IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO.

4, DECEMBER 2012 2079

Optimizing the Roles of Unit and Non-unit Protection

Methods Within DC Microgrids
Steven D. A. Fletcher, Student Member, IEEE, Patrick J. Norman, Stuart J. Galloway, Paul Crolla, Member, IEEE,
and Graeme M. Burt, Member, IEEE

Abstract—The characteristic behavior of physically compact, to the grid, providing better dynamic performance and allowing
multiterminal dc networks under electrical fault conditions can greater use of renewable sources under intermittent conditions.
produce demanding protection requirements. This represents It is also possible to transmit more dc power through a cable of
a significant barrier to more widespread adoption of dc power
distribution for microgrid applications. Protection schemes have a given voltage rating than with ac due to the higher average
been proposed within literature for such networks based around voltage level, which could facilitate a reduction in cable sizes
the use of non-unit protection methods. This paper shows how- (with associated cost savings) [5]. Additionally, dc systems are
ever that there are severe limitations to the effectiveness of such inherently free from skin effect and reactive voltage drop, fur-
schemes when employed for more complex microgrid network ther improving power transfer.
architectures. Even current differential schemes, which offer a
more effective, though costly, protection solution, must be care- However, the lack of effective solutions to date for electrical
fully designed to meet the design requirements resulting from fault protection of dc systems represents a significant barrier to
the unique fault characteristics of dc microgrids. This paper more widespread adoption of dc. The inherent challenges as-
presents a detailed analysis of dc microgrid behavior under fault sociated with dc include protecting against currents with high
conditions, illustrating the challenging protection requirements magnitude and rate of change in a fast but coordinated way, the
and demonstrating the shortcomings of non-unit approaches for
these applications. Whilst the performance requirements for the prevention of significant voltage transients when operating pro-
effective operation of differential schemes in dc microgrids are tection, and the design of dc circuit breaker technologies to op-
shown to be stringent, the authors show how these may be met erate at required speed, voltage and current levels [1], [6]. As
using COTS technologies. The culmination of this work is the a result of these issues, there has been considerable attention in
proposal of a flexible protection scheme design framework for dc recent years on developing novel protection systems for dc ap-
microgrid applications which enables the required levels of fault
discrimination to be achieved whilst minimizing the associated plications.
installation costs. One particular aspect often overlooked in the protection of
dc microgrid networks is the high sensitivity of the network re-
Index Terms—DC power systems, fault currents, microgrid,
power system protection. sponse to fault impedance. This is particularly evident from re-
cent research which proposes the use of a protection scheme
based on non-unit protection techniques without full consid-
I. INTRODUCTION eration of fault impedance [4], [7]. Implementation of such a
scheme on more complex dc microgrid architectures may lead

W ITHIN microgrid systems, dc power distribution has

the potential to offer more efficient interconnection of
distributed energy resources, such as small-scale generation,
to suboptimal fault discrimination (that is, ensuring that only the
local protection operates for a fault at a particular location in the
network) resulting in either longer fault clearance times or the
backup energy storage, and some industrial and sensitive elec- disconnection of larger parts of a network than necessary in the
tronic loads [1], [2]. Using dc distribution it is often possible event of a fault. This will be demonstrated in later sections of
to reduce the number of power conversion stages for both the this paper.
connection of distributed sources and loads, which can increase To achieve greater levels of fault discrimination within these
overall system efficiency [2], [3]. The paralleling of multiple networks, the implementation of a unit protection scheme can be
sources onto a dc bus is more straight forward than for an ac bus, necessary. However, the scope for the implementation of unit
as the requirement for tight frequency regulation of the supply protection is typically limited within distribution systems due
is removed [4]. This enables the faster connection of sources to the additional cost of the necessary communication and relay
technology. On the other hand, the development of smart grid
Manuscript received July 29, 2011; revised December 21, 2011; accepted
and microgrid concepts will lead to an increase in the amount
April 11, 2012. Date of publication May 18, 2012; date of current version De-
cember 28, 2012. This work was supported by the U.K. Engineering and Phys- of sensor and communication infrastructure within distribution
ical Sciences Research Council. Paper no. TSG-00277-2011. and low voltage networks [8], [9]. The purpose of this enhance-
The authors are with the Institute for Energy and Environment, Department
ment in infrastructure is to enable advanced automatic network
of Electronic and Electrical Engineering, University of Strathclyde, Glasgow
G1 1XW, U.K. (e-mail: [email protected]; [email protected]; monitoring and management algorithms to be embedded within
[email protected]; [email protected]; g.burt@eee. the system, to increase usage of intermittent sources and de-
strath.ac.uk).
crease network congestion. The deployment of this advanced
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. infrastructure also provides the opportunity to expand the use
Digital Object Identifier 10.1109/TSG.2012.2198499 of unit protection schemes within microgrids.

1949-3053/$31.00 © 2012 IEEE

2080 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 4, DECEMBER 2012

Fig. 1. Network diagram.

