Condition Monitoring of Power Transformers Using Advanced Machine Learning Algorithms

Abstract

An oil-filled power transformer is the most essential component in the power system network. Various online and offline condition monitoring methods have been adopted to maintain the contingency of the power supply. However, machine and deep learning (ML and DL) have added extra weight to online condition-monitoring techniques for the past few years. This thesis mainly focuses on advancing ML/DL techniques to increase the effectiveness and efficiency of the present techniques using standard and real-time data. Firstly, an advanced feature selection-based classification model is developed to classify the incipient fault inside the oil tank. Dissolved gas analysis (DGA) data are utilized to train the ML model. Next, data balancing and proper feature selection methods are used for data pre-processing, and three well-known classifiers are used to classify the faults. Another DL-based approach is also introduced with more input features. Both approaches are validated using the IEC TC 10 database. Secondly, two advanced regression models are used to investigate two important aspects of transformer health monitoring, i.e. corrosiveness due to sulphur compound and estimation of remaining useful life. Sulfur is known as one of the corrosive compounds inside the tank, which increases the corrosiveness of the oil and reduces the remaining useful life. However, standard test methods have failed to detect the presence of sulfur compounds. Two ML/DL-based regression models are developed to predict the power transformer s corrosiveness and remaining useful life using various properties. Thirdly, three advanced image processing techniques are developed to assess the ageing stages of different insulating oils, i.e., mineral, synthetic and natural ester oils. These oils are subjected to thermally accelerated ageing in the laboratory for a specified time and temperature. After that, a Fourier transform infrared spectroscopy (FTIR) analysis is carried out to analyze different functional groups in the oil.