Professional Documents
Culture Documents
Seismic Response of Critical Interdependent Networks: Leonardo Due Nas-Osorio, James I. Craig and Barry J. Goodno
Seismic Response of Critical Interdependent Networks: Leonardo Due Nas-Osorio, James I. Craig and Barry J. Goodno
SUMMARY
During the last decade, critical infrastructure systems such as electric power, water distribution, trans-
portation and telecommunications have been stressed by significant natural hazards and human errors
leading to malfunction. Earthquakes, blackouts and satellite failures have evidenced the vulnerability of
these growing networked systems threatening the continuity of the United States’ economy. The potential
for malicious perturbations to these systems is also an important current concern. However, the underlying
susceptibility of these networks to disruptive events is in large part due to the increasingly complex pattern
of interdependencies that tie these civil infrastructures together. This paper mainly investigates the effect
of seismic disruptions on the performance of real interdependent networks. Maintenance-induced mal-
functions and coordinated attacks are also studied for completeness. Several degrees of interconnectedness
are explored. Interdependent network fragility curves are introduced to display the effect of these different
strengths of coupling. Characterization of the performance displayed by the networks is utilized to propose
mitigation actions and study propagation of their effects. It is shown that minimal strategic interventions
propagated through the network can generate a more resilient interdependent entity. Copyright q 2006
John Wiley & Sons, Ltd.
KEY WORDS: interdependent networks; seismic fragility; infrastructure systems; cascading failures;
network functionality; seismic mitigation
1. INTRODUCTION
∗ Correspondence to: Leonardo Dueñas-Osorio, 325473 Georgia Tech Station, Atlanta, GA 30332-1060, 6100 Main
Street, MS-318, Houston, TX 77005-1827, U.S.A.
†
E-mail: leonardo.duenasosorio@ce.gatech.edu
reliable and constant supply of power. However, the event of 2003, besides increasing the awareness
to possible deliberate attacks, pointed out another serious problem: incomplete understanding of
the interdependencies among different infrastructures.
Recent advances have been made to characterize the topological properties (i.e. those that solely
depend upon their geometric layout) of complex independent networks. Some examples include
the western US power grid [3], general communication systems [4], railroad networks [5], Internet
routes [6], and the World Wide Web [7, 8]. Other studies focus on finding optimal flow patterns
within the networks when exposed to external disruptions and constrained by situational factors,
for instance in highway systems [9, 10]. In the context of perturbed systems, characterization of the
seismic response of independent civil infrastructures such as power and water distribution systems,
has followed two major approaches. The first approach relies upon numerical methods such as
Monte Carlo simulation [11–13] while the second approach relies on analytical seismic reliability
frameworks [14]. Although these studies represent important advances in the networks approach
to lifeline analysis, modelling interdependent networks is still an open research field. Awareness
of interdependencies has increased within several academic disciplines and institutions, including
emergency management and governmental organizations [15].
Current modelling attempts to capture network interdependencies include the use of economic
input–output frameworks [16], or complex-adaptive simulations (CAS) [17, 18]. Approaches based
on CAS treat network nodes as agents with location, capacities and memory. This paper addresses
the response of interdependent networks from another perspective. It uses geographical immediacy
as a proxy for interconnectedness, and defines the network performance based upon topological
measures derived from graph theory. Several disruption mechanisms, including earthquakes, random
failures and induced malfunctions are investigated under different network coupling strengths. Also,
recovery of network functionality is investigated based upon optimal intervention actions.
This paper is divided into seven sections. Section 2 contains fundamental definitions from
graph theory, as well as definition of the parameters used to characterize the importance of net-
work elements and their role in cascading failures. Section 3 includes the description of various
parameters for measuring network performance. These parameters depend on either network topol-
ogy or network flow. Section 4 includes the characterization of two simplified real networks for
fragility analysis. The seismic vulnerability of network elements is presented, as well as network
interconnectedness based upon geographical proximity. In Section 5 the effects of network interde-
pendencies on functionality due to seismic hazards are investigated. Random-malfunctioning and
intentional perturbations are also studied. In Section 6 a set of mitigation actions are implemented to
reduce the potential detrimental effects of complex interconnectedness. Finally, Section 7 presents
the main conclusions of this study.
