State of Charge Estimation for Lithium-Ion Battery Based on NARX Recurrent Neural Network and Moving Window Method

An accurate state of charge (SOC) estimation depends on an battery model. The influence nonlinear and unstable interference factors makes the SOC difficult. To obtain model, a method based NARX (nonlinear autoregressive network with exogenous inputs) recurrent neural moving window is proposed. This paper improves accuracy, modelling speed robustness from following three aspects. First, to overcome excessive reliance amount data in model training process, used establish external input) delay feedback functions can keep input output previous moment add it calculation next moment. Therefore, better results are achieved using small data; second, against gradient explosion vanishing that may occur process. Third, by comparing other methods under different working conditions temperatures, validity proposed verified. indicate has higher accuracy estimation. RMSE performance reduced approximately 65%, execution time shortened 50%.