Statistical learning techniques and their applications for condition assessment of power transformer

Abstract

The condition of power transformers has a significant impact on the reliable operation of the electric power grid. A number of techniques have been in use for condition assessment of transformers. However, interpreting measurement data obtained from these techniques is still a non-trivial task; correlating measurement data to transformer condition is even more difficult. This paper investigates statistical learning techniques, which is able to learn statistical properties of a system from known samples and to predict the system output for unknown samples. Within the statistical learning framework, this paper develops a support vector machine (SVM) algorithm, which can be utilised for automatically analyzing measurement data and assessing condition of transformers. Case studies are presented to demonstrate the applicability of the developed algorithm for condition assessment of power transformer.