Mid-long Term Urban Power Load Forecasting Based on Data-Driven Spatio-temporal Networks

Abstract

In order to ensure the quality of urban power grid planning and balance the power and electricity, accurate medium and long-term load forecasting becomes particularly. In view of the shortcomings of existing methods in utilizing the spatial correlation between urban areas, a prediction method based on dynamic time warping (DTW) and sp-temporal attention graph convolution (ASTGCN) is proposed. Firstly, the correlation between different regions in the target city is deeply analyzed to establish a coupling relationship., the DTW algorithm is used to construct an adjacency matrix to capture the spatiotemporal correlation between different regions in the city. Then, the ASTGC model is applied to predict the load of each region to capture the spatiotemporal characteristics of the load. Finally, the overall urban prediction load is obtained by the prediction results of each region. The experimental results show that the proposed method can capture the spatiotemporal relationship in the city more comprehensively and significantly improve accuracy of medium and long-term load forecasting.

Keywords

• mid-long term load forecasting
• correlation analysis
• spatio-temporal graphical convolutional networks