Grounding Grids Fault Diagnosis Based On DCA And PCA-BP Neural Network

Power station and substation grounding grids play a key role in the reliableoperation and safety of the staff in power system.However, the grounding grid isburied in the earth, the electric performance parameters will become worse for somespecial factors such as harsh conditions, inevitable corrosion,which will cause theground connect breaked, and directly theat the safety of operation of the power system.So an easy, accurate, undamaged mechanical corrosion test methods is needed torealize the corrosion state detection without black out and non-excavation. At present,intellingence information processing technologies like BP Neural Networks,PrincipleComponent Analysis, Designated Component Analysis and so on, have been a newhotspot and provided an effective way to slove the problem of grounding grids faultdiagnosis.This paper firstly reviews the research of grounding grids fault diagnosis, thenfocuses on the application of electrical network model in grounding grids faultdiagnosis aiming at the problems in accessing to fault sample information. In themeantime, the influencing factors that may arise in the measurement will be study,then the effective steps will be analyzed.On this basis, the basic principle of PCA and BP neural network will beintroduced in fault detection and diagnose,it illuminates that BP neural network hasgreat advantage. But at the same time,due to its restriction, the BP neural networktraining needs a lot of samples and a fairly long time of training, the networkperformance will be worse with the increase of the dimension of the training samples,all of which will make the efficiency of fault diagnose can not be optimal. This papercombines PCA with BP neural network, builds PCA-BP network model, the principalcomponent analysis could realize the linear combination of BP network input, in orderto reduce the dimension of the input vector, improve the sensitivity of the fault,eventually reach the goal of simplifying the network, shorten the training time andhave a good performance in recognition of fault diagnosis. The diagnostic exampleresults in grounding grids show the advantage of the PCA-BP neural network in faultdiagnose is obvious compared with the standard BP neural network,which verifies thevalidity of the method.In this paper,DCA is introduced to apply to grounding grids fault diagnosis forthe problem that conventional PCA can’t identify the fault mode well and multiple fault diagnosis. Observation data is projected to fault subspace spanned by faultpatterns defined in advance under condition that the system is abnormal,we canidentify faults diagnosis according to the significance detection problem forprojection energy and so the DCA overcomes the shortage of PCA compound pattern.