Protecting critical infrastructure, especially in a complex urban area or region, should focus on identifying and prioritizing potential failure points that would have the most severe consequences. Such prioritization can inform targeted planning and investment decisions, such as what infrastructure should be hardened or relocated first or what infrastructure should receive priority restoration following a disaster, among other uses. Without a prioritization process, assessment and protection programs are typically guided by intuition or expert judgement, and they often do not consider system-level resilience. While understanding how to prioritize high-consequence failure points for assessments and, for protection is essential, the complexity of infrastructure systems can quickly overwhelm. For example, in a notional region with 1,000 electric power assets, almost one million failure scenarios are associated with an N-2 contingency and nearly one billion failure scenarios are associated with an N-3 contingency. As a result, it is simply not feasible technically nor financially for system operators and government agencies to assess and prepare for all possible disruptions. Therefore, a primary goal of critical infrastructure protection and resilience programs should be to identify and prioritize the most critical contingencies affecting infrastructure systems. Achieving this goal will allow decision makers to identify high-impact isolated failures as well as cascading events, and to prioritize protection investments and restoration planning accordingly. To solve this problem, Argonne National Laboratory developed an optimization framework capable of modeling and prioritizing high-consequence failure points across critical infrastructure systems. The optimization framework can model at the system level or the interdependent “system-of-systems” level and is applicable to any infrastructure.

Download the pdf

Suggested Citation

Verner, Duane, Frederic Petit, and Kibaek Kim. “Incorporating Prioritization in Critical Infrastructure Security and Resilience Programs.” Homeland Security Affairs 13, Article 7 (October 2017). https://www.hsaj.org/articles/14091


Argonne National Laboratory (Argonne) has developed an optimization algorithm and modeling framework capable of identifying the highest-consequence failure points within critical infrastructure systems. The optimization algorithm and framework can be applied to any infrastructure at the system level or the interdependent “system-of-systems” level and can be used to model any combination of infrastructure failures. Results from the optimization modeling can be used by analysts to identify priority assets for assessments and to assist infrastructure system owners and operators and government agencies when they are making critical infrastructure protection and mitigation investment decisions.

Understanding Infrastructure Failures

A fundamental component of critical infrastructure security and resilience programs should include understanding how, why, and where systems fail. This understanding should guide decisions on where to conduct in-depth assessments as well as which protection and mitigation measures to pursue. However, a complicating factor is that infrastructure failures vary significantly. Some failures will generate significant consequences at the system or regional level, whereas effects from other failures remain local, while still others have little to no effect on the overall service provided. For illustration purposes, Figure 1 shows a 345-kV electric power transmission system between a generator substation and a remote substation.

Figure 01. Electric Transmission Lines1

In this example, the generation plant produces 1,520 MW2 of power that is transported to the remote substation via three transmission corridors. Corridor 1 combines two circuits (lines) that allow transport of a maximum of 750 MW. Corridor 2 is a single circuit that allows transport of a maximum of 400 MW. Corridor 3 combines two circuits that allow transport of a maximum of 800 MW. By design, the three corridors operate below their maximum capacity levels, which allows for the relocation of power among the remaining circuits in the event of a disruption in one of them. For example, if the Corridor 2 circuit fails, the system’s overall vulnerability will increase but it will not experience cascading system failure because the two other corridors can compensate for the loss (Figure 2).

Figure 02. Loss of Corridor 2 Circuit

Corridor 1 circuits would operate at 97% of their capability and Corridor 3 circuits would operate at 91% of their capability. Similarly, the loss of one circuit from Corridor 1 would not trigger a cascading system failure because of the ability of the remaining circuits to compensate (Figure 3).

Figure 03. Loss of One Circuit in Corridor 13

Building on the operating conditions identified in Figure 3, Corridor 3 would operate near full capacity (97%); Corridor 2 would operate at 72%; and the remaining circuit of Corridor 1 would operate at 103%, which, over time, could lead to the loss of the second circuit and therefore a failure of Corridor 1 (Figure 4).

