Waqar Ali Khan, Ashkan Pakseresht, Caslon Chua, Ali Yavari
The optimization of Hydrogen-Integrated Hybrid Renewable Energy Systems remains challenging due to conflicting objectives, nonlinear operational constraints, and the limited adaptability of conventional evolutionary algorithms. This study introduces an adaptive Non-dominated Sorting Genetic Algorithm II (NSGA-II) framework that integrates dynamic mutation, selective local search, constraint-aware offspring generation, and cross-run learning. The method is applied to a residential-scale hybrid hydrogen–renewable energy system in Broadmeadows, Melbourne, comprising photovoltaic panels, wind turbines, battery storage, proton exchange membrane electrolyzers, hydrogen tanks, fuel cells, and diesel backup. Four objectives are considered: levelized cost of electricity (LCOE), carbon dioxide (CO 2 ) emissions, non-renewable hours (NRH), and renewable energy fraction (REF) penalty. The optimization is performed within a dispatch-coupled model that enforces hourly energy balances and state-of-charge limits, supported by demand and resource forecasts generated using extreme gradient boosting models trained on long-term meteorological data. The proposed adaptive NSGA-II achieves stronger performance than the conventional NSGA-II across hypervolume, Pareto diversity, and convergence indicators. The optimal designs favor greater storage capacity, with battery capacity rising from 105 MWh to 112 MWh, hydrogen storage increasing from 0.39 tons to 1.0 ton, and electrolyzer capacity growing from 4.8 MW to 5.2 MW. These adjustments reduce LCOE from 5.9169 to 5.8083 AUD/kWh, lower annual CO 2 emissions by 45% (from 94 tons/year to 52 tons/year), improve REF from 82.34% to 88.41%, and decrease NRH from 103 to 53 h. The results demonstrate that adaptive NSGA-II produces balanced solution portfolios that support high renewable utilization, reduce reliance on fossil backup, and provide practical insights for planning hydrogen-integrated sustainable energy systems.