基于支持向量机的故障诊断.docVIP

基于支持向量机的故障诊断 摘 要 在化工生产过程中，为了准确检测故障，减少机械的损失和人员的伤亡，提出了支持向量机算法。支持向量机是基于统计学理论的方法，具有较强的逼近能力和泛化能力。但是在最近几年中，一种基于主元分析的过程监控方法已在工业过程中得到应用，主元分析方法通过正常工况下的历史数据建立的统计模型能很好地检测过程的异常变化和故障的发生。本文主要就这两种方法展开运用。在实际生产过程中，一方面，主元分析方法故障诊断能力有限；另一方面，存在着大量的历史数据，既有正常工况下的数据，又有故障数据，如何充分利用各种类别数据，提高故障诊断能力，具有十分重要的意义。 本文首先运用传统支持向量机算法对历史数据进行分类，分类结果表明该方法对于简单的数据比较容易区分，但是在数据复杂，可辨性较低的情况下，效果不明显。然后运用改进了的传统支持向量机算法对历史数据进行分类，即运用主元分析方法提取各数据的主要特征，再利用支持向量机具有的分类优势对过程数据进行在线诊断，从而提高故障诊断能力。 本文对传统支持向量机算法和改进支持向量机算法进行了仿真比较，仿真结果体现了改进支持向量机算法的优越性；改进支持向量机算法提高了传统支持向量机算法分类的正确率。该种方法在实际工程中能够提高系统的诊断性能，减少不必要的损失。 关键词：支持向量机；故障诊断；主元分析方法；田纳西-伊斯曼过程； Fault Diagnosis Based on Support Vector Machine Abstract In order to detect faults accurately, reduce mechanical losses and casualties in the chemical production process, the algorithm of support vector machines was proposed. Based on the statistics theories, support vector machine is a method of approximation ability and generalization ability. Recently, a new method of process monitoring based on principal component analysis is applied in industrial production process. The statistical model built by principal component analysis method using historic data could detect unusual changes and faults happening in the process accurately. This research is on the application of these two methods. In the actual production process, principal component analysis has certain limitations in diagnosing fault. Besides, the vast volume of historical data was collected in both normal and unusual conditions. It is of great importance to make full use of the data to improve the capacity of fault diagnosis. Firstly, this paper classified the historical data by applying the traditional support vector machine algorithm. The results showed that traditional method works well on simple data sets. However, it showed insignificant effects under a complex and low-differentiability condition. In succession, an advanced approach was used to improve the traditional method, which was approached to enhance the ability of f