Research On Load Forecasting Method Of Electric Vehicle Charging Station Based On Deep Learning

With the rapid development of economy and society,the problems of energy consumption and environmental pollution are increasingly causing widespread concern.With the advantages of energy saving and environmental protection,the number of electric vehicles is increasing and the charging demand is also increasing.The random charging behavior can cause a huge charging burden to the power grid.Therefore,the realization of accurate prediction of EV charging load can provide a reliable reference for the power dispatch of the power system and provide the necessary guarantee for the stable operation of the power system.Many current research solutions on EV charging station load prediction rely on the guidance of manual experience for charging load modeling,and the calculation process is complicated,while the spatial and temporal distribution characteristics of charging stations are not fully considered.Considering that the deep learning prediction algorithm can give full play to the data-driven advantages and overcome the defects of traditional prediction methods that rely too much on manual experience,while its powerful feature extraction capability can,to a certain extent,solve the problem of gradient explosion that tends to occur in traditional neural networks when the data volume is large.This paper introduces generative adversarial networks and gated recurrent units in deep learning algorithms,fully combines the advantages of each algorithm,and optimizes and improves the algorithm to propose a GRU-GAN-based spatio-temporal load prediction model for electric vehicles.The DBSCAN algorithm is used to aggregate the spatial dimensions of charging stations and extract the spatial information of charging stations effectively.A generator composed of gated cyclic units is used to fully extract the temporal characteristics of the load data.Convolutional neural network is used as the model discriminator and the loss function is improved.The model can make accurate predictions of future charging loads by generating adversarial networks for game training.In this paper,we use the charging data of the Boulder area in the U.S.as the actual example,and provide effective data support for the validation of the algorithm model through load data transformation,reconstruction,and normalization pre-processing.Through experimental design,experimental environment configuration and network model construction,the algorithm model proposed in this paper is trained and validated.Through comparison experiments and analysis of experimental results,it is verified that the GRU-GAN-based spatio-temporal load prediction model for EV charging stations proposed in this paper can fully consider the spatio-temporal characteristics of charging load and can achieve high prediction accuracy.Comparing with other models,it is verified that the model in this paper has a good prediction effect and takes into account the generalization ability.