【作者】 蔡林沁；

【导师】 梅涛； 孙怡宁；

【作者基本信息】 中国科学技术大学 ， 模式识别与智能系统， 2007， 博士

【摘要】 煤矿环境是典型的复杂时变系统，其中人、机、环境等诸多因素相互交错，各种信息表现出多源、异构、非结构化等特点，加之严峻的井下半封闭式空间的作业环境，致使许多传统技术的应用受到严重限制，甚至无能为力。虚拟现实技术的发展为深入认知煤矿生产环境提供了全新的思路和方法。该技术目前在国内外煤矿领域中的研究仍然处在初级阶段，对煤矿虚拟环境的建模和描述方法还缺乏系统、完善的研究成果。本文面向国家煤矿生产安全的实际需要，将煤矿环境与虚拟现实技术有机结合，系统地研究煤矿虚拟环境的建模方法，特别是将智能技术融入虚拟环境中，深入开展基于Agent技术的煤矿智能虚拟环境研究，以探索能有效保障煤矿生产安全的新途径。Agent技术充分体现了人工智能的思想与方法，非常适合于智能虚拟环境的构建。本文基于Agent技术，研究煤矿智能虚拟环境的建模方法及其关键技术，主要内容与创新点有以下几个方面：首先，建立了基于Multi-agent的煤矿智能虚拟环境的体系结构，使得该环境具有很好的灵活性、可重用性和可扩充性。文中论述了主要Agent的概念及其构造的方法，讨论了基于KQML的多Agent通信协议。随后研究了煤矿智能虚拟环境的信息知识模型。该模型基于实体对象进行了分类和组织，通过实体的各种属性来表现虚拟环境中的高层语义知识，有利于更好地支持虚拟环境中多Agent的交互和智能行为的实现。第二，针对传统煤矿VR系统在物理环境模型方面的不足，提出了建立井下合成自然环境的方法及其概念模型，并面向井下安全监控系统，建立了井下大气环境的数据模型，研究了其特征数据表达、环境数据库建立等关键技术，有效地将井下物理自然环境信息融入煤矿智能虚拟环境中，丰富了虚拟环境的知识，增强了其应用价值。第三，系统地研究了煤矿智能虚拟环境中虚拟人的建模方法，构造了具有逼真拟人外形的，拥有感知、运动、行为、认知、内部属性等广泛能力的智能虚拟矿工模型，提出了基于简化包容式结构的行为控制模型，负责基本反应行为的选择、激活与终止等功能，研究了虚拟矿工的视觉感知、路径规划、自主漫游等智能行为，增强了煤矿虚拟环境的智能性、友好性和交互性，这也是传统煤矿VR系统所欠缺的。第四，研究了基于贝叶斯决策理论的瓦斯爆炸事故致因分析方法，通过分析瓦斯事故中人、机、环境及管理因素的交互作用规律，有利于发现爆炸事件中各种隐含因素从不确定状态转变为确定状态的演化过程，对煤矿事故的处理具有重要的指导意义。最后，研发了面向井下安全训练的煤矿智能虚拟环境原型系统。该系统能够集成井下环境模型，实现井下安全检查等行为仿真，具有开放性、可扩展性等特点，充分体现了多Agent智能虚拟环境的优越性。更多还原

【Abstract】 Coal mine is a typical complex system, in which various factors such as man, machine, and environment interact with each other, and abundant information is multi-sourced, isomerous, and unstructured. Furthermore the underground envirnment is austere semi-occluded space. Therefore, many traditional technologies are severely limited so much as to be disabled. The development of Virtual Reality (short for VR) technology brings a fire-new idea and method to go deep into recognizing coal mine producing environment. At present, VR is being phased in coal mine at home and overseas. However, there still is no mature method to model and describe coal mine virtual environment.Oriented underground coal mine safety, this dissertation combines the traditional mine system with VR to systematically investigate the modeling method of mine virtual environment. Especially Intelligence is integrated into virtual environment to investigate Mine Intelligent Virtual Environment (short for MIVE) based on multi-agent so that a new approach can be found to effectively ensure coal mine safety. Agent technology sufficiently reflects the thought of Artificial Intelligence and is amore suitable method for constructing MIVE. Based on Agent technology, the dissertation focuses on the modeling methodology of MIVE and its some key aspects. The main work and innovations are as follows:Firstly, the architecture of MIVE based on multi-agent was proposed so that the virtual environment is reusable and scalable. The concepts and constructing approaches of the main agents in MIVE were described in detail, and the communication protocol using KQML was also discussed. And then the information and knowledge model was constructed. By organizing and classifying the knowledge according to the entities in virtual environment, and storing the high level knowledge by means of various kinds of properties of the entities, the knowlegde model is more effectively in favor of multi-agent interaction and can realize intelligent behaviors more easily. Secondly, to the deficiency of physical environment mode in traditional mine VR, the modeling method of underground Mine Synthetic Natural Environment, namely MSNE, was studied, and the conception model for MSNE was constructed. Oriented the mine safety surveillance and control system, the atmosphere environment data model in MSNE was established using object-oriented method. Some key technologies related to its characteristic data representation and environment database were discussed. Effectively integrated the mine physical natural environment information into the mine virtual environment, the MSNE model can enrich enormously the virtual environment knowledge and enhance its practicabilityThirdly, the modeling method of virtual human in MIVE was studied systematically and an intelligent virtual miner model was constructed. The intelligent miner has lifelike human appearance and possesses abroad ability of perception, motion, behavior, cognition and mental state. And a behavior control model based on the simpled consumption architecture was proposed to select, activate and terminate the basic reactive behavior for virtual miner. And some intelligent behaviors including vision perception, path planning and self-determining ramble also were studied. The intelligence and interaction of virtual environment are just embodied by the intelligent life, especially by the intelligent behaviors of virtual human. This is also the deficiency of the traditional coal mine VR system.Fourthly, a new causal analyzing method based on Bayesian Network was explored for the gas explosion incident. By investigating the interaction of human, machine, environment and management factors in gas explosion incident, the method can analyze the evolving process of hidden factors from uncertainty to certainty, and can guide the coal mine incident analysis and management.Finally, oriented the underground mine safety, a prototype system for multi-agent mine intelligent virtual environment was developed. The system integrated series of mine environment models mentioned above, and can achieve some initial application for underground mine safety inspect, behavior simulation etc.更多还原