基于改进粒子群算法的核函数参数优化

Optimization of kernel function parameters based on improved particle swarm optimization

针对核主元分析(Kernel Principal Component Analysis, KPCA)的性能极大地受到本身核函数参数影响问题,提出一种新型的聚类KPCA-FDA-IPSO核参数优化方法。该方法结合KPCA特征分析的相关方法,综合考虑样本的类内离散度和类间距离,并通过Fisher判别分析(Fisher Discriminant Analysis, FDA)建立的数学模型,再将传统粒子群优化算法(Particle Swarm Optimization, PSO)改进为具有继承机制的粒子群优化算法(Inheritance PSO,IPSO),并对核函数寻优。通过数据集仿真和应用研究,验证了该方法能有效地优化核函数参数并提高了KPCA的故障诊断性能。

The performance of kernel principal component analysis(KPCA) was greatly influenced by the parameters of its own kernel function,a new KPCA-FDA-IPSO kernel parameter optimization method with series clustering was proposed. In considering sample within the class of discrete degree and the distance between the classes,the method was combined with the related methods of KPCA feature analysis,and the mathematical model was established by fisher discriminant analysis(FDA). Then the traditional particle swarm optimization(PSO) algorithm was improved to the inheritance PSO(IPSO) optimization algorithm,and kernel function of KPCA was optimized by IPSO. Through the simulation and application research,it is proved that the proposed method can effectively optimize the kernel function parameters of KPCA and improve the fault diagnosis performance of KPCA.