Abstract: Tasks in the workflow of an application are generally considered to be independent of each other in current web service selection based on QoS. In practice， however， state information is often shared among some tasks in the workflow， which adds binding constraints between tasks and web services， resulting in higher time complexity and lower selection efficiency. Aiming at drawbacks of the existing methods， a genetic algorithm for stateful service selection was proposed. In the proposed algorithm， genetic operations including crossover and mutation were redefined in order to make all individuals meet state-correlate binding constraints among tasks. In addition， to prevent premature convergence， penalty function was introduced into individual evaluation strategy; moreover， similarity judgment between individuals was also included in the algorithm.The experiments results showed that with regards to stateful service selection， good solution and fast convergence rate can be obtained using the proposed algorithm; furthermore， the proposed algorithm is more efficient than the existing algorithms.

Key words: service selection, QoS(quality of service), stateful service, state-correlate binding constraints, genetic algorithm