Standard graph theory terminology used throughout the paper is introduced in this section. A set
A = {A1 , . . . , Ak } of disjoint subsets of a set A is a partition of A if A = ik = 1 Ai . A set of
all k-element subsets of A is denoted by [A]k . A graph or network is a pair G = (V, E) of sets
satisfying E ⊆ [V ]2 ; thus, the elements of E are 2-element subsets of V . The elements of V are
the vertices (i.e. nodes) of the graph G, whereas the elements of E are its edges (i.e. links). The
number of vertices of a graph G is its order denoted by |G| or n, and the number of edges is its
size denoted by G or m [19, 20].
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
SEISMIC RESPONSE OF CRITICAL INTERDEPENDENT NETWORKS 287
Several parameters can be calculated to characterize the topology (i.e. geometric configuration)
of a network. Two parameters are introduced to rank order the importance of a particular vertex
v ∈ V (G) for the connectivity of the entire network. This rank ordering is relevant because networks
tend to display avalanches or cascading failures after disruption of high rank vertices. In general,
the cascading phenomenon occurs when a particular element of a network ceases to provide its
intended function—due to either internal or external hazards. The flow traversing that node has
to be redistributed to adjacent nodes, which usually function close to their maximum capacity.
This is due to flow demand increasing faster than the expansion of the infrastructures. Some of
those adjacent nodes may be operating so close to their capacity that the additional inflow from
the redistribution can induce a failure. This process repeats itself until either the network absorbs
the disruption locally, or if the set of initial conditions are all unfavourable, until a large-scale
disruption propagates compromising a significant portion of the network and its functionality.
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
288 L. DUEÑAS-OSORIO, J. I. CRAIG AND B. J. GOODNO
4. INTERDEPENDENT NETWORKS
Two real infrastructure networks are approximated in this study to track their interdependent
response. Networks of electric power P, and water distribution W , are selected as a sample of
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
SEISMIC RESPONSE OF CRITICAL INTERDEPENDENT NETWORKS 289
critical infrastructures for modern society. The water network is simplified so that the distribution
system covers the entire area, but having removed portions of its dense end-user delivery grid. It
retains enough links to reasonably represent arterial water mains and major secondary feeders. The
networks are located in Central US in Shelby County, Tennessee, on the banks of the Mississippi
River [11–13]. The order of the systems is n p = 59 and n w = 49 for the power and water networks,
respectively. The order of the generation node subset g of P and W is n pg = 8, and n wg = 15. The
topology of the power and water networks resembles a two-dimensional grid with quadrilaterals
as fundamental building blocks or network motifs. Ideal square grids tend to be resilient network
layouts for flow distribution because there is no node more important than other nodes. Figure 1(a)
shows the physical layout of the streamlined power grid, and Figure 1(b) shows the architecture
of the streamlined water distribution network.
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
290 L. DUEÑAS-OSORIO, J. I. CRAIG AND B. J. GOODNO
Figure 2. Geographical distribution of electric power service areas. Selected water pumping plants and
water storage tanks are highlighted as service area end-users.
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
SEISMIC RESPONSE OF CRITICAL INTERDEPENDENT NETWORKS 291
within jurisdictions. The population size of the service areas is on average 25 000 people. As an
example, if power substation P25 fails (i.e. power service area No. 25), it conditions with proba-
bility pW j |P25 the failure of pumping station W12 , and storage tanks W7 , W10 , W14 , W15 . Elements
W7 , W10 , and W15 are included because their water comes from the highlighted service area. For
instance, failure of P25 implies that approximately 9% of the county population, according to the
1990 US census data [27], has the potential to experience impaired water delivery service. Reasons
include not only the typical failures in the water distribution system, but also the failures induced
by dependence on faulty power grid elements. Consequences of this scenario event will leave the
water systems incapable of fulfilling basic objectives: meeting water needs, accommodating fire
fighting, and equalizing grid operating pressures.
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
292 L. DUEÑAS-OSORIO, J. I. CRAIG AND B. J. GOODNO
0.8
P(DS > ds = Extensive | PGA)
0.6
0.4
0.2
High voltage substation
Medium voltage substation
Low voltage substation
0
0 0.5 1 1.5
(a) PGA (g)
0.9
0.8
0.7
P(DS > ds = Extensive | PGA)
0.6
0.5
0.4
0.3
0.2
Small pumping plant
0.1 Large pumping plant
Elevated steel tank
0
0 0.5 1 1.5
(b) PGA (g)
Figure 3. Fragility curves of network elements (systems) for an extensive damage limit state: (a) electric
power grid; and (b) potable water distribution networks.
