Transformer fault identification method based on improved oversampling and MSFO-SVM

Aiming at the problem of low fault identification rate caused by transformer fault sample imbalance, a transformer fault diagnosis method for improved oversampling and improved sailfish algorithm (MSFO) optimization support vector machine (SVM) is proposed. Firstly, TLR, K-means++ and the introduction of anticosine function improved oversampling (SMOTE) are used to process the fault data. Then, the Tent chaotic mapping, the new sailfish exploration mechanism and the elite inverse learning strategy of the optimal individual are used to improve the swordfish algorithm (SFO), and the superiority of the improved SFO(MSFO) algorithm is proved by comparing it with the original SFO, PSO, and GWO optimization tests. Then, an MSFO optimized SVM model is constructed. Finally, it was compared with PSO-SVM, GWO-SVM and SFO-SVM. It can be seen that when the balanced training set is used, the diagnostic accuracy of each model is improved by 6.48%, 5.56%, 5.56% and 8.34% compared with the original training set, respectively, which verifies the effectiveness of the data balancing method in this paper. In the case of the same balanced training set, the fault diagnosis rate of MSFO-SVM was the highest, indicating that MSFO-SVM had stronger fault identification ability.