Research On Load Forecast Of Electric Vehicle Charging Station Based On Deep Neural Networks

With the depletion of fossil energy and the deterioration of the natural environment,more energy-efficient and cleaner electric vehicles have become an inevitable choice for people's future trips,and the number of electric vehicle charging stations in the city has also increased.Such a large number of electric vehicle loads will definitely have a huge impact on the power grid.To eliminate the negative impact of electric vehicle loads on the grid,and convert electric vehicle loads into tools that are beneficial to the stability of the grid.The key lies in the accurate electric vehicles station load forecast.When electric vehicle charging equipment and other smart sensors record data,it is inevitable that data will be missed,duplicated and abnormally recorded due to equipment failure,power failure,and other factors.To eliminate the bad effects caused by these bad data during model training,the missing data are filled by the principle of the same data type,the adaptive sorting neighbor algorithm is used to eliminate duplicate recorded data,and the K-means algorithm is used to detect abnormal recorded data.Meanwhile,in order to solve the problem of large differences in the magnitude of various types of load forecasting data,the data is standardized by the data standardization methods.Aiming at the problems of low accuracy of traditional electric vehicle charging stations load forecasting and weak generalization ability of load forecasting models.A model that combines the random forest algorithm and the convolutional neural networks is proposed.The model uses a random forest algorithm to extract time-series load prediction data to complete the load prediction of a single charging device.The relative position information of each charging device is recorded by establishing a space-time matrix and then uses a convolutional neural network to complete the feature extraction of the two-dimensional matrix.Finally,the load forecast of the electric vehicle charging station is achieved.Taking the charging station data of Duanjiatan,Lanzhou,Gansu as an example,the RF-CNN model is compared with support vector regression machine,BP neural networks and long-short term memory neural network models.The experimental results show that the proposed charging station load forecasting method based on RF-CNN has high prediction accuracy and strong generalization ability.Accurate electric vehicle charging station load prediction plays a very important role in improving the economics and stability of the power system.Intelligent detection equipment and intelligent sensors are widely deployed on electric vehicles and generate a large amount of data.These data are complex and diverse,most electric vehicle charging station load forecasting methods cannot handle such large,complex and diverse data well.In order to fully extract the training data features and improve the load forecasting accuracy of electric vehicle charging stations,a GRU-CNN hybrid neural network model combining a gated recurrent unit(GRU)and a convolutional neural network(CNN)is proposed.The GRU module can effectively extract the feature vector of time series data,the CNN module can extract two-dimensional matrix data features.The proposed model was tested in actual examples and compared with BP,SVR and LSTM prediction models.The GRU-CNN hybrid neural network predicted the lowest MAPE and RMSE values.The experimental results show that the proposed GRU-CNN model can make fuller use of the data and achieve more accurate load forecasting of electric vehicle charging stations.