(195e) Portfolio Optimization of Carbon Capture, Utilization, and Sequestration Considering the Effect of Technological Learning

Rapid industrialization and urbanization have led to a drastic increase in anthropogenic Greenhouse Gas (GHG) emissions, posing severe environmental challenges [1]. CO₂ forms over 76% of GHG emissions [1], and thus, several CO₂ emission mitigation strategies are being developed and implemented to achieve Net Zero Emissions (NZE) by 2050 [2]. Power plants and industries are the two major stationary sources of CO₂ emissions due to the use of fossil-based fuels [1]. Thus, carbon capture is considered one of the mitigation options because it enables emission reduction while facilitating the continued use of fossil fuels to meet energy demands [3]. The captured carbon can be sequestered, utilized to enhance oil and gas recovery, or converted to value-added products. Though the capture strategy has been projected to mitigate CO₂ annual emissions by 7.6 billion tons by 2050 [3], the ongoing capture projects can capture only up to 160 million tons of CO₂ per year by 2030 [4]. Over 350 of these projects are still under development phases [5] and have high investment and operation costs [6]. Hence, cost reduction is important to accelerate the capture projects’ deployment rate to meet the requirements of the NZE scenario [7].

Technological cost reductions can be due to multiple and overlapping factors, such as investment in research and development (learning-by-researching), shared/spill-over knowledge (learning-by-copying), functionality improvement (learning-by-using), and economies-of-scale (learning-by-doing) [8]. Learning curves, such as log-linear, exponential, and S-curve, are used to mathematically represent the cost reductions occurring at different rates [9]. Each learning factor can have varying degrees of impact on the different components of capture planning, and the extent of individual impacts depends on the learning curve selected. For example, learning-by-using majorly impacts the reduction of operational costs, while learning-by-copying impacts the investment cost [10]. The technology should be implemented as early as possible to get the utmost benefit from the former learning factor. However, for the latter, deployment should be delayed to gain experience from as many other deployed projects as possible. Thus, the CO₂ capture strategy requires portfolio planning considering the effects of different learning factors to obtain optimal investment and operational decisions that maximize the cost reductions from overall technological learning.

Previous works have used mathematical models to optimize the supply chain of CO₂ capture, utilization, and sequestration (CCUS) by minimizing overall project costs to meet a capture target. These works select emission sources, capture materials and technologies, transportation routes to sequestration sites, and utilization options to minimize the overall cost of CCUS while meeting capture targets [11-13]. Our previous work focused on developing a deterministic integer programming (IP) model that assists project portfolio planning by optimizing the technologies’ deployment decisions based on the learning factors considered [14].

This work introduces a comprehensive multiperiod optimization model that integrates the emission sources, capture technologies and materials, utilization technologies, pipeline networks, and sequestration sites under a single network for planning CCUS while considering the contribution of different learning factors toward project cost reduction along the planning horizon. For the given annual CO₂ reduction targets and utilization product demands, the objective of the optimization model is to minimize the net cost involved with the CCUS network over a planning horizon. The net cost is the difference between the cost of carbon capture, compression, transportation, utilization, and sequestration, computed accounting for the cost reduction due to technological learning, and the revenue from the utilization products and the enhanced recovery of oil and gas.

A deterministic mixed-integer non-linear programming (MINLP) model has been proposed to optimize the strategic CCUS deployment, considering the effect of technological learning. The cost models that estimate the investment and operation costs of capture and utilization technologies inherently account for cost reduction due to economies of scale [11, 15]. These models lead to nonlinearity in the optimization model, making the model intractable for real-world problems where CCUS is being planned for region-wide, state-wide, or nationwide emission sources considering cost reduction due to several learning factors over longer planning horizons. In this work, two heuristics are employed to solve such large-size problems. Firstly, a rolling horizon-based decomposition approach is used to obtain a lower bound for the problem. The optimization problem is repeatedly modeled and solved for a certain period of time (forecast horizon), and then, while solving for the remaining time periods (decision horizon), the decisions made in the forecast horizon are considered fixed in these periods [16]. The solutions from these decomposed problems provide variable cuts to reduce the feasible solution space of the original problem, which can now be solved to obtain an upper bound for the problem. The gap between the upper and lower bounds is tightened using different rolling strategies and variable cuts while solving the original problem. Despite this decomposition strategy, the MINLP models can require high computational resources. Thus, a two-step iterative approach is used for solving each large MINLP model in the decomposition strategy- the MINLP model is relaxed into a MILP model to obtain a lower bound, and an initialized MINLP model with variable cuts is solved to obtain an upper bound, which is an optimal solution for the MINLP model [13]. These approaches can be used to obtain optimal solutions within the 1% optimality gap. The optimization models are formulated in Python V3.8.6 using PYOMO V6.4.1. The MILP models are solved using CPLEX V20.10, and the initialized MINLP models are solved using DICOPT V2 through GAMS V24.8.5, all on an Intel Xeon Gold 6248R 3 GHz processor with 48 cores and utilizing a maximum of 50 GB RAM.

