Transformer health index prediction using feedforward neural network according to scoring and ranking method

Abstract

This paper presents a transformer health index prediction applying feedforward neural network (FFNN) according to scoring and ranking method. Power transformer is an important asset of a power system where the function is to convert the level of electrical power and transfer it to the consumer. Outage in transmission line that is caused by the power transformer might lead to the interruption of power supply. Transformer asset management is vital to monitor the operation of transformers in the system to prevent failure. The technique in performing asset management of the transformer is health index (HI). Therefore, this paper presents the assessment of transformer HI by applying artificial neural network (ANN). The FFNN training algorithms proposed in this research to predict the transformer HI include Levenberg-Marquardt (LM), quasi-Newton backpropagation (QNBP), and scaled conjugate gradient (SCG). The HI values obtained from these FFNN techniques were compared to the scoring and ranking method to validate the proposed technique. The performance of the proposed ANN was assessed according to the correlation coefficient and mean square error (MSE). According to the findings obtained, the transformer HI can be successfully predicted by applying different training algorithm of ANN. LM, QNBP and SCG proposed in this research could identify whether the transformer condition is very good, good, fair or poor. The ANN proposed in this research also has been verified with the ranking and scoring method where it produces similar identification of the transformer health index. According to the HI, advance action could be initiated whether to perform upgrade, maintenance, replacement, monitoring, repair, and contingency control of the transformer.