Development of multi-objective co-optimization framework for underground hydrogen storage and carbon dioxide storage using machine learning algorithms

Abstract

Underground hydrogen storage is a suitable way to create a balance between energy production and consumption. However, if it is able to incorporate a solution such as Carbon Capture and Sequestration, it has brought double benefit to the environment. Hydrogen has a good potential for energy production and transfer, and its behavior in the porous media has differences compared to other gases. In order to achieve the highest recovery factor of hydrogen and carbon dioxide storage and high utility of the project, the operational parameters of underground hydrogen storage and carbon dioxide storage should be carefully determined. The optimization of these parameters is challenging, complex, and time-consuming. Artificial intelligence methods have been developed using optimization methods in this regard. Simulation and investigation of underground hydrogen storage in depleted oil reservoirs were conducted in this study. Using machine learning methods, a proxy model for the UHS process was constructed, and then the operating parameters of UHS are optimized with the help of NSGA-II algorithm under the three objective functions of Net Present Value, hydrogen recovery factor, and CCS. Multi-layer neural networks with different architectures were trained and evaluated on the database. Research results are promising, and the approach is inspiring. It was found that the selected proxy model was able to accurately predict all three objective functions, with a R-squared of 0.9988 and a relative squared error of 0.00048. When an accurate model for NPV is not available, two-objective optimization is more efficient, but if an accurate model is available, adding NPV as a third objective function is necessary to achieve economic solutions. In both optimization perspectives, two Pareto fronts including 500 optimal solutions were determined separately. The considered NPV model correlates better with the objective function of carbon dioxide storage, while it is less consistent with the objective function of hydrogen recovery.