Condition Assessment of Transformers Load Tap Changers Using Support Vector Machine

Summary

One of the widely used tests to diagnose incipient faults in transformer on-load tap changers is dissolved gas in oil analysis. Several standards and methods (i.e. International Electrotechnical Commision (IEC) 60599 and Institute of Electrical and Electronics Engineers (IEEE) C57.104 standards) have been developed to detect faults in transformers but not specifically in oil-type on-load tap changers. The main problem in applying a conventional method for assessing the condition of on-load tap changers is due to the arcing in oil through the normal operation of on-load tap changers which leads to incorrect diagnosis by these methods. An alternative way for this purpose is applying the modified Duval Triangle method to diagnose faults in on-load tap changers. Implementing an asset management tool in order to monitor the condition of assets automatically based on the historical database is a challenging issue for utilities and power companies all around the world. To achieve this, using intelligent algorithms which can learn through a training dataset is one of the best available methods to reduce the uncertainty of conventional methods. There are several intelligent algorithms, e.g., artificial neural network, fuzzy logic and wavelet network, which can be used in the area of condition monitoring of electrical equipment. In this paper, one of the most powerful machine learning methods for classification problems which is called support vector machine is applied to overcome the lack of conventional methods. The algorithm is first trained using dissolved gas in oil analysis test results based on the modified Duval Triangle and then the trained algorithm is used for automatically processing the new dataset to diagnose the condition of on-load tap changers. This algorithm is intended to be used as an ensemble tool for condition monitoring of in-service transformers and assigning a health index which represents the overall condition of the transformer.

Additional informations

Authors

Hama Hiroyuki, Kong Fei