With the development of network combination, more and more kinds of service are needed. So how to combine various kinds of service to meet the complex demands is becoming a hot research subject.

The combination of service means the experts in certain field establish a service pack template according to users' flavor. This template defines the function of each service node, execution path and Qos demand, but doesn't correspond to any specified service instance. So selecting a service instance for each service node to form the execution plan of the composite service with optimal QoS under the conditions of satisfying the global QoS constraints is the key issue of service selection.

In order to solve the above problem, researchers of Institute of Acoustics, Chinese Academy of Sciences analyzed the optimization objective and proposed an improved genetic algorithm, which was self-Adaptive Mutation Genetic Algorithm (AMGA) to solve the combinatorial optimization problem.

In the fitness function of the algorithm, the difference between aggregated QoS values of the plan and the constraints is used as a punishment. In the mutation operator, the excellent level of service instance adaptive mutation probability is used to enhance the efficiency of genetically modified and decaying exponential function is used to guarantee the convergence of the algorithm. Experimental results show that AMGA-based service selection strategy obtains better solution than the other existing Generic Algorithm based ones.