Abstract In China, many cascade hydropower reservoirs are asked to simultaneously respond the peak loads of several interconnected power grids based on the signed agreements. However, by far, there are few reports addressing the brand-new engineering problem with huge optimization difficulty caused by multilateral generation contracts, strong hydraulic-electric relationships, load feature differences and spatial-temporal coupled constraints. Here, a three-stage hybrid method is developed to satisfy this practical requirement, where the domain knowledge is firstly used to build a virtual load curve balancing the load features and electricity contracts of multiple power grids; secondly, the dynamic programming is used to determine the scheduling process of the optimized hydroplant, while the linear programming is chosen to allocate the hydropower generation among multiple power grids; finally, the quality of solution is gradually improved via iterative search. The results in two real-world cascade hydropower systems indicate that the hybrid method can achieve satisfactory scheduling results in different cases. Thus, an effective way to reduce the optimization difficulty of the large and complex problem is to break up into a series of simple and independent subproblems to be addressed by existing mature methods.