Prediction Of Li-Ion Battery Health Status Based On Improved Particle Swarm Optimization SVM Algorithm

Battery management system(BMS)refers to a critical electronic control unit in the power battery system of electric vehicles.It is capable of detecting and estimating battery status online,especially estimating state of charge(SOC)and state of health(SOH)accurately.Accurate state estimation is conducive to battery life optimization research and guarantees the safe driving of electric vehicles.Firstly,this thesis reviews the existing methods for estimating battery state of charge and state of health.At present,the estimation of battery health status has problems such as time-consuming,complicated computation,and inconvenience for online estimation.Secondly,the principle of lithium-ion battery and the relevant definition of battery SOH are introduced,the relevant factors affecting battery aging are analyzed,and the experimental platform used in this thesis and the experimental process of battery cycle life are introduced.Secondly,analyzing the principle and electrical performance characteristics of lithium battery aging attenuation,extract the characteristic quantity affecting battery aging,clarify the coupling relationship between battery health state and battery SOC,and then designs the battery dynamic stress test condition(DST)experiment,which verifies the effectiveness of the proposed algorithm in estimating SOC.An online estimation method of lithium-ion battery health based on support vector machine is proposed,and the battery charge state and health state are predicted by data-driven method.This method avoids a large number of parameter operations and obtains good prediction accuracy,but the calculation speed and accuracy of the algorithm still need to be improved.Thirdly,in order to optimize the two key parameters of support vector machine algorithm and kernel parameter,it is necessary to optimize the iterative grid search,resulting in slow operation speed,this thesis uses particle swarm optimization support vector machine(PSO-SVM)algorithm to realize the prediction of battery health.Finding the optimal two key parameters through cooperation and competition between particles,simulation and experiments show that the improved algorithm has higher prediction accuracy,faster calculation speed and better global search ability.Finally,in order to improve the stability of SOH prediction of the PSO-SVM algorithm,this thesis introduces the ensemble learning Ada Boost algorithm to optimize the PSO-SVM regression model(APSO-SVM),and combines multiple weak learners to construct a strong regression.The decentralized method solves the problem of a large error at a certain moment when a single learner(PSO-SVM)estimates the battery health state.Simulation and experimental analysis show that the method has good data adaptability and accuracy.The method proposed in this study can simplify the derivation of the corresponding formula and accelerate the convergence speed of operation.The algorithm has good sample adaptability and reliability.It has certain application significance in the research of on-line detection of BMS of electric vehicle battery pack.