An AI-Based Multi-Stage Optimization Framework for Fuel Cell-Integrated Hybrid Renewable Energy Systems

The increasing penetration of intermittent renewable energy sources necessitates capacity-sizing frameworks that integrate operational forecasting with long-term economic assessment. Here, we propose a multi-stage optimization framework for fuel-cell-integrated renewable energy systems that combines machine-learning-based forecasting of electricity demand and photovoltaic (PV) generation, hysteresis-based dynamic simulation, and component cost sensitivity analysis. Using standard meteorological and time-derived variables, the forecasting models achieved coefficients of determination (R2) of 0.866 and 0.876 for electricity demand and PV generation, respectively, on evaluation subsets obtained through random 80/20 splits. The numbers of PV panels, battery energy storage system (BESS) units, and fuel cells were optimized under predefined grid power supply ratio (GPSR) scenarios using meteorological data from 2011 to 2022, and their operation was evaluated using weather data from 2023 to 2024. Most scenarios satisfied the 10% state-of-charge (SOC) criterion. The GPSR 0.5 scenario was classified as Fail because SOC fell below this criterion for four consecutive hourly time steps; nevertheless, electricity supply was maintained without shortage throughout the validation period. The reconfigurational gain proposed in this study quantifies the difference in levelized cost of electricity (LCOE) between retaining a baseline configuration and reoptimizing capacities at identical component unit prices. At a cost ratio of 0.1 for the respective component, battery cost reduction yielded an additional 35.48% saving at GPSR 0.4, whereas fuel-cell cost reduction yielded an additional 11.4% saving at GPSR 0.5. Under the 2030 scenario, with the PV unit price held constant and battery and fuel-cell prices set to 49% and 44% of their current values, respectively, LCOE decreased to 0.1008–0.1286 USD/kWh.