Estimation the state of charge of lithium battery based on variable forgetting factor and improved extend Kalman filter

ZHANG Tao; CHEN Dongming; HOU Pengpeng; WANG Yaobin

doi:10.16186/j.cnki.1673-9787.2022030035

Abstract：Objectives To solve the problem of divergence of state-of-charge(SOC)estimation results of lithium batteries under different discharge stages and noise interference,Methods the factors and reasons affecting the estimation results were analyzed by studying and analyzing the mechanism characteristics of lithium batteries,and then,for the problem of large fluctuations in the estimation error of traditional algo-rithms,the variable forgetting factor recursive least squares(VFF-RLS)in conjunction with the adaptive squareroot unscented Kalman filter(ASRUKF)algorithm was proposed to estimate the SOC.Results Taking the dynamic stress test(DST)as an example,the maximum initial error of the opencircuit voltage of the for-getting factor recursive least squares(FFRLS)algorithm was 0.02 V,the terminal voltage error after stabili-zation was in the range of 0.004～0.010 V,the error convergence time was about 45 s,the maximum initial error of the SOC estimation was 0.3,and it gradually converged to around the theoretical value at about 400 s,and the fluctuation error after stabilization was 0.83%.Under the same conditions,the maximum initial er-ror of the VFF-RLS algorithm in the open circuit voltage experiment was 0.04 V,the terminal voltage error after stabilization was in the range of 0.003～0.007 V,the error convergence time was about 10 s,the maxi-mum initial error of SOC estimation was 0.1,and with the iteration of the algorithm,it converged to around the theoretical value within 200 s,and the maximum fluctuation error after stabilization was 0.413%.Finally,in order to ensure the universality of the application of the algorithm,the convergence of the algorithm was observed under different initial values.Conclusions The results showed that under complex test conditions,compared with the traditional algorithm,the parameter identification speed of the improved algorithm was significantly accelerated,the accuracy was improved,the fluctuation range was significantly smaller in the SOC estimation stage,and it could still converge quickly in the case of large error of the actual value,which proved that the improvement of the algorithm was feasible and could be used for actual battery research.