Research On Fuzzy Diagnosis Method For The Power Transformer Insulation Faults

Power transformer is one of the most important electrical equipments in the electrical system,and is also one of the electrical equipment resulting in most electric system accidents. Its operating state affects system's security level directly. It is an important issue for electrical department to find the potential faults of the transformer, to keep it operating safely, and to improve the reliability of power supply. Therefore , it is of great realistic significance to study the fault diagnosis technology of transformer and to increase the operating and maintaining level of transformer.Transformer fault is the result of transformer itself and its synthesized action of application environment and long-term accumulation. So the fault symptoms of transformer are various, and the relation between the fault symptoms and fault mechanics is also complicated, which make it very difficult to establish a universal fault diagnosis method for transformer. Because of the complexity of transformer fault, the diagnosis process must use multiform methods, not a single method. Therefore, we must search for principles, methods and means, which are helpful to fault diagnosis of transformer from all kinds of subjects. So it makes the fault diagnosis technology takes on a characteristic of multi-subject cross.Therefore, in the paper the contents of dissolved gases have been selected as characteristic parameters and all kinds of artificial intelligence technology is adopted to the study of fault diagnosis system for power transformer insulation. Three main parts are included in the paper.This paper establishes a model of the fuzzy system of power transformer fault diagnosis. The fuzzy-neural network module can embody effectively the fuzziness which existing in power transformer fault diagnosis and promote the network's diagnosis precision via its self-learning ability. The paper takes the BP neural network,Elman neural network and Probability neural network, which are adequate for on line monitoring,diagnosis and forecasting the power transformer's running status and faults, for examples to discuss the architecture of network,optimization to network,network's arithmetic and the number of implicit strata nod and carries lots of simulation experiments using the toolbox in matlab.Via some random selection examples, the paper compares several networks' stability which including the BP,Elman and PNN. The result indicates that Elman network can work better than BP network in detecting the transformers'fault type and reach higher precision, but the Elman needs more training times than BP. The PNN can achieve higher accuracy than BP and needs less training times.In software development, the paper takes VB as the development tools and the Access as the database. To simplify the program design, the matlab neural network toolbox is invoked directly in VB. Finally the paper take example to prove the model's validity and veracity.