Figure 04. Loss of Two Circuits in Corridor 1

A loss of Corridor 1 would impede the ability of the two other corridors to operate safely. Corridor 2’s circuit would operate at 104% of its capability, and Corridor 3’s circuits would operate at 129% of their capability. Under this scenario, the circuits could begin to heat and ultimately trip, triggering a system failure. Assuming all other risk factors are equal, this simplified example shows that the consequence of disruption of Corridor 1 is greater than disruption of Corridor 2, and, as such, Corridor 1 should receive priority when making security and risk management decisions.

Infrastructure fails in many different ways with varying consequences. This N-1 contingency test shows that this system can sustain the disruption of Corridor 2. However, in our example, the loss of one circuit in Corridor 1 would generate an overuse of the remaining circuit in the corridor and could lead to additional consequences. The N-1 contingency can be mitigated by shedding some of the load to bring the transfer capability in Corridor 1 back to 100%, which could avoid problems leading to the N-2 contingency case. The N-2 contingency test, resulting in the total loss of the two circuits in Corridor 1, would cascade to the two other corridors and lead to an overall system failure.

While this section focused on electric power, there are many similar nuances associated with failures in other infrastructure. For example, within the telecommunications sector, loss of a cellular tower does not necessarily mean that your phone will lose service, the closing of a road does not always mean that you can’t get to your destination, and so on. In other words, infrastructure system failures are not all created equal.

The Need for Prioritization

Without a prioritization process, infrastructure assessment, protection and mitigation programs are typically guided by intuition or expert judgement, and they often do not consider system-level reliability, redundancy, and overall resilience. While understanding how to prioritize high-consequence failure points for assessments and, for protection is essential, the complexity of infrastructure systems can quickly overwhelm decision-makers. For example, in a region with 1,000 electric power assets, almost one million failure scenarios are associated with an N-2 contingency, and nearly one billion failure scenarios are associated with an N-3 contingency (Figure 5). As a result, system operators and government agencies find it technically and financially prohibitive to assess and prepare for all possible disruptions.

Figure 05. Possible Failure Scenarios with an N-3 Contingency for 1,000 Electric Power Assets

Therefore, a primary goal of critical infrastructure protection and resilience programs should be to identify and prioritize critical contingencies affecting infrastructure systems. Achieving this goal will allow decision makers to identify high-impact isolated infrastructure failures, as well as cascading events, and to prioritize protection investments and resilience planning accordingly. Such an approach should also consider infrastructure interdependencies.

Considering Infrastructure Interdependencies

Interdependencies among critical infrastructure assets increase risk to individual assets and the overall system. These interconnected infrastructure components constitute a “system of systems” where the failure of one or multiple infrastructure elements can cascade and affect the resilience of the entire system and ultimately the region. Figure 6 illustrates interdependencies among seven different infrastructure sectors and subsectors.

Figure 06. Critical Infrastructure Interdependencies4

However, as highlighted in the earlier electricity example, simply identifying connections between infrastructure does not provide a sufficient understanding of why or whether a connection is critical to the operational integrity of the system. The following case study of electric power and natural gas interdependencies in Florida further illustrates this point. Because Florida is a terminal state, this case study represents one of the simplest examples of interactions between electric power and natural gas because there is no complex downstream system to consider that could further propagate the disruption. Furthermore, the natural gas system is relatively simple with only two major high-pressure transmission pipelines serving the state (i.e., Florida Gas Transmission Co, and Gulfstream Natural Gas System). Figure 07. Cascading Failure Simulation in Florida shows the results of the cascading failure simulation between natural gas and electric distribution systems in Florida.

Figure 07. Cascading Failure Simulation in Florida

The scenario postulates the occurrence of a guillotine (i.e., complete) break on a major interstate transmission pipeline supplying natural gas to the state, resulting in a 100% reduction in the flow of gas through the pipeline. The pipeline break also disrupts fuel delivery to a large number of gas-fired power plants in the state. These power plants would cease operation, leading to a statewide electricity outage with varying load curtailment intensity ranging from 10% to 100%.5

In addition, the scenario assumed that Florida has three small natural gas processing plants located in an area that would experience a 40% percent load curtailment, requiring them to curtail operations temporarily. However, because the combined output from these facilities is small relative to the total load, the associated gas curtailment would have no notable impact on gas customers in Florida.6

As discussed in the previous section, infrastructure failures are not all created equal. When interdependencies are involved, a failure in one infrastructure can cascade to other systems increasing the overall consequences. Therefore, considering interdependencies should be an integral part of critical infrastructure security and resilience programs.

