www.ietdl.

org
Published in IET Generation, Transmission & Distribution
Received on 15th July 2012
Revised on 17th February 2013
Accepted on 10th March 2013
doi: 10.1049/iet-gtd.2012.0397

ISSN 1751-8687

Dynamic formulation and approximation methods to

solve economic dispatch problems
Mohammed I. Abouheaf1, Wei-Jen Lee2, Frank L. Lewis1
1

The University of Texas at Arlington Research Institute, Electrical Engineering, University of Texas at Arlington, 7300 Jack
Newell Blvd. S., Fort Worth, Texas, TX 76118, USA
2
University of Texas at Arlington, Electrical Engineering, Energy Systems Research Center, UTA Box 19048, Arlington,
Texas 76019, USA
E-mail: [email protected]

Abstract: Economic dispatch (ED) is an optimisation tool that is used to allocate active load demands to the generating units
through optimising the fuel generation cost function subject to the different operational constraints. The high non-linearity of
the power system imposes mathematical challenges in formulating the generation cost function models, which makes the ED
problem hard to solve. This study introduces two ideas to solve issues related to the ED problem. First, a dynamic
formulation technique is developed to optimally allocate the change in the total active load demand to the generating units.
This technique is shown to be insensitive to the optimality of the initial active load distribution unlike the base point and
participation factor method. Moreover, it guarantees an optimal distribution among the generating units due the change in the
active load demand. Second, a novel approximation of the non-convex generation cost function is developed to solve nonconvex ED problem with the transmission losses. This approximation enables the use of gradient and Newton techniques to
solve the non-convex ED problem with valve point loading effect and transmission losses in an analytic approach. This
approximation is compared with some heuristic optimisation techniques.

Introduction

Economic dispatch (ED) is one of the power system energy

management tools that is used to allocate required power
generation to a number of generating units to meet the
active load demand [13]. The operation cost of the power
utilities depends on the fuel cost of the generating units. By
optimising the objective functions that depend on the fuel
cost, the ED results in fuel cost savings, [4]. The generation
cost function models are either smooth or non-smooth
based on the nature of the generating units. Sources of
non-convexity include some physical constraints such as
spinning reserve, transmission losses, prohibited operation
zones, ramp rate limit, valve point loading effect and
multiple fuel options [1]. This introduces mathematical
difculty to the generation cost function by adding the
effect of the ripples to the generation cost function [57].
This makes the ED problem a large scale non-linear
constraint optimisation problem.
Optimisation techniques are used to solve the ED
problems. These methods include Newton-based and
gradient methods, lambda iteration method, base point and
participation factors method [1], interior point algorithm,
linear programming, dynamic programming and dual
quadratic programming [812], where the generation cost
functions are assumed to be monotonically increasing
piece-wise linear functions. Heuristic optimisation
techniques provide solutions to the non-convex ED
866
& The Institution of Engineering and Technology 2013

problems. These techniques include evolutionary algorithms

which are inspired by social nature phenomena such as the
behaviour of insects colonies (ants, bees [3, 13]), genetic
algorithm (GA) [6, 14, 15] evolutionary programming [13],
differential evolution [16], particle swarm optimisation
(PSO) [17], simulated annealing [18], Tabu Search [19],
gravitational search algorithm [4] and biogeography method
[20]. These Heuristic algorithms do not always guarantee
the global best solution.
This paper introduces two ideas to solve some issues
related to the ED problem. First, a dynamic formulation
technique is developed to optimally allocate the change in
the total active load demand to the generating units. This
technique is shown to be insensitive to the optimality of the
initial load distribution and it guarantees an optimal
distribution among the generating units. This technique is
compared with base point and participation factors method.
Second, a novel approximation of the non-convex
generation cost function is developed to solve non-convex
ED problem with the transmission losses. This
approximation enables the use of gradient and Newton
techniques to solve the non-convex ED problem with valve
point loading effect and transmission losses.
This paper is organised as follows. Section 2 introduces
the classical optimisation problem. Section 3 develops the
dynamic formulation technique to allocate the change in the
total active load demand to the generating units. Section 4
veries the feasibility of the proposed dynamic formulation
IET Gener. Transm. Distrib., 2013, Vol. 7, Iss. 8, pp. 866873
doi: 10.1049/iet-gtd.2012.0397

www.ietdl.org
technique by comparing it with the base point and
participation factors method. Section 5 proposes a new
technique to approximate the incremental cost of the
non-convex ED problem. Simulation on a case study is
performed to compare the performance of the proposed
approximation to other Heuristic techniques.

