Enhancing State of Charge Prediction of Lithium-Ion Batteries Through Linear Polynomial Regression – Support Vector Machine Modeling with Temperature Varying and Open Circuit Voltage Compensation

Abstract

Accurately estimating lithium-ion batteries' State of Charge (SOC) and Open Circuit Voltage (OCV) is vital for efficient energy storage system management, especially in electric vehicles. Temperature fluctuations pose a significant challenge to precise SOC estimation. This paper presents a novel method that combines Linear Polynomial Regression (LPR) and Support Vector Machine (SVM) with temperature compensation. LPR models the temperature-dependent OCV-SOC relationship, and SVM fine-tunes SOC prediction, considering complex non-linearities. Experimental results show remarkable improvements. When comparing RMSE values, without temperature compensation, it ranges from 5.1% at - 10°C to 3.7% at 25°C. After applying the proposed method, it drops to 3.0% at - 10°C and 1.8% at 25°C. For MAE, it decreases from 4.8% at - 10°C and 3.2% at 25°C without compensation to 2.9% and 1.7%, respectively, with compensation. The LPR-SVM model outperforms other methods, with the lowest RMSE, demonstrating its effectiveness in enhancing SOC prediction accuracy across different temperatures.

Keywords: Linear Polynomial Regression, Temperature Compensation, Electrochemical Dynamics, Non-linear Regression, Support Vector Machine