Applying an Optimization Algorithm to Prioritize Infrastructure

Managing risk associated with infrastructure interdependencies requires an understanding of infrastructure failures and, especially in complex urban environments, an ability to prioritize protection and mitigation efforts. Argonne has developed an optimization algorithm for selection and prioritization of infrastructure that runs at the system-level or the interdependent “system of systems-level”. The algorithm can apply to the assessment of any infrastructure system.

The optimization algorithm assumes that the physical behavior of a system (e.g., a power network, gas pipeline, or coupled system) is described by the following optimization problem:

F(d) := minuЄU(d)f(u)


d is the 0-1 vector representing the failures at infrastructure assets,

u is the control(s) that can be manipulated to mitigate disturbances, and

f(u) is a system output metric of interest such as cost, delivered load, or deviations from a target operation.

This problem can be solved by the generalized Benders decomposition method proposed by Salmeron et al. (2009).7 This method solves the master problem maxdЄDF(d) by iteratively approximating the function F(d) with a set of linear inequalities. Set D contains a set of failure scenarios denoted by d. An element of the set D is denoted by d = (d1, d2, …, dn), where an element di of the vector is either 0 or 1 for i = 1, …, n to create a combination of the asset states. For example, d = (0,0,1,0) can model an event in which, out of n = 4 assets, the third asset is disrupted whereas the other assets are not.

The dependence of the control set U(d) on d captures the fact that the control actions available to counteract the disruption might be affected by the disruption d. The control set implicitly captures the network topology and physical laws of an infrastructure system.

Worst-case contingency analysis aims to find a contingency that causes the maximum damage to the system. The worst-case event (denoted by d(1) can be found by solving the optimization problem:

D(1) = argmaxdЄDminuЄU(d)f(u)

The second most damaging event (denoted by d(2)) can be identified by restricting the event set as D\{d(1)} and by solving the problem d(2) = argmaxdЄD\{d(1)}minuЄU(d)f(u). This procedure can be applied recursively to identify the k-th most damaging disturbance. This step is performed by restricting the disturbance set as D\{d(1), d(2), …, d(k-1)}. Our optimization algorithm systematically restricts the disturbance set by iteratively adding the linear inequalities to the worst-case interdiction problem. This approach significantly saves the computational times, as compared with an exhaustive search.

The algorithmic steps are then summarized for identifying the most damaging disturbances as follows:

  1. Create the initial set of disturbances D and the control set U(d) that is dependent on disturbance d Є D. Set k = 1.
  2. Solve the worst-case interdiction problem to find d(k) = armaxdЄDminuЄU(d)f(u).
  3. If k = K, then STOP.
  4. Update the disturbance set in order to exclude the k-th most damaging disturbance d(k).
  5. Update k = k + 1, and go to step 2.

In step 2 of this algorithm, updating the disturbance set (step 4) is also equivalent to adding a linear constraint to the Benders master problem. The optimization algorithm has been implemented in Julia script language, and CPLEX is used to solve the master and subproblems in the generalized Benders decomposition.

Argonne has applied this optimization algorithm to a test system of the California Independent System Operator (CAISO) interconnected with the Western Electricity Coordinating Council (WECC). The test system is obtained from Kim et al. (2017).8 This test system consists of 225 buses, 375 transmission lines, 135 generation units, and 40 loads.9 The algorithm ran to detect the 100 most critical substations in the system. The criticality of substations is measured based on the amount of load lost resulting from the event that a substation is disabled. In this computational test, the objective function f(u) is defined as the amount of load lost. The control set U(d) is defined by a set of constraints for the security-constrained economic dispatch problem as in Kim et al. (2017).10 Note, however, that our algorithmic approach is generic to have a user-defined objective function and additional constraints (e.g., generation cost, repair time of the failure components etc.). Figure 8 shows the results based on the test system.

