【作者】 牟盛静；

【导师】 褚健； 苏宏业；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2004， 博士

【摘要】 过程建模和过程优化是石油化工过程领域的两个重要而联系密切的研究课题。随着世界经济的全球化发展趋势，石化工业需要采用高新技术以加快发展，对生产过程提出越来越高的操作要求。现代石化工业综合自动化技术中先进控制、软测量、过程优化、调度与管理等都是以模型为基础的，而过程优化技术是现阶段和未来较长一段时期过程工业的发展重点。 论文以石化工业过程实际装置为背景，探讨了石化工业过程建模与优化中若干典型的理论与工程实际问题及其解决策略。论文内容分为两部分，第一部分对包括三个典型的工业过程，PTA氧化反应器、渣油催化裂化反应系统和复合式精馏塔进行分析、建模以及PTA氧化过程的软测量工程实施；第二部分分别提出了基于进化算法解决工业过程中普遍存在的约束优化问题和多目标优化问题的过程优化算法——基于不可行度选择遗传算法和基于邻域和存档操作遗传算法，并利用该算法对工业PTA氧化过程操作进行多目标优化研究。具体内容安排如下： 1．建立了对PTA氧化过程中产品粗TA中主要中间产物4-CBA浓度的数学模型。文章采用了加权移动平均法来处理变量之间的大时滞问题。通过偏最小二乘法的隐变量投影方法和主元分析的Hotelling T平方统计值对工业过程数据进行分类。对过程模型设置了6个装置因数修正实验室机理模型与工业实际过程的偏差，并由几组典型工业数据样本利用改进的Levenberg-Marquardt算法回归得到装置因数。工业现场数据仿真结果说明该模型能很好地预测4-CBA浓度。 2．建立了渣油催化裂化(RFCC)反应过程模型。通过将实际装置中发生裂化反应的提升管和沉降段反应器分别考虑为理想的活塞流反应器和连续搅拌式反应器，建立了简化的渣油催化裂化反应6集总组分的串行和并行反应动力学网络模型。工业现场数据验证说明该模型能很好地预测渣油催化裂化装置产率分布，并适合于在线工业计算应用。 3．提出了复合精馏塔的建模与脱瓶颈研究课题。文章提出一个醋酸溶剂脱水塔的脱瓶颈研究课题。利用稳态机理模型模拟实际操作过程，找出了限制提高塔的操作负荷以提高操作分离效果的的瓶颈位置，进而提出和比较了在不增加额外设备投资前提下解决这一瓶颈的两套方案。 4．介绍了工业PTA氧化过程软测量工程实施技术及应用效果。文章讨论了工业PTA氧化过程软测量软件开发与现场实施的重点和难点。软测量技术利用基于氧化反应机理的过程模型，结合所提出的双重校正专利技术，开发了软测量软件包，并在工业现场PTA氧化过程在线计算上得到了成功的应用，为先进控制的成功实施提供了有力保障。 5．提出了基于不可行度选择遗传算法求解工程中带约束的优化问题。首先摘要对种群中的个体定义了不可行度，并设计了退火不可行度选择操作。算法首先获得一定比例的可行解，随后进行基于不可行度的遗传算法，使搜索逐渐收敛于可行的全局最优解。仿真结果说明算法与以往的各类算法相比，在全局可行最优解的搜索效率上有很好的优势。 6.提出了基于邻域和归档操作的遗传算法处理多个目标函数优化问题，并进一步提出了基于不可行度选择的部域和归档遗传算法，以求解带约束条件的多目标优化问题。实例仿真结果说明两种算法在计算复杂度上和收敛性上都有明显的优势。 7.提出了工业PTA氧化过程中需要同时实现最大化装置产量或处理量和最小化氧化产物(粗TA)中的4一友基苯甲醛(4一CBA)浓度两个相互矛盾的目标函数优化命题，并牙}!用基于部域和归档操作的遗传算法求解和分析了该多目标优化问题。进而通过对实际工业过程操作的考察和分析，本文进一步提出了包括四级操作序列的可扩展的PTA氧化过程多目标优化问题，并进行了求解和比较，认为可扩展的多目标优化研究更适合实际工业操作优化的实现。 最后，在总结了全文工作的基础上，对工业过程建模与过程优化理论研究与工程应用的发展趋势提出了自己的观点，并给出了理论与应用的若干有待深入研究的问题。更多还原

【Abstract】 Process modeling and process optimization are two important and high associated issues in Petrochemical industries. With the development trend of globalization of world economy, the pressure to apply the advanced technologies is increasing. Process model is the basis of integrated control technologies in modern Petrochemical industries that contain advanced process control, soft-sensor, process optimization, process scheduling and management. Furthermore, process optimization is and will be a key technology in process industries now and in the near future.This dissertation takes the practical typical plants in Petrochemical industries as the research background. It focuses on several typical theory and engineering issues in petrochemical process modeling and optimization. The whole dissertation is divided into two parts. The first part comprises industrial PTA oxidation process modeling, residual fluid catalytic cracking process modeling, complex distillation modeling and analysis, the application of PTA oxidation process soft-sensor technology. In the second one, the infeasibility degree based genetic algorithm is proposed to handle constrained optimization problem in engineering cases and the neighborhood and archive based genetic algorithm and its variant are proposed to treat the multi-objective optimization problem. With that, the PTA oxidation process is regarded as a benchmark for the application of the proposed multi-objective optimization genetic algorithm. The detailed content is arranged as follows1. A mathematic model to predict the concentration of 4-carboxy-benzaldhyde (4-CBA) for an industrial Purified Terephthalic Acid (PTA) oxidation unit is developed. The model is based on a mechanism model from the results of bench-scale laboratory experiment and chemical reaction principle, which is structured into two series ideal CSTR models. Six plant factors are designed to correct the deviation between the laboratory model and the industrial practice. For the existing of substantial time delays between process variables and quality variable, the weighted moving average method is applied to make each variable be in same time slice. The analysis of process data by projection on latent variables of Partial Least Square (PLS) and analysis of Hotelling’s T-squared statistic value of Principal Component Analysis (PCA) are gave to discriminate the operating data into normal operating part and load down and load up operating part. At each operating part, the typical data are selected to regress the plant factors. The proposed model predictive result follows the tracks of the observed value quite well. Compared with the empirical Amoco model, the proposed model is regarded as to bemore suitable to be applied to industrial online soft sensor.2. A novel model for residual fluid catalytic cracking process (RFCC) is proposed. It divides the whole reactor into two part: the riser as ideal pipe flow reactor and the sett -ler as ideal CSTR. The model contains six lumps reaction kinetics with serial and parallel network. The comparison by industrial data proves that the model can predict the produ -ctive rate quite well. The proposed model is suitable to online industrial application.3. Simulation of multistage distillation column is often required for its design and operation. A de-bottleneck study for an acetic acid (HAc) dehydration column is studied in this paper. The column is consists of 4 structural packing sections at the top, a sieve tray section with smaller diameter in the middle and a sieve tray section with larger diam -eter at the bottom. By using steady state simulation, the bottleneck of increasing the effic -iency of separation is identified to be the middle sieve tray section with smaller diameter and smaller tray spacing, and the two renovation schemes without any additional invest -ment on equipment are proposed.4. The development and practical operation of an industrial soft-sensor system for a PTA oxidation unit is introduced. The soft-sensor system is based on a process model,更多还原