Research On Demand Prediction For Electric Vehicle Charging Stations Based On Transfer Learning

With the rapid development of the economy and society,the number of cars in the country continues to increase,which brings great challenges to ecological environment protection.For the consideration of environment and energy security,more and more cities begin to build electric vehicle charging station networks.Electric vehicles rely on public charging stations to provide electricity,so it is very important to accurately predict the charging demand of charging stations in new cities,which is helpful to the designation of operation strategies and the deployment of new stations.Due to the cold start problem for the demand prediction for charging stations in the new city,this paper uses transfer learning to transfer the knowledge of charging demand from the city with rich charging data to the new city.There are two core problems in the demand prediction for cross-city charging stations:how to use multi-source data to extract effective features,and how to deal with the domain shift of cross-city multi-source data.Firstly,this paper analyzes the correlation between multi-source data and charging demand,designs the external factors feature extraction module and the charging station configuration feature extraction module to model the impact of multi-source data on charging demand,and extracts the key context information in the feature graph through the spatial attention mechanism.Secondly,the domain shift problem of cross-city multi-source data is caused by the different distribution of characteristic data between the two cities.Therefore,this paper designs a domain differentiation module to learn the deep unchanging features of the charging demand pattern between the two cities.Besides,this paper designs a charging demand prediction module to predict the charging demand and charging demand heat ranking.Finally,an attention-based convolutional domain adaptive network is implemented,and three cross-city charging demand datasets are established.The proposed algorithm is fully compared with other comparison methods to verify the effectiveness of the proposed algorithm.