This section introduces the classical optimisation formulation

that is used to solve the ED problem [1, 14, 17].
Mathematically speaking, equivalent generation cost
functions are needed to model the generation fuel cost.
Then the different constraints related to the generation
capacity, losses and spinning reserve are introduced.
Practical generation cost models based on valve point
loading effects and multiple fuel options are also introduced.
Classical Optimisation

The ED problem is formulated using Lagrange dynamics. The

objective of the optimisation problem is to minimise the total
fuel generation cost function, so that
Minimise FT =

ng

Fi (Pi ),

(1)

i=1

where FT is the total generation fuel cost and it is given by

FT = F1 + F2 + F3 + . . . + Fng , Fi is the generation cost of
each unit i, Pi is the power generated by each unit i and ng
is the number of generating units.
The generation cost function Fi is approximated by the
quadratic function [1, 14], so that
Fi (Pi ) = ai + bi Pi + ci Pi2

ng

Nc

(2)

PLosses =

ng ng 

ng




Pi Bij Pj +
Boi Pi + Boo

i=1 j=1

(5)

i=1

where Bij, Boi and Boo are the transmission network power
losses coefcients. The B-loss coefcients represent the
transmission line loss and the corona loss [7].
2.2.2 Generation capacity: Each generating unit has
minimum and maximum generation capacities so that
Pimin Pi Pimax ,

(6)

where Pimin and Pimax are the designed minimum and

maximum generated power capacities of each unit i,
respectively.
2.2.3 Spinning reserve constraints: In the power
system operation, some generators are not working on their
maximum designed capacity, instead those units keep about
510% of their capacity unused [21]. This operation
enhances the security of the power system in the case of
emergencies. Here, the spinning reserve constraints are
given only for the online units without prohibition zones
[21] so that


SRi = min (Pi( max ) Pi ), SRi( max )
i without POZ

(7)

nng

SR =

SRi

(8)

i=1

li wi

(3)

i=1

where Nc is the number of the constraints, i is the Lagrange

multiplier associated with each constraint ji.
The Lagrange operator L is minimised with respect to the
generated power, whereas the constraints ji, i are satised
[1].
2.2

(4)

where PD is the total active load demand, PLosses is the

transmission losses. The transmission losses is given in
terms of Krons loss formula so that

where ai, bi and ci are the fuel cost coefcients of the

generating unit i.
Equation (2) represents the basic generation cost model.
The Lagrange operator L is formulated so that
L = FT +

(Pi ) = PD + PLosses

i=1

2 Formulation of the economic dispatch

problem

2.1

2.2.1 Generation-demand equality constraints: The

generation-demand equality constraint, implies that the sum
of the generated power is equal to the total load demand
plus the transmission losses so that

Operation constraints

The operation constraints are classied into two groups. The

rst group is related to the design specications of the
generating units and the other physical constraints such as
generation capacity, line maximum power ow, generation
ramp limits, prohibited operation zones and spinning
reserve. The second group denes an upper level of
operation control such as unit commitment and other
management plans. Here, only constraints related to the
scope of the work are considered.
IET Gener. Transm. Distrib., 2013, Vol. 7, Iss. 8, pp. 866873
doi: 10.1049/iet-gtd.2012.0397

where SRi is the spinning reserve of unit i (MW), SRi( max) is

the maximum spinning reserve of unit i, SR is the total
spinning reserve given by the generating units that does not
have any prohibited operating zones (POZ).
2.3

Generation cost functions

A simplied generation cost function (2) does not consider

the valve point loading and the multiple fuel options
effects. Here accurate generation cost models are considered.
2.3.1 Economic dispatch with valve point loading
effect: Opening
the
turbines
admission
valves
sequentially causes ripples or non-differentiable points in
the generation cost model [15]. This effect is modelled by
sinusoidal function so that
Fi (Pi ) = ai + bi Pi + ci Pi2






+ ei sin fi Pi( min ) Pi

(9)
867

& The Institution of Engineering and Technology 2013

www.ietdl.org
where ai, bi, ci, ei and fi are the fuel cost coefcients for unit i
and Pi(min) is the minimum generated power by unit i with
valve point loading effect.
2.3.2 Economic dispatch with multiple fuel options:
The generating units can work on multiple fuel options for
different regions in the operation range. This adds more
non-differential points to the generation cost function, so that

a + bi1 Pi + ci1 Pi2 , Pi( min ) Pi Pi1

i1
Pi1 Pi Pi12
ai2 + bi2 Pi + ci2 Pi2 ,
Fi (Pi ) =
:

aik + bik Pi + cik Pi2 , Pi(k1) Pi Pi( max )

