【作者】 李志明；

【导师】 孔令富；

【作者基本信息】 燕山大学 ， 计算机应用技术， 2006， 硕士

【摘要】 课题来源于大庆油田测试技术服务分公司,主要任务是开发原油中转站计量系统,并以软测量技术为中心对提高原油含水率的估算准确度进行有关的理论和应用研究。软测量技术为工业过程中难以或不能直接检测的主导变量提供了新的检测与控制手段,对于生产自动化以及产品质量控制具有重要意义,是目前测控领域的重要研究方向之一。软测量模型是软测量技术的核心,所以,进一步进行建模方法的研究具有重要意义。本文采用生命周期法、快速原型法和面向对象法相结合的程序设计方法设计并开发了原油中转站计量系统,提高了中转站的工作效率、计量精度和管理水平,为企业管理向信息化迈进奠定了基础。在此基础上分析了系统含水率估算方法准确度不高的原因,提出了采用软测量技术加以改进的思想。从经验风险最小化及结构风险最小化的角度分析了用于回归的RBFNN及SVM,阐述了它们的数学表达形式、网络结构和主要训练算法。在对RBFNN和SVM进行深入分析、对比的基础上,提出了一种适用于回归估计的SVs-RBFNN软测量建模方法。建立了RBFNN、SVM、SVs-RBFNN等三种原油含水率软测量模型,仿真结果验证了SVs-RBFNN模型与RBFNN模型相比具有较好的泛化性能。三种模型的估算效果均远远优于原有方法,表明了软测量技术对提高原油含水率估算准确度的有效性。本课题处理实际问题的一些思想和方法对信息化改造传统产业、提高工业自动化检测和控制水平具有参考意义。更多还原

【Abstract】 This project is from the Daqing Oil Field Measurement and TestTechnology Service Sub-company, its main task is to develop a crude oilintermediate repeater measurement system, improve estimation accuracy ofwater cut of crude oil by the application of soft-sensing technique and doresearches on the correlative theory and application.Soft-sensing technique provides a new approach to detect and controlprimary variables which are very difficult or impossible to be detected, hasimportant significance for production automation and quality control, and is oneof the most important research directions in the area of process control. Softsensor model is the core of soft-sensing technique, so it is necessary to researchfurther on.In this thesis, a crude oil intermediate repeater measurement system isdeveloped by a hybrid method that combines rapid-prototyping method, lifecycle method and object-oriented method. This system improves the workefficiency, measuring accuracy and management level of the crud oilintermediate repeater, establishes a foundation for greatly increasing theinformation level of the business management. The reason for inaccuracy of theoriginal estimate method is analyzed, and the idea for improving the estimateaccuracy by soft-sensing technique is proposed.The RBF neural network and Support Vector Machine used for regressionfrom the angles on empirical risk minimization and structural risk minimizationare analyzed, their mathematics expression, topology and main trainingalgorithms are expounded. Basing on the analysis and contrast between RBFneural network and Support Vector Machine, a Support Vectors-RBF neuralnetwork modeling method suitable for regression is proposed . Three soft sensormodels on water cut of crude oil is established based on RBFNN, SVM andSVs-RBFNN respectively. The simulation proves that SVs-RBFNN modelingmethod is superior to RBFNN modeling method in respect of generalizationperformance. The estimation effect of three models is greatly superior to theoriginal estimate method’s, proving that soft-sensing technique is effective inimproving estimate accuracy of water cut of crude oil.In addition, the ideas and methods of dealing with actual problems in thisresearch work may supply references to reforming conventional industries byinformation and to improving detection and control level of industrialautomation.更多还原