Linking different forms of Resilience Igor Linkov, PhD Team Lead, US Army Engineer Research and Development Center, US Army Corps of Engineers; Adjunct Professor, Carnegie Mellon University; Honorary Professor, University of Southern Denmark [email protected]1-617-233-9869 OECD NAEC, 13 April 2018
50
Embed
Linking different forms of Resilience - OECD.org - OECD · ability to anticipate, prepare for, and . adapt. to changing conditions and . withstand, respond to, and . recover . rapidly
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Linking different forms of Resilience Igor Linkov, PhD Team Lead, US Army Engineer Research and Development Center, US Army Corps of Engineers; Adjunct Professor, Carnegie Mellon University; Honorary Professor, University of Southern Denmark [email protected] 1-617-233-9869 OECD NAEC, 13 April 2018
Information: creation, manipulation and sharing information.
Cognitive: translating, sharing, and acting upon information to enable system management.
Social: interaction, collaboration and self-synchronization between individuals and entities.
Foundation for Systems Resilience
Presenter
Presentation Notes
Command and Control Research Program, Office of the Assistant Secretary of Defense. CCRP domains of warfare inform us about how to organize system components to determine what needs to measured in a system. Organizing a system into these domains helps determine what are the essential components within a system and how do those components interact among themselves and between interconnected systems.
• e.g. the International Transport Forum who
Physical Resilience
• resilience can be strengthened by implementing policies aimed at mitigating both the threats and consequences of severe crises (Economics Department)
Economic and Financial Resilience
• including resilience to climate change - minimising consequences, design for safe failure, proactive management and ideas around
Environmental Resilience
• education, labour markets and social protection systems
Social Resilience
3
Promoting a holistic approach to resilience
OECD/JRC/NIST Initiative
4 Contact: William Hynes, NAEC
Vision for Systems Resilience Real World Model Operations
Management Alternatives
5
Agenda • Risk vs. Resilience
• Terminology • Similarity and differences
• Resilience Quantification • Metrics based • Model Based
• Resilience in Transportation • Problem • Model/results • Economic Implications
• Linking with other domains/Cyber • Questions
6
Security [Robustness]—“the state of being free from threat.”
Resilience—“the capacity to recover quickly from difficulties.”
Definitions by Oxford Dictionary
Risk—“a situation involving exposure to danger [threat].”
8
“Resilience" means the ability to anticipate, prepare for, and adapt to changing conditions and withstand, respond to, and recover rapidly from disruptions.
(vi) Effective immediately, it is the policy of the executive branch to build and
maintain a modern, secure, and more resilient executive branch IT
architecture.
Calls for Increased Resilience
The Whole Picture: System Risk and Resilience
Plan Adapt
Time
Critical Functionality
System Resilien
ce
Risk Analysis
Consequence
Risk
After Linkov et al, Nature Climate Change 2014
• Further investment in risk/security will only yield marginal returns
• Governments and Industry must value and encourage resilience thinking
Buying Down Risk vs Managing Resilience?
After Bostick et al 2018
Presenter
Presentation Notes
Current and past efforts are risk-based, and we’re getting marginal return on investment for buying down risk When we start buying down risk, we get great returns for relatively small effort For most systems, we’re operating at the far right of this graph Might be time to push toward different mechanisms for minimizing loss
Business Interruption and Supply Chain
Share of Business Interruption in Total Property
Claims
How to Quantify Supply Chain Resilience?
After Linkov and Kott, 2018
Presenter
Presentation Notes
Current and past efforts are risk-based, and we’re getting marginal return on investment for buying down risk When we start buying down risk, we get great returns for relatively small effort For most systems, we’re operating at the far right of this graph Might be time to push toward different mechanisms for minimizing loss
Either aggregated as a single, or set of, number(s), or presented as a matrix
Assessment using Stakeholder Values
Use developed resilience metrics to comparatively assess the costs and benefits of different courses of action
14
Results: Project Evaluation • Baseline assessment can be used to evaluate proposed
projects
Prepare Absorb Recover Adapt
Physical 71 16 60 10
Information 63 45 21 18
Cognitive 90 49 38 27
Social 82 54 12 52
43
Prepare Absorb Recover Adapt
Physical +10 +18 +9 +32
Information +8 +17
Cognitive
Social
Prepare Absorb Recover Adapt
Physical
Information +5 +15 +22
Cognitive
Social +3 +12 +21
Prepare Absorb Recover Adapt
Physical 81 34 69 42
Information 71 45 38 18
Cognitive 90 49 38 27
Social 82 54 12 52
Prepare Absorb Recover Adapt
Physical 71 6 60 10
Information 63 50 36 40
Cognitive 90 49 38 27
Social 85 54 24 73
Project 1 Project 2
51 47
*Projects may have (+) or (-) in other matrices
Issues with Using Metrics-Based Approaches to Measure Resilience Lack of Causal Model
Changing environments and circumstances may change correlating factors
Changing business and management plans may change how previously causal factors interact
May not work in circumstances different than under those they were designed for
Network-based Resilience Theory?
