2012 IEEE International Conference on Power Electronics, Drives and Energy Systems

December16-19, 2012, Bengaluru, India

Optimum Coordination of Overcurrent Relays

Using CMA-ES Algorithm
Manohar Singh, and B.K. panigrahi Rohan Mukherjee
Electrical Engineering Department Electrical Engineering Department
Indian institute of Technology, Delhi-110016 Jadavpur University-700 032
Delhi, India kolkata
Manoharsingh33@[Link], rohanmukherjee@[Link]

Abstract— Coordination of directional overcurrent relays in a relived protection engineering from laborious calculation.
meshed power system is a challenging task for the protection Basically there are two approach are used for coordination of
engineers. In the earlier days it was done manually. Linear and OCRs, conventionally philosophy and parameter optimization
non-linear programming optimizing techniques were very techniques. In conventional technique fault analysis are
frequently used for Coordination of overcurrent relays.
conducted first and then meshed network are broken in radial
Presently, artificial intelligence (AI) techniques are applied for
form, relay at far end is set first and there after corresponding
optimal co-ordination of directional overcurrent relays (DOCR).
backup relay is set, process is repeated until all relays are
This paper discusses application of Covariance Matrix
taken into account. This process is iterative in nature. Final
Adaptation Evolution Strategy (CMA-ES) for optimal
time dial and plug setting values depends upon selection of
coordination of DOCR relays in a looped power system.
Combination of primary and backup relay is chosen by using Far
initial relays known as break points. Break points are selected

vector of LINKNET structure, to avoid mis-coordination of using graph theory approach [2].
relays. Coordination of DOCR is tested for IEEE 30 bus In late eighties, conventional optimization technique was
distribution systems using CMA-ES. The objective function used for relay coordination problem [3]. Linear programming
(OF) is formulated to minimize the operating time between technique gained good popularity but, only helpful in
backup and primary relays (coordination time interval). Results optimizing the time dial setting, as operating time of
are compared with the optimized values of Time dial setting overcurrent relay are linear function of time dial setting[4]-[5].
(TDS) and Plug setting (PS) values obtained from modified
In order to optimize the both time dial setting (TDS) and plug
differential evolution algorithm. The proposed algorithm (CMA-
setting (PS) non-linear optimization were used. Among the
ES) gives optimal coordination margin between 0.3 to 0.8 second
non-linear optimization sequential quadratic programming [6],
and no miscoordination between primary and backup pairs.
random search technique and gradient search technique [7] are
reported in literature. Conventional optimization techniques
Index Terms-- Distribution system; directional overcurrent
have a drawback, some time they may trap to local minima
relay; optimal relay coordination; CMA-ES;
and fails to give global optimal solution. Moreover, as the size
1. INTRODUCTION of system increases the convergence is also weak.
Optimization algorithms based on artificial intelligence are
Relay Coordination in a meshed power network in highly
free from these drawbacks as seen in case of conventional
tedious and time consuming affair [1]. Earlier coordination of
optimization techniques [8]. They are capable for handling
OCRs was performed manually, which was very time
large size system with highly non-linear in both objective
consuming. The use of computer in the relay coordination has
function and in constraints sets. Genetic algorithm [9]-[10],

978-1-4673-4508-8/12/$31.00 ©2012 IEEE

particle swam optimization [11]; hybrid non-linear Normal Coordination Constraints (only overcurrent relay
programming and genetic algorithm [12] are successfully coordination) [12]
applied for optimization of relay coordination problem. 1) Limits on Problem Variables:
Application of Differential evolution algorithm for overcurrent I) Bounds on time dial setting (TDS) of each relay
relay coordination is also reported in literature. [13]. TDS imin ≤ TDS i ≤ TDS imax (1)
In this paper, time multiplier setting and plug setting of
TDS imin and TDS imax are minimum and maximum value
DOCR are optimized using Covariance Matrix Adaptation
Evolution Strategy (CMA-ES). LINKNET structure (using of TDS of relay Ri . TDS imin and TDS imax are taken as 0.025
only Far vector) is used for identify the backup pairs for each and 1.2 sec.
primary relay, which eliminates the need of graph theory
II) Relay characteristic, Bounds on pickup current I p setting
approach. Application of CMA-ES algorithm for relay
of each relay
coordination gives more promising result as compare to other
min max
artificial intelligence (AI) algorithm. The proposed method is I pickup ≤ I p ≤ I pickup
(2)
tested for IEEE 30 bus system. It is seen that total summation
For standard IDMT relay, γ is 0.02 and α is 0.14
of primary relay operation time of DOCRs is optimized and
2) Limits on Primary Operation Time:
the number of mis-coordination pairs are also eliminated.
All of the primary operating time corresponding to each
possible fault location should be less than a maximum allowed
II. OPTIMAL OVERCURRENT RELAY PROBLEM
time delay and more than some minimum predefined time
FORMULATION
considering transient conditions.
In the relay coordination program, two types of tap settings,
3) Coordination criteria (Selectivity Constraint):
namely current settings (PS) and time dial settings (TDS) must
Fault is sensed by both primary as well as secondary relay
be calculated [3]. The current setting for each relay is
simultaneously. To avoid mal-operation, the backup relay
determined by two parameters, namely the minimum fault
should take over the tripping action only after primary relay
current and the maximum load current. For TDS and PS
calculation, objective function is optimized using optimization
fails to operate. If Ri is the primary relay for fault at k, and
technique. The objective function is sum of total operating R j is backup relay for the same fault, then the coordination
time of all the overcurrent relays installed at critical faults F1
constraint can be stated as;
and F2.
t i ,k − t j ,k = ∆ t (3)

