The problem of multi-stage weapon collaborative firepower distribution in land battlefield defense is a typical constrained combination optimization problem, which aims to generate a reasonable and effective firepower distribution scheme. In order to get closer to the actual operational, the confrontation game process of both sides has been introduced and a weapon firepower distribution model including the residual value of enemy combat units, combat resource consumption, and battlefield value loss of combat units is established. An improved intelligent algorithm(D- NSGA-GKM) is proposed based on the non-dominated sorting genetic algorithm III(NSGA-III) for multi-stage collaborative weapon firepower distribution. Firstly, a non-dominated sorting algorithm based on the dominance degree matrix is introduced to reduce redundant operations in the sorting process to improve the efficiency of non-dominated sorting. Then, the repair operator is introduced in the genetic operation stage to repair the infeasible solution. Finally, the genetic K-mean clustering algorithm is introduced to cluster the initial reference points automatically, the centroid of the cluster is used to replace the original reference points, and the penalization-based boundary intersection distance is introduced in the environmental selection stage to replace the vertical distance, to improve the convergence of the algorithm. The experimental results show that the D-NSGA-GKM algorithm has excellent time performance and convergence performance on the problem of multi-stage weapon cooperative fire distribution.