摘要
For fault diagnosis of power transformers, the traditional methods based on the dissolved gas analysis(DGA) DGA cannot establish accurate mathematical models, and single intelligent diagnosis method is unsatisfactory in classification ability in practical applications, resulting in lower accuracy of diagnosis results. In this paper,a transformer fault diagnosis method,which is based on PSO-ELM fusion dynamically weighted AdaBoost,is proposed. Firstly,the PSO algorithm is used to optimize the ELM model to obtain the PSO-ELM model. Then, the PSO-ELM model is iterated through the multi-class AdaBoost algorithm to obtain weak classifiers with different weights. By calculating the classification error rate of test sample for each classifier, the weight coefficient of each weak classifier is continuously adjusted,the weighted vote of all weak classifiers is eventually used for fault diagnosis based on the monitoring data. Experimental results show that,compared with BP neural network,ELM,and support vector machine(SVM), the proposed algorithm raises the diagnosis accuracy by 16.02%,9.78%,and 5.62%,respectively.
- 单位