An ensemble-based approach for short-term load forecasting for buildings with high proportion of renewable energy sources

Abstract

The increasing integration of renewable energy sources into large buildings and structures has emphasized the importance of effective control systems to optimize resource use, grid stability, and reliability. In this context, load forecasting is critical for predicting future energy demand and managing the intermittency and variability of renewable energy resources to ensure a stable performance. This study proposes an ensemble-based short-term load forecasting (STLF) method tailored for buildings. It combines parallel and series approaches for a 24-hour forecasting horizon. The parallel approach uses similar historical days to capture the general load trend, holidays, and user behavior, while the series component employs a deep learning-based neural network to address short-term changes. The final ensemble-based forecast is the culmination of these two methods. Evaluated using real load profile data sets with 15-minute resolution, the proposed method demonstrates more than 23% and 24% improvements in MAE and RMSE, respectively, compared to other methods. These results demonstrated the superior performance of the proposed method in terms of both accuracy and robustness, making it an effective solution for short-term load forecasting in buildings heavily reliant on renewable energy.