Power Transformer Fault Diagnosis Method Based on Genetic Algorithm Optimized Random Forest

Dissolved gas analysis (DGA) of oil has become an effective means of transformer incipient fault diagnosis due to its ability to avoid the influence of complex electromagnetic fields and noise from the outside environment. However, the accuracy of transformer fault diagnosis based on DGA technology needs to be improved. This paper presents genetic algorithm (GA) employed to optimize random forest (RF) parameters to improve the accuracy of transformer fault diagnosis. Firstly, by comparisons of different DGA ratio selections, the non-code ratio method of dissolved gas in oil is determined as the input of the model. The number of RF base classifiers, the decision tree's maximum depth, and the number of features for decision tree node splitting are encoded into chromosomes using the binary method. The accuracy of RF is used as the fitness of individuals to establish the GA-RF model. The diagnostic performance of the model is verified using fault instance data. The results show that the diagnostic accuracy of GA-RF is the highest when the non-code ratio of DGA is used as the input of the model and the accuracy of the GA-RF model is significantly better than that of the support vector machine (SVM), decision tree (DT), RF, and K-Nearest Neighbor (KNN) models.