hz:.4f} Hz")
+    axes[2].set_xlim(0.0, 5.0)
+    axes[2].set_title("Input Spectrum")
+    axes[2].set_xlabel("Frequency (Hz)")
+    axes[2].set_ylabel("Amplitude")
+    axes[2].grid(True, alpha=0.3)
+    axes[2].legend()
+
+    axes[3].bar(["True RMS", "Pred RMS"], [record.y_rms, pred_rms], color=["tab:green", "tab:orange"])
+    error_percent = abs(pred_rms - record.y_rms) / max(abs(record.y_rms), 1e-12) * 100.0
+    axes[3].set_title(
+        f"True RMS = {record.y_rms:.6f} | Pred RMS = {pred_rms:.6f} | Error = {error_percent:.2f}%"
+    )
+    axes[3].set_ylabel("RMS")
+    axes[3].grid(True, axis="y", alpha=0.3)
+
+    plt.tight_layout()
+    output_dir.mkdir(parents=True, exist_ok=True)
+    figure_path = output_dir / f"{file_path.stem}_prediction.png"
+    plt.savefig(figure_path, dpi=180, bbox_inches="tight")
+    plt.close(fig)
+    return figure_path
+
+
+def main() -> None:
+    args = parse_args()
+    config = make_experiment_config()
+    ckpt_path = resolve_checkpoint_path(config.data.project_root, args.checkpoint)
+    with ckpt_path.open("rb") as handle:
+        payload = pickle.load(handle)
+    model = payload["model"]
+
+    file_path = Path(args.file).resolve()
+    record = build_record_from_file(file_path, config.data)
+    pred_tr, tr_source = predict_task1_tr(
+        model,
+        feature_vector=record.features,
+        frequency_hz=float(record.frequency_hz),
+        normalization_eps=float(config.data.normalization_eps),
+        project_root=config.data.project_root,
+        prefer_curve=True,
+    )
+    pred_rms = pred_tr * record.x_rms
+
+    out_dir = evaluation_dir(config)
+    fig_path = save_prediction_figure(file_path, record, pred_rms, out_dir)
+
+    print(f"Checkpoint: {ckpt_path}")
+    print(f"Model: {payload['model_name']}")
+    print(f"Input file: {file_path}")
+    for feature_name, feature_value in zip(CORE_FEATURE_NAMES, record.features.tolist()):
+        print(f"{feature_name}: {feature_value:.6f}")
+    print(f"Predicted TR: {pred_tr:.6f}")
+    print(f"TR Source: {tr_source}")
+    print(f"Predicted RMS: {pred_rms:.6f}")
+    print(f"True RMS: {record.y_rms:.6f}")
+    print(f"Relative RMS Error (%): {abs(pred_rms - record.y_rms) / max(abs(record.y_rms), 1e-12) * 100.0:.4f}")
+    print(f"Figure saved to: {fig_path}")
+
+
+if __name__ == "__main__":
+    main()

scripts/train_final.py

@@ -0,0 +1,178 @@
+from __future__ import annotations
+
+import pickle
+from dataclasses import asdict
+from pathlib import Path
+from typing import Any
+
+import matplotlib.pyplot as plt
+import numpy as np
+import pandas as pd
+from sklearn.neighbors import KNeighborsRegressor
+from sklearn.pipeline import Pipeline
+from sklearn.preprocessing import StandardScaler
+
+try:
+    from .config import CORE_FEATURE_NAMES, ExperimentConfig, checkpoint_dir, evaluation_dir, make_experiment_config
+    from .dataset import load_all_records, records_to_frame, report_to_text
+except ImportError:
+    from config import CORE_FEATURE_NAMES, ExperimentConfig, checkpoint_dir, evaluation_dir, make_experiment_config
+    from dataset import load_all_records, records_to_frame, report_to_text
+
+
+def build_final_model(config: ExperimentConfig) -> Pipeline:
+    return Pipeline(
+        [
+            ("scaler", StandardScaler()),
+            (
+                "model",
+                KNeighborsRegressor(
+                    n_neighbors=config.model.n_neighbors,
+                    weights="distance",
+                    p=config.model.distance_power,
+                ),
+            ),
+        ]
+    )
+
+
+def serialize_for_checkpoint(value: Any) -> Any:
+    if isinstance(value, Path):
+        return str(value)
+    if isinstance(value, dict):
+        return {key: serialize_for_checkpoint(sub_value) for key, sub_value in value.items()}
+    if isinstance(value, tuple):
+        return [serialize_for_checkpoint(item) for item in value]
+    if isinstance(value, list):
+        return [serialize_for_checkpoint(item) for item in value]
+    return value
+
+
+def relative_error_percent(true_value: float, pred_value: float) -> float:
+    return abs(pred_value - true_value) / max(abs(true_value), 1e-12) * 100.0
+
+
+def build_dense_curve(model: Pipeline, result_df: pd.DataFrame, config: ExperimentConfig) -> pd.DataFrame:
+    plot_df = result_df.sort_values("frequency_hz").reset_index(drop=True)
+    dense_freq = np.linspace(
+        float(plot_df["frequency_hz"].min()),
+        float(plot_df["frequency_hz"].max()),
+        config.train.dense_curve_points,
+    )
+    dense_x_rms = np.interp(dense_freq, plot_df["frequency_hz"], plot_df["x_rms"])
+    dense_true_rms = np.interp(dense_freq, plot_df["frequency_hz"], plot_df["true_rms"])
+    dense_features = np.column_stack(
+        [
+            dense_freq,
+            dense_freq**2,
+            dense_x_rms,
+            (2.0 * np.pi * dense_freq) ** 2 * config.data.harmonic_amplitude_m,
+        ]
+    )
+    dense_pred_tr = model.predict(dense_features).clip(min=config.data.normalization_eps)
+    dense_pred_rms = dense_pred_tr * dense_x_rms
+    return pd.DataFrame(
+        {
+            "frequency_hz": dense_freq,
+            "x_rms_interp": dense_x_rms,
+            "true_rms_interp": dense_true_rms,
+            "pred_tr": dense_pred_tr,
+            "pred_rms": dense_pred_rms,
+        }
+    )
+
+
+def save_fit_plot(result_df: pd.DataFrame, dense_df: pd.DataFrame, figure_path: Path) -> None:
+    plot_df = result_df.sort_values("frequency_hz").reset_index(drop=True)
+    fig, axes = plt.subplots(2, 1, figsize=(12, 9))
+    fig.suptitle("Task1 Final Model | All Harmonic Data Fit")
+
+    axes[0].plot(dense_df["frequency_hz"], dense_df["true_rms_interp"], color="tab:blue", linewidth=2.0, label="True RMS Guide Curve")
+    axes[0].plot(dense_df["frequency_hz"], dense_df["pred_rms"], color="tab:orange", linewidth=2.0, label="Pred RMS Dense Curve")
+    axes[0].scatter(plot_df["frequency_hz"], plot_df["true_rms"], color="tab:blue", s=28, zorder=3, label="True RMS Samples")
+    axes[0].scatter(plot_df["frequency_hz"], plot_df["pred_rms"], color="tab:orange", s=22, zorder=3, label="Pred RMS Samples")
+    axes[0].set_xlabel("Frequency (Hz)")
+    axes[0].set_ylabel("RMS")
+    axes[0].grid(True, alpha=0.3)
+    axes[0].legend()
+
+    axes[1].bar(plot_df["frequency_hz"].astype(str), plot_df["relative_error_percent"], color="tab:orange")
+    axes[1].set_xlabel("Frequency (Hz)")
+    axes[1].set_ylabel("Relative Error (%)")
+    axes[1].grid(True, axis="y", alpha=0.3)
+    axes[1].tick_params(axis="x", labelrotation=45)
+
+    plt.tight_layout()
+    plt.savefig(figure_path, dpi=180, bbox_inches="tight")
+    plt.close(fig)
+
+
+def train_final() -> None:
+    config = make_experiment_config()
+    records, report = load_all_records(config)
+    print(report_to_text(report))
+
+    data_df = records_to_frame(records)
+    feature_matrix = data_df.loc[:, CORE_FEATURE_NAMES].to_numpy()
+    target_tr = data_df["target_tr"].to_numpy()
+
+    model = build_final_model(config)
+    model.fit(feature_matrix, target_tr)
+
+    pred_tr = model.predict(feature_matrix).clip(min=config.data.normalization_eps)
+    pred_rms = pred_tr * data_df["x_rms"].to_numpy()
+    true_rms = data_df["y_rms"].to_numpy()
+
+    result_df = data_df.copy()
+    result_df["pred_tr"] = pred_tr
+    result_df["pred_rms"] = pred_rms
+    result_df["true_rms"] = true_rms
+    result_df["relative_error_percent"] = [
+        relative_error_percent(true, pred) for true, pred in zip(true_rms, pred_rms)
+    ]
+    result_df = result_df.sort_values(["frequency_hz", "file_name"]).reset_index(drop=True)
+
+    mean_error = float(result_df["relative_error_percent"].mean())
+    median_error = float(result_df["relative_error_percent"].median())
+    max_error = float(result_df["relative_error_percent"].max())
+
+    ckpt_dir = checkpoint_dir(config)
+    eval_dir = evaluation_dir(config)
+    csv_path = eval_dir / config.train.fit_csv_name
+    fig_path = eval_dir / config.train.fit_figure_name
+    dense_csv_path = eval_dir / config.train.dense_curve_csv_name
+    model_path = ckpt_dir / config.train.model_name
+
+    dense_df = build_dense_curve(model, result_df, config)
+    result_df.to_csv(csv_path, index=False)
+    dense_df.to_csv(dense_csv_path, index=False)
+    save_fit_plot(result_df, dense_df, fig_path)
+
+    payload = {
+        "model_name": config.model.model_name,
+        "model": model,
+        "config": serialize_for_checkpoint(asdict(config)),
+        "feature_names": list(CORE_FEATURE_NAMES),
+        "summary": {
+            "count": len(result_df),
+            "mean_error_percent": mean_error,
+            "median_error_percent": median_error,
+            "max_error_percent": max_error,
+        },
+    }
+    with model_path.open("wb") as handle:
+        pickle.dump(payload, handle)
+
+    print(f"Final model saved to: {model_path}")
+    print(f"Fit CSV saved to: {csv_path}")
+    print(f"Dense curve CSV saved to: {dense_csv_path}")
+    print(f"Fit figure saved to: {fig_path}")
+    print(
+        f"All-data self-fit summary | count={len(result_df)} | "
+        f"mean={mean_error:.4f}% | median={median_error:.4f}% | max={max_error:.4f}%"
+    )
+    print(result_df[["file_name", "frequency_hz", "true_rms", "pred_rms", "relative_error_percent"]].to_string(index=False))
+
+
+if __name__ == "__main__":
+    train_final()

