This paper is focused on the optimal policy selection for moving target defense. Attack-defense confrontation in moving target defense environment is analyzed. Reward quantization method of moving target defense policy is proposed. Single-stage and multi-stage moving target defense game models are constructed based on the dynamic game with incomplete information. The algorithm to obtain perfect Bayesian equilibrium and the method to revise prior belief are proposed. Optimal moving target defense policies are obtained under different security situations. Finally

not only the feasibility and effectiveness of the proposed model and method are illustrated and verified in a representative example but also general rules of network defense using moving target defense policies are summarized.