Mixture Maximum Correntropy Criterion Unscented Kalman Filter for Robust SOC Estimation

Because of the existence of non-Gaussian measurement noises, designing robust estimate method of state of charge (SOC) is pivotal for managing battery power. The original unscented kalman filter(UKF) with the mean square error (MSE) criterion based SOC estimation method only performs well under the measurement noises with Gaussian assumption. To improve the estimation accuracy of the UKF under non-Gaussian measurement noise, this paper proposes a novel UKF with the mixture correntropy to accurately estimate SOC. In the proposed method, the mixture correntropy as a cost function (noted as maximum mixture correntropy criterion, MMCC) is used to substitute the MSE in original UKF framework, in which two Gaussian kernel with different kernel width are utilized as the kernel function, and we called it MMCC-UKF. Based on the mathematical model of the second-order equivalent circuit model of battery, a model-driven based novel robustness SOC estimation method is developed by using the proposed MMCC-UKF. Numerical simulations are performed to test the efficacy of the proposed MMCC-UKF based SOC estimation method under various types of non-Gaussian measurement noises.