Conclusion

We proposed a process design method that combines Bayesian optimization and a process simulator to search for design variables that satisfy the performances of an ethylene oxide plant with a small number of simulations. We verified the effectiveness of the method by comparing it with a random search. It was confirmed by the case study that the candidates for design variables that achieve the plant performances can be efficiently proposed. The reproducibility of the proposed method was also confirmed. Moreover, various candidates were obtained by increasing the number of Bayesian optimization trials. This method is expected to meet the needs of knowledgeable chemical engineers and to facilitate process designs of new plants.