东篱科研大数据发现系统（DRDS）

位置：成果数据库 > 期刊 > 期刊详情页

一致性规划中互斥常量合成方法研究

ISSN号：1000-1239
期刊名称：计算机研究与发展
时间：0
页码：1279-1287
分类：TP18[自动化与计算机技术—控制科学与工程;自动化与计算机技术—控制理论与控制工程]
作者机构：[1]吉林大学计算机科学与技术学院,长春130012, [2]长春工业大学工商管理学院,长春130012, [3]符号计算与知识工程教育部重点实验室吉林大学,长春130012
相关基金：国家自然科学基金项目（61133011,61170092,60973088,60873149）
相关项目：结构化机器学习若干关键问题研究

关键词：智能规划, 多值规划任务, 一致性规划, 常量合成, 启发式应用, intelligent planning, multi-valued planning task l conformant planning~ invariantsynthesis~ heuristics application

中文摘要：

为压缩一致性规划的状态空间，并加快一致性规划的求解速度，将常量引入到一致性规划中，定义一致性规划中的常量，形成新的知识表示“多值一致性规划任务”，定义多值一致性规划动作模型，提出一致性规划常量合成方法，给出一致性规划常量合成算法．该方法利用常量的特性在所有初始世界状态和所有实例动作中猜测、验证常量．理论分析和实验结果表明该算法能合成正确的一致性规划常量，生成多值一致性规划任务．为说明一致性规划常量的应用效果，把生成的多值一致性规划任务与规划解重用启发式结合求一致性规划解，并与规划系统CFF进行对比实验．实验结果表明求解质量和效率较高．

英文摘要：

In order to compress the state space and accelerate the speed of eonformant planning, invariants are introduced into conformant planning, lnvariants of conformant planning are defined formally, and new knowledge representation ＂multi-valued conformant planning task＂ is given. The action model of multi-valued conformant planning is defined accordingly. An invariant synthesis method is proposed for conformant planning and a conformant invariant synthesis algorithm is specified. The synthesis method guesses candidates of invariants firstly. Then the synthesis method tests candidates among all initial world states and all actions according to the properties of conformant invariants. Some candidates are given up and some candidates are modified to form new candidates for testing again. Other candidates are proved to be invariants. Theoretical analysis and experimental results show that the algorithm can synthesize correct invariants and produce multi-value conformant planning tasks. For conformant planning, multi-valued conformant task can greatly compress the state space than Boolean codes used by conformant planning. In order to specify the application of the conformant invariants, the heuristics of reusing plan is combined with the multi-valued conformant tasks for solving the conformant tasks. The comparative experiments with planning system CFF are conducted for testing the efficiency and quality of the combination. Experimental results show that this combination is better than planning system CFF in some domains.

同期刊论文项目