Abstract:

To address the challenge of accurately estimating battery state of energy under nonGaussian noise interference, an improved maximum correntropy cubature Kalman filter method based on student's t-kernel (ITMCCKF) is proposed. First, the sand cat swarm optimization algorithm is employed for rapid and accurate identification of equivalent circuit model parameters. Next, the cubature Kalman filter is applied to mitigate nonlinear estimation errors. On this basis, the maximum correntropy criterion is introduced and the traditional Gaussian kernel function is replaced with the student's t-kernel function to leverage higher-order information under non-Gaussian noise. In addition, distinct kernel parameters are set for error weight matrix calculations to enhance the ability to process non-Gaussian noise with varying distribution characteristics and improve estimation accuracy. Finally, three noise environments are constructed, combining Gaussian mixed noise, shot noise, uniform mixed noise, and Laplace noise. Tests were conducted based on the FUDS, US06, and BJDST at 25 ℃ and 0 ℃. Experimental results show that compared with MCCKF algorithms, the average root mean square estimation errors of the proposed algorithm were reduced by 50.2%, 53.8%, and 52.8%, respectively, in the three noise environments at room temperature and by 50.4%, 61.0% and 64.1%, respectively, at 0 ℃. These findings verify the generalization ability and robustness of the proposed algorithm under non-Gaussian noise conditions.

Key words: lithium battery, parameter identification, state of energy (SoE), maximum correntropy criterion, cubature Kalman filter