scripts_2/README.md

@@ -0,0 +1,43 @@
+# 结构动力特性识别脚本 (Free-Vibration Identification)
+
+## 项目简介
+本目录用于完成建筑结构动力特性识别任务。脚本基于专门的自由衰减数据文件，提取顶部响应传感器 `WSMS00007` 的 Z 向信号，识别结构一阶自振频率与阻尼比，并输出可视化诊断结果。
+
+## 技术路线
+本目录的分析流程围绕自由衰减信号展开，包含信号读取、主频识别、滤波降噪与阻尼估计四个步骤：
+
+1. **顶部响应读取**
+   从 `free_vib` 目录下的 CSV 文件中读取 `WSMS00007` 的 `value3` 列，作为结构顶部响应信号。
+2. **频域主频识别**
+   对原始自由衰减信号执行 FFT，在 `0.1Hz ~ 5.0Hz` 范围内寻找主峰，识别一阶自振频率。
+3. **带通滤波**
+   以识别出的主频为中心，构造巴特沃斯带通滤波器，对原始信号进行滤波，得到平滑的衰减振动波形。
+4. **阻尼比计算**
+   从滤波信号中提取波峰，自动选择包络线质量较高的一段，通过对数衰减率计算平均衰减量与阻尼比。
+
+## 核心模块说明
+
+- **`signal_utils.py`**
+  共享信号处理工具。负责传感器常量定义、采样率估计、稳态段截取、FFT 主频识别、抛物线插值、带通滤波与波峰筛选。
+- **`task3_identify.py`**
+  任务主入口。负责校验自由衰减数据来源、执行一阶频率与阻尼比识别，并生成包含原始波形、滤波波形和包络线的诊断图。
+
+## 使用指南
+
+**1. 运行任务脚本**
+对单个自由衰减 CSV 文件执行结构动力特性识别：
+```bash
+python scripts_2\task3_identify.py --file downloads\Non_TMD\free_vib\free_decay.csv
+```
+
+**2. 指定输出目录**
+将图像与诊断结果输出到自定义目录：
+```bash
+python scripts_2\task3_identify.py --file downloads\Non_TMD\free_vib\free_decay.csv --output-dir evaluation_outputs\task3_custom
+```
+
+**3. 调整滤波带宽**
+通过命令行参数修改主频两侧的带宽范围：
+```bash
+python scripts_2\task3_identify.py --file downloads\Non_TMD\free_vib\free_decay.csv --bandwidth 0.25
+```

scripts_2/__init__.py

@@ -0,0 +1 @@
+"""Signal-processing scripts for free-vibration identification."""

scripts_2/signal_utils.py

@@ -0,0 +1,273 @@
+from __future__ import annotations
+
+from pathlib import Path
+import re
+
+import numpy as np
+import pandas as pd
+from scipy.signal import butter, filtfilt, find_peaks
+
+
+TOP_RESPONSE_SENSOR = "WSMS00007"
+TOP_RESPONSE_AXIS = "value3"
+BASE_REFERENCE_SENSOR = "WSMS00012"
+BASE_REFERENCE_AXIS = "value1"
+TIME_COLUMN = "time"
+CODE_COLUMN = "code"
+
+
+def ensure_parent_dir(path: Path) -> Path:
+    path = Path(path).resolve()
+    path.mkdir(parents=True, exist_ok=True)
+    return path
+
+
+def parse_frequency_from_filename(file_name: str) -> float | None:
+    match = re.search(r"(\d+(?:\.\d+)?)Hz", file_name, flags=re.IGNORECASE)
+    if match is None:
+        return None
+    return float(match.group(1))
+
+
+def estimate_sampling_rate(time_values: np.ndarray) -> float:
+    time_values = np.asarray(time_values, dtype=np.float64).reshape(-1)
+    if time_values.size < 2:
+        raise ValueError("At least two time samples are required to estimate sampling rate.")
+    dt = np.diff(time_values)
+    dt = dt[np.isfinite(dt)]
+    dt = dt[dt > 0.0]
+    if dt.size == 0:
+        raise ValueError("Failed to infer a positive sampling interval from the CSV time column.")
+    return float(1.0 / np.median(dt))
+
+
+def calculate_rms(signal: np.ndarray) -> float:
+    signal = np.asarray(signal, dtype=np.float64).reshape(-1)
+    return float(np.sqrt(np.mean(np.square(signal))))
+
+
+def read_sensor_signal(
+    file_path: Path,
+    sensor_code: str = TOP_RESPONSE_SENSOR,
+    value_column: str = TOP_RESPONSE_AXIS,
+) -> tuple[np.ndarray, np.ndarray]:
+    df = pd.read_csv(file_path)
+    sensor_df = df.loc[df[CODE_COLUMN] == sensor_code, [TIME_COLUMN, value_column]].copy()
+    if sensor_df.empty:
+        raise ValueError(f"Sensor {sensor_code}/{value_column} was not found in {file_path.name}.")
+
+    sensor_df = sensor_df.sort_values(TIME_COLUMN)
+    sensor_df = sensor_df.drop_duplicates(subset=TIME_COLUMN, keep="first")
+    sensor_df[TIME_COLUMN] = sensor_df[TIME_COLUMN].astype("float64")
+    sensor_df[value_column] = sensor_df[value_column].astype("float64")
+
+    time_values = sensor_df[TIME_COLUMN].to_numpy(dtype=np.float64)
+    signal = sensor_df[value_column].to_numpy(dtype=np.float64)
+    if time_values.size < 8:
+        raise ValueError(f"Too few usable samples in {file_path.name}.")
+    return time_values, signal
+
+
+def get_middle_segment(
+    time_values: np.ndarray,
+    signal: np.ndarray,
+    start_ratio: float = 0.20,
+    end_ratio: float = 0.80,
+    steady_window_ratio: float = 0.25,
+    stride_ratio: float = 0.05,
+    min_segment_length: int = 512,
+    subwindow_count: int = 4,
+) -> tuple[np.ndarray, np.ndarray]:
+    signal = np.asarray(signal, dtype=np.float64).reshape(-1)
+    time_values = np.asarray(time_values, dtype=np.float64).reshape(-1)
+    if signal.size != time_values.size:
+        raise ValueError("Time and signal arrays must have the same length.")
+
+    length = signal.size
+    search_start = int(length * start_ratio)
+    search_end = int(length * end_ratio)
+    search_start = max(0, min(search_start, length - 1))
+    search_end = max(search_start + 1, min(search_end, length))
+
+    if search_end - search_start < min_segment_length:
+        center = length // 2
+        half = min_segment_length // 2
+        search_start = max(0, center - half)
+        search_end = min(length, search_start + min_segment_length)
+        search_start = max(0, search_end - min_segment_length)
+
+    search_length = search_end - search_start
+    window_length = max(min_segment_length, int(length * steady_window_ratio))
+    window_length = min(window_length, search_length)
+    if search_length <= window_length:
+        return time_values[search_start:search_end], signal[search_start:search_end]
+
+    stride = max(1, int(length * stride_ratio))
+    candidate_signal = signal[search_start:search_end]
+    candidate_rms = calculate_rms(candidate_signal)
+    best_score = float("inf")
+    best_slice = slice(search_start, search_start + window_length)
+
+    for window_start in range(search_start, search_end - window_length + 1, stride):
+        window_end = window_start + window_length
+        window_signal = signal[window_start:window_end]
+        split_windows = np.array_split(window_signal, subwindow_count)
+        split_rms = np.asarray([calculate_rms(chunk) for chunk in split_windows], dtype=np.float64)
+        split_peaks = np.asarray([float(np.max(np.abs(chunk))) for chunk in split_windows], dtype=np.float64)
+
+        rms_cv = float(split_rms.std() / max(split_rms.mean(), 1e-8))
+        peak_cv = float(split_peaks.std() / max(split_peaks.mean(), 1e-8))
+        window_rms = calculate_rms(window_signal)
+        score = rms_cv + 0.35 * peak_cv - 0.05 * (window_rms / max(candidate_rms, 1e-8))
+        if score < best_score:
+            best_score = score
+            best_slice = slice(window_start, window_end)
+
+    return time_values[best_slice], signal[best_slice]
+
+
+def compute_windowed_spectrum(
+    signal: np.ndarray,
+    sampling_rate: float,
+    zero_padding_factor: int = 8,
+) -> tuple[np.ndarray, np.ndarray]:
+    signal = np.asarray(signal, dtype=np.float64).reshape(-1)
+    if signal.size < 4:
+        raise ValueError("Signal is too short for FFT analysis.")
+
+    centered = signal - np.mean(signal)
+    window = np.hanning(centered.size)
+    fft_size = int(2 ** np.ceil(np.log2(max(centered.size * zero_padding_factor, centered.size))))
+    fft_values = np.fft.rfft(centered * window, n=fft_size)
+    freqs = np.fft.rfftfreq(fft_size, d=1.0 / sampling_rate)
+    magnitudes = np.abs(fft_values)
+    return freqs, magnitudes
+
+
+def parabolic_peak_frequency(freqs: np.ndarray, magnitudes: np.ndarray, peak_index: int) -> float:
+    if peak_index <= 0 or peak_index >= magnitudes.size - 1:
+        return float(freqs[peak_index])
+
+    alpha = float(magnitudes[peak_index - 1])
+    beta = float(magnitudes[peak_index])
+    gamma = float(magnitudes[peak_index + 1])
+    denominator = alpha - 2.0 * beta + gamma
+    if abs(denominator) < 1e-12:
+        return float(freqs[peak_index])
+
+    offset = 0.5 * (alpha - gamma) / denominator
+    bin_width = float(freqs[1] - freqs[0])
+    return float(freqs[peak_index] + offset * bin_width)
+
+
+def dominant_frequency_in_band(
+    signal: np.ndarray,
+    sampling_rate: float,
+    min_hz: float = 0.1,
+    max_hz: float = 5.0,
+    zero_padding_factor: int = 8,
+) -> tuple[float, np.ndarray, np.ndarray, int]:
+    freqs, magnitudes = compute_windowed_spectrum(
+        signal=signal,
+        sampling_rate=sampling_rate,
+        zero_padding_factor=zero_padding_factor,
+    )
+    band_mask = (freqs >= min_hz) & (freqs <= max_hz)
+    if not np.any(band_mask):
+        raise ValueError(f"No FFT bins fall inside {min_hz:.2f} Hz to {max_hz:.2f} Hz.")
+
+    band_indices = np.flatnonzero(band_mask)
+    peak_index = int(band_indices[np.argmax(magnitudes[band_mask])])
+    peak_frequency = parabolic_peak_frequency(freqs, magnitudes, peak_index)
+    return peak_frequency, freqs, magnitudes, peak_index
+
+
+def bandpass_filter(
+    signal: np.ndarray,
+    sampling_rate: float,
+    center_hz: float,
+    bandwidth_hz: float = 0.3,
+    order: int = 4,
+) -> np.ndarray:
+    nyquist = 0.5 * sampling_rate
+    low_hz = max(0.05, center_hz - bandwidth_hz)
+    high_hz = min(center_hz + bandwidth_hz, nyquist * 0.98)
+    if not low_hz < high_hz:
+        raise ValueError("Invalid band-pass range. Check the estimated natural frequency and sampling rate.")
+
+    b, a = butter(order, [low_hz / nyquist, high_hz / nyquist], btype="bandpass")
+    return filtfilt(b, a, np.asarray(signal, dtype=np.float64))
+
+
+def detect_decay_peaks(filtered_signal: np.ndarray, sampling_rate: float, frequency_hz: float) -> np.ndarray:
+    filtered_signal = np.asarray(filtered_signal, dtype=np.float64).reshape(-1)
+    samples_per_cycle = max(int(round(sampling_rate / max(frequency_hz, 1e-6))), 3)
+    min_peak_distance = max(1, int(0.65 * samples_per_cycle))
+    peaks, _ = find_peaks(filtered_signal, distance=min_peak_distance)
+    if peaks.size < 4:
+        peaks, _ = find_peaks(np.abs(filtered_signal), distance=max(1, min_peak_distance // 2))
+    return peaks.astype(int)
+
+
+def select_decay_peak_window(
+    filtered_signal: np.ndarray,
+    peak_indices: np.ndarray,
+    sampling_rate: float,
+    min_peaks: int = 5,
+    max_peaks: int = 12,
+) -> dict[str, np.ndarray | float | int]:
+    filtered_signal = np.asarray(filtered_signal, dtype=np.float64).reshape(-1)
+    peak_indices = np.asarray(peak_indices, dtype=int).reshape(-1)
+    if peak_indices.size < min_peaks:
+        raise ValueError("Too few peaks were detected for log-decrement damping estimation.")
+
+    amplitudes = np.abs(filtered_signal[peak_indices])
+    tail_start = int(filtered_signal.size * 0.85)
+    noise_slice = filtered_signal[tail_start:] if tail_start < filtered_signal.size else filtered_signal
+    noise_floor = max(float(np.median(np.abs(noise_slice))), 1e-8)
+    useful_mask = amplitudes > max(3.0 * noise_floor, 0.05 * float(amplitudes.max()))
+    useful_indices = peak_indices[useful_mask]
+    useful_amplitudes = amplitudes[useful_mask]
+    if useful_indices.size < min_peaks:
+        useful_indices = peak_indices
+        useful_amplitudes = amplitudes
+
+    best_window: dict[str, np.ndarray | float | int] | None = None
+    best_score = -float("inf")
+    max_window_size = min(max_peaks, useful_indices.size)
+
+    for window_size in range(min_peaks, max_window_size + 1):
+        for start in range(1, useful_indices.size - window_size + 1):
+            stop = start + window_size
+            idx_window = useful_indices[start:stop]
+            amp_window = useful_amplitudes[start:stop]
+            if np.any(amp_window <= 0.0):
+                continue
+
+            time_window = idx_window.astype(np.float64) / sampling_rate
+            log_amp = np.log(amp_window)
+            slope, intercept = np.polyfit(time_window, log_amp, 1)
+            if slope >= 0.0:
+                continue
+
+            fit_values = slope * time_window + intercept
+            ss_res = float(np.sum((log_amp - fit_values) ** 2))
+            ss_tot = float(np.sum((log_amp - np.mean(log_amp)) ** 2))
+            r_squared = 1.0 - ss_res / max(ss_tot, 1e-12)
+            amplitude_ratio = float(amp_window[0] / max(amp_window[-1], 1e-12))
+            score = r_squared + 0.02 * np.log(max(amplitude_ratio, 1.0))
+            if score > best_score:
+                best_score = score
+                best_window = {
+                    "peak_indices": idx_window,
+                    "peak_amplitudes": amp_window,
+                    "time_window": time_window,
+                    "fit_log_amplitudes": fit_values,
+                    "fit_amplitudes": np.exp(fit_values),
+                    "r_squared": r_squared,
+                    "noise_floor": noise_floor,
+                }
+
+    if best_window is None:
+        raise ValueError("Failed to isolate a clean decay envelope segment for damping estimation.")
+    return best_window

