Multivariate Load Forecasting of Integrated Energy Systems Based on Convolutional Neural Network and Soft Sharing Mechanism

Integrated energy systems can improve energy utilization efficiency and reduce carbon emissions; accurate load forecasting is an important prerequisite for its operation scheduling and energy control. In this study, a multivariate load-coupled forecasting model (CNN-MMoE-LSTM) is proposed to achieve higher accuracy in load forecasting. The model is based on convolutional neural networks (CNN), integrating multi-gate mixed expert models and long short-term memory neural network models. The model achieves deep coupling of cold, thermal, and power load relationships and significantly improves the accuracy of multivariate load forecasting through the organic combination of CNNs and soft sharing mechanism for multi-task learning. The results indicate that the proposed model can further improve the joint prediction accuracy of multiple loads. The root mean square error of electricity, heat, and cooling loads is reduced by 5.86%，5.98%, and 2.67%, respectively, compared to the long short-term neural memory network model based on multi-gate hybrid expert models (MMoE-LSTM).