Comparative Analysis of Machine Learning Algorithms Using Weather Station Grid Shor-Term Load Forecasting

The paper presents a comparative analysis of machine learning algorithms for short-term electricity consumption forecasting of a large power system. The study is based on archived data of total electricity consumption and temperature in key locations within the interregional dispatching office of Siberia. Methods such as linear regression, decision tree, k-nearest neighbors, and ensemble methods were employed for electricity consumption forecasting. The impact of using meteorological factors from different weather stations was investigated. It was found out that a large number of weather stations may not improve accuracy if the historical data depth is short. The results of each method were analyzed, and conclusions were drawn based on their effectiveness in electricity consumption prediction. Performance metrics such as RMSE, MAPE, MAE, and R2-score were used as evaluation criteria. The research findings can be valuable for identifying optimal classes of methods for developing more accurate and efficient electricity consumption forecasting models, which are crucial for energy supply optimization and resource conservation.