scripts_2/task3_identify.py

@@ -0,0 +1,381 @@
+from __future__ import annotations
+
+import argparse
+from pathlib import Path
+import sys
+
+import matplotlib
+
+matplotlib.use("Agg")
+
+import matplotlib.pyplot as plt
+import numpy as np
+
+if __package__ is None or __package__ == "":
+    sys.path.append(str(Path(__file__).resolve().parents[1]))
+
+from scripts_2.signal_utils import (
+    TOP_RESPONSE_AXIS,
+    TOP_RESPONSE_SENSOR,
+    bandpass_filter,
+    detect_decay_peaks,
+    dominant_frequency_in_band,
+    ensure_parent_dir,
+    estimate_sampling_rate,
+    read_sensor_signal,
+    select_decay_peak_window,
+)
+
+
+def parse_args() -> argparse.Namespace:
+    project_root = Path(__file__).resolve().parents[1]
+    parser = argparse.ArgumentParser(description="Identify the first natural frequency and damping ratio from free-decay data.")
+    parser.add_argument(
+        "--file",
+        type=str,
+        default=str(project_root / "downloads" / "Non_TMD" / "free_vib" / "free_decay.csv"),
+        help="Path to a free-vibration CSV inside the dedicated free_vib folder.",
+    )
+    parser.add_argument(
+        "--output-dir",
+        type=str,
+        default=str(project_root / "evaluation_outputs" / "task3_identify"),
+        help="Directory for plots and optional exported diagnostics.",
+    )
+    parser.add_argument(
+        "--bandwidth",
+        type=float,
+        default=0.3,
+        help="Half-bandwidth (Hz) of the Butterworth band-pass filter around the identified natural frequency.",
+    )
+    parser.add_argument(
+        "--show",
+        action="store_true",
+        help="Display the generated figure in addition to saving it.",
+    )
+    return parser.parse_args()
+
+
+def validate_free_vibration_source(file_path: Path) -> None:
+    if "free_vib" not in {part.lower() for part in file_path.parts}:
+        raise ValueError(
+            f"{file_path.name} is not inside a dedicated free_vib folder. Forced-vibration tails are forbidden for task 3."
+        )
+
+
+def build_envelope_curve(
+    time_values: np.ndarray,
+    peak_times: np.ndarray,
+    fit_amplitudes: np.ndarray,
+) -> tuple[np.ndarray, np.ndarray]:
+    if peak_times.size < 2:
+        return time_values, np.zeros_like(time_values)
+    envelope = np.interp(time_values, peak_times, fit_amplitudes, left=fit_amplitudes[0], right=fit_amplitudes[-1])
+    return time_values, envelope
+
+
+def log_decrement_metrics(peak_amplitudes: np.ndarray) -> tuple[float, float, float]:
+    amplitudes = np.asarray(peak_amplitudes, dtype=np.float64).reshape(-1)
+    if amplitudes.size < 2 or np.any(amplitudes <= 0.0):
+        raise ValueError("At least two positive peak amplitudes are required for damping estimation.")
+    log_decrements = np.log(amplitudes[:-1] / amplitudes[1:])
+    mean_log_decrement = float(np.mean(log_decrements))
+    damping_ratio = float(mean_log_decrement / (2.0 * np.pi))
+    damping_ratio_exact = float(
+        mean_log_decrement / np.sqrt((2.0 * np.pi) ** 2 + mean_log_decrement**2)
+    )
+    return mean_log_decrement, damping_ratio, damping_ratio_exact
+
+
+def estimate_global_decay_metrics(
+    filtered_signal: np.ndarray,
+    peak_indices: np.ndarray,
+    sampling_rate: float,
+    min_peaks: int = 5,
+) -> dict[str, np.ndarray | float]:
+    filtered_signal = np.asarray(filtered_signal, dtype=np.float64).reshape(-1)
+    peak_indices = np.asarray(peak_indices, dtype=int).reshape(-1)
+    if peak_indices.size < min_peaks:
+        raise ValueError("Too few peaks were detected for global damping estimation.")
+
+    amplitudes = np.abs(filtered_signal[peak_indices])
+    tail_start = int(filtered_signal.size * 0.85)
+    noise_slice = filtered_signal[tail_start:] if tail_start < filtered_signal.size else filtered_signal
+    noise_floor = max(float(np.median(np.abs(noise_slice))), 1e-8)
+    useful_mask = amplitudes > max(3.0 * noise_floor, 0.05 * float(amplitudes.max()))
+    useful_indices = peak_indices[useful_mask]
+    useful_amplitudes = amplitudes[useful_mask]
+    if useful_indices.size < min_peaks:
+        useful_indices = peak_indices
+        useful_amplitudes = amplitudes
+
+    time_window = useful_indices.astype(np.float64) / sampling_rate
+    log_amp = np.log(np.maximum(useful_amplitudes, 1e-12))
+    slope, intercept = np.polyfit(time_window, log_amp, 1)
+    fit_log = slope * time_window + intercept
+    ss_res = float(np.sum((log_amp - fit_log) ** 2))
+    ss_tot = float(np.sum((log_amp - np.mean(log_amp)) ** 2))
+    r_squared = 1.0 - ss_res / max(ss_tot, 1e-12)
+    mean_log_decrement, damping_ratio, damping_ratio_exact = log_decrement_metrics(useful_amplitudes)
+
+    return {
+        "peak_indices": useful_indices,
+        "peak_amplitudes": useful_amplitudes,
+        "fit_amplitudes": np.exp(fit_log),
+        "r_squared": float(r_squared),
+        "mean_log_decrement": float(mean_log_decrement),
+        "damping_ratio": float(damping_ratio),
+        "damping_ratio_exact": float(damping_ratio_exact),
+    }
+
+
+def estimate_tail_from_max_peak_metrics(
+    filtered_signal: np.ndarray,
+    peak_indices: np.ndarray,
+    sampling_rate: float,
+    min_peaks: int = 5,
+) -> dict[str, np.ndarray | float]:
+    filtered_signal = np.asarray(filtered_signal, dtype=np.float64).reshape(-1)
+    peak_indices = np.asarray(peak_indices, dtype=int).reshape(-1)
+    if peak_indices.size < min_peaks:
+        raise ValueError("Too few peaks were detected for max-peak-tail damping estimation.")
+
+    amplitudes = np.abs(filtered_signal[peak_indices])
+    max_peak_pos = int(np.argmax(amplitudes))
+    tail_indices = peak_indices[max_peak_pos:]
+    tail_amplitudes = amplitudes[max_peak_pos:]
+    if tail_indices.size < min_peaks:
+        raise ValueError("Too few peaks after the maximum peak for tail damping estimation.")
+
+    tail_start = int(filtered_signal.size * 0.85)
+    noise_slice = filtered_signal[tail_start:] if tail_start < filtered_signal.size else filtered_signal
+    noise_floor = max(float(np.median(np.abs(noise_slice))), 1e-8)
+    useful_mask = tail_amplitudes > max(3.0 * noise_floor, 0.05 * float(tail_amplitudes.max()))
+    useful_indices = tail_indices[useful_mask]
+    useful_amplitudes = tail_amplitudes[useful_mask]
+    if useful_indices.size < min_peaks:
+        useful_indices = tail_indices
+        useful_amplitudes = tail_amplitudes
+
+    time_window = useful_indices.astype(np.float64) / sampling_rate
+    log_amp = np.log(np.maximum(useful_amplitudes, 1e-12))
+    slope, intercept = np.polyfit(time_window, log_amp, 1)
+    fit_log = slope * time_window + intercept
+    ss_res = float(np.sum((log_amp - fit_log) ** 2))
+    ss_tot = float(np.sum((log_amp - np.mean(log_amp)) ** 2))
+    r_squared = 1.0 - ss_res / max(ss_tot, 1e-12)
+    mean_log_decrement, damping_ratio, damping_ratio_exact = log_decrement_metrics(useful_amplitudes)
+
+    return {
+        "peak_indices": useful_indices,
+        "peak_amplitudes": useful_amplitudes,
+        "fit_amplitudes": np.exp(fit_log),
+        "r_squared": float(r_squared),
+        "mean_log_decrement": float(mean_log_decrement),
+        "damping_ratio": float(damping_ratio),
+        "damping_ratio_exact": float(damping_ratio_exact),
+    }
+
+
+def save_task3_figure(
+    file_path: Path,
+    output_dir: Path,
+    time_values: np.ndarray,
+    raw_signal: np.ndarray,
+    filtered_signal: np.ndarray,
+    peak_indices: np.ndarray,
+    selected_peak_indices: np.ndarray,
+    selected_peak_amplitudes: np.ndarray,
+    fit_amplitudes: np.ndarray,
+    natural_frequency_hz: float,
+    local_damping_ratio: float,
+    global_peak_indices: np.ndarray,
+    global_peak_amplitudes: np.ndarray,
+    global_fit_amplitudes: np.ndarray,
+    global_damping_ratio: float,
+    tail_peak_indices: np.ndarray,
+    tail_peak_amplitudes: np.ndarray,
+    tail_fit_amplitudes: np.ndarray,
+    tail_damping_ratio: float,
+) -> Path:
+    output_dir = ensure_parent_dir(output_dir)
+    peak_times = time_values[selected_peak_indices]
+    envelope_time, envelope = build_envelope_curve(time_values, peak_times, fit_amplitudes)
+
+    fig, axes = plt.subplots(3, 1, figsize=(12, 10), sharex=False)
+    fig.suptitle(f"Task 3 | Free-Vibration Identification | {file_path.name}")
+
+    axes[0].plot(time_values, raw_signal, color="tab:gray", linewidth=1.0, label="Raw top response")
+    axes[0].set_title(f"Top Response ({TOP_RESPONSE_SENSOR} / {TOP_RESPONSE_AXIS})")
+    axes[0].set_ylabel("Acceleration")
+    axes[0].grid(True, alpha=0.25)
+    axes[0].legend()
+
+    axes[1].plot(time_values, filtered_signal, color="tab:blue", linewidth=1.0, label="Band-pass filtered")
+    axes[1].scatter(time_values[peak_indices], filtered_signal[peak_indices], color="tab:orange", s=14, label="Detected peaks")
+    axes[1].set_title(f"Filtered around f_n = {natural_frequency_hz:.4f} Hz")
+    axes[1].set_ylabel("Acceleration")
+    axes[1].grid(True, alpha=0.25)
+    axes[1].legend()
+
+    axes[2].plot(time_values, filtered_signal, color="tab:blue", linewidth=1.0, alpha=0.65, label="Filtered signal")
