Multiobjective optimization of ammonia-based hydrogen-storage systems using thermodynamic and neural-network models

Ammonia, owing to its high energy density and long-term storage potential, is regarded as a promising medium for storing green hydrogen produced by renewable-powered electrolysis. However, its synthesis is energy-intensive and generates a substantial amount of waste heat. To address this issue, this study proposes a novel cascaded waste heat recovery system that employs thermal oil as an intermediate heat-transfer fluid, thereby achieving thermal decoupling between the green ammonia synthesis process and bottoming cycle. The bottoming cycle integrates a recompression–reheating supercritical CO2 cycle in series with an organic Rankine cycle, facilitating separate recovery of multi-grade waste heat for cascaded utilization while maintaining high operational flexibility. A thermodynamic model is developed to assess the system performance and environmental impacts. A multilayer perceptron-surrogate trained on simulation data models the thermodynamic input–output mapping from decision variables to performance objectives and serves as a fast evaluator within the nondominated-sorting-genetic algorithm-III to accelerate optimization. The inverted-generational-distance index is applied to evaluate the Pareto-front quality. Under optimal conditions, the system delivers a net power output of 8807.77 kW, with a thermal efficiency of 32.85 % and exergy efficiency of 68.64 %. These results demonstrate that the proposed framework facilitates deep recovery and efficient utilization of waste heat, enhances operational flexibility, and offers a scalable pathway for energy optimization in ammonia-based hydrogen-storage systems.