Correlated load forecasting in active distribution networks using Spatial-Temporal Synchronous Graph Convolutional Networks

Load forecasting becomes increasingly challenging as power distribution networks evolve towards active distribution networks with high-penetration renewables. In the context of active distribution networks, the load can be principally referred to as a mixture of power consumption devices as well as renewables-based distributed energy resources behind the meters. Accordingly, more hidden information (e.g., correlations) should be mined from historical load observations to relieve the significant challenges resulting from behind-the-meter renewables. Here, a novel spatial-temporal graph representation method is proposed to characterise and present spatial and temporal correlations of historical load observations. The graph-structured data is then fed into a model denoted as Spatial-Temporal Synchronous Graph Convolutional Network (STSGCN) for performing load forecasting by extracting the inherent spatial-temporal features of historical load observations. Finally, numerical experiments are performed on a real-world load dataset. The results show that the proposed method manages to capture spatial-temporal correlations of load observations in the forecasting process while outperforming the state of the art in terms of overall forecasting accuracy.