An ARIMA-NARX Model to Predict Li-Ion State of Charge for Unknown Charge/Discharge Rates

State of charge (SOC) prediction for Li-ion batteries is an essential feature of a battery management system (BMS). This paper proposes two Autoregressive Integrated Moving Average (ARIMA) models to independently forecast cell current and voltage, respectively and a Nonlinear Autoregressive neural network (NARX-net) model. The battery parameters corresponding to an unknown higher C-rate are forecasted using the parameters corresponding to known C-rates obtained experimentally using a 3.7V, 3.5Ah Li-ion battery. Four algorithms are then used to train a NARX-net to predict SOC for an unknown higher C-rate, performances of which are compared with the experimentally obtained SOC for C/10. The resulting proposed model combining ARIMA and NARX-net predicts SOC with very low error values.