基于知识识别的规划方法研究

发布时间：2018-06-07 06:53

本文选题：知识识别 + BDI规划　；参考：《东华大学》2016年博士论文

【摘要】：智能规划指的是某智能体从一特定问题的初始状态出发,寻找达到解决该问题的目标状态的动作序列。它又称为自动规划,是人工智能研究领域中的一个重要分支,同时也是涵盖知识表示与推理、人机交互和认知科学等多领域的交叉学科研究。本文主要围绕智能规划技术提出两种不同规划方法来解决规划实际应用问题。为了解决智能系统开发效率较低、成本高的问题,本文提出了一种基于知识识别的BDI(Belief,Desire,Intention)规划方法,该方法能实现智能系统规划知识复用,提高规划效率;提出了基于Agent编程框架下的规划识别模型,同时引入工作流网表示的行为模型发现、推导行为序列模式;提出了从工作流网转化为BDI Agent能识别的规划结构,即规划体、上下文环境识别方法。本文提出的BDI规划方法是智能规划方法上一种新的尝试,实验结果初步证明了该基于知识识别的BDI规划方法的可行性,并具有一定的现实指导意义。目标识别是一种特殊的规划识别,它是对规划识别的补充和完善。论文提出了一种基于Agent编程框架下的目标识别方法。在规划识别工作基础上,根据是否存在目标库,对BDI目标进行识别。通过实验,分析出对算法性能有重要影响的因素,即工作流网中的扩展节点、选择分支因素与平行分支因素,并给出实验结果。基于知识识别的BDI规划,是一种推导识别面向BDI Agent编程范例的规划,规划的识别过程也是学习BDI Agent程序的过程。为了解决网络系统自私节点合作收敛问题,本文提出了一种基于智能规划的协作策略,它结合博弈论和规划技术分析行为策略进行决策,逐步实现合作收敛这一目标状态。本文实验分析不同静态拓扑环境下,网络节点能促使自私节点逐步实现合作,使网络系统达到稳定状态。文中实验证明了一定条件下,提出的基于智能协作方法性能超出了其他存在的确定性更新策略规划方法,包括IBN,IBS以及WSLS等。论文研究成果为推动智能规划和规划识别的研究提供了新的解决问题的方法和思路。总之,智能规划是一种问题求解技术,基于知识识别的BDI规划方法研究(包括规划识别和目标识别),能够实现BDI规划知识复用,提高智能系统的开发效率;基于智能规划的协作策略,能够指导、调整无线传感网络节点的行为策略,实现网络中自私节点合作收敛这一目标状态,有助于提高网络的性能和效率。
[Abstract]:Intelligent planning refers to the action sequence of an agent from the initial state of a specific problem to find the target state to solve the problem. It is also called automatic planning, which is an important branch in the field of artificial intelligence. It is also an interdisciplinary study covering knowledge representation and reasoning, human-computer interaction and cognitive science. In this paper, two different planning methods are proposed to solve the practical planning problems. In order to solve the problem of low efficiency and high cost of intelligent system development, this paper presents a method of BDI Bel Beliefa design planning based on knowledge recognition, which can realize the reuse of intelligent system planning knowledge and improve the planning efficiency. In this paper, a programming identification model based on Agent programming framework is proposed. At the same time, the behavior model of workflow net representation is found, the behavior sequence pattern is deduced, and the planning structure, which can be recognized by BDI Agent, is proposed. Context environment recognition method. The BDI programming method proposed in this paper is a new attempt in intelligent planning method. The experimental results show that the BDI planning method based on knowledge recognition is feasible and has a certain practical significance. Target recognition is a special kind of planning recognition, which is complementary to and perfect for planning recognition. In this paper, a method of target recognition based on Agent programming framework is proposed. On the basis of planning and recognition, the BDI target is recognized according to the existence of target database. Through experiments, the factors that have important influence on the performance of the algorithm are analyzed, that is, the extended nodes in the workflow network, the branch factors and the parallel branching factors are selected, and the experimental results are given. The BDI programming based on knowledge recognition is a kind of programming that deduces and recognizes BDI Agent programming paradigm, and the process of planning identification is also a process of learning BDI Agent program. In order to solve the problem of cooperative convergence of selfish nodes in network systems, a cooperative strategy based on intelligent planning is proposed in this paper, which combines game theory and planning technology to analyze behavior strategies to make decisions, and realize the goal state of cooperative convergence step by step. In this paper, it is experimentally analyzed that network nodes can promote the selfish nodes to cooperate step by step and make the network system stable under different static topology environment. Experiments show that under certain conditions, the proposed intelligent collaboration method outperforms other existing deterministic updating policy planning methods, including IBS and WSLS. The research results of this paper provide a new method and train of thought for the research of intelligent planning and planning identification. In a word, intelligent planning is a kind of problem solving technology. The research of BDI planning method based on knowledge recognition (including planning identification and target recognition) can realize the reuse of BDI planning knowledge and improve the development efficiency of intelligent system. The cooperative strategy based on intelligent planning can direct and adjust the behavior strategy of wireless sensor network nodes and realize the target state of selfish node cooperation convergence in the network which is helpful to improve the performance and efficiency of the network.
【学位授予单位】：东华大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP18

【参考文献】