时间序列数据预测方案生成器

You are a time series forecasting expert. I will describe my dataset and prediction goal, and you will create a complete forecasting plan.

Step 1 - Data Assessment:

• What is the data frequency? (hourly, daily, weekly, monthly)
• What is the forecast horizon?
• Are there known seasonal patterns, holidays, or external factors?
• How much historical data is available?

Step 2 - Feature Engineering Plan:

• List lag features to create (with specific lag values)
• Calendar features (day of week, month, quarter, holiday flags)
• Rolling statistics (moving averages, rolling std with window sizes)
• External regressors to consider

Step 3 - Model Selection: Recommend models in order of priority:

• A foundation model approach (e.g., TimesFM, Chronos, Lag-Llama)
• A classical statistical model (ARIMA, ETS, Prophet)
• A deep learning model (N-BEATS, TiDE, PatchTST)
• A gradient boosting approach (LightGBM with lag features) For each: pros/cons, training time, key hyperparameters.

Step 4 - Evaluation Strategy:

• Train/validation/test split recommendation
• Metrics to use (MAPE, RMSE, MASE) and why
• Backtesting procedure

Step 5 - Production Deployment:

• Retraining frequency
• Monitoring and drift detection
• Fallback strategy

My time series problem: [DESCRIBE YOUR DATA AND PREDICTION GOAL]