Considering the problem which the artificial bee colony is inclined to fall into local optimum. Firstly, the onlooker bees′ number is doubled, the roulette wheel selection mechanism is adopted to update with its half to keep evolution along the high concentration of nectar sources; and the roulette wheel selection mechanism of the reverse with others is used to maintain population diversity. Secondly, employed bees is transformed corresponding to all the nectar sources which the times of counter without updating is greater than the threshold value into scout bees, then are these nectar sources updated. Finally, the center position of nectar sources is calculated which are gotten after each iteration, a new solution is produced in the neighborhood that is formed by the center position and every nectar source, then compared thevalues of fitness to choose optimum. Experiments show that the improved colony algorithm not only has better convergence precision and good robustness, but also increases the chance of the algorithm to jump out of local optimum, enhances the ability of search, and has a better optimization performance.