Enhanced Solar Power Prediction using Polynomial and Gradient-based Variational Techniques

Abstract

Accurate prediction of solar photovoltaic (PV) energy output plays a pivotal role in enhancing energy management, planning, and grid reliability. While conventional regression techniques provide interpretability, they often struggle with modelling the nonlinear dependencies present in real-world PV systems. Meanwhile, purely black-box models may yield high accuracy but lack the transparency needed for practical engineering deployment. This study explores the use of polynomial regression for modelling the  AC power output from rooftop PV system, trained on measured environmental and PV system parameters. To further refine the  predictions and enhance generalizability, we integrate a Projected Gradient Descent (PGD) approach for regularization and optimization. The combination of feature expansion and iterative optimization offers a robust modelling framework with strong predictive capability. Experimental results demonstrate that the polynomial regression model achieves excellent agreement with actual measurements, capturing fine-grained variations in power output with high accuracy. This highlights the proposed approach as a computationally efficient and interpretable method suitable for real-time solar forecasting applications