SOC Prediction Of Ev Battery Based On DE-GWO-BP Neural Network

The development of electric vehicle(EV)plays an important role in maintaining national energy security,reducing vehicle exhaust pollution and ensuring social sustainable development.As an important parameter of electric vehicle power battery,state of charge(SOC)plays an important role in estimating endurance mileage,preventing overcharge and overdischarge of battery and balancing management of battery pack.Therefore,accurate prediction of SOC of electric vehicle power battery is an important task of battery management system(BMS).Based on the real-time collection and upload of the parameters of the power battery to the cloud platform by BMS,in order to achieve the prediction of the SOC of the electric vehicle battery,the prediction method of the SOC of the electric vehicle battery is studied according to the data of the power battery with the sampling period of 10 s.Firstly,the influencing factors of SOC are analyzed theoretically,and the average voltage,current and temperature of power battery are selected as input variables to predict SOC.At the same time,according to the characteristics of the current,the neural network SOC prediction model of the charging and discharging process of the power battery is established.In order to further improve the accuracy and convergence speed of standard BP neural network in SOC prediction of electric vehicle battery,a hybrid improved grey wolf optimization algorithm(GWO)and differential evolution algorithm(DE)optimization BP neural network(DE-GWO-BP)is proposed.Through a non-linear convergence factor,the global search and local search are balanced,and a variable proportion weight is used to make grey wolf dynamic index of leadership Introduction group advance.At the same time,the idea of particle swarm optimization is introduced into the gray wolf algorithm to make the gray wolf individual interact with the individual historical optimal value information,so as to avoid the algorithm falling into local optimal.Finally,we use DE algorithm to mutate,cross and select the parent population,increase the diversity of the population,and avoid GWO algorithm falling into search stagnation.The initial weight and threshold of BP neural network are optimized by DE-GWO to reduce the prediction error of neural network.The standard BP neural network,GWO-BP neural network and DE-GWO-BP neural network are tested with 1914 sets of discharge data and 2504 sets of charging data.The convergence speed and accuracy are compared and analyzed.The experimental results show that the average relative error of DE-GWO-BP neural network is 0.85% in thedischarge process and 0.52% in the charging process,which is significantly higher than that of standard BP neural network and GWO-BP neural network,greatly improving the prediction accuracy,and the convergence speed of DE-GWO algorithm is faster than that of standard BP neural network and GWO algorithm.