Solar Power Prediction using Meta-Residual Booster with Explainable AI

The key innovations of this research are:

1. Two-stage residual learning: Explicit error correction where the second model learns to predict residuals of the first, rather than modeling the target directly.
2. Meta-feature augmentation: Self-referential features (rolling statistics, prediction lags, magnitude) derived from base model outputs that capture temporal error patterns.
3. Self-aware prediction architecture: The model leverages its own intermediate predictions as additional input signals, creating a feedback loop for error refinement.
4. Interpretable ensemble: Built on SHAP explainability, enabling identification of both original feature contributions and meta-feature importance.