Estimating transformer oil parameters using artificial neural networks

In this paper the correlation between dielectric strength, the water content and oil CO₂/CO ratio with insulation resistance in oil-filled power transformers is studied using artificial neural networks. This correlation allows and improves the condition assessment of transformer insulation using the Megger test. This is because dielectric strength, water content and CO₂/CO ratio are important parameters for determining the deterioration state of the transformer insulation. The neural network model is built using tests´ data for nineteen power transformers. The data collected is the high voltage, medium voltage, and low voltage to ground insulation resistance, oil breakdown voltage, water content and oil CO₂/CO ratio. The results propose an efficient model with a breakdown voltage, water content, and oil CO₂/CO ratio prediction rates of 95%, 82.8%, and 87.3% respectively.