System’s critical functionality (K)
Network topology: nodes (𝓝𝓝) and links (𝓛𝓛)
Network adaptive algorithms (𝓒𝓒) defining how nodes’ (links’) properties and parameters change with time
A set of possible damages stakeholders want the network to be resilient against (𝑬𝑬)
𝑅𝑅 = 𝑓𝑓 𝓝𝓝,𝓛𝓛,𝓒𝓒,𝑬𝑬 17
Ganin et al., 2016
Network-based Approaches Resilience can be quantified with a network science approach by considering the different domains as interdependent multiplex networks.
Resilience: Transportation Network
Washington, DC January 20, 2016 1 inch of snow melted and turned into ice. • 767 car accidents. • Hours of traffice delays • Traffic jams took days to disentangle!
Washington, DC 1937
Network Resilience: Introduction
• video
21
Transportation Networks in 40 Cities
40 US Cities with Different Traffic Delays
v. 3, Dec 2017
22
Transportation Networks in 40 Cities
23
Efficiency and Resilience don’t always correlate.
Ganin et al., 2017
Efficiency
Resi
lienc
e
Lack of Resilience: Financial Implications
Input-Output Close analysis of
inter-industry relationships
General Equilibrium Estimate of long-run stability of
the economy allows for analysis of policy decisions
Econometrics Advanced statistical analyses
underpinning the model
Economic Geography Effects of geographic
concentration of labor and industry
Integrated REMI economic modelling approach
Regional Economic Modeling (REMI)
1 Month of 5% Network Disruption: Atlanta
• 770 jobs lost (0.07%) • $125 million 2009 dollars in GDP lost (0.09%) • $66 million current dollars in disposable personal income lost (0.09%)
Resilience Related Delays not equal Efficiency Related Delays
Hours
Cost
of D
elay
($)
Predictable Delay (Low-efficiency)
Unpredictable Delay (Lack of Resilience)
Aggregate Yearly Cost of Travel Delays
Lack
of R
esili
ence
Cos
t (M
illio
n o
f $)
Inefficiency Cost (Millions of $)
Why Bother?: Managing Resilience is Different than Efficiency
Current System
Design to Maximize Efficiency
Design to Maximize Resilience
Military examples Civil examples
A highly networked system is governed by domains of warfare that organize system components and establish a basis for measurement [1].
Modern infrastructure system are dependent on each other. Nodes pertaining to one infrastructure system affect nodes from the others and vice versa.
Illustration by L. Dueñas-Osorio et al [2]. 1. D.S. Alberts and R.E. Hayes. Power to the edge. CCRP, 2005. 2. L. Dueñas-Osorio, A. Kwasinski. Quantification of lifeline system interdependencies after the 27 February 2010 Mw 8.8 Offshore Maule, Chile, Earthquake. Earthquake Spectra, 2012.
Real Networks are Interdependent
Presenter
Presentation Notes
In reality, systems depend on each other. A military example is shown on the left where we have physical, information, and social and cognitive domains. On the right, we present civil infrastructure dependencies.
Random and Scale-free Networks We consider two types of undirected networks: random and scale-free The number of nodes in both networks is 200,000 and the number of links is 510,000 Average degree is 5.1
Links are distributed between nodes with equal probability
Links distribution favors highly connected nodes
Presenter
Presentation Notes
We looked at two typical network topologies. The first one is described in the left panel and the second one is on the right. Each network contained 200,000 nodes and 510,000 links resulting in the average degree of 5.1. By topology here we mostly mean node degree distribution. On the left you may see the so-called random network. It is formed by distributing links at random with equal probability among all the nodes. To the right we have a scale-free topology. Here the degree distribution is much broader. Even though here we still have the same average degree, there are many nodes with small degrees but also some nodes of very large degrees. Such nodes are known as hubs.
Importance of Connectedness 32
Conceptual Model Graph representation
In undirected networks, typically there is a giant connected component (GCC) that fills most of the network – green nodes and links on the panel to the right. In certain infrastructure systems only nodes connected to the GCC can function normally.