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
SEISMIC RESPONSE OF CRITICAL INTERDEPENDENT NETWORKS 293
represent leaks [11]. The diameter of the pipelines ranges from = 0.15 to 1.22 m in order
to include transmission and distribution links. Rrate is useful to characterize the probability of
having pipeline ruptures since it allows estimation of the mean occurrence rate of breaks. This
characterization is done by introducing a spatial Poisson process to define the probability that the
number of pipe breaks Br equals r within a given pipeline segment length L:
(Rrate ∗ L)r −Rrate ∗L
P(Br = r ) = e (5)
r!
In this study, it is assumed that the occurrence of at least one break will impair the functionality
of a pipeline segment. Therefore, the probability of pipeline break occurrence simplifies to
P(Br >0) = 1 − P(Br = 0) = 1 − e−Rrate ∗L (6)
Equation (6) permits simulation of failure of pipeline intersections given a PGV level of seismic
hazard. These intersection nodes are a fundamental part of the topological characteristics of the
network in the graph theory approach of this study. In essence these junctions lump the incident
pipeline fragilities. Non-functionality or failure, Fv , of a junction, v, is reached when all pipeline
segments of v have at least one break. This is analogous to the probability of failure as prescribed
by the reliability of parallel systems:
P f = P(Fv ) = P(Fv1 ∩ Fv2 ∩ . . . ∩ Fvd(v) ) (7)
1
d(G) = 2, <L> = 10 km
d(G) = 4, <L> = 10 km
d(G) = 2, <L> = 1 km
d(G) = 4, <L> = 1 km
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1 1.2
PGV (m/s)
Figure 4. Fragility curves for brittle pipeline junctions as a function of PGV, average vertex degree d(G),
and average pipeline segment lengths L.
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
294 L. DUEÑAS-OSORIO, J. I. CRAIG AND B. J. GOODNO
where the vertex degree, d(v), corresponds to the maximum number of incident segments of the
intersection under consideration. This relationship, along with the fragility curves of Figure 3, fully
characterize the vulnerability of the constitutive elements of the introduced networks. Samples of
generic pipeline junction fragility curves for a non-functional junction damage state DS (i.e. at least
one break per pipeline segment) are presented in Figure 4. Typical main water distribution system
parameters (e.g. length and vertex degree) are utilized to develop their fragility. The network used
in this study has a vertex degree d(v) that ranges from 1 to 5, with an average of d(G) = 3, where
G is an entire graph. Also, pipeline segment lengths, L v j , consistent with Figure 1, vary from 1
to 15 km, with an average L = 6.9 km.
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
SEISMIC RESPONSE OF CRITICAL INTERDEPENDENT NETWORKS 295
Figure 5. Hazard contours for most significant event of de-aggregated probabilistic seismic hazard in
Shelby County, Tennessee, U.S.A. Plots are presented in: (a) PGA; and (b) PGV.
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
296 L. DUEÑAS-OSORIO, J. I. CRAIG AND B. J. GOODNO
0.8
Power
Water
0.2
0
0 0.5 1
Connectivity Loss
Figure 6. Complement of interdependent CDFs for the water and power connectivity
loss CL at different PGA levels.
The objective here is to understand the fundamentals of infrastructure fragilities and their interde-
pendent effect—across networks and not within them.
The probability of failure of a particular network element given a seismic hazard intensity level
is then compared with a uniform random variable X ∼ U (0, 1). If the value of X is larger than the
probability of failure of the element, then the element is removed. This is done for all elements
at each hazard level and repeated 10 000 times to determine the distribution of the response of
the networks. The performance of the network after each trial is measured by the connectivity
loss, CL —which is a proxy for electric circuit modelling and mass and energy conservation. The
complement of the cumulative distribution is displayed in Figure 6, where each point of a hazard
level comes from integrating the distribution of the response, CL , obtained from the Monte Carlo
simulation. After all trials are completed for a particular hazard level, the proportion of trials
that exceed certain network damage state, DS (e.g. DS>20%CL , DS>50%CL , or DS>80%CL )
is the limiting probability of network failure given a seismic intensity value PGA or PGV. This
process is performed at several levels of PGA and PGV (e.g. reflecting various hazard rates)
allowing construction of infrastructure fragility curves. In this study the response data from the
simulations leads to the development of fragility curves in terms of the CL response metric. This
parameter is chosen because of its compromise between computational complexity and ability to
capture network flow distribution. The worst-case running time as a function of n and m for CL
is O(n 2 m).