(10)

that
DPD =

Dynamic formulation of the ED problem

This section develops a new method to allocate the changes in

the total active load demand to the online generating units.
First an assumption is made such that the current instances
of the generated powers satisfy the optimal power
distribution for the ED problem.
Assumption: Consider a power system of ng generating units.
Assume that for a total active load demand PD, each
generating unit i has an optimal power generation Pi. The
change in the generated power of each generator i due to
the change in the total active load demand PD is denoted
by Pi. The generation cost function for generated power
(Pi + Pi) is given by





2
Fi Pi + DPi = ai + bi Pi + DPi + ci Pi + DPi

(12)

Equation (13) yields



L/DPi = Fi Pi + DPi /DPi l = 0




Fi Pi + DPi /DPi = l

 

 


l = Fi Pi + DPi /DPi = bi + 2ci Pi + DPi

(15)

Taking the summation of (15) for all the generators yields


ng
ng 
ng






bi + 2ci Pi
1 dFi
 

=
DPi
dDPi i=1
2ci
i=1 2ci
i=1
Rearranging this equation yields

ng

g


 

  
1/ 2ci = DPD +
bi + 2ci Pi / 2ci

i=1

i=1

(16)

Rearranging (16) yields



l = DPD +

ng

 n

ng
g




   
 
Pi +
bi / 2ci /
1/ 2ci

i=1

i=1

 

i=1

   

 
bi / 2ci /
1/ 2ci
= PD New +
ng

i=1

ng

i=1

(17)
where PD New = DPD +
demand.

ng
i=1

Pi is the new total load

Equation (15) yields




DPi = l bi + 2ci Pi /2ci

(18)

Then, the optimal generated power PiNew for each unit i

because of the new total load demand PDNew is given by

  
PiNew = l bi / 2ci

(19)

(13)

where l is the Lagrange multiplier or system incremental cost.

The change in the total active load demand PD is given so

Equation (17) reveals that the system incremental cost

depends only on the total required load demand (new load
demand) and the generating units cost coefcients.





+
b
P
+
c
P
+
e

sin
f

P
a


i1
i1 i
i1 i
i1
i1
i( min )
i




Pi( min ) Pi Pi1





ai2 + bi2 Pi + ci2 Pi2 + ei2 sin fi2 Pi1 Pi



Fi (Pi ) =
Pi1 Pi Pi12

:
:
:
: 




+
b
P
+
c
P
+
e

sin
fik Pi(k1) Pi
a

ik
ik i
ik i
ik




P
P P
i(k1)

868
& The Institution of Engineering and Technology 2013

(14)

Equations (12) and (14) yield

Optimising the Lagrange operator with respect to DPi so that



DPi

i=1

where aik, bik and cik are the cost coefcients of each unit i
using fuel type k.
2.3.3 Economic dispatch with valve point loading
effects and multiple fuel options: The generation cost
function because of valve point effects and multiple fuel
options is denoted as hybrid cost function [15]. The
hybrid cost functions result from combining the generation
cost models (9) and (10), so that (see (11))
where aik, bik, cik, eik and fik are the fuel cost coefcients for
each unit i.

ng

(11)

i( max )

IET Gener. Transm. Distrib., 2013, Vol. 7, Iss. 8, pp. 866873

doi: 10.1049/iet-gtd.2012.0397

www.ietdl.org
Equation (19) reveals that the optimal dispatch among the
generating units is found by only knowing the total required
load demand (new load demand) and the cost coefcients
of the generation units. This is not possible with base point
and participation factors method.
Equations (17) and (19) imply that the assumption given in
this section is not needed anymore, because the incremental
cost and the new optimal power for each generation unit i
do not depend on the initial instances (previous instances)
of the generated power.

Rearranging this equation yields (see (22))

Subtracting (21) from (22) yields (see (23))
The correction factor j for each generating unit j is dened
as

 

sj = DPj (proposed) DPj (base point) / DPj (base point)
Or


   
 
ng
ng
PDOld + i=1,i=j
bi / 2ci 2cj Pj i=1
1/ 2ci


sj =
PD New PD Old

4 Comparison between the proposed

method and the base point participation
factors method

(24)

This section veries the feasibility of the dynamic

formulation method. The developed method is compared
with base point and participation factors method. The base
point and participation factors method allocate the change
in the total load demand PD to each generating unit [2].
The change in the generated power Pj by each generating
unit j because of the change in total load demand PD is
given by
DPj (base point) =

