Fault Diagnosis Method of Power Transformer Winding Looseness Based on CEEMDAN-BWO- SVM

In order to effectively realize the classification and identification of power transformer winding looseness faults, This paper presents a diagnosis method for power transformer winding looseness faults, which is based on CEEMDAN and the BWO- SVM model. Firstly, a 10 kV power transformer winding looseness fault diagnosis platform is built to collect, store and analyze vibration signals of transformer tank's surface in real time. Then, we utilized CEEMDAN method to decompose the vibration signal and obtain several groups of IMF components. The characteristic component is selected based on the Pearson correlation coefficient between each IMF component and the initial vibration signal. Power spectrum entropy of characteristic component is calculated to form feature vectors. Finally, Parameters of SVM are optimized by BWO. The BWO - SVM model undergoes training with the training set data, and its validity is confirmed through the test set data. The outcomes indicate that, in comparison with the conventional feature extraction approach and parameter optimization method, the fault recognition accuracy of the proposed method reaches 93.75%. This implies that this method can effectively conduct feature extraction and state recognition for power transformer winding loose faults.