TY - JOUR
T1 - Efficient prediction of hydrogen storage performance in depleted gas reservoirs using machine learning
AU - Mao, Shaowen
AU - Chen, Bailian
AU - Malki, Mohamed
AU - Chen, Fangxuan
AU - Morales, Misael
AU - Ma, Zhiwei
AU - Mehana, Mohamed
PY - 2024/5
Y1 - 2024/5
N2 - Underground hydrogen (H2) storage (UHS) has emerged as a promising technology to facilitate the widespread adoption of fluctuating renewable energy sources. However, the current UHS experience primarily focuses on salt caverns, with no working examples of storing pure H2 in porous reservoirs. A key challenge in UHS within porous rocks is the uncertainty in evaluating storage performance due to complicated geological and operational conditions. While physics-based reservoir simulations are commonly used to quantify H2 injection and withdrawal processes during storage cycles, they are computationally demanding and unsuitable for providing rapid support to UHS operations. In this study, we develop efficient reduced-order models (ROMs) for UHS in depleted natural gas reservoirs using deep neural networks (DNNs) based on comprehensive reservoir simulation data sets. The ROMs can accurately forecast UHS performance metrics (H2 withdrawal efficiency, produced H2 purity, produced gas-water ratio) across various geological and operational conditions and are over 22000 times faster than physics-based simulations. Then, we employ the ROMs for sensitivity analysis to assess the impact of geological and operational parameters on UHS performance and conduct uncertainty quantification to characterize potential performance and associated probabilities. Lastly, we present a field case study from the Dakota formation of the Basin field in the Intermountain-West (I-WEST) region, USA. Based on the ROMs’ predictions, Dakota formation is favorable for UHS due to its high H2 withdrawal efficiency and purity, and low water production risk. By optimizing operational parameters, we can further improve the storage performance in Dakota formation and reduce the uncertainty in UHS performance prediction. This study introduces an efficient ROM-based approach to assess and optimize UHS performance, supporting the development of effective UHS projects in depleted gas reservoirs.
AB - Underground hydrogen (H2) storage (UHS) has emerged as a promising technology to facilitate the widespread adoption of fluctuating renewable energy sources. However, the current UHS experience primarily focuses on salt caverns, with no working examples of storing pure H2 in porous reservoirs. A key challenge in UHS within porous rocks is the uncertainty in evaluating storage performance due to complicated geological and operational conditions. While physics-based reservoir simulations are commonly used to quantify H2 injection and withdrawal processes during storage cycles, they are computationally demanding and unsuitable for providing rapid support to UHS operations. In this study, we develop efficient reduced-order models (ROMs) for UHS in depleted natural gas reservoirs using deep neural networks (DNNs) based on comprehensive reservoir simulation data sets. The ROMs can accurately forecast UHS performance metrics (H2 withdrawal efficiency, produced H2 purity, produced gas-water ratio) across various geological and operational conditions and are over 22000 times faster than physics-based simulations. Then, we employ the ROMs for sensitivity analysis to assess the impact of geological and operational parameters on UHS performance and conduct uncertainty quantification to characterize potential performance and associated probabilities. Lastly, we present a field case study from the Dakota formation of the Basin field in the Intermountain-West (I-WEST) region, USA. Based on the ROMs’ predictions, Dakota formation is favorable for UHS due to its high H2 withdrawal efficiency and purity, and low water production risk. By optimizing operational parameters, we can further improve the storage performance in Dakota formation and reduce the uncertainty in UHS performance prediction. This study introduces an efficient ROM-based approach to assess and optimize UHS performance, supporting the development of effective UHS projects in depleted gas reservoirs.
UR - http://www.scopus.com/inward/record.url?scp=85186268083&partnerID=8YFLogxK
U2 - 10.1016/j.apenergy.2024.122914
DO - 10.1016/j.apenergy.2024.122914
M3 - Article
SN - 0306-2619
VL - 361
JO - Applied Energy
JF - Applied Energy
M1 - 122914
ER -