IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 36, NO.

5, SEPTEMBER/OCTOBER 2000

1235

Time Coordination Method for Power System

Protection by Evolutionary Algorithm
C. W. So and K. K. Li

AbstractA typical industrial power network may consist of

hundreds of pieces of equipment and even more protection relays
to protect the system are required. Each protection relay in the
power network needs to be coordinated with the relays protecting
the adjacent equipment. The overall protection coordination is,
thus, very complicated. A new concept of protection coordination
by time is introduced in this paper to formulate all the system relays
and system equipment operation into a set of optimization equations and constraints. Its purpose is to search for an optimal protection setting to minimize the system disturbance time as well as
the time of interruption of the power supply. An evolutionary algorithm is applied as a constraint satisfaction optimization tool to
search for the optimal relay setting. This method can find the best
protection relay coordination, which cannot be achieved by traditional methods. This is the most significant achievement of the
paper.
Index TermsEvolutionary algorithm, power system protection,
protection coordination.

I. INTRODUCTION

TYPICAL industrial power network may consist of hundreds of pieces of equipment, and even more protection
relays to protect the system are required. The protection system
consists of various relays with different operating principles to
tackle different types of faults. Very often, two or more relays
with different operating principles depending on the voltage
level and importance may be required to protect the equipment.
Each protection relay in the power system needs to be coordinated with the relays protecting the adjacent equipment. The
overall protection coordination is, thus, very complicated. Unfortunately, in a practical industrial power network, it is almost
impossible to obtain a protection setting that can satisfy the coordination between all adjacent relays. In fact, there exists a certain number of blind spots of coordination in the protection of
the power system.
In the traditional method of protection coordination, protection relays are classified according to their type, such as overcurrent and distance protection, and coordination is carried out on
each type individually. The effect of coordination on other protection systems is not usually considered. Work done on relay
setting coordination on individual types of relays is shown in
Paper ICPSD 9949, presented at the 1999 Industry Applications Society Annual Meeting, Phoenix, AZ, October 37, and approved for publication in the
IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS by the Power Systems Protection Committee of the IEEE Industry Applications Society. Manuscript submitted for review Ocotber 8, 1999 and released for publication March 17, 2000.
This work was supported by Hong Kong Polytechnic University.
The authors are with the Department of Electrical Engineering, Hong Kong
Polytechnic University, Kowloon, Hong Kong (e-mail: paulso@clp.com.hk;
eekkli@polyu.edu.hk).
Publisher Item Identifier S 0093-9994(00)07617-9.

Fig. 1.

Duplication of protection schemes.

[1][5]. This paper presents how a sophisticated optimization

technique is applied to adjust the relay settings of overcurrent
and distance relays, and examines the satisfaction of the relay
operation times within the coordination margin by considering
the effect on all protection systems.
II. PROBLEM FORMULATION
A typical protection scheme is shown in Fig. 1. Very often, two
or more relays are required at various ends to sense and exchange
system information in order to provide a high fault clearing time
so as to increase the reliability of the protection system.
To clear a fault, the faulty equipment will be tripped by either
relay which detects the fault more quickly and sends an opencircuit command faster. The operation time of the equipment
may be expressed as
(1)
is the operation time of the protection scheme prowhere
is the minimum fault clearing
tecting the equipment, and
) and is equal to the sum of
time of the protection scheme (
the minimum possible time of the protective relays plus the operating time of the fault-clearing device (breaker operation time).
If the faulty equipment cannot be isolated, the backup protection located in the adjacent equipment will take up the responsibility of the system recovery as soon as possible. The sequence
of backup protection tripping actions will be carried out stage
by stage until the system disturbance is cleared. Fig. 2 shows
the backup stages for clearing the system disturbance.
When the fault clearance by the main protection has failed,
the first stage backup protection will be initiated. If one element of the first stage backup protection fails, the second stage
backup protection will be started. A time delay exists between
the main protection and the first stage backup, and between the

00939994/00$10.00 2000 IEEE

1236

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 36, NO. 5, SEPTEMBER/OCTOBER 2000

TABLE I
RELAY SETTING METHODS

Fig. 2. Backup protection for the faulty equipment, assuming first-stage

backup protection failure.

