基于多分类支持向量机的有杆抽油泵故障诊断研究  

Study on the fault diagnosis of rod-pumping unit based on multi-class support vector machines

作　　者：王凯[1] 

机构地区：[1]西安理工大学机械及精密仪器学院,陕西西安710048

出　　处：《西安石油大学学报（自然科学版）》2010年第1期91-95,共5页Journal of Xi’an Shiyou University（Natural Science Edition）

摘　　要：针对传统方法对有杆抽油泵系统进行故障诊断存在的缺陷,提出了基于支持向量机的抽油泵故障诊断方法.为解决高维、非线性分类问题,通过引入核运算技巧,分析非线性软间隔分类学习机具体算法,并得到使分类间隔最大的最优分类超平面方程.提出了特殊情况下的样本数据标准化处理具体算法,采用基于网格搜索的交叉验证法来选择模型参数,避免了参数选择的盲目性和随意性.采用一对多SVM分类器对抽油泵工况进行了多分类仿真试验,并与BP网络、RBF网络、最小距离法等加以比较,试验结果表明一对多SVM分类法理论严谨,方法可行,自适应好,可在线运行,在解决有杆抽油泵故障诊断问题中表现出了良好的性能,有助于提高采油效率,实现远程采油控制智能化,建设数字油田.Because there are many defects in the traditional methods for the fault diagnosis of rod-pumping unit,a rod-pumping unit fault diagnosis method based on Support Vector Machine (SVM) is proposed.To solve the problem of higher dimensional non-linear classification,the specific algorithms of non-linear soft margin classification learning machine are analyzed and the optimal classification hyperplane equation with maximal classification margin is deduced by introducing kernel technique.The concrete algorithm for data preprocessing under special circumstances is presented.To choose optimal the model parameters,the cross validation method based on grid-search is used so to avoid the blindness and arbitrary and casualness in the selection of the parameters.The multi-classification simulation experiment for oil-well pumping operation modes is carried out using one versus the rest SVM.The experiment result is compared with BP-ANN,RBF-ANN,minimum distance method on the same experiment object.It is shown that one versus the rest SVM classification method has the advantages of strict theory,better feasibility,powerful generalization ability,stronger self-adaptation,running on-line etc.It shows good performance in the fault diagnosis of rod-pumping unit.In addition,it is helpful to enhancing oil-production efficiency,realizing the intelligence of remote oil-production control and building digital oilfield.

关 键 词：有杆抽油泵 故障诊断 支持向量机 最优分类超平面 核运算 参数寻优 

分 类 号：TP182[自动化与计算机技术—控制理论与控制工程] TE938[自动化与计算机技术—控制科学与工程]