Hindawi Publishing Corporation

Mathematical Problems in Engineering

Volume 2010, Article ID 906935, 11 pages
doi:10.1155/2010/906935

Research Article
Ant Colony Search Algorithm for
Optimal Generators Startup during Power
System Restoration

A. Ketabi1 and R. Feuillet2

1
Department of Electrical Engineering, University of Kashan, Kashan 87317-51167, Iran
2
Laboratoire d’Electrotechnique de Grenoble, INPG/ENSIEG, 38402 Saint Martin d’Hères Cedex, France

Correspondence should be addressed to A. Ketabi, aketabi@kashanu.ac.ir

Received 16 September 2009; Revised 7 December 2009; Accepted 14 January 2010

Academic Editor: Jihuan He

Copyright q 2010 A. Ketabi and R. Feuillet. This is an open access article distributed under
the Creative Commons Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.

Generators startup sequence plays a significant role in achieving a suitable and effective restoration
strategy. This paper outlines an ant colony search algorithm in order to determine the generator
starting times during the bulk power system restoration. The algorithm attempts to maximize
the system generation capability over a restoration period, where the dynamic characteristics of
different types of units and system constraints are considered. Applying this method for the 39-bus
New England test system, and comparing the results with backtracking-search and P/t methods,
it is found that proposed algorithm improved generation capability.

1. Introduction
In recent years, power systems are operated fairly close to their limits primarily due to
economic competition and deregulation. At the same time, they have increased in size and
complexity. Both factors increase the risk of major power outages 1. After a blackout, power
needs to be restored as quickly and reliably as possible, and consequently detailed restoration
plans are necessary 2, 3. While most restoration plans have been developed on a trial-and-
error principle, there is a tendency to employ a more systematic planning and analysis of
restoration procedures 4–10.
In the case of a total system outage, system restoration must begin from the black-
start units. Black-start units are units that do not require off-site power to start, such as:
diesel generator sets and hydroelectric units 5. As the Black-start units themselves can
only supply a small fraction of the system load, these units must be used to assist in the
starting of larger units, which need their station service loads to be supplied by outside
2 Mathematical Problems in Engineering

Table 1: Startup characteristics of different types of units.

Unit type Crank power Critical maximum interval Critical minimum interval
Black-start CT or Hydro No No No
Drum Yes Yes No
SCOT Yes No Yes

power sources. Full restoration of system load can only occur when these larger units can
come on line. In order to achieve a suitable and effective restoration strategy, deciding on the
generators startup sequence is an important task 7–10. It should be pointed out that a proper
sequence of generators start-up can increase the system MW outputs and keep the constraints
satisfied. Optimal generators start-up strategy in system restoration is a multistage decision
optimization problem.
A number of studies have been carried out to determine generators start-up sequence
7–10 using heuristic methods which do not guarantee their optimality. In this paper, from
the global optimization point of view, a novel approach based on the Ant Colony Search
ACS algorithm is proposed.
The goal of the proposed method is to maximize the total system generation capability
over a restoration period whilst considering the corresponding static and dynamic constraints
including the cranking power, critical maximum interval, and critical minimum interval
constraints.
In the following sections, first the problem formulation is presented. The proposed
ACS algorithm is then described. Finally, the simulation results for a 39-bus New England
test system are illustrated and compared with those obtained by backtracking search and p/t
methods.

2. Problem Formulation
During total blackout, the initial power source to crank non-black-start generators must be
found. The initial source of power is provided by starting black-start generator quickly. For
non-black-start generators, different physical characteristics and the starting requirements in
each power station need to be considered. Table 1 shows startup characteristics of different
types of units 7. If a unit with a critical maximum interval is not started within the interval,
there may be a delay of hours before the unit becomes available again. A unit with a critical
minimum interval cannot be started until the interval ends. Hence, consideration of unit’s
constraints is essential for start-up sequence planning.
Figure 1 shows start up timing for a typical unit 11, where we have

t0 : restartup time,

t1 : synchronization time,

t2 -t1 : time to reach minimum load,

t3 -t2 : time to reach maximum load.

Finding proper cranking priorities for non-black-start units can increase the system
MW output. The goal is to maximize G, the total system generation capability in MWH, which
Mathematical Problems in Engineering 3

P
Pmax

Pmin

t0 t1 t2 t3
Time

Figure 1: Startup timing for a typical unit.

can be produced during the restoration period as described below:

n T

G Pi t − ti dt, 2.1
i1 ti

where ti is the start time for the ith generating unit whose generating capability is Pi t − ti ; T
is the planning ending time.
The constraints to be considered in this problem are as follows:

a maximum MW output of units,

b reactive power over and under excitation limits from generator capability curve,
c start-up times,
d start-up and house-load MW requirement,
e ramping rates.

