ISSN 1000-1239 CN 11-1777/TP

• 人工智能 •

### 求解约束可满足问题的eSTR算法优化

1. 1(符号计算与知识工程教育部重点实验室(吉林大学) 长春 130012); 2(吉林大学计算机科学与技术学院 长春 130012); 3(吉林大学软件学院 长春 130012) (120801104@qq.com)
• 出版日期: 2016-07-01
• 基金资助:
国家自然科学基金项目(61170314,61272208)；吉林省自然科学基金项目(20140101200JC)

### Optimizing eSTR Algorithm for Solving Constraint Satisfaction Problems

Wang Ruiwei1,3, Li Zhanshan1,2,3, Li Hongbo1,2

1. 1(Key Laboratory of Symbolic Computation and Knowledge Engineering (Jilin University), Ministry of Education, Changchun 130012);2(College of Computer Science and Technology, Jilin University, Changchun 130012);3(College of Software, Jilin University, Changchun 130012)
• Online: 2016-07-01

Abstract: Table constraints, i.e., constraints given in extension by listing all allowed (or forbidden) tuples, are very straight forward and easy to understand, which are intensively studied in constraint programming (CP). Because such constraints are presented in many real world applications from areas such as design, databases, configuration and preferences modeling. However, With the growth of number of constraints and number of tuples, the space cost for table constraints and time cost of consistency checking have become key topics. eSTR is an algorithm which extends simple tabular reduction (STR) to higher-order consistency. After in-depth analysis of eSTR algorithm, this paper proposes two kinds of optimized methods for eSTR: PW\+{sup} data structure and minimal constraint scope, and then we prove that the constraint network enforce Pair-Wise Consistency and Generalized Arc-Consistency (PWC+GAC) with minimal constraint scope is equivalent to original constraint scope. At the same time, minimal constraint scope can reduce further space cost of eSTR2 algorithm by deleting columns of the tables in constraints, and PW\+{sup} data structure can reduce the CPU running time by avoiding some unnecessary checking of Pair-Wise-support (PW-support), since tuples in table of the constraint may not lose any PW-supports on the tables of other neighbour constraints. The experimental results show that combining our methods with eSTR2 algorithm can obviously outperform eSTR2 and eSTR2w on many instances of different problems, since it reduces the space cost and CPU running time of eSTR algorithm.