+    axes[2].scatter(peak_times, selected_peak_amplitudes, color="tab:red", s=28, label="Selected envelope peaks")
+    axes[2].plot(envelope_time, envelope, color="tab:green", linewidth=1.8, label="Fitted positive envelope")
+    axes[2].plot(envelope_time, -envelope, color="tab:green", linewidth=1.2, linestyle="--", label="Fitted negative envelope")
+    global_peak_times = time_values[global_peak_indices]
+    axes[2].scatter(global_peak_times, global_peak_amplitudes, color="tab:purple", s=14, alpha=0.65, label="Global peaks")
+    axes[2].plot(global_peak_times, global_fit_amplitudes, color="tab:purple", linewidth=1.2, linestyle="-.", label="Global envelope fit")
+    tail_peak_times = time_values[tail_peak_indices]
+    axes[2].scatter(tail_peak_times, tail_peak_amplitudes, color="tab:brown", s=14, alpha=0.65, label="Max-peak tail peaks")
+    axes[2].plot(tail_peak_times, tail_fit_amplitudes, color="tab:brown", linewidth=1.2, linestyle=":", label="Max-peak tail fit")
+    axes[2].set_title(
+        "Damping ratio | "
+        f"local zeta={local_damping_ratio:.5f} | "
+        f"global zeta={global_damping_ratio:.5f} | "
+        f"max-tail zeta={tail_damping_ratio:.5f}"
+    )
+    axes[2].set_xlabel("Time (s)")
+    axes[2].set_ylabel("Acceleration")
+    axes[2].grid(True, alpha=0.25)
+    axes[2].legend()
+
+    plt.tight_layout()
+    figure_path = output_dir / f"{file_path.stem}_task3_identify.png"
+    plt.savefig(figure_path, dpi=180, bbox_inches="tight")
+    return figure_path
+
+
+def analyze_free_vibration(
+    file_path: Path,
+    output_dir: Path,
+    bandwidth_hz: float = 0.3,
+    show: bool = False,
+) -> dict[str, float | int | str | Path]:
+    file_path = Path(file_path).resolve()
+    validate_free_vibration_source(file_path)
+
+    time_values, raw_signal = read_sensor_signal(file_path, sensor_code=TOP_RESPONSE_SENSOR, value_column=TOP_RESPONSE_AXIS)
+    sampling_rate = estimate_sampling_rate(time_values)
+    natural_frequency_hz, freqs, magnitudes, peak_index = dominant_frequency_in_band(
+        raw_signal,
+        sampling_rate=sampling_rate,
+        min_hz=0.1,
+        max_hz=5.0,
+        zero_padding_factor=16,
+    )
+    filtered_signal = bandpass_filter(
+        raw_signal,
+        sampling_rate=sampling_rate,
+        center_hz=natural_frequency_hz,
+        bandwidth_hz=bandwidth_hz,
+        order=4,
+    )
+    peak_indices = detect_decay_peaks(filtered_signal, sampling_rate=sampling_rate, frequency_hz=natural_frequency_hz)
+    peak_window = select_decay_peak_window(filtered_signal, peak_indices, sampling_rate=sampling_rate)
+
+    selected_peak_indices = np.asarray(peak_window["peak_indices"], dtype=int)
+    selected_peak_amplitudes = np.asarray(peak_window["peak_amplitudes"], dtype=np.float64)
+    fit_amplitudes = np.asarray(peak_window["fit_amplitudes"], dtype=np.float64)
+    mean_log_decrement, damping_ratio, damping_ratio_exact = log_decrement_metrics(selected_peak_amplitudes)
+    global_metrics = estimate_global_decay_metrics(
+        filtered_signal=filtered_signal,
+        peak_indices=peak_indices,
+        sampling_rate=sampling_rate,
+    )
+    tail_metrics = estimate_tail_from_max_peak_metrics(
+        filtered_signal=filtered_signal,
+        peak_indices=peak_indices,
+        sampling_rate=sampling_rate,
+    )
+
+    figure_path = save_task3_figure(
+        file_path=file_path,
+        output_dir=output_dir,
+        time_values=time_values,
+        raw_signal=raw_signal,
+        filtered_signal=filtered_signal,
+        peak_indices=peak_indices,
+        selected_peak_indices=selected_peak_indices,
+        selected_peak_amplitudes=selected_peak_amplitudes,
+        fit_amplitudes=fit_amplitudes,
+        natural_frequency_hz=natural_frequency_hz,
+        local_damping_ratio=damping_ratio,
+        global_peak_indices=np.asarray(global_metrics["peak_indices"], dtype=int),
+        global_peak_amplitudes=np.asarray(global_metrics["peak_amplitudes"], dtype=np.float64),
+        global_fit_amplitudes=np.asarray(global_metrics["fit_amplitudes"], dtype=np.float64),
+        global_damping_ratio=float(global_metrics["damping_ratio"]),
+        tail_peak_indices=np.asarray(tail_metrics["peak_indices"], dtype=int),
+        tail_peak_amplitudes=np.asarray(tail_metrics["peak_amplitudes"], dtype=np.float64),
+        tail_fit_amplitudes=np.asarray(tail_metrics["fit_amplitudes"], dtype=np.float64),
+        tail_damping_ratio=float(tail_metrics["damping_ratio"]),
+    )
+    if show:
+        plt.show()
+    plt.close("all")
+
+    return {
+        "file_path": str(file_path),
+        "sensor_code": TOP_RESPONSE_SENSOR,
+        "axis": TOP_RESPONSE_AXIS,
+        "sampling_rate_hz": float(sampling_rate),
+        "natural_frequency_hz": float(natural_frequency_hz),
+        "fft_peak_bin_hz": float(freqs[peak_index]),
+        "fft_peak_magnitude": float(magnitudes[peak_index]),
+        "detected_peak_count": int(len(peak_indices)),
+        "selected_peak_count": int(len(selected_peak_indices)),
+        "mean_log_decrement": mean_log_decrement,
+        "damping_ratio": damping_ratio,
+        "damping_ratio_exact": damping_ratio_exact,
+        "global_selected_peak_count": int(len(np.asarray(global_metrics["peak_indices"], dtype=int))),
+        "global_mean_log_decrement": float(global_metrics["mean_log_decrement"]),
+        "global_damping_ratio": float(global_metrics["damping_ratio"]),
+        "global_damping_ratio_exact": float(global_metrics["damping_ratio_exact"]),
+        "global_envelope_r2": float(global_metrics["r_squared"]),
+        "max_tail_selected_peak_count": int(len(np.asarray(tail_metrics["peak_indices"], dtype=int))),
+        "max_tail_mean_log_decrement": float(tail_metrics["mean_log_decrement"]),
+        "max_tail_damping_ratio": float(tail_metrics["damping_ratio"]),
+        "max_tail_damping_ratio_exact": float(tail_metrics["damping_ratio_exact"]),
+        "max_tail_envelope_r2": float(tail_metrics["r_squared"]),
+        "envelope_r2": float(peak_window["r_squared"]),
+        "figure_path": str(figure_path),
+    }
+
+
+def print_report(result: dict[str, float | int | str | Path]) -> None:
+    print("=== Task 3: Structural Dynamic Identification ===")
+    print(f"Input file: {result['file_path']}")
+    print(f"Top response sensor: {result['sensor_code']} / {result['axis']}")
+    print(f"Sampling rate: {result['sampling_rate_hz']:.4f} Hz")
+    print(f"Natural frequency f_n: {result['natural_frequency_hz']:.6f} Hz")
+    print(f"Local mean log decrement delta: {result['mean_log_decrement']:.6f}")
+    print(f"Local damping ratio zeta (delta / 2pi): {result['damping_ratio']:.6f}")
+    print(f"Local damping ratio exact: {result['damping_ratio_exact']:.6f}")
+    print(f"Global mean log decrement delta: {result['global_mean_log_decrement']:.6f}")
+    print(f"Global damping ratio zeta (delta / 2pi): {result['global_damping_ratio']:.6f}")
+    print(f"Global damping ratio exact: {result['global_damping_ratio_exact']:.6f}")
+    print(f"Max-tail mean log decrement delta: {result['max_tail_mean_log_decrement']:.6f}")
+    print(f"Max-tail damping ratio zeta (delta / 2pi): {result['max_tail_damping_ratio']:.6f}")
+    print(f"Max-tail damping ratio exact: {result['max_tail_damping_ratio_exact']:.6f}")
+    print(f"Detected peaks: {result['detected_peak_count']}")
+    print(f"Selected peaks for local envelope: {result['selected_peak_count']}")
+    print(f"Selected peaks for global envelope: {result['global_selected_peak_count']}")
+    print(f"Selected peaks for max-tail envelope: {result['max_tail_selected_peak_count']}")
+    print(f"Local envelope linearity R^2: {result['envelope_r2']:.6f}")
+    print(f"Global envelope linearity R^2: {result['global_envelope_r2']:.6f}")
+    print(f"Max-tail envelope linearity R^2: {result['max_tail_envelope_r2']:.6f}")
+    print(f"Figure saved to: {result['figure_path']}")
+
+
+def main() -> None:
+    args = parse_args()
+    result = analyze_free_vibration(
+        file_path=Path(args.file),
+        output_dir=Path(args.output_dir),
+        bandwidth_hz=float(args.bandwidth),
+        show=bool(args.show),
+    )
+    print_report(result)
+
+
+if __name__ == "__main__":
+    main()

scripts_3/README.md

@@ -0,0 +1,33 @@
+# 激振频率逆向预测 (Forced-Vibration Frequency Prediction)
+
+## 项目简介
+本目录用于完成强迫振动频率逆向预测任务。脚本基于建筑顶部响应传感器 `WSMS00007` 的 Z 向数据，从稳态振动段中识别主频，反推出地震台输入的简谐波频率，并输出频率误差与可视化结果。
+
+## 技术路线
+本目录的分析流程覆盖稳态段提取、主频精细识别与结果校验三个部分：
+
+1. **稳态中段截取**
+   对强迫振动时程执行滑动窗口搜索，从中间区域选出波动最稳定的响应片段，用于频谱分析。
+2. **顶部响应主频识别**
+   读取 `WSMS00007` 的 `value3` 列，对稳态段进行加窗 FFT 与补零计算，并结合抛物线插值输出高精度激振频率。
+3. **文件名真值校验**
+   从样本文件名中提取频率标签，计算预测值与真值之间的绝对误差。
+
+## 核心模块说明
+
+- **`task4_predict_freq.py`**
+  任务主脚本。负责稳态段截取、频谱分析、抛物线插值、误差计算与频域图像生成。
+
+## 使用指南
+
+**1. 运行任务脚本**
+对单个强迫振动样本预测激振频率：
+```bash
+python scripts_3\task4_predict_freq.py --file downloads\Non_TMD\val\harmonic_5mm_0.75Hz.csv
+```
+
+**2. 指定输出目录**
+将图像保存到自定义目录：
+```bash
+python scripts_3\task4_predict_freq.py --file downloads\Non_TMD\val\harmonic_5mm_0.75Hz.csv --output-dir evaluation_outputs\task4_custom
+```

scripts_3/__init__.py

@@ -0,0 +1 @@
+"""Signal-processing scripts for forced-vibration frequency prediction."""

