【作者】 赵秀粉；

【导师】 王时龙；

【作者基本信息】 重庆大学 ， 机械制造及其自动化， 2003， 硕士

【摘要】 市场需求的变化和竞争的加剧，要求现代制造系统以较快的响应速度、较低的成本生产出高质量的产品以满足顾客的需求，提高企业的竞争力。但是随着现代制造系统内部和外部环境的改变，制造系统变得越来越复杂，对制造系统的控制也变得越来越困难。因此，本论文将人工智能领域中的多代理（Multi-agent）技术应用于机械制造领域，建立基于多代理技术的机械制造系统的动态模型，实现制造系统的分布自治控制，解决传统的集中式控制或递阶式控制的一些固有弊端。本论文在分析代理技术的基础上，把工件和制造设备等映射为一个个的代理，建立了四种机械制造系统的动态模型：地域递阶结构模型、面向对象结构模型、临近关系结构模型和总线结构模型，并详细介绍了模型中各个代理的功能以及系统的运行机制。结合制造系统的实际情况，要完成代理之间的通信，采用无线的方式是最优的选择，而蓝牙技术是一种短距离无线通信的技术规范，因此可以采用蓝牙技术来实现多代理之间的通信。本文在介绍蓝牙技术的基础上，研制出了基于蓝牙技术的无线通信平台，实现数据和消息的无线收发，并建立基于蓝牙通信平台的制造系统控制结构模型。在多代理系统中，代理要适应周围环境的变化并对环境的变化做出反应，自学习能力是很重要的因素之一，本论文在分析强化学习（Reinforcement Learning）的基础上，提出了改进的强化学习算法，即将强化学习和历史经验相结合，既能提高学习的效率和收敛速度，又能满足实时性的要求。本论文对该学习算法进行了详细地分析，指出了该算法的优点和不足。更多还原

【Abstract】 In order to satisfy the various demands of market rapidly and the severity of competition, the manufacturing enterprises are required to not only yield products with high-quality and low-cost, but also adapt the unpredictable changing environment automatically and increase rates of new product introduction. However, with the change of the interior and exterior environment in modern manufacturing system, the manufacturing system is becoming more and more complex. To solve the contradiction, this paper introduces the concept of agent in the area of Artificial Intelligent into manufacturing field and builds up some dynamic modeling of manufacturing system based on multi-agent system. These modeling can fulfill the distributing and autonomy of system and settle some problems of the traditional centralized and hierarchical structure.On the basis of analyzing Multi-agent technology thoroughly, this paper establishes a mapping of work piece and manufacturing equipments to a series of agents, and builds up four dynamic models of manufacturing system: domain based hierarchical structure, topology of cascading agent structure, proximity relation structure and bus-based network structure. This paper also illuminates the function of agents in these models and the mechanism of these models.Considering the actual situation of manufacturing, using wireless communication technology is the best way to accomplish the communication of multi-agent system. Bluetooth is the criterion of short distance of wireless communication. Therefore, we can use Bluetooth to realize the communication of agents. This paper successfully develops the communication plat based on Bluetooth, through which can receive and transmit the datum and information wirelessly and sets up manufacturing system control constitution based on the Bluetooth communication plat.In the multi-agents systems about manufacturing, whether an agent can adapt in the undefined environment and learn from the background knowledge is essential. To improve the adaptation of multi-agent system, we can strengthen the ability of learning from the environment. This paper analyzes the Reinforcement Learning in the field of Artificial Intelligent and puts forward improved reinforcement learning arithmetic that combines Reinforcement Learning with background knowledge, which can speed the Reinforcement Learning and satisfy the demand of real time. This paper detailedly analyzes the arithmetic of learning and finally uses a simple example to illuminate the feasibility of the improved reinforcement learning arithmetic.更多还原