高斯径向基核函数参数的GA优化方法

GA based parameter optimization of Gauss kernel function

核Fisher判别法KFDA(Kernel Fisher Discriminant Analysis)在模式分类应用中通常采用高斯径向基函数做核函数,但高斯径向基函数中参数σ的选取对模式分类的效果影响较大。参数σ的选取目前仅凭经验,缺乏自动选取方法。提出采用遗传算法GA(Genetic Algorithm)实现自动优化参数σ使KFDA具有自适应性的方法,用GA优化参数σ所确定的高斯径向基核函数应用于KFDA时,模式分类的可分性测度大。该方法在电机滚动轴承故障分类实验表明优于其他KFDA分类效果。

KFDA（Kemel Fisher Discriminant Analysis） usually adopts the Gauss radial basis function as its kernel function for mode classification ,in which the selection of parameter influences the result greatly. The value of the parameter is normally determined by experience. It is proposed to use GA in the automatic optimization of the parameter to improve the adaptability of KFDA,which makes the classification resolution higher. Its application in fault classification of motor rolling bearing shows its superiority.