Distribution Transformers Failure Prediction using Machine learning

Abstract

Power grids consist of distribution transformers that act as step-down transformers to convert high voltage to low voltage which is required for residential, industrial and commercial use. As there is a significant number of distribution transformers in the distribution network, the status (burned or not) of these transformers is crucial to the reliability of these power grids. The rise of the fourth industry expands the potential of predictive maintenance and opens the road for effective and well-executed maintenance processes. This paper proposes a methodology for predictive maintenance of distribution transformers using actual data collected by Compania Energética de Occidente. As a result of the transformer's status prediction, maintenance or replacement might be required. Hence predictive maintenance helps enchases the reliability of power distribution systems and decrease unwanted costs. This research employs different machine learning (ML) models and compares various scenarios using different data scaling and feature selection methods. The results generated by the most effective machine learning algorithm are utilized to establish various cost scenarios which would aid in making better decision through managerial implications.

Date of Award	Apr 2023
Original language	American English
Supervisor	Maher Maalouf (Supervisor)