Hybrid Deep Learning Algorithm for the State of Charge Prediction of the Lithium-Ion Battery for Electric Vehicles

An accurate and dependable evaluation of the State of Charge (SOC) is required to maximize battery life and safety. The primary goals of this research are to identify dual-polarization parameters and estimate SOC. Dynamic identification of model parameters and estimation of battery SOC is achieved by co-estimating recursive Chicken Swarm Optimization (CSO) and Grey Wolf Optimization (GWO) algorithms from real-time current and voltage measurement data. A dual-polarization model’s projected voltage is nearly the same as the actual voltage to better depict the dynamic properties of the battery and the identification process. Adaptive noise variance updating techniques applied to the extended improved SOC estimates. As a result, the proposed technique is validated using Dynamic Stress Test (DST) data and a Federal Urban Driving Schedule (FUDS). During FUDS testing, an estimated error of less than 2% and a root-mean-square error of less than 0.01085 are observed. We discovered that the approach can withstand erroneous beginning SOCs and other measurement noise covariance in the robustness study.