Multi-birth Optimization Based on Ergodic Multi-scale Cooperative Mutation Self-Adaptive Escape PSO for Transformer Fault Diagnosis and Location

Abstract

In terms of traditional oil-immersed transformer fault diagnosis and location, major works focus on data feature selection and classifier optimization currently. They are studied as two independent directions due to the difference of solving method. In this paper, ergodic-MAEPSO (EMAEPSO) is proposed, which inherits the ability of classifier parameter optimization from PSO, and by introducing ergodic comparison into Multi-scale Cooperative Mutation Self-adaptive Escape PSO (MAEPSO) to realize feature selection. Based on EMAEPSO, the idea of Multi-birth Optimization by merging two different scale problems simultaneously, feature selection and classifier parameter optimization, is presented to improve the accuracy of transformer fault diagnosis and location. Additionally, considering the scarcity of the fault dataset in some cases, the Random Seed of SMOTE is included into the Multi-birth Optimization for further improvement of diagnostic model. To this end, for the purpose of verifying the generalization and reliability of the idea of Multi-birth Optimization, different types of classifiers are carried out for comparison. Experimental results show that the model optimized by Multi-birth Optimization based on EMAEPSO has a higher diagnostic accuracy, no matter which type of classifier is involved.