python_algorithm/train_corrector.py

import pandas as pd
import joblib
from apscheduler.schedulers.blocking import BlockingScheduler
from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split

# === 配置路径 ===
csv_path = 'C:\\Users\\Administrator\\Desktop\\defrost\\feedback_data.csv'  # 你的csv
model_save_path = "defrost_time_corrector.pkl"  # 模型保存路径

# === 特征列定义 ===
feature_columns = [
    "w", "rho_coal", "rho_ice", "C_coal", "C_ice", "L", "k_coal", "k_ice", "h",
    "T_air", "T_initial", "T_m", "a", "b", "c"
]

# 定义定时任务的训练函数
def train_and_save_model():
    print("🔄 定时任务开始：重新训练模型...")

    # === 1. 读取CSV并预处理 ===
    try:
        df = pd.read_csv(csv_path, parse_dates=["t_formula", "t_real"], encoding='utf-8')
        print(f"✅ 成功读取CSV文件，共{len(df)}条数据")
    except Exception as e:
        print(f"❌ 读取CSV失败: {e}")
        return

    # 强制转换为 datetime 类型
    df["t_real"] = pd.to_datetime(df.get("t_real"), errors="coerce")
    df["t_formula"] = pd.to_datetime(df.get("t_formula"), errors="coerce")

    # 删除无法转换时间的行
    before_drop = len(df)
    df = df.dropna(subset=["t_real", "t_formula"])
    after_drop = len(df)
    if before_drop != after_drop:
        print(f"⚠️ 删除了 {before_drop - after_drop} 行非法时间数据")

    # 计算真实解冻时长（小时）
    df["t_real_hours"] = (df["t_real"] - df["t_formula"]).dt.total_seconds() / 3600

    # 再次检查并去除无效目标值
    if df["t_real_hours"].isna().any():
        invalid_count = df["t_real_hours"].isna().sum()
        df = df.dropna(subset=["t_real_hours"])
        print(f"⚠️ 删除了 {invalid_count} 行无效的解冻时长")

    # 确保字段类型正确（如果这两列存在）
    for col in ["material_name", "manufactured_goods"]:
        if col in df.columns:
            df[col] = df[col].astype(str)

    # 检查有没有缺失特征
    missing_features = [col for col in feature_columns if col not in df.columns]
    if missing_features:
        print(f"❌ 缺少必要特征列: {missing_features}")
        return

    # === 2. 智能训练模型 ===
    X = df[feature_columns]
    y = df["t_real_hours"]

    if len(X) >= 10:
        # 数据够多，做train_test_split
        X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
        print(f"📚 数据量 {len(X)}，已划分训练集和测试集")
    else:
        # 数据少，直接全量训练
        X_train, y_train = X, y
        X_test, y_test = None, None
        print(f"⚠️ 数据量太少（{len(X)}条），直接全量训练")

    # 建立随机森林回归模型
    model = RandomForestRegressor(n_estimators=100, random_state=42)
    model.fit(X_train, y_train)

    # 保存模型
    joblib.dump(model, model_save_path)
    print(f"✅ 模型训练完成，已保存为 {model_save_path}")

# === 启动定时任务调度器 ===
if __name__ == '__main__':
    # 先执行一次
    train_and_save_model()

    # 然后设置每小时定时执行
    scheduler = BlockingScheduler()
    scheduler.add_job(train_and_save_model, 'interval', hours=1)
    print("⏰ 启动定时任务调度器：每小时自动训练模型...")
    scheduler.start()