Figure 8. Result of the Optimization Algorithm for the Test System of CAISO Interconnected with the WECC

In this example, a total of 36 substations resulted in significant load loss and failures; the other substations did not cause any load loss. The optimization algorithm terminated after the detection of zero-load substation failure. Government analysts and infrastructure owners and operators can use this type of information to protect the highest consequence failure points within infrastructure systems.


Protecting critical infrastructure, especially in complex urban areas, should focus on identifying and prioritizing potential failure points that would have the most severe consequences. Applying a technique like this optimization algorithm can inform this prioritization process. For example, the algorithm can identify the highest-consequence failures resulting from a cyber-attack against a specific critical infrastructure system, or identify the most consequential failures affecting complex interdependent infrastructure systems supporting a large urban area, regardless of the cause of disruption. Infrastructure system owners and operators, and government agencies can use results from optimization modeling to identify priority assets for in-depth security and resilience assessments, and to inform investment decisions related to critical infrastructure protection and mitigation.

Argonne is currently refining the optimization algorithm framework described within this paper through the Resilient Infrastructure Initiative, which is funded through Laboratory Directed Research and Development (LDRD) resources.11 The list of critical assets resulting from the optimization algorithm can be analyzed further by infrastructure impact models such as EPfast12 for electric power. Because of the computational complexity of assessing high numbers of infrastructure connections and associated failure scenarios, these studies are performed on Blues, a 350-node, high-performance computing cluster at Argonne.

About the Authors

Duane Verner is the Resilience Analysis Group Leader within the Global Security Sciences Division at Argonne National Laboratory. He oversees staffing and technical assignments, including critical infrastructure vulnerability assessments, modeling, and dependency analyses. He has provided methodology development and project implementation support to the U.S. Department of Homeland Security Regional Resiliency Assessment Program since its inception in 2009. Duane is vice-chair of the National Academies Transportation Research Board’s (TRB) Committee on Critical Transportation Infrastructure Protection and a member of the TRB Military Transportation Committee. He regularly contributes to the international resilience research community through publications and trans-Atlantic collaboration. Prior to his position with Argonne, he was a project manager for a private sector engineering firm in New York City, working in the transportation, homeland security, and defense sectors. He may be reached at dverner@anl.gov

Frédéric Petit is a Research Scientist specializing in critical infrastructure interdependencies and resilience at Argonne National Laboratory. With a background in earth sciences and civil engineering, Dr. Petit has focused on risk management and business continuity since 2002. Dr. Petit leads the development of methodologies for the assessment of preparedness, mitigation, response, recovery, and overall resilience capabilities of facilities, communities, and regions. He also lends his expertise to work on risk, vulnerability and threat analysis of critical infrastructure. Dr. Petit received his PhD from the École Polytechnique de Montreal in Civil Engineering, focusing on vulnerability analysis techniques for critical infrastructure cyber dependencies. Dr. Petit is member of various program committees for conferences, such as the Symposium on Risk Management and Cyber-Informatics (RMCI) and the National Symposium on Resilient Critical Infrastructure. He serves as Regional Director for North America of the International Association of Critical Infrastructure Protection Professionals (IACIPP) and is member of the International Advisory Board for the SmartResilience Project. He may be reached at fpetit@anl.gov

Kibaek Kim is an assistant computational mathematician in the Mathematics and Computer Science Division at Argonne National Laboratory. He holds Ph.D. and M.S. degrees from Northwestern University and a B.S. degree from Inha University in Korea, all in industrial engineering. He currently serves as a reviewer for several peer-reviewed journals, including Operations Research, Mathematical Programming, Computational Optimization and Applications, European Journal of Operational Research, and IEEE Transactions on Power Systems. His research interests are in modeling and parallel algorithms for large-scale optimization problems in applications to network design, planning, and operations. He may be reached at kimk@anl.gov


The work presented in this paper was partially supported by Argonne National Laboratory under U.S. Department of Energy contract number DE-AC02-06CH11357. The submitted manuscript has been created by UChicago Argonne, LLC, Operator of Argonne National Laboratory (“Argonne”). Argonne, a U.S. Department of Energy Office of Science laboratory, is operated under Contract No. DE-AC02-06CH11357. The U.S. Government retains for itself, and others acting on its behalf, a paid-up nonexclusive, irrevocable worldwide license in said article to reproduce, prepare derivative works, distribute copies to the public, and perform publicly and display publicly, by or on behalf of the Government.

