ANN based temperature compensation for variations in polarization and depolarization current measurements in transformer

Abstract

Nowadays, more than a quarter of the faults that occur in oil-filled transformers are due to the aging and deterioration of their oil-paper insulation. Therefore, condition assessment of the oil-paper insulation of oil-filled transformers, which are installed in the power grids, is very important. One of the non-destructive methods, which are currently used to distinguish the insulation condition, is polarization and depolarization current (PDC) technique. Factors such as oil conductivity, paper conductivity, moisture content, temperature and insulation aging alter PDC test results. This paper proposes a method based on artificial neural network (ANN) integrated genetic algorithm to modify the effect of temperature variation in the results of PDC tests. For this purpose, PDC tests are performed on two different transformers at temperatures of 22 °C, 30 °C, 40 °C, 50 °C, 60 °C and 70 °C with the help of Megger-IDAX 300 device, which is capable of performing PDC tests up to 70 °C. To achieve the desired temperatures, the transformers are placed in the thermal furnace and with special thermal cables; the terminals of the devices are connected to the transformers inside the furnace. According to the results of depolarization current, which are measured at different temperatures, the parameters of the insulation model have been estimated by using genetic algorithm (GA). The insulation model parameters corresponding to 22 °C temperature in each transformer are selected as target parameters. Insulation model parameters related to higher temperatures are applied as inputs to the artificial neural network (ANN). This is done in order to train the ANN. The purpose of artificial neural network training is to transmit parameters related to higher temperatures on target parameters. Finally, with the help of the transferred model parameters, the polarization and depolarization current curves are drawn and transferred on the target curves (curves of 22 °C). The transfer error is calculated by the mean square error (MSE) index. Comparison of the proposed model results with the traditional method, which is based on the Arrhenius equation, in correcting the effect of temperature on the PDC test results, confirms the prediction accuracy of the proposed method.