Abstract: In order to solve the problems of irrational planning and imperfect scheduling of feeder bus routes, a loop feeder bus route optimization method based on improved NSGA-II was proposed. Firstly, combining with the bi-level programming theory, the upper layer model was constructed with the minimization of passenger travel time cost, the maximization of bus enterprise operation revenue and feeder bus service rate as the objective functions and the number of stops of the feeder bus line, the length of the line and the frequency of departure as the constraints. And the lower layer feeder passenger flow distribution model was constructed with the Logit model. Secondly, the initialized population of NSGA-II was optimized using Floyd algorithm, and the model solution flow was designed for the proposed model. Finally, taking Harbin City Subway Line 1 Medical University First Hospital Subway Station as a case study, the proposed multi-objective bi-level programming model and algorithm were applied for solution and compared with the original NSGA-II algorithm and the NSGA-II algorithm based on Logistic chaos mapping. The research results show that the improved NSGA-II algorithm based on Floyd algorithm has a faster convergence speed and better performance in solving multi-objective bi-level programming models. The solution results can obtain multiple mutually non-dominated optimal solutions at the Pareto front. Different solution sets correspond to different objective function values, but all of them can achieve the optimal overall benefit of the connecting bus network, and the solution results are expressed by the compromise optimal solution set.