Artificial fish is a hot spot in fish bionics research. Aiming at the low efficiency of artificial intelligence fish path planning in three-dimensional virtual marine environment, an artificial intelligence fish path planning method based on improved A* algorithm is proposed. First and foremost, after each path node is selected in the path search of A* algorithm, the OPEN table is emptied. Then, collision detection algorithm is used to screen out unnecessary path nodes in the optimization path to further optimize the fish swimming path. The results show that compared with the traditional A* algorithm, the improved A* algorithm has the same efficiency, and the path length is reduced by 5%. The research shows that the improved A* algorithm can accurately plan the swimming path of artificial intelligence fish in the three-dimensional virtual ocean space, improve the authenticity of artificial intelligence fish simulation and provide a new idea for the simulation of fish in fishery.