Reinforcement learning-driven stochastic optimization and blockchain-enabled demand response to determine the techno-economic feasibility of hydrogen fueling stations in Australia
Paul C. Okonkwo, Samuel Chukwujindu Nwokolo, Edson L. Meyer, Chinedu Christian Ahia, Ibrahim B. Mansir
Abstract
This study aims to assess the feasibility of hydrogen refueling in the Central Business District (CBD) corridors of Australia, utilizing a reinforcement-learning-based stochastic framework integrated with blockchain-enabled demand response, with the objective of ensuring reliable dispatch of hybrid PV-wind-battery-electrolyze systems. The aim is to assess the techno-economic viability amidst tariff and weather uncertainties across six locations—North Terrace (Adelaide), Elizabeth Quay (Perth), Salamanca Place (Hobart), Southbank (Melbourne), Eagle Street (Brisbane), and Martin Place (Sydney)—utilizing district-scale solar and wind projections, performance-based costing, and multi-vector integration. This contribution integrates reinforcement learning (RL) control, digitally auditable demand response (DR) settlement, and spatially resolved siting into a cohesive decision framework, validated through physics-constrained simulations and bio-inspired optimizers, facilitating geography-aware configuration options and risk-sensitive investment evaluation. The scope encompasses PV–WT–Battery, PV–Battery, and WT–Battery configurations; electrolyze and tank dimensioning; demand scenarios; and sensitivity-robustness assessments. Highlights indicate that PV–WT–Battery consistently reduces lifespan metrics while achieving RF = 100 %. Elizabeth Quay represents the cost frontier (NPC ≈ $28.04 ×10 ³; LCOE ≈ $0.01747/kWh; LCOH ≈ $0.603/kg), with Eagle Street in proximity, North Terrace and Southbank at moderate elevations, and Martin Place at the upper limit; wind-dominant Salamanca Place displays a configuration inflection where 750 kW PV surpasses 1000 kW due to wind-solar covariance and part-load penalties. Reinforcement learning combined with dynamic pricing reduces LCOH variability by synchronizing electrolyze setpoints, compressor operations, and battery-tank buffers during low-marginal-cost periods, while blockchain settlement provides verifiable green-hydrogen credentials. Demand elasticity assessments reveal that −15 % scenarios yield around 10–12 % decreases in NPC/LCOE/LCOH and surpluses of about 25–30 %, whereas + 15 % scenarios result in cost increases of approximately 5–10 %. The grouping of hydrogen valleys near ports and transit hubs improves utilization and financial viability; alternative logistics (LH₂, LOHC) and subterranean storage expand site possibilities for densely populated urban areas. The strategic considerations support PV-biased trihybrids in the sun-drenched western and subtropical regions, diversified portfolios for Adelaide, Melbourne, and Sydney, and optimized wind-driven operations in Hobart, providing a replicable framework for achieving verifiable, net-zero-aligned HRS in Australian cities. Results are applicable across various climates, precinct types, and market conditions.