Physics informed models show limited gains in solar power forecasting
The Utility of Physics-Informed Machine Learning in Solar Power Production Forecasting
Machine Learning
Summary
Predicting how much solar power will be produced is challenging, and some people think combining physics formulas with machine learning might help. The authors tested this idea with big data from thousands of solar installations across Europe. They found that adding physics knowledge to modern deep learning models did not significantly improve predictions compared to just using data alone. This means simpler data-driven methods may be enough for forecasting solar power one day ahead.
What this means in practice
- •For solar energy operators: Use efficient data-driven models to forecast photovoltaic power without adding complex physical models, saving development time and resources.
- •For energy grid managers: Rely on scalable machine learning forecasts trained on large datasets instead of hybrid physics-informed models to predict solar power fluctuations across regions.
Authors
Peter Pavlík, Franz M. Rohrhofer, Bernhard C. Geiger
Abstract
Physics-informed machine learning (PIML) and hybrid modeling have attracted significant attention as approaches for combining physical knowledge with the flexibility of deep learning. In this work, we investigate whether these approaches deliver their commonly claimed benefits for large-scale photovoltaic (PV) power forecasting. We develop two hybrid forecasting architectures that integrate the ADR physical PV efficiency model with the TSMixer neural forecasting architecture through feature augmentation and residual learning. The proposed models are evaluated on a dataset comprising over 3 million samples from 3,713 PV installations across 33 European countries. We assess convergence speed, forecasting accuracy, transferability across geographical regions, and data efficiency. Although the physics-informed models seem to exhibit slightly faster convergence during training, they do not achieve consistent improvements in test-set accuracy, generalization, or performance under limited training data compared to the purely data-driven baseline. These findings suggest that, for modern deep forecasting architectures trained on large datasets, explicitly incorporating the ADR physical model provides limited practical benefit for day-ahead PV forecasting.