Machine learning Based Short-Term Solar Generation Forecasting using CPOA-SVM

Abstract

Accurate solar generation forecasting is crucial for optimizing the operation of Renewable Energy Source (RES)-integrated power grids. This study presents a novel hybrid Chaotic Pelican Optimization Algorithm (CPOA)-Support Vector Machine (SVM) model for accurate hourly day-ahead solar power forecasting, which optimizes the SVM's hyperparameters to improve prediction accuracy and reduce forecasting errors. The model uses time-related, historical, and meteorological data, with performance evaluated using metrics like RMSE, MAE, and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>. Three experimental cases were examined: Case 1 (using all features), Case 2 (only meteorological variables), and Case 3 (using a subset of top-ranked features via MRMR and RReliefF). The CPOA-SVM model outperformed other SVM-based algorithms in all cases. CPOA-SVM in Case 3 outperformed the other models, achieving a testing RMSE of 65.14, MAE of 40.21, R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.9937, and sMAPE of 7.61%. It showed significant improvement over Case 1, with a 9.22% reduction in RMSE and a 12.76% reduction in MAE. Case 2 shows a completely poor performance compared to the other two cases. The study highlights the importance of intelligent feature selection and metaheuristic optimization in enhancing forecasting accuracy, demonstrating CPOA-SVM's potential for real-time solar generation forecasting in smart grids.