This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

Modernizing Distribution
System Restoration to Achieve
Grid Resiliency Against
Extreme Weather Events:
An Integrated Solution
By C h e n C h e n , Member IEEE, J i a n h u i Wa ng , Senior Member IEEE, and D a n Ton

ABSTRACT | Recent severe power outages caused by extreme the grid modernization efforts could benefit distribution system
weather hazards have highlighted the importance and urgency restoration, and we propose an integrated solution in the form
of improving the resilience of the electric power grid. As the of a decision support tool to achieve the goal. The advantages
distribution grids still remain vulnerable to natural disasters, the of the solution include improving situational awareness of the
power industry has focused on methods of restoring distribution system damage status and facilitating survivability for customers.
systems after disasters in an effective and quick manner. The The paper provides a comprehensive review of how the existing
current distribution system restoration practice for utilities methodologies in the literature could be leveraged to achieve the
is mainly based on predetermined priorities and tends to be key advantages. The benefits of the developed system restoration
inefficient and suboptimal, and the lack of situational awareness decision support tool include the optimal and efficient allocation
after the hazard significantly delays the restoration process. of repair crews and resources, the expediting of the restoration
As a result, customers may experience an extended blackout, process, and the reduction of outage durations for customers, in
which causes large economic loss. On the other hand, the
response to severe blackouts due to extreme weather hazards.
emerging advanced devices and technologies enabled through
grid modernization efforts have the potential to improve the
KEYWORDS | Automated feeder switch; distributed generator;
distribution system restoration strategy. However, utilizing
distribution system restoration; extreme weather event; grid
these resources to aid the utilities in better distribution system
modernization; microgrid; resiliency; situational awareness;
restoration decision making in response to extreme weather
survivability
events is a challenging task. Therefore, this paper proposes an
integrated solution: a distribution system restoration decision
support tool designed by leveraging resources developed for I. I N T RODUC T ION
grid modernization. First, we review the current distribution
Recent severe power outages caused by extreme weather
restoration practice and discuss why it is inadequate in
hazards have highlighted the importance and urgency of
response to extreme weather events. Then, we describe how
improving the resilience of the electric power grid. People
will never forget that Superstorm Sandy in 2012 left over
Manuscript received September 2, 2016; revised January 10, 2017; accepted
March 11, 2017. This work was supported by the U.S. Department of Energy Office 8 million customers without power across 15 states and
of Electricity Delivery and Energy Reliability. The submitted manuscript was Washington, DC on the east coast of the United States [1], [2],
created by UChicago Argonne, LLC, Operator of Argonne National Laboratory
(Argonne). Argonne, a U.S. Department of Energy Office of Science laboratory, is and Hurricane Irene in 2011 resulted in more than 6.5 million
operated under Contract No. DE-AC02-06CH11357.
C. Chen is with the Energy Systems Division, Argonne National Laboratory,
people loosing power [1]. A recent Congressional Research
Argonne, IL 60439 USA (e-mail: morningchen@anl.gov) Service study estimates that the inflation-adjusted cost of
J. Wang is with the Department of Electrical Engineering, Southern M ­ ethodist
University, Dallas, TX, USA, and the Energy Systems Division, Argonne National
weather-related outages in the United States is $25–70 bil-
Laboratory, Argonne, IL 60439 USA (email: jianhui.wang@ieee.org) lion annually [3]. Besides the United States, large-scale power
D. Ton is with the U.S. Department of Energy Office of Electricity Delivery and
Energy Reliability, Washington, DC 20585 USA (email: Dan.Ton@hq.doe.gov) outages caused by extreme weather events occurred in other
Digital Object Identifier: 10.1109/JPROC.2017.2684780

0018-9219 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
Proceedings of the IEEE 1
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

countries over the last decade. For example, the most severe
ice storm seen in the last couple of decades with large-scale
power interruptions following the incident were reported in
China during the ice storm of 2008. More than 200 million
people were left without electricity, and the direct costs of
the event were estimated to be more than $2.2 billion [4].
The Cyclone Dagmar of 2011 left about 570 000 customers
(from a total of 3.2 million electricity customers) experienc-
ing a power outage in Finland [5]. In January 2005, a severe
storm swept across Northern Europe, from Ireland to Russia.
More than 500 000 homes were left without power, with Fig. 1. Weather and climate disaster events with losses exceeding
Denmark and Southern Sweden being hit particularly hard. $1 billion in the United States, 1980–2016. (Source: NOAA’s NCEI [12].)
Five nuclear power plants had to be shut down due to saltwa-
ter seeping into electricity distribution plants [6]. Recently,
in September 2016, fierce storms and lightning strikes in climate disasters since 1980 where overall damages/costs
Australia left the entire state of South Australia with 1.6 mil- reached or exceeded $1 billion with adjustment of consumer
lion people without power overnight [7]. price index (CPI), as shown in Fig. 1. The 1980–2015 annual
Extreme weather is the number one cause of electric average is 5.2 events; the annual average for the most recent
power outages in the United States [1]. According to the five years (2011–2015) is 10.8 events. In 2015, there were
analysis by Hines et al. [8], among 933 events causing power ten weather and climate disaster events with losses exceed-
outages from 1984 to 2006, as shown in Table 1, almost ing $1 billion each across the United States, as shown in
44% were weather related. According to the database of grid Fig. 2. In 2016, there were eight weather and climate disas-
disturbance events maintained by the U.S. Department of ter events with losses exceeding $1 billion (as of July 2016),
Energy (DOE) [9], around 78% of the reported 1333 elec- including two flooding events and six severe storm events.
tric grid disruptions from 1992 to 2011 were weather related The electric distribution grids still remain vulnerable
[3]. Regarding Europe, natural disasters have become the to extreme weather events. Fig. 3 shows examples of dis-
primary threats for the continuity of electricity service. tribution-line and substation damage caused by wind and
After the central European floods of 2002, 2010, and 2013, flooding, respectively, during Superstorm Sandy [1], [13].
European winter storms and the heavy snow of 2012, and the On the other hand, customers’ expectations for the continu-
northeast European hurricane of 2011, both the European ity of electricity services have increased with the evolution
Union and its member countries started questioning the of modern society’s reliance on electricity, which creates
security level of their electric power supply [10]. pressure on utilities to enhance the grid resiliency against
The number of outages caused by extreme weather is extreme weather events. While system hardening and resil-
expected to rise as climate change increases the frequency ience investments (vegetation management, underground-
and intensity of hurricanes, blizzards, floods, etc. [1], [11]. ing, elevating substations, etc.) at the planning stage are
According to data provided by the National Centers for very important, system restoration in response to the disas-
Environmental Information (NCEI) of the National Oceanic trous event is also indispensable to achieving resilient elec-
and Atmospheric Administration (NOAA) [12], as of July tric distribution systems.
2016, the United States has sustained 196 weather and

Table 1 Causes of Major Blackouts in the United States [8]

Fig. 2. The location of ten weather and climate disaster events

with losses exceeding $1 billion across the United States in 2015.
(Source: NOAA’s NCEI [12].)

2 Proceedings of the IEEE

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

distribution system restoration practices and discuss their

limitations for restoring extreme-weather-related outages.
Then, we describe the proposed integrated restoration deci-
sion support tool. With the integrated solution, two advan-
tages can be achieved: 1) improving situational awareness
of the system damage status; and 2) facilitating customers’
survivability using local generation sources. The paper also
gives a comprehensive review of how the existing studies
Fig. 3. Examples of vulnerability of distribution system to extreme in the literature could be leveraged to achieve these two
weather events: (a) downed wires during Superstorm Sandy [1]; objectives and pave the way for modernizing distribution
(b) substation flood in the Dumbo area of Brooklyn, NY, USA during system restoration practice in response to extreme weather
Superstorm Sandy [13].
events. The main contribution of this paper is to provide a
framework for an integrated solution and potential method-
Current distribution system restoration is based on pre- ologies, as well as discussion about some open problems on
determined priorities [14]; this approach is not adaptive how we can exploit the benefits from grid modernization
with respect to the status of power grid damage under evolv- efforts to improve distribution system restoration under
ing weather events and available restoration capabilities and extreme weather events for utilities.
resources, and thus tends to be inefficient and suboptimal. Note that although most of the studies are based on
In addition, lack of situational awareness of distribution cases in the United States, including historical hazard data,
grids after the disaster strikes poses great challenges to utili- statistical impacts of the extreme weather events on power
ties’ operators, and greatly delays the restoration process. As grids, and grid modernization efforts, the methodologies
a result, customers may experience extended outages which discussed in the paper are for general cases and have the
cause large economic loss. These challenges come from potential to tackle the challenges of distribution system res-
the unique features of power outages caused by extreme toration under extreme weather in other countries.
weather events, such that the current restoration schemes The remainder of the paper is organized as follows.
that are designed for typical power outages may not be suit- Section II describes the current distribution system restora-
able. In this sense, it is imperative to improve the distribu- tion practices and the corresponding limitations. Section III
tion system restoration strategies for utilities in response to discusses how the grid modernization can be leveraged to
disastrous events. improve distribution restoration and describes the proposed
It has been observed that the grid modernization efforts integrated solution in the form of a distribution system res-
in the power industry have the potential to enhance the toration decision support tool. Section IV reviews the exist-
grid resiliency. In the United States, grid modernization ing methodologies for using weather information for dam-
was initiated by the Department of Energy (DOE) under age assessment, and the multisource data fusion method
the American Recovery and Reinvestment Act of 2009 [15]. to improve situational awareness. Section V reviews exist-
Under the largest program, the Smart Grid Investment ing studies on how to utilize local distributed resources to
Grant (SGIG), DOE, and the electricity industry have jointly enhance survivability for customers. We conclude our dis-
invested $8 billion in 99 cost-shared projects involving more cussion in Section VI.
than 200 participating electric utilities and other organiza-
tions. In the European Union, during the period from 2010 II. OV ERV I E W OF C U R R E N T
to 2020, cumulative investments in smart grid technolo- DIST R IBU T ION S YST E M R E ST OR AT ION
gies will reach 56.6 billion [16]. The investments for grid PR AC T IC E A N D C H A L L E NGE S
modernization are also observed in other countries, includ-
ing China, Japan, South Korea, Australia, and Brazil [16]. A. Current Distribution System Restoration Practice
Among these efforts, distribution automation (DA) and Distribution system restoration is one of the core function-
advanced metering infrastructure (AMI) enable more alities in the distribution management system (DMS) of the
observability and controllability of distribution systems, utilities. After a power outage occurs as a result of a sustained
which can be leveraged to improve the restoration process, fault, the outage management system (OMS) will receive the
and these benefits have been shown through some pilot pro- customers’ trouble call to pinpoint the faulted area and dis-
jects [17]. However, integrating these emerging technolo- patch crews to repair the fault. Before the repair is completed,
gies to provide the utility operators an improved restoration the outage load will be restored by altering the topological
strategy is a challenging task. structure of the distribution grid (i.e., via network reconfigu-
To achieve this goal, this paper presents an integrated ration by controlling the switch devices manually or auto-
solution: a decision support tool that can assist utilities matically) to connect to another feeder and/or another lateral
with distribution system restoration after extreme weather on the same feeder. After the faulted section is repaired, the
events. First, we provide an overview of the current network is reconfigured to the normal state. Fig. 4 illustrates

Proceedings of the IEEE 3

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

1) large transmission lines;

2) substations;
3) public safety calls;
4) main or three-phase backbone feeders;
5) emergency services and facilities critical to public
health and safety;
6) single-phase lines serving large blocks of customers;
7) lines serving neighborhoods and multiple customers;
8) individual customers.
The time needed to restore a distribution system fol-
lowing a major event is highly dependent on a quick and
accurate assessment of system damage [14]. Damage assess-
ment scouts, also called field checkers or spotters, evaluate
storm damage before line crews are dispatched. The role of
a damage assessor is to patrol the feeders to identify trouble
spots, evaluate the extent of the damage, and develop initial
estimates of resources needed for restoration. The assess-
ment generates critical information that helps to define the
scope of the work, prioritize efforts, and assign resources.
The typical damage assessment process is as follows [14]2:
• scores of damage evaluators or damage assessment
Fig. 4. Illustration of distribution service restoration by network teams are sent out to survey all the feeders and taps;
reconfiguration: (a) the fault at Feeder 1 causes a power outage
• assessors record and tally the number of broken
for downstream customers; (b) using network reconfiguration, the
outage load is restored by Feeder 2.
poles, spans of wire down, damaged transformers,
etc., and the location of the damage;
• when teams return to the dispatch location (usually
the restoration process using a simple two-feeder system. This at the end of the day), the information is passed on
scheme, which is called distribution service restoration, can to someone to prioritize the trouble and dispatch
reduce outage duration for customers and enhance the system crews to perform the repairs.
reliability, and is an integral part of fault location, isolation,
and service restoration (FLISR) in the modern DMS [18]. The Once the damage assessment is complete, coordinators
methodologies of the distribution service restoration have been in the control center have a fairly good idea of the size
extensively studied in the literature; they include expert sys- and extent of the damage and the resources required.
tems [19], [20], fuzzy logic [21], [22], multiagent systems [23], Crews are then dispatched to repair damage and restore
[24], heuristic search [25], and optimization [26]. These meth- service on the basis of certain rules and guidance that
ods can effectively isolate the fault and restore as much load as consider various factors (e.g., ease of repair or access to
possible after the general outage occurs in the distribution grid. the damaged area). Besides the utilities’ own personnel,
The utility’s restoration activity in response to storms the U.S. electric industry has developed effective mutual
and other extreme weather events is generally guided by assistance programs, in which utilities call in crews from
predetermined priorities. These provide a high-level guide across a region to help restore downed lines, poles, and
to the order in which systems should be restored, based on transformers [28]. The mutual assistance framework
importance and criticality. According to the survey by the has also been set up in Europe to speed up restoration
Distribution System Testing Application and Research con- after a major power disturbance by facilitating arrange-
sortium, the restoration priorities are very similar across the ments among the members for access to spare parts
utilities, and the general order is listed as follows [14]1: and workforces [27]. The host utility needs to provide
logistical support in the form of lodging, food, and fuel;
1
Note that the restoration priorities are from the survey regarding
must have on hand the necessary spare materials and
the cases for the United States. However, the predetermined priority- equipment; and must have the capability to efficiently
based restoration scheme is a general rule for power distribution grids in dispatch crews to the affected areas [14].
other countries as well. For example, as shown in [27], the recovery
2
measures of the floods in 2014 in Southeastern Europe include the reha- Note that the damage assessment process is based on the survey
bilitation of damaged and destroyed power lines and equipment on a regarding the cases for the United States. But it is a general practice for resto-
priority basis. In another example of severe icing in 2014 in Slovenia, the ration after extreme weather in other countries. For example, during the
fault handling was carried out by working from high to low voltage [27], severe icing event in 2014 in Slovenia, the loss of communications systems
which is also based on a predetermined priority. resulted in manual assessment of the situation after the disaster [27].

