The Integrated SVM Network Fault Diagnosis Based On MultiBoost

The essence of the network fault diagnosis based on data-driven is a pattern recognitionproblem. The crucial problem of this case is constructing a magnificent classifier in thesystem. SVM (Support Vector Machine) for its excellent non-linear classified capability andperfect generalization performance has been achieved good results in classified methods. Inview of this, SVM algorithm and its improvement have been more widely used in networkfault diagnosis.When utilizing SVM algorithm, we need to set the value of regularization parameter Cwhich required by the soft margin classification and the Gaussian width (σ) of Gaussiankernel function. Generally, the different values of the parameters (C, σ) have a large impacton the performance of classifier model. At an early time, parameters (C, σ) are setted mainlybased on exhaustive empirical or repetitive experiments, but these methods have someindeterminacy and randomness. In this thesis, genetic algorithm is applied to parametersoptimization of SVM, and the standard genetic algorithm has improved as a adaptive geneticalgorithm—HMGA (Hormone Modulation Genetic Algorithm). HMGA can get the best Cand σ.Ensemble learning method can be used to integrate multiple SVMs and improve theaccuracy of classified model. MultiBoost algorithm is one of ensemble learning method whichcan be viewed as combining AdaBoost (Adaptive Boost-AdaBoost) with Bagging (Bootstrapaggregating-Bagging). AdaBoost has a strong capacity for reducing the bias. Bagging canweakly reduce bias but strongly reduce the variance of classified model. MultiBoost algorithmis a combination of AdaBoost and Bagging, which can not only effectively reduces the biasbut the variance is also reduced effectively. Thereby, MultiBoost can reduce the classificationerror availably.To improve the classification accuracy and reduce classification error for network faultdiagnosis system, this thesis propose an HMGA adjusting SVM approach based onMultiBoost—MHSVM (MultiBoost Hormone Modulation Genetic Algorithm SVM). InMHSVM, firstly, the parameters (C, σ) of SVM classifier will be optimized by HMGA.Then, they will be used as base learners of MultiBoost. Finally, some simulation experiments also verify the feasibility and advantages of classification accuracy of this method.