This paper first investigates the capability of non-unit pro- TABLE I

tection methods to achieve effective protection discrimination NETWORK PARAMETERS
within dc microgrid systems by analytically describing the cur-
rent, voltage, , and impedance response for a range
of fault locations and impedances. The particular difficulties of
achieving fast acting and discriminative protection using non-
unit methods are subsequently demonstrated using an analytical
case study. Differential current behavior within dc microgrids is
then analyzed, facilitating the identification of, and the proposal Previous work has highlighted that the fast discharge of
of solutions for, the challenging measurement and signal syn- capacitors used as filters on the dc terminals of the VSC can
chronization requirements for unit schemes. Finally, a protec- damage both the capacitors themselves and any other sensitive
tion system design framework for dc microgrid applications is components in the fault path [7]. Considerable short term
proposed which provides optimal protection discrimination and electromagnetic forces on conductors can also be induced, cre-
operating speed whilst minimizing system installation costs. ating risks of physical damage to mountings or insulation [10].
Furthermore, [6] and [11] illustrate the potential for voltage
II. QUANTIFICATION OF DC PROTECTION SYSTEM reversal if dc side faults are not cleared within an adequate time
OPERATING REQUIREMENTS frame. The voltage reversal can cause significant currents to
Cuzner [1] outline the key design criteria for any protection flow through converter freewheeling diodes, causing damage
system and these relate to the operability and cost of a protec- to these devices.
tion system. From the criteria presented, the performance of the The fault current withstand of VSCs is low compared to more
protection system is assessed on its ability to provide continuity robust thyristor based converter topologies [4], [7], therefore
of supply to loads where other parts of the network are experi- current must be limited or interrupted much more quickly to
encing faults. prevent damage to internal components when supplying fault
The dc microgrid network used as the basis for comparison current.
within this paper is presented within Fig. 1, with Table I pre- The typical topology of VSC devices is such if the back-bi-
senting relevant network parameters. This network has been de- asing dc voltage is lost after the occurrence of a fault, the an-
rived from example architectures within literature [4], [7] and tiparallel diodes across the switching devices will begin to con-
is supplied by a voltage source converter (VSC). Only a single duct, meaning the converter is unable to block the flow of cur-
source has been considered within Fig. 1 to simplify analysis rent to the fault [6], [12]. For these converter types, it neces-
and illustration however it is anticipated that findings will be sary for network protection to act quickly to prevent damaging
applicable to networks with multiple sources. To ensure min- currents from flowing through the diodes, within 2 ms in some
imum disruption to the network presented within Fig. 1 in the cases [4].
event of a fault, protection devices to must operate in Alternative VSC topologies contain their own internal pro-
a coordinated way, such that only the device immediately up- tection functionality, which enables the interruption of current
stream from the fault operates. flow through the converter. An example topology capable of this
However there are a number of factors which influence the is provided within [7], where antiparallel diodes are replaced
time-frame within which the network protection has to coordi- with emitter turn-off devices. Internal converter protection can
nate its devices operation. Many of these factors center around be sensitive to overcurrent, overvoltage or undervoltage [13],
the use of use of a VSC as the main network supply. [14]; however as the only source of fault current within Fig. 1
FLETCHER et al.: OPTIMIZING THE ROLES OF UNIT AND NON-UNIT PROTECTION METHODS WITHIN DC MICROGRIDS 2081

TABLE II overcurrent, highlighting the unique challenges for the type of

REQUIRED TRIPPING TIMES FOR UNDERVOLTAGE THRESHOLD OF 200 V FOR A network considered.
1-m FAULT AT VARIOUS FAULT LOCATIONS
The following sections will demonstrate the challenges in
achieving discriminatory protection system operation within the
time frames outlined using of non-unit methods.

