Research On Support Vector Machine Based On Quantum Particle Swarms Optimization And Its Application

As an important utility theory and algorithm in small sample technique, support vector machine (SVM) is researched widely, because SVM has the simple structure and strong generalization ability. In this dissertation, some problem of SVM are analyzed about its kernels and mixtures of kernels, and Quantum-behaved particle swarm optimization algorithm and its two improved algorithms are used to optimize the parameters of SVM based on mixtures of kernels.The main research in this dissertation can be classed as follows:(1) The basic theory of quantum computation is introduced. The concept and theoretical knowledge of quantum evolutionary (QEA) algorithm are introduced at the same time. QEA and particle swarm optimization algorithm are combined to produce quantum particle swarm optimization (QPSO). The bascule knowledge of QPSO is introduced, and last the behavior of QPSO and PSO are compared.(2) The two types of basic kernel function of SVM are researched. Because SVM basic kernel function has some faults, the RBF and PLOY are combined by some weighted to constitute a hybrid kernel function, and the characteristics of mixed kernel function is analyzed. The SVM based on mixed kernel function (CSVM) and the SVM based basic kernel function are compared.(3) The parameters are very important for the SVM, so the QPSO is used to optimize the parameters of the CSVM, then the reasonable parameters can be chosen, so the performance of SVM can be improved.(4) The QPSO has the disadvantage of algorithm divergence and premature convergence algorithm. For overcoming the algorithm divergence, the QPSO based on convergence factor is proposed, in order to overcome the premature convergence of the algorithm, two different strategies that can increase the diversity of population are put forward to avoid the premature convergence algorithm. The improved QPSOs are used to optimize the CSVM.(5) The QPSO-CSVM is used in the compressive capacity test of the concrete. The result indicates that the two CSVMs based on improved QPSOs have the better function than the SVM.