【作者】 李笕列；

【导师】 荣冈；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2009， 博士

【摘要】 随着流程工业综合自动化程度的提高和计算机技术的深入发展,利用工厂生产,作业流程中的数据和模型进行决策支持成为许多有识之士的共识。现有的决策方法常常将数据、模型以及求解工具糅合在一起针对某个特定问题进行求解,这样就导致了决策过程具有一次性和抛弃性的特点。本文在对流程企业综合自动化框架和功能进行分析的基础上,对流程企业决策支持中模型和数据管理的框架设计和应用进行了深入的研究,利用关系和子程序相结合的方法建立了流程企业的模型库,并提出了一种图形化的流程企业的模型和数据管理方法,以面向应用为目标对模型和数据管理平台进行了研究和原型开发。在这一主线上,本文的主要研究内容和创新点如下:(1)在流程企业综合自动化功能框架的基础上,提出了通过建立描述模型库和应用模型库对基础模型进行管理的方法,针对流程企业层次化管理的特点提出了图形化的模型管理方法,并开发了相应的模型管理平台,解决了目前流程企业中对模型的快速搭建和多层次应用的需求。(2)从流程企业的数据特性和模型管理的需求出发,提出了一个用于存储层次化数据和支持模型管理的企业数据模型。从概念,逻辑和物理层面上对该数据模型进行了研究,并在流程企业数据管理平台中进行了实现;提出了一种在不同层次间进行数据映射的结构,为流程企业数据管理平台中数据的有效传递和层次模型的关联奠定了基础。(3)提出了一种利用数据管理平台中冗余数据进行数据校正的算法,该算法结合了图论方法和贝叶斯方法,通过分析时间和空间上的冗余度,生成显著误差候选集,对各测量数据的可靠性进行计算;利用优化框架对带显著误差候选集的数据进行校正,有效减少计算中的0,1变量,提高了算法的效率和准确率。(4)充分发挥模型和数据管理平台在建模和模型协作方面的优势,提出了一种调度层仿真的策略;利用模型-模型接口协同操作动稳态仿真模型对炼油流程进行了调度层仿真的应用,解决了对调度仿真效率和准确性的需求。最后在总结全文的基础上,提出了流程企业模型和数据管理中有待深入研究的几个问题。更多还原

【Abstract】 In the last decades, advantages of using models and data to support decision making have became more and more recognizable for process industries due to the improvement of computer technology and the spread of computer integrated manufacturing system. Previous model and data management researches have encountered problems such as model base structure, integration of model and data management system and model representation. After a survey of researches on model management, data rectification, data management and related applications under the framework of enterprise control system integration, we eliminate these problems by showing how to build model bases for process industry and how to design a model and data management system. The main content and major contributions in this dissertation are listed as follows:(1) Previous model base structure and model management approach have encountered problems when they are applied in process industry. After analysis of the current model base structure and model management approach, a new idea of building descriptive model base and analysis model base is proposed. In order to facilitate the modeling process, interfaces between data, models and tools are designed and developed along with the prototype of visualized model management system.(2) Data management system in the process industries is composed of real-time database and relational database. Guided by the standard of enterprise control system integration, conceptual, logical and physical data model for hierarchical process data storage and application is developed with the consideration of future extension. Interfaces between real-time data and relational data are implemented and became the foundation of effective data exchange.(3) Redundant data in the data management system are used in the data rectification. The improved MILP method uses both temporal and spatial redundancies to generate gross error candidates and calculate prior probabilities. By reducing the binary variables in the MILP framework, efficiency and accuracy of the original MILP algorithm is remarkably improved.(4) An effective scheduling simulation strategy for refinery is proposed and implemented on the model and data management system. With the help of model mapping, dynamic-steady simulation and expert rules, this strategy is able to validate optimization methods and generate simulation data in the hierarchical levels. And the using of model and data management system makes the modeling and analyzing process more convenience and adjustable.At the end of this dissertation, promising future researches on model and data management are introduced based on the conclusion of this dissertation.更多还原