III. ANALYSIS OF NON-UNIT PROTECTION METHOD

EFFECTIVENESS
Non-unit protection does not protect a clearly bounded zone
of the power system and will operate whenever its threshold
is violated; non-unit schemes have inherent backup capabili-
ties and will act to protect the system if a neighboring protec-
it is essential that the converter protection coordinates with pro- tion system fails to operate [16]. Recent research [4], [7] out-
tection devices to to ensure that only the appropriate pro- lines approaches for the use of non-unit based protection scheme
tection device operates prior to converter protection operation. methods on microgrid and other multiterminal dc applications.
Operational standards do exist for ac and HVDC systems Whilst non-unit approaches proved effective for the applica-
which describe the requirements for converter connection in the tion considered, the architecture of these networks was such that
event of network fault conditions. For example, [14] stipulates there was not the requirement to coordinate upstream and down-
that in the event of a network undervoltage, converters are re- stream devices in any significant way.
quired to remain connected for a minimum of 140 ms to avoid This section investigates the capability of non-unit protection
sympathetic tripping for faults elsewhere in the network [15]. methods to achieve effective protection selectivity, within the
However it is difficult to see how these requirements apply to times derived within the previous section, for the more complex
less robust converter types, where connection for this period dc system illustrated within Fig. 1. The section first describes
of time may result in the flow of damaging current magnitudes the current, voltage, , , and impedance profiles as
flowing through the converter. measured at the generator converter terminals to enable a better
Whilst converter undervoltage protection is typically not as understanding of the network fault response. To enable this anal-
important as overcurrent for preventing device damage, for a ysis to be more easily understood and usefully employed, sim-
dc system the undervoltage is a consequence of filter capacitor plifying assumptions have been made, which are based on pre-
discharge, which in itself may cause problems. An undervoltage vious work by the authors and presented in [6]. The key findings
will be followed by an overcurrent condition on the ac side of from this analysis can then be considered for specific protection
the converter, as more current is drawn to attempt to recover schemes and an example case study, based on an overcurrent
the dc voltage. The dc side undervoltage can also be linked to scheme, is presented to illustrate the challenges faced.
the operation of ac side protection, which may monitor both dc
voltage and current to determine its operation [13]. A. Analysis of Dc Microgrid Fault Response
Given that the dc voltage is linked to a number of aspects Under fault conditions, the initial current response of the net-
of the network and converter protection, it is useful to consider work generally takes one of two forms depending on circuit
the voltage response when deriving protection system operating damping conditions [6]. For an underdamped circuit (defined
criteria. The dc voltage response has the added advantage of below), current can be defined as
being least dependent on ac network conditions and configu-
ration, and hence provides a dc side solution which could be (1)
deployed within multiple applications. For these reasons, this
paper assesses the potential for current fault detection methods
where is the initial voltage on filter capacitance . In
to coordinate with a converter undervoltage threshold for the
(1), the term represents the damped resonant frequency and
network described within Fig. 1.
is defined as
To derive a fixed operating point, an undervoltage threshold
of 200 V (half the nominal system voltage) has been selected. It (2)
should however be noted that the observations in the following
sections are relevant for various voltage thresholds. Table II is the damping factor (or Neper frequency) and is defined as
highlights the time at which this voltage threshold is reached
following the occurrence of a 1-m fault at the six fault loca- (3)
tions indicated in Fig. 1.
From Table II it is clear that, for the range of low impedance the term is the resonant radian frequency and is defined as
faults considered, the rapid loss of voltage at the converter ter-
minals creates particularly challenging times for the operation (4)
of protection if it is to act to prevent the undervoltage occur-
ring. The times identified are much shorter than required for ac and and are the total resistance and inductance between the
converter connection [14], although they are in fact similar in discharging capacitor and the fault. Note that (1) assumes that
magnitude to the requirements derived in [4] for prevention of initial current (at ) in the circuit is zero as the dominant
2082 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 4, DECEMBER 2012

part the fault current characteristic is usually due to the initial and
voltage across the converter filter capacitance [6].
The damping conditions in the network depend on the terms (12)
and . The circuit is considered underdamped if ,
and overdamped if . For the overdamped case, current respectively.
is defined as Expressions for the rate of change of voltage are found from
the derivatives of (9) and (10), which are
(5)
(13)
again neglecting initial current, and where
and
(6)
(14)
Current magnitude is often not used in itself to discriminate
fault location but is also coupled with time, with a common respectively.
form being based on the response of the network. The Finally the network impedance under fault conditions can be
expressions for the two damping conditions, neglecting initial found from the division of the voltage expressions given in (9)
current, can be derived by squaring and integrating (1) and (5), and (10) by the equivalent current expressions presented in (1)
respectively. These are and (5). In this manner, the underdamped response impedance
can be expressed as

(15)

and the overdamped impedance is equal to

(7)
(16)

for underdamped conditions and Having derived these expressions it can be seen how the var-
ious circuit parameters shape the transient response. For ex-
ample, shows how the resistance and inductance parameters
affect the exponential decay. illustrates how the combina-
tion of all circuit parameters can affect oscillatory frequency and
hence the peak current time. Finally (6) shows how the damping
(8) and frequency terms equate to give the roots of the characteristic
equation and hence establish the appropriate equations to use in
for overdamped network conditions. the analysis.
Whilst it is more common to use current to detect faults, al- The equations can be analyzed in detail to describe a number
ternative measures such as voltage [17], [18], [19], of aspects of the different fault responses. However, for the pur-
and impedance [20] can be used for fault detection and are there- poses of this paper, there are two key characteristics which the
fore worthy of consideration. Conveniently, these can be de- analysis reveals.
rived from the current response, as described below. First, the equations illustrate the impact of damping on
The voltage response is proportional to the integral of current the network response, emphasizing the high sensitivity of the
and, over the faulted period, can be represented by different responses to fault impedance. This is a result of the
relatively short interconnecting cables (and hence small cable
(9) impedance) within microgrids. The effect of this is that the
response is less dependent on location for impedance faults,
making discrimination more difficult. This leads to the design
for underdamped conditions and
of protection schemes based more on time based grading and
less on magnitude based grading, which is undesirable for the
(10) intended application given the tight requirements on protection
system operating time. Both of these characteristics are exam-
for overdamped conditions. ined in further detail within [21].
To develop expressions for rate of current change, the deriva- One exception to this is the response under initial con-
tives of (1) and (5) are taken which gives ditions, where a similarity exists for both under and overdamped
fault conditions at the same fault location (that is, for variable
impedance fault) [21]. A fault detection method is currently
(11)
being developed by the authors based on this characteristic [22].
FLETCHER et al.: OPTIMIZING THE ROLES OF UNIT AND NON-UNIT PROTECTION METHODS WITHIN DC MICROGRIDS 2083

This approach would however be limited to the protection of a

single line, and so could not replace network wide protection.
Second, a comparison of all of equations, with respect to
damping, highlights that they display similar behavioral char-
acteristics. This is an unsurprising finding given that they are
all derived from the capacitive response. As such, any conclu-
sions regarding the effectiveness of non-unit protection methods
drawn on the basis of current based methods will be relevant for
all potential non-unit measurands.
The following section will demonstrate the challenges in im-
plementing overcurrent techniques to provide effective protec-
tion to the network illustrated in Fig. 1, looking specifically at
the network’s current and responses to a range of fault con-
ditions.

