Abstract: In view of the interference of outliers appearing in observed voltage data during the process of SOC estimation of traction battery, an improved UKF algorithm is proposed, which corrects the observed noise model into the scaled-contaminated normal distribution model and utilizes Bayesian theorem to calculate the posterior probability of outliers, which can be used as weighting coefficients to adaptively adjust the filter gain and state covariance. This method can effectively overcome the problem of outlier interference, but when there are errors in initial SOC setting, it may mistakenly regard the normal value of observed voltage data as the outliers and only adjust it with small filter gain control quantity, leading to the slow convergence or even divergence of the algorithm. Therefore, a sub-optimal fading factor based on the strong tracking principle is introduced in the initial stage of the algorithm to track the target quickly, remedying the inadequacies of the simple anti-outlier method mentioned above. The test results show that the improved UKF algorithm has strong robustness, good tracking speed and accuracy, providing a new method for outlier interference resistance in estimating the SOC of traction battery

Key words: state of charge, outliers, unscented Kalman filtering algorithm, Bayes theorem, strong tracking principle