On the feature selection for battery state of health estimation based on charging-discharging profiles

Correctly evaluating the health status of the battery is of great significance for ensuring the safety of electric vehicles, and avoiding potential failures of electric vehicles. Recently, the data-driven methods have raised interest in evaluating battery the battery state of health (SOH) based on the statistical theory. However, the accuracy of the battery state of health estimation algorithms is greatly affected by the model input selection. Because of the limitation for battery data type, it is meaningful to extract the useful data information from the raw data. In this work, we extract health indicators from the battery current, voltage, temperature data based on the laboratory measured experimental data, which can inform model input choices, thus improving the accuracy in battery health estimation. Then, grey relation analysis is used to quantify the correlation between health indicators and battery capacity degradation, and using this quantified result as the basis for the selection of model variables for battery modeling. According to the correlation degree value which calculated by grey relation analysis, it shows that most health indicators are more related to the battery heath. The value of correlation degree for most features are above 90%, and the lowest value is 69%. Finally, the performance of the estimated model based on these health indicator is evaluated.

Automated summary

This study extracts health indicators from battery current, voltage, and temperature data based on laboratory measured experimental data to improve the accuracy in battery health estimation. Grey relation analysis quantifies the correlation between health indicators and battery capacity degradation, showing that most health indicators are closely related to battery heath, with correlation degree values mostly above 90%.