Predictive State of Charge (SoC) Modeling using Machine Learning Algorithms in Lithium-Ion NMC Batteries

Abstract

The State of Charge (SoC) estimation is crucial in lithium-ion batteries to prevent excessive charging and discharging, impacting the battery's safety, stability, and efficiency. Conventional techniques are the most frequently employed method for estimating SoC. However, they are less accurate in predicting SOC due to their computational sensitivity and difficulty adapting to complex environments. This study proposed four machine learning models: Linear Regression, Multilayer Perceptron, Decision Tree, and Random Forest that were applied for SoC prediction on lithiom-ion NMC batteries. The models' performance was evaluated based on Correlation Coefficient and error values (Mean absolute error or MAE and Root mean square error or MRSE). Based on the result, the Random Forest model exhibited the best performance with a Correlation Coefficient of 1, and MAE and MRSE values of 0.2052 and 0.2712, respectively. Conversely, the Linear Regression model demonstrated the worst performance, with a Correlation Coefficient of 0.9534 and MAE and MRSE values of 5.9064 and 8.2602, respectively.