B. Illustration of Detection Challenges Based on an All

Overcurrent Protection Scheme
Fig. 2. Current (top) and (bottom) response for 1-m (left) and 500-m
To assess the capabilities of an overcurrent protection scheme (right) faults at .
to deliver the required levels of performance, this section looks
at the coordination of pairs of upstream and parallel downstream
devices, relating them to the previously derived operating re- representative of low and high impedance fault conditions. It
quirements, and highlighting how these operating requirements can be seen from Fig. 2 that for the two fault types, the peak
differ depending on the connection of downstream devices. The fault current is vastly different, emphasizing the dominance of
merits of specific current-time graded protection schemes are the fault impedance relative to the total fault path impedance.
not analyzed, as is perhaps more standard, as the authors be- However in both cases the steady state output of the converter
lieve the issues are more clearly demonstrated with a study of tends to the same level as the converter attempts to maintain
network response rather than detailed device characteristics. It output voltage to nominal levels. The magnitude of this steady
is worth noting however that a relay operated on the extremely state current will depend on either ac side fault level or con-
inverse current-time characteristic (designed for fast operating verter rating (if the converter is capable of limiting current for
conditions) would behave in a similar manner to that a device dc faults). Whilst the network voltage will not decay to the
operated on [16]. Furthermore, [23] does discuss the poten- same extent for impedance fault conditions (and potentially not
tial issues in coordinating current-time characteristics for net- reducing below the defined voltage threshold), the operating
works with large capacitive sources. requirement relates to the converter’s ability to supply this fault
Whilst it is standard practice to coordinate protection device current.
operation beginning with the furthest downstream device, the This causes a problem in setting the overcurrent threshold for
section instead first assesses the coordination of upstream de- . For example, if an initial threshold is set for as the at
vices because of the challenges associated with operating close the undervoltage threshold (set in the previous section as 0.9 ms,
to the capacitive source and the impact this has on downstream at which point equals ), expanding the plot
protection operation. These challenges are illustrated in the fol- for the 500-m fault within Fig. 2 will show that it takes 1.18 s
lowing sections. after fault inception to reach the same value. This would
1) Coordination of With and : To achieve good lead to the converter supplying fault current for longer than de-
performance when coordinating with and , the protec- sired, and hence there is a requirement to lower this operating
tion system must ensure that: any faults on line are quickly threshold from this initial level. However to maintain coordina-
discriminated and cleared and remains stable for faults on tion with downstream devices, there is a limited degree to which
downstream lines but provides backup in the event that or this can be achieved. To assess the scope for the reduction, con-
fail to operate. sider the current and for 1 m fault at and shown in
As will be shown in later figures, the detection and discrim- Fig. 3. Note that due to faults and being the same distance
ination of a low impedance fault at is reasonably straight from the converter, and suitably low impedance, the responses
forward given the excessive overcurrent produced compared to to either fault is equivalent.
more distant faults. Therefore the objective for the protection From a comparison of Fig. 3 and Table II it can be determined
system for close up faults is to operate sufficiently quickly to that the undervoltage threshold crossing at 2.2 ms corresponds
prevent damage at the point of fault and to components sup- to an value of . Relating this value to the previous
plying fault current. Instead, the key coordination challenge in fault case, is reached 0.16 s following the inception
setting the overcurrent threshold at relates to the network of fault . Whilst this is perhaps longer than is desir-
fault response for higher impedance faults. To illustrate why this able, it is reasonable to assume that the converter could supply
is the case, consider the plot shown in Fig. 2. current for this shorter time given the slower decay of dc side
Fig. 2 illustrates the response of the network to 1- and voltage. Therefore one protection setting option would be to re-
500-m faults at , values which have been chosen to be duce the threshold at to this level. However to maintain a
2084 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 4, DECEMBER 2012

TABLE IV
SUMMARY OF OPERATING THRESHOLD TIMES OF , ,
AND FOR A FAULT AT OR

TABLE V
SUMMARY OF OPERATING THRESHOLD TIMES OF
AND FOR A FAULT AT

Fig. 3. Current (left) and (right) response for 1-m fault at and .

