Investigation of a Decision Tree Approach for Charge Balancing in Lithium-Ion Batteries

In multi-cell battery configurations, cells naturally develop charge imbalance due to variations in manufacturing, aging, and operating conditions. Effective cell balancing is important to mitigate these issues by equalizing the charge and discharge rates across all cells, preventing overcharging or deep discharging of any single cell to enhances the overall performance, reliability and maximizes its lifespan. This research investigates the application of machine learning (ML) models to enhance the balancing in batteries. Conventional algorithms often struggle with constraints related to time and efficiency. To address these issues, in this research, ML models are trained to select a balancing resistor taking into consideration the degree of imbalance to improve the balancing time. Classification and Regression Trees (CART) algorithm are employed to develop models that mimic and optimize over conventional methods on eight cells in 4S2P configuration. The model's performance was evaluated using the Fl score, yielding results of approximately 0.99 for both the training and testing datasets across all targets. Subsequently, the model was implemented in MATLAB Simulink. The results demonstrated the model's ability to apply its learned knowledge effectively and showed improvements in balancing time compared to conventional techniques.