Parallel neural network combined with adaptive Kalman filter for co-estimation of SOH and SOC of lithium-ion batteries

Accurate estimation of a battery's states, particularly the state of health (SOH) and state of charge (SOC), can affect its reliability and safety. This paper proposes a parallel neural network (PNN) structure to estimate the SOH and SOC of lithium-ion batteries, where back propagation neural network and Elman neural network are considered. The PNN is first constructed to estimate the SOH, then, the estimated SOH is applied as one of the inputs for estimating the SOC considering the correlation between the SOH and SOC. In order to enhance the robustness of the model, the SOC is further optimized by utilizing an adaptive Kalman filter (AKF). The experimental results exhibit that the proposed method has good performance than those single networks. The mean absolute error (MAE) and root mean square error (RMSE) of the SOH estimated by PNN with NASA battery dataset can be as low as 1.01 % and 1.27 %, while with Oxford battery dataset are close to 0.46 % and 0.51 %. Moreover, the MAE/RMSE of the SOC estimation with NASA and Oxford battery dataset at specific SOH values are within 0.56 %/0.57 % and 0.13 %/0.15 %, respectively.