State-of-charge estimation and prediction by machine learning models using experimental dataset of lithium-ion batteries based on ionic liquid modified LiFSI electrolyte

Accurate state-of-charge (SOC) estimation and prediction in Lithium-ion battery (LIBs) remains a critical issue due to nonlinear battery behavior and environmental fluctuations. This work explains the development of Machine learning (Support vector machines, Decision tree, Gradient boosting) models to estimate and predict accurate SOC of LIBs wherein the experimental data was used from LIBs with LiFSI and pyridinium-based ionic liquid electrolytes. To achieve the optimal model performance, key hyperparameters were tuned using max_features_input, max_depth of models, max_split_data, etc. Feature importance revealed that discharge capacity was the most influential feature, confirming the model's reliability. Optimized Decision tree model attained an exceptional SOC accuracy with low Root mean square error (RMSE) = 0.001298, Mean square error (MSE) = 0.00000168, and R² = 0.999948 in 0.301193 s for experimental data. These findings demonstrate the potential of machine learning approach to enhance SOC estimation and the longevity, safety, and capacity of LIBs.