How to rationally arrange maintenance of generators is an important task in the dispatch and operation of hydro-thermal power systems. On a long timescale, the randomness of natural water inflow makes the generator maintenance schedule (GMS) essentially a stochastic optimization problem. The scenario-based method is usually used to describe the randomness, but it is difficult to solve efficiently the high-dimensional optimization problem with this method. This paper establishes a coupled multi-scenario GMS model of hydro-thermal power systems, applies a multi-disciplinary collaborative optimization (MCO) method to decouple the nonanticipative and the coupling constraints on maintenance variables between scenarios. Thus, the dimension of the multi-scenario GMS model is reduced and the MCO-based structure has inherent parallelism. In addition, in the MCO-based system-level optimization problem, an absolute value penalty term is introduced to replace the quadratic penalty term to ensure that the problem is a mixed integer linear programming model. This helps improve computational efficiency. Finally, a simulation calculation on a real provincial hydro-thermal power system is carried out to verify the effectiveness of the model and algorithm proposed. This work is supported by the National Natural Science Foundation of China (No. 52077083).