【作者】 操小明；

【导师】 王波；

【作者基本信息】 重庆大学 ， 计算机应用技术， 2006， 硕士

【摘要】 软测量技术为解决工业控制系统在线检测过程中,一些控制参数无法通过检测仪表进行直接检测或由于检测仪表价格昂贵难以应用等实际难题提供了有效技术手段。基于统计学习理论的支持向量机是一种新型的学习方法,它遵照结构风险最小化原则,同时支持向量机算法是一个凸二次优化问题,能够保证找到的极值解就是全局最优解,从而在统计样本量较少的情况下获得良好的统计规律和更好的泛化能力,为解决小样本、非线性、高维数等学习问题提供了一个框架,帮助解决了许多其他学习方法难以解决的问题。粗糙集理论是一种研究不完整、不确定知识和数据的表达、学习、归纳的理论方法。其在不需额外的先验知识的情况下,通过描绘知识表达中不同属性的重要性,进行知识表达空间约简,去掉冗余信息,简化信息的表达空间维数。由于实际生产中获得的数据往往是有限的,并且一个复杂的过程控制系统中,往往需要大量传感器监测过程参数,传感器的数量直接影响系统的投资。如何减少一些测量不重要参数的传感器,在有限的小样本基础上保证有较高的预测精度,这是本文的重点。文章在分析了目前常用的几种软测量建模方法的基础上,结合支持向量机和粗糙集的各自特点,提出了一种基于粗集-支持向量机的软测量建模方法。利用粗糙集属性约简方法减少属性数;利用支持向量机解决小样本、非线性系统的建模问题。本文沿着“软测量技术粗糙集理论支持向量机理论粗糙集-支持向量机混合方法的提出将粗糙集-支持向量机混合方法应用到软测量中”的思路对作者在课题研究中所取得的成果进行介绍。其中粗糙集-支持向量机混合方法和将该混合方法应用到软测量建模是本论文的重点。在课题研究过程中,除了分析分析说明了该混合方法的特点、使用步骤以及为何将它应用到软测量建模中外,主要使用工具软件ROSETTA、LIBSVM以及MATLAB分别对粗糙集理论属性约简的算法和支持向量机分类、回归建模方法进行了模拟仿真,并以某市污水处理厂的实测数据,针对污水处理过程控制参数及水质参数进行了基于粗集-支持向量机混合方法的软测量建模研究,对污水处理重要参数COD、TN、TP以及SVI等难以在线测量的参数或虽能在线测量但检测费用高的参数进行了测量,将获得的预测结更多还原

【Abstract】 Soft Sensor Technique is an effective approach which is used to solve some actual puzzles, caused by the unavailability of measurers in the controlling process variables measuring or the high price of measurers, in industrial process control systems.Support Vector Machine (SVM) is a novel powerful machine learning method based on statistical learning theory which observes the principle of structural risk minimization (ERM). Furthermore, the SVM algorithm is a quadratic programming problem that promises the extremal solution is the global optimum and thus makes it possible to have good learning and generalization performance under the situation of small-samples. SVM provides a framework to solve the learning problems characterized by small samples, nonlinearity, high dimension and local minima, which are difficult to be dealt with by other learning methods.Rough set theory is a theoretical method which is applied to study the expression, learning and inducing of incomplete and uncertain data. It can reduce the knowledge- expression space, cancel the redundant information via describing the importance of different attribute in knowledge expression without prior knowledge.Since data acquired in practical production are always limited and a great number of sensors are demanded in a complicated process control system, the number of sensors has a direct influence on the investment. The key point of this thesis is how to decrease some sensors testing unimportant variables and insures a high predictive classification accuracy on the bases of limited small-samples.This thesis proposes a soft-sensor modeling method based on RS-SVM after analyzing some common soft-sensor modeling methods and considering the advantage of SVM and RS. refining the number of properties by using the attribute reduction theory of RS and taking advantage of the good generalization performance of SVM to modeling for the Soft-sensor problem of small examples, nonlinear and high dimensions.This dissertation is along the thinking of " Soft Sensor Technique Rough Set Theory Support Vector Machine Theory the combination method of RS and SVM the application of RS-SVM method in Soft-Sensor " to introduce the achievement concluded in the subject study. The most essential part is the last two parts, namely, the combination method of RS and SVM, as well as the application of RS-SVM method in更多还原