View metadata, citation and similar papers at core.ac.

uk brought to you by CORE

provided by Elsevier - Publisher Connector

Engineering Science and Technology, an International Journal 19 (2016) 2035–2042

Contents lists available at ScienceDirect

Engineering Science and Technology,

an International Journal
journal homepage: www.elsevier.com/locate/jestch

Full Length Article

Modified radial movement optimization (MRMO) technique for

estimating the parameters of fuel cost function in thermal power plants
M. Vanithasri ⇑, R. Balamurugan, L. Lakshminarasimman
Department of Electrical Engineering, Annamalai University, Annamalai Nagar 608002, Tamil Nadu, India

a r t i c l e i n f o a b s t r a c t

Article history: Accurate estimation of coefficients of fuel cost function in thermal power plants is an important problem
Received 25 April 2016 for economical operation of a power system. In this paper, a novel stochastic method called modified
Revised 21 July 2016 radial movement optimization (MRMO), is developed for estimating fuel cost coefficients of thermal gen-
Accepted 21 July 2016
erating units. Radial movement optimization is a novel global optimization technique used to solve the
Available online 1 August 2016
complex optimization problems. In this work, four case studies are considered for implementing the RMO
technique in fuel cost parameter estimation problems. The coefficients of fuel cost curve functions in
Keywords:
thermal power plants are found for first, second and third order models. Valve point effect is included
Radial movement optimization
Economic dispatch
in a third order model for estimating the cost coefficients for 13 unit systems using the proposed method.
Optimization The results show that the RMO settles at local optima. Hence the algorithm is modified and termed as
Thermal power plant MRMO to converge on better results. From the results, it is inferred that the MRMO outperforms other
techniques like particle swam optimization (PSO), teaching learning based optimization (TLBO), least
error square (LES), artificial bee colony (ABC) algorithm based on the low total error.
Ó 2016 Karabuk University. Publishing services by Elsevier B.V. This is an open access article under the CC
BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction characteristics function of generating units must be estimated

accurately. Many factors affect the parameters of this function,
The production cost for electricity can be divided into three cat- such as the working ambient operating temperature and aging of
egories namely, facility construction, ownership cost and operating the generating units. Many utilities use obsolete data to represent
costs. Of these three, operating cost is the most significant, and fuel cost function in their ED calculations. This naturally reduces
hence the emphasis is on the economics of the operation. Economic significance of ED computations from the actual optimal operating
dispatch (ED) is a vital task in power system operation. It can be state of a given generating plant. Thus, a realistic approximation of
formulated as an optimization problem with the objective of min- the fuel cost function to the actual cost curve by periodically esti-
imization of total generation cost by satisfying the operating con- mating the cost function parameters is crucial in order to improve
straints. Finding solution to this nonlinear objective function the final accuracy of the results in solving ED problems [1]. Hence a
with large number of constraints is complex. One of the most powerful and reliable estimation technique is essential for the esti-
important optimization in a power system problem is to find an mation of the parameters of the fuel cost curve function.
accurate value of the thermal unit fuel cost curve parameters. So far, only few works are available on the estimation of fuel
The accuracy of the estimated coefficients affect the final accuracy cost curve parameters. In the past decades, several techniques
of the ED process. Hence, it is vital to employ an accurate and reli- are employed to solve the complex fuel cost problem. Each method
able estimation technique for finding the input–output curve has its own merits and demerits. Different mathematical models
coefficients. are used for representing the fuel cost function. Various static esti-
Various mathematical models are used for representing the fuel mation techniques, such as recursive mathematical method [2],
cost function. Fuel cost function is mainly classified as smooth and least square error (LSE), Gauss–Newton method (Bard algorithm),
non-smooth models. The coefficients in the thermal input–output Marquardt algorithm, Powell regression algorithm [3], weighted
least squares multiple linear regression method [4], sequential
regression technique [5], least absolute value approximations [6],
⇑ Corresponding author.
Gram–Schmidt orthonormalization and least-squares [7], have
E-mail address: vanithasimman@gmail.com (M. Vanithasri).
been employed for finding the input output characteristics of fuel
Peer review under responsibility of Karabuk University.

http://dx.doi.org/10.1016/j.jestch.2016.07.012
2215-0986/Ó 2016 Karabuk University. Publishing services by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
2036 M. Vanithasri et al. / Engineering Science and Technology, an International Journal 19 (2016) 2035–2042

