摘要：

Among the evaluation indicators of the remaining useful life of on-board power batteries, accurate capacity estimation is one of the dominant factors for the reliability of electric vehicles. During the battery lifespan, taking into account that a large amount of charging data can provide periodical updates for empirical model, a sequential capacity estimation framework combining data-driven method and empirical model is put forward. The feature variables of incremental capacity are extracted as the training data set of back propagation neural network, framing the online estimation model; simultaneously the Arrhenius fading model is employed for continuous estimation. However, open-loop Arrhenius model suffers from on-board parameter mismatch. To further improve convergence and precision, sequential estimation algorithm is proposed in this paper: the estimation of model parameter adopting the particle filter leads to a closed-loop parameter update; afterwards the updated capacity is fused once again with online identification by the extended Kalman filter. Finally, the results of experiments under alternating temperature indicate that the error of sequential capacity estimation converges to within 2.5%. In addition, the results of robustness analysis prove the stability of the sequential capacity estimation strategy.