【作者】 郭庆强；

【导师】 李歧强；

【作者基本信息】 山东大学 ， 控制理论与控制工程， 2010， 博士

【摘要】 炼油生产企业是流程工业的典型代表,激烈的市场竞争使炼油企业提高利润和减小成本的压力越来越大。生产调度作为炼油企业过程综合自动化系统(Computer Integrated Processing System, CIPS)的中间环节,处于企业运营的核心位置,是企业信息集成和任务集成的核心,是企业在现有条件下提高资源利用率、降低能耗、节约成本、提高效益、增强竞争能力的重要手段。目前的炼油过程调度往往忽略了由于生产活动改变而导致的炼油装置长过渡过程,得到的是粗放化的调度结果。因而,研究炼油过程动态优化和调度,对提高生产调度的质量和企业效益具有重要的意义。炼油生产过程具有连续、平稳、高能耗、装置复杂且有较长的过渡过程等突出特点。当物料切换、装置开停等计划内事件或原料中断、装置故障等计划外随机事件发生时都会引起装置的过渡过程。本文研究了计划内事件引起的装置过渡过程,在调度模型中包含装置过渡过程,从而实现了调度与控制的有机衔接。对计划外随机事件,如何有效结合调度专家的经验,在保证调度方案可行的同时,具有一定的全局最优性,是目前炼油企业动态调度问题中的一个难点,也是本文的另一个研究点。本文的主要内容包括以下几个方面：(1)介绍了炼油企业生产调度在综合自动化系统体系结构中的地位和作用,指出调度与控制的结合点在于稳态优化。概括了炼油过程中各生产环节的生产调度研究现状,并对当前炼油企业生产调度建模和优化的方法进行了比较研究。认为将数学规划、启发式方法和人工智能相结合,在优化性和实时性之间找到一个令人满意的平衡点,是解决动态优化调度的关键问题。并结合炼油过程的特点,指出了炼油厂过程生产调度建模中存在的主要问题。(2)根据对装置过渡过程的分析,给出了面向调度的装置过渡过程模型的一般形式。对模型的特性分析表明,过渡过程模型可实现调度与控制之间的不同时间域转换,能够得到优化控制的设定值,可以得到调度所需要的动态收率,完全能够满足调度对模型的输入输出要求。以催化裂化装置为例,在七集总模型基础上化简得到催化裂化装置的过渡过程模型,并验证了其有效性。通过催化裂化过渡过程模型,得到了部分指导装置优化控制的指令,为实现调度与控制之间的连接打下了基础。(3)炼油装置生产活动的改变常常存在比较长的过渡过程,根据过渡过程的这一特点,采用连续时间建模方式,以最大利润为目标,建立起包含过渡过程的生产调度模型。通过开关变量的设置可以在平稳态时将调度模型中的很多约束简化处理,不会增加模型的复杂性。仿真结果表明,所建模型可以实现更加精细化的调度,有效提高企业经济效益。根据模型的特点,基于对惯性权重和加速系数的改进策略,设计了改进的两阶段PSO算法。仿真实验表明该算法在求解TSP问题时具有较好的稳定性和鲁棒性。(4)针对动态调度中规则获取困难的问题,基于粗集在知识获取方面的独特优势,把变精度粗集应用于生产调度中,描述了在大量历史经验数据中获取调度规则的过程及其应用实例。对调度中存在的多决策属性并存的决策表分解问题,提出了有因果关系决策属性的决策表分解方法。针对粗集应用中的决策表属性约简问题,提出了有约束指导的属性约简算法。对现场调研和根据生产数据得到的炼油厂调度规则进行了归纳和总结,并举例说明了规则在动态调度中的应用。(5)对于流程工业来说,由于生产工艺基本固定,物料流动的先后顺序基本确定,装置也没有频繁的切换。基于流程模拟的思想,结合现代计算机软件技术,将数学规划、启发式方法和人工智能有效结合,设计了动态调度优化系统,在一定程度上解决了石油炼制行业动态调度的建模困难,并在保证调度实时性的同时获得较好的调度优化性。由调度人员先根据经验确定好装置的选择和物料生产顺序,再根据软件内嵌的优化模型进行优化,从而保证调度可行基础上的优化。动态调度优化系统已经编制成软件,在齐鲁石化炼油厂试用并获得好评。更多还原

