In the actual flexible job shop scheduling problem, not only the work piece requires processing time, but also transferring work piece by using AGV between the various machines also requires time. Therefore, studying flexible job shop scheduling problem with AGV constraint is more practical and significant. Aiming at this problem, the mathematical model of flexible job shop scheduling problem with AGV constraint is established. Then, a multi-segment encoding method is proposed to directly eliminate some genes that have no help for evolution process. The adaptive crossover probability, mutation probability and multi-population evolution mechanism are proposed to enhance fast convergence and improve global optimization results. Finally, simulation example shows the effectiveness and feasibility of the proposed algorithm in this paper.