Reconstructive Algorithm Based on Extension Rule for Solving #SAT Incrementally

DOI：10.13328/j.cnki.jos.004827

 作者 单位 E-mail 贾凤雨 吉林大学计算机科学与技术学院, 吉林 长春 130012符号计算与知识工程教育部重点实验室(吉林大学), 吉林 长春 130012 欧阳丹彤 吉林大学计算机科学与技术学院, 吉林 长春 130012符号计算与知识工程教育部重点实验室(吉林大学), 吉林 长春 130012 张立明 吉林大学计算机科学与技术学院, 吉林 长春 130012符号计算与知识工程教育部重点实验室(吉林大学), 吉林 长春 130012 limingzhang@jlu.edu.cn 刘思光 吉林大学计算机科学与技术学院, 吉林 长春 130012符号计算与知识工程教育部重点实验室(吉林大学), 吉林 长春 130012

#SAT问题是人工智能中的重要问题,在人工智能领域被广泛应用.在对基于扩展规则的模型计数求解方法CER深入研究的基础上,重构CER中使用的计算公式,并对其正确性进行了证明;提出极大项相交集和扩展极大项相交集的概念,并给出根据两者关系重用极大项相交集计算结果的增量求解方法,且对广义互补子句集对应的所有扩展极大项相交集进行剪枝,有效避免了计算所有极大项相交集对应极大项个数时的冗余求解;提出构建记录子句间互补关系的互补表方法,给出重用极大项相交集基础子句集互补结果的增量互补判定方法,较好地避免了判断子句间和各极大项相交集的基础子句集互补关系时的重复计算.实验结果表明:RCER方法易于实现,扩展性强,比CER方法效率更高,尤其是在互补因子较低时,效率提升更为显著.

#SAT problem is an important problem and used widely in the field of artificial intelligence. By the in-depth study of model counting algorithm CER based on extension rule, the calculation formula used in CER is reconstructed and its correctness is proved in this paper. Then, the concepts of maximum term intersection and extended maximum term intersection are put forward. An incremental method for #SAT is also proposed, and the extended maximum term intersections are computed preferentially based on the solution of the maximum term intersection. Pruning is performed on the extended maximum term intersections corresponding to the generalized complementary base clause sets to avoid the redundant calculation. A method is provided to decide if a clause set is generalized complementary or not by creating a complementary table for recording the complementarity of all pairs of clauses. Using this complementary table, the base clause set of extended maximum term intersection can be decided incrementally. Based on this method, the redundant calculation for deciding the complementarity of clauses and clause sets is avoided better. Experimental results show that the presented RCER algorithm runs faster than CER algorithm, especially in the instances where complementary factor is small.
HTML  下载PDF全文  查看/发表评论  下载PDF阅读器