DPD 1/Fj

n

g 




1/Fj
/

(20)

i=1

where Fj = 2 Fj /p2j , Fj = 2cj . Then



ng


1
 
DPj (base point) = DPD / 2cj
i=1 2ci

5 Approximate cost function and gradient

technique to solve ED problem with valve point
loading effect

Rearranging this equation yields



g




1
 
DPj (base point) = PDNew PDOld / 2cj
2c
i
i=1


(21)

where PDold is the instance of the previous total load demand,

PD = (PDNew PDold).
Here, a correction factor is developed to show the
dissimilarities between the developed dynamic formulation
and the base point and participation factors methods.
Combining (17) and (18) yields that the change in the
generated power for each generating unit j DPj (using
dynamic formulation) so that (see equation at the bottom of
the page)

DPj =

DPj (proposed) =

Simulation is performed using the case study data given in

Table 1 [1]. Table 2 uses arbitrary (non-optimal) power
distribution satisfying a total load demand (850 MW) as
initial power distribution. Table 2 shows that the generation
fuel cost given by the dynamic formulation is less than the
generation cost given by base point participation factors
method for two new active load demands (870, 900 MW).
Moreover, Table 2 shows that the dynamic formulation is
insensitive to the previous instances of power distribution,
unlike the base point and participation factors. Table 3 uses
optimal power distribution satisfying a total load demand
(850 MW). Table 3 shows that the generation cost given by
the dynamic formulation method is less than or almost
equal to the cost given by the base point and participation
factors method.

Admission valves are operated sequentially in large scale

turbines to give the increase in the production rate [17]. The
admission of these valves adds ripples to the generation
cost function [17]. Moreover, the use of multiple fuel
Table

1 Case study: generation

generation capacities for three units
Unit

1
2
3

cost

coefficients

and

Pmin

Pmax

561
310
78

7.92
7.85
7.97

0.001562
0.00194
0.00482

100
100
50

600
400
200


 n

ng
g






bi
PD New +
1/2ci (bj + 2cj Pj ) /2cj
/
2ci
i=1
i=1

PD New +

ng     ng   
ng   
b
/
2c
1/
2c
P

b
+
2c
i
i
j
i
j j
i=1
i=1
i=1 1/ 2ci
ng   
2cj i=1 1/ 2ci

DPj (proposed) DPj (base point)


 n 


 


ng
ng
ng
g








bi
1
1
1
  + 2cj Pj
 
 
bj
= PD Old +
/ 2cj
2ci
2ci
2ci
2ci
i=1
i=1
i=1
i=1
IET Gener. Transm. Distrib., 2013, Vol. 7, Iss. 8, pp. 866873
doi: 10.1049/iet-gtd.2012.0397

(22)

(23)

869

& The Institution of Engineering and Technology 2013

www.ietdl.org
Table 2 Correction factors and the generation cost values
using non-optimal initial generation distribution (comparison
between base point participation factors and the proposed
dynamic formulation)
Initial demand (PDold = 850 MW), initial random generation
distribution (P1 = 350 MW, P2 = 350 MW, P3 = 150 MW)
new demand (PDNew)
870 MW
900 MW
1
4.5959
1.8384
2
2.0358
0.8143
9.1241
3.6496
3
Base point method
MW
P1, MW
P2, MW
P3, MW
Cost, $/h

870
359.3931
357.5629
153.0440
8384.7

900
373.4828
368.9073
157.6100
8660.7

Proposed method
870
402.5629
342.1667
125.2704
8377.6

900
416.6526
353.5110
129.8364
8653.6

options increases the number of minima to the generation cost

function. This section develops an approximation for the
non-convex generation cost function to solve the
non-convex ED problems with the transmission losses [14,
17, 22]. The incremental cost is approximated by a
piece-wise linear function. Newton and gradient techniques
use this approximation to solve the non-convex ED problem.
The proposed approximation is used to solve the ED
problem with valve point loading effect (EDVP), where the
cost function (9) is considered. Moreover, the proposed
approximation is used to lay out the solution framework for
the hybrid ED problem with the hybrid cost function (11).
Rearranging the non-convex generation cost function (9)
yields
 
 
Fi Pi = ai + bi Pi
 2
 

+ ci Pi +hi ei sin fi Pi min Pi

(25)

where i = sgn(sin( fi (Pimin Pi))) or



hi =

 


1, 0 mod
 fi Pi Pi min
 , 2p
 p
1, p mod fi Pi Pi min , 2p 2p

Fig. 1 Piecewise linear approximation of the non-convex

generation cost function

Pi, together with Fig. 1 and (26) yield



 

dFi
= l = bi + 2ci Pi hi ei fi cos fi Pi min Pi
(27)
dPi
Equations (26) and (27) yield



