Abstract: The 3.5 V/20 A•h lithium iron phosphate battery was taken as the research object.Aiming at the online

estimation of its state of charge（SOC），the second-order Thevenin equivalent RC circuit model was established，and a joint SOC estimation algorithm was proposed by combining BP neural network，dynamic forgetting factor recursive least square（DFFRLS）method and adaptive square root unscented Kalman filter（ASRUKF）algorithm. The open circuit voltage- state of charge（OCV-SOC）curve was fitted by BP neural network instead of polynomial to improve the fitting accuracy. The model parameters were identified online by DFFRLS. Combined with ASRUKF algorithm，SOC joint estimation was conducted. The research shows that the proposed joint estimation algorithm effectively eliminates the artificial error caused by the initial value of noise covariance and overcomes the non-positive semidefinite problem of state covariance matrix caused by filtering divergence，so as to obtain the optimal SOC estimation value. Under the experimental condition of cyclic dynamic pressure test（DST），the joint estimation algorithm was compared with other traditional algorithms. The results show that the proposed SOC joint estimation algorithm has better rapidity，convergence and accuracy.

      Key words: battery；BP neural network；dynamic forgetting factor recursive least square（DFFRLS）algorithm；

adaptive square root unscented Kalman filter（ASRUKF）algorithm；SOC joint estimation