first and second stage backup protection. The formation of the

first and second stage obviously depends on the fault location.
Ordinary protection system has no communication on information exchange for the location of fault occurrence and the stage
of each relay. Therefore, the relay setting is the only parameter
to control the tripping in the desirable sequence. A new method
of coordination method called time coordination is developed
in this paper to formulate all the system relays and system equipment operation into a set of optimization equations and constraints. Its purpose is to search for an optimal protection setting
to minimize the system disturbance time as well as the time of
interruption of customer supply.
III. COORDINATION OF BACKUP PROTECTION
In this paper, the following backup protections are considered:
definite time-delay overcurrent (TDOC);
Inverse definite multiple time lag overcurrent (IDMTL);
directional inverse definite multiple time lag overcurrent
(DIDMTL);
distance protection zone 2 (DIST2);
distance protection zone 3 (DIST3).
The relay setting methods and operating conditions are summarized in Table I where
I
system current flow at the relaying point;
Zf
measured impedance at the relaying point;
Tdelay
delay time setting;
Isetting
is the current setting;
TM
time multiplier;
CSM
current setting multiplier;
CT
current transformer at the relaying point;
TZONE2 time delay for distance zone 2;
TZONE3 time delay for distance zone 3;
Z2REACH reach for distance zone 2 fault;
Z3REACH reach for distance zone 3 fault;
where
is the per-unit
current; and are two constants.

In the coordination of backup protection, the fastest relay operated will decide the equipment isolation time. The two adjacent pieces of equipment are formed into a coordination pair.
They are protected by the backup relays and will be operated
in sequence. The operation time difference between the coordination pair must conform to the grading margin. The number
of system constraints should be the product of the number of
combination of coordination pairs, the number of combinations
of busbar faults, the number of possible system configurations,
and the variation of fault types. For the purpose of demonstration, a single-phase-to-ground fault is considered in this paper.
The effectiveness of the relay setting is evaluated by the objective value as defined in (2). A smaller objective value indicates a good coordinated protection system
objective

(2)

is the relay operation time including the breaker opwhere

is
erating time at that particular system configuration .
the difference in the coordination margin between the coordination pair of equipment and the preferred value of coordination margin. In the case of distance relays, it is the difference
of the computed zone coverage and the corresponding preferred
is the number of constraint
coverage of the protected line.
violations among all the coordination pairs for equipment .
, , , and are coefficients governing the amount of contribution to the overall objective value. relates to the relay operation time; relates to the deviation from the preferred setting;
and and relates to the number of constraint violations. The
range of constants ranges from 0 to 1 and is set according to the
user preference. For instance, the effect of relay operation time
is reduced to zero if is set to zero. If, is set close to 1, the
computed relay setting will attempt to align the preferred setting as close as possible. In some system configurations, for example, a system with a weak source, too much emphasis should
not be placed on the minimization of the constraint violations,
therefore, should be set closer to 0. The speed of minimization of constraint violation is controlled by . The number of
constraint violations for some complicated systems may be up
to 10 000, and is used to moderate the dominate effect of constraint violation to the objective functions and is typically set to
0.1. Typical values of , , and used in this paper are 0.9, 0.8,
and 0.6, respectively.

SO AND LI: POWER SYSTEM PROTECTION BY EVOLUTIONARY ALGORITHM

1237

Fig. 4. Simplified industrial network.

duce the next generation. Several sets of relay settings are altered by a process called mutation to produce several sets of
new relay settings as the next generation. The mutation is based
on Gaussian normal distribution noise as shown in

(3)

Fig. 3. Time coordination algorithm flow chart.

Note: In each system configuration, each coordination pair of

equipment will be iterated by , ,
IV. TIME COORDINATION ALGORITHM
The key steps of the time coordination algorithm are shown
in Fig. 3. The time coordination is classified as constraint satisfaction optimization and each stage of the coordination process
is discussed in the following.
A. Initialize Relay Settings
Several sets of relay settings will be randomly generated.
Each set of relay settings will be checked against all system constraints and configurations. The effectiveness of the relay settings is calculated by (2). Any violation of system constraints
will be penalized and this will be reflected in the increased objective value.
B. Evolutionary Algorithm
The evolutionary algorithm (EA) is a stochastic parallel
search method for multivariables. It can find out the optimum
relay settings in a fixed number of generations with maximum
satisfaction of coordination constraints. The industrial power
network under study is very complex. It contains a large
amount of variables and numerous constraints. The optimal
relay settings may be considered as the global optimum point
on the solution searching space. Conventional searching
methods, such as steeper-decant and nonlinear programming
[1][3], perform a single-point search. In most cases, the global
optimum probably will not be reached just for one trial due to
being trapped by the local optimums.
EA employs a multipoint search methodology. The chances
of reaching the global optimum are, therefore, much higher.
It is based on the modified evolutionary programming technique [6][9] by applying mutations of relay settings to pro-

where
element of relay settings;
scale factor for EA mutation;
offset for EA mutation;
effectiveness of the relay settings which can be
reflected in its objective value;
Gaussian normal distribution noise.
The scale factor and the offset are to control the performance of the EA and are typically set to 1 and 0, respectively.
These newly generated relay settings should be passed to constraints checking and objective calculation for the next EA generation.
C. Constraint Checking
is the number of constraint violations. Each relay
In (2),
settings will be checked under different coordination pairs. If
the time difference of the coordination pair is less than the cowill be incremented by one. More
ordination time margin,
constraint violations of relay settings will result in a larger objective value and will lead to less chance of surviving in the next
generation. In the constraint checking, the relay operation time
and coordination margin
also be calculated. The objective value can be calculated for the next generation.
D. Termination
The process will stop after a fixed number of generations. The
number of generations required to carry out the optimum relay
setting depends on the pattern and the number of initial relay
settings.
V. SIMULATION
The simplified industrial network for this study is shown in
Fig. 4. The system parameters are listed in Tables II and III. For
the demonstration of the significance of the time-grading algorithm, all transmission lines at both ends are protected by inverse

