Backpropagation and Radial Basis Function Neural Network to Predict Top-Oil Temperature of Distribution Transformer

Abstract

The aim of transformer monitoring is to observe the performance of the transformer in order to do predictive maintenance to prevent transformer’s aging or damage. Damage or aging of isolation is a frequent problem in transformers. One cause of such insulation damage is the temperature rise in the transformer. Monitoring can also determine the remaining life of transformer through hot-spot temperature, which is obtained through top-oil and bottom-oil temperatures approximated by a particular function. Therefore, this research conducted a study on monitoring the temperature of transformer oil (top-oil) based on current, loading, and power factor for modeling using Backpropagation Neural Network (BPNN). For comparison, modeling also used Radial Basis Function Neural Network (RBFNN). The methods obtain prediction which results in transformer oil temperature by conducting training and testing of data verified by measuring top-oil temperature. The results of prediction from different capacities of transformers using both methods are then compared. Performance of the methods is shown by Mean Absolute Percentage Error (MAPE) value.