Multi-objective path planning of large-scale grid maps has large nodes and a large number of targets, and it’s difficult for existing algorithms to balance the speed and quality of getting Pareto front. Therefore, it has certain theoretical significance to study efficient optimization algorithms based on Pareto front. Firstly, the weighted graph modeling method based on cost vectors is proposed, and optimization algorithms for solving large-scale problems are studied accordingly, which significantly saves time and costs compared to graph search algorithms. Secondly, in response to the problem of low quality based on Pareto front solutions, an improved multi-objective evolutionary algorithm is proposed, which includes a new initialization strategy. With the ideas of angle and shift-based density, individual and environment selection strategies are designed. The improvement measures comprehensively consider population diversity and convergence, improving the solving efficiency. Finally, the effectiveness of the improved algorithm was verified through simulation experiments.