Research On Adaptive And Dynamic Web Service Composition Based On AI Planning

As one of the core research field of service-oriented computing, Web service composition can utilize the reuse value of Web service, meet users’ complex requests, and add value to original services. However, with the skyrocketing number of Web services, the phenomenon begins to appear that the quality of services become discrepant and the function turns to homogeneous, which makes the efficiency and quality of service composition declined sharply. Meanwhile, the uncertainty in dynamic environment also reduces the execution success rate of composition results. Therefore, the study of adaptive and dynamic Web service composition that aims to improve the intelligence, reliability and effectiveness of service composition is an crucial research topic with strong theoretical and practical significance.In recent years, along with breakthroughs of ability for describing problems and solving algorithms, research on AI planning has been rapidly developed. Since the similarities in the description model and other issues between service composition and AI planning, it is possible to combine the two domain; on the other hand, the achievements in AI planning such as classic planning and planning in uncertainty also provide a theoretical basis for realizing adaptive dynamic Web service composition through AI planning. Therefore, using AI planning technology to solve variety of encountered difficulties in service composition and improve the adaptive capacity in real-world is a very promising and challenging research field.Based on the above analysis, this thesis uses AI planning as the core technology, regards “core AI planning algorithms—convert/composition model—Web services oriented optimization and improvement” as the research roadmap, and takes dynamic and adaptive Web service composition for research objects. The mainly work in this thesis are as follows:① Analyzed the research status of dynamic Web service composition and adaptive service composition in uncertainty, where AI planning techniques showed a certain superiority; secondly, studied the key technologies and research status of AI planning, and elaborated the research on service composition through planning; finally, confirmed the research roadmap of this thesis.② Proposed a dynamic heuristic planning algorithm based on macros, which guided the searching direction and accelerated the planning process through learning potential domain control knowledge. By means of analyzing the dependency relationship between actions in pre-existing planning experience, extracted discrete actions that can combined as a macro and extended the source of macros, based on which designed corresponding macro learning algorithms; secondly, while utilizing generated macros, studied a dynamic weight measurement maximizing the valued of macros on premise of validity. Consequently achieved an improved dynamic macro-oriented heuristic planning algorithm.③ Proposed a dynamic Web service composition model based on the presented heuristic planning algorithm to realize automated service composition, which could make use of maximum reuse value and improve the composition efficiency. Firstly, analyzed and compared the description model for Web service and AI planning, then studied a converting model between the two; secondly, based on the analysis of reuse value of service composition results, introduced the proposed dynamic macro-oriented heuristic planning into service composition, and studied a further optimization of the planning algorithm according to the specialty of Web service; finally studied and designed the dynamic Web service composition model based on heuristic planning.④ Proposed a Web service composition method based on extended MDP and heuristic Q-learning, which enhanced the adaptive capacity in uncertain environment and increased the execution success rate of composition results. Firstly, analyzed the uncertain factors in Web service environment and studied the MDP which could help to automatically choose the most promising strategy; secondly, extended the classic MDP aiming to handle the execution failure caused by invalid Web service, based on that the adaptive service composition model was presented; finally, in consideration of the enlarged learning space with extended MDP, proposed a heuristic Q-learning algorithm to improve the learning process through learning the substitution relationship between services.