Study On SOC Estimation Of Ternary Lithium Battery Based On Genetic Algorithms And Improved BP Neural Network

With the development of the time-table for banning the sale of traditional fuel vehicles in various countries and the increasingly serious environmental pollution problem,many big cities begin to use electric taxis,drip-drip taxis and other means of transportation.Electric vehicle is the trend in the future.Developing an efficient,safe and low-cost power battery management system is one of the key factors restricting the development of electric vehicle.Accurate estimation of state of Charge(SOC)is not only the basis of balancing battery packs in battery management system,but also can improve the efficiency of power batteries and prolong the service life of power batteries.The SOC of batteries can not be obtained directly by measurement,but can only be obtained indirectly by other parameters and methods.Therefore,combined with the performance requirements of electric vehicles for power batteries,this paper takes the commonly used ternary lithium-ion batteries as the research object,and makes a thorough study on how to improve the estimation accuracy of the residual power of electric vehicles power batteries.This paper mainly focuses on the following aspects:(1)The definition of SOC for lithium-ion batteries is described,the factors affecting SOC estimation are analyzed,and the performance parameters of lithium-ion batteries are analyzed by using the built experimental platform,and the advantages and disadvantages of several commonly used SOC estimation methods are analyzed.(2)Through the comparative analysis of the advantages and disadvantages of different estimation methods of SOC,the data-driven model-BP neural network is used to estimate the SOC of batteries.The charging and discharging data of experimental batteries at different rates are collected using the built experimental platform.The BP neural network model algorithm is built in Python environment,and the network is continuously trained by the collected data.Practice,use the network model after training to complete the estimation experiment of battery SOC,and analyze the experimental results.(3)Aiming at the shortcomings of traditional BP algorithm such as slow convergence speed,low estimation accuracy and easy to fall into local minimum,the performance parameters of batteries are carefully analyzed.On the basis of voltage,current and temperature of batteries,the internal resistance of batteries is introduced as the input of neural network model.On this basis,an improved adaptive momentum term BP neural network based on Python programming is proposed.The network algorithm dynamically adjusts the magnitude of momentum factor by using the MSE of the actual output value and expected value of the model,completes the improvement of SOC estimation by traditional BP neural network,overcomes the shortcomings of slow convergence speed and low estimation accuracy of traditional BP algorithm.(4)The selection of BP neural network weights and thresholds has great influence on the performance of the model after training.In pursuit of higher SOC estimation accuracy,the genetic algorithm in bionic algorithm is used to optimize the weights and thresholds of the improved BP algorithm.A genetic algorithm based on Python programming is built to optimize the algorithm flow of BP neural network,and the data collected by the experimental platform are used to modify the algorithm flow separately.The network model before and after training is used to estimate SOC under constant current condition and DST condition respectively.The experimental results fully show that the improved momentum term BP algorithm optimized by genetic algorithm has a significant improvement in convergence speed and estimation accuracy,and can achieve the purpose of accurately predicting battery SOC.