多分类SVM的代价敏感加权故障诊断方法

为了在支持向量机(support vector machine,简称SVM)中合理引入代价敏感机制来降低故障误诊断的代价,提出一种多分类SVM的代价敏感加权故障诊断方法。该方法通过对多分类SVM的硬判决得票矩阵进行代价敏感加权,将故障误诊断的代价作为权重融入SVM的硬判决,并分析硬判决的得票数和得票权重,从而构造出各故障的概率分配,最终实现多分类故障的SVM代价敏感加权诊断及概率输出。实验结果表明,多分类SVM代价敏感加权处理的诊断结果更趋向于高代价故障,所提方法能够有效降低故障误诊断的代价。

Weighted Cost-Sensitive Fault Diagnostics of Multi-classification Support Vector Machine

To reduce the cost of mistaken diagnoses by properly introducing the cost-sensitive mechanism in the support vector machine(SVM), a weighted cost-sensitive fault diagnostic of the multi-classification SVM is proposed. The votes matrix of the hard decision of the multi-classification SVM is weighted in a cost-sensitive way. The cost of the mistaken diagnosis is used as the weight and fused into the hard decision, and the number of votes of the hard decision and its weight are analyzed so that the probability distribution of every fault can be constructed. Finally, the cost-sensitive weight diagnosis and its probability output of the SVM for the multi-classification faults are realized. The experimental results show that the diagnostic results of the cost-sensitive weight for the multi-classification SVM tend to diagnose high-cost faults, and the proposed method can effectively decrease the cost of fault diagnosis.