This study presents a genetic algorithm integrated with dynamic programming to address the challenges of the hydropower unit commitment problem, which is a nonlinear, nonconvex, and discrete optimization, involving the hourly scheduling of generators in a hydropower system to maximize benefits and meet various constraints. The introduction of a progressive generating discharge allocation enhances the performance of dynamic programming in fitness evaluations, allowing for the fulfillment of various constraints, such as unit start-up times, shutdown/operating durations, and output ranges, thereby reducing complexity and improving the efficiency of the genetic algorithm. The application of the genetic algorithm with dynamic programming and progressive generating discharge allocation at the Manwan Hydropower Plant in Yunnan Province, China, showcases increased flexibility in outflow allocation, reducing spillages by 79%, and expanding high-efficiency zones by 43%.

Large-scale hydro–wind–solar systems require scenario sets that represent cross-resource dependence without losing station-level and intra-day characteristics. A remaining challenge is to connect cluster-level joint modeling with station allocation under sum constraints and hourly reconstruction consistent with daily totals. This study develops an integrated framework: wind and photovoltaic plants are grouped by output correlation; season-specific C-vine copulas model daily wind–solar–runoff dependence; a centered log-ratio transformation and Gaussian R-vine copula allocate cluster totals to stations; and a stratified typical-day library reconstructs hourly profiles. The framework is evaluated using 872 observed days of hourly data from a grid-connected system in Southwest China comprising 53 wind plants, 47 photovoltaic plants, and one hydropower-station runoff series; 1000 joint scenarios are generated. Against independent marginal sampling, Spearman correlation-matrix errors fell from 0.419–0.432 to 0.318–0.395 (8.7–24.2%), and Kendall errors from 0.306–0.325 to 0.236–0.298 (8.3–22.8%). Compared with historical-share resampling, station allocations were more diverse but less faithful to historical share distributions; compared with a unified typical-curve library, intra-day autocorrelation errors decreased across all resource-season cases. The resulting multi-scale scenarios can serve as generation-side inputs for stochastic scheduling and operational risk analysis.