Presenter
Presentation Notes
As we discussed earlier it is crucial that a command and control network stays connected during a disruption. This slide illustrates how our abstract graph model can relate to a communication network.
Connecting Two Networks 33
Presenter
Presentation Notes
My final figure displays the agreement between theory and simulations for the fraction of nodes always persistent in the connected part of the network. This figure highlights novel behavior of coupled scale-free networks. It is known that when robustness is concerned such networks perform worse than coupled random ones. In this case we observe, that with respect to stability scale-free topology outperforms the random one.
Connecting Several Networks: Cyber Resilience Domains
Cyber Attacks on Transportation
Washington, DC
After Ganin et al., 2018 (under review)
Resilient
Smart
Top-Down Decision Analysis/Social Science
Bottom-Up Risk Assessment/ Physical Sci
Goal Identification and Problem Framing
-
What are the goals, alternatives, and
constraints?
Decision Model -
What are the criteria and metrics, How do we measure decision-
maker values
Metrics Generation and Alternative Scoring
-
How does each alternative score along our identified criteria and
metrics?
Data Collection -
What are fundamental properties/mechanisms
associated with each alternative?
Physical/Statistical Model -
What is the hazard? What is exposure?
Risk Characterization -
What are the risks relative to a threshold? How do they compare to
other alternatives?
Modeling
Data Collection
Management
Risk-Resilience Integration
Linkov et al., 2014
Sustainability as component of Resilience Sustainability and Resilience as Separate Goals
Additional slides
US Army Engineer Research and Development Center
Field Offices
Laboratories
Over 1000 engineers and scientists 28% PhDs; 43 % MS
Construction Engineering Research Laboratory (Champaign, IL)
Topographic Engineering Center (Alexandria, VA)
Cold Regions Research Engineering Laboratory (Hanover, NH)
2500 Employees
Research Laboratories of the
Corps of Engineers
Risk and Decision Science Team Boston, MA)
Presenter
Presentation Notes
Before we move on to discuss transportation I would like to stress that the results I will discuss today where in no way obtained by me along but instead where the outcome of an intense teamwork. So, I would like to say a few words about our team. The core of our team is based off the US Amy Engineer Research and Development Center or ERDC The center consists of multiple offices and labs and our team originates from Boston, MA What unites us is the common interest in understanding Risk and Resilience in networked systems.
How Extensive are Current US Agency Resilience Plans?
45
Larkin, Fox-Lent, Linkov et al., 2015
Static vs. Dynamic Analysis
FHWA: TranSight is among the “best equipped to estimate productivity impacts”*
• Construction spending • O&M spending • Travel time savings • Emissions savings • Safety improvements • Fuel expenditures • Non-fuel VOCs • Network speed improvements • Access to labor • Access to intermediate inputs
Static Analysis Dynamic Analysis
Network Disruption Model How to allocate links between nodes to improve the system’s response to links disruption and ensure the optimal connectedness of nodes.
Example: Random network with 100 nodes and 257 links
Normal state Random disruption of 70% of the links
Presenter
Presentation Notes
Classical way to look at the robustness in network theory is to simulate a disruption of a set of links and or nodes and see what happens to the largest connected component of the graph. An example to the left illustrates a toy model of a random network with the degree of 5.1. On the right we have removed a fraction of links.
Disruption and Connectedness
48
Disruption of links (left panel) results in a formation of a new giant connected component (right panel)
Disruption of 70% of the links Giant connected component after the disruption (green nodes)
Presenter
Presentation Notes
As a result, the network connected part has been modified. To the right we have shown all the nodes that are now no longer a part of the largest connected component in black. We also removed the links between such nodes for clarity.
Is the Connected Component Stable? 49
As the links disruption is random, in another realization of link disruption different links will be taken out. This stochasticity means that the GCC will be different, even though the size of disruption (70% of links) stays the same.
Giant connected component after the disruption (1st example)
Giant connected component after the disruption (2nd example)
Presenter
Presentation Notes
In addition to the different responses to disruptions we know that the set of nodes which are disconnected is not fixed as links are deactivated at random. The two panels illustrate this point. The question becomes what is the stable connected part of the network. Which nodes are always connected when we hit the graph with multiple disruptions?
Not Knowing Sub-Suppliers Makes it Difficult to Understand Risk and Resilience •Which supply chain is more
risky or less resilient?
If you only know the top 2 layers you may get the wrong answer