The capabilities of the developed model are demonstrated with a case study of planning CCUS for Alabama state, assuming different capture targets over a planning period of 25 years, discretized into five equal time periods. The emission sources under study can be categorized based on industry type, such as chemical, metal, mineral, oil and gas, power, pulp and paper, and other industries. The annual CO₂ emission rates vary from a few tons to 20 million tons, and the CO₂ composition in the flue gas varies from 2.5 to 99 mol.% [17]. Four sequestration sites have been selected. Two of these sites are offshore sandstone formations, and at the other two onshore sites, enhanced recovery of oil and methane is possible [18]. Two post-combustion capture technologies are considered- absorption technology using aqueous monoethanolamine (ABS-MEA), which is a commercial technology, and pressure swing adsorption technology using methyl viologen exchanged zeolite Y (PSA-MVY), which is still under pilot-scale studies and demonstration. Three sustainable utilization paths that produce ethanol (by combined reforming), acetic acid, and dimethyl ether (by dry reforming) are considered. Of these products, ethanol has the highest demand, and dimethyl has the highest selling price. The effect of cost reduction in investment on portfolio planning decisions is analyzed by considering technological learning from shared knowledge and functionality improvement for capture technologies and learning from investment in research and development for the utilization technologies. The case study results for multiperiod planning of CCUS without technological learning reveal that only large emission sources are preferred for capture in the early planning horizon. However, the preliminary results of the planning considering technological learning exhibited the scope for cost reduction by implementing more facilities at early planning periods to accelerate technological learning.

References

1. EPA (2024). Draft Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990-2022.
2. IEA (2022). The Need for Net Zero Demonstration Projects.
3. IEA (2021). Net zero by 2050: A roadmap for the global energy sector.
4. IEA (2022). Large-scale CO₂ capture projects in industry and transformation, actual vs. Net Zero Scenario, 2020-2030.
5. Global CCS Institute (2023). CCS Facilities Database.
6. Yan, Y., Borhani, T. N., Subraveti, S. G., Pai, K. N., Prasad, V., Rajendran, A., ... & Clough, P. T. (2021).
7. de Kleijne, K., Hanssen, S. V., van Dinteren, L., Huijbregts, M. A., van Zelm, R., & de Coninck, H. (2022).
8. Upstill, G., & Hall, P. (2018).
9. Anzanello, M. J., & Fogliatto, F. S. (2011).
10. Elia, A., Kamidelivand, M., Rogan, F., & Gallachóir, B. Ó. (2021).
11. Hasan, M. F., Boukouvala, F., First, E. L., & Floudas, C. A. (2014).
12. Roh, K., Al‐Hunaidy, A. S., Imran, H., & Lee, J. H. (2019).
13. Santhamoorthy, P., Subramani, H. J., & Cremaschi, S. (2023). FOCAPO/CPC.
14. Santhamoorthy, P., & Cremaschi, S. (2024). ESCAPE34/PSE24
15. Chung, W., Lim, H., Lee, J. S., Al-Hunaidy, A. S., Imran, H., Jamal, A., ... & Lee, J. H. (2022).
16. Glomb, L., Liers, F., & Rösel, F. (2022).
17. EPA (2022). Greenhouse Gas Emissions from Large Facilities.
18. Bauer, J., Rowan, C., Barkhurst, A., Digiulio, J., Jones, K., Sabbatino, M., Rose, K., & Wingo, P. (2018). NATCARB.