【作者】 王旭；

【导师】 荣冈；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2008， 博士

【摘要】 过程实际测量数据不可避免地带有误差。数据校正是一种利用冗余信息,剔除原始数据中的显著误差,降低随机误差对测量值的影响,并设法估计出未测量变量的技术。生产过程的测量数据是流程工业关于过程状态的重要信息源,因此基于生产过程机理和统计分析方法的数据校正已成为流程工业过程优化的关键技术。随着MES系统在流程工业的推广应用,该技术必然会在更广阔的领域得到应用。本文在综述国内外数据校正研究发展概况后,在分层次统一建模方法研究、多层次数据校正及其工业应用、测量网络设计等方面做了一些工作,主要内容如下:1)针对以往流程工业物料平衡模型建模过程中所存在的问题,以全厂数据校正为背景,提出了企业物料平衡模型分层次建模思想,并对各个层次间的相互约束关系做了数学描述,建立了分层次物流平衡模型的统一建模方法。2)依据全厂物料平衡分层建模理论框架,可以从不同的层次描述数据校正问题,其中精确的物料平衡模型是数据校正技术的基础。实际生产中,生产方案的切换动态改变着物料的流向,使得物料平衡模型发生改变。依据专家经验选择贝叶斯网络关键变量,利用大量的历史数据学习出贝叶斯网络。在应用中,首先通过对关键变量的收率数据进行稳态检验,判断是否发生了生产方案的变化;在发生方案切换时,利用贝叶斯网络的诊断功能实现对生产方案的确定;在此基础上建立准确的物料平衡模型,增强了数据校正的可行性。3)测量数据中所含的测量误差是影响石化生产过程监控效果的重要因素之一。根据企业物料平衡模型分层次建模的统一描述,本文提出了一种全厂两层物料平衡数据校正策略。该方法对不同层次的测量数据分别进行数据校正计算,将某一层次模型中精度较高的测量数据的数据校正计算结果作为约束条件应用于其他层次的数据校正计算,增加了数据校正问题的总体冗余性,有效地去除了测量数据中存在的显著误差。4)分析了某石化企业物流数据的分类与特点,运用全厂两层物料平衡数据校正方法,并给出了针对石化企业实际情况的数据校正算法流程和运行结果分析。通过全厂物流数据的实际应用,证明了该算法在处理带有显著误差的变量和未测量变量估算方面的有效性。5)根据全厂物流分层建模理论框架,装置层提供的数据是其他各个层次建模的基础,因此在装置层将数据校正技术与测量网络的设计和改进相结合能够有效地提高过程数据的可靠性、精确性和完整性,更好的支撑层次化建模中的上层应用。我们提出了一种应用于装置层的传感器测量网络的设计方法。该方法根据图论原理和工业应用的实际需求,综合考虑了可靠性指标和投资约束条件,可以在传感器数量不足以使整个系统都可观测的情况下,提供一个有效的解决方案。该算法能够在一些关键变量必须被测量,并且另一些变量由于技术原因不能直接测量时给出一个测量网络的设计方案。最后在总结全文的基础上,指出了数据校正技术有待深入研究的若干个问题。更多还原

【Abstract】 Measurements are inevitablely influenced by errors. Making full use of redundancy information, data rectification technique eliminates the gross errors in the raw data, reduces the random measurement errors and estimates the unmeasured variables. Reliable process data are the key to efficient operation of chemical plants. Data reconciliation is a technique used to reduce the impact of measurement noise by integrating information from both measurements and process models. With the development and application of Manufacturing Execution System (MES) in process industry, data reconciliation has attracted more attention. After a survey of major issues in data reconciliation, problems on multi-layer modeling, data reconciliation algorithms and industrial application, sensor network design strategy are discussed and developed. The main contributions in this dissertation are listed as follows:1) There are many difficulties in data reconciliation industry applications which is strongly dependent on the models, and a method of measurement network multi-layer modeling is proposed for mass balance of a whole factory. The constraints of mass balance model among different layers are discussed and the mathematic formulation is also modeled.2) According to the multi-layer modeling framework of a whole factory, the data rectification can be defined in different layers. The precise mass balance model is the basis of data rectification. However, the process network changes dynamically due to frequently occurring scheduling events. A new method was proposed to update the mass balance model in view of application. Bayesian network was used, its structure was selected according to expert’s experiences and its variables were trained by historical data. Consequently, scheduling events were identified based on diagnostic function of Bayesian network, and an updated simplified model was finally established. In this way, the feasibility of data rectification was enhanced.3) The error of measurements can lead to significant deterioration in the quality of petrochemical process monitoring. According to the multi-layer modeling framework, a whole factory two-layer data rectification strategy for mass balance is presented. The approach utilizes the high precision measurements’ data rectification result in one layer as constraints to correct other layer’s measured variables and improve the whole system’s redundancy, and this approach can eliminate the gross error more effectively.4) Given the characteristic of measurement data in Sinopec JiaJiang Refinery, the whole factory two-layer data rectification approach is used. The practical data rectification utilization is introduced and the result of data rectification is analyzed. It is proven by the results of measurement data in the Refinery that with the proposed method the gross error measurements are able to effectively be processed and an estimation of unmeasured variables is provided.5) The flow-rate data of unit-layer are the basis of other layers modeling according to the multi-layer modeling framework of a whole factory. In order to obtain the highly reliable and precise data in unit-layer and other layers, sensor network design and retrofit is introduced in unit-layer. A procedure for the design of a sensor network is reported, the strategy uses the graph theory and takes the reliability objective and cost constraint into account. According to the requirements of industrial application, the approach presented in this paper can obtain an efficient solution even though there are not enough sensors to guarantee the whole system’s observability. The proposed algorithm can also provide a solution in such situation where some important variables have to be measured and a portion of them cannot be directly measured technically.In the end, data reconciliation researches are summerized and some problems are presented so as to demonstrate the orientation to deeply study in the next stage.更多还原