Abstract: NSGA-II is widely used in multi-objective evolutionary optimization for its high convergence and time efficiency.However，the population maintenance based on crowding distance in NSGA-II has not worked well in maintaining the diversity of solution sets.This paper proposes an improved strategy to dynamically maintain diversity by setting a self-adaptive threshold value，and the better diversity individuals have more chances to survive.Comparing new algorithm to NSGA-II and ε-MOEA in five test problems，the results show that improved algorithm efficiently promotes the diversity and achieves efficient convergence at the same time.