To deliver cranking power from black-start to non-black-start units, it is necessary to build

transmission paths between them. The number of required switching as well as limitations
of the units MVAR are the two most important issues in path selection. Wherever units
MVAR limitations permit, the shortest path i.e., the path requiring the minimal number
of switching action between the supplying unit and the non-black-start unit is selected. A
heuristic algorithm, called A∗ search 12, is used to identify the shortest path. This algorithm
also examines sum of line charging for the selected path as a constraint. This must be smaller
than the limit MVAR absorbing capabilities of generating units.
Load restoration and power supply must coincide with each other to guarantee the
stability of system frequency and voltage, because the restoration of power plants and that of
loads are synchronous.

3. Ant Colony System Algorithm

Ant Colony Optimization ACO method handles successfully various combinatorial
complex problems. Dorigo, inspired by the behavior of real ant colonies, proposed ACO
4 Mathematical Problems in Engineering

S/S S1 S2 S3 S4 Sn
G1 Black start

G2

G3

G4

Gn

Figure 2: Search space of ant sequence determination.

method for the first time in his Ph. D. thesis 13. The method is postulated on the basis
of a set of artificial ants which cooperate in solving a problem by exchanging information via
pheromone deposited on a graph. It can be used to solve both continuous and discontinuous,
nonconvex, nonlinear constrained optimization problems. The characteristics of ACO include
positive feedback, distributed computation, and utilization of constructive greedy heuristic.
Recently a large number of different ACO algorithms have become available. All of
these algorithms contain a strong exploitation of the best solutions found during the search;
the most successful of which add explicit features to the search in order to avoid premature
stagnation. The main differences between the various ant system extensions consist of the
techniques used to control the search process. ACS is reported to be the most aggressive of
the ACO algorithms 13 returning the best solution quality for very short computation times.
In power systems, the ACO has been applied to solve the optimum generation
scheduling problems 14, unit commitment 15, economic dispatch 16, and the
constrained load flow 17.
Once the search space of generators start-up problem is established using multi-
process decision making concept, the ACS algorithm can be applied to problems such as
traveling salesman problem TSP. TSP defines the task of finding a tour of minimal total cost
given a set of fully connected nodes and costs associated with each pair of nodes. The tour
must be closed and contain each node exactly once. This can be represented as a sequence of n
items, where the actual order of the sequence determines a particular solution to the problem.
Thus, in general, the search space consists of all n! permutations.
The sequence in which generators are started up during system restoration is
important since different sequences yield different MWH outputs. Figure 2 shows the search
space of ant sequence determination where the first generator is a black-start unit. For the kth
sequence Sk , the generator will be started up at tSk ; the exact time of which depends on the
system condition and the generator start-up MW requirement. The cost function for sequence
S is


n
CS  T CSi−1 , Si , 3.1
i1
Mathematical Problems in Engineering 5

where T CSi−1 , Si  is the transition cost between the sequence i − 1 and i, defined as follows:

T
T CSi−1 , Si   − PSi − P crSi  · dt, 3.2
tSi

where tSi , PSi , and P crSi are starting time, MW output, and cranking power of generator at
ith sequence, respectively. It should be noted that the cost function C in 3.1 is in fact equal
to G in quantity at 2.1 but with negative sign and C should, therefore, be minimized.
The two main phases of the ACS algorithm constitute the ants’ solution construction
and the pheromone update.
In this algorithm, ants find solutions starting from a start node and move to
feasible neighbor nodes in the process of ants’ generation and activity. During the process,
information collected by ants is stored in the so-called pheromone trails. During the process,
ants can also release pheromone while building the solution local pheromone trail update
or after the solution is built global pheromone trail update. An ant-decision rule, made
up of the pheromone and heuristic information, governs ants’ search toward neighbor nodes
stochastically. Pheromone evaporation is a process of decreasing the intensities of pheromone
trails over time. This process is used to avoid local convergence and explore more search
areas.

