advantage to the proposed algorithm, an entity preferred in Dh : Load demand at the hth hour (set to 10% of Dh),
many industrial applications. Rh : Spinning reserve at the hth hour,
Pi (min) : Minimum generation limit of ith unit,
II. UNIT COMMITMENT PROBLEM Pi (max) : Maximum generation limit of ith unit,
A. Problem Formulation MUi : Minimum up-time of ith unit,
MDi : Minimum down-time of ith unit.
In this section, we first formulate the UC problem. The
Xi on(t) : Duration during which ith unit is continuously on.
objective of the UC problem is the minimization of the total
production costs over the scheduling horizon. Therefore, the Xi off(t) : Duration during which ith unit is continuously off.
objective function is expressed as the sum of fuel and start-up
costs of the generating units. For N generators, the operation B. Data Specifications
cost is defined mathematically as shown in eqn. (1). A benchmark system comprising of 10 generators is adopted
N
TPC N = ∑ [F ( P
i =1
i ih ) + ]
STi (1 − U i ( h−1) ) U ih (1) as the test bed in this work. The data specifications are given
in Table I with the demand over 24-hour period available in
The operating cost accumulates over the total number of Table II. The same set of data was used in [3-9].
operating hours, H, where H = 24 which represents 24 hours TABLE I
GENERATOR SYSTEM OPERATOR DATA
of operation for each unit of generator. Therefore eqn. (1) is
Unit 1 Unit 2 Unit 3 Unit 4 Unit 5
rewritten as: Pmax (MW) 455 455 130 130 162
H N Pmin (MW) 150 150 20 20 25
TPC HN = ∑ ∑ [F ( P
h =1 i =1
i ih ) + ]
STi (1 − U i ( h −1) ) U ih (2)
α ($/h) 1000 970 700 680 450
β ($/MWh) 16.19 17.26 16.60 16.50 19.70
In eqn. (1), TPCN is the total production cost for N units of γ ($/MWh2) 0.00048 0.00031 0.002 0.00211 0.00398
Min Up (h) 8 8 5 5 6
generators whereas TPCHN in eqn. (2) denotes the total Min Down (h) 8 8 5 5 6
production cost for N units of generators over H number of Hot start cost ($) 4500 5000 550 560 900
operating hours. Owing to the operational requirements, the Cold start cost ($) 9000 10000 1100 1120 1800
Cold start hrs (h) 5 5 4 4 4
minimization of the objective function is subject to the Initial status (h) 8 8 -5 -5 -6
following constraints: Unit 6 Unit 7 Unit 8 Unit 9 Unit 10
Pmax (MW) 80 85 55 55 55
(a) Power balance constraint Pmin (MW) 20 25 10 10 10
α ($/h) 370 480 660 665 670
N
β ($/MWh) 22.26 27.74 25.92 27.27 27.79
∑P U
i =1
ih ih = Dh (3)
γ ($/MWh2) 0.00712 0.0079 0.00413 0.00222 0.00173
Min Up (h) 3 3 1 1 1
(b) Spinning reserve constraint Min Down (h) 3 3 1 1 1
N Hot start cost ($) 170 260 30 30 30
(4)
∑P
i =1
i (max)U ih ≥ Dh + R h Cold start cost ($)
Cold start hrs (h)
340
2
520
2
60
0
60
0
60
0
Initial status (h) -3 -3 -1 -1 -1
(c) Generation limit constraint
Pi (min) ≤ Pih ≤ Pi (max) (5)
TABLE II
(d) Minimum up-time constraint (MU) LOAD DEMANDS FOR 24 HOURS
Hour Dh Hour Dh
X i on (t ) ≥ MU i (6)
1 700 13 1400
(e) Minimum down-time constraint (MD) 2 750 14 1300
3 850 15 1200
Xi
off
(t ) ≥ MDi (7) 4 950 16 1050
where the notations used are as follows: 5 1000 17 1000
6 1100 18 1100
7 1150 19 1200
TPC : Total production cost of the power generation, 8 1200 20 1400
Fi (Pih) : Fuel cost function of the ith unit with generation 9 1300 21 1300
10 1400 22 1100
output, Pih, at the hth hour. Usually, it is a quadratic 11 1450 23 900
polynomial with coefficients αi, βi and γi as follows: 12 1500 24 800
Fi ( Pih ) = α i + β i Pih + γ i Pih 2 (8)
N : Number of generators, C. Error Calculation
H : Number of hours, The error can be a good measure to determine the robustness
Pih : The generation output of the ith unit at the hth hour, of an algorithm. The error can give an idea of the degree of
STi : Start-up cost of the ith unit, constraint violations. In this work, the error is calculated
Uih : The on/off status of the ith unit at the hth hour, and based on the following formula:
Uih = 0 when off, Uih = 1 when on,
DRPT2008 6-9 April 2008 Nanjing China
3
TABLE VI [13, 14]. Some of them are based on penalty functions. They
METHODOLOGICAL PRIORITY LIST SOLUTION, TOTAL COST=$563977.1 differ, however, in how the penalty function is designed and
Unit
1 2 3 4 5 6 7 8 9 10
Start Hourly
Err
Power applied to infeasible solutions. As scheduling of generators is
h Cost Cost (MW) a very risky task, we generate feasible solution which meets
1 455 245 0 13683.13 0 700
the (3) – (7) constraints.
