Adaptive Feature Selection for Probabilistic Multi-Energy Load Forecasting
Yi Ge, Wenjia Zhang, Guojing Liu, Zesen Li, Hu Li
Abstract
Accurate probabilistic forecasting of the multi-energy loads can provide essential uncertainty information about future loads for the management of integrated energy systems. The selection of appropriate features lays a critical foundation to achieve accurate forecasting, but such an issue is not thoroughly studied for probabilistic load forecasting, especially for multi-energy loads. In this paper, we propose an adaptive feature selection framework for probabilistic multi-energy load forecasting by considering different operation patterns to select pattern-specific features. Specifically, we develop a ProbLassoNet model by integrating the multi-quantile regression model with the residual-connecting-Lasso operation to capture both linearity and nonlinearity for effective feature selection. We conduct experiments on an open dataset and validate that the proposed method can significantly improve probabilistic multi-energy load forecasting by distinguishing important features from redundant features. We also provide a comprehensive analysis of important features and multi-energy relationships in different periods, which can serve as a reference for further research on multi-energy load forecasting.