A transformer condition recognition method based on dissolved gas analysis features selection and multiple models fusion

Abstract

Intelligent condition identification is the development trend of transformer condition recognition, However, the existing intelligent condition recognition has the disadvantage of single model and low recognition accuracy. In order to overcome the disadvantage, a transformer condition recognition method based on dissolved gas analysis features selection and multiple models fusion is proposed. At first, artificial bee colony (ABC) algorithm is employed to select a set of optimal dissolved gas analysis (DGA) features. Then, according to the selected feature set, support vector classifier (SVC) model, naive bayes classifier (NBC) model, and back-propagation neural network (BPNN) model are respectively established. Finally, the Dempster–Shafer​ (D-S) evidence theory is used to combine the output results of the three recognition models. The identification performances of the optimal DGA features set and the multiple models fusion are evaluated. The results reveal that the proposed method has the highest accuracy and consistency of 89.39% and 90.49% respectively. With the input vector of proposed optimal DGA features set, the overall performance of multiple model fusion method is superior than the SVC, NBC, BPNN, extreme learning machine (ELM) and K- Nearest-Neighbor (KNN). The accuracy and consistency are increased by 4.54%, 4.52%, and 10.42%, 14.68%, and 3.03%, 3.67%, and 10.6%, 13.45%, and 24.24%, 27.57% respectively. Also, the overall performance of multiple model fusion method with optimal DGA features set is increased significantly compared to multiple model fusion method with IEC Ratio, Rogers Ratio, Doernenburg Ratio, optimal DGA features set selected by particle swarm optimization (PSO) and harris hawks optimization (HHO). The accuracy and consistency are increased by 4.54%, 7.43%, and 6.06%, 7.33%, and 10.6%, 13.58%, and 1.51%, 1.39%, and 9.09%,10% respectively. This proves the effectiveness and the feasibility of the proposed method.