4 Proceedings of the IEEE

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

Successful restoration efforts also include effective com- outages. If the distribution system experiences a complete
munications, both internal and external. The internal com- loss of electricity, or some isolated islands are formed as a
munications are essential to manage and coordinate res- result of the extreme weather event, customers may experi-
toration activities, while the external communications are ence extended outages until the faulted areas are repaired
equally important to create public trust and reinforce the and energized again. This process may take a week or
perception of a successful effort [14]; e.g., providing accu- longer, as actually happened during Superstorm Sandy.
rate estimated time of restoration (ETR) will alleviate the This scenario poses new challenges for utilities to achieve
frustration of the customers experiencing the blackout. survivability for customers, which refers to the ability to
maintain some basic level of electrical functionality to indi-
B. Inadequacy of Current Restoration Practice for vidual consumers or communities in the event of a complete
Extreme Weather Events loss of electrical service from the distribution system [28].
Actually, the concept of assisting customers with surviv-
While the current distribution system restoration ability features is relatively new to the electric industry.
schemes work well for typical power outages, the story may Historically, many customers such as hospitals, banks, and
change for extreme weather events. The challenges come data centers have assumed responsibility for their own sur-
from the unique features of power outages due to extreme vivability, relying on backup generators and uninterruptible
weather events, as shown in Table 2 from our previous work power supplies (UPS) [28]. However, owing to modern soci-
[29]. These features are highly related to the characteristics ety’s high dependence on electricity and the rapid evolution
of the extreme weather event. For example, a storm may of customer expectations regarding electricity supply conti-
topple trees at several locations that snap utility poles to
nuity (e.g., for communications and modern conveniences)
cause multiple faults and cause a widespread outage, and
after an extreme weather event, the utilities are facing pres-
these locations depend on the path of the storm. On the
sure to achieve survivability for customers before the com-
contrary, a typical outage is usually caused by one or a few
pletion of repairs on faulted areas.
random faults. Sometimes even bulk power systems suffer
damage as a result of the disaster, which in turn causes out- 2) Lack of Situational Awareness: Lack of situational
ages that conventional restoration mechanisms may not be awareness after disastrous events is another challenge to
able to deal with in an effective manner. In addition, natural distribution system operators, which largely delays the res-
disasters may destroy other infrastructures which are inter- toration process and causes large economic costs to custom-
dependent with power grids (e.g., transportation network, ers. Unlike the transmission systems, most of the current
communications network, gas pipelines) so that the resto- distribution systems are “blind” in terms of monitoring and
ration process will face even more difficulties. Conversely, control capabilities beyond the distribution substation. Even
a typical power outage usually does not suffer from these with some observability enabled by AMI or DA, data after a
issues. These differences will result in the limitations of natural disaster may be unavailable or questionable, since
the current restoration practices in response to extreme the devices as well as the underlying communications net-
weather events, as discussed below. work may also be damaged. To pinpoint the faulted areas,
the current OMS usually depends on customer trouble calls,
1) Degraded Customers’ Survivability: The distribution
which are slow and inaccurate, or even unavailable owing
system restoration strategies designed for typical outages may
to damaged telecommunication systems. The existence of
not be suitable for recovery from natural-disaster-induced
so-called “nested outages,” i.e., an outage within an outage,
Table 2 Differences Between Typical Outages and Natural-Disaster- which are more common in power outages due to extreme
Induced Outages [29] weather events, creates additional difficulties in identifica-
tion of outage areas.
On the other hand, the damage status information is
usually captured by field assessors, which is also a very slow
process, and the crews cannot be dispatched until the dam-
age information is collected and analyzed. Furthermore, as
indicated in the white paper by GTM Research [30], the
separation of the damage assessment system from the OMS
during restoration results in data silos for the two processes,
which impact the ability to achieve situational awareness in
a timely manner. In this sense, utilities have little knowl-
edge about what components/networks are damaged and
how severely they are damaged after natural disasters, and
thus it is difficult for utilities to schedule repair crews and
restore the distribution system effectively and efficiently.

Proceedings of the IEEE 5

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

III. A N I N T EGR AT ED SOLU T ION FOR

DIST R IBU T ION S YST E M R E ST OR AT ION
It has been observed that the progress of grid modernization
efforts in recent years has provided the potential to improve
distribution system resiliency. However, integrating these
emerging technologies and devices during the restoration
process is not an easy task for utilities. In this section, the
grid modernization efforts toward grid resiliency are intro-
duced. Then, an integrated solution to help utilities’ deci-
sion making for distribution system restoration is described.

A. Grid Modernization Efforts Toward Grid

Resiliency
The investments in grid modernization efforts have
been conducted in many countries in the world. In the
United States, as a result of the American Recovery and
Reinvestment Act of 2009, the DOE, in collaboration with
the power industry, launched the SGIG program [31]–[33],
aiming at modernizing electric power grids through deploy-
ment of advanced devices and smart grid techniques. Fig. 5 Fig. 6. Geographical distribution of investments and project
shows the SGIG expenditures by categories of technolo- categories of smart grid technologies in the European Union [16].
gies and systems. In the European Union, the investments
in smart grid technologies are from several funding agen-
cies including the European Recovery Fund, the European high currents or low/no voltage [34]. Fault analysis appli-
Regional Development Fund, and the European Educational cations can utilize the data provided by fault indicators to
Research Association [16]. Fig. 6 shows the geographical achieve improved accuracy in locating and identifying faults
distribution of investments and project categories of smart and their causes. Fig. 7(a) shows a fault indicator designed
grid technologies in the European Union. and manufactured by Schweitzer Engineering Laboratories
Among them, the investments in electric distribution (SEL) [35], and Fig. 7(b) shows the deployment of fault indi-
systems and advanced metering infrastructures provide the cators on overhead distribution lines.
potential to enhance distribution system restoration in the Automated feeder switches are also emerging as a result
following ways in response to extreme weather events. of the development of DA. These switches can open and
close a feeder section in response to control commands,
1) Distribution Automation: The development of DA which provide controllability of the distribution network
enables the deployment of field devices such that the vis-
ibility and controllability of the distribution system behind
the substation is enhanced. One important field device is
the remote fault indicator, which is a low-cost sensor that
can detect electric signatures associated with faults, such as

Fig. 7. Examples of field devices for distribution automation:

(a) overhead fault indicator [35]; (b) deployment of overhead fault
indicator; (c) automated feeder switch [36]; and (d) automatic
Fig. 5. The U.S. DOE Smart Grid Investment Grant program [32]. switch control system [37].

6 Proceedings of the IEEE

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

to isolate faults and reconfigure faulted segments of the B. An Integrated Distribution System Restoration
distribution feeder to restore power. Fig. 7(c) shows Decision Support Tool
the IntelliRupter PulseCloser Fault Interrupter designed Although the new devices and technologies discussed in
and manufactured by S&C Electric Company [36], Section III-A could potentially help the utilities to improve
which integrates fault interruption, fault isolation, the restoration process, it will be a challenge to deter-
and circuit restoration in a single, easy-to-install pack- mine how to best utilize these resources, especially during
age, and Fig. 7(d) shows the corresponding automatic extreme weather events. A decision support tool that inte-
switch control system [37]. For example, in the SGIG grates these resources and technologies will be helpful for
project implemented by the Electric Power Board (EPB) decision making by the utilities’ operators during restora-
in Chattanooga, TN, USA, more than 1400 automated tion. The tool leverages weather information/forecasts and
feeder switches have been installed with the correspond- system fragility assessment to improve utilities’ situational
ing communication and information management [17]. awareness during restoration processes. Advanced opti-
In a July 2012 derecho that affected half of EPB’s cus- mization models with user-defined multiobjective metrics
tomers, EPB’s response was up to 17 h faster because of will be developed to schedule repair crews and restoration
the automated feeder switches, which restored power resources and reconfigure the distribution grids in real time.
to 40 000 customers instantly [17]. The tool is adaptive to system variations, as a feedback loop
With the development of grid modernization, the increas- will be designed for decision updates. The structure of the
ing penetration of distributed generation technologies and
proposed decision support tool includes four modules, as
microgrids provides localized power resources, which may
shown in Fig. 8.
provide huge benefits during restoration in response to
extreme weather events. As recommended in the report by • The data fusion module leverages the existing trou-
GridWise Alliance [2], these distributed energy resources ble call systems, weather information/forecasts, sys-
(DERs) can be leveraged to serve critical loads when the util- tem fragility assessment models, and data from dis-
ity grid is down after extreme weather events. The automated tribution field measurements to improve situational
feeder switches and the associated control mechanism make awareness and estimate and predict distribution
it possible to utilize these DERs to ensure the survivability of system damage status after extreme weather events
the customers and reduce outage sizes and durations. The such as a hurricane, flooding, nor’easter, ice storm,
resilience benefits of the DERs was demonstrated by Sendai or wildfire.
Microgrid (SM) in Japan during the 2011 great east Japan • The crew and resource dispatch module captures
earthquake and tsunami. When the earthquake occurred, the key attributes and constraints of repair crews and
Tohoku Electric Power Company stopped supplying power associated restoration resources, and integrates
to the area surrounding SM, resulting in an almost three-day them into restoration models.
outage. Nevertheless, SM was able to continuously supply • The advanced control module utilizes the resources
power to some small critical loads within the campus and enabled by distribution automation, e.g., automated
provided full heat and power service for almost two and a half feeder switches and DERs, to reconfigure distribu-
of the three blackout days [38]. tion grids and pick up loads.
2) Advanced Metering Infrastructure: The smart meters
deployed in the AMI are equipped with outage notification
capabilities that allow the devices to transmit a “last-gasp”
alert when power to the meter is lost [34]. The information
can be integrated into the OMS to provide additional way
to pinpoint the outage area and help to assess the damage.
Smart meters can also transmit “power on” notifications
to operators when power is restored, or even allow utili-
ties to “ping” meters in the affected areas to assess the out-
age boundary and verify the restoration progress, enabling
field crews to be deployed more efficiently, thus reduc-
ing the restoration time. This benefit was shown in the
SGIG project at PECO in Philadelphia, PA, USA, during
Superstorm Sandy, where smart meter operations helped
PECO avoid more than 6000 truck rolls as power restora-
tion was confirmed by pinging meters without having to Fig. 8. The overall structure of the integrated distribution system
send repair crews [17]. restoration decision support tool.

Proceedings of the IEEE 7

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

• The user-defined multiobjective optimization mod- 2) Facilitating survivability for customers in the event
ule exploits advanced optimization models to coor- of a complete loss of electrical service from the util-
dinate repair crew scheduling, resource allocation, ity grid. The advanced control module will integrate
distribution grid reconfiguration, and load pickup the new devices and technologies enabled by smart
according to different metrics specified by users. grids and distribution automation to reconfigure the
The tool is adaptive to system variations, as a feed- distribution grid and pick up load in an optimal way.
back loop of updated information will be integrated As a result, the DERs can be utilized to serve the
into the decision-making process. critical loads after extreme weather events to reduce
the customers’ outage size and duration.
The benefits of the proposed integrated distribution
system restoration support tool include the optimal and In Section IV, we will discuss how different sources of
efficient allocation of repair crews and resources, the information can be integrated to estimate the system dam-
expediting of the restoration process, and the reduction age status to improve the situational awareness for utilities.
of outage time for customers in response to severe black- Several existing technical approaches will be described,
outs due to extreme weather hazards. Thus, the economic and a data fusion framework will be proposed to integrate
costs of power outages can be reduced. It will further these sources of information. In Section V, we will describe
benefit the whole electricity industry, including utility how the DERs and automated feeder switches can be lever-
companies, customers, independent system operators/ aged to reconfigure the grid and pick up local loads, which
regional transmission organizations (ISOs/RTOs), regu- provide the potential to reduce customers’ outage duration
latory agencies, and vendors. The anticipated benefits and facilitate survivability. Some case studies are provided
to various entities and stakeholders are summarized as to demonstrate the approaches, and some special technical
follows. considerations will be discussed.

