Surrogate-Based Constrained Single-Objective Bayesian Optimization for Joule-Heated Methane Dry Reforming Reactors

Joule heating coupled with porous media offers rapid in situ heat generation and a more uniform temperature distribution for DRM reactions, whereas the strong coupling of structural and operating parameters renders conventional trial-and-error optimization inefficient and costly. To tackle this challenge, a multi-physics numerical model for a centrally resistively heated DRM reactor is constructed by coupling Joule heating, flow and heat transfer, multicomponent species transport, and apparent reaction kinetics. A fully connected feedforward neural network (FC-FNN) surrogate model is further established to enable rapid prediction of reactor performance. Seven critical input variables are selected, including reaction temperature, gas hourly space velocity (GHSV), etc., and 9216 simulation samples are generated via numerical simulations for surrogate model training and parametric analysis. The results demonstrate that the Bayesian-optimized FC-FNN model exhibits high predictive accuracy for the core reactor performance metrics, achieving coefficients of determination (R2) above 0.960 for CH4 conversion, CO2 conversion, energy efficiency, and the H2/CO molar ratio. Subsequently, an unknown-constrained alternating Bayesian optimization (UCA-BO) algorithm is adopted to maximize the average reactant conversion under the constraints of H2/CO molar ratio ≥ 0.9 and bed-temperature non-uniformity ≤ 5%. Under the optimized operating conditions, the average reactant conversion reaches 94.12%, accompanied by a H2/CO molar ratio of 0.98, a bed-temperature non-uniformity of 4.53%, and a system energy efficiency of 74.83%. Posterior verification with the trained surrogate model gives slightly lower values (91.28%, 0.91, 4.86%, and 72.36%), all of which remain within the imposed constraints, confirming feasibility. Overall, the proposed integrated framework of numerical simulation, surrogate modeling, and Bayesian optimization enables efficient parameter screening and comprehensive performance optimization of resistively heated porous-media DRM systems, providing a methodological basis for the structural design and operating-parameter optimization of related reactor systems.