Hybrid Model for Estimating State of Charge of Lithium-Ion Batteries

The objective of this study is to develop hybrid models to estimate the state of charge (SoC) of lithium-ion batteries (LiB). An LiB dataset from Bayerische Motoren Werke (BMW) i3 60Ah containing physical variables such as voltage, current, and temperature was studied, and a fourth variable was derived, which was the charge obtained by integrating the current over time. The parameters of the equivalent circuit of an RC branch (ECM1RC) were estimated, and Machine Learning (ML) models were trained and evaluated for regression, such as the Extreme Learning Machine (ELM), Decision Tree (DT), and a combination of these, through hybrid and ensemble models and through Linear Regression (LR) to increase the individual performance of the models. After 5 folds of crossvalidation and subsequent testing on a Raspberry Pi 4, a fitting curve R2 of 0.9999 and RMSE of 0.0018% were obtained for the ECM1RC+DT hybrid model using 100,232 testing data. In addition, using the same models retrained using the National Aeronautics and Space Administration (NASA) 18650 LiB data, an R2 of 1.0, and an RMSE of 0.0% were obtained for the ECM1RC+DT using 201,501 testing data.