TABLE III
SUMMARY OF OPERATING THRESHOLD TIMES OF , ,
AND FOR A FAULT AT OR the undervoltage cut off, is . As this is greater than
the revised threshold for in the previous section, there is
a need to reduce this level. To maintain consistency with the
previous section, the threshold for has been reduced to
(half that of ). A summary of the impact of
this on required operating time and time margins is shown in
Table IV.
suitable time margin between the operating points of upstream Table IV highlights that the difference in required operating
and downstream protection (to enable device coordination), it is time for the initial undervoltage thresholds is already very tight
also necessary to reduce the thresholds of and . This how- and the impact of the reduced operating threshold compounds
ever brings its own problems given the need for to coordinate this problem, making the setting of devices extremely difficult.
with further downstream devices and hence reducing the scope As with the previous case, given that the required operating time
for threshold reduction. The necessity to reduce thresholds to is already small, there is little scope for accelerating protection
achieve acceptable operating times does indicate that options to operation through threshold reduction. However for complete-
ride through the initial capacitive discharge, as suggested in [1] ness, and to quantify challenges further downstream protection,
and [23], are limited. the following section assesses the options for coordination of
To continue this example, consider the potential for circuit with and .
breaker coordination when reducing the threshold setting of 3) Coordination of With and : In a similar manner
and to (half the original setting). Table III sum- to the previous section, the potential for device coordination
marizes the times at which the thresholds will be reached for the is assessed through comparison of the initial and revised over-
initial and revised protection settings. current thresholds. The initial threshold for or was
Table III highlights that whilst the initial protection settings , which is again greater than revised upstream
were challenging to meet because of the short time frame, a levels, and so in line with previous sections the it has been
sufficient time margin existed between upstream and down- reduced to (half of ). A summary of initial and
stream protection to ensure coordinated protection operation. revised operating times for a fault at is presented in Table V.
However given the requirement to reduce the upstream Table V shows a similar trend to the previous section in terms
threshold to achieve reasonable operating times, the subsequent of both required operating time and time difference between up-
impact means that the time margin between device operations stream and downstream devices. Therefore the device coordina-
has reduced to a level such that on protection coordination is tion challenges are similar to those reported previously.
extremely difficult to achieve. This is in part due to the typical
delay time between detection and circuit breaker operation, and C. Discussion of Results
this is discussed in more depth in later a section. The results presented in the previous sections have demon-
In order to increase the time margin between different de- strated the significant challenges which exist in the coordina-
vice operations, there may be some scope for reduction in the tion of protection in compact dc power systems, within the time
threshold of , albeit limited, given that it does not need to co- constraints imposed by converter interfaces, using overcurrent
ordinate with further downstream devices. This is not the case based protection schemes.
for , so further reduction in its threshold is not necessarily In each scenario it was illustrated that the time margin be-
an option. To quantify this, the following sections investigate tween upstream and downstream protection operation was pro-
the impact of upstream device coordination on the response of hibitively small, creating a risk of upstream protection opera-
downstream protection. tion for downstream faults. This was in part due to the tight op-
2) Coordination of With and : The potential for erating requirements from the network voltage response. How-
threshold reduction can be examined from analysis of down- ever the need for reduction in the threshold of (to achieve
stream faults . The initial threshold for , derived from reasonable operating times under impedance fault conditions)
FLETCHER et al.: OPTIMIZING THE ROLES OF UNIT AND NON-UNIT PROTECTION METHODS WITHIN DC MICROGRIDS 2085

has a cascading effect on the downstream device settings and major challenge which remains for the implementation of a dif-
hence reduces operating margins. From this, it is worth noting ferential protection scheme within dc microgrids is the capa-
in Table IV and V the time difference between the initial up- bility to achieve fault detection within the time frame described
stream and the revised downstream threshold is twice that of the in previous sections.
difference between the two initial settings. This suggests that if In addition to the technical constraints, the economics of the
the constraint of lowering the upstream threshold is removed, a protection system implementation must also be considered [1].
greater opportunity for device coordination exists, provided op- The requirements for device communication means the instal-
erating requirements are met. lation costs of a current differential scheme can be substantial
It is also worth considering how the difference in required op- compared to an overcurrent scheme, particularly at the distri-
erating time compares to that of the physical operating speeds bution level, although this might be somewhat reduced for mi-
of circuit breakers. Previous work by the authors [6] highlights crogrid systems due to their relatively small geographical area.
that the requirement for fast acting protection can limit the range However, given the likely increase in the amount of sensor and
of protection devices which can be employed in dc microgrid communication infrastructure within distribution networks [9]
networks. For example, the operating time of dc electro-me- as smart grid and microgrid [8] concepts develop, much of the
chanical circuit breakers (EMCBs) can be from around 3 ms required infrastructure may already be in place. The deployment
[24], which exceeds the time difference in the scenarios de- of this advanced infrastructure therefore provides an opportu-
scribed in the previous section, meaning coordination is not nec- nity to expand the use of unit (communication-based) protection
essarily possible using the methods presented. Solid state and schemes within microgrids.
hybrid circuit breaker technologies offer a potential alternative To explore this opportunity, this section considers two as-
to EMCBs, as these breakers can operate extremely quickly after pects to the implementation of current differential protection.
the detection of a fault, however there are greater limitations on The first aims to quantify these implementation challenges and
the operating voltage and current levels of these devices than the second considers the impact of unit protection implemen-
for EMCBs. tation on overall protection. These aspects are described in the
It must also be appreciated that dc current breaking cannot following subsections.
be achieved instantaneously and there is a finite time when cur-
rent is driven to zero. During this period current will continue to A. Challenges and Proposed Solutions for the Implementation
flow through upstream devices [25] and this could, depending of Fast Acting Unit Protection Schemes
on network conditions, cause an upstream device to operate be- There are two main challenges for the implementation of unit
fore the fault is fully cleared. protection within the highly transient environment described in
Acknowledging these shortfalls, it can be concluded that the the previous sections. The first is, can currents be compared and
non-unit methods analyzed are unlikely to be able to achieve the fault location determined within the required time frame? The
required levels of protection discrimination for the derived op- second is, can the current measurements at different points in the
erating requirements. Within future dc networks it is likely that network be accurately synchronized to ensure correct protection
a higher level of fault discrimination will be desirable, particu- system operation? These issues are discussed in the following
larly if dc is to be proven a viable alternative to ac distribution. sections.
For these future networks, it has been demonstrated that for this 1) Challenges in Meeting the Operating Requirement: Cur-
to be achieved, more robust protection approaches are required. rent differential protection applied in ac systems typically has a
The following section investigates the potential for unit protec- target operation time of 1–2 cycles, which often represents an
tion schemes to provide this required protection performance operation time of 20 ms [28], [29]. In comparison, the oper-
and proposes commercial off the shelf (COTS) based solutions ating times derived in Section II are much shorter and hence
for the key challenges identified. This analysis subsequently en- alternative implementation methods must be deployed in order
ables the development of a protection framework for dc micro- to meet these operating times.
grid applications. One factor which prevents the reduction in operating time
of an ac current differential system is the requirement for indi-
IV. APPLICATION OF UNIT PROTECTION vidual phase current measurement and phasor comparison [26].
Unit protection protects a clearly bounded zone of the power This requirement does not exist for dc implementation, where
system and will not operate for faults external to this zone. In only current magnitudes need to be compared. Furthermore, as
contrast to non-unit schemes, it does not provide backup to ad- dc current will be measured using a current transducer (such as
jacent elements of the system [26] hence non-unit protection is a Hall Effect device) rather than via a current transformer, the
often deployed alongside unit protection to provide the neces- measurement will be in the form of voltage which facilitates
sary backup functionality. A common form of unit protection easier integration with digital processing devices. As [6] sug-
is current differential protection, which operates by comparing gests, the sampling rates available on modern microprocessors
all currents’ magnitudes and/or relative directions at the bound- are such that numerous current samples could be summed and
aries of a specified element within a network [26]. a fault detected well within the derived operating requirements.
The nature of the current differential protection method is Therefore the dominant factor for fault detection within dc net-
such that it is far less susceptible to the effects of variable fault works is the communication delay between different devices. A
levels and impedances than non-unit methods [27], facilitating dc current differential scheme for the network described within
effective protection coordination in the network. However, the this paper must therefore be supported by a high bandwidth
2086 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 4, DECEMBER 2012

