In secondary path identification of power transformer Active Noise Control (ANC) system, the background noise makes identification results can’t reflect the characteristic of secondary path, which will affect the stability and convergence of the ANC system. Firstly, this paper analysed the effect of identification error on convergence coefficient taking Fx-LMS algorithm as the example. Then, a new denoising method, of which by combining the time-sharing measurement method with signal correlation analysis, is proposed so that the influence of background noise is almost suppressed. The simulation results demonstrate that the accuracy of identification is significantly improved, showing the effectiveness of the proposed method.