Robust Data-Driven Battery State of Charge Estimation for Hybrid Electric Vehicles

This article presents a method to estimate the State of Charge (SOC) in Lithium-ion batteries of Hybrid Electric Vehicles (HEVs) with Artificial Neural Networks (ANNs). The inputs of the SOC estimation algorithm are the measured values of current, voltage, and temperature. In the article, two different battery packs are considered for a power-split full HEV. The training and validation datasets needed for developing the ANNs are generated exploiting a numerical model of two different configurations of an HEV performing real-world driving missions or the Worldwide Harmonized Light Vehicle Test Procedure (WLTP) cycle, while the testing dataset is collected experimentally on battery cells. Specifically, the capacity values for the considered battery pack sizes are 1.82 kWh and 1.06 kWh. The proposed method uses a Nonlinear AutoRegressive with eXogenous input (NARX) recurrent ANN, which has been observed to have reasonable computational cost in prior research. The performance of the investigated technique is demonstrated by estimating the SOC with a low estimation error for both the considered battery sizes. Coulomb counting is used to compute the reference value of the SOC during the real charge/discharge cycles. An analysis of the robustness of the proposed estimation method to offset errors on the measured input current is also performed.