• Utilities can use the tool to expedite the restora-

tion process and utilize the distribution automation I V. I M PROV I NG SI T UAT IONA L
devices and technologies to reduce the outage dura- AWA R E N E S S
tions and sizes during extreme weather events. As shown in Fig. 8, the data fusion module utilizes multiple
• Electricity customers will benefit from the reduced sources of information to improve the situational awareness
sizes and durations of power outages during extreme of the damage status. One important source is weather infor-
weather events. mation, which can bring in useful knowledge for estimating
• ISOs/RTOs will be able to use the methodology to the damage to the distribution grid due to natural disasters.
coordinate transmission-level restoration processes The integration of weather information with the customers’
with distribution-level utilities to improve the over- trouble call information as well as the measurements from
all system restoration. field devices has the potential to achieve more accurate sys-
• Regulatory agencies will have a quantitative method tem damage status. A probabilistic data fusion framework is
to evaluate the resilience enhancement of the distri- proposed as one of the possible options to explore the inte-
bution system due to the development of advanced gration of multiple sources of information to improve the
control and optimization technologies. situational awareness.
• Vendors may be able to use the tool to specify the
requirements of certain devices that will be utilized A. Leveraging Weather Information for Damage
in the restoration and enhance the development of
Assessment
outage management systems.
To integrate the weather information into the distribution
This integrated solution for distribution system restora- damage assessment, an important task is to understand how
tion decision making has two key advantages compared to different weather metrics impact the distribution grid; several
the existing solutions. existing studies address this issue. In general, the method-
ologies can be divided into two categories: component-level
1) Improving situational awareness of system dam- damage assessment and system-level damage assessment.
age status. The data fusion module in the tool will
integrate multiple sources of information, includ- 1) Component-Level Damage Assessment: The compo-
ing weather information/forecasts, system fragility nent-level damage assessment method aims to estimate how
assessment, customers’ trouble calls, smart meters, the weather impacts each individual component of the distri-
and distribution SCADA (protective relays, fault bution grid. It applies a probabilistic damage model, which
indicators, etc.) to better estimate the damage sta- uses damage/fragility curves to represent the probability
tus after the extreme weather event. As a result, the of damage as a function of weather metrics such as wind
process of restoration can be faster. speeds and flood water depth. For example, the fragility of a

8 Proceedings of the IEEE

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

distribution line exposed to extreme winds is determined by (FEMA) HAZUS Hurricane Model [42]. This method
the failure of at least one of its support structures, i.e., dis- considers substations as a single unit instead of employ-
tribution poles. Based on the observations made by Quanta ing fragility curves for individual substation elements as
Technology [39], the failure probability of an existing distri- approximation.
bution support structure can be approximated by an expo- Flooding is a major concern for substations. Flooding
nential function using historical data, as follows: becomes a problem for substations when the amount of water
exceeds the capacity of the drainage network. Flooding can
​​λp​  ​​ = a ⋅ exp (b ⋅ ​w​s​​)​ (1) cause severe damage to substation equipment, and may lead
to interruptions in service continuity and widespread out-
where λ​​ p​  ​​​is the pole failure rate, w​
​​ s​​​is the sustained wind ages. The approach of estimating the impacts of flooding on a
speed, and ​a​and ​b​are tuning parameters based on histori- distribution substation is based on the metric of water depth
cal data. Using historical data from Florida Power & Light in flooding, which dictates whether a substation remains in
results in ​a = 0 . 0001​and ​b = 0 . 0421​ [39]. Besides the operation. The expected loss of functionality for substations
empirical methods using historical data, other methods from inundation is determined on a site-specific basis as a
based on physical models are also available. For example, function of the depth of flooding. The functionality thresh-
Winkler et al. [40] used wind force and maximum rated line olds and damage functions will be obtained from the FEMA
perpendicular stress resistance to approximate the probabil- HAZUS Flood Model for each type of substation [43]; exam-
ity of line failure. The wind force on line ​k​, ​​F​wind,k​​​, is calcu- ple information for low-voltage substations is shown in Fig. 9.
lated with the standard American Society of Civil Engineers The severe weather forecasts can be combined with
(ASCE) design equation using wind gust speed (​​w​s​​​) and line geographic information system (GIS) data on electric dis-
cross-sectional area ​​A​c​​​ tribution components to determine the projected weather
conditions at a specific location. Using the damage curves,
​​F​wind,k​​ = Q ​kz​ ​​ ​IFW ​ ​​ ​Cf​​​ ​A​c​​ ​w​  2s​  ​​
​ ​​ ​GWRF (2)
the potential damage and availability of these grid compo-
where other parameters in (2) accounting for air density nents would be determined specifically for a given weather
(​Q​), terrain correction (​​k​z​​​), hazard importance (​​IFW
​ ​​​), wire threat, coupled with the forecasted weather conditions.
strain (​​GWRF
​ ​​​), and drag coefficient (​​C​f​​​) are defined by the The historical and forecasting weather data are available in
ASCE Engineering Practice Report 113 [41]. Line damage shapefile format (a popular geospatial vector data format for
probability is calculated as the ratio of the maximum per- GIS) from the National Weather Service (NWS) [44] and
pendicular force that the line can endure (​​F​brk,k​​​) and the NOAA [45] for the following weather metrics:
line wind loading (​​F​wind,k​​​) for a distribution line ​k​ • precipitation;
​Fw​ ind,k​​  (​ws​​​) • temperature;
( )
​​Pr​wind,k​​ = min​ γ  ​ ________
​Fb​​​ rk, k
​, 1.0 ​​ (3) • depth of flood water;
• hurricane and storm surge.
where factor ​γ​is used to scale the line fragility estimates to
match recorded failure data. The distribution grid suffers Besides these publicly available weather data, advanced
from significant vulnerability to damage caused by flying weather forecasting models, e.g., the Weather Research
debris, as the distribution lines are close to the ground and and Forecasting (WRF) Model [46], can also be utilized to
often in close proximity to trees. Tree windthrow probabil- obtain weather forecasting with finer granularity.
ity therefore affects the fragility of the distribution lines sig-
nificantly. The study in [40] also discussed the flying debris 2) System-Level Damage Assessment: Unlike the com-
models and how they will affect the distribution line failure ponent-level damage assessment methods that investigate
probability.
The damage probability for substations in terms of wind-
storms can be represented via log-normal fragility curves
[40]. These curves generate the probability of damage for
a given wind gust speed (​​w​s)​​​ while taking into account the
local terrain and structural characteristics of the substation
under consideration. The general form of the fragility curves
is shown as

( )​dx​
2
−∞ − ​( ln x − μ)​​  ​
1  ​  exp​ ​ _________
​Pr(damage| ​w​s​​​=​x)  =​​∫ ​  ​  ​____
​​​​  ___  ​ (4)
x ​√ 2π ​  σ 2
2 ​σ​​  ​
where parameters μ​ ​and ​σ​represent the logarithmic mean
and standard deviation of the fragility curve, which can be
taken from the Federal Emergency Management Agency Fig. 9. Damage curve for substations in terms of flood water depth.

Proceedings of the IEEE 9

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

the individual component failure risk, system-level damage values at that node. This process is continued until each end
assessment methods use characteristics of the overall sys- node reaches purity (SSE = 0) or until a minimum number
tem, the hazard, and the area to estimate the failure risk. of records are reached at a given node. Another tree-based
Multivariate regression-based statistical models are usually method is Bayesian additive regression trees (BART). The
applied for this type of approach. Guikema et al. [47] pre- BART model consists of a large number of small trees with
sent several statistical models developed in previous work each tree constrained by a prior to restrict each tree’s con-
[48]–[51], as well as tree-based data mining models, for esti- tribution to the final model, making each individual tree
mating the number of poles that will need to be replaced for a “weak learner.” Fit and inference in BART are achieved
an approaching hurricane. through a Markov chain Monte Carlo algorithm [29], [52].
The generalized linear model (GLM) is a generaliza- To evaluate the predictive accuracy, a holdout valida-
tion of a standard linear regression that allows regression tion analysis is applied. The data set will be partitioned into
analysis of count data. A GLM consists of three components: training and validation sets (e.g., 80% of the data is the
1) a conditional distribution for the count events given the training set and 20% of the data is the validation set as a
parameter(s) of the distribution; 2) a “link” equation relat- general rule). The mean absolute error (MAE) and mean
ing the distribution parameter(s) to a function of the explan- squared error (MSE) are employed for the validation and
atory variables; and 3) an equation specifying the function comparison
of the explanatory variables to be used in the link function n
[47]. For example, a negative binomial GLM to model dis- ​  n1  ​ ​ ∑ ​|​  ​ ​​y ​​î ​​  − ​y​i​​  |,​
​MAE = __ (8)
i=1
tribution system damage in grid cell i​​can be described as
follows: n
1
​MSE = ​ __ 2
−1 n ​ ​ ∑ ​​(​  ​​y ​​î ​​  − ​y​i​​)​​  ​​ .​ (9)

​ ) Γ(​α​​  −1​) ( α ​ ​​  −1​  + ​λi​  ​​ )

Γ(​yi​​​  + α)
_____________ ​α​​  −1​  ​ ​​​ 
​α​​  ​ i=1
​​f​Y​​  (​yi​​​  |α, ​λi​  ​​)~ ​   ​  ⋅ ​​ ​ _____ ​
Γ(​yi​​​​+1 The comparison of accuracy of various statistical-based
methods to predict the system damage is performed in refer-
​yi​​​ ence [47], where MAE and MSE based on detailed pole-level
( α​ ​​  ​  + ​λi​  ​​ )
​ ​  ​​
λ
⋅ ​​ _____
  ​  −1 i  ​ ​​​  ​​ (5) damage data on the basis of 150 random holdout samples
are shown. The results show that no statistical method can
achieve 100% accuracy. Certain methods are superior to
​log (​λi​  ​​) = β​ 0​  ​​  + ​∑​ ​​βj​  ​​​ ​xij​ ​​​ (6)
j others in terms of MAE and MSE in these case studies, but
where ​​yi​​​​denotes the number of damages (count data) we cannot derive a general conclusion on which method is
in grid cell ​i​, ​​λi​  ​​​is the parameter of the negative binomial the best since the result may be different for another geo-
distribution, the vector ​​ x​i​​ = ​[ ​x​i1​​, ​xi2
​ ​​, …, ​xin
​ ​​  ]​​  ′​​ represents graphic location with a different set of historical data.
the explanatory variables of size n​ ​in grid cell i​​, the vector The statistical models highly rely on the historical data
​β = ​[ ​β0​  ​​, ​β1​  ​​, …, ​βn​  ​​  ]​​  ′​​is the regression parameters to be esti- available. For the damage assessment, the distribution sys-
mated, and ​α​is the overdispersion parameter of the nega- tem damage data set (e.g., pole replacement data) is the
tive binomial distribution that is observed in power system response variable. The explanatory data required for statisti-
performance data. cal models include power-system-related data and environ-
The generalized additive model (GAM) allows a non- mental data. The former consists of the topology of the grid,
linear relation between the parameters of the conditional the number and availability of protective devices, the loca-
distribution and the covariates [47]. For example, the tion and number of customers, etc. Environmental data may
link function shown in (6) can be changed to represent a vary with the disaster scenario. Examples of such parameters
nonlinear smoothing function. and available sources of the data are listed as follows [29]:
• land and geometric characteristics of the area such
​log (​λ i​  ​​) = β​ 0​  ​​  + ​∑​ ​s​  (​x​ij​​) .​ (7) as land use and land cover data, soil moisture lev-
i
els, elevation characteristics, land slopes, and com-
The tree-based data mining methods represent the relation-
pound topographic index [53], [54];
ship between the response variable of interest (e.g., the
• disaster variables such as hurricane duration and
number of damaged poles) and the explanatory variables
intensity, approaching angle, landfall position, and
through the use of recursive binary partitioning of the data
translation velocity [55];
set. For example, the classification and regression trees
• climate characteristics, such as standardized precip-
(CART) method used in [47] develops a single tree to cap-
itation index (SPI) and annual and monthly precipi-
ture the relationship between the response variable and the
tation [56].
explanatory variables. In building the regression trees, the
algorithm chooses the variable to split the data set and the Note that statistical models in the literature are based on
value of that variable at which to split the data to maximize the stationarity assumption (i.e., power system components
the decrease in the sum of squared errors (SSE) of the fitted stay unchanged over the time horizon). However, there will

