A review of deep learning methods for multi-energy load joint forecasting in integrated energy systems

Abstract To accommodate the large-scale integration of renewable energy, and enhance the utilization efficiency of multiple energy types, such as electricity, gas, cooling, and heat, the Integrated Energy System (IES) has emerged in recent years. The forecasting of multiple loads, is a key challenge in guiding the operational strategies of IES, and the development of deep learning (DL) technology, with its advantages in efficiency and accuracy, provides an effective solution. This review first explains the uniqueness and challenges of IES multi-load forecasting, which involves predicting load time series while accounting for the temporal characteristics of each load and their interdependencies. It then summarizes traditional forecasting methods and analyses the advantages of DL-based methods, focusing on key aspects of the capability of dealing with load time series characteristics, load coupling, multi-task learning, and privacy protection. Finally, future challenges and trends in DL for IES multi-load forecasting are discussed.