Abstract:

The reinforcement learning mechanism was applied to the processbased Web service composition and the Web service composition algorithm with continual selfadaptability was proposed herein so as to enable continuous adaptation to the dynamic Web environment. The algorithm integrates the exploitation of past data about the Web service performance with the continual explor
ation of new options according to QoS actual performance and approaches the optimal Web service composition policy corresponding to the process model gradually. Compared with other similar methods, the algorithm can adjust the Web service composition solution so as to adapt to the dynamic Web continually by means of perceiving the change of Web service and its performance on the Web, and  exploiting the acquired data about the past performance of individual services at any runtime. The relation between exploitation and exploration is accounted for by discussing the range in which the entropy takes its values. Two kinds of experiments were performed to verify the algorithm’s adaptability to the static and dynamic environment.

Key words: composition policy, Web environment, process model, service choice;