Short-Term Multivariate Load Forecasting for Integrated Energy Systems Based on BIGRU-AM and Multi-Task Learning

Abstract

Complex strong coupling relationships may exist among multiple loads in an integrated energy system, and accurate multivariate load forecasting plays an important role in energy scheduling and operation planning of integrated energy systems. Therefore, this paper proposes a short-term multivariate load forecasting model for integrated energy systems based on bidirectional gated cycle unit (BIGRU), multi-task learning (MTL) and attention mechanism (AM). Firstly, considering the coupling characteristics between the electric, heating and cooling loads of the integrated energy system, the PCC method is used to screen the input variables of the prediction model and reduce the dimensionality of the input features of the prediction model; then the attention mechanism is introduced into the BIGRU model based on multi-task learning, which further mines the coupling features between loads through multi-task learning, compensates for the insufficient mining and utilization of coupling features between loads, and achieves the differentiated extraction of important features in the shared layer by each sub-task through the attention mechanism, which improves the prediction accuracy of the model; finally, using the historical data of electric, cooling and heating loads of Arizona State University Tempe campus as the actual arithmetic example, the reasonableness and validity of the model were verified by simulation comparison.