一种有效的分层加权库编译方法  

Efficient Compilation Approach on Stratified Weighted Bases

作　　者：赖永[1,2] 刘大有[1,2] 

机构地区：[1]吉林大学计算机科学与技术学院,吉林长春130012 [2]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

出　　处：《软件学报》2012年第10期2550-2563,共14页Journal of Software

基　　金：国家自然科学基金(61133011;61170092;60973088;60873149);吉林大学研究生创新基金(20111060)

摘　　要：提议对加权库进行分层,一方面符合人类的思维习惯,另一方面能够提高推理效率.首先说明现有的针对非分层加权库的编译方法也适用于编译分层加权库,但是,由于存在较多冗余信息而效率不高.提出一种新的编译方法,能够在编译过程中去除冗余信息,并提出两种优化技术提高时间效率.该方法与现有方法相同,当软约束权值改变时无需重新编译.选择ROBDD为目标语言,使用随机问题对该方法进行测试.结果表明:对于非分层加权库,该方法的空间效率高于已存在方法;对于分层加权库,该方法的时间和空间效率均高于已存在方法,且当层数越多时,该方法的效率越高.To be in accordance with the thinking habits of human and improve the efficiency of reasoning,this study argues to stratify weighted bases.The study shows that the existing compilation approach to non-stratified weighted bases can also be applied to COMPILE stratified weighted bases;however,its time and space costs are relatively high because of redundant information in the compilation results.The paper proposes a novel compilation approach,which can remove the redundant information in the process of compilation,and presents two optimization techniques to further improve the time efficiency.As with the existing approach,re-compiling a stratified weighted base is not required whenever the weights associated with soft constraints change with time.The approach is tested by compiling random instances into ROBDD-normal bases,and the preliminary experimental results show that the time and space costs of this approach are lower than the existing approach for most instances.

关 键 词：偏好 惩罚逻辑 分层加权库 知识编译 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]