2 455 295 0 14554.50 0 750
3 455 370 25 900 17709.45 0 850
4 455 455 40 0 18597.67 0 950
A. Satisfying Power Demand via Lambda iteration
5 455 390 130 25 560 20580.02 0 1000
6 455 360 130 130 25 1100 23487.04 0 1100
7 455 410 130 130 25 0 23261.98 0 1150
In this paper, the calculation for economic load dispatch
8 455 455 130 130 30 0 24150.34 0 1200 (ELD) is done using the lambda-iteration method. Therefore,
9 455 455 130 130 85 20 25 860 28111.06 0 1300 the power demand is automatically satisfied using this method.
10 455 455 130 130 162 33 25 10 60 30117.55 0 1400 The ELD calculation is stopped when the allowable error,
11 455 455 130 130 162 73 25 10 10 60 31976.06 0 1450 which indicates that the total power generated minus the
12 455 455 130 130 162 80 25 43 10 10 60 33950.17 0 1500
13 455 455 130 130 162 33 25 10 0 30057.55 0 1400
power demand is less than the number of participating
14 455 455 130 130 85 20 25 0 27251.06 0 1300 generators divided by ten. In mathematical form, the lambda
15 455 455 130 130 30 0 24150.34 0 1200 iteration is stopped when the following criteria is satisfied
16 455 310 130 130 25 0 21513.66 0 1050
17 455 260 130 130 25 0 20641.83 0 1000 N (13)
18 455 360 130 130 25 0 22387.04 0 1100
TPowh − Dh <
1000
19 455 455 130 130 30 0 24150.34 0 1200
20 455 455 130 130 162 33 25 10 490 30547.55 0 1400 where TPowh is the total power generated at hour h whereas
21 455 455 130 130 85 20 25 0 27251.06 0 1300
22 455 455 145 20 25 0 22735.52 0 1100
Dh is the demand at hour h. Note that the lambda-iteration is
23 455 420 25 0 17684.70 0 900 carried out per hour basis. The main advantage of this method
24 455 345 0 15427.42 0 800 is the very fast speed in calculating the optimal scheduling of
Total 4090 563977.1 0 27.1 K generators.
the load demand without considering MU and MD constraints. convergence and is more reliable. As for now, MPL is the
Then, the MD constraint checking is done. After finishing, the most effective deterministic method in comparison to the one
MU constraint checking follows, based on the priority list, existing in solving UC problem.
starting from the cheapest generator. This is rather simple yet
efficient constraint satisfaction techniques for MU and MD IX. REFERENCES
constraints proposed in this work. [1] C. K. Pang, G. B. Sheble, and F. Albuyeh, “Evaluation of Dynamic
Programming Based Methods and Multiple Area Representation for
VII. NUMERICAL COMPARISONS Thermal Unit Commitment,” IEEE Trans. Power App. Syst., vol. PAS-
100, no.3, pp. 1212-1218, 1981.
The results of MPL are compared to other existing methods [2] F. Zhuang and F. D. Galiana, “Toward a more Rigorous and Practical
both in numerical values and time recorded as shown in Table Unit Commitment by Lagrangian Relaxation,” IEEE Trans. on Power
Systems, vol. 3, no. 2, pp. 763-772, 1988.
IX. The characteristic of the methods are given in the second [3] C.C.A. Rajan and M.R. Mohan “An evolutionary programming-based
column under “Char”. For the deterministic methods, denoted tabu search method for solving the unit commitment problem,” IEEE
by D in the table, the solutions for best, average and worst are Trans. Power Syst., vol. 19, no. 1, pp. 577-585, Feb 2004.
the same as the total cost recorded is the same for each run. [4] T.O. Ting, M.V.C. Rao and C.K. Loo, “A Novel Approach for Unit
Commitment Problem via an Effective Hybrid Particle Swarm
From this comparison, MPL records the best total cost
Optimization”, IEEE Trans. Power Syst., vol 21, no. 1, pp. 1-8, Feb
($563977) compared to other deterministic methods such as 2006.