Figure 6 presents a revealing plot for the complement of the cumulative distribution functions
of CL . This figure shows that the water network CL displays different behaviour depending on the
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
SEISMIC RESPONSE OF CRITICAL INTERDEPENDENT NETWORKS 297
hazard level (i.e. approximately 10, 5 and 2% probability of exceedance in 50 years). The behaviour
changes from a power law for low PGA hazard levels, to a linear relation for medium PGA levels, to
a logarithmic function for high PGA levels. The effect of interdependencies induces an analogous
behaviour. The curve for the water network at low seismic hazard, but with 100% interdependence
with the power grid or P(W j |Pi ) = 1.0, highlights the potential that interconnectedness has for
inducing extra fragility. The complement CDF for the power CL at low seismic hazard levels
is superimposed in the plot. These observations on water and power networks strengthen the
notion that both network topology and network element vulnerability have a significant role on
the performance of the systems.
The power grid proves to be highly vulnerable, despite being the backbone of modern produc-
tivity. One of the reasons is that electrical substations which are composed of numerous vulnerable
subsystems have a steep fragility relationship. Also, as compared to the water network, the order of
the power grid generation subset n Pg ⊆ G P is smaller and its sparseness is higher. Power demand
and generation have experienced a significant growth in the last 25 years, whereas, transmission
lines (i.e. edges), growing at slower rates, have been forced to be used beyond their original
design capacities. These factors have the effect of reducing the reliability of the power system,
and of increasing the potential for massive blackouts due to transmission overload and subsequent
cascading failures [1, 31].
The distributions of the network performance connectivity loss, CL , for several hazard levels
provide sufficient information to establish network functionality levels. Fragility curves for the net-
works can be established fitting the probabilities of exceeding specific limit states (i.e. functionality
levels) to a lognormal distribution. Connectivity loss levels of CL = 20, 50 and 80% represent three
limiting states to measure the ability of the network to function properly. More precisely, they
quantify the likelihood of the distribution nodes to decrease their capacity to be connected to
generation nodes given a particular seismic hazard. Figure 7 presents the fragility curves for the
potable water and power grid infrastructure systems as a function of the maximum expected PGA
in the area and expressed in term of their connectivity loss, CL .
The fragility of the power grid, and to some extent the fragility of the water network, reflect
the response of a system that undergoes a phase transition: it either functions or not. Under
disturbances, these transmission and distribution systems try to maintain their global functionality
by maximizing the size of the remaining connected portions of the network. As elements fail, this
global functionality becomes more and more dependent on fewer links. Once those few links are
gone, the network suddenly fragments. This phenomenon is also observed in the growth of abstract
random graphs [3].
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
298 L. DUEÑAS-OSORIO, J. I. CRAIG AND B. J. GOODNO
0.8
P(Water DS > dsi | PGA)
0.6
0.4
0.2
20% Connectivity loss
50% Connectivity loss
80% Connectivity loss
0
0 0.5 1 1.5
(a) PGA (g)
0.8
P(Power DS > dsi | PGA)
0.6
0.4
0.2
20% Connectivity loss
50% Connectivity loss
80% Connectivity loss
0
0 0.5 1 1.5
(b) PGA (g)
Figure 7. Infrastructure fragility curves for: (a) the water distribution network; and (b) electric power grid.
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
SEISMIC RESPONSE OF CRITICAL INTERDEPENDENT NETWORKS 299
0.6
0.4
1
P(Water DS > 50% CL | PGA and Pw|p)
0.8
0.6
0.4
1
P(Water DS > 80% CL | PGA and Pw|p)
0.8
0.6
0.4
Figure 8. Interdependent fragility curves for water network at different performance levels:
(a) CL = 20%; (b) CL = 50%; and (c) CL = 80%.
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
300 L. DUEÑAS-OSORIO, J. I. CRAIG AND B. J. GOODNO
1
Random P(Wj|Pi) = P(Wj)
Random P(Wj|Pi) = 0.2
0.4
0.2
0
10-2 10-1 10
0
0.25
Interdependent Effect on CL- Water
0.15
0.1
0.05
0
0 0.2 0.4 0.6 0.8 1
(b) Fraction of Removed Power Grid Elements
Figure 9. Interdependent network response for random-induced malfunctioning and intentional disruptions:
(a) connectivity loss CL ; and (b) interdependent effects on CL .
