Sustainable Hydrogen Production from Sour Gas: Integrating Machine Learning for Process Optimization and Prediction

Abstract

A Machine Learning (ML)-based surrogate model is developed in this thesis to conduct datadriven optimization of a hydrogen production process. When screening literature it has been found that Hydrogen Sulfide Methane Reformation (HSMR) is one of the most promising routes for hydrogen production from sour gas. As a pre-requisite for developing a ML-based surrogate model, a first principle-based hydrogen production simulation of HSMR process was developed on Aspen Hysys V14. Secondly, various machine learning (Linear Regression, Support Vector Regression, Random Forest Regression, Extreme Gradient Boosting) and deep learning (Artificial Neural Networks) models were trained to ascertain the best surrogate model. Given the strong dependence of data generation strategy (sampling techniques and sample sizes) on the quality of surrogate model, three different sampling techniques (Sobol, Stratified LHS, Importance sampling) with various sample sizes (100, 200, 500, 750, 1000) were also studied. The study systematically generated data for ten significant input features and six target variables using developed simulation of the HSMR process. ML Models were trained on thoroughly pre-processed datasets, a remarkable RMSE of 0.03598 and R² of 0.99904 in predicting the process outputs using Artificial neural network (ANN) was achieved. Stratified LHS technique was able to represent sample space uniformly without demanding bigger datasets. Finally, we conducted data-driven optimization of an ANN model using evolutionary algorithms within the Python environment; this optimization aimed to identify optimal process parameters for HSMR Process. A comparison between first principle based and data-driven optimization suggested that ML-based surrogate models are efficient not only for accurate prediction but also in effective optimization of HSMR process.

Date of Award	19 Jul 2024
Original language	American English
Supervisor	Ali Almansoori (Supervisor)