Multiobjective mixed-integer Bayesian optimization for the implementation of hydrogen as maritime fuel through liquid organic hydrogen carriers (LOHCs)

Hydrogen is proposed as a fuel for maritime decarbonization, but its storage is challenging. Liquid Organic Hydrogen Carriers (LOHCs) offer a solution for storing hydrogen under ambient conditions enabling to reuse oil infrastructure. The dehydrogenation module, coupled with a Solid Oxide Fuel Cell, is modeled using a first-principles model. Detailed three-phase reactor modeling accounts for heat, mass and momentum and represents relevant phenomena in LOHCs reactors. Two different stages are set: process design and operation. In the first stage, unit design parameters and operating conditions are considered and several objectives are proposed. Additionally, several reaction technologies are suggested (mixed-integer formulation). With equipment defined, operation is optimized. Expensive high-complexity models hinder optimization. Hence, Machine Learning algorithms and specifically Bayesian Optimization (BO) can be leveraged for LOHCs optimization. Process design involves the optimization of a mixed-integer, single (SO) and multiobjective (MO) constrained formulation. Then, SO and MO constrained optimization are carried out for operation. To handle integers, adapted kernels and multiple acquisition function optimizers are used. These strategies perform comparably or better than approaches ignoring integers. Then, these are encouraged for an effective mixed-integer optimization. Design stage reached a trade-off with 1990 $/kWel and 5.95 l/kWel equipment investment and volume respectively and 37.49% efficiency. Detailed reactor models differ significantly from simplified baselines showing the need to accurately represent these units. Then, operation MO was carried out. Pareto Front was built with several operational points. Thus, the use of LOHCs for hydrogen storage in maritime sector is explored and promoted.