10 Proceedings of the IEEE

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

be upgrading of the system components over time, espe- and circuit breakers to identify the most probable faulted
cially due to the grid modernization efforts. One method to section of the distribution grid, e.g., the methods in [58] and
tackle this issue is to include the power system component [59] based on expert systems, and the method in [60] based
change information in explanatory variables in statistical on overcurrent direction and network structure. Recent work
models, so that the inference results can capture the effects [61] proposed an automatic and fast faulted line section iden-
of system change over time. tification based on the data from fault indicators, which have
The system-level assessment method can also be utilized been widely installed in distribution systems because of their
by utilities to achieve efficient and effective emergency lower cost. The information can be further leveraged to esti-
response planning in advance of an approaching disaster. mate the fault location once the faulted line section is identi-
The assessment can be used to inform the public about the fied, as discussed in, e.g., [62]–[64], and references therein.
estimation of the scale of the outage and duration of the res- In addition, the continuous measurements from field devices
toration, which is also an important task for utilities before can be exploited to identify the faulted segment. For example,
and during restoration. the method in [65] uses current magnitude to identify mul-
tiple faulted segments and further uses voltage sag to reduce
the candidate faulted segments. However, in the event of dis-
B. Other Sources of Information for Damage
asters, the data from these field devices may become unavail-
Assessment
able owing to damage to the devices or the underlying com-
The system damage status can be estimated utilizing munication system (i.e., SCADA), so that fault identification
other sources besides the weather information, includ- by these methods, based on incomplete data, may provide
ing the traditional customer trouble calls, data from fault- ambiguous results. In addition, these methods rely on the
indicating devices enabled by DA for feeder information, electric signature at the time of the fault, so further damage to
and smart meter data enabled via AMI on the customers’ the outage/faulted area cannot be identified or located.
end. Each source of these data will provide information on
system outage, fault, and damage from different temporal 3) Data From Smart Meters in AMI: With the development
and spatial dimensions; thus, integrating them can achieve of AMI, smart meters are capable of outage notifications via
improved situational awareness of the system status after underlying two-way communications systems. These noti-
extreme weather events. fications provide useful information to pinpoint the outage
and damage from the customers’ end. Many studies appear
1) Customer Trouble Calls: Traditionally, distribution grids in the literature on identifying power outages using smart
have little observability and controllability beyond the substa- meters. Chen et al. [66] proposed a fuzzy-petri-nets-based
tions. When a power outage occurs, utilities rely on trouble method to detect outage events using AMI data. Liu and
calls from customers to pinpoint the outage area. The OMS Schulz [67] designed a knowledge-based system to locate
will match the phone number from each call to the specific distribution system outages utilizing wireless automated
customer locations, so that the distribution transformer and meter reading systems. Sridharan and Schulz [68] proposed
protective devices on the corresponding feeder can be identi- probabilistic and fuzzy model-based filter algorithms to pro-
fied. When sufficient trouble calls are collected, the OMS pre- cess outage data from automated metering system. Smart
dicts the tripped protective devices and faulted line sections. meters could also help with identifying the fault location
Through call center automation techniques, including inter- using their voltage measurements [69], and during the res-
active voice response (IVR) and automated callbacks [30], the toration process, smart meters could provide verification
collection of the trouble calls could be expedited. Laverty and and confirmation of the restoration [70]. However, like the
Schulz [57] proposed a topology-based algorithm that associ- distribution SCADA system, the underlying communication
ates trouble calls with each other to provide more accurate network for AMI is vulnerable to natural disasters, leading
information about outages during heat storms. However, the to the availability issue for the smart meter data.
accuracy of the method based on trouble calls relies on the
availability of trouble calls from customers, which is an issue 4) Data From μ​ ​PMU: The microphasor measurement units
during natural disasters, since they could also impact the tel- (​μ​PMU), a new synchrophasor measurement device designed
ecommunication systems. for distribution systems and funded by DOE Advanced
Research Project Agency—Energy (ARPA—E) [71], could
2) Data From Fault-Indicating Devices: The data from fault- provide more accurate and finer-granularity measurements to
indicating devices could capture the electric signatures asso- monitor the distribution grid status. Fig. 10 shows the μ​ ​PMU
ciated with faults and the control center could get the data developed by the Power Standards Lab (PSL). A scoping study
through the distribution SCADA. The fault-indicating devices by Lawrence Berkeley National Laboratory [72] shows the
generally include circuit breakers (CBs), feeder terminal effectiveness of ​μ​PMUs for distribution system restoration.
units (FTUs), fault indicators (FIs), etc. There are some exist- With deployment of more μ​ ​PMUs in distribution systems, the
ing studies on how to utilize information on the tripped relay situational awareness could be further improved.

Proceedings of the IEEE 11

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

elements, as shown in Fig. 11. These are: 1) fragility assess-

ment that correlates various weather conditions with the
impacts on functionalities of individual grid components;
2) a feeder-level situational assessment model that deter-
mines or infers the states of grid components based on
field measurements; 3) a customer-level outage assessment
model based on customers’ trouble calls and smart meter
data in OMS to identify outage areas; 4) an automated inter-
face that a) synthesizes the outcomes of the fragility curves
Fig. 10. μ
​ ​PMU by the Power Standards Lab [73]: (a) PQube 3; (b) and the situational assessment model to determine the state
outdoor pole mount with the wireless 4G modem. of major components in the grid and b) updates the compo-
nent models in the grid according to the component states
to represent their operation conditions; and 5) a grid analy-
5) Data From Social Media: In addition to these field
devices, with customers actively engaged in social media sis model that contains models of grid components includ-
such as Facebook and Twitter, the ever-growing digital ing generation and load as well as geological information of
media network can be considered as social sensors to iden- the equipment and topology of the grid such that the grid
tify the location and extent of an outage without adding status can be determined from the component states. In the
new measurement and communication instruments [74]. data fusion module, an inferring scheme will be developed
The usefulness of social media in outage detection has been to synthesize the available measurements for situational
recognized by the power industry [75]. Sun et al. [74] pro- assessment. To build the analytical grid model, a GIS-based
pose a probabilistic framework for the detection and loca- distribution power flow model is preferred to provide con-
tion of power outages by using the abundant amount of data nectivity and determine the feasibility of power flow.
collected from Twitter, which can be utilized as an addi-
tional source of information to improve utilities’ situational 2) Data Fusion Methodologies: To apply a data fusion
awareness. method, the damage estimation from each source of infor-
mation should be unified into a generic probabilistic model.
For the fragility curve-based methods that assess the impact
C. Multisource Data Fusion Schemes of extreme weather to distribution grid in the component
As discussed before, each source of data could provide level, the probabilistic model is directly given by the fragil-
useful information for estimating the system damage status ity curve that estimates the damage probability given the
from different perspectives; however, due to features of nat- weather metrics. If the parameters of the physical model of
ural disasters, these data may be incomplete. Therefore, a the component are available, the probability of the damage
data fusion method that combines the input data from vari- given the weather metrics can be obtained according to the
ous sources, including weather, measurements from field physical model, for example, the probability of failure is rep-
devices, customers’ trouble calls, smart meters, and even resented by (2) and (3).
social media, to estimate and predict the distribution grid For the statistical regression-based methods which esti-
status is desired. There are several studies in the literature mate the damage count in a given area, we aim to convert
that combine some of these sources for outage and fault iden- the system-level statistical information to component-level
tification. Sun et al. [76] utilized fault-indicating devices to damage probability. Assume that the damage count ​​Y​i​​​ in
identify faulted line sections in medium-voltage feeders and a given area is described by Poisson distribution, that is,
trouble call systems to locate faulted areas for low-voltage ​​Yi​​​  ~Poisson(​λi​  ​​)​ for type ​i​component with failure rate
distribution systems, and the interaction between these ​​ i​  ​​ = 𝔼{ ​Yi​​​  }​, where ​​Y​i​​​is the response variable representing
λ
two schemes is discussed. Liu and Schulz [77] proposed a
knowledge-based system to locate distribution system out-
ages using comprehensive data from customer trouble calls,
a wireless automated meter reading system, and distribution
SCADA. Jiang et al. [78] proposed a multihypothesis method
for identification of the faulted section on a feeder or lateral
using data from smart meters and protective devices.

1) Data Fusion Framework: In the proposed distribu-

tion system restoration decision support tool introduced
in Section III-B, the data fusion module aims to design a
multisource data fusion scheme to improve the situational
awareness of system damage status. It consists of five major Fig. 11. Multisource data fusion module.

12 Proceedings of the IEEE

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

the historical data regarding damage count. In the GLM

method, we apply the logarithmic link function

​log (​λi​  ​​) = β​ 0​  ​​  + ​  ∑ ​​​βp​  ​​​ ​Xp​ ​​​ (10)
p∈

where ​​Xp​ ​​​is the explanatory variable describing the value of​
p​th weather metric of all weather metrics (by set ​​). Using Fig. 12. Linear pool data fusion model.
historical weather data, the coefficient β​​ p​  ​​​ can be estimated
using regression methods. With the model, the failure rate
data fusion models can be exploited to combine local poste-
in a given area can be estimated using the weather informa-
rior probabilities. For example, the linear pool model [79],
tion/forecasts. The component-level failure probability can
as shown in Fig. 12, applies linear combinations of local pos-
then be computed using the exponential distribution. Note
terior probabilities
that for each type of component in the given area, the failure
M
probability of each component is the same, because the fail- p(y| ​x​​  (1)​, ​x​​  (2)​, …, ​x​​  (M)​) ∝ ​  ∑ ​wm
​ ​ ​​​  ⋅ p(y| ​x​​  (m)​).​ (11)
ure rate describes the damage properties of this area from m=1
the statistical perspective. The weight ​​w​m​​​ reflects the significance attached to the
The fault section identification algorithms using meas- ​ t​h information source. It can be used to model the reli-
m
urement data from fault-indicating devices (described in ability or trustworthiness of an information source if such
Section IV-B2) can yield a unique faulted line section by parameters are available, and we can use equal weights if
assuming each line section is equipped with a fault-indi- such information is not available.
cating device. However, due to the damage sustained from The independent pool model [80] applies the product of
the natural disaster, some devices may also be damaged, local posterior probabilities
or in the case that not all line sections are equipped with a M
fault-indicating device, the algorithm will result in multiple p(y| ​x​​  (1)​, ​x​​  (2)​, …, ​x​​  (M)​) ∝ ​  ∏ ​p​  (y| ​x​​  (m)​)​
​ (12)
m=1
faulted line section candidates. We can convert this ambigu-
ity into a probabilistic model to allocate a probability to each which is illustrated in Fig. 13.
instance of line section damage, so that a generic probabilis- We can see that both models have flexibility regarding
tic damage assessment model from fault-indicating devices the number of sources ​M​, i.e., it can tackle any number of
can be derived. For the fault section identification using sources. This feature is very useful as the data fusion frame-
smart meters and other sources of information, the similar work can be easily applied for different electric utilities with
approach can be applied to model the multiple identifica- diverse sources of information. The proposed probabilistic
tion results into a generic probabilistic damage assessment data fusion framework provides an optional solution so that
model. multiple sources of information can be complementary to
Given the probabilistic model of damage estimation each other to improve the accuracy of damage assessment.
using different sources, the objective of data fusion is to Note that there are still several technical challenges in
combine the probabilistic information. We assume that obtaining situational awareness accurately using the exist-
​M​information sources are available and the observations ing approaches and the proposed data fusion framework in
from the ​mt​h source are arranged in the vector ​​x​​  (m)​​. Data practice, such as model uncertainty as a result of the extreme
fusion aims to compute the global posterior probability​ weather, limited amount of historical data, nonstationarity
p(y| ​x​​  (1)​, ​x​​  (2)​, …, ​x​​  (M)​)​, given the information contributed feature of the problem, data quality issues of measurements,
by each source, where ​y​ indicates the event of damage to a etc., which are still open problems in this research area.
component. Generally, each source of information can con-
tribute two conditional probabilities: a local posterior prob- V. FACI L I TAT I NG SU RV I VA BIL I T Y
ability p​ (y| ​x​​  (m)​)​and a likelihood function p​ (​x​​  (m)​  |y)​. In this As shown in Table 2, the lack of power availability during
problem, however, the local posterior probability ​p(y| ​x​​  (m)​)​ outages due to extreme weather events poses challenges
is what each source of information can contribute, which for conventional restoration strategies, which are based on
is actually the result of the probabilistic model of damage
assessment for each source, while it is intractable to obtain
the likelihood function because the relation between the
damage and the observations is through complicated algo-
rithms (e.g., statistical regression or certain fault section
identification algorithms), and we cannot obtain the reverse
of these algorithms to compute the likelihood function.
With the local posterior probability ​p(y| ​x​​  (m)​)​ of the dam-
age from each source of information computed, probabilistic Fig. 13. Independent pool data fusion model.

