Multi-Objective Bayesian Optimization for Design and Operating of Fluidized Bed Reactor

Treating the explosive waste using fluidized bed reactor, the design and operating conditions have highly effect to emission of pollutants (e.g. nitrogen oxides). Although it is possible to reduce the amount of pollutants through additional unit processes and extreme design and operating conditions, there are many practical difficulties because it causes an increase in cost. In addition, because of the explosive properties of waste, designing the process through actual experiments has many risks. Therefore, computational fluid dynamics (CFD) is used to simulate the reactor with high accuracy and to observe the internal temperature characteristics. While CFD shows high accuracy, it is difficult to obtain sufficient data for optimization because it requires a long computation time. Bayesian optimization repeats surrogate model optimization and infill criteria optimization and adaptively constructs the surrogate model. It shows good performance for time-consuming or expensive experiments. This study is to identify the design and operating conditions that minimize nitrogen oxides and cost through multi- objective Bayesian optimization. Multi-objective optimization problems generally do not have a single global optimization solution, but multiple solutions. This set of solutions forms a pareto front, which derives various solutions and gives decision- makers many options.