Smart grid is being developed to modernize the electricity grid in order to increase power quality. Accurate short-term load forecasting (STLF) is crucial for improving the reliability and energy efficiency of power utility networks. Operational planning decisions depend primarily on load forecasting; overestimation leads to wasted energy and costs, whereas underestimation leads to energy shortages or even blackouts. However, there is no universal model for solving all forecasting problems. This study focuses on the regularized greedy forest (RGF) algorithm, which learns a forest by considering the current tree structure with regularization. In this work, the RGF model is combined with the eXtreme gradient boosting and light gradient boosting machine models, which are gradient-boosting...[ Read more ]

Smart grid is being developed to modernize the electricity grid in order to increase power quality. Accurate short-term load forecasting (STLF) is crucial for improving the reliability and energy efficiency of power utility networks. Operational planning decisions depend primarily on load forecasting; overestimation leads to wasted energy and costs, whereas underestimation leads to energy shortages or even blackouts. However, there is no universal model for solving all forecasting problems. This study focuses on the regularized greedy forest (RGF) algorithm, which learns a forest by considering the current tree structure with regularization. In this work, the RGF model is combined with the eXtreme gradient boosting and light gradient boosting machine models, which are gradient-boosting frameworks, to form a more robust ensemble model using the Bayesian optimization technique. The results show that the proposed ensemble model is suitable for STLF problems. It is practical and reliable and can provide accurate day-ahead short-term hourly load forecasting for Hong Kong. It achieves the best performance among the tree-based models and the deep learning models in different scenarios.