Proceedings of the IEEE 13

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

B. Utilizing Microgrids With Fixed Boundary

This type of method is based on already-installed micro-
grids, whose boundaries are fixed. In this case, a microgrid
can act in the island mode to serve its own critical loads by
utilizing local generation and storage facilities even when
the utility grid is unavailable. This operation mode requires
special control for the frequency and voltage, since no sup-
port comes from the utility grid. The power electronics
inverters in this case act as voltage source to control the
frequency and voltage [82]. Moreira et al. [90] recommend
the following sequence of actions for a microgrid central
controller to perform service restoration: 1) sectionalize
Fig. 14. Microgrid architecture comprising microsources, storage the microgrid around each microsource (MS) with black-
devices, loads, and control devices [82]. start (BS) capability; 2) build the low-voltage (LV) network
utilizing storage devices; 3) synchronize small islands ener-
gized by the MSs; 4) connect the controllable loads to the
the assumption that most utility power sources are work- LV network; 5) connect noncontrollable MSs or MSs with-
ing and stay connected. In this case, DERs could provide out BS capability; 6) connect other loads; 7) change the con-
valuable resources to facilitate survivability of custom- trol mode of MS inverters; and 8) synchronize the micro-
ers. However, the challenge is how to manage these local grid with the medium-voltage network. In [91], Resende et
power resources to serve the critical loads in an effective al. further propose a novel distribution system architecture
and efficient manner [81]. Microgrids, which are small- that allows the coordination among multiple microgrids
scale power systems typically on the medium- or low-volt- for service restoration, and the corresponding sequence of
age distribution feeder that include distributed load and actions is defined.
generation together with storage and protection devices, In addition to the critical loads within its boundary, a
provide an applicable solution. Fig. 14 shows a typical microgrid can also be utilized to pick up critical loads outside
microgrid architecture that consists of different types of its boundary using its available surplus power. In this case,
components [82]. the microgrids act as virtual feeders to pick up critical loads,
and this scheme is especially useful for areas where no other
suitable restoration path or source is available [29]. To pick
A. Using Microgrids for Grid Resiliency
up the external loads, feasible paths should be formed to con-
A key advantage of microgrids in terms of resiliency nect the microgrid to the load; the automated feeder switches
is that microgrids add active network components (i.e., could help to expedite this process. Li et al. [92] proposed a
DERs), which provide operational flexibility and reduce con- spanning tree search algorithm to find the feasible path using
ventional power vulnerabilities caused by centralized gen- automated feeder switches to maximize the restored load
eration and control architecture and long distances between and minimize the number of switching operations without
power sources and loads [83]. Microgrids provide an effec- violations of operational constraints. Xu et al. [93] further
tive solution to managing these distributed resources so that extended the work to propose an optimal placement strat-
utilizing them for system restoration after extreme weather egy for automated feeder switches to enhance the resiliency.
events is possible. Furthermore, the uneven damage distri- Mohagheghi and Yang [94] proposed a mathematical model
bution of the disaster on the distribution grid increases the to utilize microgrids to alleviate the outage in the absence of
resilience when applying microgrids for load restoration, as a suitable restoration path/source. Besides picking up exter-
the chances of all microgrids being damaged are very low nal loads, microgrids could also have the potential to provide
[83]. The value of microgrids for enhancing customers’ sur- ancillary services such as blackstart to the bulk power system
vivability and grid resilience has been recognized in several restoration. For example, Castillo [95] developed a stochas-
tic mixed-integer linear program (MILP) to assess the impact
recent studies in the literature [84]–[87]. They are being
of coordinating microgrids as a blackstart resource to the
adopted by some state governments and industries, and the
regional grid or RTO after a natural disaster.
technical, regulatory, and financial barriers to implementa-
tion are being studied, e.g., [81], [88], and [89].
Existing methods of using microgrids to enhance distri- C. Dynamic Formation of Microgrids
bution restoration can be categorized as two types, accord- This type of method does not require existing micro-
ing to the conditions of the microgrids: 1) microgrids with grids; instead, the microgrids can be formed dynamically
fixed boundary; and 2) dynamic formation of microgrids, by controlling the automated feeder switches. Because of
which will be discussed below. the greater flexibility this scheme could provide compared

14 Proceedings of the IEEE

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

to the methods based on fixed-boundary microgrids, it will this microgrid, then line (​ i, j)​belongs to microgrid k​ ​. If line​
be adaptable to the different outage and damage scenarios (i, j)​belongs to any one of the microgrids in K​​, the switch on
of natural disasters. However, this scheme also poses chal- this line (if it exists) should be in the closed state. Thus, the
lenges for controlling and coordinating these smart switches branch-node constraints to specify switch state on line ​(i, j)​
without violating the operational constraints. in terms of node clustering variables v​ ​​ ik​​​can be expressed as

1) MILP Formulation of Microgrid Formation: In our ​​c​ij​​  = ​  ∑ ​​​vh​ k​​​,  h = ​
ζk​  ​​  (i, j),  (i, j)   ∈ ℒ​ (16)
k∈K
previous work [96], a microgrid formation mechanism to
restore critical loads in the absence of a utility grid is pro- where ​h = ​ ζk​  ​​  (i, j)​denotes the children node of line (​i, j)​
posed. We consider a radial power distribution system con- regarding microgrid ​k​.
sisting of ​N​nodes denoted by the set of nodes N ​ : = { 1, …, N}​, Additional microgrid load pickup constraints specify the
and ​L​power lines represented by the set of edges ℒ ​ : = { (i, j) condition on whether the load at node ​i​can be energized
⊆ N × N}​. Let ​​c​ij​​ ∈ { 0, 1}​denote the binary decision vari- by microgrid ​k ∈ K​, which can be represented by satisfy-
ables indicating whether the switch associated with line ​(i, j)​ ing the following two conditions simultaneously: 1) node ​i​
is open (​​c​ij​​ = 0​) or closed (​​c​ij​​ = 1​), and let ​​s​i​​ ∈ 0, 1​denote the belongs to microgrid k​ ​, i.e., ​​vi​k​​  = 1​; and 2) the switch asso-
binary decision variables representing whether the switch ciated with the load is closed so that the load is connected to
connecting the load and node i​​is open (​​s​i​​ = 0​) or closed node ​i​, i.e., s​​ ​i​​  = 1​. These two conditions formulated as ​​v​ik​​  ⋅ ​
(​​si​​​ = 1​). By controlling these switches, K ​ ​self-adequate micro- s​i​​  = 1​, which is a quadratic constraint. To address the non-
grids can be formed to maximize the total prioritized loads linearity issue, we define auxiliary binary decision variables​​
to be restored. The microgrid formation problem is mod- ​  ​​  ∈ { 0, 1}​ as ​​γik
γik ​  ​​  = ​​ ​​  ⋅ ​s​i​​,   ∀ i ∈ N, k ∈ K​, and the quad-
vik
eled as an MILP problem. To model node clustering con- ratic equality constraints can be converted to the following
straints that specify which microgrid a node belongs to, we three linear inequality constraints:
define auxiliary binary decision variables ​​v​ik​​ ∈ { 0, 1}​indicat-
ing whether node i​ ∈ N​belongs to microgrid k​ ∈ ​ (​​vik ​ ​​ = 1​ ​  ​​  ≤ ​
​​γik vik
​ ​​,   ∀ i ∈ N,  k ∈ K​ (17)
if node ​i​belongs to microgrid ​k​, and v​​ i​k​​ = 0​if otherwise),
​  ​​  ≤ ​
​​γik si​​​,   ∀ i ∈ N,  k ∈ K​ (18)
and the node clustering constraints can be expressed as
​  ​​  ≥ ​
​​γik ​ ​​  + ​s​i​​  − 1,  ∀ i
vik ∈ N,  k ∈ K.​ (19)
​​  ∑ ​​​vi​k​​​  = 1,  ∀ i ∈ N​ (13)
k∈
Besides these logical constraints discussed above, other
to represent that if the load at node ​i ∈ N​can be energized
operational constraints are modeled, including voltage
by microgrids, this node can only belong to one of the K ​​
range constraints at each node, and real and reactive power
microgrids.
output range constraints for DGs. The objective function
For node ​i​at which the DG k​ ​is installed (i.e., i​ = k​),
aims to maximize the total priority weighted load picked up,
node ​i​will belong to microgrid ​k​, which can be written as
where the weights indicate the importance and priority of
the following equality constraints:
the loads for restoration. The detailed formulation can be
​​v​ik​​  = 1,  i = k,  ∀ i ∈ N,  k ∈ K.​ (14) found in our previous work in [96].
With this scheme, DGs can be utilized to serve critical
The radial topological features are captured in the load locally to facilitate the customers’ survivability after
microgrid connectivity constraints. Specifically, for a radial extreme weather events. Fig. 15 illustrates the formation of
distribution network, each microgrid is a subtree network three microgrids based on the IEEE 37-node test system [96].
with the root node being the node where the DG is installed. Our work in [97] proposed a self-healing strategy by section-
In that sense, one node can belong to microgrid k​ ​only if alization of the distribution system into microgrids, with the
its parent node (for this microgrid) belongs to microgrid ​k​, consideration of the uncertainty of renewable generation.
which can be expressed as
2) Multi-Time-Step Sequential Restoration: A feasible
​​v​ik​​  ≤ ​
vjk
​ ​​,   ∀ k ∈ K,  ∀ i ∈ N\ { k},  j = ​
θk​  ​​  (i)​ (15) service restoration plan should include a sequence of con-
trol actions (e.g., switching, DG dispatching) that the sys-
where ​​θk​  ​​  (i)​denotes the parent node of node ​i​ regarding tem operator can follow to restore the affected customers
microgrid ​k​. step by step. However, the previous work focuses on how
In addition, the microgrid branch-node constraints spec- to generate a “snapshot” of the optimal reconfiguration of
ify the relation between nodes and lines in a microgrid. If the grid (e.g., the microgrid formation scheme in [96]).
both nodes ​i​and ​j​belong to microgrid ​k​, i.e., ​​vi​k​​  = ​vjk
​ ​​  = 1​, The snapshot configuration must be verified by switching
then the line connecting nodes i​​and j​​should also belong to order management (SOM) in the control center. If some
microgrid k​ ​. Together with (14), we can further derive that if constraints happen to be violated, the solution should be
the children node of (​ i, j)​(regarding microgrid k​ ​) belongs to updated and verified by SOM again. This iterative process

Proceedings of the IEEE 15

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

the problem into two steps: 1) optimal switching operation

sequence searching by branch and bound algorithm; and
2) validating the switching sequence by simulation.
Our work in [101] proposes an innovative multi-time-
step distribution restoration scheme to coordinate switch-
ing action and DG dispatching in an integrated manner. An
MILP model, which can be solved efficiently, is formulated
to maximize the energized loads while satisfying various
operational constraints. To satisfy network topological con-
straints, we can derive several logical constraints governing
the binary decision variables, for example, a switchable line
can only be energized when both end nodes are energized;
a switchable load can only be restored when the node it
connects to is energized; a node can be energized when a
black start DG connects to it, or an energized line connects
to it; and a non-black-start DG can only be started when it
connects to an energized node. In addition, the sequential
constraints should be satisfied, that is, each line should be
energized only when at least one of its ends is energized at
a previous interval. Other constraints include the power
flow model, transformer and line capacity constraints, DG
power output constraints, voltage range constraints, ramp
rate limit constraints, frequency response rate constraints,
etc., which can be found in [101] in detail.
To illustrate the sequential restoration results, we test
the algorithm on the modified IEEE 13 node test feeder. The
single line diagram of the test feeder is shown in Fig. 16. The
system is assumed to be completely de-energized due to an
Fig. 15. A case study of dynamic microgrid formation based on extreme-weather-related event. Two lines and one load are
the IEEE 37-node test system [96]: (a) the original IEEE 37-node assumed to be damaged and cannot participate in the resto-
test system with three DGs; and (b) three microgrids are formed ration. There are three DGs connected to nodes 650, 646,
energized by these DGs. and 680. Two case studies are considered: in case 1, only
DG1 connected to the substation node is used for service
will continue until a feasible solution is found. There are restoration; in case 2, three DGs will participate in the res-
some existing studies on how to find an optimal and feasible toration. Table 3 summarizes the restoration sequence for
switching sequence for the traditional restoration scheme loads and lines at each step. Fig. 17 illustrates the step-by-
described in Section II-A, such as the generic algorithm [98] step restoration sequence energized by three DGs (case 2).
and dynamic programming [99]. However, for dynamic for-
3) Mobile Generator Dispatch With Microgrid Formation:
mation of microgrids, the operation of DGs and the switch-
In addition to the static DGs, a mobile generator, which is a
ing action should be carefully coordinated; i.e., during the
restoration process, it is important to ensure secure and
safe operation when performing the switching actions and
ramping the DGs up and down. Pham et al. [100] divided
Table 3 Restoration Sequence for Loads and Lines in Cases 1 and 2

Fig. 16. Single line diagram of modified IEEE 13 node test feeder
with three DGs.