1238

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 36, NO. 5, SEPTEMBER/OCTOBER 2000

TABLE II
GENERATOR PARAMETERS

TABLE IV
RELAY SETTING RULES (DIST 1 IS THE ZONE 1 SETTING OF DISTANCE RELAY

TABLE III
TRANSMISSION LINE PARAMETERS

Fig. 5. Sample network.

definite time phase fault overcurrent (IDMTL OC), inverse definite time earth fault overcurrent (IDMTL EF), earth fault distance zone 2 (DIST 2), and earth distance zone 3 (DIST 3). The
EA method used in this paper is applied to the relays associated with the transmission lines connecting the various buses as
shown in Fig. 4. This is adequate to demonstrate the application
of EA method to relay coordination. The relays associated with
generators G1G4 are not considered in this case study. If required, they can be included in the optimization process without
any problem.
The setting ranges of the type of relays are as follows:
IDMTL OC CSM from 50% 200% in step of 1%, time
multiplier (TM) from 0.1 1.0 in step of 0.01;
IDMTL EFCSM from 20% 80% in step of 1%, TM
from 0.1 1.0 in step of 0.01;
DIST 2reaching from 100% 999% in step of 1%, time
delay from 400 ms 1000 ms in step of 1 ms;
DIST 3reaching from 100% 1999% in step of 1%,
time delay from 400 ms 2000 ms in step of 1 ms.
The EA stochastically changes the relay setting in order to
minimize the objective value. As the settings are generated in
random, some unwanted relay settings may be generated, resulting in wasting of computation. To minimize this waste, rules
are established to control the value of the relay settings. Certain
types of relays will impose a desirable pattern. For example,
zone 2 reach of distance relays are normally equal to 150% of
zone 1 reach, and zone 3 reach is equal to 200% of zone 1 reach.
In the study, as time coordination methods can find out the optimum reach in zone 2 reach and zone 3 reach, the rule set in the
simulation is that zone 1 reach must be less than zone 2 reach
and zone 2 reach must be less than zone 3 reach. The rules employed in the simulation are listed in Table IV.

Fig. 6. Relay setting optimization direction.

VI. COORDINATION IMPROVEMENT

Consider the simple network shown in Fig. 5. The backup
protection from Bus A to Bus B is used as an example. The relaying point at Bus A has IDMTL and distance zone 2 and 3
(DIST 2 and DIST 3). According to the EA statistical data, the
relay settings of IDMTL, DIST B, and DIST 3 are improving in
each generation in the directions shown in Fig. 6, so as to optimize the grading discrimination and coverage. Its major constraint is the setting of relays at Bus B as shown in Fig. 7.
The time coordination method will keep the grading margin
between relays at Bus A relative to relays at other buses at a
minimum. One example is the grading margin GM1 from DIST
2 at Bus A to DIST 1 at Bus B, and GM2 from DIST 3 at Bus A
to DIST 2 at Bus B. For the fault occurring far away from Bus
A, the operation time of IDMTL at Bus A should also be graded
with IDMTL at Bus B, for example, GM3. In fact, the relay
setting will be checked against all single-phase busbar faults in
all combinations of transmission line and generator outages. The
relays at Bus A are, therefore, constrained by other adjacent bus
relays according to the fault current direction.
The zone 2 and zone 3 reach of distance relays are normally
set as the effective impedance from Bus A to Bus B plus a certain percentage of the next shortest line impedance. A typical

SO AND LI: POWER SYSTEM PROTECTION BY EVOLUTIONARY ALGORITHM

1239

TABLE V
BEST RELAY SETTINGS AMONG 200 INITIALIZED RELAY SETTINGS

Fig. 7.

Overall characteristic of relays at Bus A and Bus B relays.

Fig. 8.

Distance zone 2 and zone 3 reach for radial-fed circuits.

