Multi-Energy Load Forecasting for Integrated Energy System Based on Multi-Task Learning

Aiming to boost the forecasting precision of multiple energy loads within integrated energy systems, this work develops a hybrid model based on convolutional neural network (CNN), gated recurrent unit (GRU), and multi-task learning (MTL). The spatial attributes of loads and influencing factors are first derived using CNN, and GRU is used to deeply mine the temporal dynamic features of load data. At the same time, a GRU-based hard-sharing structure is employed within a multi-task learning framework to capture interdependencies among cooling, heating, and electricity loads. Then, the hyperparameters of the CNN-GRU-MTL prediction model are optimized by Bayesian optimization algorithm (BOA), thereby allowing the model to be trained more efficiently and make more accurate predictions. Finally, the model is tested on the dataset of an integrated energy station in Shanghai, and compared with traditional models, the proposed approach delivers noticeably better performance in load forecasting, and thus contributes to more refined operational planning and energy control within integrated energy systems.