16 Proceedings of the IEEE

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

Fig. 18. DistFlow model for a radial distribution grid.

necessary in the optimization formulation. The iteratively

based distribution power flow model (e.g., forward and
backward sweep method in [104]) cannot be directly inte-
grated into the optimization models. Instead, the DistFlow
model proposed by Baran and Wu [105], [106] provides a
viable solution for a radial distribution grid. For the radial
distribution grid shown in Fig. 18, the DistFlow model can
Fig. 17. Illustration of step-by-step restoration sequence for case 2. be represented by
​P​  2ij​ ​  + ​Q​  2ij​ ​
​​P​ij​​  − ​  ∑ ​​​Pjk pj​​​  + ​r​ij​​ ​ ______
​ ​​​  = ​ 2
 ​​ (20)
truck-mounted generator with the merits of mobility and large k∈​Nj​​​ ​V​  i​  ​
capacity (up to several MW), could be quickly connected to ​P​  2ij​ ​  + ​Q​  2ij​ ​
provide replacement power to critical or isolated loads, thus qj​​​  + ​x​ij​​ ​ ______
​​Q​ij​​  − ​  ∑ ​​​Qj​k​​​  = ​ 2
 ​​ (21)
providing more flexibility to facilitate customers’ survivability. k∈​Nj​​​ ​V​  i​  ​
As suggested in the GridWise report [2], electric utilities should ​P​  2ij​ ​  + ​Q​  2ij​ ​
increase the use of mobile generators to provide temporary rij​ ​​ ​Pij​ ​​  + ​x​ij​​ ​Qij​ ​​)− (​r​  2ij​ ​  + ​x​  2ij​ ​) ​ ______
​​V​  2i​  ​  − ​V​  2j​  ​  = 2(​ 2
 ​​ (22)
​V​  i​  ​
power during extreme weather events. However, currently
where ​​P​ij​​​ and ​​Q​ij​​​represent the real and reactive power
they are not well utilized. For example, before Superstorm
flow from node ​i​to j​​, ​​V​j​​​is the voltage at node j​​, and ​​r​ij​​​ and​​
Sandy struck, 400 industrial-size truck-mounted emergency
response generators were prepared by the FEMA, but only a x​ij​​​denote the resistance and reactance of line (​i, j​), respec-
fraction of them were providing power even three days after tively. DistFlow can be further linearized by omitting the
Sandy made landfall [102]. Our work in [103] integrates the loss term (quadratic term) and approximating the voltage
dispatch of mobile generators into the microgrid formation by term, e.g., [96], [105], and [107], to make the computation
proposing a two-stage dispatch framework: more efficient. While the DistFlow model is for a balanced
• in the first stage (i.e., prior to a natural disaster), system, further studies [108]–[110] extended it to a three-
prepositioning is conducted to placing utilities’ phase unbalanced system by a linearization approximation,
mobile generators across their staging locations to or convex relaxation which can be solved efficiently by the
ensure the earliest possible response after the natu- semidefinite programming (SDP) technique.
ral disaster strikes;
2) Cold Load Pickup Issue: Cold load pickup (CLPU) is a
• in the second stage (i.e., after the natural disaster
practical issue when restoring loads. CLPU refers to the obser-
strikes), real-time allocation is optimized to send
vation that when a load is being restored after an extended
mobile generators from staging locations to allo-
outage, a much higher demand than the pre-outage level
cated locations. Upon arrival, they are connected
to the grid and form microgrids to pick up critical will result because a large concentration of thermostatically
loads. controlled loads (such as air conditioners, heaters, and refrig-
With the proposed two-stage framework, the mobile erators) will start at the same time, as illustrated in Fig. 19. A
generators can be better utilized to serve critical loads after study by the IEEE PES Power System Relay Committee [111]
extreme weather events. discussed various aspects of CLPU issues. Schneider et al.
[112] described several methods to quantify the CLPU profile.
This special load profile caused by the loss of demand
D. Special Considerations diversity should be integrated into the restoration control
1) Distribution Power Flow Model: One important sequence. In [113], Kumar et al. integrated the CLPU con-
operation constraint is to make sure the voltage at each straints into restoration using the generic algorithm. Our
node is within certain ranges; thus the power flow model is previous work [101] applied a typical delayed exponential

Proceedings of the IEEE 17

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

Fig. 19. Illustration of the CLPU issue.

CLPU curve and used linearization to integrate it into the Fig. 20. The load demand under CLPU conditions.
MILP formulation. Fig. 20 shows a typical delayed expo-
nential CLPU curve. The outage occurs at t​​​0​​​, and the load
microgrids. However, the communication infrastructure
is restored at t​​​1​​​. Due to the loss of diversity, the undiversi-
may also be damaged during extreme weather events, pos-
fied loading factor at ​​t​1​​​ is ​​SU​ ​​​. Then, the load starts to gain
ing challenges for restoration. An alternative communica-
diversity at ​​t​2​​​ and decreases exponentially. The post-outage tion solution is needed in this case. While satellite commu-
diversified loading factor is ​​S​D​​​, which is normally equal to nication will usually not be impacted by extreme weather
the pre-outage loading level. It should be noted that the res- events, its cost is an issue, especially for device-level com-
toration time (i.e., t​​​1​​​) for each load is not predetermined. munications. Another possible solution is to utilize a peer-
Given a CLPU curve, we can calculate the load demand to-peer ad hoc-type communication network, which does
at each sampling time. In this work, we assume that the not rely on infrastructure [114]. In previous work [96], we
CLPU curve is equally sampled, and the sampling interval is designed a distributed information collection scheme using
​ t​, which is the interval used for the restoration. Assuming
Δ
only local communications, which can be achieved by this
total ​N​samples are collected from the CLPU curve of load peer-to-peer communication architecture. The proposed
​l​, denote L​​ l​​​  (k)​as the load demand at the ​k​th sample, and method employs the average consensus algorithm discussed
​Δ ​Ll​​​  (k)​as the load difference between two consecutive sam- in [115] to achieve global information discovery by iterative
ples. Δ​  ​Ll​​​  (k)​can be active power Δ​  ​Pl​​​  (k)​, reactive power local information update, as shown in (26)
​  ​Ql​​​  (k)​, or apparent power Δ​  ​Sl​​​  (k)​. Then, Δ​  ​Ll​​​  (k)​can be cal-
Δ
culated as ​​z​  k+1 z​  ki​  ​  + ​ ∑ ​​​εij​  ​​​  ⋅ (​z​  kj​  ​  − ​z​  ki​  ​)​
i​  ​  = ​ (26)

{ ​L​l​  ​​  (k)  − ​L​l​​  (k − 1),​  1 ≤ k ​ ≤ N​​​

j∈​ℛ​i​​
0, k = 1
Δ ​Ll​​​  (k)   = ​ ​  
​ (23) where ​​z​  ki​  ​​represents the information vector at node i​ ​at step k​ ​,
​​R​i​​​represents the set of direct neighboring nodes of node i​​,
where ​​L​l​​  (k)​can be calculated using the exponential decay and ε​​ ij​  ​​​is the step size, which is related to the graph charac-
rate in Fig. 20. Then, the CLPU load (i.e., ​​P​  Ll,t ​​​  and ​​Q​  Ll,t ​​​  for teristics of the network. With proper design of ​​z​  ki​  ​​ regarding
active and reactive power) can be calculated in an accumu- the actual status of the devices, the global information can
lative manner, which could be formulated as eventually be obtained at each node using this simple linear

( )
t iteration, with convergence guaranteed. The details of the dis-
​​P​  Ll,t ​ ​  = ​
P​  Ll​  ​​ S​ ​​  U​ ​x​  Ll,t ​ ​  − ​ ∑ Δ​ ​ ​ ​Pl​​​  (k) ​x​  l,t−k+1
L
 ​ ​ ​​ (24)
k=1 tributed information collection scheme can be found in [96].

( )
t
​​Q​  Ll,t ​ ​  = ​
Q​  Ll​  ​​ S​ ​​  U​ ​x​  Ll,t ​ ​  − ​ ∑ Δ​ ​ ​ ​Ql​​​  (k) ​x​  l,t−k+1
L
 ​ ​ ​​ (25) V I. CONCLUSION A N D DISC US SION
k=1
where ​​x​  Ll,t ​​​  is
the binary decision variable indicating whether The increasing frequency and severity of extreme weather
the load ​l​is energized at time step t​ ​in the MILP formulation events, the vulnerability of distribution grids to natural dis-
of the sequential restoration model in [101]. ​​P​  Ll​  ​​ and ​​Q​  Ll​  ​​ are asters, and the higher expectations of customers regarding
pre-outage active and reactive power of load ​l​, respectively. electricity service continuity during disasters are the main
With (23) and (24), the time sequence of load demand motivating forces for utilities to improve their restoration
considering CLPU can be integrated in the multi-time- strategies in response to extreme weather events. This paper
step sequential restoration optimization as discussed in describes the current distribution system restoration prac-
Section V-C2. tices and discusses their inadequacy for power outages during
extreme weather events. To leverage the grid modernization
3) Distributed Information Collection: Collecting infor- development initiated by DOE, this paper presents an inte-
mation on grid status, e.g., load information and switch grated solution based on a decision support tool that could
states, is a prerequisite for conducting restoration using assist utilities with decision making for distribution system

18 Proceedings of the IEEE

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

restoration in response to extreme weather events. The solu- are needed to facilitate vital investments and upgrades in
tion features two advantages: 1) improving situational aware- grid modernization technologies [2]. Proper rate structures
ness of system damage status; and 2) facilitating survivability and incentives should be designed to encourage DGs and
of customers. Many existing methodologies in the literature grid modernization investments.
can be utilized to achieve these two advantages, and this
paper provides a detailed review of these studies. B. Technical Challenges
There still exist many challenges, from both regulatory
and technical perspectives, for modernizing distribution The damage caused by extreme weather events will
system restoration to cope with extreme weather events. affect not only electric power systems, but also other infra-
Here we give a brief discussion based on our observations. structure, such as communications networks, natural gas
pipelines, transportation networks, etc. The interdepend-
ence between these infrastructures and electric power sys-
A. Regulatory Challenges tems will have significant impacts on the utilities’ restora-
Although DGs could have the potential to enhance tion process. For example, damage to the transportation
supply continuity and customers’ survivability during the network may result in delayed repair-crew dispatch; the
outages after disasters, current interconnection standards operations of certain DGs (e.g., internal combustion engine
(primarily the IEEE Standard 1547 [116]) require customer- generators or microturbines) heavily rely on the availability
owned DGs to be disconnected during disturbances, to of fuel; and situational awareness and damage assessment
guarantee the safety of utility crews and the power quality will be highly degraded if the underlying communications
of the grids. These requirements were developed when the networks experience failures. Analysis and quantification of
penetration of distributed resources was low; now, how- this interdependence and how it can be integrated into the
ever, with the ever-increasing penetration of distributed distribution system restoration process is challenging and
resources, these requirements in the standard will underuti- needs further R&D efforts.
lize these local resources during restoration due to extreme Utilizing DERs to enhance supply continuity during
weather events. In this sense, there is a need to adjust the restoration changes the way traditional distribution sys-
standard to accommodate scenarios in which the benefits tems operate. For example, frequency and voltage need
of the local resources can be leveraged during restoration in to be regulated by these local resources rather than from
response to extreme weather events. bulk power systems, so the coordination and control
Investments in grid modernization technologies could schemes need to be carefully designed. The lower iner-
provide the potential to enhance grid resiliency and improve tia compared to the bulk power system will pose chal-
system restoration; however, current efforts are still heavily lenges for control schemes to guarantee system stability
dependent on government investment, as the resilience and and dynamic performance. In addition, the protective
reliability benefits of these investments are not reflected devices and the corresponding control logic need sub-
in the electricity tariff. In addition, there is no scheme to stantial modifications to avoid serious problems as result
compensate for the DG owners for providing service dur- of power injection from DERs. Tackling these challenges
ing blackouts resulting from extreme weather events. In in a cost-effective way is not an easy task and requires
this sense, government policy makers and electric utili- R&D efforts with interdisciplinary expertise, including
ties should consider the societal benefits when performing power systems operation and control, power system pro-
cost-benefit analyses of grid modernization investments. tection, power electronics, communications, operation
Appropriate changes in policy and regulatory environments research, etc. 

R EFER ENCES [3] R. J. Campbell, “Weather-related power [6] PHYS.org, “The cost of blackouts in Europe,”
[1] U.S. Department of Energy, “Economic outages and electric system resiliency,” Apr. 2016. [Online]. Available: http://phys.
benefits of increasing electric grid resilience Congr. Res. Service, Aug. 2012. [Online]. org/news/2016-04-blackouts-europe.html
to weather outages,” Aug. 2013. [Online]. Available: http://www.fas.org/sgp/crs/misc/ [7] BBC News, “South Australia storms leave
Available: http://energy.gov/sites/prod/ R42696.pdf state without power,” Sep. 2016. [Online].
files/2013/08/f2/Grid Resiliency Report_ [4] S. Küfeoğlu, S. Prittinen, and M. Lehtonen, Available: http://www.bbc.com/news/world-
FINAL.pdf “A summary of the recent extreme weather australia-37495535
[2] The GridWise Alliance, “Improving electric events and their impacts on electricity,” Int. [8] P. Hines, J. Apt, and S. Talukdar, “Trends in
grid reliability and resilience: Lessons Rev. Electr. Eng., vol. 9, no. 4, p. 558, Aug. the history of large blackouts in the United
learned from Superstorm Sandy and other 2014. States,” in Proc. IEEE Power Energy Soc.
extreme events,” Jun. 2013. [Online]. [5] S. Küfeoğlu and M. Lehtonen, “Cyclone General Meeting, Jul. 2008, pp. 1–8.
Available: http://www.gridwise.org/ Dagmar of 2011 and its impacts in Finland,” [9] U.S. Department of Energy, “Electric
documents/ImprovingElectricGridReliability in Proc. IEEE PES Innov. Smart Grid Technol. Disturbance Events (OE-417) Annual
andResilience_6_6_13webFINAL.pdf Conf. Eur., Oct. 2014, pp. 1–6. Summaries,” Jul. 2016.