If you would like more information regarding this paper, please contact Duane Verner at dverner@anl.gov.


1 The percentages represent the line transfer capabilities.

2 About 5% of power is lost during transmission because of energy dissipated in the conductors and the equipment used for transmission. Thus, from a starting generation capability of 1,520 MW, a maximum of about 1,450 MW of power arrives at the substation. For the purpose of illustration, the example assumes that electric power is divided equally among the transmission circuits that remain operable.

3 For the purposes of illustration, the example assumes that electric power is divided equally among the transmission circuits that remain operable. In a real case, it would be expected that Corridor 2 would operate at higher capacity to compensate.

4 Adapted from J. Phillips, et al. , State Energy Resilience Framework, Argonne National Laboratory, Global Security Sciences Division, (2016) ANL/GSS-16/4, Argonne, Ill, USA, available at https://www.energy.gov/sites/prod/files/2017/01/f34/State%20Energy%20Resilience%20Framework.pdf, accessed February 14, 2017.

5 E. Portante et al. , “Modeling Electric Power and Natural Gas Systems Interdependencies,” The CIP Report, Center for Infrastructure Protection and Homeland Security, George Mason University School of Law, Washington, D.C., USA, May–June 2016, available at http://cip.gmu.edu/2016/06/03/modeling-electric-power-natural-gas-systems-interdependencies/, accessed February 14, 2017.

6 Ibid.

7 J. Salmeron, K. Wood, and R. Baldick, “Worst-Case Interdiction Analysis of Large-Scale Electric Power Grids,” IEEE Transactions on Power Systems 24.1: (2009) 96–104.

8 Kibaek Kim, et al., “Data Centers as Dispatchable Loads to Harness Stranded Power,” IEEE Transactions on Sustainable Energy 8.1 (2017): 208-218.

9 Ibid.

10 Ibid.

11 Argonne Energy and Global Security, undated, Resilient Infrastructure, available at https://www.anl.gov/egs/group/resilient-infrastructure, accessed February 14, 2017.

12 E.C Portante et al., “EPfast: A Model for Simulating Uncontrolled Islanding in Large Power Systems,” Proceedings of the Winter Simulation Conference, 2011 Winter Simulation Conference.

Copyright © 2017 by the author(s). Homeland Security Affairs is an academic journal available free of charge to individuals and institutions. Because the purpose of this publication is the widest possible dissemination of knowledge, copies of this journal and the articles contained herein may be printed or downloaded and redistributed for personal, research or educational purposes free of charge and without permission. Any commercial use of Homeland Security Affairs or the articles published herein is expressly prohibited without the written consent of the copyright holder. The copyright of all articles published in Homeland Security Affairs rests with the author(s) of the article. Homeland Security Affairs is the online journal of the Naval Postgraduate School Center for Homeland Defense and Security (CHDS).

1 thought on “Incorporating Prioritization in Critical Infrastructure Security and Resilience Programs”

  1. Priority acceleration piece is missing. After the Fukishima Nuclear Meltdown, the routine business accelerates to criticality like the airline industry network. So if the network is down a week, the priority stays low then the backlog goes critical and severe consequences. Then after disaster recovery, it returns to routine.

    The same for the telephone service. This can be perceived as routine, but after a disaster every one calls those affected by the disaster to see if they are okay. So it means any critical call get stopped by the bulk traffic. So you have to properly size the telephone service and prioritize the traffic to get the critical calls through. This involves priority acceleration. Should the routine call, like food service, be stopped, the backlog goes critical and must restored. Then after disaster recovery, it returns to routine.

    This hsaj article provides additional insght on topics pertaining to the Argonne model similar to:

    White, Richard. “Towards a Unified Homeland Security Strategy: An Asset Vulnerability Model (AVM) .” Homeland Security Affairs 10, Article 1 (February 2014).  https://www.hsaj.org/articles/254

Leave a Comment

Your email address will not be published.

Scroll to Top