Ultra-short-term Multi-region Power Load Forecasting Based on Spearman-GCN-GRU Model

To improve the prediction accuracy of multi-region power load, an ultra-short-term multi-region power load forecasting model based on Spearman-GCN-GRU is proposed with focus on the spatial-temporal correlation analysis of multi-region power data. Firstly, the Spearman correlation coefficient is used to analyze the spatial-temporal correlation of power load in different regions and construct the Spearman adjacency matrix. And then, the graph convolutional network (GCN) and gated recurrent unit (GRU) are used to respectively extract the spatial and temporal features from the data. Finally, the multilayer perceptron (MLP) is used to decode and output the prediction results. Through comparison with the distance adjacency matrix-based models, the Spearman-GCN-GRU model is proved to be feasible. In terms of prediction accuracy, the Spearman-GCN-GRU model are optimal in common evaluation indexes compared with traditional statistical models and neural network models. Specifically, in terms of the root mean square error (RMSE), the Spearman-GCN-GRU model exhibits a respective decrease of 13.90%, 11.66%, and 8.36% compared to the GRU, GCN and deep neural network (DNN) models, demonstrating its superior predictive performance.