The basic genetic algorithm suffers from problems such as precocity and low search efficiency when solving multi-objective optimization problems in large-scale computing environments. Aiming at these problems, this paper introduces various improvement strategies such as neighborhood operation, adaptive strategy, chaos optimization and cooling into the classical genetic algorithm, and designs an improved genetic algorithm process that organically combines various improvement strategies. The improved genetic algorithm and other existing large-scale multi-objective optimization algorithms are tested using LSMOP test problems, and the improved genetic algorithm has better convergence and diversity than other algorithms on both two-objective and three-objective LSMOP test problems. The PF curves of the seven algorithms are plotted separately for the two-objective on LSMOP6 and the three-objective on LSMOP5 when the decision variable is 200, and the images show that the improved genetic algorithm has the most uniform population distribution. The experimental results confirm the effectiveness of the improved genetic algorithm in solving large-scale multi-objective optimization problems.