Load Prediction of Electric Vehicle Charging Station Based on ASC-GCN-LSTM

In this paper, an attribute-augmented spatio-temporal-graph convolutional network-long short term memory (ASC-GCN-LSTM) is proposed for charging station forecasting model. The proposed model successfully achieves accurate load prediction of charging stations. Firstly, the spatio-temporal information graph is defined to describe the spatiotemporal relationship among the loads of charging stations. a spatio-temporal feature extraction network and utilize graph convolutional neural networks are constructed to extract the spatial and temporal dimension information of infographics. Secondly, by comprehensively considering external environmental factors, such as temperature, precipitation, date and point of interest (POI), node features with enhanced attribute characteristics, are established in combination with the temporal load of the charging station. Finally, the extracted features are fused to conduct load forecasting. The results of the calculation examples show that the proposed ASC-GCN-LSTM model has better prediction performance compared with GCN-LSTM, LSTM and multi-layer perceptron (MLP).