基于支持向量机的控制图模式识别

为了提高控制图模式识别效果，提出混合核函数支持向量机的模式识别方法。在模型构造中采用一对一多类分类支持向量机，并利用遗传算法优化混合核函数支持向量机参数。仿真和应用结果表明，混合核函数支持向量机对各种模式控制图的总体识别率，I型错判均优于单独核函数、概率神经网络和小波概率神经网络，且具有良好的泛化能力，适合生产现场实时在线工序质量控制。

Control chart patterns recognition based on support vector machine

To improve patterns recognition performance of control chart,a pattern-recognition method based on hybrid kernel was presented.In the structure modeling,the one-against-one-algorithm multi-class classification Support Vector Machine(SVM) was applied,and genetic algorithm was used to optimize the parameters of SVM.The simulation results and application show that the performance of SVM with hybrid kernel is superior to single common kernel,probabilistic neural network and Wavelet Probabilistic Neural Network(WPNN) in the aggregate classification rate and type I error.What's more,it has simpler structure and quicker convergence that it could be applied in the on-line process quality control.