Predictive Modeling of Solar Photovoltaic Power Generation: A Comparative Evaluation of Machine Learning Algorithms Under Volatile Micro-Climatic Conditions

Abstract

The accelerating integration of solar photovoltaic (PV) systems into modern power grids has introduced unprecedented challenges in grid stability due to the stochastic nature of solar irradiance. Accurate short-term power forecasting is a critical operational requirement for energy dispatching, yet traditional physical and statistical models frequently fail to capture rapid generation transients caused by local micro-climatic shifts. This paper presents a comprehensive comparative evaluation of traditional machine learning (ML) algorithms—including Support Vector Regression (SVR), Random Forest (RF), and Artificial Neural Networks (ANN) - against advanced sequential architectures like Long Short-Term Memory (LSTM) networks. Furthermore, this study introduces a novel theoretical methodology: an Adaptive Hybrid Ensemble Framework that integrates localized micro-climatic variables such as real-time Cloud Cover Indices (CCI) and Aerosol Optical Depth (AOD). Through rigorous analytical projections, we demonstrate how this hybrid approach theoretically minimizes error metrics during volatile weather conditions, significantly outperforming standalone models and providing a robust pathway for real-time smart grid management.