| In recent years, with the rapid development of economy, the scale of power grid expandsunceasingly, and the safe operation of electric power circuit is more and more important. Toidentify the power circuit fault accurately is an extremely important constituent, and plays avery vital role in power supply security. Departure from the basic theory of artificial neuralnetwork, on the basis of the power cable fault, this paper carries out in-depth study in powercable fault system modeling and fault simulation, and practices the classification of cable faultshort-circuit type.Firstly, the thesis introduces the research status and development trend of the power cablefault and the neural network technology, studies the traditional pattern recognition technology,especially on the research and analysis of the K-means clustering method. It not onlycomprehends the K-means clustering application performance and range of applications, butalso identifies the type of cable fault by applying of the K-means clustering method.Secondly, the paper studies the multi-class classification algorithm of artificial neuralnetworks, first of all, First of all, it summarizes the commonly used currently in traditionalneural network classification algorithm, including the error back propagation BP algorithmand the Hebb learning method which is out of the led by any instructor. BP algorithm is basedon the gradient descent method which is a kind of learning methods under the guidance ofteachers. At the same time, The thesis introduces the learning vector quantization (LVQ)neural network algorithm, which is a kind of learning algorithm to train the competitive layerin a condition for teachers. Comparing with the teachers condition of neural networkalgorithm, LVQ neural network algorithm could identifies the required object types moreaccurately. LVQ neural network using the competition rules for the Winner-Taker-All, whichadjusts the weight only for the winning neurons, and is "blocking" for any other neurons, sothis article improves the LVQ neural network in the Winner-Taker-All weight adjustmentmethod. At the same time, on the basis of cable fault identification system, the thesisintroduces the system voltage signal acquisition and regulation circuit design, and dataacquisition card selection and parameter setting, etc. And at the end, by applying theimproved LVQ neural network into the recognition system of cable fault, the paper works outthe identification classification of the cable fault successfully through the simulationexperiment.Finally, the paper puts forward the application the LVQK classifier model, which iscombined by k-means clustering and the improved learning vector quantization neuralnetwork. It takes the use of the improved learning vector quantization neural network onweights adjustment and K-means clustering iterative calculation to adjust the cluster centroidclassifier, and identifies the cable short-circuit four fault types and achieves good results withthe combination of LVQK algorithm. And thesis implements multiple class classification byusing the improved LVQ neural network, so the method improves the randomness oftraditional LVQ network classifier and enhances the accuracy and reliability of object recognition classification. At the end, the method is applied to the power cable faultidentification system built in the laboratory, implements the four kind of fault for power cableof the optimal classification, at the same time improves the efficiency of the classifier. |
