【作者】 张烈平；

【导师】 张云生；

【作者基本信息】 昆明理工大学 ， 机械设计及理论， 2011， 博士

【摘要】 随着流程工业生产过程向着生产规模大型化、生产装置复杂化和控制任务多目标优化等方向发展,流程工业生产过程呈现出分散化、异构化的趋势,在功能、时间或空间以及相互关系上表现出极强的分布性、层次性、复杂性,导致现有的控制系统不能满足现代流程工业生产过程的控制要求。因此,如何使流程工业生产过程中各控制系统能够独立自主地进行决策控制,又能相互协调配合,共同完成复杂的生产控制任务,成为流程工业过程控制中急需解决的问题。本文以流程工业过程控制的实际控制需求和生产发展的需要为研究背景,将多Agent技术引入流程工业分布式控制领域,提出如何构建流程工业分层分布式控制的多Agent模型这一命题,并就其中的协作机制、任务分配方法和任务执行策略等关键问题进行研究,以实现流程工业生产过程按任务和功能分派的分布式智能控制。本文主要的研究内容如下：第一、在分析多Agent系统自主、联合、分布及智能特征的基础上,设计和构建了一种面向流程工业过程控制的分层分布式多Agent模型。模型顶层是系统管理Agent,执行控制的中间层,由各车间调度控制Agent组成,底层则由各现场Agent组成,并通过OPC接口实现数据和信息的交互。通过引入Agent组的概念,将多个现场Agent和车间调度控制Agent组成功能Agent组,采用Internet中基于域的模式来实现对分布在Ethernet/Intranet上各Agent组的有效管理。给出了各个体Agent的主要结构和功能,选择基于点到点和改进的基于协助Agent的通信模式作为Agent之间的通信机制,设计了二级协调控制工作模式作为系统分布协调控制机制。构建的多Agent模型,做到了组织结构上分层,控制上分布协调,具有很好的集成性、异构性、灵活性和开放性。第二、为解决功能Agent组中各现场Agent分布式控制的控制形式问题,在详细分析Agent协作的基本概念、协作方式和博弈论方法的基础上,将博弈论引入到多现场Agent协作中,构建了基于改进博弈论的多现场Agent协作模型。通过引入衰减因子,对收益函数进行改进,使协作Agent开始更重视对目标的包围,而当协作Agent对目标形成一定的包围态势后,则将与目标之间的距离作为策略选择的依据,从而加快协作任务的完成。以典型的追捕-逃跑问题为例,采用MATLAB软件对基于改进博弈论的多Agent协作机制进行了仿真,仿真结果验证了其在多Agent协作之间的有效性。第三、为解决流程工业分层分布式控制多Agent系统中的车间调度控制Agent的任务分配问题,在对多Agent系统任务调度策略和粒子群算法研究的基础上,提出基于改进粒子群算法的多Agent系统任务调度方法。考虑到基本粒子群算法在实际应用过程中存在容易出现早熟收敛和全局收敛性比较差的不足,将基于惯性权重的粒子群算法应用到流程工业分布式控制多Agent系统的任务调度中。算法采用动态惯性权重,使惯性权值在粒子群算法搜索过程中线性变化,以提高粒子群算法的搜索性能。通过流程工业生产企业具体生产任务的调度仿真,并与蚁群系统算法调度结果比较,验证了给出调度方法的有效性和优越性。第四、为解决流程工业分层分布式控制Agent系统中现场Agent的有效控制问题,在分析模糊PID控制和遗传算法的基础上,提出了将自适应遗传优化模糊PID控制作为现场Agent控制策略的思想。通过选择适应度函数,并设计惩罚功能,以及采用自适应调整遗传算法控制参数的策略,实现遗传算法优化模糊PID控制的比例因子和量化因子。采用Matlab对给出的控制算法进行了仿真,结果表明,给出算法的控制品质有较大的改善和提高,可以作为现场Agent的控制策略,以实现流程工业生产过程的智能优化控制。第五、在上述研究的基础上,以流程工业中洗衣粉生产过程控制为应用背景,针对目前洗衣粉生产过程控制还大多采用集中监控方式,构建了洗衣粉生产过程控制多Agent系统结构。在分析基于JADE的多Agent系统平台及通信机制的基础上,选择遵循FIPA标准的JADE做为系统开发平台,结合Java软件、通用工控软件MCGS、MATLAB软件、S7-PLC可编程控制器和OPC技术,就洗衣粉生产过程控制多Agent系统中Agent之间的通信、任务调度功能、控制系统功能的实现进行了探讨,给出了具体的实现技术。论文以分布式人工智能中的多Agent技术为基础,系统地分析和研究了基于多Agent的流程工业分层分布式控制系统的组织结构、协作机制、调度算法和控制策略等关键问题,并以洗衣粉生产过程控制为应用背景,对流程工业分层分布式控制多Agent系统的实现进行了探讨。论文的研究成果为流程工业分布式控制探讨了理论与技术基础,同时也提供了工程化的设计和实现方法,具有一定的理论意义和实用价值。更多还原

