A Novel Dissolved Gas Analysis Approach using Support Vector Machine based on Data Generation Method

Abstract

In this paper, new data are generated from existing gas data and a method of gas analysis based on machine learning based oil-filled transformer is proposed to solve the problems based on the IEC 60599 diagnostic criteria and to improve the diagnostic performance. It is difficult to acquire a large amount of data at one time and it is difficult to measure the performance with existing data. In IEC 60599, if diagnosis criteria do not exist or it is on the duplication area, it is difficult to make a decision without any expert’s analysis. In the proposed data generation method, new data is generated for each class from existing gas data and combined with existing data. The proposed method based on a machine learning based dissolved gas analysis of a power transformer using support vector machines addresses the issue inherent in IEC 60599 diagnosis criteria. To validate performance of the proposed method, the effectiveness of the machine learning based oil-filled transformer gas analysis method using newly generated gas data was verified. As a result, compared with the existing IEC diagnostic reference method, applying the machine learning based method using the generated data showed higher classification performance.