Advanced modelling and optimization of steam methane reforming: From CFD simulation to machine learning - Driven optimization.

Computational fluid dynamics simulations were utilized to investigate the steam methane reforming process with the aim to improve its efficiency. Key parameters examined for their impact on process performance included surface heat flux (73–108 kW/m2), tube length (1–16 m), steam-to-carbon ratio (1.4–4), and flow rate (0.22–0.38 kg/s). To analyze the simultaneous effects of these variables while reducing computational costs, Deep Neural Networks (DNN) were employed. An optimized DNN was designed to achieve acceptable performance, featuring an input layer with four neurons that represent reformer length, flow rate, heat flux, and steam-to-carbon ratio. The network includes four hidden layers with 32, 16, 8, and 8 neurons respectively, and concludes with an output layer comprising seven neurons for residual methane, water vapor, produced hydrogen, carbon dioxide, carbon monoxide, wall temperature, and gas outlet temperature. The results indicated that the proposed model achieved high accuracy, exceeding 99%, in predicting both training and test data. Following the DNN modeling, an optimization algorithm based on the random search method was developed. This algorithm searches a wide range of parameters to identify the optimal conditions for simultaneously maximizing hydrogen production and minimizing reformer length. [Display omitted] • Efficiency Enhancement: Improved steam methane reforming efficiency via CFD analysis. • Critical Parameters: Analyzed heat flux, tube length, S/C ratio, and flow rate impacts. • Machine Learning Approach: Employed deep neural networks to optimize multiple variables. • High Accuracy Prediction: Achieved over 99% accuracy in training and test data predictions. • Optimization Algorithm: Developed an algorithm using random search techniques. [ABSTRACT FROM AUTHOR]