基于Vague-Sigmoid核函数的PSVM故障诊断算法研究

摘要

Support vector machine (SVM) is widely applied in fault diagnosis field because it has many virtues than the traditional methods. But the standard SVM consumes more system memory and training time, proximal support vector machines ( PSVM) has the characters of faster training speed and less system memory needed, it is suitable for the fault diagnosis of large scale of samples. But the classification accuracy is not very good to the samples nearby the separating plane. In order to resolve this problem of PSVM, Vague-Sigmoid kernel function is utilized to improve the diagnosis accuracy of PSVM to the samples nearby the separating plane in this paper. The simulation is done to testify that this method is effective.%支持向量机因其相比于传统算法具有良好的分类性能,而广泛地应用于故障诊断研究中.但标准SVM存在训练时间长,占用内存大的不足.近似支持向量机(Proximal Support Vector Machines,PSVM)算法具有训练速度快占用内存少的特点,特别适用于大量数据的故障诊断.但其对于分类超平面附近点的诊断精度略显不足.针对此类问题文中将耗时较少的VagueSigmoid核函数应用于PSVM,用以提高其对于在分类面附近样本的分类精度,仿真证明获得了较好的效果.