Abstract
Accurate photovoltaic (PV) power forecasting is essential for reliable grid dispatch and renewable energy integration, yet it remains challenging due to weather variability, day-night transitions, regime-dependent dynamics, and strict physical constraints. We propose PARA-PV, a Physics-Aware Retrieval-Augmented framework that embeds physical knowledge throughout the forecasting process.
Framework Design
The framework first encodes multivariate PV observations into patch-level representations. Through a physics-aware retrieval-augmented learner, it retrieves historical patches and analog trajectories consistent with the current window in temporal shape, power level, PV operating state, and intra-day period, yielding a physically grounded base forecast.
To supplement local memory with broader temporal knowledge, the base forecast is then calibrated against a frozen Chronos time-series foundation-model prior through a lightweight residual adapter, adapting general temporal regularities to PV-specific dynamics without overriding the physically grounded prediction.
Distribution Shift Correction
Due to persistent residual conditional distribution shifts when weather and diurnal regimes change, a physics-aware distribution shift correction module subsequently adjusts the preliminary forecast using power, weather, timestamp, and day/night conditions, applying gated mean-shift and scale corrections selectively.
Loss Function Design
Finally, a physics-constrained loss function partitions the samples into peak, ramping, night-time, and regular regimes, adaptively reweighting their error contributions to prevent the dominant regular regime from suppressing the learning of operationally critical states. Our code is available at GitHub.
Blogger's Review: The PARA-PV framework presents an innovative approach to tackle the complexities of PV power forecasting by integrating physical knowledge with historical data, showcasing the potential of applying deep learning and physical models in renewable energy fields. Its design offers new directions for future energy forecasting research.