A beautiful marriage between POMDPs and subsurface applications : decision making for subsurface systems
Abstract/Contents
- Abstract
- Climate change is a pressing global issue, with the global average temperature rising to unprecedented levels, threatening the planet's habitability. Subsurface applications, such as subsurface remediation, carbon capture and storage (CCS), geothermal energy, and subsurface storage of renewable energy, hold great potential in addressing climate change and reducing greenhouse emissions. However, decision-making in subsurface operations is challenging due to the uncertainties inherent in subsurface environments and the long-term consequences of actions. As a result, we need a robust decision-making method capable of handling a wide range of scenarios in the subsurface. This dissertation examines widely adopted decision-making approaches in subsurface applications, identifying their limitations and proposing the use of Partially Observable Markov Decision Processes (POMDPs) as a more effective alternative. The thesis aims to bridge the gap between the subsurface and POMDP communities by providing comprehensive guidance on formulating and solving POMDPs for subsurface applications. Two examples, groundwater contaminant remediation and CCS, are presented to demonstrate the superiority of POMDPs performance compared to baseline approaches. This work also advocates for the increased adoption of POMDPs in subsurface applications, with the potential to revolutionize decision-making processes and contribute significantly to mitigating the effects of climate change. Chapter 2: A Review of Decision-Making Approaches. This chapter delves into the commonly used decision-making approaches in subsurface literature. We take a closer look at one-shot, model predictive control, and Markov decision process and examine their formulation. We highlight their limitations and how they might not necessarily lead to optimal decisions for subsurface applications. To facilitate a clear comparison, we use dynamic decision networks to visually compare these approaches. Chapter 3: Partially Observable Markov Decision Processes (POMDPs). We introduce POMDPs as a solution to the limitations of previous approaches. We explain the formulation of POMDPs and use dynamic decision networks to visualize them. We review belief updates or Bayesian inversion methods for POMDPs and propose a new inversion method. To bridge the communities between subsurface and POMDPs, we summarize similar terminologies and present them in a table. After formulating a subsurface problem as a POMDP, we review two cutting-edge online solvers for POMDPs. Chapter 4: Groundwater Management Example. We showcase a groundwater management example formulated as a POMDP and solved using a belief state planner. All components of the POMDP formulation are thoroughly illustrated. We compare our proposed POMDP approach with conventional optimization approaches and a hand-crafted heuristic solution method and demonstrate the superiority of using POMDP in this example. Chapter 5: Safe CCS Example. We provide another example to highlight the importance of our proposed approach for subsurface applications. We discuss how our proposed approach ensures long-term safety for CCS operations compared to other baseline approaches. We also demonstrate the flexibility of the proposed approach by introducing three different monitoring strategies and examining their impact on decision quality. Chapter 6: Conclusion and Future Opportunities. We summarize our main findings and contributions. We outline potential future research directions to extend the applicability of our proposed approach to make a significant impact in addressing climate change.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2023; ©2023 |
| Publication date | 2023; 2023 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Wang, Yizheng, (Machine learning scientist) |
|---|---|
| Degree supervisor | Caers, Jef |
| Thesis advisor | Caers, Jef |
| Thesis advisor | Mukerji, Tapan, 1965- |
| Thesis advisor | Tartakovsky, Daniel |
| Degree committee member | Mukerji, Tapan, 1965- |
| Degree committee member | Tartakovsky, Daniel |
| Associated with | Stanford Doerr School of Sustainability |
| Associated with | Stanford University, Department of Earth and Planetary Sciences |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Yizheng Wang. |
|---|---|
| Note | Submitted to the Department of Earth and Planetary Sciences. |
| Thesis | Thesis Ph.D. Stanford University 2023. |
| Location | https://purl.stanford.edu/pz111qs2822 |
Access conditions
- Copyright
- © 2023 by Yizheng Wang
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...