Predicting Photovoltaic Solar Energy Generation using Capsule Network Architecture

Abstract

The growing reliance on photovoltaic (PV) solar energy as a sustainable source of electricity requires accurate forecasting models to ensure efficient integration into the grid. Traditional methods, including statistical approaches and conventional deep learning models, often struggle with the inherent variability and complex dependencies in solar energy data. This study proposes an approach to predicting photovoltaic solar energy generation using a Capsule Network (CapsNet) architecture. In the study, solar energy generation data from the publicly available UNISOLAR dataset are used. Through extensive experimentation, we demonstrate that CapsNet outperforms traditional machine learning models in prediction of solar energy generation, with R^2=0.94, R M S E=1.95 and MAE=0.97. Among the various techniques, CapsNets shows particular promise for capturing complex relationships in solar data and providing highly accurate predictions. The results underscore the potential of CapsNets in enhancing the reliability and efficiency of solar energy predicting, contributing to more effective energy management and grid stability. © 2024 IEEE.