Power Transformer Fault Diagnosis Based-on Dissolved Gas Analysis

Large-scale power transformer is an important and very expensive device in power system. Its reliability is closely related to the security of power system. Presently, equipment maintenance system has being changed from schedule maintenance to condition maintenance in power system. In order to know the operating state of a power transformer, the information, which can be acquired through monitoring, must be analyzed so that the diagnosis can be made out. In this background, the motive of this work is to give technology support for condition maintenance of power transformers.This paper introduces the theory and methods of power transformer fault diagnosis based on DGA. On the basis of that, a fault diagnosis system, which combines the neural network with the expert system, is developed. In this system, expert system is a multi-expert system, which uses many knowledge modules to diagnose power transformer faults. Neural network is only one of these modules. So, it is a big problem how to make an integrated decision based on the decisions presented by several experts. In order to solve this problem, weighted majority algorithm is introduced.In this paper, simulations are carried out for different neural network structures, different activation transfer functions and different training functions so that the best diagnosis scheme of neural network can be determined. Through simulation, the proper activation transfer function and training functions are found. Also the probabilities of accurate diagnosing by neural network are compared for different quantities of training samples. The result is: the larger the quantity of training samples is, the higher probability is achieved. Through numerous simulations, the different structure performances of neural networks are also compared, and some basic rules are found. The aforementioned works for designing a neural network may be helpful to the construction of similar networks.As for the software development, Visual Basic is used as developing tool, and Microsoft Access is used as database. In order to simplify the program designing, the method of calling Matlab neural network toolbox in VB is used. And also, calling Microsoft Word in VB is used to print out the diagnosis reports, which can satisfy different print requests of the users.