Battery state of charge estimation for electric vehicle using Kolmogorov-Arnold networks

Abstract

Accurate estimation of the state of charge (SoC) in electric vehicle (EV) batteries is essential for effective battery management and optimal performance. This study investigates the application of Kolmogorov-Arnold Networks (KAN) for SoC estimation, comparing its performance against Artificial Neural Networks (ANN) and a hybrid Barnacles Mating Optimizer-deep learning model (BMO-DL). The dataset, derived from simulations involving a lithium polymer cell model (ePLB C020) in an electric car similar to Nissan Leaf EV, encompasses 68,741 instances, divided into training and testing sets. Three KAN models were developed and evaluated based on root mean square error (RMSE), mean absolute error (MAE), maximum error (MAX), and coefficient of determination (R2). Residual analysis indicates that KAN-Model 1 performs the best, with residuals closely clustered around zero and no significant patterns, suggesting reliable and unbiased predictions. KAN-Model 2 also performs well but exhibits some nonlinear trends in the residuals. ANN and BMO-DL models show larger deviations and less consistent performance. These findings highlight the potential of KAN for enhancing SoC estimation accuracy in EV applications.