A cost effective artificial intelligence based transformer insulation health index

Abstract

Reliable power systems are characterized by their ability to provide the consumer load with minimal service interruption. Such systems comprise of healthy operational transformers that are continuously being monitored for their functionality and operational conditions. Transformer Asset Management (TAM) defines the monitoring activities and responsive actions taken to protect the reliability of power systems against the transformers failure. Failure of transformers primarily occurs due to the aging of the oil-paper insulation system. One of the recently used transformer assessment techniques in the TAM industry is the Health Index (HI). The HI value mainly indicates the condition of the transformer in question based on the insulation strength. A set of insulation condition tests/parameters for transformer oil samples is used to compute the HI value. The combined cost for conducting such tests would be expensive for a vast number of transformer oil samples. The objective of the presented work is to reduce the HI cost by reducing the number of required condition tests. Such an objective is achieved with the use of artificial neural networks (ANN) for HI prediction, and feature-based exhaustive technique for eliminating the least significant tests.