Power Distribution Transformer Fault Diagnosis with Unbalanced Samples Based on Neighborhood Component Analysis and k-Nearest Neighbors

In the power distribution transformer fault diagnosis method based on machine learning techniques, the imbalance of the number of cases among various fault categories will lead to a decrease in the accuracy of the fault diagnosis. In order to improve the accuracy and operating efficiency of the power distribution transformer fault diagnosis model, a fault diagnosis model combining the nearest neighbor component analysis with the introduction of correction factors and k-nearest neighbor learning is proposed. Firstly, a correction factor is introduced to the objective function of the neighborhood component analysis algorithm (NCA) to reduce the impact of sample imbalance on model training. Then, the k-nearest neighbors (kNN) classifier is used to map and transform the input data to reduce the distance between samples of the same type, thereby improving the classification performance of kNN. Finally, use the Bayesian optimization algorithm to tune the hyperparameters of the model, and obtain the model parameter set with the highest accuracy of the test set. The analysis results of the calculation examples based on the transformer fault case library show that the proposed model has improved computational efficiency and accuracy of classification results compared with the traditional machine learning diagnosis model. The research of this paper can provide references for the fault diagnosis of power distribution transformer.