DP, LR and LRGA. The total cost is competent to the one [5] H. Mori and O. Matsuzaki, “Application of priority-list-embedded Tabu
obtained by HPSO, PL EA and EPL. HPSO in [4] records a search to unit commitment in power systems,” Inst. Elect. Eng. Japan,
vol. 121-B, no. 4, pp. 535-541, 2001.
slightly lower total cost, $563942.3, due to the reason that the [6] D. Srinivasan and J. Chazelas, “A priority list-based evolutionary
variable for power are encoded in floating point rather than algorithm to solve large scale unit commitment problem,” in Proc of Intl
integer as considered in this case. However, HPSO has Conf on Power System Technology, vol 2, pp. 1746-1751, 2004.
expensive computational requirements. [7] T. Senjyu, K. Shimabukuro, K. Uezato and T. Funabashi, “A Fast
Technique for Unit Commitment Problem by Extended Priority List,”
The time recorded for MPL is exactly 0.037s and is the
IEEE Trans. Power Syst., vol. 18, no. 2, pp. 882 – 888, May 2003.
same for each run. This is indeed a very fast method, which is [8] S. A. Kazarlis, A. G. Bakirtzis, and V. Petridis, “A Genetic Algorithm
the best compared to technique such as EPL, recording a solution to the unit commitment problem,” IEEE Trans. Power Syst.,
duration of 0.72s. This comparison is valid as the computer vol. 11, no. 1, pp. 83 – 92, Feb. 1996.
[9] K. A. Juste, H. Kita, E. Tanaka and J. Hasegawa, “An Evolutionary
with the same speed, Intel Pentium4 CPU (1.5 GHz) is used Programming to the Unit Commitment Problem”, IEEE Trans. Power
for simulation, similar to the one in [7]. Syst., vol. 14, no. 4, pp. 1452-1459, 1999.
[10] A. J. Wood and B. Wollenberg, Power Generation Operation and
Control, 2nd ed. New York: Wiley, 1996.
TABLE IX
COMPARISON OF RESULTS ON 10-UNIT SYSTEM
Total cost [$] X. BIOGRAPHIES
Method Char Time [s]
Best Average Worst Yang Tingfang was born in Luxi, Hunan, P.R.China
DP [8] D — 565825 same same on August 9, 1975. He received his B.S degree and
LR [8] D — 565825 same same M.Sc. degree both from from Huazhong University of
LRGA [15] D — 564800 same same Science & Technology. His interests including power
GA [8] ND 221 565825 — 570032 quality and fault location. He is presently with the
EP [9] ND 100 564551 565352 566231 School of Electrical & Information Engineering,
HPSO [4] ND 62 563942.3 564772.3 565785.3 Changsha University of Science and Technology,
PL EA [6] ND — 563977 — 565451 Hunan Province, 410076, P. R. China.
EPL [7] ND 0.72 563977 — —
ACSA [16] ND — 564049 — — T. O. Ting obtained his B. S. degree from University
MPL D 0.037 563977 same same Sarawak Malaysia, Sarawak, Malaysia, M.E. degree from Multimedia
D = Deterministic University, Malacca, Malaysia and Ph.D. degree from The Hong Kong
ND = Non Deterministic Polytechnic University, Kowloon, Hong Kong. His research interests include
optimization techniques, evolutionary computation and power system
applications.
VIII. CONCLUSION Su Sheng received the B.S degree from Wuhan University of Hydraulic and
Application of MPL is a new approach in solving Unit Electrical Engineering, China in 1998 and M.S degree from Wuhan
Commitment problem. MPL incorporates more intelligent University, China in 2002. He is currently working as a Research Assistant in
strategy with priority list as the backbone. The MPL is simple The Hong Kong Polytechnic University. His research interests include load
and more efficient than conventional priority list method. All modeling and Artificial Intelligence.
the associated constraints are met in the simulation results.
The simple and efficient technique to satisfy the MD
constraint is new in this work. The Economic Dispatch (ED) is
solved using the lambda iteration method. The simplicity of
the MPL and fast calculation of ED leads to a methodological
and competent method in comparison with conventional
method. Application to large scale system proves the
superiority of MPL in terms of computational speed. MPL
provides deterministic solution, a criterion preferred for
industrial applications. Also, it does not encounter premature