Artificial neural network based implementation of Oommen's curve

An ANN based implementation of Oommen's curve to study the online estimation of moisture in paper insulation of transformer using temperature and moisture in oil as input is presented. It is based on a multilayer feed forward network (MLP) with one hidden layer in addition to input and output layer. The implementation, analysis, results and applications of the scheme is discussed. The results confirm that the estimated output of the ANN follow the desired output of the Oommen's curve very closely. It is found that the error remains within ±2% of full scale for temperature of 40°C to 100°C. The implementation has the potential to diagnose the incipient fault in real time based on estimation of moisture in paper insulation.

I. Introduction

Transformer is one of the main parts in the power system and hence its health must be carefully monitored. Power transformer reliability is essential for an adequate operation of power systems. The operational reliability of power transformers depends primarily on the transformer oil and paper insulation condition. It's well known, that the presence of moisture has significant influence on ageing of paper insulation and also on the dielectric strength of transformer oil [1]. At constant temperature, moisture equilibrium between oil and paper insulation exists. The change of load conditions causes a change of temperature and as a result there is a migration of moisture between oil and paper insulation. These interactions are very important for the breakdown phenomena in oil-paper insulation. The disturbance of moisture equilibrium results in a significant reduction of the electrical strength of the oil [2]. When the transformer is in equilibrium condition, the moisture content in oil and its temperature provide a quick way of examining the moisture content in paper. Thus by measuring the oil temperature and its moisture content failure of the transformer can be predicted. Oommen's curve which represents the moisture content of paper insulation with the moisture content in transformer oil at different temperature is one the most widely used curves for the manufacturing of different types of transformers. In this paper, the Oommen's curve is implemented by artificial neural network (ANN) to perform online estimation of moisture in paper using temperature and moisture in oil as input. Therefore this ANN based paper moisture estimation scheme can be utilized effectively to predict the failure of transformer caused due to high moisture content in paper insulation.