 
 
dFi
= l = bi + 2ci Pi + ei fi cos mod fi Pi min Pi , p
dPi
(28)
where

 


 
 
d ei sin fi Pi min Pi
= ei fi cos mod fi Pi min Pi , p
dPi
and mod is the modulus after division function.
Since the non-convex part is periodic as shown in Fig. 1b,
then ei ficos(mod( fi (Pijmin Pi), )) can be approximated by a
piece-wise linear function with different slopes in the period
[0, ] as shown in Fig. 1c. It is worth to say that, this
approximation works for all periods related to generator i.
This approximation yields

(26)

Fig. 1a shows graphically, the non-convex part of the

generation cost functions (9) and (25). Fig. 1b shows the
derivative of the non-convex cost function, which is
periodic. Fig. 1c shows the piecewise linearisation of this
non-convexity for one interval [0, ].
Taking the derivative of the generation cost function (25)
with respect to the generated power for each generating unit


 
 
ei fi cos mod fi Pi min Pi , p
 



= gi Pi ni p/fi + hi

(29)

where gi and hi are approximation constants and ni = (Pi

Pimin)/(/fi) is an integer number of periods for each
generator i. Here, the approximation constants gi and hi are
specic to each generator i and they can be saved to a look
up table for easier retrieve.

Table 3 Correction factors and the generation cost values using optimal initial generation distribution (comparison between Base
Point Participation Factors and the proposed dynamic formulation)
Initial demand (PDOld = 850 MW), initial optimal generation distribution (P1 = 393.17 MW, P2 = 334.604 MW, P3 = 122.226 MW)
870 MW
900 MW
new demand (PDNew)
17.4
6.9
1
32.4
12.9
2
133.9
53.6
3
Base point method
MW
P1, MW
P2, MW
P3, MW
Cost, $/h

870
402.5631
342.1669
125.2700
8377.614813332205

870
& The Institution of Engineering and Technology 2013

Proposed method

900
416.6528
353.5113
129.8360
8653.603253856323

870
402.5629
342.1667
125.2704
8377.614813331242

900
416.6526
353.5110
129.8364
8653.603253855362

IET Gener. Transm. Distrib., 2013, Vol. 7, Iss. 8, pp. 866873

doi: 10.1049/iet-gtd.2012.0397

www.ietdl.org
The generated power is given by

Using (29) in (28) yields






dFi 
= gi + 2ci Pi + hi gi ni p/fi + bi
dPi

(30)

Since (30) represents an approximation of the non-convex

generation cost function, it is useful to extend the
formulation of the optimisation problem to include the
transmission losses. The benet of using the proposed
approximation is that it enables using analytic techniques to
solve EDVP with the transmission losses, which is difcult
to achieve with the heuristic techniques.
The transmission losses formula [1, 7, 17] is given by
PLosses = P T [B]P + BTo P + Boo

(31)

where P is a vector of the generated power (MW) of the all

generating units. B, Bo and Boo are the B loss coefcients.
Matrix B is symmetric such that B = B T
Considering the classical optimisation problem, the
equality constraint is given by

w = (PD ) + PLosses (Pi )

ng

(Pi ) = 0

(32)

i=1

The Lagrange operator is given by

L = FT + lw

(33)

Taking the derivative of (31) with respect to the power vector

P yields
dP losses
= 2[B]P + Bo
dP

(34)

Rearranging (30) for all the generating units yields

 
 
 
dF
= m1 n x1 + m2 n xn [P ]ng x1
g
g g
dP ng x1

(35)

m1 =

b1 g1 n1 (p/f1 ) + h1
:

:
bng gng nng (p/fng ) + hng

2c1 + g1 .
0

:
:
:
m2 =

0
. 2cng + gng

and

Applying (dL/dP) = 0 on (33) yields



 
 
dF
dP Losses
= l [1]ng x1
dP
dP ng x1
ng x1
Using (34) and (35) in (36) yields
 
 
m1 n x1 + m2 n xn [P ]ng x1
g
g g


= l [1]ng x1 2[B]ng xng [P]ng x1 [Bo ]ng x1
IET Gener. Transm. Distrib., 2013, Vol. 7, Iss. 8, pp. 866873
doi: 10.1049/iet-gtd.2012.0397

1

m2 n xn +2l[B]ng xng
g g


 

l[1]ng x1 m1 n x1 l[Bo ]ng x1
g

gi (Pi ),hi (Pi )

(38)
The constraint (32) yields
PD New + PLosses (P)

ng

(P) = 0

(39)

i=1

where PD New is the new active load demand.

