Abstract: Aiming at the problems of privacy protection, computational efficiency, and predictive accuracy in existing forecasting methods for electric vehicle charging stations, a cloud-edge collaborative and graph neural network based load forecasting approach was proposed. A privacy preserving module based on embedding is developed in the cloud, which reconstructs the input data through embedding transformation to prevent potential privacy leakage risks. A method for generating representation with graph structure based on clustering is proposed to provide additional spatiotemporal information and achieve more accurate forecasting. Personalized graph neural network forecasting models are designed for clients based on cloud's graph structure representation, enabling collaborative training of electric vehicle charging stations in different regions while protecting privacy. Experimental results on the Perth dataset demonstrate that the model outperforms benchmark methods in predictive accuracy and that the cloud-edge collaborative framework proposed in this study significantly enhances the performance of graph neural network algorithms in the task of load forecasting for electric vehicle charging stations.

Key words: cloud-edge collaborative, graph neural network, electric vehicle charging stations, load forecasting