In high-speed railways, the sunset-departure and sunrise-arrival trains (SDSA-trains) have become the primary choice of many inter-city travelers. However, optimization of SDSA-trains is very challenging due to the consideration of regular maintenance at night in high-speed railways. By considering the disturbances of maintenance activities in real-world, this paper proposes a two-stage stochastic programming model to optimize a robust train scheduling for SDSA-trains. Specifically, in the first stage, we aim to decide the departure time from the origin and the arrival time at destination station for SDSA-trains, while minimizing the total arrival time. In the second stage, we consider the maintenance disturbances to generate a robust train schedule by adjusting the stop patterns, train orders, and the assignment of tracks. Some linearization techniques are used for the nonlinear constraints. Finally, a real-world instance based on the Beijing-Guangzhou high-speed railway corridors is conducted to show the effectiveness of our approach.