The plots in Figure 8 introduce a set of interdependent fragility curves. They display the effect that
seismic-induced failures on the power grid have on the water distribution network. Four different
values of the strength of coupling are considered (i.e. magnitude of the interdependence). These
values range from independence where P(W j |Pi ) = P(W j ) to complete interdependence where
P(W j |Pi ) = 1.0. Figure 8(a) shows the interdependent effect for a connectivity loss, CL = 20%.
Other performance levels are presented in Figure 8(b) for CL = 50%, and in Figure 8(c) for
CL = 80%.
These plots clearly show the increase in system fragility as the strength of coupling grows
larger. This effect is amplified as the monitored network performance level, CL , increases. For
P(W j |Pi ) = 0.20 the median value of the fragility distribution decreases approximately 60% with
respect to the median of the independent system at CL = 20%. The decrements in the median are
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
SEISMIC RESPONSE OF CRITICAL INTERDEPENDENT NETWORKS 301
65 and 70% for the damage states CL = 50 and 80%, respectively. The maximum interdependent
effect is Ie = 100%. This parameter is defined as the absolute difference between the independent
and interdependent response, normalized by the maximum independent response. Accounting for
interdependencies increases the complexity of the systems. This adds new combinations to the
possible failure modes inducing a substantial increase to the overall infrastructure fragility.
Seismic hazards are extreme events that expose the potential for detrimental interdependent
effects. However, simple random malfunctioning can also reveal that interconnectedness is able to
induce unforeseen responses. Power and water networks are more often subjected to other hazards
such as lightning, vehicle collisions, falling trees, equipment failure, or vandalism. For those
cases, network performance can be measured and the effects of their interconnectedness quantified.
Figure 9(a) displays the connectivity loss CL of the water network as its elements and the power grid
elements fail at random. Hence, this plot is a function of the fraction of randomly removed nodes
of the power grid. Different levels of interconnectedness are considered. The strength of coupling
accelerates the loss of connectivity between distribution and generation nodes. Figure 9(b) presents
the interdependent effects of the power grid failures on the water network functionality in terms of
its connectivity loss, CL . Results are shown for the cases where both networks experience random
failures, and where the water experiences random failures while the power grid undergoes vertex
degree induced malfunctions. The effect of the interdependencies on the response is significant for
removal fractions less than 40%. The magnitude of the interdependent effect at its peak amounts
for 22% of the independent response. This is a non-negligible percentage, and it occurs where the
fraction of removals is low. This is precisely the range in which the networks operate and where
common failures generate element malfunctions—which are a small number as compared to the
order n of entire networks.
An interesting outcome of the interdependent effect plot comes from the targeted removals in
the power grid (i.e. vertex degree based). Picking the most connected nodes of this system appears
to cause no significant interdependent effects on the water network. This is because the most
connected nodes of the power grid are not necessarily the nodes that facilitate the interaction with
other systems. At the interacting interface, the water network depends on power grid nodes which
are not necessarily highly connected.
6. MITIGATION ACTIONS
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
302 L. DUEÑAS-OSORIO, J. I. CRAIG AND B. J. GOODNO
starts with local modification of network topology by adding bypasses or alternative routes around
congested nodes. The objective is to induce a less vulnerable flow regime in network performance
as measured by the connectivity loss, CL . The location for re-meshing the network is obtained from
the vertex betweenness, Bv , rank ordering. This sorting criterion assesses the amount of load or
total flow that passes through the nodes at an unperturbed network state. The vertices to select for
lessening their congestion are those where more than 65% of all possible flow paths pass through.
Flow paths go from the generation set, g, to the distribution set, d. For the power grid, being one
of the core enabling systems for network interdependencies, its intact flow state indicates that ele-
ments P2 , P5 , and P6 have the highest betweenness. Hence, these nodes are selected for remedial
actions. The fundamental premise is to create bypasses around the most heavily loaded nodes to
decongest them. The new edges used for the bypass and subsequent decongestion are constrained
in their length by edge expansion limitations. Growth of the US high voltage transmission lines
for the 2002–2011 period is projected to be 5% of the existing line-miles, despite the growth in
demand of 20% [34]. Approximately 17% of the transmission lines of the power grid of this study
are high-voltage lines (e.g. >350 kV). This leads to exploration of two network growth scenarios
(i.e. bypasses), consistent with observed power grid growth trends. The first scenario corresponds
to approximately 10 km (i.e. 5% growth) of new lines around element P2 and partially around P5 ;
and the second scenario is approximately 40 km (i.e. 20% growth) of new lines around elements
P2 , P5 , and P6 .
