Multi-objective optimization of hydrogen production based on integration of process-based modeling and machine learning

Abstract

Hydrogen production through various methods has emerged as a crucial aspect of sustainable energy systems. This study presents a novel approach to optimize hydrogen production in a steam methane reforming reactor using a multi-objective framework coupled with the integration of process-based modeling and machine learning. Conventionally, process-based modeling is conducted using complex physical and chemical equations, often involving differential equations, governing the reactor. However, this approach requires significant computational costs. One emerging solution is to model the reactor using machine learning based on operational data. This approach solves the computational cost problem, but machine learning models are black boxes in nature thus the model lacks interpretability. Incorporating physical laws governing the reactor makes the model to be interpretable while at the same time reducing the computational cost. In this study, process-based modeling is utilized to seamlessly incorporate physical laws into machine learning. The developed machine learning-based surrogate model is then used as a background model in the optimization framework, providing rapid predictions of process responses to enable multi-objective optimizations in terms of the levelized cost of hydrogen and specific carbon emissions. The machine learning-based surrogate model improves the computational efficiency by 3,627 times compared to the process-based model. The levelized cost of hydrogen and specific carbon emissions will simultaneously be minimized by finding the appropriate operating conditions in the reactor. Furthermore, the optimization results provide the trade-offs between these objectives to offer decision-makers valuable insights for designing and operating hydrogen production facilities.