recommended for feeders where coordination with downstream

devices is likely to be difficult within an acceptable time frame,
but non-unit protection is recommended for all other cases in
order to balance required system performance against installa-
tion cost. This framework is applicable to dc microgrids of var-
ious configurations.
To illustrate the underlying philosophy of this scheme, con-
sider the example microgrid network illustrated in Fig. 1. If
a current differential scheme were to be applied between the
supply converter output and the first parallel connection point
(prior to and ) in place of an overcurrent scheme, it would
serve to remove the constraint of reducing the threshold to
achieve acceptable operating times under impedance fault con-
ditions. This could be achieved as the unit protection zone would
be insensitive to external faults and hence not operate even with
Fig. 4. Protection scheme approach decision tree. high current throughput. The subsequent effect of this would be
to enable the remainder of the protection settings within the net-
work to return to the initial values derived from the time of un-
communications system in order to operate effectively within dervoltage, increasing the time margin between the operation of
the derived operating times. Exact specification of these require- different devices. However this still leaves very tight operating
ments is an area of future work. time requirements, particularly where devices have to coordi-
2) Challenges in the Implementation of Unit Systems When nate with other downstream protection devices.
Operating Under High Rate of Change Fault Conditions: To Additionally, Section III-B shows that the time margin be-
avoid the occurrence of errors in the current differential sum, tween adjacent devices from onwards is similar, and this is
and hence ensure that the scheme performs correctly, current due to the uniform fault separation and cable parameters within
measurements must be synchronized in time [26]. There are a the network. To adhere to requirement of operating protection
number of potential sources of poor time synchronization. These prior to a network undervoltage, it has been shown that the only
include timing errors between communicating devices and non- means of increasing this time margin is to decrease the down-
synchronous current sampling. Given the high rate of current stream threshold. This is possible between and , however
change in the fault response considered within this paper (ini- due to the connection of additional parallel loads downstream
tially 17 A s for a fault at for example), accurate de- for , potential reduction in the overcurrent threshold at
vice synchronization becomes even more important to ensure is limited. The application of unit protection at each of these
the protection scheme does not mal-operate. parallel connection points would not only ensure accurate fault
To minimize the impact of this on the protection system oper- detection for internal zone faults but also that there is sufficient
ation in practice, protection relays often communicate both cur- time available for the operation of protection devices for load
rent and time information to ensure current measurements are connection points. Protection of these parts of the network could
compared in the same point in time. Synchronization using this be achieved through the use of simpler non-unit techniques.
approach can be most effectively achieved where devices are
synchronized through GPS time stamping [28], where devices V. CONCLUSION
are synchronized with an accuracy of . It is anticipated The development of effective protection system solutions is a
that this degree of accuracy would be sufficient however if the critical step in the development of high performance dc micro-
time difference was to increase then there is a possibility of large grid systems. The key contribution of this paper is to identify
errors occurring within the differential sum. the means with which to achieve fast and effective protection
system operation, whilst seeking to minimize installation costs,
B. Protection System Design Framework for DC Microgrid against a set of very strict operating requirements. The paper has
Applications demonstrated the limitations of non-unit protection methods to
In this section, a protection system design framework is achieve effective fault discrimination within derived operating
proposed which provides clear design guidance with which to times and the use of current differential protection is introduced
achieve fast and effective protection system operation, whilst as a potential solution. However this has implications for system
seeking to minimize installation costs, against a set of very cost and complexity. Following the analysis of these protec-
strict operating requirements. It achieves this by proposing the tion methods, the potential roles of unit and non-unit protection
use of unit protection upstream within a network in order to methods are defined within the example microgrid network. Ex-
ease the constraints on downstream non-unit protection. Fig. 4 trapolating this analysis, a design framework is proposed for dc
provides a pictorial overview of the design framework. microgrid systems which provides a means of optimizing pro-
The proposed framework provides guidance on the required tection scheme design to achieve required fault discrimination
protection approach for a feeder based on its position within and operating speed in an economic way. Whilst of simple ap-
the network and the surrounding elements of this network. In pearance, the clarity provided by this framework is only possible
this manner, unit protection methods (with non-unit backup) are as a result of the underpinning analysis.
FLETCHER et al.: OPTIMIZING THE ROLES OF UNIT AND NON-UNIT PROTECTION METHODS WITHIN DC MICROGRIDS 2087

