www.melanexjou.fora.pl

[loiwtjsf]

Support vector machine based on superconducting fault current limiter pattern recognition


Fruit, followed by linear kernel support vector machine, and finally the neural network perceptron. From Figure 6, Figure 7, Figure 8 of the optimal separating surface of the envelope of the normal data sample space can be seen, RBF classification of the nuclear surface are most close to the normal sample space, that is, the amount for the two features, the algorithm in smaller values the results can be identified, so that rapid and accurate identification of the experimental training results table l Table1ResultofSVMtrainingRBF nuclear: K (xi, xj) = e, a = 6 is indeed the linear characteristics of the nuclear balance between the amount of two However, two features on the amount of money in order to have greater recognition results, the accuracy was not affected but the fast discount, and for the perceptron neural network, and its surface to form one-sided situation classification, mainly reflects the current value of this feature Biomass,[link widoczny dla zalogowanych], fast and good but the accuracy is not good. But can be realized, the neural network the most easily achieved, simply construct two neurons can be achieved. Linear kernel support vector machine is relatively small because the number of support vectors,[link widoczny dla zalogowanych], and for the linear kernel, a simple vector inner product operation, fpga sufficient resources can be achieved, Figure 9 is the use of this method for identification of fault current, when the three-phase short circuit fault current is detected,[link widoczny dla zalogowanych], the use of radial basis function kernel support vector machine. Figure 8, the decision boundary shoreline SVM RBF kernel the optimal classification of the training surface Fig. 8TheoptimizedclassificationsurfacebySVMwithRBFkernel cut into the superconducting current limiter, and three-phase current limit waveform cut into the signal waveform, but the best of the RBF kernel, the computing needs of feedforward nonlinear Gaussian kernel mapping lookup table structure, you need a lot of resources, and the number of multi-vector,[link widoczny dla zalogowanych], therefore, to implement more complex, difficult, resource-constrained. Supplement Ho Yi,[link widoczny dla zalogowanych], et al: Support vector machine based on superconducting fault current limiter pattern recognition 427.4 Results wife. References: After experimental data, this method can be diagnosed 3 short-circuit current, but the results are different, RBF kernel support vector machine is best, followed by the linear kernel, then the perceptron neural network, and in the algorithm on the achievability is just the opposite. Therefore, the algorithm can be realized and the effect of the two contradictory aspects, need to compromise choice. Therefore, limited resources, the linear kernel support vector machine is a comprehensive two best choices. The paper linked to for the design of larger networks prototype, providing a reference design based on the theory and method. For network troubleshooting applications in other areas, also has reference value. But as a new type of power system products, its design and manufacture of many technical difficulties to be overcome, but also requires further research. Figure 9 using a linear kernel support vector machines and three-phase short-circuit current limiter cut waveform Fig. 9Three-phasecurrentwavesandthetriggeredsignalwaveofSCFCLbymethodofSVMwithlinearkernelfunction 【1] KCILINV'KOVALCVI, KMGLOVSeta1. ModelofTHSthree-phasesaturatedcorefaultcurrentlimiter [J]. IEEETransactionsonAppliedsuperconductivity, 2000,10 (1) :836-839. 【2] HAGANMDEMUTHHB, BEALEMH. Neuralnetworkdesign [M]. Beijing: Mechanical Industry Press, 2002. 【3] VAPNIKVN. Thenatureofstatisticallearningtheory 【M]. NewYork: Spfinge ~ veflag, 1995. 【4] BURGESC. Atutorialonsupportvectormachinesforpatternrecognition [J]. DataMiningandKnowledgeDiscovery, 1998,2 (2) 121-167. [5] MOOREA. Supportvectormachine 【EB / OL]. [link widoczny dla zalogowanych] CS. cmu. edu / ~ awm / tutorials, 2001.

相关的主题文章：


[link widoczny dla zalogowanych]

[link widoczny dla zalogowanych]

[link widoczny dla zalogowanych]