Journal of Intelligent & Fuzzy Systems 26 (2014) 2577–2590 DOI:10.3233/IFS-130929 IOS Press 2577 Fault prognosis in power transformers using adaptive-network-based fuzzy inference system Jafar Zarei * , Mokhtar Shasadeghi and Abdolrahman Ramezani Faculty of Electrical and Electronic Engineering, Shiraz University of Technology, Shiraz, Iran Abstract. In this paper a new method for estimating dissolved gases of power transformers in the terms of operation time of power transformers using adaptive-network-based fuzzy inference system is presented. Then, transformer condition is evaluated via dissolved gas analysis (DGA) according to IEC 60599 standard. Prominent features of the proposed method are high accuracy, reliability, and high speed convergence in real-world application in spite of the limited numbers of data. The proposed methodology has been evaluated for several real cases while the transformers are operating, and the efficiency of the proposed method is verified. Keywords: Adaptive-network-based fuzzy inference system (ANFIS), power transformers, type-2 fuzzy logic systems, diagnostic reasoning 1. Introduction Power transformers are one of the most important and expensive equipment mainly used in the power system transmission. An acceptable performance monitoring, prediction of failure rate and planning for mending transformers is vital to prevent power outages. There- fore, every year, high cost is spent for servicing, and preventive maintenance of these equipment [1, 2]. When a fault occurs within a transformer, its temper- ature rises and different gases are produced and arisen according to the type and severity of the fault. At this moment, molecular bounds of hydrocarbon oil have been broken. During chemical reactions, different gases are generated and will be dissolved in the oil. Thus, the transformer condition can be evaluated by analyzing the solved gases. * Corresponding author. Jafar Zarei, Faculty of Electrical and Electronic Engineering, Shiraz University of Technology, Shiraz, Iran. E-mail: zarei@sutech.ac.ir. The classical methods of diagnosis are: the Duval’s triangle [3], the key gas method, the Roger ratios [4], IEC method [5], IEEE method, etc. At the present time, one of the mainly used on-line methods of power transformers monitoring is dissolved gas anal- ysis (DGA). Another technique which uses gas ratios analysis according to IEC 60599 standard is evaluation of the condition of the paper-oil system of power trans- formers [6]. The gas ratios analysis is powerful, simple and highly adaptive to use, since the rules in this method can be tuned with practical experience. However, the rules of the DGA method are developed based on sin- gle faults. Therefore, when multiple faults occur, only the major fault is detectable using these rules. On the other hand, the structure of the IEC codes, known as the gas ratios analysis, are quantized to the crisp boundaries of 0, 1. In practice, when simultaneous faults occur the boundaries become fuzzy. Therefore, these codes could lead to errors and sudden changes in detection moving across crisp boundaries from one fault to another. Fuzzy logic provides an approximate but an effec- tive means of describing the behavior of systems that 1064-1246/14/$27.50 © 2014 – IOS Press and the authors. All rights reserved