Power transformers condition monitoring using dissolved gas analysis and hidden markov prediction model

Abstract

The reason of the power transformer (PT) monitoring is to prevent the failure of the PT. There are many methods to detect the failure of the PT. The methods include Conventional Monitoring System (CMS), Polarization Depolarization Current (PDC) Analysis, and Hidden Markov Model (HMM). The CMS gives the current condition of PT but it cannot give reliable failure prediction. The PDC involves complicated setup at site and the measurement is done when the PT is off line (shutdown) which is not preferable. HMM is a prediction model based on dissolved gas analysis (DGA) database. Its accuracy is believed to be further improved when more DGA data are available with the passing of time. The main focus of this project is to obtain the PT failure time estimates with an error of less than or equal to 10%. Mathematical models were used to predict the PT condition at several stages by knowing the current DGA data. Result shows the accuracy of 90% in transformer level prediction, means that 9 accurate results out of 10 transformers tested. The technique can be used to predict the transformer deterioration level and to prevent transformer failure which can lead to tremendous losses to company. The result will assist the maintenance personnel to make various maintenance decisions with cost effective way.