where, ti , j is the operating time of the primary relay R j , for

fault at k; ti ,k is the operating time for the backup relay Ri ,

for the same fault at k, ∆t is the coordination time interval

(CTI) generally taken as 0.2 sec [14].

III. CMA_ES WITH DIRECTIONAL TARGET TO

BEST SCHEME
Fig.1. Near end far end fault
 The Covariance Matrix Adaptation Evolution µ
xW( g ) = ∑ i =1 wi xi(:λg ) , where wi > 0 for all i = 1....µ
Strategy (CMA-ES) is an improved strong
µ
minimization strategy. By virtue of this algorithm
and ∑ i =1
wi =1, and i : λ denotes the i -th best

the full covariance matrix of the normal mutation individual.

search distribution is adapted. The main feature of The Algorithm depends of two types of adaptation
the Covariant Matrix Adaptation (CMA) with for the mutation parameters:
Evolution Strategy (ES) is the ability of being Adaptation of covariance matrix C ( g ) .
invariant to landscape transformations and scaling Adaptation of global step size σ ( g ) .
modulation [15]. The CMA-ES is also invariant to Directional Target to Best Scheme[16]:
applications of rotation, reflection and translation, Each member of the population is updated
besides maintaining order and monotonicity. It according to the equation:
offers no discrepancy in behavior towards varied
X i( g +1) = X i( g ) + A.( Best _ Individual − X i( g ) ) (6)
nature of functions and is easily generalizable [9- r
Where, A is mathematically defined as:
10]. uuuuur
If rand be a normally distributed random vector,
Complexity of algorithm is largely reduced with
uuuuur
update schemes of CMA-ES and thus it offers an i.e. rand = randn(1, No _ of _ Variables )
uuuuur
extremely prospective mode of minimization in r rand
then A = uuuuur and this is nothing but the unit
single-objective function landscapes. || rand ||
In CMA-ES λ individuals for the next are vector in the direction of the random normal vector
uuuuur
updated by virtue of the equation: rand .The fitnessi is thereby updated.
2
xk( g +1) ~ N ( xW( g ) , σ ( g ) C ( g ) ) This scheme efficiently guides the particles towards
(4)
Where N ( µ , C ) implies a normally distributed globally best position in the fitness landscape. The

random vector with mean µ and covariance matrix unit random vector in the direction of best
individual of the population ensures better
[Link] update equation for these λ individuals can
exploitation of thee landscape and converging on
be best approximated as:
2
global optima.
N ( x w( g ) ,σ ( g ) , C ( g ) ) ~ xw( g ) + σ ( g ) B ( g ) D ( g ) N (0,1) (5)

The recombination point xW( g ) is the weighted mean IV. IMPLEMENTATION OF PROPOSED
ALGORITHM
of the selected individuals and is mathematically
A. CASE STUDY
defined as A Case study shown in Fig. 2 consists of 15 lines, 14 buses,
3 transformers, and 2 generators. LINKNET structure is used
to store the network configuration. It is very helpful for data
generation, updating and retrieval of data during the relay and backup pairs in a complex looped network. There are 30
coordination problem formulation. Another important digital overcurrent relays with IDMT characteristics and
application of LINKNET structure is determination of primary
tripping direction away from the buses is installed in fault is known as near end fault. If relay on the other end of
system. Faults are generated on each bus ends. If fault occurs line clears the same fault, then it is known as far end fault.
near a bus and relay on the same bus clears it, then this type of

