Hybrid rough set and support vector machine for faults diagnosis of power transformer

The available testing data for power transformer are always incomplete and biased, which make the task of power transformer fault diagnosis difficult. This paper presents a method for fault diagnosis of power transformers using the hybrid of Rough Set (RS) and Support Vector Machine (SVM). The approach uses the results of Dissolved Gas-in-oil Analysis (DGA) as the necessary attributes to classify different faults of power transformer. In virtue of the ability rough set has to decrease the amount of data and get rid of redundancy, the method can reduce amount of training data and overcome SVM's defect of slow running speed when processing large dataset. At the same time, by the aid of SVM algorithm the method can classify the core of property set so as to have extensiveness and high identification rate, and avoid disturbance. In particular, the historical data of transformer faults was used for validation and the results show the practical viability of the proposed approach.