Electric Vehicle Battery States Estimation During Charging Process by NARX Neural Network

The electric vehicle battery state prediction in real time is an important issue to avoid the risks of burning the battery due to over-charging or dead batteries that are caused by aging. Based on the past works, it is found that the State of Charge (SOC) can be predicted, while predicting the State of Health (SOH) is a difficult challenge. Usually, the SOH is predicted after the end of the driving or the charging cycle under constant conditions; this method is practically impossible because the battery can reach the end of the battery life before achieving the prediction process. In this paper, a SOH prediction method is proposed based on SOC prediction because there is a relation between the SOC and the SOH as indicated by deriving a mathematical model. The prediction process of battery age is achieved during the beginning of the battery charging process under constant conditions of charging, in which a SOC estimation has been implemented by the nonlinear auto-regressive with exogenous input neural network (NARX) with two initial values of SOC, the default value (0%) and practical value (10%) and two charging current rates (0.5C and 1C). The proposed method has been simulated by MATLAB, which several scenarios have been achieved to validate the proposed method. The root-mean-square error (RMSE) values are very promising for both predicting SOC and SOH that are 0.5% and 0.018%, respectively.