In order to pick out the service composition that meets the complex application requirements of users and the high comprehensive performance in the large-scale Web service environment, we propose an improved fireworks algorithm mixing chaotic mechanism and Levy mutation. Firstly, the chaos theory is used to generate the initial fireworks population and avoid uneven dispersion of individuals which can result in repeated local optimization. Then the Levy mutation operator is introduced to the search process to enhance the global search capability of the algorithm and avoid premature convergence. Finally, the elite selection strategy is adopted to reduce the time expenses of the algorithm in the process of selecting next generation fireworks population. Experimental results verify the optimization performance and stability of the algorithm .