Enhanced NARX Neural Network Model for Specific Time-Series Prediction. Case Study: Battery State of Charge in Electric Vehicles

The paper presents an enhanced Nonlinear Autoregressive Exogenous (NARX) neural network model designed to address the complexities of specific time-series variables, enabling accurate predictions even for variables with different initial values. A serious problem that is not explicitly expressed in literature. A case study on predicting lithium-ion battery State of Charge (SOC) for electric vehicles (EVs) application is featured in this research. In fact, the battery SOC is a time-series variable, and its initial value can exhibit fluctuations due to diverse factors. The obtained simulation results demonstrate the capacity of the enhanced NARX neural network to provide satisfactory estimations, compared to conventional algorithm, even when dealing with different initial battery SOC values.