In recent years, low frequency noise produced by transformers has attracted many attention and became a hot research point in noise source control. In this paper, bubble sort is adopted to automatically identify the frequency components with most energy. Based on the identification and with the application of perturbation technique and merging algorithm, secondary sound source is automatically produced to neutralize the primary sound source made by transformers, in this way the noise from transformers can be eliminated. The noise data of an actual 110 kV substation is measured and analyzed theoretically. Simulation results show that the proposed method is effective and correct.