Dissolved Gas Analysis and Application of Artificial Intelligence Technique for Fault Diagnosis in Power Transformers: A South African Case Study

Abstract

In South Africa, the growing power demand, challenges of having idle infrastructure, and power delivery issues have become crucial problems. Reliability enhancement necessitates a life-cycle performance analysis of the electrical power transformers. To attain reliable operation and continuous electric power supply, methodical condition monitoring of the electrical power transformer is compulsory. Abrupt breakdown of the power transformer instigates grievous economic detriment in the context of the cost of the transformer and disturbance in the electrical energy supply. On the condition that the state of the transformer is appraised in advance, it can be superseded to reduced loading conditions as an alternative to unexpected failure. Dissolved gas analysis (DGA) nowadays has become a customary method for diagnosing transformer faults. DGA provides the concentration level of various gases dissolved, and consequently, the nature of faults can be predicted subject to the concentration level of the gases. The prediction of fault class from DGA output has so far proven to be not holistically reliable when using conventional methods on account of the volatility of the DGA data in line with the rating and working conditions of the transformer. Several faults are unpredictable using the IEC gas ratio (IECGR) method, and an artificial neural network (ANN) has the hindrance of overfitting. Nonetheless, considering that transformer fault prediction is a classification problem, in this work, a unique classification algorithm is proposed. This applies a binary classification support vector machine (BCSVM). The classification precision is not reliant on the number of features of the input gases dataset. The results indicate that the proposed BCSVM furnishes improved results concerning IECGR and ANN methods traceable to its enhanced generalization capability and constructional risk-abatement principle.