A Comparative Analysis of Short-Term Solar Power Forecasting Using Machine Learning Methods

Integrated power systems encounter a multitude of challenges due to the variability observed in solar energy. As a result, it is requisite to make precise predictions regarding solar power. This study conducted a comparative analysis of various models including categorical boosting (CatBoost), random forest (RF), adaptive boosting (AdaBoost), gradient boosting (GB), extreme gradient boosting (XGBoost), and gated recurrent network (GRU). The objective of this paper is to determine the most accurate model for predicting global horizontal irradiance (GHI) in Canberra, Australia. This research used data collected at 60-minute, 30-minute, 15-minute, 10-minute, and 5-minute intervals throughout the year to ensure each model’s consistency in performance. As the number of samples increased in the dataset, each model’s performance improved significantly. This research made it quite evident that GRU outperformed all other models in producing the best results for this dataset.