ISSN 1000-1239 CN 11-1777/TP

• 论文 • 上一篇    下一篇

从WSBPEL程序中学习Web服务的不确定动作模型

饶东宁1,2 蒋志华3 姜云飞2 吴康恒2   

  1. 1(广东工业大学计算机学院 广州 510090) 2(中山大学信息科学与技术学院软件研究所 广州 510275) 3(暨南大学信息科学与技术学院计算机科学与技术系 广州 510632) (rdn2006@163.com)
  • 出版日期: 2010-03-15

Learning Non-Deterministic Action Models for Web Services from WSBPEL Programs

Rao Dongning1,2, Jiang Zhihua3, Jiang Yunfei2, and Wu Kangheng2   

  1. 1(Faculty of Computer, Guangdong University of Technology, Guangzhou 510090) 2(Institute of Software, School of Information Science and Technology, Sun Yat-sen University, Guangzhou 510275) 3(Department of Computer Science, School of Information Science and Technology, Jinan University, Guangzhou 510632)
  • Online: 2010-03-15

摘要: 智能规划是一种比较有前途的Web服务组合(WSC)方法.用规划进行WSC需要Web服务(WS)的动作模型,而让工程师来写它却很困难.考虑到现存WSC解决方案多用Web服务业务流程语言(WSBPEL)手工编写,可从现存方案中提取动作模型.由于WS本身有不确定性,且现存方案中蕴含对WS的语义要求,所以学习的应是体现流程语义且包含条件效果的不确定动作模型.为此,先将WSBPEL程序转成保留流程语义的标签转换系统(LTS);然后将动作模型学习技术扩展到包括条件效果的不确定规划(NDP),并从LTS中学习动作模型.实现了ARMS-WS系统,它可从WSBPEL程序中学习WS的不确定动作模型.

关键词: 人工智能, 人工智能规划, 动作模型, 不确定规划, Web服务组合

Abstract: AI planning is a promising method for Web service composition (WSC). But it encounters knowledge engineering bottleneck for there is no predefined action models for Web services (WS) in planning language under most circumstances. For example, using planning to find WSC solutions needs action models for WS, which is difficult for engineers to write. Considering that most existing WSC solutions are manually written in business process language for Web services (WSBPEL), it is preferred to learn action models from existing WSC solutions. In the mean time, WS may have multiple outcomes. For this non-deterministic nature of WS and the hidden semantic requirements for WS in existing WSC solutions, the corresponding action model for WS should be a non-deterministic action with condition effects and it should embody semantic requirements too. So firstly WSBPEL programs are translated into a labeled transition system (LTS). Then the scope of learning action models is extended into non-deterministic planning (NDP) with condition effects, and non-deterministic action models are learned from LTS. A system called ARMS-WS is implemented, and experimental results show that non-deterministic action models can be learned from WSBPEL programs. This work helps avoid writing planning description for existing Web services, and makes planning tools more applicable for real-world WSC problems.

Key words: AI, AI planning, action model, non-deterministic planning, Web service composition