A stochastic optimization framework for short-term peak shaving in hydro-wind-solar hybrid renewable energy systems under source-load dual uncertainties

The rapid growth of global electricity demand, driven by industrialization and urbanization, poses significant challenges to power system operators, particularly the increasing disparity between peak and off-peak loads, which exacerbates grid stability issues. To mitigate these challenges and transition towards sustainable energy systems, hydro-wind-solar hybrid systems offer a promising solution for efficient, cost-effective, and environmentally energy production. This study introduces a novel stochastic optimization framework for short-term peak shaving in a hybrid renewable energy system comprising hydro, wind, and solar power sources. The framework explicitly accounts for dual uncertainties, namely those associated with energy supply (hydrological runoff, wind, and solar power) and electricity demand, which complicate grid stability and real-time dispatch decisions in hybrid systems. The framework integrates deep learning models, specifically deep convolutional generative adversarial networks (DCGAN) for simulating wind and solar power generation uncertainties, and the martingale model for hydrological runoff and load demand uncertainties. A stochastic optimization model for peak shaving is developed within the framework of stochastic programming. This model is designed to minimize the peak-valley variation of the residual load, factoring in uncertainties from both the energy supply and demand perspectives. The particle swarm optimization algorithm is employed for model solving, and a scenario reduction algorithm based on probability distance is adopted to balance scenario scale and computational efficiency. Numerical experiments on a hydro-wind-solar hybrid system in the upper reaches of the Yellow River in China demonstrate the effectiveness of the framework. The results show that the DCGAN model effectively encapsulates the probabilistic distributions of wind and solar energy, demonstrating robust generalization with an error rate below 2 % for all samples without assuming specific random variable distributions. The integrated operation of hydro-wind-solar energy systems significantly reduces peak-to-valley load differences and yields notable improvements in overall operational benefits compared to individual energy source optimization or stand-alone hydropower scheduling. Crucially, the scenario reduction algorithm preserves solution quality while substantially reducing computational burden. Furthermore, the marginal compromise of a 0.2 % reduction in hydropower generation is justified as a rational and economical trade-off for achieving a stable and controllable load process. Collectively, these findings underscore the framework's potential to optimize both energy generation and grid stability. Consequently, the insights gained contribute to advancing the development of more robust and responsive power system management strategies, thereby supporting the transition towards sustainable and resilient energy infrastructures.