Proceedings of the IEEE 19

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

[Online]. Available: https://www.oe.netl.doe. [24] C. P. Nguyen and A. J. Flueck, “Agent based benefit modeling,” Quanta Technol.,
gov/OE417_annual_summary.aspx restoration with distributed energy storage Raleigh, CA, USA: Tech. Rep., May 2008.
[10] S. Küfeoğlu, “Economic impacts of electric support in smart grids,” IEEE Trans. Smart
Grid, vol. 3, no. 2, pp. 1029–1038, Jun. 2012. [40] J. Winkler, L. Dueñas-Osorio, R. Stein, and
power outages and evaluation of customer D. Subramanian, “Performance assessment
interruption costs,” Ph.D. dissertation, Aalto [25] A. L. Morelato and A. Monticelli, “Heuristic
of topologically diverse power systems
Univ., Espoo, Finland, 2015. search approach to distribution system
restoration,” IEEE Trans. Power Delivery, subjected to hurricane events,” Rel. Eng. Syst.
[11] A. Staudt and R. Curry, “More extreme
vol. 4, no. 4, pp. 2235–2241, Oct. 1989. Safety, vol. 95, no. 4, pp. 323–336, Apr. 2010.
weather and the U.S. energy infrastructure,”
National Wildlife Federation, Virginia, VA, [26] S. Khushalani, J. M. Solanki, and N. N. [41] Substation Structure Design Guide, ASCE
USA, Tech. Rep., 2011. [Online]. Available: Schulz, “Optimized restoration of Manuals and Reports on Engineering
https://www.nwf.org/~/media/PDFs/Global- unbalanced distribution systems,” IEEE Practices, Amer. Soc. Civil Eng., Reston, VA,
Warming/Extreme-Weather/Final_NWF_ Trans. Power Delivery, vol. 22, no. 2, USA, 2008.
EnergyInfrastructureReport_4-8-11.ashx pp. 624–630, May 2007.
[42] Federal Emergency Management Agency
[12] National Centers for Environmental [27] “Protecting electricity networks from natural
Information (NCEI), “Billion-dollar weather hazards,” Org. Secur. Co-Oper. Eur., 2016. (FEMA), “Multi-hazard loss estimation
and climate disaster: Overview,” Jul. 2016. methodology hurricane model: Hazus-
[28] Electric Power Research Institute (EPRI),
[Online]. Available: http://www.ncdc.noaa. “Enhancing distribution resiliency: MH2.1.” [Online]. Available:
gov/billions/ Opportunities for applying innovative http://www.fema.gov/media-library-
[13] L. Warren and B. Stebner, “There will be technologies,” Jan. 2013. [Online]. Available: data/20130726-1820-25045-9850/hzmh2_1_
light coned promises to restore power to http://www.epri.com/abstracts/Pages/ hr_tm.pdf
Manhattan by Friday...but everyone else will ProductAbstract.aspx?Product
[43] Federal Emergency Management Agency
have to wait a week,” Nov. 2012. [Online]. Id=000000000001026889
Available: http://www.dailymail.co.uk/news/ (FEMA), “Multi-hazard loss estimation
[29] Y. Wang, C. Chen, J. Wang, and R. Baldick, methodology flood model: Hazus-MH.”
article-2225308 “Research on resilience of power systems
[14] L. A. Freeman, G. J. Stano, and M. E. under natural disasters—A review,” [Online]. Available: http://www.fema.gov/
Gordon, “Best practices for storm response IEEE Trans. Power Syst., vol. 31, no. 2, media-library-data/20130726-1820-25045-
on U.S. distribution systems,” pp. 1604–1613, Feb. 2016. 8292/hzmh2_1_fl_tm.pdf
Distrib. Syst. Test. Appl. Res., Tampa, FL, [30] GTM Research, “The new face of damage [44] NWS, “National Weather Service data in
USA, Tech. Rep., Mar. 2010. assessment,” Nov. 2015. [Online]. Available: shapefile format.” [Online]. Available: http://
[15] U.S. Department of Energy, “Recovery Act: http://www.ourenergypolicy.org/ www.nws.noaa.gov/gis/shapepage.htm
Smart Grid Investment Grant (SGIG) wp-content/uploads/2016/01/
GE-Whitepaper-2015-Web-Final.pdf [45] NOAA, “Severe weather data.” [Online].
Program,” 2016. [Online]. Available: http://
energy.gov/oe/information-center/recovery- [31] U.S. Department of Energy, “Smart Grid Available: http://www.ncdc.noaa.gov/data-
act-smart-grid-investment-grant-sgig- Investment Grant Program,” Progress Rep., access/severe-weather
program Jul.2012. [Online]. Available: http://www. [46] The Weather Research & Forecasting Model.
[16] V. Giordano, F. Gangale, G. Fulli, and M. S. energy.gov/oe/downloads/smart-grid- [Online]. Available: http://www.wrf-model.
Jimenez, “Smart grid projects in Europe: investment-grant-program-progress-report- org/index.php
Lessons learned and current developments,” july-2012
Eur. Commission Joint Res. Center-Inst. [32] U.S. Department of Energy, “Smart Grid [47] S. D. Guikema, S. M. Quiring, and S.-R.
Energy, JRC Ref. Rep., 2011. Investment Grant Program,” Progress Rep. II, Han, “Prestorm estimation of hurricane
Oct. 2013. [Online]. Available: https://www. damage to electric power distribution
[17] U.S. Department of Energy, “Smart grid
investments improve grid reliability, smartgrid.gov/files/SGIG_progress_ systems,” Risk Anal., vol. 30, no. 12, pp.
resilience, and storm responses,” Smart Grid report_2013.pdf 1744–1752, Oct. 2010.
Investment Grant Program Nov. 2014[Online]. [33] U.S. Department of Energy, “2014 Smart [48] H. Liu, R. A. Davidson, D. V. Rosowsky, and
Available: http://energy.gov/sites/prod/ Grid System Report,” Report to Congress, J. R. Stedinger, “Negative binomial
files/2014/12/f19/SG-ImprovesRestoration- Aug. 2014. [Online]. Available: http://www.
energy.gov/sites/prod/files/2014/08/f18/ regression of electric power outages in
Nov2014.pdf
SmartGrid-SystemReport2014.pdf hurricanes,” J. Infrastruct. Syst., vol. 11, no. 4,
[18] J. Wang, “Advanced distribution
pp. 258–267, Dec. 2005.
management systems for grid [34] U.S. Department of Energy, “Reliability
modernization: DMS functions,” Argonne improvements from the application of [49] H. Liu, R. A. Davidson, and T. V.
Nat. Lab., Lemont, IL, USA, Tech. Rep. distribution automation technologies— Apanasovich, “Spatial generalized linear
ANL/ESD-15/17, Sep. 2015. Initial results,” Smart Grid Investment Grant mixed models of electric power outages due
[19] C.-C. Liu, S. J. Lee, and S. S. Venkata, “An Program, Dec. 2012. [Online]. Available: to hurricanes and ice storms,” Rel. Eng. Syst.
expert system operational aid for restoration http://energy.gov/sites/prod/files/
Safety, vol. 93, no. 6, pp. 897–912, Jun. 2008.
and loss reduction of distribution systems,” DistributionReliability-Report_
IEEE Trans. Power Syst., vol. 3, no. 2, Dec2012Final.pdf [50] S.-R. Han, S. D. Guikema, S. M. Quiring, K.-
pp. 619–626, May 1988. [35] Schweitzer Engineering Laboratories (SEL), H. Lee, D. Rosowsky, and R. A. Davidson,
[20] C.-S. Chen, C.-H. Lin, and H.-Y. Tsai, “A rule- “AR360: Overhead AutoRANGER fault “Estimating the spatial distribution of power
based expert system with colored Petri net indicators.” [Online]. Available: https:// outages during hurricanes in the gulf coast
models for distribution system service selinc.com/products/AR360/ region,” Rel. Eng. Syst. Safety, vol. 94, no. 2,
restoration,” IEEE Trans. Power Syst., vol. 17, [36] S&C Electric Company, “Intellirupter pp. 199–210, Feb. 2009.
no. 4, pp. 1073–1080, Nov. 2002. pulsecloser fault interrupter.” [Online].
[51] S.-R. Han, S. D. Guikema, and S. M.
[21] S. I. Lim, S. J. Lee, M. S. Choi, D. J. Lim, and Available: http://www.sandc.com/products/
switching-overhead-distribution/ Quiring, “Improving the predictive accuracy
B. N. Ha, “Service restoration methodology of hurricane power outage forecasts using
for multiple fault case in distribution intellirupter-pulsecloser/
systems,” IEEE Trans. Power Syst., vol. 21, [37] S&C Electric Company, “6800 series generalized additive models,” Risk Anal., vol.
no. 4, pp. 1638–1644, Nov. 2006. automatic switch control.” [Online]. 29, no. 10, pp. 1443–1453, Oct. 2009.
[22] S.-J. Lee, S.-I. Lim, and B.-S. Ahn, “Service Available: http://www.sandc.com/products/ [52] R. Nateghi, S. D. Guikema, and S. M. Quiring,
restoration of primary distribution systems automation-control/6800-series-automatic- “Comparison and validation of statistical
based on fuzzy evaluation of multi-criteria,” switch-control.asp
methods for predicting power outage durations
IEEE Trans. Power Syst., vol. 13, no. 3, [38] C. Marnay, H. Aki, K. Hirose, A. Kwasinaski, in the event of hurricanes,” Risk Anal., vol. 31,
pp. 1156–1163, Aug. 1998. S. Ogura, and T. Shinji, “Japan’s pivot to no. 12, pp. 1897–1906, Dec. 2011.
[23] J. M. Solanki, S. Khushalani, and N. N. resilience: How two microgrids fared after the
Schulz, “A multi-agent solution to 2011 earthquake,” IEEE Power Energy Mag., [53] Multi-Resolution Land Characteristics
distribution systems restoration,” IEEE Trans. vol. 13, no. 3, pp. 44–57, May-June 2015. Consortium. National Land Cover Database
Power Syst., vol. 22, no. 3, pp. 1026–1034, [39] L. Xu and R. Brown, “Undergrounding (NLCD). [Online]. Available: http://www.
Aug. 2007. assessment phase 3 report: Ex ante cost and mrlc.gov/