3.1. Sequence Construction

Sequences are constructed by applying the following simple constructive procedure to each
ant: 1 choose a start unit black-start unit at which the ant is positioned; 2 use pheromone
and heuristic values to construct a sequence on probability basis, by iteratively adding units
that the ant has not yet visited. This will continue until all units have been visited. In ACS,
when located at unit i, ant k moves to a unit j chosen according to the so called pseudorandom
proportional rule, given by

⎧ ⎧  ⎫
⎪
⎪ ⎨max τiu · ηiu β
⎬
⎪
⎨arg if q ≤ q0 ,
j ⎩ u ∈ Sk ⎭ 3.3
⎪
⎪
⎪
⎩
J otherwise,

where τiu is the pheromone trail, ηiu  1/tciu is the problem-specific heuristic information,
and β is a parameter representing the importance of heuristic information, q is a random
number uniformly distributed in 0, 1, q0 is a prespecified parameter 0 ≤ q0 ≤ 1 and 1 − q0 
can be interpreted as trail evaporation, Sk is the set of units currently not passed by ant k, and
J is a random variable selected according to the probability distribution given by

⎧
⎪
⎪ τij ηij
β
⎪
⎨ if j ∈ Sk ,
 β
Pijk  u∈Sk τiu  ηiu 3.4
⎪
⎪
⎪
⎩0 otherwise.
6 Mathematical Problems in Engineering

In other words, with probability q0 the ant makes the best possible move as indicated by
the learned pheromone trails and the heuristic information in this case, the ant is exploiting
the learned knowledge, while with probability 1 − q0  it performs a biased exploration of
the arcs. Tuning the parameter q0 allows modulation of the degree of exploration and the
choice of whether to concentrate the search of the system around the best-so-far solution or
to explore other sequences.

3.2. Global Pheromone Trail Update

In ACS, following each iteration, only one ant the best-so-far ant is allowed to add
pheromone such that the new pheromone trail becomes a weighted average between the
old pheromone value and the amount of pheromone deposited. Thus, the update in ACS is
implemented by the following equation:

τij ←− 1 − ρ τij  ρΔτijbs ∀ i, j ∈ Sbs , 3.5

where i, j’s are the edges belonging to Sbs , the best minimum cost sequence since the
beginning of the trial, 0 < ρ < 1 is a parameter governing global pheromone decay, and change
in pheromone is given by

1
Δτijbs  , 3.6
Cbs

where Cbs is the best-so-far cost. It is important to note that in ACS the pheromone trail
update, both evaporation and new pheromone deposit, only applies to the best sequence.

3.3. Local Pheromone Trail Update

In addition to the global pheromone trail updating rule, in ACS the ants use a local
pheromone update rule that they apply immediately after having crossed a sequence i, j
during the sequence construction:

τij ←− 1 − ξτij  ξτ0 , 3.7

where 0 < ξ < 1 is a constant governing the local pheromone decay process and τ0 is the
initial value of the pheromone trails. The effect of the local updating rule is that each time an
ant uses a sequence i, j its pheromone trail τij is reduced, so that the sequence becomes less
desirable for the following ants. In other words, ants do not converge to the generation of a
common sequence.
The flowchart of the ACS algorithm for optimal generators start-up is shown in
Figure 3.
Mathematical Problems in Engineering 7

Start

Input units start-up

characteristics

Initialization and parameters

selection for the ACS

Iteration counter i  1

Construction step counter j 1

All ants increment their

partial solution by adding
one more using 3.3, 3.4
j j 1
Locally update the
pheromone trails using 3.7

Construction No
finish?

Yes

Record the best solution of this

iteration and globally update the ii1
pheromone trails using 3.5

Iteration No
terminate?

Yes

Output
solution

End

Figure 3: The flowchart of the ACS algorithm for optimal generators start-up.

3.4. Parameter Settings

Good convergence behavior of ACS algorithm can be achieved by suitable selection of
parameters. The parameters that affect the computation of ACS algorithm directly or
indirectly include ant number, ρ, ξ, β, and q0 .
8 Mathematical Problems in Engineering

30 37
25 26 28 29
2
24
38
27
1
3 18 17
39 16 35

15 21 22
9 4 14
12
23
8 5 19
6
13 36
20
11 33

31 10
34
7
32

Figure 4: One line diagram of 39 bus New England test system.

The settings are done for 39-bus New England system whose characteristics are
given in next section. They are tested for each parameter taking several values within a
boundary limit, all others being constant. More than 10 simulations for each setting are
performed in order to achieve some statistical information about the average evolution.
The range of interval considered for each parameters are ρ{0.1 0.9}, ξ{0.1 0.9}, β{0 10}, and
q0 {0 1}.
The initial trail level is set as τ0  1. The number of ants allowed to pass through the
search space is taken as the generators’ number. The final combination of parameters ρ, ξ, β,
q0  that provided the best results is 0.1, 0.1, 2, 0.98.

