改进最小二乘支持向量机算法及其在故障诊断中的应用

Improved least squares support vector machine algorithm and its application in fault diagnosis

针对人为因素对故障诊断的干扰,提出了一种基于二叉树结构的遗传算法改进可变惩罚因子的最小二乘支持向量分类机(BTGAVPF-LSSVCM)故障诊断方法。首先为减少支持向量机惩罚因子选取受研究人员经验的影响,建立可变惩罚因子的支持向量分类机(VPF-SVCM),并证明了算法的对偶问题;其次,针对支持向量机不易求解的问题,利用最小二乘法求解VPF-SVM,提出VPF-LSSVCM算法,并推导其计算公式;然后,利用遗传算法搜索VPFLSSVCM核参数,提出GAVPF-LSSVCM算法;最后,根据故障诊断实际问题,构建二叉树结构的GAVPF-LSSVCM算法。通过数值仿真实验结果表明,相比支持向量机穷举法,所提出的BTGAVPF-LSSVCM算法诊断精度提高了近14.3%。

To minimize the interference of man-made factors on fauh diagnosis, BTGAVPF-LSSVM fault diagnosis method was proposed. First, in view of that the penalty factor selection of Support Vector Machine （SVM） was established, and the dual problem of the algorithm was proven. Secondly, because of the SVM is not easy to solve, using the least squares solution of VPF-SVCM, VPF-I_SSVCM algorithm was put forward. Then, by using the genetic algorithm to search the parameters of VPF-LSSVCM, GAVPF-LSSVCM algorithm was put forward. Finally, for the practical problems of fault diagnosis, GAVPF- LSSVM algorithm of binary tree structure was built. Numerical simulation experiment results show that compared with the original SVM algorithm, the precision of the proposed BTG.AVPF-LSSVCM algorithm diagnosis is increased by nearly 14.3%.