基于自适应UKF的锂离子动力电池状态联合估计

针对由静态的电池模型参数而造成的状态估计累计误差、噪声统计特性的时变不确定性等实用化的问题，基于无迹卡尔曼滤波(unscented Kalman filter， UKF)框架设计了一种自适应UKF的电池状态联合估计算法.在无迹变换(unscented transform，UT)时，对量测方程进行准线性化处理，降低了循环迭代过程中的计算开销；利用带遗忘因子的Sage-Husa自适应估计方法对过程噪声的统计特性参数进行递推估计与修正，提高了UKF估计算法的自适应容错能力；实时跟踪滤波的收敛性，若呈发散趋势时，通过自适应衰减因子对误差协方差进行调整以抑制滤波发散，保证了滤波过程的数值稳定性；采用联合估计策略对一阶Thevenim电池欧姆内阻模型参数进行在线更新，以确保动态测试工况下电池模型的准确性，从而提高了电池荷电状态(state of charge，SOC)以及电池健康状态(state of health，SOH)的估计精度.实验与仿真结果验证了该电池状态联合估计算法的可行性与有效性.

State Co-estimation Algorithm for Li-Ion Power Batteries Based on Adaptive Unscented Kalman Filters

On the practical issues such as accumulative state estimation errors caused by the inaccurate battery model， and time-varying and unknown noise statistical characteristics， a state co-estimation algorithm for Li-ion power batteries based on adaptive unscented Kalman filter (UKF) framework is proposed. During the iteration process， the measurement equation called by unscented transform (UT) each time is quasi-linearized， which could greatly reduce the computational complexity. The statistical characteristic parameters of unknown noises can be estimated by introducing Sage-Husa adaptive estimation method with forgetting factor， which could improve the adaptive fault tolerance of UKF estimation algorithm. If the divergence trend occurs in the filtering process， the error covariance will be penalized by employing adaptive attenuation factor to prevent the divergence of the filter， accordingly guaranteeing the numerical stability of filtering process. By employing co-estimation strategy， the inner ohmic resistance of the 1st-order Thevenin battery model， which could characterize the state of health (SOH) indirectly， can be estimated and adjusted in real time， and therefore a high-precision estimation of the state of charge (SOC) is able to be obtained owing to the accuracy of the battery model during the process of dynamic charge-discharge cycling tests. The comparative experimental results verify the feasibility and effectiveness of the adaptive co-estimation algorithm.