Abstract: An improved generalized genetic algorithm is proposed in this paper. Heterogeneous strategy is proposed to increase the population diversity and to avoid premature while the convergence efficiency is ensured. This algorithm is used for multi-parameter dynamic optimization of the structure which is under random loads and has stress constrains. Numerical examples demonstrated that heterogeneous strategy can improve the probability of convergence to global optimal solution and the improved generalized genetic algorithm can be used to solve complicated structural dynamic optimization problems effectively.