Equation (38) gives a closed form for the optimal power
distribution in terms of the system transmission losses, fuel
cost coefcients and approximation parameters. The given
approximation enables the use of Newton method and
gradient techniques to solve the non-convex ED with valve
point loading effect and transmission losses.
Similar results are obtained for the generation systems that
use multiple admission valves and multiple fuel options. The
generation cost model is then described by the hybrid
generation cost functions (11). A similar setup for the
problem with hybrid cost function can be done using the
similar procedure described before. The only difference is
with the values of the indices m1 and m2. These indices
reect the effect of using multiple fuel types so that

p
+ h1,j
b1,j g1,j n1,j

f1,j

and
m1 =

+ hng,i
bng,i gng,i nng,i
fng,i

2c1,j + g1,j .
0

:
:
:
m2 =

. 2cng ,i + gng ,i

This adds more importance of this approximation in the

multiple fuel options case or the hybrid ED problem.
To test the feasibility of this approximation, Newton
method is used to solve the ED problem with valve point
loading effect and transmission losses. The results are
compared with some other techniques. First, ED with valve
point loading effect is considered, and then the transmission
losses is considered.
Newton method reveals that the iteration equations for each
generating unit are given by



where

[P ]ng x1 =



 l+1  l 
Pi
= Pi df /dPi l 1 f (P, l) l



(l)l+1 = (l)l df /d l 1 f (P, l)
l

(36)

where l is the iteration index and f (P, )|i is the constraint

calculated at each step l so that


ng






(P) Plosses (P) PD New
f (P, l) l =


i=1

(37)

(40)

Here, the values of df/dPi and df/d are derived. Equation (36)
871

& The Institution of Engineering and Technology 2013

www.ietdl.org
for each generator i yields


Table

4 Case study: generation cost

generation capacities of six thermal units




p
gi + 2ci Pi + hi gi ni + bi
fi


ng


Bij Pj Boi
=l 12
j=1

Rearranging this equation yields



ng

p
Bij Pj + hi gi ni + bi
gi + 2ci + 2lBii Pi + 2l
fi
j=1,j=i

a, $

b, $/MW

c, $/MW2

Pmin

Pmax

1
2
3
4
5
6

240
200
220
200
220
190

7
10
8.5
11
10.5
12

0.007
0.0095
0.009
0.009
0.008
0.0075

300
200
150
150
150
150

0.031
0.042
0.063
0.063
0.063
0.063

100
50
80
50
50
50

500
200
300
150
200
120

An expression for d f/d is derived so that

 T
df
dP df
=
dl
dl dP

The generated power for each generator i is given by

Taking the derivative of each generated power with respect to

yields (see equation at the bottom of the page)
Simplifying this equation yields (see equation at the bottom
of the page)
Rearranging this equation yields







gi + 2ci 1 Boi + 2Bii hi gi ni p/fi + bi
dPi
=

2
dl
gi + 2ci + 2lBii
 ng
2(gi + 2ci ) j=1,j=i
Bij Pj

(42)
2
gi + 2ci + 2lBii

and

Unit

= l lBoi


 



 ng
l 1 Boi hi gi ni p/fi + bi 2l j=1,j=i
Bij Pj


Pi =
gi + 2ci + 2lBii
(41)

coefficients

(45)

A numerical example is used to verify the feasibility of this

approximation. The data of this example are given in Table 4.
The transmission power losses are expressed in terms of
Krons loss formula. The losses coefcients Bij, Boi and Boo
are given as follows (see equation at the bottom of the
next page)
Table 5 shows that the proposed approximation achieved
the lowest fuel generation cost compared with PSO [20],
GA [23] and RGA [24]. The proposed approximation
shows lower generation cost compared with other Heuristic
techniques such as genetic algorithms as emphasised in
Table 5. The proposed approximation is used to perform
Newton method simulation to solve the ED problem for
active load demands (PD = 1080, 1100, 1220 and 1240
MW) as shown in Table 6.

