In order to ensure the rationality of urban rail transit stations location and avoid low sharing rate after construction, a bi-level model was established based on the selected route and candidate stations. The upper level aimed to achieve maximum ridership, and a GWR model was selected for ridership forecast at candidate stations. The lower level aimed to achieve minimum comprehensive cost per passenger, and the urban rail transit comprehensive transportation cost was defined to include both ridership cost and operating cost. Several heuristic algorithms were compared, and the simulated annealing algorithm was selected to solve the bi-level model. Harbin Metro Line 1 was taken as a case. The established model was used to optimize the location of the stations on this route. Results showed that there were 191 553 passengers per day at the optimized stations. Compared with the ridership of 177 010 at the current stations, there is an increase of 14 543 passengers with an increasing rate of 8.2%. The cost of energy consumption of the optimized stations was 15 972 yuan, which is 1 529 yuan less than that of the current stations with a decreasing rate of 8.7%. Therefore, the established bi-level model of urban rail transit station location can guarantee both social and economic benefits.