4. Case Study
In this section, the proposed method is used for restoration of 39-bus New England system
see Figure 4. The system includes 39 buses, 10 generators, 35 lines, and 12 transformers, and
generators in buses 30 and 35 are black-start units, and thus there are 2 subsystems islands
for simultaneous restoration.
The results of proposed method are compared with results of P/t and backtracking
search method. In P/t method, the priority for generators start-up is determined based on
the ratio of MW capability and the time required for the plant to be synchronized with the
system, whose, the generator that its P/t is greater than others has priority for start-up 7.
Backtracking search method determines problem solutions by systematically searching the
solution space using a tree organization 9.
Figure 5 shows the performances of ACS algorithm during starting sequence
optimization.
Tables 2, 3, and 4 show the results of generators start-up sequence using the proposed
method, backtracking search method, and P/t method, respectively.
Mathematical Problems in Engineering 9

×104
−4.4
−4.5
−4.6

Cost function
−4.7
−4.8
−4.9
−5
−5.1
−5.2
0 5 10 15 20 25 30 35 40 45 50
Iteration number

Figure 5: Performances of ACS algorithm during starting sequence optimization.

×103
6

4
MW

0
0 100 200 300 400 500 600 700 800
Time min.

Proposed method
Backtracking method
P/t method

Figure 6: Generation capability of 39-bus New England system.

Figure 6 shows generation capability of the system after system restoration for the
three generators start-up sequences.
During the first 800 minutes of system restoration, the system generation capability
for the proposed method, backtracking search method, and P/t method are 51402, 48535, and
46946 MWH, respectively. This indicates that determination of optimal sequence for start-up
can increase generation capability of the system during system restoration.
These simulation results and other test cases show the improved effectiveness and
accuracy of the proposed method in identifying the optimal generator’s start-up sequence.
During restoration of large power system, it is advantageous in most cases to split the
power system into subsystems in order to allow parallel restoration of islands, and to reduce
the overall restoration time. Within each subsystem, the proposed method can be used to
determine optimal starting sequence. Then the subsystems are interconnected and remaining
loads are picked up and the system performs its transition to the normal state.
10 Mathematical Problems in Engineering

Table 2: Generator start-up sequence and selected path with the proposed method.

Unit no. Bus no. Cranking unit Start time Selected path for energization
1 30 Black start 0 : 00 —
6 35 Black start 0 : 00 —
4 33 6 0 : 10 35-22, 22-21, 21-16, 16-19, 19-33
10 39 1 0 : 15 (30-2), (2-1), (1-39)
7 36 6 0 : 20 22-23, 23-36
2 31 6 0 : 30 16-15, 15-14, 14-4, 4-5, 5-6, 6-31
3 32 4 0 : 40 33-19, 14-13, 13-10, 10-32
5 34 4 0 : 45 19-20, 20-34
8 37 10 2 : 00 2-25, 25-37
9 38 10 3 : 20 1-27, 27-26, 26-29, 29-38

Table 3: Generator start-up sequence and selected path with the backtracking search method.

Unit no. Bus no. Cranking unit Start time Selected path for energization
1 30 Black start 0 : 00 —
6 35 Black start 0 : 00 —
4 33 6 0 : 10 35-22, 22-21, 21-16, 16-19, 19-33
7 36 6 0 : 20 35-22, 22-23, 23-36
8 37 1 0 : 20 30-2, 2-25, 25-37
5 34 4 0 : 40 19-20, 20-34
3 32 4 0 : 50 16-15, 15-14, 14-13, 13-10, 10-32
2 31 4 0 : 50 10-11, 11-6, 6-31
10 39 8 1 : 40 2-1, 1-39
9 38 8 3 : 30 25-26, 26-29, 29-38

Table 4: Generator start-up sequence and selected path with the P/T method.

