Short-Term Multi-Energy Load Forecasting Based on Feature Reconstruction and TCN-Informer

With the diversification of user energy and coordinated operation development of the integrated energy system, the coupling relationships between different energies becomes increasingly complex, which brings great challenges to short-term multi-energy load forecasting. Considering inadequate correlation analysis of influence factors and insufficient model learning ability to implicit coupling relationships, a short-term multi-energy load forecasting method is proposed based on feature reconstruction and TCN-informer. Firstly, the day-scale feature is screened and optimal feature sets are reconstructed via correlation analysis between load and influence factor time series in specific date. Secondly, combining TCN and informer network, the multi-energy load forecasting model is built via multi-task learning architecture. The input layer is composed of TCN, which can extract key information and temporal correlation of the input data. The shared layer is composed of Informer network, which can excavate inner relation between different loads effectively. Finally, simulation data sets generate from integrated energy system of the Tempe campus at Arizona State University. Prediction indicator comparison confirms effectiveness and superiority of feature reconstruction and TCN-informer multi-energy load forecasting model.