Loading...

Optimization of Underground Hydrogen Storage (UHS) Operation using Artificial Intelligence

Akhtarkavan, Saeb | 2025

18 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58850 (06)
  4. University: Sharif University of Technology
  5. Department: Chemical and Petroleum Engineering
  6. Advisor(s): Ayatollahi, Shahab; Mahani, Hassan
  7. Abstract:
  8. Growing global concerns over climate change and the continued reliance on fossil fuels have increased the importance of renewable energy sources and the need for large‑scale energy storage. Within this context, underground hydrogen storage has emerged as a promising solution for managing fluctuations in renewable energy production and demand. Effective deployment of such systems requires a detailed understanding of hydrogen flow behavior in porous media, particularly in the presence of in‑situ reservoir fluids, which introduces significant complexity. Developing accurate modeling approaches and optimized operational strategies is therefore essential. A major challenge is the dependence on physics‑based models that directly solve the governing flow and transport equations. Although these simulations provide valuable physical insights, their high computational cost and long execution times limit their practicality for extensive scenario analysis and optimization. Reduced‑order (or surrogate) models based on artificial intelligence offer an attractive alternative, enabling accurate prediction of system behavior while dramatically improving computational efficiency. In this thesis, a base-case physical model for hydrogen storage in a depleted gas reservoir was developed using CMG‑GEM, and the dynamic reservoir response during cyclic injection and production was analyzed. The base‑case results show stable cyclic behavior, with average reservoir pressure oscillating between 21,000 and 26,000 kPa. Under these conditions, the average hydrogen purity in the produced stream is 83.5%. Using the CMOST module, 136 simulation scenarios were generated by varying geological and operational parameters, and a comprehensive dataset was constructed through Latin Hypercube Sampling. After evaluating multiple machine‑learning architectures, the multilayer perceptron (MLP) was selected as the optimal model due to its accuracy, stability, and ability to capture the nonlinear system behavior. Two surrogate models were developed to predict ultimate hydrogen recovery and produced hydrogen purity. These models successfully reproduced the numerical simulation results, achieving correlation coefficients (R2) of 0.9991 for hydrogen recovery and 0.9825 for purity. Computational efficiency improved substantially: while each physics‑based simulation required approximately 12 minutes, the surrogate models generated predictions in about 1 second—more than a 720‑fold speedup. Sensitivity analysis using the Sobol method identified operational parameters—particularly injection and production rates—as the dominant factors influencing system performance. In the optimization phase, the Particle Swarm Optimization (PSO) algorithm was applied to three cases: maximizing recovery, maximizing purity, and a combined multi‑objective optimization. Increasing the production rate while reducing the injection rate enhanced recovery (up to 87.0%) but lowered purity to 71.6%. Conversely, increasing the injection rate and reducing the production rate improved purity (up to 96.7%) while decreasing recovery to 79.6%. The multi‑objective optimization produced a balanced solution, increasing recovery and purity simultaneously to 80.48% and 84.58%, respectively. Validation against the physics‑based model confirmed that the surrogate models predict system responses with a relative error below 0.5%. These findings demonstrate that machine‑learning‑based reduced‑order models, combined with intelligent optimization algorithms, offer powerful, accurate, and computationally efficient tools for the design, analysis, and management of underground hydrogen storage systems, and can play a significant role in future decision‑making processes
  9. Keywords:
  10. Underground Hydrogen Storage ; Artificial Intelligence ; Surrogate Model ; CMG-GEM Software ; Surrogate Model Optimization

 Digital Object List

 Bookmark

No TOC