Multiload forecasting method for microgrid based on STSGCN

Amidst the growing diversification of user-side energy demands, accurate multi-load forecasting has emerged as a critical factor for effective planning and optimization of microgrids. As versatile distributed energy systems, microgrids necessitate adept management and dispatch of various loads - spanning electricity, heat, cooling, and energy storage. Nevertheless, these loads exhibit intricate spatiotemporal correlations and heterogeneities, posing formidable challenges for load forecasting. Current methods predominantly concentrate on time series correlations, neglecting the spatiotemporal data's heterogeneity, thereby inadequately serving practical applications. To tackle these challenges, this paper introduces a novel multi-load prediction model for microgrids, rooted in the Spatial-Temporal Synchronous Graph Convolutional Network (STSGCN). This model integrates Graph Convolutional Network (GCN) and Long Short-Term Memory (LSTM), synergistically capturing both the correlations and heterogeneities within spatiotemporal data. Simulation outcomes underscore the model's effectiveness, with the STSGCN model demonstrating a significant reduction in average absolute error - approximately 13% compared to traditional LSTM, 7% versus GCN, and 5% GCN-LSTM prediction models. This underscores the model's potential to revolutionize multi-load forecasting in microgrids.