Three-dimensional tactical training Airspace planning is a complex combinatorial optimization problem. it is very important to construct the corresponding model and design the efficient and fast algorithm to improve the efficiency of airspace utilization and training. Aiming at the characteristics of airspace, this paper divides the whole airspace into cubical units. Based on this, the training airspace planning model is constructed, and an improved genetic algorithm is used to find the optimal solution, which excludes a large number of infeasible solutions and improves the convergence rate. The experimental results show that the method is feasible and stable, and has a strong practical value, which can effectively solve the planning problem of 3D tactical training airspace.