Research On Power Transformer Faults Diagnosis And Prediction Based On LS-SVM

As the very expensive hub equipment in power system, it plays an important role in the safety and reliability of power system.Its failure will not only affect the reliability of power supply, but also pose a serious threat to the safe operation of power system.Therefore, monitoring, diagnosng and predicting the transformer running state effectively, have a great significance.The least squares support vector machine LS-SVM performed well on structural risk minimization principle of statistical learning theory. It has good generalization ability in little samples and can avoid falling into local minimum. Therefore it has become a powerful tool of intelligent fault diagnosis and prediction gradually. In this article, it is studied for the application of least squares support vector machine LS-SVM in the transformer fault diagnosis and prediction. The main contents are as follows:(1) This article applys the least squares support vector machine LS-SVM to the transformer fault diagnosis model and proposed the solving steps which based on the relationship between Dissolved Gas Analysis (DGA) and the transformer fault. It analyzes the transtormer diagnosis model established by the four types of classification algorithms. They are 1-v-1, 1-v-r, MOC and ECC. The results proves that the model of MOC algorithms has a higher rate of correct judgments.(2) For the LS-SVM parameter selection problem, this article applys the adaptive optimization method to choose the rational diagnosis model parameters. It provides an effectiver way of solving the practical application of LS-SVM.(3) The application of LS-SVM regression algorithm in power Transformer fault prediction is studied. LS-SVM is used to predict the dissolved gase in transformer oil.The results proves that the prediction model is effective and superior.(4) On this basis, this article further combines the transformer fault diagnosis with prediction based on LS-SVM. The predicted value obtained from prediction models based on LS-SVM is the input for the diagnosis model of LS-SVM. And then the potential fault is judged. In this way, it combines the transformer fault prediction and fault diagnosis. So the whole system is abounded. It is more complete and can be more effectively to prevent the transformer fault.