ACKNOWLEDGMENT [23] R. Cuzner, D. MacFarlin, D. Clinger, M. Rumney, and G. Castles, “Cir-

cuit breaker protection considerations in power converter-fed DC sys-
tems,” in IEEE Elect. Ship Tech. Symp., Apr. 2009, pp. 360–367.
This work has been carried out as part of the Rolls-Royce [24] “Secheron high-speed dc circuit-breaker for rolling stock type UR26,”
UTC program. 2011. [Online]. Available: http://www.secheron.com
[25] S. D. A. Fletcher, P. Norman, S. Galloway, and G. Burt, “Solid state
circuit breakers enabling optimised protection of dc aircraft power sys-
tems,” presented at the 14th Eur. Conf. Power Electron. Appl. (EPE),
Birmingham, U.K., Sep. 2011.
REFERENCES [26] “Unit protection of feeders,” in Network Protection and Automation
Guide ALSTOM, 10 [Online]. Available: www.alstom.com/grid
[1] R. Cuzner and G. Venkataramanan, “The status of DC micro-grid pro- [27] P. Karlsson and J. Svensson, “Fault detection and clearance in DC dis-
tection,” in Proc. IEEE Ind. Appl. Soc. Annu. Meeting (IAS), Oct. 2008, tributed systems,” presented at the Nordic Worskshop Power Ind. Elec-
pp. 1–8. tron., Stockholm, Sweden, Aug. 2002.
[2] K. George, EPRI Solutions, Inc., “DC power production, delivery and [28] N. Villamagna and P. Crossley, “A symmetrical component-based GPS
utilization: An EPRI white paper,” 2006. signal failure-detection algorithm for use in feeder current differential
[3] K. Shenai and K. Shah, “Smart DC micro-grid for efficient utilization protection,” IEEE Trans. Power Del., vol. 23, no. 4, pp. 1821–1828,
of distributed renewable energy,” presented at the IEEE Energytech, Oct. 2008.
Cleveland, OH, May 2011. [29] SEL-inc, “SEL-311L line current differential protection and automa-
[4] D. Salomonsson, L. Soder, and A. Sannino, “Protection of low-voltage tion system,” 2011. [Online]. Available: http://www.selinc.com
DC microgrids,” IEEE Trans. Power Del., vol. 24, no. 3, pp.
1045–1053, Jul. 2009.
Steven D. A. Fletcher received the B.Eng. (Hons) degree in electrical and elec-
[5] T. Kaipia, P. Salonen, J. Lassila, and J. Partanen, “Application of
low voltage DC-distribution system—A techno-economical study,” tronic engineering from the University of Strathclyde, Glasgow, U.K., in 2007,
presented at the 19th Int. Conf. Elect. Distrib., Vienna, Austria, May where he is currently working toward the Ph.D. degree.
2007. He is currently a Research Assistant with the Institute for Energy and En-
[6] S. Fletcher, P. Norman, S. Galloway, and G. Burt, “Determination of vironment, University of Strathclyde. His research interests include the tran-
protection system requirements for DC unmanned aerial vehicle elec- sient analysis and protection of dc systems for microgrid, marine, and aerospace
trical power networks for enhanced capability and survivability,” IET applications.
Elect. Syst. Transp., vol. 1, no. 4, pp. 137–147, 2011.
[7] M. E. Baran and N. R. Mahajan, “Overcurrent protection on voltage-
source-converter-based multiterminal dc distribution systems,” IEEE
Trans. Power Del., vol. 22, no. 1, pp. 406–412, Jan. 2007.
Patrick J. Norman received the B.Eng. (Hons) degree in electrical and mechan-
[8] G. Venkataramanan and C. Marnay, “A larger role for microgrids,”
ical engineering and Ph.D. degree in electrical engineering from the University
IEEE Power Energy Mag., vol. 6, no. 3, pp. 78–82, May-Jun. 2008.
[9] V. Sood, D. Fischer, J. Eklund, and T. Brown, “Developing a commu- of Strathclyde, Glasgow, U.K.
nication infrastructure for the smart grid,” in Proc. IEEE Elect. Power He is a Full-Time Research Staff Member with the Institute for Energy
Energy Conf. (EPEC), Oct. 2009, pp. 1–7. and Environment, University of Strathclyde. He has published over 35 peer
[10] Short-Circuit Currents in DC Auxiliary Installations in Power Plants reviewed journal and conference publications. His current research interests lie
and sSubstations—Part 2: Calculation of eEffects, IEC 61660-2:1997, in the modeling and simulation, design, control, protection of aircraft secondary