Fig.2. IEEE 30 bus distribution system [17]

B. OPTIMISED RELAY SETTING
Time dial setting (TDS) and plug setting (PS) value are evolution (DE) but CMA-ES gives more minimized value
optimized using CMA-ES with selected control parameters. 15.8929.
Optimal settings are tabulated in Table. I. The minimized
value of objective function is 18.59 using differential

Table.I. Optimal TDS and PS setting of overcurrent relays for IEEE 30 bus system

DE CMA DE CMA
Relay Relay
TDS PS TDS PS TDS PS TDS PS
1 0.15 0.25 0.1 1.5 16 0.9 1.35 0.9 0.85
2 0.05 0.25 0 0.25 17 0.05 0.25 0.05 0.25
 3 0.05 0.25 0.05 0.25 18 1.05 1.45 1.05 0.75
4 0.05 0.25 0.05 0.25 19 1.1 0.7 0.9 0.85
5 0.05 1.3 0.05 1.3 20 0.05 0.3 0.05 0.25
6 0.1 0.25 0.1 0.25 21 1.1 0.5 0.65 1.15
7 0.05 0.9 0.05 0.85 22 0.15 1.15 0.1 1.1
8 0.05 0.6 0.1 0.3 23 0.1 0.45 0.1 0.45
9 0.05 0.25 0.05 0.25 24 0.05 1.5 0.05 1.5
10 0.05 0.25 0.05 0.25 25 0.05 0.25 0.05 0.25
11 0.05 1.5 0.05 1.5 26 0.3 0.25 0.25 0.3
12 0.05 1.45 0.05 1.2 27 0.05 0.25 0.05 0.25
13 0.05 0.9 0.15 0.35 28 0.45 0.25 0.15 1
14 0.05 0.25 0.05 0.25 29 0.2 1 0.25 0.35
15 0.4 0.3 0.15 1.2 30 0.1 0.25 0.1 0.25

Coordination time intervals (CTI) are tabulated in Table. II. differential evolutionary. q is backup relay for each primary
Positive value for each margin means there is sufficient time relay p. Maximum value of CTI is 0.8 and minimum value is
margin for the backup relay operate and coordination of 0.2 using CMA-ES optimization. While for same value of
overcurrent relays to overcurrent relays are well maintained. objective function the upper limit of CTI are violated when
Result is compared with optimized value obtained from differential evolution algorithm is used.

[Link] Coordination time interval for IEEE 30 bus system

CTI CTI
p q p q
CMS-ES DE CMS-ES DE
1 7 0.21207 0.24852 16 19 0.23157 0.20143
2 1 0.86713 0.91471 17 30 0.20716 0.20716
3 7 0.95704 1.18960 18 16 0.23767 0.22753
4 1 0.75871 0.95482 19 21 0.21624 0.30107
5 1 0.55315 0.74926 20 18 2.41672 3.19732
6 1 0.60860 0.80472 21 18 0.20654 0.26113

