Electric vehicle battery state of charge estimation using metaheuristic-optimized CatBoost algorithms

Abstract

State of Charge (SoC) estimation plays a crucial role in battery management systems for electric vehicles, directly impacting their operational efficiency and reliability. This study presents a hybrid approach combining the CatBoost algorithm with metaheuristic optimization techniques to enhance SoC estimation accuracy and robustness. The methodology was validated using an extensive dataset collected from 72 real-world driving trips of a BMW i3 (60 Ah), comprising 1053,910 instances of battery and vehicle operation metrics. A comprehensive data preprocessing pipeline was implemented, including missing value treatment, outlier removal, and feature normalization using Min-Max scaling. Three distinct metaheuristic algorithms were investigated: Barnacles Mating Optimizer (BMO), Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Whale Optimization Algorithm (WOA), each integrated with CatBoost to optimize critical parameters including learning rate, tree depth, regularization, and bagging temperature. Experimental results demonstrate that the BMOCatBoost approach achieved superior performance with best-case metrics of RMSE = 6.1031, MAE = 4.1303, and R² = 0.8211, outperforming both PSOCatBoost, GA-CatBoost, and WOA-CatBoost implementations. The framework's effectiveness was validated through rigorous testing, establishing its potential for real-world electric vehicle applications. This research contributes to the advancement of battery management technology, offering promising implications for electric vehicle energy management and broader energy storage applications.