Abstract: A new optimization model of double-level hierarchical structure has been developed in this study for power-generation scheduling of cascade reservoirs by the principle of large-scale system decomposition-coordination to solve the problem of curse of dimensionality in the standard dynamic programming algorithm. At the level of its system decomposition, multi-thread technology is applied to parallelization of standard dynamic programming that solves an optimization problem for each subsystem of a single reservoir. At the coordination level, different single-reservoir subsystems are coupled and the corresponding information is exchanged iteratively with the level of decomposition to obtain an optimal solution of the system. This model along with the parallelization algorithm was applied to the optimal scheduling of a three-reservoir cascade system in the Lixian River basin. Results show that the method is effective and can improve the efficiency of large system optimization, and thus it would provide theoretical reference for the optimization of complex engineering systems.