Computational Intelligence and Machine Learning
Gaussian Process Regression for State of Charge Prediction: Overcoming Temperature Variability in Li-ion Battery Applications
Authors: Md Ismail Hossain, KhandokerMainul Islam, Tarifuzzaman Riyad, Md Rafsan Jany, Al Shahriar Zishan and Mohammad Nasir Uddin
Publishing Date: 30-04-2025
ISBN: 978-81-975670-5-6
Abstract
Accurately predicting the State of Charge (SOC) in lithium-ion batteries is essential for efficient energy management in electric vehicles. Due to the nonlinear characteristics of battery behavior, which depend heavily on temperature and operating conditions, SOC estimation presents a significant engineering challenge. This study proposes a data-driven approach using Gaussian Process Regression (GPR) to predict SOC, leveraging a comprehensive dataset with voltage, current, temperature, and average historical measurements. The model was trained and optimized through Bayesian hyperparameter tuning, and itsperformance was evaluated across four temperature conditions (-10°C, 0°C, 10°C, and 25°C) to assess robustness. The results demonstrate high prediction accuracy, with an overall Root Mean Square Error (RMSE) below 0.02 and Maximum Absolute Error (MAE) below 0.1 across all temperature settings. These metrics confirm the model’s reliability and adaptability to varied thermal environments, highlighting the potential of GPR for practical SOC estimation in automotive applications. This approach provides a practical alternative to traditional electrochemical models, supporting advancements in battery health monitoring and energy management.
Keywords
Battery, SOC, Unscented Kalman Filter, Kalman Filter, Nonlinear state, Degrading system.
Cite as
Md Ismail Hossain, KhandokerMainul Islam, Tarifuzzaman Riyad, Md Rafsan Jany, Al Shahriar Zishan and Mohammad Nasir Uddin, "Gaussian Process Regression for State of Charge Prediction: Overcoming Temperature Variability in Li-ion Battery Applications", In: Sandeep Kumar and Kavita Sharma (eds), Computational Intelligence and Machine Learning, SCRS, India, 2025, pp. 121-130. https://doi.org/10.56155/978-81-975670-5-6-10