scripts_3/task4_predict_freq.py

@@ -0,0 +1,221 @@
+from __future__ import annotations
+
+import argparse
+from pathlib import Path
+import sys
+
+import matplotlib
+
+matplotlib.use("Agg")
+
+import matplotlib.pyplot as plt
+import numpy as np
+from scipy.signal import find_peaks
+
+if __package__ is None or __package__ == "":
+    sys.path.append(str(Path(__file__).resolve().parents[1]))
+
+from scripts_2.signal_utils import (
+    TOP_RESPONSE_AXIS,
+    TOP_RESPONSE_SENSOR,
+    dominant_frequency_in_band,
+    ensure_parent_dir,
+    estimate_sampling_rate,
+    get_middle_segment,
+    parse_frequency_from_filename,
+    read_sensor_signal,
+)
+
+
+def parse_args() -> argparse.Namespace:
+    project_root = Path(__file__).resolve().parents[1]
+    parser = argparse.ArgumentParser(description="Predict the harmonic input frequency from top-response steady-state data only.")
+    parser.add_argument(
+        "--file",
+        type=str,
+        default=str(project_root / "downloads" / "Non_TMD" / "val" / "harmonic_5mm_0.75Hz.csv"),
+        help="Path to a forced-vibration CSV such as data from Non_TMD val/test.",
+    )
+    parser.add_argument(
+        "--output-dir",
+        type=str,
+        default=str(project_root / "evaluation_outputs" / "task4_predict_freq"),
+        help="Directory for plots and optional exported diagnostics.",
+    )
+    parser.add_argument(
+        "--show",
+        action="store_true",
+        help="Display the generated figure in addition to saving it.",
+    )
+    return parser.parse_args()
+
+
+def save_task4_figure(
+    file_path: Path,
+    output_dir: Path,
+    time_segment: np.ndarray,
+    signal_segment: np.ndarray,
+    freqs: np.ndarray,
+    magnitudes: np.ndarray,
+    predicted_frequency_hz: float,
+    true_frequency_hz: float | None,
+) -> Path:
+    output_dir = ensure_parent_dir(output_dir)
+    fig, axes = plt.subplots(2, 1, figsize=(12, 8))
+    fig.suptitle(f"Task 4 | Forced-Vibration Frequency Prediction | {file_path.name}")
+
+    axes[0].plot(time_segment, signal_segment, color="tab:blue", linewidth=1.0)
+    axes[0].set_title(f"Steady middle segment from {TOP_RESPONSE_SENSOR} / {TOP_RESPONSE_AXIS}")
+    axes[0].set_xlabel("Time (s)")
+    axes[0].set_ylabel("Acceleration")
+    axes[0].grid(True, alpha=0.25)
+
+    band_mask = (freqs >= 0.1) & (freqs <= 5.0)
+    axes[1].plot(freqs[band_mask], magnitudes[band_mask], color="tab:purple", linewidth=1.1, label="Windowed FFT")
+    axes[1].axvline(
+        predicted_frequency_hz,
+        color="tab:red",
+        linestyle="--",
+        linewidth=1.6,
+        label=f"Predicted = {predicted_frequency_hz:.5f} Hz",
+    )
+    if true_frequency_hz is not None:
+        axes[1].axvline(
+            true_frequency_hz,
+            color="tab:green",
+            linestyle=":",
+            linewidth=1.6,
+            label=f"Filename truth = {true_frequency_hz:.5f} Hz",
+        )
+    axes[1].set_title("High-resolution spectrum from top steady-state response")
+    axes[1].set_xlabel("Frequency (Hz)")
+    axes[1].set_ylabel("Magnitude")
+    axes[1].grid(True, alpha=0.25)
+    axes[1].legend()
+
+    plt.tight_layout()
+    figure_path = output_dir / f"{file_path.stem}_task4_predict_freq.png"
+    plt.savefig(figure_path, dpi=180, bbox_inches="tight")
+    return figure_path
+
+
+def pick_fundamental_peak_index(
+    freqs: np.ndarray,
+    magnitudes: np.ndarray,
+    min_hz: float = 0.1,
+    max_hz: float = 5.0,
+    relative_threshold: float = 0.20,
+    relative_prominence: float = 0.08,
+) -> int:
+    band_mask = (freqs >= min_hz) & (freqs <= max_hz)
+    band_indices = np.where(band_mask)[0]
+    if band_indices.size == 0:
+        raise ValueError("No FFT bins fall inside the requested frequency band.")
+
+    band_magnitudes = magnitudes[band_mask]
+    max_magnitude = float(np.max(band_magnitudes))
+    height_threshold = relative_threshold * max_magnitude
+    prominence_threshold = relative_prominence * max_magnitude
+    local_peaks, _ = find_peaks(
+        band_magnitudes,
+        height=height_threshold,
+        prominence=prominence_threshold,
+    )
+
+    if local_peaks.size == 0:
+        # Fallback for very smooth spectra: choose the strongest bin in band.
+        return int(band_indices[int(np.argmax(band_magnitudes))])
+
+    # Fundamental-first rule: choose the left-most valid peak.
+    return int(band_indices[int(local_peaks[0])])
+
+
+def analyze_forced_vibration(
+    file_path: Path,
+    output_dir: Path,
+    show: bool = False,
+) -> dict[str, float | int | str]:
+    file_path = Path(file_path).resolve()
+    time_values, raw_signal = read_sensor_signal(file_path, sensor_code=TOP_RESPONSE_SENSOR, value_column=TOP_RESPONSE_AXIS)
+    time_segment, signal_segment = get_middle_segment(time_values, raw_signal)
+    sampling_rate = estimate_sampling_rate(time_segment)
+    _, freqs, magnitudes, _ = dominant_frequency_in_band(
+        signal_segment,
+        sampling_rate=sampling_rate,
+        min_hz=0.1,
+        max_hz=5.0,
+        zero_padding_factor=32,
+    )
+    peak_index = pick_fundamental_peak_index(
+        freqs=freqs,
+        magnitudes=magnitudes,
+        min_hz=0.1,
+        max_hz=5.0,
+        relative_threshold=0.20,
+        relative_prominence=0.08,
+    )
+    predicted_frequency_hz = float(freqs[peak_index])
+    true_frequency_hz = parse_frequency_from_filename(file_path.name)
+    absolute_error_hz = None if true_frequency_hz is None else abs(predicted_frequency_hz - true_frequency_hz)
+
+    figure_path = save_task4_figure(
+        file_path=file_path,
+        output_dir=output_dir,
+        time_segment=time_segment,
+        signal_segment=signal_segment,
+        freqs=freqs,
+        magnitudes=magnitudes,
+        predicted_frequency_hz=predicted_frequency_hz,
+        true_frequency_hz=true_frequency_hz,
+    )
+    if show:
+        plt.show()
+    plt.close("all")
+
+    return {
+        "file_path": str(file_path),
+        "sensor_code": TOP_RESPONSE_SENSOR,
+        "axis": TOP_RESPONSE_AXIS,
+        "sampling_rate_hz": float(sampling_rate),
+        "segment_sample_count": int(signal_segment.size),
+        "segment_start_time": float(time_segment[0]),
+        "segment_end_time": float(time_segment[-1]),
+        "predicted_frequency_hz": float(predicted_frequency_hz),
+        "fft_peak_bin_hz": float(freqs[peak_index]),
+        "fft_peak_magnitude": float(magnitudes[peak_index]),
+        "true_frequency_hz": None if true_frequency_hz is None else float(true_frequency_hz),
+        "absolute_error_hz": None if absolute_error_hz is None else float(absolute_error_hz),
+        "figure_path": str(figure_path),
+        "note": "Prediction uses only the top response sensor WSMS00007/value3. Base sensor WSMS00012/value1 is not used to infer the answer.",
+    }
+
+
+def print_report(result: dict[str, float | int | str]) -> None:
+    print("=== Task 4: Excitation Frequency Reverse Prediction ===")
+    print(f"Input file: {result['file_path']}")
+    print(f"Top response sensor: {result['sensor_code']} / {result['axis']}")
+    print(f"Sampling rate: {result['sampling_rate_hz']:.4f} Hz")
+    print(f"Steady segment samples: {result['segment_sample_count']}")
+    print(f"Steady segment time range: {result['segment_start_time']:.4f} s to {result['segment_end_time']:.4f} s")
+    print(f"Predicted excitation frequency: {result['predicted_frequency_hz']:.6f} Hz")
+    if result["true_frequency_hz"] is not None:
+        print(f"Filename frequency truth: {result['true_frequency_hz']:.6f} Hz")
+        print(f"Absolute error: {result['absolute_error_hz']:.6f} Hz")
+    else:
+        print("Filename frequency truth: unavailable")
+    print(f"Figure saved to: {result['figure_path']}")
+    print(result["note"])
+
+
+def main() -> None:
+    args = parse_args()
+    result = analyze_forced_vibration(
+        file_path=Path(args.file),
+        output_dir=Path(args.output_dir),
+        show=bool(args.show),
+    )
+    print_report(result)
+
+
+if __name__ == "__main__":
+    main()

scripts_4/README.md

@@ -0,0 +1,45 @@
+# scripts_4: TMD 小样本适配
+
+本目录实现二阶段方案：
+
+1. 基础模型（`Non_TMD` 训练得到的 `task1_final_model.pkl`）先给出基线预测；
+2. 用极少量 `TMD` 样本拟合频率相关比例系数 `k(f)`；
+3. 输出 `TMD` 预测：`y_tmd = k(f) * y_base`。
+
+## 为什么这样做
+
+- 仅 5 个 TMD 点直接重训主模型容易过拟合；
+- 保留基础模型泛化能力，只学习域偏移；
+- 通过 `shrinkage` 把校正系数向 `1.0` 收缩，降低小样本波动影响；
+- 频率外推时会自动向 `1.0` 回归，避免越界夸张。
+
+## 脚本说明
+
+- `train_tmd_adapter.py`：拟合并保存适配器，同时输出 LOOCV 报告。
+- `predict_tmd_single.py`：加载基础模型 + 适配器，对单个 TMD 文件推理。
+- `adapter.py`：频率校正器定义与序列化工具。
+
+## 使用方法（请使用虚拟环境 Python）
+
+```bash
+.venv/bin/python scripts_4/train_tmd_adapter.py
+```
+
+```bash
+.venv/bin/python scripts_4/predict_tmd_single.py \
+  --file downloads/TMD/val/harmonic_5mm_1.25Hz_TMD.csv
+```
+
+可选参数示例：
+
+```bash
+.venv/bin/python scripts_4/train_tmd_adapter.py \
+  --shrinkage 0.25 \
+  --extrapolation-decay-hz 0.3
+```
+
+## 输出文件
+
+- 适配器：`checkpoints_final/task1_final/task1_tmd_adapter.pkl`
+- 报告：`evaluation_outputs/task1_final/tmd_adapter_report.json`
+

