基于支持向量机的故障诊断问题研究

发布时间：2018-03-25 01:34

本文选题：支持向量机　切入点：故障诊断　出处：《东北石油大学》2011年硕士论文

【摘要】：支持向量机借助于最优化方法解决机器学习中的问题,在解决少样本数据集以及非线性问题上有独特优势,适用于解决抽油机故障这种少样本问题。本文将支持向量机方法应用于油田抽油机故障诊断。具体完成了以下工作: 对智能故障诊断常用方法和局限做了详细分析和研究,并深入探讨了抽油机故障这一少样本问题的常用解决办法。研究了支持向量分类机,就多类分类问题进行探讨,重点分析了“一对一方法”和“一对余方法”。对核函数进行了改进,一是采用自适应核函数,利用样本数据对原核函数进行修正,得到相应的核函数,在一定程度上提高了正确率和分类速度,减少了支持向量的数量;二是提出一种基于改进的Gauss核函数的混合核函数,该核函数兼具单一Gauss核函数和多项式核函数的性能,具备全局性核函数学习能力强和泛化能力强的双重优点。将支持向量机应用于油田抽油机故障诊断,并对诊断结果进行比较,讨论相关参数对诊断结果的影响。
[Abstract]:Support vector machine (SVM) has a unique advantage in solving the problem of machine learning with the help of optimization method, and it has a unique advantage in solving small sample data sets and nonlinear problems. This paper applies the support vector machine (SVM) method to the fault diagnosis of oil field pumping units. In this paper, the common methods and limitations of intelligent fault diagnosis are analyzed and studied in detail, and the common solutions to the problem of pumping unit faults with fewer samples are discussed in depth. The support vector classification machine is studied, and many kinds of classification problems are discussed. The one-to-one method and one-to-one method are analyzed. The kernel function is improved. Firstly, the kernel function is modified by using the sample data, and the corresponding kernel function is obtained. To some extent, the accuracy and classification speed are improved, and the number of support vectors is reduced. Secondly, a hybrid kernel function based on improved Gauss kernel function is proposed, which has the performance of a single Gauss kernel function and a polynomial kernel function. It has the dual advantages of strong learning ability and generalization ability of global kernel function. The support vector machine (SVM) is applied to fault diagnosis of oil field pumping units. The results of diagnosis are compared and the influence of relevant parameters on the diagnosis results is discussed.
【学位授予单位】：东北石油大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TH165.3;TP18

【参考文献】