【Abstract】 Refinery production enterprise is a typical representative of process industry, which meets the growing pressure of improving profits and reducing cost from market competition. As the intermediate links of a Computer Integrated Processing System (CIPS), refinery production scheduling is the core of enterprise information integration and task integration, as well as the core of business operations. Also, scheduling plays an important role in improving resource utilization, reducing energy consumption, saving costs, improving efficiency and enhancing competitiveness for the enterprises in current conditions. At present, the long transition process of refinery unit, results from changes in production, and is often overlooked in refining scheduling, which results in an extensive scheduling result. Therefore, it is important to research on dynamic optimization and scheduling of the refining process to improve the scheduling quality and business benefits.There are many salient features in the refining process, such as continuous, smooth, high energy consumption, installation of complex and long transition process, etc. And transition process will be brought on when the material switches, devices stop or interruption of raw materials, equipment failures and other unplanned occurrence of random events. This paper studies the device transition process caused by the planned events, and achieves organic convergence of scheduling and control through including the device transition process in the scheduling model. Generally speaking, it is difficult to globally optimize scheduling combined the easily experience of experts effectively when unforeseen events happen, which is also a research point in this paper.This article includes the following aspects:(1) It is pointed out that steady-state optimization is the point of combining scheduling with control after introducing the roles of scheduling. Current modeling and optimization methods of refinery production scheduling are compared and their research status is outlined in various refining process. The key issue of dynamic optimal scheduling is to combine mathematical programming, heuristic methods and artificial intelligence together to find a satisfactory balance point between optimization and real-time. The main problems in production scheduling modeling are pointed out after the characteristics of the refining process are analyzed.(2) The general form of the transition process model for scheduling is achieved based on the analysis of the transition process unit. Analysis shows that the transition process model can realize the different time-domain conversion between scheduling and control, optimize the control settings, get the dynamic yield that scheduling requires, and fully be able to meet the scheduling requirements of the model input and output. The transition process model of fluid catalytic cracking unit (FCC) is achieved based on the seven lumped model, whose validity is verified by an example. Part of the optimal control instruction has been obtained through the transition process model of FCC, which lays foundation for the combination between scheduling and control.(3) A long transition process often happens when a production activity is changed for oil refining unit. According to the characteristics of the transition process, the production scheduling model including the transition process is established as the goal of maximum profit, using continuous-time modeling approach. Through the switch variable settings, scheduling model will be simplified owing to a lot of restrictions in a stable state, which will not increase the complexity of the model. The simulation results show that the model can achieve a more refined scheduling and effectively improve the economic efficiency of enterprises. According to the model’s characteristics, an improved two-stage PSO algorithm (TSPSO) is designed based on the improved strategy of inertia weight and acceleration coefficients. Simulation results show that the algorithm has a good stability and robustness ability in solving the TSP problem.(4) As it is difficult to acquire rules in dynamic scheduling, the rough set was applied to production scheduling based on its unique advantages in the knowledge acquisition. The process of acquiring production scheduling rules from a large number of data based on rough set is described and an example is given. A decision-making table decomposition method is proposed for the causal relationship between decision attribute, which solved the co-existence of multi-attribute in scheduling decision table. Applications for the rough set decision table attribute reduction problem, we introduced the reduction algorithm under instruction. Found the scheduling rules in the historical production data solve the problems of rule-based scheduling. Scheduling rules were summed up and concluded which were obtained from on-site research and production data, and illustrates the application of rules in dynamic scheduling.(5) Since the production process is fixed basically, and there are no frequent device switches, the order of material flow is generally fixed in the process industry. The mathematical programming, heuristic methods, and artificial intelligence are effectively integrated to design dynamic scheduling optimization system, based on the thought of process simulation and modern computer software technology. At a certain extent, the system can solve the problem of modeling for dynamic scheduling in oil refining industry, and ensure to obtain better optimization and satisfy the real-time requirement. According to the optimization model embedded in the software, the system ensures optimal scheduling on a viable basis, after determining the device choice and material sequence based on personnel experience. Dynamic scheduling optimization system has been developed into trial software in the Qilu Petrochemical refinery and received high praise.更多还原