Research On Agent Based Web Service Composition Technology

Since the function of a single Web service is limited, it is necessary to compose multiple Web services to provide powerful function. Web service composition has been identified as one of the core problems of service oriented computing. The research of Web service composition has made progress in recent years. However, most of the Web service composition approaches are centralized, which suffers from single point of failure and performance bottlenecks. With the increase of available Web services on the Internet and the Web service requesters, distributed Web service composition is inevitable.This paper studies the agent-based Web service composition technology from the aspects including service agent model, distributed Web service composition algorithm, quality optimization of the composite Web service and self-optimization mechanism for service agent. The contributions that we have made in the paper are highlighted as follows:(1) Service agent modelThe service agent model is the integration result of Web service and agent. Each service agent manages a group of related Web services and adaptively provides services for the user. The service agent model uses the operation template to abstract the features of Web service operations with similar functions, which shields the heterogeneities of distributed Web services. Meanwhile, the service agent model uses the plan to specify the process of how to use operation templates to achieve a certain business goal. Therefore, the common users only need to focus on what they want to achieve rather than how to achieve.(2) Distributed Web service composition algorithm based on service agentBased on the service agent model, a distributed Web service composition algorithm is proposed, which formalizes the Web service composition as graph search problem according to the dependence relations among service agents and the minimal cover based search strategy is proposed. The proposed distributed Web service algorithm based on the service agent model has two distinguished features. Firstly, the distributed Web service composition algorithm is based on the distributed decision making of the autonomous service agents, which has better scalability. Secondly, the proposed distributed Web service composition algorithm is effective for its ability to produce the high quality solution at a low cost of communications.(3) An optimization approach for the composite Web service based on the multi-objective genetic algorithmThe optimization algorithm for the composite Web service is based on the extended quality model for Web service. The extended quality model takes account of various aspects and multi representation forms of non-functional attributes of web service, which has strong description ability. In order to satisfy the multifarious quality requirement, the quality optimization of the composite Web service is formalized as the problem of multi-objective optimization with constraints and an approach of multi-objective genetic algorithm is proposed. The proposed multi-objective genetic algorithm for the optimization of the composite Web service can effectively address the fuzziness existing in the Web service quality description and quality requirement. Simulation experiments demonstrate that the proposed multi-objective genetic algorithm can quickly obtain the approximate global optimal solution.(4) The self-optimization mechanism for service agent based on the fuzzy multi-attribute group decision makingA fuzzy multi-attribute group decision making algorithm for service agent self-optimization is proposed for the fuzzy description of the Web service performance. The proposed algorithm uses the fuzzy operations and social choice function to synthesize the service performance evaluation of several decision makers, which supports the improvement of service agent performance step by step.The above four aspects are are investigated deeply in this paper. Finally, a prototype system is implemented, which demonstrates the effectiveness and feasibility of the theory and algorithms proposed in this paper.