Multi-level optimization of green hydrogen-integrated gas turbine power plants: a hierarchical AI framework for minimum LCOE and emissions

Hybrid natural energy sources have gained high attention, specifically concentrated solar power (CSP) that is combined with other traditional energy sources. The integration of green hydrogen into natural gas-fired power plants presents a critical pathway for decarbonizing the energy sector while maintaining grid reliability. However, the optimal hydrogen blending ratio varies dynamically with ambient conditions, load demand, and renewable availability. This paper presents a novel hierarchical artificial intelligence framework that optimizes hydrogen blending in gas turbines under five time-varying factors: ambient temperature between 10 and 50°C, relative humidity from 15 to 85%, solar irradiance from 0 to 1050 W/m², load demand ranges from 40 to 100%, and blade cooling effectiveness includes 0.5 to 0.85. Using Monte Carlo simulation (n = 2000), Bayesian optimization, and multi-level optimization, we demonstrate that the optimal H₂ ratio exhibits counter-intuitive behaviour: it decreases monotonically from about 42% at 10°C to 22% at 50°C, indicating that cooler ambient temperatures are distinctly more favourable. The proposed framework achieves an 11.3% reduction in Levelized Cost of Energy, an 18.8% NOx reduction, and a 25% CO₂ reduction at optimal H₂ ratios of between 18% and 22%. Bayesian optimization outperforms grid and random search methods by 4 to 6% in final LCOE. The 3D Pareto front reveals that simultaneous minimization of LCOE, NOx, and CO₂ is possible within a narrow H₂ ratio window range from 15 to 25%, beyond which trade-offs become severe.