Table 5 Comparison of Newton method with the proposed

The gradient of the constraint f with respect to the power

generated by each unit i is given by
df
dP
= 1 Losses
dPi
dPi

(43)

Using (36) and (43) yields

1

 dFi
dPLosses 
= 1/l
dPi
dP
 i

= (1/l) bi + 2ci Pi

 
 
+ei fi cos mod fi Pi min Pi , p

approximation and some heuristic techniques to Solve EDVP for

active load demand (PD = 1263 MW)
Unit

GA [23]

RGA [24]

PSO [20]

Approximation

P1, MW
P2, MW
P3, MW
P4, MW
P5, MW
P6, MW
losses, MW
T. G. P.a
T.G.C.b

474.807
178.636
262.209
134.283
151.904
74.181
13.022
1276.03
15 459

420.2342
199.4412
263.7234
120.0030
167.2319
105.1250
13.2627
1275.8
15 461

432.9639
170.5198
261.9009
116.9111
190.4102
103.4931
13.142
1276.2
15458.56

467.5062
168.2134
268.1626
117.8149
167.3637
87.3330
13.3938
1276.4
15 457

(44)

T.G.P., Total generated power (MW)

T.G.C., Total generation cost ($/h)




ng
gi + 2ci + 2lBii 1 Boi 2 j=1,j=i
Bij Pj
dPi
=

2
dl
gi + 2ci + 2lBii






ng
2Bii hi gi ni p/fi + bi + 4lBii j=1,j=i
Bij Pj 2lBii 1 Boi
+

2
gi + 2ci + 2lBii






ng
dPi (gi + 2ci ) 1 Boi 2 j=1,j=i Bij Pj + 2Bii hi gi ni p/fi + bi
=

2
dl
gi + 2ci + 2lBii
872
& The Institution of Engineering and Technology 2013

IET Gener. Transm. Distrib., 2013, Vol. 7, Iss. 8, pp. 866873

doi: 10.1049/iet-gtd.2012.0397

www.ietdl.org
Table 6 Using Newton method with the proposed
approximation to solve EDVP for active load demands (PD = 1080,
1100, 1220, 1240 MW)
Unit
P1, MW
P2, MW
P3, MW
P4, MW
P5, MW
P6, MW
losses
T.G.P.
T.G.C., $/h

1080

1100

1220

1240

378.1635
68.3878
267.5805
94.9455
193.8721
88.1038
11.0533
1091.1
13 800

460.3753
163.2718
258.9505
100.6888
56.2446
71.3900
10.9211
1110.9
14 227

473.7645
137.4892
298.4074
68.7877
199.7594
56.1158
14.3240
1234.3
15 650

466.1173
192.7883
284.6906
68.1629
168.1531
74.0882
14.0004
1254
16 008

Conclusions

This paper proposed and developed two techniques to solve

issues related to the ED problem. First, a dynamic
formulation technique is developed to optimally allocate the
change in the active load demand to the generating units.
This technique achieves the lowest fuel generation cost
compared with base point and participation factors method.
Moreover, this method is found to be insensitive to the
previous instances of power distribution (either optimal or
non-optimal), which is impossible to achieve with base
point and participation factors method.
Second, a mathematical approximation is developed to solve
the non-convex ED problem with transmission losses. This
approximation simplies the incremental cost to a piecewise
linear function. This approximation enables the use of
Newton and other gradient techniques to solve non-convex
ED problem, which was hard analytically. Furthermore, a
closed form of the optimal power distribution is given in
terms of the system incremental cost, fuel cost coefcients,
fuel types and the developed approximation parameters.

Acknowledgments

This work was supported by NSF grant ECCS-1128050,

ARO grant W91NF-05-1-0314, AFOSR grant FA 9550-091-0278, China NNSF grant 61120106011, and China
Education Ministry Project 111 (No.B08015).

References

1 Wood, A.J., Wollenberg, B.F.: Power generation, operation and

control (Wiley, New York, 1996, 1984, 2nd edn.)
2 Raol, C.S., Nagaraju, S.S., Raju, P.S., Reddy, M.R.: Optimized integral
controller for economic dispatch in a two area system based on
Hooke-Jeeves algorithm, J. Eng. Appl. Sci., 2007, 2, (6), pp. 4956
3 Niknam, T., Mojarrad, H.D., Meymand, H.Z., Firouzi, B.B.: A new
honey bee mating optimization algorithm for non-smooth economic
dispatch, Energy, 2011, 36, (2), pp. 896908

