Optimization Study For QoS-Aware Service Composition Based On Reinforcement Learning

In a service-oriented architecture, a single web service can not meet user increasingly complex demands, which gave birth to the simple combination of existing services to build a complex value-added services meeting users’ demands, namely services composition.With the rapid development of web services technology, more and more services have the same function with different QoS levels resulting in screening services in service composition process. QoS-aware service composition became a hot area of servcie composition with the goal to maximize the users satisfaction. On the other hand, network-based web service have inherent dynamics, which require adaptability for composition process. Meanwhile, the complexity of business processes and the rapid growth of candidate services led to the large-scale problems for service composition.In light of the above considerations, we propose a QoS-aware service composition optimization based on reinforcement learning. It mainly include two aspects, one is a reinforcement learning algorithm optimization based on directed exploration for service composition, which calculates the recency of states and uses the learning experience to guide the exploration process, so as to speed up the algorithm convergence. The other is a reinforcement learning algorithm optimization based on gaussian process for service composition, which is a kernel function approximation techniques, and it can predict the distribution of the objective function value with strong communication skills and generalization ability. Experimental results show that the optimization algorithms we proposed have validity, adaptability and scalability.