Abstract:

Carbon capture and storage is a climate change mitigation technology that involves collecting and injecting CO2 emissions from fossil fuel burning power plants, or other point source emitters, into deep underground geologic formations such as brine aquifers. In carbon storage operations, structurally trapped CO2 that is mobile (i.e., able to flow as a free-CO2 phase) will be susceptible to leakage if the cap rock is compromised. Thus, the environmental and economic risks associated with a sequestration project can be reduced by facilitating other storage mechanisms and minimizing the amount of structurally trapped CO2. Data assimilation is also essential if we are to reduce the uncertainty in the CO2 plume location and to quickly identify any leakage through the cap rock.

In this work we develop and apply computational optimization procedures to minimize the risk of CO2 leakage and to perform data assimilation in order to identify the location of the CO2 plume and detect CO2 leakage through the cap rock. The risk of leakage is quantified both in terms of the mobile CO2 in the formation and in terms of the total mobility of free CO2 at the top of the storage aquifer. Risk minimization is accomplished by determining optimum locations and time-varying injection rates for a set of horizontal CO2 injection wells. Both Hooke-Jeeves Direct Search and Particle Swarm Optimization algorithms are used for this purpose. A brine cycling procedure, in which brine is periodically produced at the bottom of the aquifer and reinjected at the top of the aquifer, is also considered, and the parameters associated with this operation are optimized. For data assimilation (or history matching), aquifer geology is represented in terms of a relatively small number of parameters using a Karhunen- Loeve (K-L) expansion. Sensor and CO2 injection-well data provide the measurements to be matched. A procedure for optimizing the placement of monitoring wells and the weights of the various types of measured data, with the goal of maximizing the efficacy of the history matching procedure, is also presented.

Optimization results for both deterministic and uncertain aquifer models (in the latter case, the aquifer is represented using multiple realizations) are presented for a variety of cases, and reduction in the risk of leakage is consistently achieved. Specifically, by optimizing well placement and control (with known geology), the mobile CO2 fraction is reduced from around 0.32 to 0.22, and the total mobility is decreased by around 39%. For cases with uncertain geology, the reduction in mobile CO2 from optimization is only 7%, highlighting the need for a-priori geologic characterization. Optimizing brine cycling processes leads to further risk reduction, and a plot of risk of leakage versus pore volume of brine injected (which is related to cost) provides a Pareto front for a bi-objective optimization involving these two variables as objectives. The data assimilation procedure is shown to improve predictions for the CO2 plume location relative to results from prior geological models. Specifically, in a series of tests, this procedure reduces the average error in the predicted CO2 mobility in the top layer of the model (which is the quantity of interest) by 46% relative to the error using the prior model.

Finally, we investigate the early detection of leaks in the cap rock using pressure data. We introduce a three-region model to quantify the amount of leakage for a large number of leakage cases (some including multiple leaks). A data assimilation method is applied to determine leakage locations and permeabilities for a number of cases, with pressures at sensor wells and injection wells providing the measured data. Particle Swarm Optimization is used for the minimizations associated with this data assimilation problem. A data-rich scenario with nine sensor wells (completed in the overlying aquifer and storage formation) and a data-scarce scenario with four sensor wells (completed only in the overlying aquifer) are considered. Results indicate that the history matching procedure effectively locates leakage positions in cases with a single leak, for both the data-rich and data-scarce scenarios. For cases with multiple leaks, however, the procedure is less reliable, though the data-rich scenario is shown to provide better matches than the data-scarce scenario.

Press the Back button in your browser.

Copyright 2013, David A. Cameron: Please note that the reports and theses are copyright to their original authors. Authors have given written permission for their work to be made available here. Readers who download reports from this site should honor the copyright of the original authors and may not copy or distribute the work further without the permission of the author, David A. Cameron.