0.0017

0.0012

0.0012
0.0014

0.0007
0.0009
Bij =
0.0001 0.0001

0.0005 0.0006
0.0002 0.0001

4 Duman, S., Guvenc, U., Yorukeren, N.: Gravitational search algorithm

for economic dispatch with valve point effects, Int. Rev. Electr. Eng.,
2010, 5, (6), pp. 28902895
5 Amjady, N., Nasiri-Rad, H.: Economic dispatch using an efcient
real-coded genetic algorithm, IET Gener. Trans. Distrib., 2009, 3,
(3), pp. 266278
6 Amjady, N., Nasiri-Rad, H.: Solution of non-convex and non smooth
economic dispatch by a new adaptive real coded genetic algorithm,
Exp. Syst. Appl., 2010, 37, (7), pp. 52375245
7 Lukman, D., Walshe, K., Blackburn, T.R.: Loss minimisation in
industrial power system operation. Australasian Universities Power
Engineering Conf. (AUPEC2000), Brisbane, Australia, 2000, pp. 1520
8 Lin, W.M., Chen, S.J.: Bid-based dynamic economic dispatch with an
efcient interior point algorithm, Electr. Power Energy Syst., 2002,
(24), pp. 5157
9 Granelli, G.P., Montagna, M.: Security-constrained economic dispatch
using dual quadratic programming, Electr. Power Syst. Res., 2000, 56,
(1), pp. 7180
10 Lin, C.E., Viviani, G.L.: Hierarchical economic dispatch for piecewise
quadratic cost functions, IEEE Trans. Power Appl. Syst., 1984,
PAS-103, (6), pp. 11701175
11 Yang, H.T., Chen, S.L.: Incorporating a multi-criteria decision
procedure into the combined dynamic programming/production
simulation algorithm for generation expansion planning, IEEE Trans.
Power Syst., 1989, 4, (1), pp. 165175
12 Chen, C.L., Wang, C.L.: Branch-and-bound scheduling for thermal
generating units, IEEE Trans. Energy Convers., 1993, 8, (2), pp.
184189
13 Yang, H.T., Yang, P.C., Huang, C.L.: Evolutionary programming based
economic dispatch for units with non-smooth fuel cost functions, IEEE
Trans. Power Syst., 1996, 11, (1), pp. 112118
14 Walters, D.C., Sheble, G.B.: Genetic algorithm solution of economic
dispatch with valve point loading, IEEE Trans. Power Syst., 1993, 8,
(3), pp. 13251331
15 Chinag, C.L.: Improved genetic algorithm for power economic dispatch
of units with valve-point effects and multiple fuels, IEEE Trans. Power
Syst., 2005, 20, (4), pp. 16901699
16 Coelho, L.S., Mariani, V.C.: Combining of chaotic differential
evolution and quadratic programming for economic dispatch
optimization with valve-point effect, IEEE Trans. Power Syst., 2006,
21, (2), pp. 98996
17 Chaturvedi, K.T., Pandit, M., Srivastava, L.: Particle swarm
optimization with crazy particles for nonconvex economic dispatch,
Appl. Soft Comput., 2009, 9, (3), pp. 962969
18 Wong, K.P., Fung, C.C.: Simulated annealing based economic dispatch
algorithm, IEE Proc. C, 1993, 140, (6), pp. 509515
19 Lin, W.M., Cheng, F.S., Tsay, M.T.: An improved Tabu search for
economic dispatch with multiple minima, IEEE Trans. Power Syst.,
2002, 17, (1), pp. 108112
20 Vanitha, M., Thanushkodi, K.: An efcient technique for solving the
economic dispatch problem using biogeography algorithm,
Eur. J. Sci. Res., 2011, 50, (2), pp. 165172
21 Ciornei, I., Kyriakides, E.: A GA API solution for the economic
dispatch of generation in power system operation, IEEE Trans. Power
Syst., 2011, 27, (1), pp. 19
22 Min, K., Lee, S., Moon, Y.: An economic dispatch algorithm as
combinatorial optimization problems, Int. J. Con. Aut. and Sys.,
2008, 6, (4), pp. 468476
23 Gaing, Z.L.: Particle swarm optimization to solving the economic
dispatch considering the generator constraints, IEEE Trans. Power
Syst., 2003, 18, (3), pp. 11871195
24 Guvenc, U., Duman, S., Saracoglu, B., zturk, A.: A hybrid GA-PSO
approach based on similarity for various types of economic dispatch
problems, Electronics and Electrical Engineering Journal 2011, 2,
(108), pp. 109114

0.0007

0.0001

0.0009
0.0031

0.0001
0

0
0.001

0.0024
0.0006

0.0006 0.0008

0.0005 0.0002

0.0006 0.0001

0.001 0.0006

0.0006 0.0008

0.0129 0.0002

0.0002

0.015

Bo = 0.001[0.3908 0.12970.70470.05910.2161 0.6635], Boo = 0.0056

IET Gener. Transm. Distrib., 2013, Vol. 7, Iss. 8, pp. 866873
doi: 10.1049/iet-gtd.2012.0397

873

& The Institution of Engineering and Technology 2013