Local mitigation actions increase the redundancy of networked systems. This is because new
edges around congested vertices introduce additional cycles, which diversify the path portfolio
used in flow transfer. To test this observation the network is subjected to a seismic hazard of 10%
probability of exceedance in 50 years. Figure 10(a) presents the impact of bypasses on the power
grid for two transmission line growth levels. The response improves but not as significantly as
expected. The steep fragility curves of the power grid have the explanation: for the seismic hazard
under consideration (i.e. 475 year return period, which induces a PGA ∼ 0.30 g), the vulnerability
of electric substations prevents new edges to be more effective, because failure occurs prematurely
at most network nodes. This calls for seismic retrofit of vertices.
The effect of seismic retrofit is reported in the same figure. Improvements are introduced
to the three identified vertices P2 , P5 , and P6 , which are high voltage electrical substations.
The fragility of an electrical substation can be reduced by providing adequate anchoring of
its various subelements. It is assumed that the mean and the standard deviation of the LN
(PGA) increase by 15% after anchoring of their subsystems. This results in a new median
PGAm = 0.25, and a new dispersion = 0.4, as compared to the original PGAm = 0.20, and
= 0.35 of Figure 7(a). Network improvements are more significant when seismic retrofit is
implemented. However, this result provides an unfair view on the significance of node bypass-
ing. For more modest disruptions such as lighting, vandalism, or aging, the nodes involved in
the disruption can be effectively avoided through the new bypasses. They enhance local absorp-
tion of the perturbation. They contain it in a space that does not compromise large portions of
the system.
Power network growth for the seismic perturbation under consideration is not sufficient to
propagate visible benefits to the water network (Figure 10(b)). The performance of the water
system is also reported when the interdependency is set to be P(W j |Pi ) = 0.5. If the power grid
is retrofitted, then the beneficial effects on the water network response are more evident. This
indicates that a good mitigation policy must combine reduction of the vulnerability to external
hazards, and reduction of the internal congestion around funneling nodes.
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
SEISMIC RESPONSE OF CRITICAL INTERDEPENDENT NETWORKS 303
0.8
0.6 0% Growth
5% Growth
20% Growth
0.4 20% + Retrofit
0.2
0
0 0.2 0.4 0.6 0.8 1
(a) Connectivity Loss
Complement CDF of Connectivity Loss
0.8
0% Growth
0.6 5% Growth
20% Growth
20% + Retrofit
Water | 0%
0.4
Water | 20%
Water | 20% + R.
0.2
0
0 0.2 0.4 0.6 0.8 1
(b) Connectivity Loss
Figure 10. Impact of growth and seismic retrofit on: (a) power grid response;
and (b) water network response.
1 h
IROv = ai vi (8)
h i=1
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
304 L. DUEÑAS-OSORIO, J. I. CRAIG AND B. J. GOODNO
where ai is the weighting factor given to each criterion of node importance, and vi is the rank
of node v in each criterion. Sorting the IROv for all v ∈ V (G) in ascending order provides a
simple prioritization list to implement mitigation actions. This strategy is expected to propagate
improvements through the networks as an inverse avalanche in the same way some disruptions can
generate cascading failures after exceeding a particular threshold of stability [35, 36].
7. CONCLUSIONS
Infrastructure systems are entities whose networked nature makes them suitable for mathematical
analysis using concepts and tools from graph theory. One of the first tasks to understand networks
is to measure their properties in terms of global connectivity, local clustering, and overall shape.
These parameters aid in the characterization of networked systems. Their rank ordering ranges
from pure topological metrics to parameters that simultaneously account for topology and flow
patterns.
The response of networked systems is investigated for two types of disruptions: simultaneous
removal of several elements and systematic removal of network elements—one at a time. The
explicit modelling of network interdependencies poses a challenge on the selection of the parameters
that capture the phenomenon. This study proposes a framework to capture the interdependence
among infrastructures in a single entity by using conditional probabilities of failure. This matrix
of conditional probabilities accounts for the location of the interdependency, the direction of the
interaction, and the magnitude of the strength of coupling, pW j |Pi , which can be tuned from
independence to interdependence. Geographical immediacy is the key aspect for defining the
locations where the interfaces of the networks interact.
