Smart Energy Management System: Oil Immersed Power Transformer Failure Prediction and Classification Techniques Based on DGA Data
Oussama Laayati, Hicham El Hadraoui, Mostafa Bouzi, Ahmed Chebak
Abstract
The power transformer is the key element in the electrical grid. The failures of the power transformer impact critically the grid, can cause energy loss and blackouts. In energy production, transmission, distribution, and industrial applications, the oil immersed power transformer is the most used. The maintenance of this key equipment is highly important which can be done with different techniques such as thermal and vibration analysis, frequency analysis using wavelet transform and dissolved gas analysis. The application of predictive maintenance of the power transformer represents an important feature of the Smart Energy Management System in micro grids, that can reduce the percentage of failure occurrence while increasing the availability of the power transformer and prevents blackouts. This paper represents different failure classification techniques based on dissolved gas analysis data mainly logistic regression, multiclass jungle, multiclass decision tree and artificial neural network. The application can diagnosis the power transformer failures based on the parts-per-million of the different gas generated in the oil. The results of applying different types of classification algorithms shows the best technique to be part of a bigger system of monitoring and diagnostic of different installed equipment in a micro grid. The implementation of such application in real time energy management system requires different type of sensors and the interaction of offline database, the paper also shows the steps to integrate the algorithm in the Smart Energy Management System.