Day-Ahead Demand Response Potential Forecasting Model Considering Dynamic Spatial-Temporal Correlation Based on Directed Graph Structure

The day-ahead demand response (DR) potential forecasting can provide reference information for load aggregators (LAs) to participate in bidding offers in the electricity market and reduce decision-making risks. However, most of the current DR potential forecasting methods have the following two problems: 1) lack of consideration of spatial correlation between different customers; 2) lack of consideration of causal relationships between input feature nodes. When there are drastic changes in DR potential, critical information may not be utilized effectively, resulting in poor forecasting accuracy. Based on this, this article proposes a directed graph structure based day-ahead demand response potential forecasting model considering the dynamic spatial-temporal correlation. Firstly, the residential customers are clustered according to the DR potential pattern. Secondly, the important feature nodes affecting the DR potential of LAs are extracted, and a directed edge structure is established by analyzing the causal relationships between different clusters of nodes, and the Pearson correlation coefficient (PCC) is used to characterize the dynamic spatio-temporal correlations between similar feature nodes to establish a directed graph structure. Finally, the directed graph structure is used to train an online forecasting model. Case study shows that the proposed model can tap into the dynamic spatio-temporal correlation properties of residential customers and improve the forecasting accuracy of DR potential for LAs.