On-line Detection And Fault Diagnosis System Of Gas Chromatograph In Transformer Oil Based On Artificial Neural Network

Power transformer, one of the most important devices, has a very significant influence to the security and stability of power system.Under normal situations,transformer of type of the oil immerse can produce a little amount of gases,such as low mark hydrocarbon and carbon dioxide,carbon monoxide etc.These gases can be dissolved in oil.But this development process is very slow.When there are inner nature faults,for example over hot or discharge,the velocityof producing the gases will be proved.The relations between the constituent and content of the gases and the type of the inner fault and the fault's serious degree are very consanguineous. Therefore,the means of analysing the constituent and content of the gases which is dissolved in transfer is an important method to judge the latent faults. In recent years,a great deal of research work has been done by the inside and outside researchers.But except few developed countries such as Japan which have some reports of on-line monitoring equipment.In our country,the standard still stay around the monitor of a single type of gas. How to find the fault reason of the transformer quickly after the fault is a good way to increase work efficiency and lighten the economy losing. Therefore, the study of the transformer fault diagnosis has a very important meaning.In this paper, several common methods of out-line monitoring the gas solved in oil are introduced at first, then, presents the principle and solution of on-line monitoring system with eigenvalue of gas solved in oil. Discussed the mechanism of macromolecule permeation pellicle. Researched the characteristics and its important effect on gas monitoring of SnO2-majored N type semiconductor gas sensitivity elements. Gived the hardware and software design of the microcomputer control system. At the same time, presents that ANN is suitable for transformer fault diagnosis. Four different methods,such as standard BP algorithm,additional mome-ntum method,adapt learn rate method and L-M ruleare are used to train the same Neural net,and a comparation of the different error is made.The possibility and superiorities of the algorithm of improvement are further confirmed.Make sure that the improved algorithm can be carried on training.On this foundation,discussed the measure of the component and consistencyof the mixed gases based on neural net mode identify.And combined an real example,and aimed at standard algorithm's shortages of the restrain speed and the train accuracy,used an improved algirithm to identify the mixed gases,at the same time ,gived the matrix and results of the training.The result expressed that the improved algorithm greatly improved the net's refrain spe-ed and train accuracy.At last,taked the air density as the importation,take the fault's nature as the exportation,give a real example of transfer fault diagnosis.Under different numbers of neural units,combined the study refrainspeed and the diagnosis results of four kinds of diagnosis algorithm.The result expressed that aiming at the same fault ,using neural net diagnosis method is puperior to other judge method.Neural net has a very big potential in transfer fault diagnosis,select fit net modle,continuously renew study sample,can adquire the increasingly perfectResult.