TABLE VI
OPTIMUM RELAY SETTINGS

Fig. 9. Distance zone 2 and zone 3 reach for ring-fed circuits.

example is shown in Fig. 8. The reach of DIST 2 at Bus A is

assumed equal to the protected line impedance plus 30% of the
. The reach of
next shortest line impedance, i.e.,
DIST 3 is also assumed equal to the protected line impedance
.
plus 80% of the next shortest line impedance, i.e.,
If Bus C is interconnected to Bus D by a line, as shown in
Fig. 9, the reach of DIST 2 will be
, and the reach of DIST 3 will be
. The reach of DIST 2 and DIST 3
is, therefore, dependent on the protected line impedance . For
a long transmission line, the impedance will be larger and the
effectiveness of the ring circuit will be reduced. The percentage
of zone 2 and zone 3 reach will, therefore, be smaller. For the
system under study, all transmission lines are interconnected
and the effects of interconnections are considered for zone 2
and zone 3 reach of distance relays.

VII. RESULT AND DISCUSSION

Relay coordination is carried out using the EA with the
process previously described. Table V shows the best set of
initial relay settings from 200 sets of initialized relay settings.
Table VI shows the optimized relay settings by using the EA
after 2000 generations. The performance of these relay settings

1240

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 36, NO. 5, SEPTEMBER/OCTOBER 2000

TABLE VII
PERFORMANCE OF RELAY SETTING COMPARISON

is indicated by the sum of relay operating time, the count of

constraint violations, and the objective value. After the computation, the objective value is significantly improved. It means
that all relays have optimum operation for all single-phase
busbar faults and the coordination time margin is always the
minimum.
Constraint violations always occur in low-fault-level conditions. As the IDMTL relay operation in low fault current is slow,
some marginal IDMTL operations cannot coordinate with adjacent IDMTL or distance relays, and this will result in failure
in constraint checking. The time coordination algorithm can reduce the number of constraint failures significantly. The performance of relays before and after the setting computation is
shown in Table VII.
VIII. CONCLUSION
The time coordination method has successfully been applied
to coordinate the backup relay settings, disregarding the operation principles. For numerous noncontinuous variables, such as
backup relay settings for an industrial power network, the EA
effectively optimizes the relay grading margins and minimizes
the coordination constraint violations. The relay settings are optimized and directed toward the desirable pattern. The time coordination method is the milestone of an automatic protection
setting coordination calculation.
REFERENCES
[1] B. Chattopadhyay, M. S. Sachdev, and T. S. Sidhu, An on-line relay
coordination algorithm for adaptive protection using linear programming technique, IEEE Trans. Power Delivery, vol. 11, pp. 165171,
Jan. 1996.

[2] A. J. Urdaneta, H. Restrepo, S. Marquez, and J. Sanchez, Coordination

of directional overcurrent relay timing using linear programming, IEEE
Trans. Power Delivery, vol. 11, pp. 122129, Jan. 1996.
[3] S. J. Lee and C. C. Liu, Intelligent approach to coordination identification in distance relaying, in Proc. Int. Conf. Intelligent Systems Applications to Power Systems, Orlando, FL, Jan. 28Feb. 2, 1996, pp. 6267.
[4] C. W. So, K. K. Li, K. T. Lai, and K. Y. Fung, Application of genetic algorithm for overcurrent relay coordination, in Proc. IEE 6th
Int. Conf. Developments in Power System Protection, Nottingham, U.K.,
Mar. 1997, pp. 6669.
[5] I. Dabbaghchi, R. D. Christie, G. W. Rosenwald, and C. C. Liu, AI
application areas in power systems, IEEE Expert, vol. 12, pp. 5866,
Jan./Feb. 1997.
[6] D. E. Goldberg, Genetic algorithm approach: Why, how, and what
next?, in Proc. Conf. Adaptive and Learning System: Theory and
Applications, New Haven, CT, May 1985, pp. 247253.
, Genetic Algorithm in Search, Optimization and Machine
[7]
Learning. Reading, MA: Addison-Wesley, 1989.
[8] D. B. Fogel, An analysis of evolutionary programming, in Proc. 1st
Conf. Evolutionary Programming, Evolutionary Programming Society,
La Jolla, CA, 1992, pp. 4351.
[9] J. H. Kim and H. Myung, Evolutionary programming techniques for
constrained optimization problems, IEEE Trans. Evol. Comput., vol. 1,
pp. 129140, July 1997.

C. W. So (M96) was born in Hong Kong in 1969.

He received the B.Eng. degree in 1996 from Hong
Kong Polytechnic University, Hong Kong, where he
is currently working toward the Ph.D. degree.
He is currently an Engineer II with CLP Power
Ltd., Hong Kong, where he is responsible for protection and substation automation. His research interests are power system protection, applications of artificial intelligence, substation automation, and power
system computer programming.

K. K. Li (M76SM91) received the M.Sc. degree

from the University of Manchester Institute of
Science and Technology, Manchester, U.K., and the
Ph.D degree from City University, London, U.K.
He has worked as a Protection Engineer in Hong
Kong. He is currently an Associate Professor in the
Department of Electrical Engineering, Hong Kong
Polytechnic University, Hong Kong. His research interests are power system protection and applications
of artificial intelligence in power systems.