Research On Nonlinear Autoregressive Estimation Of Power Battery State Of Charg

In recent years,the rapid growth in the number of traditional vehicles has exacerbated oil resource consumption and environmental pollution.Therefore,people have been looking for a more environmentally friendly way to travel,and electric vehicles have entered the public’s vision.The wide application of electric vehicles can deal with resource crisis and environmental pollution,and meet the demand of time development of carbon peak and carbon neutralization.As a significant part of electric vehicle power battery pack,battery management system（BMS）can realize the function of real-time data acquisition,information communication,energy management,protection alert and state estimation.Among them,high-precision state of charge（SOC）estimation has a significant impact in feedbacking driving information to users,balancing the state of single battery,preventing overcharge/overdischarge and improving service life of power battery.Therefore,facing the technical bottleneck of state of charge estimation,a state of charge estimation method considering the characteristic of input data and improving the evaluation index of fitness function is designed,hoped to improve the accuracy of state of charge estimation under complex working conditions.This thesis focuses on the following research work:Firstly,the factor affecting the variation of charge state of power battery are analyzed.Aiming at the problem of input data selection,the factor that may affect the variety of state of charge is analyzed on the simulation platform,and the cause of relevant factors affecting the variety of state of charge is determined from the mechanism.Finally,the ambient temperature,battery voltage and battery current are determined as the input data of state of charge estimation.On this basis,the correlation analysis of temperature time sequence,battery voltage time sequence,battery current time sequence and state of charge time sequence are carried out to determine the initial value of model sequence delay of input and output data.Secondly,the state of charge estimation model is established.In view of the time sequence characteristics of input and output data in the process of state of charge estimation,a nonlinear autoregressive（NARX）neural network with external input is designed as the state of charge estimation model,which can take into account the input and output data at the previous time.Meanwhile,in view of the deficiency that only weight parameters,such as weights and biases,are considered in traditional neural network,a state of charge estimation model considering neural network topology parameters,such as input delay,output delay,number of hidden layer neurons,and weight parameters is designed,and the result is optimized by improved beetle antennae search（IBAS）.The collaborative identification of topology parameters and weight parameters of the state of charge estimation model is realized by modifying the evaluation index of fitness function,Finally,the validity of the state of charge estimation model is verified under complex working conditions.The relevant data obtained from the dynamic stress test（DST）condition and the world-wide harmonized light vehicles test cycle（WLTC）condition is inputted into the state of charge estimation model.The simulation result shows that the NARX neural network optimized by the IBAS algorithm can significantly improve the accuracy of state of charge estimation.Under the DST condition and the WLTC condition,the root mean square errors are 3.38×10^-3 and 2.92×10^-3.Taking root mean square error as standard,compared to the NARX neural network method optimized by the traditional beetle antennae search algorithm,the estimation accuracy is improved by 42.4%and 45.7%.The maximum errors are only 0.76%and 1.27%respectively.Compared with the NARX neural network optimized by the traditional beetle antennae search algorithm,the estimation accuracy is significantly improved.It is proved that the method proposed in this thesis can effectively improve the estimation accuracy of the state of charge,and can ensure good estimation performance under complex working conditions.