This paper presents an algorithm for automatic generation of train dispatches in metro lines using model predictive control (MPC) with receding horizon. Train trajectories are optimized with reduced computational effort according to a moving horizon scheme, allowing transition between periods with large variation of passenger demand. The model is based on linear programming, it considers all operational constraints and give a trade-off solution between operational costs and service quality to passengers. Piecewise-linear functions are used for directly or indirectly modelling of waiting time of passengers at stations, onboard passenger comfort, train trip duration and number of trains in service. The performance of the proposed methodology is illustrated using a metro line similar to North-South line of Sao Paulo underground.