Haw, Jia Yong (2023) Machine learning algorithm to determine stray gassing in transformer oil based on dissolved gases. Masters thesis, Universiti Tun Hussein Onn Malaysia.

Preview	Text 24p HAW JIA YONG.pdf Download (504kB) | Preview
	Text (Copyright Declaration) HAW JIA YONG COPYRIGHT DECLARATION.pdf Restricted to Repository staff only Download (347kB) | Request a copy
	Text (Full Text) HAW JIA YONG WATERMARK.pdf Restricted to Registered users only Download (3MB) | Request a copy

Abstract

Transformer is part of the most important electrical equipment in the distribution power system network. However, transformer will exhibit faults under long operation which may often be confused with the stray gassing (SG) phenomenon. Nevertheless, this SG phenomenon is still being left out by many researchers and the root causes are yet being identified. The current SG test and interpretation methods used by the industries is also very inefficient and inaccurate, where the existing Duval Pentagon Method (DPM) interpretation method is proven with a low accuracy of only 58.7% to interpret stray gassing condition. Therefore, an accurate and fast interpretation tool is required to solve the dissolved gas analysis (DGA) interpretation and gassing pattern of transformer materials should be investigated to help in transformer root cause determination. This research work involves the use of ensemble-based machine learning (ML) algorithms to improve the interpretation accuracy of DPM. Three ML models are developed, and all the models are also showing promising results with more than 70.0% of interpretation accuracy to interpret three different transformer conditions. The random-under-sampler (RUS)boosted trees model is the best model with the interpretation accuracy of 81.2%. Besides, three different transformer materials, which are insulation paper, core metal, and gasket were experimented with uninhibited and inhibited transformer oil under heat to understand the gassing behaviour caused by the materials. The findings indicate that for uninhibited oil, the insulation paper effects the generation of hydrogen (H2), carbon monoxide (CO) and carbon dioxide (CO2) gases, the core metal effects generation of CO gases, and the gasket effects the generation of H2, methane (CH4), ethylene (C2H4), CO and CO2 gases. For the inhibited oil, the insulation paper effects the generation of H2 and CO2 gases, the core metal effects the generation of H2, CO and CO2 gases, and the gasket effects the generation of H2, C2H4, CO and CO2 gases. This research contributes to the use of data resampling method to design ML models with high interpretation accuracy and also contributes to the findings of gassing characteristics for gasket material

Actions (login required)

View Item