Design of a green ammonia production process by machine learning

Green ammonia is attracting attention as a way of realizing a decarbonized society. In this study, we designed a green ammonia production process under limited power supply. In conventional process optimization, candidates for the design variables are selected based on the experience of the process engineer and calculation results, and optimization is performed by repeated simulations. However, as the number of design variables increases, an enormous number of simulations are required to search for optimal candidates. In this study, we used Bayesian optimization to select candidates for the design variables with a high probability of achieving the target values of the objective variables. Candidate selection and simulation were repeated to maximize green ammonia production with limited power supply. The process was compared with previously reported green ammonia production processes, which showed that the process designed in this study produced the maximum amount of green ammonia at ≤ 10 MW.