Research on Transformer Fault Diagnosis Method Based on Improved Random Forest Algorithm

Fault diagnosis of transformer is crucial for its overhaul and maintenance work. In this paper, the first random forest is used to assess the importance of the indicators, derive the weight of each indicator, and furthermore according to the Pearson correlation coefficient, and then delete the indicators with high correlation degree, to get the simplified and optimized evaluation indicators, which reduces the workload of collecting indicators, and also alleviates the number of inputs to the random forest, and improves the efficiency of transformer fault diagnosis. The optimal parameters of the random forest are obtained by using the grid search method to find the optimal parameters of the decision tree number, and finally the random forest classification prediction is carried out in accordance with the optimal decision tree number parameters, which effectively improves the accuracy of transformer fault diagnosis.