A two stage stochastic programming model of rolling stock circulation plan based on flexible train composition mode and robust passenger flow control strategy is proposed to satisfy the dynamic, non-equilibrium and stochastic passenger demands. The first stage is the optimization model of the train composition assignment and rolling stock circulation plan, and the objective is to minimize the operating costs of urban metro system. The second stage is the optimization model of the robust passenger flow control strategy on all the stations of the urban rail transit line, which aims at minimizing the total passenger waiting time. The original model is reformulated as a mixed-integer linear programming model through some linearization methods, which can be directly solved by some optimization softwares (e.g., CPLEX/GUROBI). The numeral experiments demonstrate that the flexible train composition mode can reduce the operating costs by 30.2% with only 0.5% increase of the total passenger waiting time, which shows that the flexible train composition mode can reasonably reduce the operating costs while meeting the passenger demands. In addition, the robust passenger flow control strategy can avoid the conservative optimization results, which can be adapted to the stochastic passenger demands in the actual operation environment.