An ANFIS Based Fuzzy Synthesis Judgment for Transformer Fault Diagnosis

Abstract

In conventional Fuzzy Synthesis Judgment (FSJ), the shape parameters of fuzzy membership functions are subjectively confirmed by man beforehand, and fuzzy relationship matrix is also established in advance, thus, the applicable object of fuzzy inference is a confirmed system beforehand. However, while the size of data set is larger, and the space distribution of the samples is more complicated, it is difficult to establish such a model to trace the characteristic change of input-output data. But Adaptive Neuro-Fuzzy Inference System (ANFIS) can resolve the above problem better. However, ANFIS is applied to realize FSJ, there two problems required to be improved. One is the innovation of the firing strength of a rule; the other is to change the learning of TSK linear parameters as the learning of fuzzy relationship matrix. In this paper the proposed ANFIS not only resolves the two problems, but also can be like FSJ to implement reasoning, and learn the shape parameters of fuzzy subjection function and fuzzy relation matrix from the given data set, automatically. In the end, transformer fault diagnosis results indicate that the proposed method is effective and ubiquitous, and is an ideal pattern classifier.