Research And Application Of Power Battery SOC Estimation Based On Deep BPNN

An electric vehicle is a clean and efficient new type of vehicle driven by power battery,which can reduce air pollution.The state of charge(SOC)represents the remaining power of the battery,which needs to be estimated by the battery management system based on the relevant information collected by the electric vehicle sensor and cannot be directly measured,which is very different from the fuel vehicle.Insufficient estimation of the SOC will cause the battery to be over charged or over discharged during use,which directly affects the battery life and efficiency.Therefore,the accurate estimation of SOC has important significance.After summarizing the current SOC estimation methods,this thesis will aim to improve the accuracy of SOC estimation,take the power battery as the research object,and use the information collected by the sensors during the actual operation of the electric vehicle as the data source,and conduct research on SOC estimation based on deep backpropagation neural networks(BPNN).The specific work is as follows:(1)Considering that the relevant fields in the data set are highly non-linearly related to SOC,a SOC estimation model based on deep BPNN is designed.Compared with traditional BPNN,deep BPNN has a deeper hidden layer,more neurons,and combines the popular deep learning related algorithms,which makes the training effect of the model better and the estimation accuracy higher.And an improved loss function Sickle-L1 is proposed to prevent the over-fitting of the estimated value and the label value during training,and further improve the robustness of the trained model.The experimental results show that the SOC estimation model based on deep BPNN has high estimation accuracy.(2)Because the parameters collected by different types of electric vehicles,the degree of linear correlation between the parameters,and the data noise are different,it is necessary to manually adjust the hyperparameters of the model on different types of electric vehicles to achieve better estimation results,which lacks good portability.A deep BPNN hyperparameter optimization method based on Parallel Bayesian Optimization(PBO)is proposed to automatically search for optimal hyperparameter combinations.PBO can use the collected data to decide the next sampling,and use parallel sampling to further improve the execution speed.The experimental results show that while PBO greatly improves the optimization efficiency,the final hyperparameter combination is also of higher quality.(3)A deep BPNN estimation model training system is designed and implemented.The system can use the offline or online data of electric vehicles to train the power battery SOC estimation model based on deep BPNN,and support model evaluation and export.In addition,the system also includes login management,user management,data query and visualization functions.