基于充电电压片段和融合方法的锂离子电池SOC-SOH-RUL联合估计

A Joint Estimation Framework of SOC-SOH-RUL for Lithium Batteries Based on Charging Voltage Segment and Hybrid Method

对锂离子电池的荷电状态(state of charge,SOC),健康状态(state of health,SOH)和剩余使用寿命(remaining useful life,RUL)进行准确估计是锂离子电池安全稳定运行的重要保障,该文提出一种结合充电电压片段和等效电路模型(equivalent circuit model,ECM)-数据驱动(data driven method,DDM)融合方法的锂离子电池SOC-SOH-RUL联合估计框架,实现对电池全生命周期的SOC、SOH和RUL的联合估计。首先提取与电池当前容量关联度最高的恒流充电电压曲线片段的上升时间作为健康特征(health factor,HF),利用外部训练集电池的老化数据,离线建立电池老化的最小二乘支持向量机(least squares support vector machine,LSSVM)模型。应用阶段,通过采集待测电池充电电压片段提取HF并代入老化模型进行SOH估计;对该电压区段进行ECM拟合,用阻容参数辨识值和容量估计值建立状态方程和观测方程,结合无迹卡尔曼滤波算法(unscented Kalman filter,UKF)进行SOC估计;用高斯过程回归(Gaussian process regression,GPR)对当前循环次数以前的DV随循环次数的变化进行映射,并借助老化模型预测容量的退化轨迹,实现RUL估计。实验结果表明,所提方法能够联合实现SOC、SOH和RUL的长期稳定估计。

Accurate estimations of the state of charge(SOC), the state of health(SOH) and the remaining useful life(RUL) of the lithium-ion batteries is an important guarantee for the safe and stable operation of the batteries. This paper proposes an estimation based on the partial charging voltage segment and the fusion of the equivalent circuit model(ECM)and the data-driven method(DDM) to achieve the accurate estimations of SOC, SOH and RUL of a longer life cycle of the battery. Firstly, the rising time of the voltage segment which has the highest correlation with the present capacity is extracted as the health factor(HF), and the battery aging model is established offline by least squares support vector machine(LSSVM) with a training set. In the application stage, the HF of the battery to be tested is extracted by fitting the charging voltage segment to estimate the SOH with the LSSVM model.Then, the identified resistance and capacitance parameters acquired by fitting the voltage segment with the ECM and the estimated capacity are used to form the state equation and observation equation, which are combined with the unscented Kalman filter(UKF) algorithm to estimate the SOC. The Gaussian process regression(GPR) is used to map the changes of the DV before the current number of cycles, and the RUL estimation is realized by predicting the trend of DV and SOH combining with the LSSVM model. Experimental results show that the proposed method is able to realize the long-term stable estimations of SOC, SOH and RUL jointly.