任务级行动序列问题中的定性偏好研究  

Research on qualitative preference in planning of task-level Course-of-action

作　　者：王炎娟[1,2] 姚莉[1] 刘斌[1] 

机构地区：[1]国防科技大学信息系统工程重点实验室,湖南长沙410073 [2]北京航天飞行控制中心,北京100094

出　　处：《智能系统学报》2014年第5期551-559,共9页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金资助项目(70971134)

摘　　要：关注了一类典型行动序列,研究如何在动作集合上存在定性偏好,且偏好集合存在不一致性时开展规划。所考虑的行动序列问题称为任务级COA,以抽象层次的动作为基本要素,所考虑的定性偏好包括静态偏好和时序偏好,所讨论的规划目的是获得最大满意度的COA方案。首先建立了偏好与约束的归一化形式描述,在此基础上形成了COA方案设计算法;进一步,使用计算辩论技术排除偏好集合中的不一致性,形成用户接受度最高的COA方案。文中建立的以定性推理为基础的规划框架,实现了偏好解耦,能够适应不同的领域问题,是以定量计算为基础的传统规划算法的有效补充。通过快速响应卫星成像的COA案例,演示了算法的可行性。This paper focuses on a special type of course-of-action. Specifically,performing study on planning with the existence of qualitative preferences and functions on the actions and owns the inner inconsistence. The courseof-action that is taken into consideration is called ‘task-level＇course-of-action（ COA）,with abstracted action as basic element. The qualitative preferences in discussion include static preferences and temporal preference. The objective of planning is a COA plan with satisfaction. Firstly,a unified formulated description is established for constraints and preferences,based on which an algorithm for COA planning is developed. Furthermore,computational argumentation is utilized to exclude inconsistence in the set of preferences,to maximize the user＇s satisfaction for COA planning. The planning framework based on qualitative deduction is an effective add-in for conventional planning scheme based on quantitative computation. The property of preference-decoupling makes itself adaptable to applications in different domain. A case study on scheduling responsive imaging satellites is proposed to demonstrate the effectiveness of the scheme.

关 键 词：行动序列 规划 定性偏好 时序偏好 计算辩论技术 偏好解耦 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]