Machine Learning Based Residual Life Assessment for Health Indexing of Distribution Transformers

Distribution Transformers, as defined by the Bureau of Indian Standards, cover relatively smaller transformers with capacity up to 2500 kVA and a maximum voltage rating of 33 kV (phase-phase) in the primary & supplying at 415 V to the end consumers. Since these transformers serve as a direct connection between the Distribution Company (DISCOM) & the consumers, ensuring reliability of operation of these transformers is critical to the customer satisfaction, revenue optimization & branding. However, the huge & scattered population of these assets with relatively lesser per unit cost, makes the conventional maintenance of these assets cost & manpower intensive. In a regulated framework, the need of the hour is cost optimization while ensuring maximum system reliability. Rising unpredictability in climatic conditions have added to the challenges of utilities, as the utilities are seeing a surge in peak demand exceeding their generous forecasts, making it difficult to plan CAPEX investments. The paper focuses on the use of Machine Learning Algorithms for devising a technique to assess the current health of the Distribution transformers by predicting their residual life, thereby, helping utilities in making informed decisions over maintenance, repair, refurbishment, investment, quality assessment, and control.