The response of the networks to widespread initial failure, as induced by earthquakes, shows a
particular type of behaviour as the intensity of the hazard increases. Their observed probabilistic
response resembles a step function. As the hazard increases, there is a critical moment in which some
of the remaining elements have the responsibility of connecting various fragments of the disrupted
network. If they are absent, the entire network collapses. This sudden change in response is observed
in both independent and interdependent networks. The effect of the strength of coupling also induces
a sudden chance in state. The change in the response of the systems between independence and low
interdependence is larger than the change between low strength of interdependence and complete
interdependence. This information is captured in the interdependent fragility curves, which are
introduced as a tool to enhance current loss estimation and risk assessment methods.
For systematic removals, two parameters, vertex degree, d(v), and vertex betweenness, Bv ,
could be utilized to determine the importance of network vertices, and to define the location for
potential initiation of network avalanches. When the removal of the elements of the power grid
focuses on its important elements, then the effect on the water network is smaller than in the case
where the removal of the power grid elements occurs at random. This contradicts the expected
result: if one systematically disrupts a network in the most damaging possible way, one expects
to see larger effects on other networks. The explanation of this contradiction is that the nodes that
are critical to maintaining the integrity of one network are not necessarily the nodes that maintain
the flow at the interface between networks.
The insights gained about response distributions, the importance of elements, and the interde-
pendent effects can be utilized to devise effective mitigation actions. These minimal interventions
have the potential to induce significant improvements in network response. This study explored
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
SEISMIC RESPONSE OF CRITICAL INTERDEPENDENT NETWORKS 305
the ability of mitigation actions to decrease critical nodes’ congestion and seismic vulnerability.
Both are useful depending on the nature of the disruption. Small disruptions are controlled with
bypasses and large disruptions are contained if their elements are less fragile. The results indicate
that an effective mitigation policy should account for the simultaneous importance of every node
to maintain network connectivity, to facilitate network flow, to be instrumental in flow transfer
between networks at their coupling interfaces, and to reduce overall vulnerability. A simplified
method for interdependent rank ordering is proposed. By identifying critical interdependent net-
work elements, future corrective measures will ensure propagation of benefits during their life
cycle and throughout the same networked medium in which cascading failures propagate.
Finally, the search for mitigation actions also calls for more sophisticated representation of the
transient dynamics of networked systems. Physics-based models and time-dependent simulations
may provide the tools to identify the conditions that lead to cascading failures. Having identified
those conditions, one can take preventive actions which will ultimately ensure no widespread
failure within or across infrastructures.
ACKNOWLEDGEMENTS
The work reported here has been funded in part by the Mid-America Earthquake Center through National
Science Foundation grant EEC-9701785. However, all results, conclusions and findings are solely those
of the authors and do not necessarily reflect the views of the sponsors.
REFERENCES
1. Hauer JF, Dagle JE. White paper on review of recent reliability issues and system events. Consortium for Electric
Reliability Technology Solutions, CERTS. PNNL-13150. United States Department of Energy, 1999.
2. PSOTF. Final report on the 2003 blackout in the United States and Canada. U.S.–Canada Power Systems Outage
Task Force, PSOTF. U.S. Department of Energy and Natural Resources of Canada, 2004.
3. Watts DJ. Small Worlds. Princeton University Press: Princeton, NJ, 1999; 262.
4. Guimera R, Arenas A, Diaz-Guilera A, Giralt F. Dynamical properties of model communication networks.
Physical Review E 2002; 66:026704, 8. DOI: 10.1103/PhysRevE.66.026704.
5. Parongama S, Dasgupta S, Chatterjee A, Sreeram PA, Mukherjee G, Manna SS. Small-world properties of the
Indian railway network. Physical Review E 2003; 67(3):036106, 5. DOI: 10.1103/PhysRevE.67.036106.
6. Zhou S, Mondragón RJ. Accurately modeling the internet topology. Physical Review E 2004; 70(6):066108, 8.
DOI: 10.1103/PhysRevE.70.066108.
7. Albert R, Jeong H, Barabasi AL. Error and attack tolerance of complex networks. Nature 2000; 406:378–382.
DOI: 10.1038/35019019.
8. Broder A, Kumar R, Maghoul F, Raghavan P, Rajagopalan S, Stata R, Tomkins A, Wiener J. Graph structure in
the web. Computer Networks 2000; 33:309–320. DOI: 10.1016/S1389-1286(00)00083-9.
9. Sánchez-Silva M, Daniels M, Lleras G, Patiño D. A transportation network reliability model for the efficient
assignment of resources. Transportation Research Part B 2005; 39:47–63. DOI: 10.1016/j.trb.2004.03.002.
