Short-Term Load Forecasting Using STL Decomposition and Stacking Ensemble Learning

This paper proposes a short-term load forecasting model combining Seasonal-Trend decomposition using Loess (STL decomposition) and Stacking ensemble learning. The method decomposes power load data into trend, seasonal, and residual components with STL, predicts each component using deep learning models, and combines the results using Stacking ensemble learning to improve forecasting accuracy. Grey Relational Analysis (GRA) is used to identify key meteorological features correlated with load variations, enhancing the model's performance. Evaluated with real-world data from the non-metallic mineral industry in Shandong, China, the STL-Stacking model outperforms traditional models, offering higher accuracy and better handling of complex load patterns.