Optimizing Solar Irradiance Prediction: Feature Selection for All-Sky Image Processing Using a Hybrid Prediction Method

Abstract

The forecasting of solar irradiance is crucial for photovoltaic solar energy generation, as production is subject to intermittency due to climatic conditions, such as cloud cover, wind and, temperature. Based on the Hybrid Prediction Method (HPM), this study investigated the influence of a set of all-sky image processing features on the HPMs Artificial Neural Network prediction accuracy. Using correlation-based attribute selection, three predictive models with different input feature sets were evaluated. The results show that, when considering all horizons together and paired, the Medium set of 6 features achieves prediction accuracy statistically similar to the Complete set with 9 features, reducing the computational time (14.4%) and model input dimensionality (33.3%). However, when comparing individual horizons, the Complete set outperforms the Medium set at 5- and 15-minute horizon, while maintain similar accuracy at the 1-minute horizon. The Reduced set, with three features, consistently underperformed. This study provides news insights into the optimization of solar irradiance forecasting using HPM, contributing to advances in photovoltaic energy forecasting.