[CCoE Notice] Cullen College Dissertation Announcement (ECE)

[Hutchinson, Inez A]

[Dissertation Defense Announcement at the Cullen College of Engineering]
Large-scale Carbon Monitoring
WITH  DATA-DRIVEN OPTIMIZATION AND MACHINE LEARNING-ENHANCED SIGNAL PROCESSING
Yuan Zi

April 23, 2024; 9:00 AM - 11:00 AM (CST)
Location: ECE Large Conference room, D N328, Eng Bld 1
Zoom: https://urldefense.com/v3/__https://uh-edu-cougarnet.zoom.us/j/99146061274__;!!LkSTlj0I!FCLRp-tKYKYThdEuaM0HBZ555IzxypC1fFKLgWDlxLSpAV-P8Q1Dq-O0-1SQKINV959zqSJe01BXMbu5nCrFPzgB9p0$ <https://urldefense.com/v3/__https://uh-edu-cougarnet.zoom.us/j/99146061274__;!!LkSTlj0I!A5yZmVOA6gENs_Juiy1AM6Y3STGfwLUj5_hcXgJhhckeksGZKDdM3lxUvSoZ0KbD9opHoG4Y8RMjQtLWSGliDfS3eIgHgK8$>

Committee Chair:
Dr. Jiefu Chen, Ph.D.

Committee Co-Chair:

Dr. Zhu Han, Ph.D.

Committee Members:
Dr. Lei Fan, Ph.D. | Dr. Saurabh Prasad, Ph.D. | Prof. Jie Zhang, Ph.D. | Dr. Xuqing Wu, Ph.D.

Abstract

This study offers an examination of data-driven optimization techniques for comprehensive carbon monitoring on a large scale, integrating strategies for the strategic placement of induced passive seismic sensors, and methane point sensors, and leveraging machine learning for remote methane detection. The first chapter underscores the pivotal importance of passive-seismic monitoring in geological carbon storage initiatives, advocating for secure and continuous surveillance of passive-seismic activities during CO2 injections. It introduces advancements in subsurface characterization through machine learning enhanced Subsurface Imaging and Pattern Recognition, and devises an optimal strategy for placing ground geophone networks using stochastic optimization. Specifically, it utilizes P-median stochastic programming to balance the monitoring capabilities against budgetary constraints, aiming to optimize performance within financial limits with a particular emphasis on prompt alarm detection and accurate localization. The methodology considers site-specific passive-seismic scenarios to mitigate uncertainties, significantly enhancing detection capacities over traditional grid configurations. The second chapter transitions to exploring sensor placement optimization techniques for expansive field environments, constrained by budgetary considerations. It critically addresses the limitations inherent in conventional stochastic programming, particularly its inadequacy in managing worst-case scenarios and environmental unpredictabilities. By introducing distributionally robust optimization, the study enhances the robustness of sensor network detections, aiming to minimize the detection time expectancy amidst a range of uncertainties such as unpredictable leakage rates, sensor delays, and variable wind conditions. Demonstrated through atmospheric simulations tailored to specific methane emission scenarios, this approach is shown to outperform stochastic programming-based methods in ensuring superior worst-case operational outcomes. In the third chapter, the dissertation pivots to a fundamental machine-learning approach for the detection of methane remotely. By harnessing extensive machine learning algorithms, it articulates the development of a remote methane detection system that incorporates fundamental machine learning models and prompt engineering to significantly improve monitoring efficiency. Utilizing traditional signal processing technique matched filter for prior extraction without necessitating custom model training, the developed anomaly detection system is capable of providing precise and timely alerts for methane emissions across extensive areas in a sensor-agnostic manner. This approach not only complements but also synthesizes the surface and subsurface optimization monitoring strategies delineated in preceding chapters, culminating in a comprehensive framework for large-scale carbon monitoring. Collectively, this thesis enriches the domain of large-scale carbon monitoring by introducing the strengths of multi-modality data, data-driven optimization, and machine learning. This fusion not only augments the precision, efficiency, and reliability of passive seismic and methane point sensor placements but also advances real-time remote methane detection capabilities. The insights and methodologies delineated across the chapters furnish robust tools for enhancing decision-making processes within the realms of geological carbon storage, environmental surveillance, and business resilience.

[Engineered For What's Next]



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://Bug.EGR.UH.EDU/pipermail/engi-dist/attachments/20240409/55893b09/attachment.html