scripts_4/__init__.py

@@ -0,0 +1,2 @@
+"""Task4: Tiny-sample TMD adaptation on top of baseline RMS model."""
+

scripts_4/adapter.py

@@ -0,0 +1,101 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+from pathlib import Path
+import pickle
+
+import numpy as np
+
+
+def _safe_float(value: float) -> float:
+    return float(np.asarray(value, dtype=np.float64).reshape(-1)[0])
+
+
+@dataclass
+class FrequencyScaleAdapter:
+    frequencies_hz: np.ndarray
+    scale_values: np.ndarray
+    shrinkage: float = 0.2
+    extrapolation_decay_hz: float = 0.25
+    min_scale: float = 0.4
+    max_scale: float = 2.5
+
+    def __post_init__(self) -> None:
+        freqs = np.asarray(self.frequencies_hz, dtype=np.float64).reshape(-1)
+        scales = np.asarray(self.scale_values, dtype=np.float64).reshape(-1)
+        if freqs.size == 0:
+            raise ValueError("frequencies_hz cannot be empty.")
+        if freqs.size != scales.size:
+            raise ValueError("frequencies_hz and scale_values must have the same length.")
+        order = np.argsort(freqs)
+        freqs = freqs[order]
+        scales = np.clip(scales[order], self.min_scale, self.max_scale)
+        # Pull scales toward 1.0 to reduce overfitting with tiny TMD samples.
+        scales = 1.0 + (scales - 1.0) * (1.0 - float(np.clip(self.shrinkage, 0.0, 0.95)))
+        self.frequencies_hz = freqs
+        self.scale_values = scales
+
+    def _inside_range_weight(self, frequency_hz: float) -> float:
+        left = self.frequencies_hz[0]
+        right = self.frequencies_hz[-1]
+        f = _safe_float(frequency_hz)
+        if left <= f <= right:
+            return 1.0
+        distance = min(abs(f - left), abs(f - right))
+        decay = max(_safe_float(self.extrapolation_decay_hz), 1e-6)
+        return float(np.exp(-distance / decay))
+
+    def scale_at(self, frequency_hz: float) -> float:
+        f = _safe_float(frequency_hz)
+        boundary_interp = float(np.interp(f, self.frequencies_hz, self.scale_values))
+        in_range_weight = self._inside_range_weight(f)
+        scale = 1.0 + in_range_weight * (boundary_interp - 1.0)
+        return float(np.clip(scale, self.min_scale, self.max_scale))
+
+    def predict(self, base_rms: float, frequency_hz: float) -> float:
+        return max(_safe_float(base_rms), 0.0) * self.scale_at(frequency_hz)
+
+    def to_payload(self) -> dict:
+        return {
+            "frequencies_hz": self.frequencies_hz.tolist(),
+            "scale_values": self.scale_values.tolist(),
+            # scale_values are already post-shrinkage values after __post_init__.
+            "shrinkage": 0.0,
+            "extrapolation_decay_hz": float(self.extrapolation_decay_hz),
+            "min_scale": float(self.min_scale),
+            "max_scale": float(self.max_scale),
+            "already_shrunk": True,
+        }
+
+    @classmethod
+    def from_payload(cls, payload: dict) -> "FrequencyScaleAdapter":
+        shrinkage = float(payload.get("shrinkage", 0.2))
+        if bool(payload.get("already_shrunk", False)):
+            shrinkage = 0.0
+        return cls(
+            frequencies_hz=np.asarray(payload["frequencies_hz"], dtype=np.float64),
+            scale_values=np.asarray(payload["scale_values"], dtype=np.float64),
+            shrinkage=shrinkage,
+            extrapolation_decay_hz=float(payload.get("extrapolation_decay_hz", 0.25)),
+            min_scale=float(payload.get("min_scale", 0.4)),
+            max_scale=float(payload.get("max_scale", 2.5)),
+        )
+
+
+def save_adapter(adapter: FrequencyScaleAdapter, output_path: Path, extra: dict | None = None) -> None:
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+    payload = {
+        "adapter_type": "frequency_scale_interp_v1",
+        "adapter": adapter.to_payload(),
+        "extra": extra or {},
+    }
+    with output_path.open("wb") as handle:
+        pickle.dump(payload, handle)
+
+
+def load_adapter(adapter_path: Path) -> tuple[FrequencyScaleAdapter, dict]:
+    with adapter_path.open("rb") as handle:
+        payload = pickle.load(handle)
+    adapter = FrequencyScaleAdapter.from_payload(payload["adapter"])
+    extra = payload.get("extra", {})
+    return adapter, extra

scripts_4/predict_tmd_single.py

@@ -0,0 +1,78 @@
+from __future__ import annotations
+
+import argparse
+import importlib.util
+import pickle
+from pathlib import Path
+import sys
+
+PROJECT_ROOT = Path(__file__).resolve().parents[1]
+
+try:
+    from .adapter import load_adapter
+except ImportError:
+    from adapter import load_adapter
+
+
+def _load_module(module_name: str, file_path: Path):
+    spec = importlib.util.spec_from_file_location(module_name, file_path)
+    if spec is None or spec.loader is None:
+        raise ImportError(f"Cannot load module from {file_path}")
+    module = importlib.util.module_from_spec(spec)
+    sys.modules[module_name] = module
+    spec.loader.exec_module(module)
+    return module
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="Predict TMD RMS using baseline model + frequency adapter.")
+    parser.add_argument("--file", type=str, required=True)
+    parser.add_argument("--checkpoint", type=str, default="checkpoints_final/task1_final/task1_final_model.pkl")
+    parser.add_argument("--adapter", type=str, default="checkpoints_final/task1_final/task1_tmd_adapter.pkl")
+    parser.add_argument("--data-dir", type=str, default="downloads/TMD")
+    return parser.parse_args()
+
+
+def main() -> None:
+    args = parse_args()
+    config_module = _load_module("config", PROJECT_ROOT / "scripts" / "config.py")
+    dataset_module = _load_module("task1_dataset", PROJECT_ROOT / "scripts" / "dataset.py")
+    make_experiment_config = getattr(config_module, "make_experiment_config")
+    build_record_from_file = getattr(dataset_module, "build_record_from_file")
+
+    config = make_experiment_config()
+    config.data.data_dir = args.data_dir
+    config.data.__post_init__()
+
+    ckpt_path = (PROJECT_ROOT / args.checkpoint).resolve()
+    adapter_path = (PROJECT_ROOT / args.adapter).resolve()
+    with ckpt_path.open("rb") as handle:
+        payload = pickle.load(handle)
+    model = payload["model"]
+    adapter, extra = load_adapter(adapter_path)
+
+    file_path = Path(args.file).resolve()
+    record = build_record_from_file(file_path, config.data)
+
+    pred_tr = max(float(model.predict([record.features.tolist()])[0]), config.data.normalization_eps)
+    pred_base = pred_tr * record.x_rms
+    pred_tmd = adapter.predict(pred_base, record.frequency_hz)
+
+    true_rms = float(record.y_rms)
+    base_err = abs(pred_base - true_rms) / max(abs(true_rms), 1e-12) * 100.0
+    tmd_err = abs(pred_tmd - true_rms) / max(abs(true_rms), 1e-12) * 100.0
+
+    print(f"Checkpoint: {ckpt_path}")
+    print(f"Adapter: {adapter_path}")
+    print(f"Adapter extra: {extra}")
+    print(f"Input file: {file_path}")
+    print(f"Frequency (Hz): {record.frequency_hz:.6f}")
+    print(f"x_rms: {record.x_rms:.6f}")
+    print(f"True y_rms: {true_rms:.6f}")
+    print(f"Base pred y_rms: {pred_base:.6f} | error={base_err:.4f}%")
+    print(f"Adapted pred y_rms: {pred_tmd:.6f} | error={tmd_err:.4f}%")
+    print(f"Applied scale k(f): {adapter.scale_at(record.frequency_hz):.6f}")
+
+
+if __name__ == "__main__":
+    main()

