Demand Response Potential Prediction based on GAT Considering Spatial Correlation

To address the challenges of renewable energy's intermittent nature and prediction errors, this study proposes a demand response (DR) potential prediction model based on Graph Attention Networks (GAT). By integrating historical power data and meteorological information into an undirected graph structure, the model transforms demand response prediction into a node regression problem. GAT technology enables dynamic allocation of attention weights, capturing complex interactions between different loads. This approach considers both spatial correlations among variables and leverages the graph attention mechanism to enhance prediction accuracy. Experimental results show that the GAT model outperforms traditional neural networks, supporting effective power system management and optimization. This advancement facilitates better integration of renewable energy sources while improving system flexibility and stability.