Given the shortcomings of the traditional unscented Kalman filter (UKF) algorithm of harmonic state estimation with time-varying noise and abnormal data, an algorithm based on adaptive square root unscented Kalman filter (SRUKF) is proposed for power system harmonic state estimation. First, an improved Sage-Husa noise estimation method is proposed for real-time estimation of noise covariance in view of the time-varying noise. Second, an abnormal data correction method is proposed in view of the abnormal data interference. A correction coefficient is introduced to reduce the influence of abnormal data in state estimation. Finally, an IEEE14-node system is built to validate the estimation performance of the adaptive SRUKF algorithm. It has been applied to the dynamic harmonic state estimation of a power system. The simulation results show that the proposed algorithm has good estimation performance with the interference of time-varying noise and abnormal data. This work is supported by the National Natural Science Foundation of China (No. 61903384 and No. 51477124).