Forecasting Solar Energy Generation Using Machine Learning Techniques and Hybrid Models Optimized by War SO

Abstract

Due to threats caused by climate change and energy security, the attainment of adequate and sustainable energy resources is becoming of great importance. There exist promising alternatives to the traditional source, such as solar and wind. However, there are high obstacles to their penetration into a power grid because of the variability and uncertainty in renewable sources. In this regard, it becomes quite necessary to accurately forecast the models so that one can optimize energy generation and guarantee grid stability. This work studies the application of several machine learning algorithms, including Cat Boost, AdaBoost, and Light GBM, to solar energy generation forecasting. The approach has been applied based on data from two solar stations over a period of two years, where the performance of each stand-alone algorithm and a hybrid model that will be optimized with War SO optimizer is analyzed and presented. The standalone CatBoost model demonstrated superior performance, achieving an R² of 0.9106 and RMSE of 4.06 MW in the 30 MW farm. Hybrid models further improved accuracy, with the AdaBoost-War SO model reaching an R² of 0.9836 and RMSE of 1.75 MW. These results confirm the efficiency of utilizing machine learning approaches toward enhancing accuracy in renewable energy forecasting, and therefore hybrid models play an important role in energy prediction with higher accuracy