scripts_4/train_tmd_adapter.py

@@ -0,0 +1,164 @@
+from __future__ import annotations
+
+import argparse
+import json
+import pickle
+from pathlib import Path
+import sys
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(PROJECT_ROOT / "scripts"))
+
+from config import make_experiment_config
+from dataset import build_record_from_file
+
+try:
+    from .adapter import FrequencyScaleAdapter, save_adapter
+except ImportError:
+    from adapter import FrequencyScaleAdapter, save_adapter
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="Fit a tiny-sample TMD adapter on top of the Non_TMD baseline .pkl model.")
+    parser.add_argument("--checkpoint", type=str, default=None, help="Path to baseline model .pkl.")
+    parser.add_argument("--tmd-dir", type=str, default="downloads/TMD", help="Directory that contains TMD harmonic csv files.")
+    parser.add_argument("--output", type=str, default="checkpoints_final/task1_final/task1_tmd_adapter.pkl")
+    parser.add_argument("--report-json", type=str, default="evaluation_outputs/task1_final/tmd_adapter_report.json")
+    parser.add_argument("--shrinkage", type=float, default=0.20, help="Scale shrinkage toward 1.0.")
+    parser.add_argument("--extrapolation-decay-hz", type=float, default=0.25)
+    parser.add_argument("--min-scale", type=float, default=0.40)
+    parser.add_argument("--max-scale", type=float, default=2.50)
+    return parser.parse_args()
+
+
+def resolve_checkpoint_path(checkpoint_arg: str | None) -> Path:
+    if checkpoint_arg:
+        return Path(checkpoint_arg).resolve()
+    return PROJECT_ROOT / "checkpoints_final" / "task1_final" / "task1_final_model.pkl"
+
+
+def load_baseline_model(model_path: Path):
+    with model_path.open("rb") as handle:
+        payload = pickle.load(handle)
+    return payload["model"], payload.get("summary", {})
+
+
+def base_predict_rms(model, features: np.ndarray, x_rms: float, eps: float) -> float:
+    pred_tr = max(float(model.predict([features.tolist()])[0]), eps)
+    return pred_tr * float(x_rms)
+
+
+def collect_tmd_rows(model, tmd_dir: Path, eps: float) -> list[dict]:
+    config = make_experiment_config().data
+    config.data_dir = str(Path(tmd_dir).relative_to(PROJECT_ROOT))
+    config.__post_init__()
+    rows: list[dict] = []
+    for file_path in sorted(config.data_root.rglob("harmonic_5mm_*Hz_TMD.csv")):
+        record = build_record_from_file(file_path, config)
+        base_pred = base_predict_rms(model, record.features, record.x_rms, eps)
+        scale = float(record.y_rms / max(base_pred, eps))
+        rows.append(
+            {
+                "file_name": file_path.name,
+                "frequency_hz": float(record.frequency_hz),
+                "true_y_rms": float(record.y_rms),
+                "base_pred_y_rms": float(base_pred),
+                "scale_true_over_base": float(scale),
+            }
+        )
+    if not rows:
+        raise RuntimeError("No TMD harmonic files found.")
+    return rows
+
+
+def loocv_error(rows: list[dict], shrinkage: float, extrapolation_decay_hz: float, min_scale: float, max_scale: float) -> list[float]:
+    errors: list[float] = []
+    eps = 1e-12
+    for holdout in range(len(rows)):
+        train_rows = [row for i, row in enumerate(rows) if i != holdout]
+        adapter = FrequencyScaleAdapter(
+            frequencies_hz=np.asarray([row["frequency_hz"] for row in train_rows], dtype=np.float64),
+            scale_values=np.asarray([row["scale_true_over_base"] for row in train_rows], dtype=np.float64),
+            shrinkage=shrinkage,
+            extrapolation_decay_hz=extrapolation_decay_hz,
+            min_scale=min_scale,
+            max_scale=max_scale,
+        )
+        sample = rows[holdout]
+        pred = adapter.predict(sample["base_pred_y_rms"], sample["frequency_hz"])
+        err = abs(pred - sample["true_y_rms"]) / max(abs(sample["true_y_rms"]), eps) * 100.0
+        errors.append(float(err))
+    return errors
+
+
+def main() -> None:
+    args = parse_args()
+    ckpt_path = resolve_checkpoint_path(args.checkpoint)
+    model, baseline_summary = load_baseline_model(ckpt_path)
+    rows = collect_tmd_rows(model, (PROJECT_ROOT / args.tmd_dir).resolve(), eps=1e-6)
+    rows = sorted(rows, key=lambda x: x["frequency_hz"])
+
+    adapter = FrequencyScaleAdapter(
+        frequencies_hz=np.asarray([row["frequency_hz"] for row in rows], dtype=np.float64),
+        scale_values=np.asarray([row["scale_true_over_base"] for row in rows], dtype=np.float64),
+        shrinkage=args.shrinkage,
+        extrapolation_decay_hz=args.extrapolation_decay_hz,
+        min_scale=args.min_scale,
+        max_scale=args.max_scale,
+    )
+
+    adapted_rows = []
+    for row in rows:
+        adapted_pred = adapter.predict(row["base_pred_y_rms"], row["frequency_hz"])
+        adapted_err = abs(adapted_pred - row["true_y_rms"]) / max(abs(row["true_y_rms"]), 1e-12) * 100.0
+        adapted_rows.append({**row, "adapted_pred_y_rms": adapted_pred, "adapted_err_pct": adapted_err})
+
+    loocv_errors = loocv_error(
+        rows=rows,
+        shrinkage=args.shrinkage,
+        extrapolation_decay_hz=args.extrapolation_decay_hz,
+        min_scale=args.min_scale,
+        max_scale=args.max_scale,
+    )
+    report = {
+        "checkpoint": str(ckpt_path),
+        "baseline_summary": baseline_summary,
+        "samples": adapted_rows,
+        "adapter_scales_after_shrinkage": [
+            {"frequency_hz": float(f), "scale": float(s)}
+            for f, s in zip(adapter.frequencies_hz.tolist(), adapter.scale_values.tolist())
+        ],
+        "metrics": {
+            "base_mean_err_pct": float(
+                np.mean(
+                    [
+                        abs(row["base_pred_y_rms"] - row["true_y_rms"]) / max(abs(row["true_y_rms"]), 1e-12) * 100.0
+                        for row in rows
+                    ]
+                )
+            ),
+            "adapted_mean_err_pct": float(np.mean([row["adapted_err_pct"] for row in adapted_rows])),
+            "loocv_mean_err_pct": float(np.mean(loocv_errors)),
+            "loocv_max_err_pct": float(np.max(loocv_errors)),
+        },
+    }
+
+    output_path = (PROJECT_ROOT / args.output).resolve()
+    save_adapter(adapter, output_path, extra={"report_metrics": report["metrics"], "checkpoint": str(ckpt_path)})
+
+    report_path = (PROJECT_ROOT / args.report_json).resolve()
+    report_path.parent.mkdir(parents=True, exist_ok=True)
+    report_path.write_text(json.dumps(report, ensure_ascii=False, indent=2), encoding="utf-8")
+
+    print(f"Adapter saved to: {output_path}")
+    print(f"Report saved to: {report_path}")
+    print(pd.DataFrame(adapted_rows).to_string(index=False))
+    print(f"LOOCV mean error (%): {report['metrics']['loocv_mean_err_pct']:.4f}")
+    print(f"LOOCV max error (%): {report['metrics']['loocv_max_err_pct']:.4f}")
+
+
+if __name__ == "__main__":
+    main()

src/config.py

@@ -1,30 +0,0 @@
-import os
-
-# Data Configuration
-# 使用基于当前文件的绝对路径拼接，以防止你在不同目录下运行报错
-DATA_DIR = os.path.join(os.path.dirname(os.path.dirname(__file__)), 'downloads')
-SEQ_LEN = 512 # 滑动窗口的长度 (时间序列步数)
-STEP_SIZE = 20       # 滑动窗口的步长
-BATCH_SIZE = 256
-
-# Features Configuration
-INPUT_SENSOR = 'WSMS00012'
-OUTPUT_SENSORS = ['WSMS00007', 'WSMS00008', 'WSMS00009', 'WSMS00010', 'WSMS00011']
-INPUT_AXIS = 'value1'   # 底部传感器输入轴（课程要求：X轴）
-OUTPUT_AXIS = 'value3'  # 目标传感器输出轴
-
-# Model Configuration
-CHANNELS = [64, 64, 128, 128, 256, 256] # TCN 各层通道数
-KERNEL_SIZE = 5
-DROPOUT = 0.2
-
-# Training Configuration
-LEARNING_RATE = 1e-4
-EPOCHS = 50
-WEIGHT_DECAY = 1e-3
-ENABLE_EARLY_STOP = False
-EARLY_STOP_PATIENCE = 15
-
-# Device
-import torch
-DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu'

src/dataset.py

@@ -1,188 +0,0 @@
-import os
-import glob
-import pandas as pd
-import numpy as np
-import torch
-from torch.utils.data import Dataset, DataLoader
-from sklearn.preprocessing import StandardScaler
-from config import *
-
-class MultiOutputStandardizer:
-    """Per-channel standardization that ignores missing labels via masks."""
-    def __init__(self, n_outputs):
-        self.n_outputs = n_outputs
-        self.mean_ = np.zeros(n_outputs, dtype=np.float32)
-        self.scale_ = np.ones(n_outputs, dtype=np.float32)
-        self.fitted = False
-
-    def fit(self, y_sequences, mask_sequences):
-        means = []
-        scales = []
-        for c in range(self.n_outputs):
-            valid_values = []
-            for y_seq, m_seq in zip(y_sequences, mask_sequences):
-                valid = m_seq[:, c] > 0.5
-                if np.any(valid):
-                    valid_values.append(y_seq[valid, c])
-            if len(valid_values) == 0:
-                means.append(0.0)
-                scales.append(1.0)
-                continue
-            vals = np.concatenate(valid_values, axis=0)
-            mean = float(np.mean(vals))
-            std = float(np.std(vals))
-            if std < 1e-6:
-                std = 1.0
-            means.append(mean)
-            scales.append(std)
-
-        self.mean_ = np.asarray(means, dtype=np.float32)
-        self.scale_ = np.asarray(scales, dtype=np.float32)
-        self.fitted = True
-
-    def transform(self, y):
-        if not self.fitted:
-            raise RuntimeError("MultiOutputStandardizer must be fitted before transform.")
-        return (y - self.mean_) / self.scale_
-
-    def inverse_transform(self, y):
-        if not self.fitted:
-            raise RuntimeError("MultiOutputStandardizer must be fitted before inverse_transform.")
-        return y * self.scale_ + self.mean_
-
-
-class BuildingDataset(Dataset):
-    def __init__(self, file_paths, seq_len, step_size, scaler_X=None, scaler_Y=None, fit_scaler=False):
-        self.seq_len = seq_len
-        self.X_data = []
-        self.Y_data = []
-        self.M_data = []
-        self.scaler_X = scaler_X if scaler_X is not None else StandardScaler()
-        self.scaler_Y = scaler_Y if scaler_Y is not None else MultiOutputStandardizer(len(OUTPUT_SENSORS))
-
-        raw_X = []
-        raw_Y = []
-        raw_M = []
-
-        for f in file_paths:
-            # 读取数据
-            df = pd.read_csv(f)
-            # 使用长表格式: code, type, time, value1, value2, value3
-            
-            # 提取 012 的输入轴作为基准
-            df_in = df[df['code'] == INPUT_SENSOR][['time', INPUT_AXIS]].rename(columns={INPUT_AXIS: 'input_signal'})
-            if len(df_in) == 0:
-                # 若无输入传感器(如自由衰减数据)，则补零
-                df_in = pd.DataFrame({'time': df['time'].unique()})
-                df_in['input_signal'] = 0.0
-            
-            # 提取 007~011 的 Z 轴并按时间戳逐步合并 (使用 left join 确保以 df_in 的时间为基准)
-            df_merged = df_in
-            for sens in OUTPUT_SENSORS:
-                df_out_sens = df[df['code'] == sens][['time', OUTPUT_AXIS]].rename(columns={OUTPUT_AXIS: f'out_{sens}'})
-                df_merged = pd.merge(df_merged, df_out_sens, on='time', how='left')
-                
-            df_merged = df_merged.sort_values('time').reset_index(drop=True)
-            
-            if len(df_merged) == 0:
-                print(f"Warning: Skipping file {f} due to no overlapping timestamps across required sensors.")
-                continue
-
-            x_seq = df_merged['input_signal'].values.reshape(-1, 1).astype(np.float32)
-
-            # 提取所有 target 传感器列与可用性掩码
-            out_cols = [f'out_{sens}' for sens in OUTPUT_SENSORS]
-            y_seq = np.zeros((len(df_merged), len(OUTPUT_SENSORS)), dtype=np.float32)
-            m_seq = np.zeros((len(df_merged), len(OUTPUT_SENSORS)), dtype=np.float32)
-            for c, col in enumerate(out_cols):
-                series = df_merged[col]
-                observed = ~series.isna()
-                m_seq[:, c] = observed.astype(np.float32)
-                if observed.any():
-                    filled = series.interpolate(method='linear').bfill().ffill()
-                    y_seq[:, c] = filled.fillna(0.0).values.astype(np.float32)
-                else:
-                    y_seq[:, c] = 0.0
-            
-            raw_X.append(x_seq)
-            raw_Y.append(y_seq)
-            raw_M.append(m_seq)
-
-        if len(raw_X) == 0:
-            raise ValueError("未能从文件中构造出有效序列，请检查数据路径与传感器编码配置。")
-            
-        # 拼接所有文件数据进行 fit
-        X_all = np.vstack(raw_X)
-        
-        if fit_scaler:
-            self.scaler_X.fit(X_all)
-            self.scaler_Y.fit(raw_Y, raw_M)
-            
-        # 切分窗口
-        for x_seq, y_seq, m_seq in zip(raw_X, raw_Y, raw_M):
-            x_seq_scaled = self.scaler_X.transform(x_seq)
-            y_seq_scaled = self.scaler_Y.transform(y_seq)
-            y_seq_scaled = np.where(m_seq > 0.5, y_seq_scaled, 0.0).astype(np.float32)
-            
-            for i in range(0, len(x_seq_scaled) - seq_len + 1, step_size):
-                x_win = x_seq_scaled[i:i+seq_len]
-                y_win = y_seq_scaled[i:i+seq_len]
-                m_win = m_seq[i:i+seq_len]
-                if np.sum(m_win) <= 0:
-                    continue
-                self.X_data.append(x_win)
-                self.Y_data.append(y_win)
-                self.M_data.append(m_win)
-                
-        self.X_data = np.array(self.X_data)
-        self.Y_data = np.array(self.Y_data)
-        self.M_data = np.array(self.M_data)
-
-    def __len__(self):
-        return len(self.X_data)
-
-    def __getitem__(self, idx):
-        return (
-            torch.tensor(self.X_data[idx], dtype=torch.float32),
-            torch.tensor(self.Y_data[idx], dtype=torch.float32),
-            torch.tensor(self.M_data[idx], dtype=torch.float32),
-        )
-
-def get_dataloaders(condition='Non_TMD', include_free_vib=False):
-    """
-    condition: 'Non_TMD' 或者是 'TMD'
-    include_free_vib: 是否在训练集中加入自由振动与自由衰减数据
-    """
-    base_dir = os.path.join(DATA_DIR, condition)
-    
-    # 手动区分的子目录
-    train_files = glob.glob(os.path.join(base_dir, 'train', '*.csv'))
-    if include_free_vib:
-        train_files += glob.glob(os.path.join(base_dir, 'free_vib', '*.csv'))
-        
-    val_files = glob.glob(os.path.join(base_dir, 'val', '*.csv'))
-    test_files = glob.glob(os.path.join(base_dir, 'test', '*.csv'))
-    
-    print(f"[{condition}] Train files: {len(train_files)}, Val files: {len(val_files)}, Test files: {len(test_files)}")
-    
-    if len(train_files) == 0:
-        raise ValueError(f"错误: 在 {base_dir}/train 目录下未找到训练文件！请检查路径是否正确。")
-    if len(val_files) == 0:
-        raise ValueError(f"错误: 在 {base_dir}/val 目录下未找到验证文件！请检查路径是否正确。")
-    
-    train_dataset = BuildingDataset(train_files, SEQ_LEN, STEP_SIZE, fit_scaler=True)
-    val_dataset = BuildingDataset(val_files, SEQ_LEN, STEP_SIZE, 
-                                  scaler_X=train_dataset.scaler_X, 
-                                  scaler_Y=train_dataset.scaler_Y, fit_scaler=False)
-    # 若某条件（如 TMD）下没有 test 数据，可以处理一下防止报错
-    test_loader = None
-    if len(test_files) > 0:
-        test_dataset = BuildingDataset(test_files, SEQ_LEN, STEP_SIZE, 
-                                       scaler_X=train_dataset.scaler_X, 
-                                       scaler_Y=train_dataset.scaler_Y, fit_scaler=False)
-        test_loader = DataLoader(test_dataset, batch_size=BATCH_SIZE, shuffle=False)
-                                   
-    train_loader = DataLoader(train_dataset, batch_size=BATCH_SIZE, shuffle=True)
-    val_loader = DataLoader(val_dataset, batch_size=BATCH_SIZE, shuffle=False)
-    
-    return train_loader, val_loader, test_loader, train_dataset.scaler_X, train_dataset.scaler_Y

