基于时空图卷积网络的电动汽车充电需求预测

Electric Vehicle Charging Demand Forecasting Based on Spatiotemporal Graph Convolutional Networks

为提高电动汽车充电需求预测的准确性,减少热点区域交通压力,提出一种融合图卷积网络(GCN)与长短期记忆网络(LSTM)的时空图卷积网络模型(GCN+LSTM)。该模型将充电站作为图的节点,并通过地理位置的接近程度定义节点间的连接。通过GCN迭代聚合相邻节点信息,模型能捕捉充电站之间的空间关联。同时,LSTM对充电需求的时间序列特征进行分析,利用历史数据预测未来的充电趋势。通过构建充电站间的栅格地图,模型实现了高效的数据处理和特征提取。实验结果表明,与其他传统网络模型相比,GCN+LSTM模型在7 d、30 d预测任务中,整体上均展现出更低的平均绝对误差(MAE)、均方根误差(RMSE)以及平均绝对百分比误差(MAPE),显示出更优的预测性能。

In order to enhance the accuracy of electric vehicle charging demand forecasting and alleviate traffic pressure in hotspot areas,this article proposes a spatiotemporal graph convolutional network model(GCN+LSTM)that integrates Graph Convolutional Network(GCN)and Long Short-Term Memory(LSTM)networks.The model treats charging stations as nodes in a graph,with connections between nodes defined by their geographical proximity.By iteratively aggregating information from adjacent nodes through GCN,the model captures spatial correlations between charging stations.Meanwhile,LSTM analyzes the time series characteristics of charging demand,using historical data to predict future charging trends.By constructing a raster map between charging stations,the model achieves efficient data processing and feature extraction.The experimental results indicate that compared to other traditional network models,the GCN+LSTM model demonstrates overall lower Mean Absolute Error(MAE),Root Mean Square Error(RMSE),and Mean Absolute Percentage Error(MAPE)in both 7-day and 30-day forecasting tasks,showing superior predictive performance.