基于超声时域特征及随机森林的磷酸铁锂电池荷电状态估计

State of charge (SOC) is an important monitoring parameter in the battery management system. Due to the flat open circuit voltage and SOC curve, SOC of LiFeO₄ (LFP) batteries is not sensitive to changes in electrical signals. Therefore, it is difficult to accurately estimate the SOC of LFP batteries. Ultrasonic wave signals can detect changes in the physical properties of electrode materials, and establish a structure-activity relationship to characterize the battery state. In this paper, a SOC estimation method of LFP batteries is proposed based on high-correlation ultrasound features and a low-complexity regression model. Firstly, the consistency and correlation between commonly used ultrasonic features and SOC are analyzed under different conditions such as ultrasonic transmission frequency, current rate, and temperature. Secondly, the time domain ultrasound features of high-correlation are further extended based on the structural features of ultrasound envelope line. After the comparison of data-driven and model-driven methods, an accurate estimation method of SOC is proposed based on random forest model. The experimental results show that the root mean square error and mean absolute error of SOC estimation under different dynamic conditions are lower than 1.9% and 1.6%, respectively, which verifies the reliability and accuracy of this method.