International Electrotechnical Commission, 1997. power offtake and distribution systems, and microgrid and shipboard power
[11] J. Yang, J. Fletcher, and J. O’Reilly, “Multiterminal dc wind farm col- systems.
lection grid internal fault analysis and protection design,” IEEE Trans.
Power Del., vol. 25, no. 4, pp. 2308–2318, Oct. 2010.
[12] L. Tang and B.-T. Ooi, “Locating and isolating DC faults in multi-
terminal DC systems,” IEEE Trans. Power Del., vol. 22, no. 3, pp.
Stuart J. Galloway received the Bachelors degree in mathematical sciences
1877–1884, Jul. 2007.
[13] J. Candelaria and J.-D. Park, “Vsc-hvdc system protection: A review of from the University of Paisley, Paisley, U.K., in 1992, and the M.Sc. degree in
current methods,” in Proc. IEEE/PES Power Syst. Conf. Expo. (PSCE), nonlinear modeling and Ph.D. degree in numerical analysis from the University
Mar. 2011, pp. 1–7. of Edinburgh, Edinburgh, U.K., in 1993 and 1998, respectively.
[14] National Grid Electricity Transmission plc, “The grid code,” no. 4, He is currently a Senior Lecturer with the Institute for Energy and Environ-
revision 1, May 16, 2010 [Online]. Available: http://www.national- ment, University of Strathclyde. Since 1998, he has been researching optimiza-
grid.com/uk/Electricity/Codes/gridcode/ tion problems in power systems, electricity markets, and novel electrical archi-
[15] K. I. Jennett, C. D. Booth, and L. Martin, “Analysis of the sympathetic tectures relating to aero and marine electrical systems.
tripping problem for networks with high penetrations of distributed
generation,” presented at the Int. Conf. Adv. Power Syst. Autom. Prot.,
Beijing, China, Oct. 2011.
[16] “Overcurrent protection for phase and earth faults,” in Network Protec-
Paul Crolla (M’06) received the M.Sc. degree in physics from the University
tion and Automation Guide ALSTOM, 2011, ch. 9 [Online]. Available:
www.alstom.com/grid of Strathclyde, Glasgow, U.K., in 2005.
[17] R. M. Tumilty, M. Brucoli, G. M. Burt, and T. C. Green, “Approaches He is currently a Research Assistant with the Protection, Automation and
to network protection for inverter dominated electrical distribution sys- Control Team, Institute for Energy and Environment, University of Strathclyde.
tems,” in Proc. IET 3rd Int. Conf. Power Electron., Mach. Drives, Mar. His main research interests lie in the development of power hardware-in-the-
2006, pp. 622–626. loop applications and power systems laboratory demonstration.
[18] E. Cinieri, A. Fumi, V. Salvatori, and C. Spalvieri, “A new high-speed
digital relay protection of the 3-kvdc electric railway lines,” IEEE
Trans. Power Del., vol. 22, no. 4, pp. 2262–2270, Oct. 2007.
[19] S. Foster, L. Xu, and B. Fox, “Control of an LCC HVDC system for Graeme M. Burt (M’95) received the B.Eng. degree in electrical and electronic
connecting large offshore wind farms with special consideration of grid engineering and the Ph.D. degree from the University of Strathclyde, Glasgow,
fault,” in IEEE Power Energy Soc. Gen. Meeting, Jul. 2008, pp. 1–8.
U.K., in 1988 and 1992, respectively, following research into fault diagnostic
[20] “Distance protection,” in Network Protection and Automation Guide
ALSTOM, 11 [Online]. Available: www.alstom.com/grid techniques for power networks.
[21] S. Fletcher, P. Norman, S. Galloway, and G. Burt, “Analysis of the He is currently a Director of the Institute for Energy and Environment, Uni-
effectiveness of non-unit protection methods within dc microgrids,” versity of Strathclyde, where he also directs the University Technology Centre
presented at the IET Conf. Renew. Power Gen., Edinburgh, U.K., Sep. in Electrical Power Systems sponsored by Rolls-Royce. He is a Professor of
2011. electrical power engineering, and has particular research interests in the areas
[22] S. D. A. Fletcher, P. J. Norman, S. J. Galloway, and J. E. Hill, “Protec- of integration of distributed generation, power system modeling and real-time
tion system for an electrical power network,” U.K. Patent application simulation, power system protection and control, microgrids, and more-electric
GB1102031.0, Feb. 2011. systems.