Conférence : 2023 IEEE 13th International Conference on Electronics and Information Technologies, 25 septembre 2023
For an effective and economical deployment of battery-powered electric vehicles, mobile phones, laptops, and medical gadgets, the State of Charge (SoC) of the batteries must be properly assessed. It permits a safe operation, have a longer usable battery life, and prevent malfunctions. In this context, the battery management systems provide diverse SoC estimation solutions. However, the Machine Learning (ML) based SoC estimation mechanisms are becoming popular because of their robustness and higher precision. In this study, the features set is prepared using the intended battery cell charge/discharge curves for voltage, current, and temperature. Utilizing statistical analysis and the shape context, the attributes are extracted. Following that, three credible machine learning (ML) algorithms—decision trees, random forests, and linear regression—process the set of mined attributes. The applicability is tested using the Panasonic Lithium-Ion (Li-Ion) battery cells, publicly provided by the McMaster University. The feature extraction and the ML based SoC prediction modules are implemented in MATLAB. The “correlation coefficient”, “mean absolute error”, and “root mean square error” are used to assess the prediction performance. The results show an outperformance of the random forest regressor among the intended ones by attaining the correlation coefficient value of 0.9988.