To address the problem of reduced accuracy or failure in traditional non-intrusive harmonic impedance measurement methods caused by large fluctuations in background harmonics or abrupt changes in harmonic impedance, a novel estimation method for system harmonic impedance is proposed. First, the distance correlation coefficient is used to filter out data with strong correlation between harmonic voltage and current amplitudes, thereby reducing the impact of background harmonic fluctuations on the impedance estimation results. Then, the Bayesian change-point detection algorithm is employed to identify abrupt changes in the coarse estimates of harmonic impedance, and data are grouped accordingly based on the detected change points. Finally, a mean shift parameter is incorporated into the harmonic impedance regression model. By introducing a threshold rule for the penalty function and the Bayesian information criterion (BIC), robust regression is performed on the grouped data to obtain the optimal estimation of the harmonic impedance, thus mitigating the influence of outliers on the estimation results. Simulation results indicate that the identification of impedance change points in the filtered data is more accurate, and the precision of the grouped impedance estimation results is higher. This provides a new approach for harmonic impedance estimation in scenarios with background harmonic fluctuations and impedance changes.