Complexity And Application Of Padéapproximation First Weight Function Neural Network

The purpose of this paper is deriving the structure of the first kind of Padé weight functionneural network based on the theoretical basis of the first kind of weight function neural network,then deriving the algorithm complexity of the first kind of Padé weight function neural network, andgiving the error analysis. Then verify the theoretical results by experiment, shows algorithmcomplexity of the first kind of Padé weight function neural network algorithm complexity is betterthan traditional neural network.The research of this paper is the first kind of Padé weight function neural network algorithmcomplexity. As a special weight function neural networks, the first kind of Padé weight functionneural network combination the advantages of weight function neural network and the Padéapproximation. This paper introduces steps to comput the first kind of Padé weight function neuralnetwork combined with Newton interpolation and Padé approximation based on the theory of firstweight function neural network. According to the definition of the algorithm complexity, get thealgorithm complexity of first kind of Padé weight function neural network by analysis the operationnumber of the key steps, and verify the results in Matlab.In the paper, deduced the formula of the first kind of Padé weight function neural networkalgorithm complexity, conclusion is the algorithm complexity related of the dimension of sampleinput and output dimensions and the number of sample points. First kind of Padé weight functionneural network algorithm complexity has the quadratic polynomial growth relationship with thenumber of samples while the number of input and output dimension have been set, and has thepolynomial growth relationship with the number of iutput dimension while the number of samplesand number of output dimension have been set, and has the polynomial growth relationship with thenumber of output dimension while the number of samples and number of iutput dimension havebeen set by the result of experiment.The simulation experiment verifys three aspects: the complexity formula of the first kind ofPadé weight function neural network is accurate; the error of the first kind of Padé weight functionneural network is very small; The first kind of Padé weight function neural network has a bettertraining speed than traditional neural network (BP, RBF). So the first kind of Padé weight functionneural network algorithm has obvious advantages because it has a low complexity and small error,takes less training time compared to the traditional network on complex issues. The research part of application of this paper is network intrusion detection, introduing the basicprinciple of the intrusion detection system based on the the first kind of Padé weight function neuralnetwork. Several experimental results show that the intrusion detection system based on the the firstkind of Padé weight function neural network has high efficiency and correct rate.