Unit no. Bus no. Cranking unit Start time Selected path for energization
1 30 Black start 0 : 00 —
6 35 Black start 0 : 00 —
7 36 6 0 : 20 35-22, 22-23, 23-36
8 37 1 0 : 20 (30-2), (2-25), (25-37)
5 34 6 0 : 30 22-21, 21-16, 16-19, 19-20, 20-34
4 33 7 1 : 10 19-33
2 31 7 1 : 10 16-15, 15-14, 14-4, 4-5, 5-6, 6-32
3 32 7 1 : 20 14-13, 13-10, 10-32
10 39 8 1 : 40 2-1, 1-39
9 38 8 3 : 30 25-26, 26-29, 29-38
Mathematical Problems in Engineering 11

5. Conclusion
A method for unit start-up sequence determination during system restoration is described
in this article. Once the search space of generators start-up problem is established using
multiprocess decision-making concept, the ACS algorithm is used to determine the units
starting times to maximize the generation capability dispatch during restoration period. The
dynamic characteristics of different types of units and system constraints are also considered.
The proposed method has been tested on the 39-bus New England test system in order to
determine the optimal start-up sequence. The results are then compared with those obtained
by the backtracking search method and P/t method, which indicated improved effectiveness
and accuracy of the proposed method.

References
1 G. Andersson, P. Donalek, R. Farmer, et al., “Causes of the 2003 major grid blackouts in North America
Europe, and recommended means to improve system dynamic performance,” IEEE Transactions on
Power Systems, vol. 20, no. 4, pp. 1922–1928, 2005.
2 M. M. Adibi and L. H. Fink, “Power system restoration planning,” IEEE Transactions on Power Systems,
vol. 9, no. 1, pp. 22–28, 1994.
3 J. J. Ancona, “Framework for power system restoration following a major power failure,” IEEE
Transactions on Power Systems, vol. 10, no. 3, pp. 1480–1485, 1995.
4 Y. Kojima, S. Warashina, M. Kato, and H. Watanabe, “Application of knowledge engineering
techniques to electric power system restoration,” in Proceedings of IEEE Workshop on Artificial
Intelligence for Industrial Applications, pp. 320–325, 1988.
5 M. M. Adibi, T. L. Volkmann, and M. Northern, “Remote cranking of steam electric stations,” IEEE
Transactions on Power Systems, vol. 11, no. 3, pp. 1613–1618, 1996.
6 F. P. de Mello and J. C. Westcott, “Steam plant startup and control in system restoration,” IEEE
Transactions on Power Systems, vol. 9, no. 1, pp. 93–101, 1994.
7 C.-C. Liu, K.-L. Liou, R. F. Chu, and A. T. Holen, “Generation capability dispatch for bulk power
system restoration. A knowledge-based approach,” IEEE Transactions on Power Systems, vol. 8, no. 1,
pp. 316–325, 1993.
8 T. Nagata, H. Sasaki, and R. Yokoyama, “Power system restoration by joint usage of expert system
and mathematical programming approach,” IEEE Transactions on Power Systems, vol. 10, no. 3, pp.
1473–1479, 1995.
9 A. Ketabi, H. Asmar, A. M. Ranjbar, and R. Feuillet, “An approach for optimal units start-up during
bulk power system restoration,” in Proceedings of the Large Engineering Systems Conference on Power
Engineering, pp. 190–194, 2001.
10 Y. Liu and X. Gu, “Skeleton-network reconfiguration based on topological characteristics of scale-free
networks and discrete particle swarm optimization,” IEEE Transactions on Power Systems, vol. 22, no.
3, pp. 1267–1274, 2007.
11 British Electricity International, Modern Power Station Practice, vol. C, D, and G, Pergamon Press,
Oxford, UK, 1991.
12 S. L. Tanimoto, The Elements of Artificial Intelligence: An Introduction Using LISP, Computer Science
Press, New York, NY, USA, 1987.
13 M. Dorigo and T. Stützle, Ant Colony Optimization, MIT Press, Cambridge, Mass, USA, 2004.
14 Y. U. In-Keun, C. S. Chou, and Y. H. Song, “Application of the ant colony search algorithm to short-
term generation scheduling problem of thermal units,” in Proceedings of the International Conference on
Power System Technology, vol. 1, pp. 552–556, 1998.
15 M. Y. El-Sharkh, N. S. Sisworahardjo, A. Rahman, and M. S. Alam, “An improved ant colony search
algorithm for unit commitment application,” in Proceedings of IEEE Power Systems Conference and
Exposition, pp. 1741–1746, 2006.
16 Y. H. Hou, Y. W. Wu, L. J. Lu, and X. Y. Xiong, “Generalized ant colony optimization for economic
dispatch of power systems,” in Proceedings of IEEE International Conference on Power System Technology,
vol. 1, pp. 225–229, 2002.
17 J. G. Vlachogiannis, N. D. Hatziargyriou, and K. Y. Lee, “Ant colony system-based algorithm for
constrained load flow problem,” IEEE Transactions on Power Systems, vol. 20, no. 3, pp. 1241–1249, 2005.