State of Energy Estimation of a Lithium-Ion Battery Based on Different Regression Techniques

Battery Management Systems (BMS) are rapidly evolving due to the rising utilization of lithium-ion batteries (LiBs), especially in electric vehicles (EVs). One of the crucial metrics in BMS is the State of Energy (SOE) which plays a vital role in assessing battery performance and ensuring effective energy management. This study examines the application of four machine learning regression methods namely Random Forest (RF), Support Vector Regression (SVR), Decision Trees (DT), and Gradient Boosting (GB) for a precise estimate of SOE. The evaluation employs empirical data obtained from the Braunschweig City Driving Cycle (BCDC) at three ambient temperatures: 15 °C, 25 °C, and 45 °C. Two scenarios were built using the dataset: the first scenario uses the smoothed SOE as the trained model, while the second scenario evaluates model performance using a fresh feature set derived from the power dataset. This work concludes that instance 2 has superior performance at 25 °C in which RF model consistently outperforms other methods in predicting SOE. RF attains the lowest RMSE values under all temperature circumstances, demonstrating its exceptional capacity to model complex non-linear relationships while preserving robust generalization via ensemble averaging.