Extended Kalman Filter Algorithm Based On Neural Network Optimization For State Of Charge Estimation Of Lithium-Ion Battery

Since the 21st century,the world has faced extremely serious environmental pollution and energy crisis.Based on the background,new energy vehicles have been vigorously developed.Due to its excellent performance,lithium batteries have gradually become one of the most important energy sources for electric vehicles.With the use of the power battery,if the state of charge(SOC)is accurately estimated,the utilization efficiency of the battery energy will be improved,in addition,the service life of the battery will be prolonged as well.However,the battery SOC cannot be obtained by measurement directly,which can be calculated by other parameters,such as voltage,current and so on.Therefore,from modeling the battery and estimation algorithms,the paper focuses on the lithium battery cells,and analyzes the commonly used methods for estimating the battery SOC.The specific research of the paper includes:(1)According to the basic principle and parameter characteristics of lithium battery,the second-order Thevenin model is selected as the equivalent circuit model.The equivalent circuit model of the lithium battery is built in the MATLAB/SIMULINK workspace,and the parameters of the lithium battery model are researched in the other paper.Which helps configuring the unknown parameters and obtaining a nonlinear relationship between the open circuit voltage(OCV)of the battery and the SOC.The simulation of constant current discharge condition,pulse discharge condition and UDDS condition is carried out,from which,the terminal voltage and current data of the battery are collected.With these data,the SOC of the three kind of working conditions is calculated by the ampere-hour integral method.The value is seemed as the theoretical value of SOC.(2)The BP neural network algorithm is used to estimate the SOC of the battery.The voltage and current are selected as the input of the BP neural network,and the SOC is used as the output of the neural network.After determining the parameters of the input layer and output layer of the neural network,and the excitation function of each layer as well,amounts of the simulation experiments are carried out to determine the number of nodes in the hidden layer of the neural network.Training the BP neural network with the sample data,then the BP neural network is used to perform the constant current discharge condition,the pulse discharge condition and the UDDS condition.The SOC is estimated and analyzed for the performance of the neural network.(3)The EKF algorithm is used to estimate the SOC of the battery.The principle of the KF algorithm is analyzed.The SOC of the battery is estimated by the EKF algorithm after the state variables of the system are selected and the state equations are listed.The EKF algorithm is used for the three kind of operating conditions of the battery,then the SOC is estimated and compared with the theoretical value to verify the validity of the EKF algorithm.Changing initial value of the SOC and adding noise to the current analyzes the stability of the algorithm.(4)Based on the EKF algorithm,BP algorithm is added for optimization.Firstly,the structure of EKF-BP algorithm is used.The error of EKF algorithm is used as the output of BP neural network,and the Kalman filter coefficient,voltage value and SOC value are selected as input of BP neural network.The number of nodes in the hidden layer is gotten from a large number of simulation experiments.Using the trained EKF-BP algorithm estimates the SOC of the three kind of operating conditions of the battery,and then change the initial value of the SOC and add noise to the current,compared with the EKF algorithm,to study the performance of the optimized algorithm.