Stochastic Co-Optimization and Fast Frequency Regulation of Variable-Speed Pumped Storage Hydropower Integrated with Floating Solar Photovoltaic Plants

Abstract

Fixed-speed pumped storage hydropower operates at constant power input during pumping, limiting its utility for mitigating rapid solar photovoltaic generation ramp events. Floating solar photovoltaics co-located on reservoir surfaces reduce evaporative water loss and share grid interconnection infrastructure, but cloud transients introduce severe sub-hourly power volatility. Variable-speed pumped storage hydropower using doubly fed induction machines (DFIMs) provides bidirectional power modulation, yet existing control strategies struggle to balance hydrodynamic cavitation limits with sub-second frequency response requirements, and existing scheduling models rarely define how a levelized cost of energy is to be computed for a co-optimized, storage-plus-generation hybrid in the first place. We present a two-layer stochastic co-optimization and adaptive electromechanical control architecture for an integrated 200-MW variable-speed pumped storage hydropower facility and 150-MWp floating solar photovoltaic plant. The upper layer uses a multi-period stochastic programming model, with an explicit capital-recovery-factor-based levelized-cost definition, to co-optimize energy arbitrage, evaporation savings, and spinning reserve schedules under meteorological uncertainty. The lower layer deploys a nonlinear governor-inverter controller that modulates machine rotor speed to deliver primary frequency response within 250 ms while restricting turbine suction head drop, via an explicitly signed frequency-deviation droop law, to prevent runner cavitation. Simulations show the hybrid framework reduces system levelized cost of energy by 18.4%, conserves 1.42 million cubic meters of reservoir water annually, and improves grid frequency nadir by 42.3% relative to fixed-speed operation during major generation-trip contingencies, while an unconstrained variable-speed controller that ignores the cavitation limit achieves a marginally better nadir at the cost of a negative Thoma margin, i.e., active cavitation.