Artificial Intelligence Based Technique for the Classification of Incipient Faults in Power Transformer

Abstract

Power transformer has been identified as crucial and vital equipment in power system. Any disturbance such as faults will result in immense impact to the whole power system. This paper presents the development of an Evolutionary Programming (EP) – Taguchi Method (TM) – Artificial Neural Network (ANN) based technique for the classification of incipient faults in power transformer using Dissolved Gas Analysis (DGA) method based on historical industrial data. It involved the development of ANN model and embedding TM and EP as the optimization techniques in order to enhance the system accuracy and efficiency. ANN is a powerful computational technique that mimics how human brain process information. It has great ability to learn from experiences and examples, hence greatly suitable for classification, pattern recognition and forecasting purposes. In designing the ANN model, there are parameters which need to be chosen wisely. However, there is no systematic ways and guidelines to select the optimal ANN parameters. It is greatly dependent on the design knowledge and experiences of the experts. The process of finding suitable parameters is become difficult, tedious and time consuming, thus optimization technique is needed to overcome the shortcoming. In this study, TM and EP were employed as the optimization techniques to improve the ANN-based model. The findings obtained from the proposed technique have proved the effectiveness of both TM and EP in optimizing the ANN model. As a result, a reliable EP-TM-ANN based system has been successfully developed that can classify incipient faults in power transformer.