Integrated Energy System Load Forecasting with Spatially Transferable Loads

In the era of dual carbon, the rapid development of various types of microgrid parks featuring multi-heterogeneous energy coupling presents new challenges in accurately modeling spatial and temporal load characteristics due to increasingly complex source-load characteristics and diversified interaction patterns. This study proposes a short-term load forecasting method for an interconnected park-level integrated energy system using a data center as the case study. By leveraging spatially transferable load characteristics and the heterogeneous energy correlation among electricity-cooling-heat loads, an optimal feature set is selected to effectively characterize the spatial and temporal coupling of multi-heterogeneous loads using Spearman correlation analysis. This optimal feature set is fed into the multi-task learning (MTL) combined with the convolutional neural network (CNN) and long- and short-term memory (LSTM) network model to generate prediction results. The simulation results demonstrate the efficacy of our proposed approach in characterizing the spatial and temporal energy coupling across different parks, enhancing track load "spikes" and achieving superior prediction accuracy.