【Abstract】 As the developing direction of industrial production process to large scale, production equipment to complex and controlling task to multi-objective optimization, the process industry presents decentralized and heterogeneous tendency in the production process, and also presents distribution, hierarchy and complexity in function, in time or space as well as in reciprocity. So it is difficult to make the traditional control system to meet the requirements of the modern process industrial control. Therefore, how to keep the control system can make decisions and control independently and coordinate with each other to realize complex process control has become the anxious problem of process control in process industry. For the purpose of this paper, a novel multi-Agent architecture based on the actual control requirements of industrial process control and the needs of the development of production was proposed, and the proposition how to build the hierarchical distributed control multi-Agent system of process industry was also put forward. Research on key issues such as the cooperation mechanism, task scheduling method and task execution strategy was studied, to achieve distributed intelligent control of process industry, which was assigned by task and foundation. The main contents of this paper are as following:Firstly, based on the analysis of independence, union, distribution and intelligent characteristics of the multi-Agent system, a process industry control oriented hierarchical distributed multi-Agent model was constructed and designed. On the top of model is system management Agent and execution grade of model is workshop. Agent, which is consisted of some job-shop scheduling control Agents. The bottom of the model is established by the scene Agent and they realize the interaction of data and information through OPC interface. With the introduction of the concept of Agent group, the domain-based internet model was used to achieve the effective management of the Agent groups, what are distributed on the Ethernet/Intranet and composed of multiple scene Agents and workshop scheduling control Agent. The structure and main function for each individual Agent were designed, a communication structure based on point to point and improved based on Agent-based were adopted to achieve the communication among Agents in this system, and a two coordinated control was taken as the distributed coordinated control mechanism of the whole system. This multi-Agent control model has achieved hierarchical organizational structure, distributed coordination on control, and has a good characteristic of integration, heterogeneity, flexibility and openness.Secondly, to solve the problem of coordination among agents which are belonged to the same Agent group, game theory was introduced to the multi-Agent system for distributed control Agent collaboration. Based on the detailed analysis of basic concepts, collaboration and knowledge of game theory, the collaboration model of multi-Agent was constructed which was based on game theory. Through the improvement of benevolent function by introduction of attenuation factor, the lay siege on the target was paid more attention at beginning of Agents collaboration, and the distance between the targets was taken as the basis of strategy selection after collaboration Agents have maken a certain siege situation on the target, which can speed up the completion of the collaborative task. By using MATLAB, the cooperation mechanism of multi-Agent system was tested through the pursuit-evasion problem. These simulation results indicated that the multi-Agent cooperation mechanism based on improved game theory can be effectively used to cooperation of multi-Agent.Thirdly, in order to solve task scheduling problem of process industry distributed control system and on the basis of task scheduling strategy of multi-Agent system and particle swarm algorithm, task scheduling strategy based on the improved particle swarm algorithm was proposed. Considering the basic particle swarm algorithm has the shortages of premature convergence and the global convergence in actual application, the particle swarm algorithm based on the self-adapting inertia was applied to the task scheduler process of process industry multi-Agent system. Tthrough making linear transform of inertia weights in search process of particle swarm algorithm, dynamic self-adapting inertia was used to improve the performance particle swarm optimization algorithm. Through comparing the simulation results with ant colony system of specific production scheduling in process industry, the effectiveness and superiority of scheduling method were fully substantiated.Fourthly, in order to solve the control strategy of process industry distributed control Agent system, based on analysis of fuzzy PID control and genetic algorithm, the conception was put forward that adaptive genetic algorithm was used to optimize the fuzzy PID parameters, which was taken as control strategy of on-site Agent. By selecting the adaptive function and designing the penalty function, as well as adopting to adjust control parameters of genetic algorithm, the proportionality factor and the quantification factor of fuzzy PID were optimized by genetic algorithm. The control algorithm was validated by MATLAB. Simulation results show that the design algorithm has a better control quality and it can be used as the control strategy to realize intelligent optimization control of process industry process.Fifthly, on the basis of above researches, the detergent manufacturing process control of multi-Agent system structure was constructed. Based on the analysis of component and communication mode of multi-Agent system on JADE, the JADE platform which follows the FIPA standard was chosen as multi-Agent system developing platform. With Java, MATLAB, MCGS, S7-PLC programmable controller and OPC technology, the implement technology of Agent communication, task scheduling method and control system technology for detergent manufacturing were explored, and the specific implement technology were also given.Aiming at the problem of information isolated island in process control system of detergent manufacturing and at the basis of multi-Agent technology of distributed artificial intelligence, the key issues such as organizational structure, coordination mechanism, scheduling algorithm and control strategy, were analyzed and studied systematically in this paper, which was based on distributed control multi-Agent system of process industry. Then, taking detergent manufacturing process control as the application background, the implementation technologies were also analyzed and studied. The researche results can provide the theory and technology function for process industry distributed control, also can provide the engineering design and realization method, and have an important theoretical and practical value.更多还原