A comparative study of neural network efficiency in power transformers diagnosis using dissolved gas analysis

This paper presents a comparative study of neural network (NN) efficiency f or the detection of incipient faults in power transformers. The NN was trai ned according to five diagnosis criteria commonly used for dissolved gas an alysis (DGA) in transformer insulating oil. These criteria are Doernenburg, modified Rogers, Rogers, EEC and CSUS. Once trained, the neural network wa s tested by using a new set of DGA results. Finally, NN diagnosis results w ere compared with those obtained by inspection and an annalist. The study s hows that NN rate of successful diagnosis is dependant on the criterion und er consideration, with values in the range. of 87-100%.