src/evaluate.py

@@ -1,50 +0,0 @@
-import torch
-import numpy as np
-import matplotlib.pyplot as plt
-from config import *
-from dataset import get_dataloaders
-from model import BuildingTCN
-
-def evaluate_model():
-    _, _, test_loader, scaler_X, scaler_Y = get_dataloaders()
-    if test_loader is None:
-        raise ValueError("当前数据配置下没有可用的测试集。")
-    
-    model = BuildingTCN(input_size=1, output_size=5, num_channels=CHANNELS, 
-                        kernel_size=KERNEL_SIZE, dropout=DROPOUT).to(DEVICE)
-    model.load_state_dict(torch.load('best_model.pth', map_location=DEVICE))
-    model.eval()
-    
-    all_preds = []
-    all_targets = []
-    
-    with torch.no_grad():
-        for inputs, targets, masks in test_loader:
-            inputs = inputs.to(DEVICE)
-            outputs = model(inputs)
-            
-            all_preds.append(outputs.cpu().numpy())
-            all_targets.append(targets.numpy())
-            
-    # Concatenate results
-    all_preds = np.concatenate(all_preds, axis=0)
-    all_targets = np.concatenate(all_targets, axis=0)
-    
-    # 取一个 batch 的第一条序列进行可视化
-    sample_pred = scaler_Y.inverse_transform(all_preds[0])
-    sample_target = scaler_Y.inverse_transform(all_targets[0])
-    
-    plt.figure(figsize=(15, 10))
-    for i, sens in enumerate(OUTPUT_SENSORS):
-        plt.subplot(5, 1, i+1)
-        plt.plot(sample_target[:, i], label='True', alpha=0.7)
-        plt.plot(sample_pred[:, i], label='Pred', alpha=0.7, linestyle='--')
-        plt.title(f'Sensor {sens} Z-axis Response (Test: Earthquake)')
-        plt.legend()
-        
-    plt.tight_layout()
-    plt.savefig('test_results.png')
-    print("Evaluation done. Result saved to test_results.png")
-
-if __name__ == '__main__':
-    evaluate_model()

src/model.py

@@ -1,76 +0,0 @@
-import torch
-import torch.nn as nn
-from torch.nn.utils import weight_norm
-
-class Chomp1d(nn.Module):
-    def __init__(self, chomp_size):
-        super(Chomp1d, self).__init__()
-        self.chomp_size = chomp_size
-
-    def forward(self, x):
-        return x[:, :, :-self.chomp_size].contiguous()
-
-class TemporalBlock(nn.Module):
-    def __init__(self, n_inputs, n_outputs, kernel_size, stride, dilation, padding, dropout=0.2):
-        super(TemporalBlock, self).__init__()
-        self.conv1 = weight_norm(nn.Conv1d(n_inputs, n_outputs, kernel_size,
-                                           stride=stride, padding=padding, dilation=dilation))
-        self.chomp1 = Chomp1d(padding)
-        self.relu1 = nn.ReLU()
-        self.dropout1 = nn.Dropout(dropout)
-
-        self.conv2 = weight_norm(nn.Conv1d(n_outputs, n_outputs, kernel_size,
-                                           stride=stride, padding=padding, dilation=dilation))
-        self.chomp2 = Chomp1d(padding)
-        self.relu2 = nn.ReLU()
-        self.dropout2 = nn.Dropout(dropout)
-
-        self.net = nn.Sequential(self.conv1, self.chomp1, self.relu1, self.dropout1,
-                                 self.conv2, self.chomp2, self.relu2, self.dropout2)
-        self.downsample = nn.Conv1d(n_inputs, n_outputs, 1) if n_inputs != n_outputs else None
-        self.relu = nn.ReLU()
-        self.init_weights()
-
-    def init_weights(self):
-        self.conv1.weight.data.normal_(0, 0.01)
-        self.conv2.weight.data.normal_(0, 0.01)
-        if self.downsample is not None:
-            self.downsample.weight.data.normal_(0, 0.01)
-
-    def forward(self, x):
-        out = self.net(x)
-        res = x if self.downsample is None else self.downsample(x)
-        return self.relu(out + res)
-
-class TemporalConvNet(nn.Module):
-    def __init__(self, num_inputs, num_channels, kernel_size=2, dropout=0.2):
-        super(TemporalConvNet, self).__init__()
-        layers = []
-        num_levels = len(num_channels)
-        for i in range(num_levels):
-            dilation_size = 2 ** i
-            in_channels = num_inputs if i == 0 else num_channels[i-1]
-            out_channels = num_channels[i]
-            layers += [TemporalBlock(in_channels, out_channels, kernel_size, stride=1, dilation=dilation_size,
-                                     padding=(kernel_size-1) * dilation_size, dropout=dropout)]
-
-        self.network = nn.Sequential(*layers)
-
-    def forward(self, x):
-        return self.network(x)
-
-class BuildingTCN(nn.Module):
-    def __init__(self, input_size, output_size, num_channels, kernel_size=3, dropout=0.2):
-        super(BuildingTCN, self).__init__()
-        self.tcn = TemporalConvNet(input_size, num_channels, kernel_size, dropout=dropout)
-        self.linear = nn.Linear(num_channels[-1], output_size)
-
-    def forward(self, x):
-        # x shape: (batch, seq_len, input_size)
-        # TCN needs shape: (batch, input_size, seq_len)
-        x = x.transpose(1, 2)
-        y = self.tcn(x)
-        # y shape: (batch, num_channels, seq_len)
-        # linear needs shape: (batch, seq_len, num_channels)
-        y = y.transpose(1, 2)
-        return self.linear(y)

src/train.py

@@ -1,77 +0,0 @@
-import torch
-import torch.nn as nn
-import torch.optim as optim
-from config import *
-from dataset import get_dataloaders
-from model import BuildingTCN
-
-def masked_l1_loss(pred, target, mask):
-    diff = torch.abs(pred - target) * mask
-    denom = mask.sum().clamp(min=1.0)
-    return diff.sum() / denom
-
-def train_model():
-    train_loader, val_loader, _, _, _ = get_dataloaders()
-    
-    # 初始化前向模型 (1 -> 5)
-    model = BuildingTCN(input_size=1, output_size=5, num_channels=CHANNELS, 
-                        kernel_size=KERNEL_SIZE, dropout=DROPOUT).to(DEVICE)
-                        
-    optimizer = optim.AdamW(model.parameters(), lr=LEARNING_RATE, weight_decay=WEIGHT_DECAY)
-    scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, 'min', patience=8, factor=0.5)
-    
-    best_val_loss = float('inf')
-    no_improve_epochs = 0
-    early_stop_patience = EARLY_STOP_PATIENCE
-    
-    for epoch in range(EPOCHS):
-        model.train()
-        train_loss = 0.0
-        
-        for inputs, targets, masks in train_loader:
-            inputs = inputs.to(DEVICE)
-            targets = targets.to(DEVICE)
-            masks = masks.to(DEVICE)
-            
-            optimizer.zero_grad()
-            outputs = model(inputs)
-            loss = masked_l1_loss(outputs, targets, masks)
-            
-            loss.backward()
-            torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=1.0)
-            optimizer.step()
-            
-            train_loss += loss.item()
-            
-        # 验证
-        model.eval()
-        val_loss = 0.0
-        with torch.no_grad():
-            for inputs, targets, masks in val_loader:
-                inputs = inputs.to(DEVICE)
-                targets = targets.to(DEVICE)
-                masks = masks.to(DEVICE)
-                outputs = model(inputs)
-                loss = masked_l1_loss(outputs, targets, masks)
-                val_loss += loss.item()
-                
-        train_loss /= len(train_loader)
-        val_loss /= len(val_loader)
-        
-        scheduler.step(val_loss)
-        
-        print(f"Epoch {epoch+1}/{EPOCHS} | Train Loss: {train_loss:.4f} | Val Loss: {val_loss:.4f}")
-        
-        if val_loss < best_val_loss:
-            best_val_loss = val_loss
-            torch.save(model.state_dict(), 'best_model.pth')
-            print("  --> Saved Best Model")
-            no_improve_epochs = 0
-        else:
-            no_improve_epochs += 1
-            if ENABLE_EARLY_STOP and no_improve_epochs >= early_stop_patience:
-                print(f"Early stopping at epoch {epoch+1}")
-                break
-
-if __name__ == '__main__':
-    train_model()