10. Peeta S, Yu JW. A hybrid model for driver choices incorporating en-route attributes and real-time information
effects. Networks and Spatial Economics 2005; 5(1):21–40. DOI: 10.1007/s11067-005-6660-9.
11. Hwang H, Shinozuka M. Seismic performance assessment of water delivery systems. Journal of Infrastructure
Systems 1998; 4(3):118–125.
12. Shinozuka M, Rose A, Eguchi RT. Engineering and Socioeconomic Impacts of Earthquakes. MCEER
Monograph 2. Multidisciplinary Center for Earthquake Engineering Research: Buffalo, New York, 1998.
13. French S, Jia X. Estimating societal impacts of infrastructure damage with GIS. URISA Journal 1997; 9(1):31–43.
14. Li J, He J. A recursive decomposition algorithm for network seismic reliability evaluation. Earthquake Engineering
and Structural Dynamics 2002; 31:1525–1539. DOI: 10.1002/eqe.174.
15. USDHS. The National Strategy for the Physical Protection of Critical Infrastructures and Key Assets. United
States Department of Homeland Security, 2003.
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe
306 L. DUEÑAS-OSORIO, J. I. CRAIG AND B. J. GOODNO
16. Haimes YY, Jiang P. Leontief-based model of risk in complex interconnected infrastructures. Journal of
Infrastructure Systems 2001; 7(1):12.
17. Rinaldi SM, Peerenboom JP, Kelly TK. Critical infrastructure interdependencies. IEEE Control Systems 2001;
21(6):11–25.
18. Amin M. Towards self-healing energy infrastructure systems. IEEE Computer Applications in Power 2001;
14(1):20–28.
19. Diestel R. Graph Theory. Springer: New York, NY, 2000; 312.
20. Bollobás B. Modern Graph Theory. Springer: New York, NY, 1998; 394.
21. Barabási AL. Linked. Plume–Penguin Group: New York, NY, 2003; 294.
22. Watts DJ, Strogatz SH. Collective dynamics of small—world networks. Nature 1998; 393:440–442. DOI:
10.1038/30918.
23. Newman MEJ. Scientific collaboration networks. II. Shortest paths, weighted networks and centrality. Physical
Review E 2001; 64(1):016132, 7. DOI: 10.1103/PhysRevE.64.016132.
24. Albert R, Albert I, Nakarado GL. Structural vulnerability of the North American power grid. Physical Review
E 2004; 69(2):025103, 4. DOI: 10.1103/PhysRevE.69.025103.
25. Holme P, Kim BJ, Yoon CN, Han SK. Attack vulnerability of complex networks. Physical review E 2002;
65:056109, 14. DOI: 10.1103/PhysRevE.65.056109.
26. Newman MEJ. The structure of complex networks. Society for Industrial and Applied Mathematics 2003;
45(2):167–256.
27. USCB. Census 1990. United States Census Bureau, 1990. <http://www.census.gov/main/www/cen1990.html>
28. FEMA. HAZUS-MH technical manual: earthquake model. Multi-hazard Loss Estimation Methodology. United
States Department of Homeland Security, Federal Emergency Management Agency, Washington, DC, 2003.
29. O’Rourke M, Ayala G. Pipeline damage due to wave propagation. Journal of Geotechnical Engineering 1993;
119(9):1490, 9.
30. Harmsen SC, Frankel AD, Petersen MD. Deaggregation of U.S. seismic hazard sources: the 2002 update.
Open-File Report 03-440, United States Geological Survey, 2003.
31. USDOE. The Changing Structure of the Electric Power Industry: An Update. Energy Information Administration,
United States Department of Energy, Washington, DC, 2000.
32. Vanzi I. Structural upgrading strategy for electric power networks under seismic action. Earthquake Engineering
and Structural Dynamics 2000; 29(1053):20.
33. Latora V, Marchiori M. Vulnerability and protection of infrastructure networks. Physical Review E 2005;
71(1):015103, 4. DOI: 10.1103/PhysRevE.71.015103.
34. Bush R. The future of transmission in North America. Transmission and Distribution World 2003; 55(1):7–26.
35. Watts DJ. Six Degrees. Norton & Company, Inc.: New York, NY, 2003; 374.
36. Callaway D, Newman MEJ, Strogatz S, Watts D. Network robustness and fragility: percolation on random graphs.
Physical Review Letters 2000; 85(25):5468–5471.
Copyright q 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2007; 36:285–306
DOI: 10.1002/eqe