Transformer Fault Diagnosis Based On Probabilistic Neural Networks

Power transformer is the most important and expensive equipments in power system, faults of which will lead to power system accidents. Transformer maintenance in the past mainly relies on regular preventive maintenance. Along with the development of power industry, power transformer in a high-capacity, high voltage direction, the traditional preventive maintenance have been unable to meet current needs in a safe and economical way. It is of great realistic significance to study the fault diagnosis technology of transformer and to increase the operating and maintaining level of transformer.Dissolved Gases Analysis (DGA) is an important mean to diagnose the internal fault of transformer and it offers an important basis to find the general incipient faults in the transformer indirectly. Thus, the IEC three-ratio used abroad recently has been placed in an important role in the preventive tests of power transformer equipment in our country. But there are two shortcomings with it, coding defect and critical value criterion defect. With the advantages of distributed parallel processing, adaptive, self-learning, associative memory, Artificial Neural Network has broken a new path to fault diagnosis for knowledge engineers. Now, BP (Back-Propagation) network are used in majority transformer fault diagnosis system. But because the structural characteristics of BP, its network isn' t convergent and often easily run into local optimum when a larger samples and a higher precision are needed.In the paper transformer fault diagnosis system is established based on probabilistic neural network aiming at the shortcomings of BP. The classification problems are resolved by using linear learning algorithm to replace the former work done by the nonlinear way. The high accuracy of nonlinear algorithm is maintained at the same time. And the corresponding weights of the network are the distribution of the samples. The network does not need to be trained, which can meet the real-time processing requirements. In this paper, a large number of simulations have been done based on the two different networks. Simulations show that the transformer fault diagnosis system based on the probability neural network is better than BP network system in the speed, accuracy and the ability for samples' superaddition. Finally, the friendly interface of the transformer fault diagnosis system with excellent performance based on the probabilistic neural network is developed by JAVA.