Multi-Timescale Optimization Scheduling of Regional Integrated Energy System Based on Source-Load Joint Forecasting

Abstract

The uncertainty of renewable energy output randomness and multiple load demand uncertainty significantly increases the complexity of optimization and scheduling in regional integrated energy systems (RIES), rendering traditional optimization methods insufficient. This study proposes a multi-scale optimization and scheduling strategy for RIES based on source-load forecasting. Firstly, a forecasting model is constructed using a Temporal Convolutional Network (TCN) and Multi-Head Attention (MHA) mechanism, with the assistance of Multi-Task Learning (MTL) to achieve joint source-load forecasting considering the coupling characteristics of sources and loads. This improves the accuracy of forecasting. Secondly, to reduce the impact of source and load randomness, a multi-time scale optimization model is constructed, which formulates unit output plans in two stages: day-ahead and intra-day. The day-ahead optimization stage obtains the reference values for system optimization and operation strategies, while the rolling optimization strategy is implemented in the intra-day stage to correct the day-ahead optimization and scheduling strategies. Case simulations verify the effectiveness of the proposed method, showing a 4.47% reduction in economic indicators and improved utilization efficiency of RIES energy, guaranteeing dynamic user demand and enhancing energy stability.

Keywords: Regional integrated energy system, Source-load forecasting, Multi-task learning, Multi-timescale, Rolling optimization