7 8 0.21464 0.22584 22 29 0.23019 0.21177

8 7 0.75635 1.07038 23 26 0.20744 0.26825
9 5 0.20394 0.20394 24 26 0.69703 0.77918
10 6 0.23239 0.23239 25 23 0.20964 0.20964
11 13 0.25396 0.28180 26 29 0.35743 0.71393
12 5 0.23032 0.20902 27 29 0.57185 0.98916
13 12 0.27358 1.49137 28 22 0.20475 0.21777
14 5 0.28473 0.28473 29 15 0.21755 0.22672
15 28 0.22757 0.29969 30 28 0.43501 0.79060
 CONCLUSION
Overcurrent relay coordination in an interconnected [8] C. W. So, and Li K. K. “Overcurrent relay coordination by
evolutionary programming”, Electric Power System Research,
distribution system with heavily meshed distribution system
vol.53, pp. 83-90, 2000
becomes complex problem. Multiple backup relays are
[9] Farzad Razavi, Hossein Askarian Abyaneha, Majid Al-Dabbaghb,
identified with the help of LINKNET structure. Coordination Reza Mohammadia and Hossein Torkaman, ”A new
problem becomes more complex, when relay setting are comprehensive genetic algorithm method for optimal overcurrent
relays coordination,” Electric Power Systems Research, vol.78, pp.
optimized for near and end far end faults simultaneously for
713-720, 2008.
bidirectional power feed networks. Proposed CMA-ES
[10] Abbas Saberi Noghabi, Javed Sadeh and Habib Rajabi Mashhadi
algorithm is applied for IEEE 30 bus distribution system and ,“considering different network topologies in Optimal overcurrent
result are compared with other AI optimization algorithms. relays Coordination using Hybrid GA,” IEEE Trans. on power
delivery,vol.24,No.4, pp.1857-1862, Oct., 2009.
The performance of proposed algorithm is better as compare
[11] Mohamed M. Mansour, Said F. Mekhamer, and Nehad El-Sherif
to DE algorithm proposed earlier.
El-Kharbawe, “A Modified Particle Swarm Optimizer for the
REFERANCES Coordination of Directional Overcurrent Relays, IEEE Trans. on
[1] Hossein Askarian Abyaneh, Majid Al-Daddagh, Hashemi power delivery, vol. 22, no. 3, pp-1400-1410,July, 2007
kamangar,Seyed hesameddin Hossein sadeghi and Rana Abdul [12] P. P. Bedekar and S.R. Bhide, “Optimum coordination of
Jabbar Khan,“ A new Optimal approach for coordination of directional overcurrent relays using the hybrid GA-NLP
overcurrent relays in interconnected power system”, IEEE Trans. approach”, IEEE Trans. on Power Del., vol. 26, no. 1, pp.109-119,
on Power Del., vol. 15, no. 2, pp. 430-435, Apr. 2003. Jan. 2011.
[2] M. J. Damborg, R. Ramaswami, S. S. Venkata and J. M. [13] Radha Thangaraj, Millie Pant and Kusum Deep, “Optimal
Postforoosh, “Computer aided transmission protection system coordination of over-current relays using modified differential
design,” IEEE Trans. on Power Apparatus and Systems, [Link]- evolution algorithms,” Engineering Applications of Artificial
1, no.1, pp-51-59, January 1984 Intelligence, vol.23,pp. 820-829, 2010.
[3] A. J. Urdaneta, Nadira Ramon and Luis G.P. Jimenez, “Optimal [14] Luis G. Perez and Alberto J. Urdaneta, “Optimal coordination of
coordination of directional relays in interconnected power system” directional overcurrent relays considering definite time backup
IEEE [Link] Power Del., vol. 3, no. 3, pp. 903-911, July 1988. relaying’, IEEE Trans. on Power Del., vol.14, no.4 1996, pp.1276-
4] Abbas Saberi Noghabi, Habib Rajabi Mashhadi and Javed Sadeh, 1284, Oct. 1996
“Optimal Coordination of directional overcurrent relays [15] N. Hansen, “The CMA evolution strategy: a comparing
considering different network topologies using interval linear review, towards a new evolutionary computation”. Advances on
programming,” IEEE Trans. on power estimation of distribution algorithms, Springer, pp. 1769–1776,
delivery,vol.25,No.3,pp.1348-1254, July, 2010. 2006.
[16] Anne Auger and Nikolaus Hansen, “A Restart CMA Evolution
[5] Chattopadhyay B., Sachdev M.S., and Sidhu T.S., “An online relay Strategy with Increasing Population Size”, Proceedings of the
coordination algorithm for adaptive protection using linear IEEE Congress on Evolutionary Computation, CEC 2005
programming technique”, IEEE Trans. on Power Del., vol.11, no. [17] Power Systems Test Case Archive. [Online]. Available:
1, pp.165-173, Jan.1996. [Link]
[6] D. Birla, R. P. Maheshwari and H. O. Gupta, “A new nonlinear [18] Chao-Rong Chen, Cheng-Hung Lee , Chi-Juin Chang, “Optimal
directional overcurrent relay coordination technique and banes and overcurrent relay coordination in power distribution system using a
boons of near end faults based approach”, IEEE Trans. on Power new approach”, Electrical Power and Energy Systems ,vol.45, pp.
Del., vol. 21,no.3, pp.1176-1182, June 2006. 217–222, 2013
[7] V. Ravikumar Pandi, H. H. Zeineldin and Weidong Xiao, [19] José A. Sueiro , Eloy Diaz-Dorado, Edelmiro Míguez, José Cidrás,
“Determining Optimal Location and Size of Distributed “Coordination of directional overcurrent relay using evolutionary
Generation Resources Considering Harmonic and Protection algorithm and linear programming,” Electrical Power and Energy
Coordination Limits,” accepted in IEEE Trans. on Power System, Systems, vol 42 ,pp 299–305,2012.
2012. [20] Matlab, the Language of Technical Computing. [Online].
Available: [Link]