20 Proceedings of the IEEE

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

[54] U.S. Geological Survey, “STATSGO State [70] R. A. Fischer, A. S. Laakonen, and N. N. disaster recovery,” IEEE Power Energy Mag.,
Soil Geographic (STATSGO) data base.” Schulz, “A general polling algorithm using a vol. 12, no. 3, pp. 67–76, May-June 2014.
[Online]. Available: http://www.soilinfo.psu. wireless AMR system for restoration   [86] M. Montoya, R. Sherick, P. Haralson,
edu/index.cgi?soil_data&statsgo confirmation,” IEEE Trans. Power Syst., R. Neal, and R. Yinger, “Islands in the
[55] National Weather Service: National vol. 16, no. 2, pp. 312–316, May 2001.
storm: Integrating microgrids into the
Hurrican Center, “National Oceanic and [71] DOE ARPA-E, “Measuring phase angle larger grid,” IEEE Power Energy Mag.,
Atmospheric Administration (NOAA) change in power lines.” [Online]. Available: vol. 11, no. 4, pp. 33–39, July-Aug. 2013.
Hurricane Database.” [Online]. Available: http://arpa-e.energy.gov/?q=slick-sheet-
project/measuring-phase-angle-change-   [87] L. Che, M. Khodayar, and M. Shahidehpour,
http://www.nhc.noaa.gov/ “Only connect: Microgrids for distribution
power-lines
[56] National Weather Service: Advanced system restoration,” IEEE Power Energy Mag.,
Hydrologic Prediction Service, “National [72] J. H. Eto et al., “Scoping study on research vol. 12, no. 1, pp. 70–81, Jan.-Feb. 2014.
Oceanic and Atmospheric Administration and development priorities for distribution-
system phasor measurement units,”   [88] A. R. Hopper et al., “Maryland resiliency
(NOAA) Precipitation Database.” [Online].
Lawrence Berkeley Nat. Lab., Berkeley, CA, through microgrid task force report,”
Available: http://water.weather.gov/precip/
USA, Tech. Rep., Dec. 2015. Maryland Energy Admin., Baltimore, MD,
[57] E. Laverty and N. N. Schulz, “An improved USA, Task Force Rep., Jun. 2014.
algorithm to aid in post-heat storm [73] Power Standards Lab (PSL), “PQube 3:
World’s smallest & most precise micro   [89] A. M. Cuomo, R. L. Kauffman, J. B.
restoration,” IEEE Trans. Power Syst., vol. 14,
synchrophasor.” [Online]. Available: http:// Rhodes, A. Zibelman, and J. M. Hauer,
no. 2, pp. 446–451, May 1999.
www.powersensorsltd.com/PQube3.php “Microgrids for critical facility resiliency in
[58] E. Vazquez M, O. L. Chacón M, and H. J. New York state,” New York State Energy
[74] H. Sun, Z. Wang, J. Wang, Z. Huang, N.
Altuve F, “An on-line expert system for fault Res. Develop. Authority, Albany, NY, USA,
Carrington, and J. Liao, “Data-driven power
section diagnosis in power systems,” IEEE Final Rep., Dec. 2014.
outage detection by social sensors,” IEEE
Trans. Power Syst., vol. 12, no. 1, pp. 357–362, Trans. Smart Grid, vol. 7, no. 5, pp. 2516–
Feb. 1997.   [90] C. L. Moreira, F. O. Resende, and J. A. P.
2524, Sep. 2016. Lopes, “Using low voltage microgrids for
[59] C. Fukui and J. Kawakami, “An expert [75] D. Keseris, “Social media could provide early service restoration,” IEEE Trans. Power Syst.,
system for fault section estimation using warning of power outages,” The Telegraph, vol. 22, no. 1, pp. 395–403, Feb. 2007.
information from protective relays and Apr. 2014. [Online]. Available: http://www.
circuit breakers,” IEEE Trans. Power Delivery,   [91] F. O. Resende, N. J. Gil, and J. A. P. Lopes,
telegraph.co.uk/technology/news/10789017/ “Service restoration on distribution
vol. 1, no. 4, pp. 83–90, Oct. 1986. Social-media-could-provide-early-warning- systems using multi-microgrids,” Eur.
[60] Y. Liao, “Identification of faulted feeder of-power-outages.html Trans. Electr. Power, vol. 21, no. 2,
section in distribution systems using [76] K. Sun, Q. Chen, and Z. Gao, “An automatic pp. 1327–1342, Mar. 2011.
overcurrent information from switching faulted line section location method for
devices,” Int. J. Emerg. Electr. Power Syst., electric power distribution systems based on   [92] J. Li, X.-Y. Ma, C.-C. Liu, and K. P.
vol. 15, no. 2, pp. 171–176, 2014. multisource information,” IEEE Trans. Power Schneider, “Distribution system restoration
Delivery, vol. 31, no. 4, pp. 1542–1551, with microgrids using spanning tree
[61] J.-H. Teng, W.-H. Huang, and S.-W. Luan,
Aug. 2016. search,” IEEE Trans. Power Syst., vol. 29,
“Automatic and fast faulted line-section
no. 6, pp. 3021–3029, Nov. 2014.
location method for distribution systems [77] Y. Liu and N. N. Schulz, “Knowledge-based
based on fault indicators,” IEEE Trans. Power system for distribution system outage   [93] Y. Xu, C.-C. Liu, K. P. Schneider, and
Syst., vol. 29, no. 4, pp. 1653–1662, Jul. 2014. locating using comprehensive information,” D. T. Ton, “Placement of remote-controlled
IEEE Trans. Power Syst., vol. 17, no. 2, switches to enhance distribution system
[62] M. Kezunovic, “Smart fault location for
pp. 451–456, May 2002. restoration capability,” IEEE Trans. Power
smart grids,” IEEE Trans. Smart Grid, vol. 2,
[78] Y. Jiang, C.-C. Liu, M. Diedesch, E. Lee, and Syst., vol. 31, no. 2, pp. 1139–1150,
no. 1, pp. 11–22, Mar. 2011.
A. K. Srivastava, “Outage management of Mar. 2016.
[63] Y. Gong and A. Guzmán, “Integrated fault
location system for power distribution distribution systems incorporating   [94] S. Mohagheghi and F. Yang, “Application of
feeders,” IEEE Trans. Ind. Appl., vol. 49, no. 3, information from smart meters,” IEEE Trans. microgrids in distribution system service
Power Syst., vol. 31, no. 5, pp. 4144–4154, restoration,” in Proc. IEEE Innov. Smart Grid
pp. 1071–1078, May/Jun. 2013.
Sep. 2016. Technol., Jan. 2011, pp. 1–7.
[64] Y. Liao, “Generalized fault-location methods
[79] M. Stone, “The opinion pool,” Ann. Math.   [95] A. Castillo, “Microgrid provision of
for overhead electric distribution systems,” Stat., vol. 32, no. 4, pp. 1339–1342, 1961.
IEEE Trans. Power Delivery, vol. 26, no. 1, blackstart in disaster recovery for power
pp. 53–64, Jan. 2011. [80] J. Manyika and H. Durrant-Whyte, Data system restoration,” in Proc. IEEE Smart
Fusion Sensor Management: A Decentralized Grid Commun., Oct. 2013, pp. 534–539.
[65] N. Kang, M. J. Mousavi, and M. Zarghami, Information-Theoretic Approach, ser. Ellis
“Faulted segment identification in   [96] C. Chen, J. Wang, F. Qiu, and D. Zhao,
Horwood Series in Electrical and Electronic “Resilient distribution system by
multiphase distribution systems using Engineering. Upper Saddle River, NJ, USA:
sequence component modelling,” Electr. microgrids formation after natural
Prentice-Hall, 1995. disasters,” IEEE Trans. Smart Grid, vol. 7,
Power Compon. Syst., vol. 44, no. 1,
pp. 18–28, Nov. 2015. [81] “Microgrids for disaster preparedness and no. 2, pp. 958–966, Mar. 2016.
recovery: With electricity continuity plans
[66] S.-J. Chen, T.-S. Zhan, C.-H. Huang, J.-L.   [97] Z. Wang and J. Wang, “Self-healing
and systems,” International Electrotechnical
Chen, and C.-H. Lin, “Nontechnical loss and Commission (IEC), white paper, Mar. 2014. resilient distribution systems based on
outage detection using fractional-order self- [Online]. Available: http://www.iec.ch/ sectionalization into microgrids,” IEEE
synchronization error-based fuzzy Petri nets whitepaper/microgrids/ Trans. Power Syst., vol. 30, no. 6, pp. 3139–
in micro-distribution systems,” IEEE Trans. 3149, Nov. 2015.
[82] J. A. P. Lopes, C. L. Moreira, and A. G.
Smart Grid, vol. 6, no. 1, pp. 411–420, Jan. Madureira, “Defining control strategies for   [98] I. Watanabe and M. Nodu, “A genetic
2015. microgrids islanded operation,” IEEE Trans. algorithm for optimizing switching
[67] Y. Liu and N. N. Schulz, “Integrated fuzzy Power Syst., vol. 21, no. 2, pp. 916–924, sequence of service restoration in
filter for distribution outage information,” May 2006. distribution systems,” in Proc. Congr. Evol.
Electr. Power Syst. Res., vol. 63, no. 3, [83] A. Kwasinski, V. Krishnamurthy, J. Song, and Comput., 2004, pp. 1683–1690.
pp. 177–184, Oct. 2002. R. Sharma, “Availability evaluation of micro-   [99] R. Perez-Guerrero, G. T. Heydt, N. J. Jack,
[68] K. Sridharan and N. N. Schulz, “Outage grids for resistant power supply during B. K. Keel, and A. R. Castelhano, “Optimal
management through AMR system using an natural disasters,” IEEE Trans. Smart Grid, restoration of distribution systems using
intelligent data filter,” IEEE Trans. Power vol. 3, no. 4, pp. 2007–2018, Dec. 2012. dynamic programming,” IEEE Trans. Power
Delivery, vol. 16, no. 4, pp. 669–675, Oct. [84] K. P. Schneider, F. K. Tuffner, M. A. Delivery, vol. 23, no. 3, pp. 1589–1596, Jul.
2001. Elizondo, C.-C. Liu, Y. Xu, and D. Ton, 2008.
[69] F. C. L. Trindade, W. Freitas, and J. C. M. “Evaluating the feasibility to use microgrids [100] T. T. H. Pham, Y. Besanger, and N. Hadjsaid,
Vieira, “Fault location in distribution as a resiliency resource,” IEEE Trans. Smart “New challenges in power system
systems based on smart feeder meters,” Grid, vol. 8, no. 2, pp. 687–696, Mar. 2017. restoration with large scale of dispersed
IEEE Trans. Power Delivery, vol. 29, no. 1,   [85] C. Abbey et al., “Powering through the storm: generation insertion,” IEEE Trans. Power
pp. 251–260, Feb. 2014. Microgrids operation for more efficient Syst., vol. 24, no. 1, pp. 398–406, Feb. 2009.

Proceedings of the IEEE 21

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Chen et al .: Modernizing Distribution System Restoration to Achieve Grid Resiliency Against Extreme Weather Events

[101] B. Chen, C. Chen, J. Wang, and K. Butler- IEEE Trans. Power Delivery, vol. 4, no. 2, “Evaluating the magnitude and duration
Purry, “Multi-time step service restoration pp. 1401–1407, Apr. 1989. of cold load pick-up on residential
for advanced distribution systems and [107] H.-G. Yeh, D. F. Gayme, and S. H. Low, distribution using multi-state load
microgrids,” IEEE Trans. Smart Grid, to be “Adaptive VAR control for distribution models,” IEEE Trans. Power Syst., vol. 31,
published. circuits with photovoltaic generators,” no. 5, pp. 3765–3774, Sep. 2016.
[102] D. Barrett, “Few big FEMA generators IEEE Trans. Power Syst., vol. 27, no. 3, [113] V. Kumar, R. K. H. C, I. Gupta, and H. O.
humming,” Wall Street J., Nov. 2012. pp. 1656–1663, Aug. 2012. Gupta, “DG integrated approach for service
[Online]. Available: http://www.wsj.com/ restoration under cold load pickup,” IEEE
[108] E. Dall’Anese, H. Zhu, and G. B. Giannakis,
articles/SB10001424052970204707104578 Trans. Power Delivery, vol. 25, no. 1, pp.
093192471666514 “Distributed optimal power flow for smart
microgrids,” IEEE Trans. Smart Grid, vol. 4, 398–406, Jan. 2010.
[103] S. Lei, J. Wang, C. Chen, and Y. Hou, no. 3, pp. 1464–1475, Sep. 2013. [114] S. M. George et al., “Distressnet: A
“Mobile emergency generator pre- wireless ad hoc and sensor network
positioning and real-time allocation for [109] H. Ahmadi, J. R. Martí, and A. von Meier,
architecture for situation management in
resilient response to natural disasters,” “A linear power flow formulation for three-
disaster response,” IEEE Commun. Mag.,
IEEE Trans. Smart Grid, 2016, DOI: 10.1109/ phase distribution systems,” IEEE Trans.
vol. 48, no. 3, pp. 128–136, Mar. 2010.
TSG.2016.2605692. Power Syst. vol. 31, no. 6, pp. 5012–5021,
Nov. 2016. [115] R. Olfati-Saber, J. A. Fax, and R. M.
[104] W. H. Kersting, Distribution System Murray, “Consensus and cooperation in
Modeling and Analysis. Boca Raton, FL, [110] L. Gan and S. H. Low, “Convex relaxations networked multiagent systems,” Proc. IEEE,
USA: CRC Press, 2012. and linear approximation for optimal power vol. 95, no. 1, pp. 215–233, Jan. 2007.
flow in multiphase radial networks,” in Proc.
[105] M. E. Baran and F. F. Wu, “Optimal sizing [116] IEEE Guide for Design, Operation, and
of capacitors placed on a radial distribution Power Syst. Comput. Conf., 2014, pp. 1–9.
Integration of Distributed Resources Island
system,” IEEE Trans. Power Delivery, vol. 4, [111] F. Friend, “Cold load pickup issues,” in Systems With Electric Power Systems,
no. 1, pp. 735–743, Jan. 1989. Proc. 62nd Annu. Conf. Protective Relay Eng., Coordinating Committee 21 on Fuel Cells,
[106] M. E. Baran and F. F. Wu, “Network 2009, pp. 176–187. Photovoltaics, Dispersed Generation, and
reconfiguration in distribution systems for [112] K. P. Schneider, E. Sortomme, S. S. Energy Storage, IEEE Standards 1547.4-
loss reduction and load balancing,” Venkata, M. T. Miller, and L. Ponder, 2011, Jul. 2011.

ABOUT THE AUTHORS

Chen Chen (Member, IEEE) received the B.S. Associate Editor of the Journal of Energy Engineering and an editorial
and M.S. degrees in electrical engineering from board member of Applied Energy. He is the Editor-in-Chief of the IEEE
Xi’an Jiaotong University, Xi’an, China, in 2006 Transactions on Smart Grid and an IEEE PES Distinguished Lecturer. He is
and 2009, respectively, and the Ph.D. degree in also the recipient of the IEEE PES Power System Operation Committee
electrical engineering from Lehigh University, Prize Paper Award in 2015.
Bethlehem, PA, USA, in 2013.
During 2013–2015, he worked as a Postdoc-
toral Researcher at the Energy Systems Division, Dan Ton received the B.S. degree in electrical
Argonne National Laboratory, Argonne, IL, USA. engineering and the M.S. degree in business
He is currently a Computational Engineer with management from the University of Maryland,
the Energy Systems Division at Argonne National Laboratory. His primary College Park, MD, USA.
research is in optimization, communications and signal processing for He is the Program Manager of Smart Grid
smart electric power systems, cyber–physical system modeling for smart R&D within the U.S. Department of Energy
grids, and power system resilience. (DOE) Office of Electricity Delivery and Energy
Reliability (OE). He is responsible for develop-
Jianhui Wang (Senior Member, IEEE) received ing and implementing a multiyear R&D pro-
the Ph.D. degree in electrical engineering from gram plan for next-generation smart grid tech-
Illinois Institute of Technology, Chicago, IL, USA, nologies to transform the electric grid in the United States, through
in 2007. public/private partnerships. He has served as Acting Deputy Assis-
Currently, he is an Associate Professor with tant Secretary of Power Systems Engineering Division within the U.S.
the Department of Electrical Engineering, South- Department of Energy (DOE) Office of Electricity Delivery and Energy
ern Methodist University, Dallas, TX, USA. He Reliability (OE). In that capacity, he was responsible for managing
also holds a joint appointment as Section Lead development of projects for “next generation” electricity delivery tech-
for Advanced Power Grid Modeling at the Energy nologies and supporting activities to accelerate their introduction to
Systems Division at Argonne National Labora- the marketplace. Key activities in the Power Systems Engineering Divi-
tory, Argonne, IL, USA. He has held visiting positions in Europe, Australia, sion focus on smart grid research and development, energy storage,
and Hong Kong including a VELUX Visiting Professorship at the Technical and cybersecurity for energy delivery systems, all in support of the
University of Denmark (DTU). OE’s mission to drive electric grid modernization and resiliency. Previ-
Dr. Wang is the secretary of the IEEE Power & Energy Society (PES) ously, he managed the Renewable Systems Integration program within
Power System Operations, Planning & Economics Committee. He is an the DOE Solar Energy Technologies Program.

22 Proceedings of the IEEE