cost function in thermal power plants. Dynamic methods like Kal- 3. Radial movement optimization
man filter [8], square root filter (SRF) [9] have also been utilized for
solving the ED problems. These mathematical methods need high An efficient tool, for global optimization of multivariable com-
computational time and yield poor performance due to gross plex system, called radial movement optimization (RMO) is devel-
errors. Also these techniques rely on initial condition and convex- oped by Rasoul Rahmani and Rubiyah Yusof (2014) [25]. The
ity to find the global optimum; methods based on these assump- swarm based stochastic optimization technique strikes similarity
tions do not guarantee to find the global optimum when with other evolutionary methods like PSO and differential evolu-
considering practical of generator constraints [10,11]. tion (DE). The search space in RMO, is a vector space wherein a
Many heuristic and stochastic optimization methods such as location vector associated with a particle in space proposes a solu-
the evolutionary programming [12] genetic algorithm (GA) [13], tion to the problem. The movement style of the particle in RMO dif-
particle swarm optimization (PSO) [14–16] are employed to find ferentiates it from other optimization methods. The particles get
the fuel cost function. In recent years, global optimization methods scattered from the center point and gets constantly updated at
such as, artificial bee colony (ABC) algorithm [1], Lagrange’s multi- each step in RMO. In a three dimension search space, the particles
plier method [17], modified artificial bee colony (MABC) [18], pat- with different velocities are seen spread along the radii of a sphere.
tern search technique [19], Cuckoo search [20,21], teaching The location of the particle is evaluated by an objective function to
learning based optimization (TLBO) [22], simulated annealing return an objective value and in turn location of the best particle is
(SA) [23] and Flower pollination algorithm [24] are applied to esti- obtained. The unique feature of this method is that the radial
mate the fuel cost coefficients. In this paper, a simple and efficient movement and the updation of particles enhance the exploration
stochastic optimization algorithm is employed to find the input and exploitation capabilities to converge on global optimal values.
output parameters of thermal power plants. RMO and MRMO are RMO has an edge over other search techniques as it requires a
applied to four different case studies to validate the proposed smaller memory and due to its ability to carry out better and den-
approach for estimating the fuel cost function of thermal generat- ser search around the target [25]. It is made possible due to the fact
ing units. RMO technique is modified as MRMO to give better accu- that the particle start moving from an updated point at every iter-
racy of the fuel cost function. The results obtained are compared ation and the location and velocity for particle are not transferred
with ABC, PSO, LSE and TLBO algorithms. between iteration. The other search techniques in contrast are suit-
able only for small and medium sized problems as the computa-
2. Fuel cost curve models tional cost increases with size. This technique is also robust in
the sense that the global best vector in the updating process pre-
In an ED problem, the fuel – cost function can be represented in sents the starting of algorithm in a local optimum.
terms of smooth function. The fuel cost curve for thermal generat- The objective function returns a fitness value for particle at the
ing unit (i) can be described as a polynomial function as, end of each generation step. The location of the best fitness value
for each generation obtained among all evaluated particle is stored
X
L as the best solution, Rbest. Similarly the best solution obtained
F i ðPti Þ ¼ aoi þ aji Piti þ r ci i ¼ 1; 2; . . . N ð1Þ among all the generations is stored as Gbest. Either the number of
j¼1 generations or a definition of a small error for Gbest from a desired
value, or both serves as the termination criteria depending on the
where choice of test function. The algorithm of RMO with components is
Fi – fuel cost function of the ith generating unit well explained in reference [25].
Pti – electrical power output of the ith generating unit, MW
aoj, aji – cost coefficients of ith generating unit
3.1. Choosing coefficients of RMO
rci – error related to ith generating unit
L – equation order
The two coefficients, C1 and C2 of RMO has a direct implication
N – number of generating units
on the values of Gbest and Rbest respectively. These two coefficients
allow the user to fix the optimization level as required for specific
The smooth cost functions are modeled in linear, quadratic and
applications. As far as C1, the coefficient of Gbest is concerned, larger
cubic forms. They are expressed as follows:
is the C1 if the search space is large and if the problem is nonlinear.
For linear model (First order), L = 1
Higher values of C1 increase the pace of convergence but reduce
the quantity of optimization and can be typically fixed in the range
F i ðPti Þ ¼ aoi þ a1i Pti þ rci ð2Þ
of 0.4–0.9 but should not exceed unity. C2, the coefficient of Rbest,
can be chosen in line with the range of C2 but should be greater
For quadratic form (Second order), L = 2
than C1. Higher values of C1 and C2 are preferred if a fast solution
F i ðPti Þ ¼ aoi þ a1i Pti þ a2i P2ti þ r ci ð3Þ is desired but lower values are usually better for unknown func-
tions. But when the value of C1 and C2 is greater than one, it not
For cubic form (Third order), L = 3 only increases the length of the up vector but also retorts the con-
F i ðPti Þ ¼ aoi þ a1i Pti þ a2i P2ti þ a3i P3ti þ r ci ð4Þ vergence of the algorithm to settle at a global optimum.

3.2. Modified radial movement optimization for parameters estimation

2.1. Fuel – cost function: non smooth
of fuel cost functions
The valve point effects produce ripples in the input–output
The algorithm for modified radial movement optimization for
characteristics and this effect can be incorporated in the fuel cost
parameters estimation of fuel cost functions is shown in Fig. 1.
function by adding a sinusoidal term [18] and is
" #
X
L
j
3.2.1. Step – I: initialization
F i ðPti Þ ¼ aoi þ aji Pti þ jei sinðf i ðPi;min  Pti ÞÞj þ r ci ð5Þ The initial particles in the search space are randomly generated
j¼1
within their limits. The control variables are fuel cost coefficients
 M. Vanithasri et al. / Engineering Science and Technology, an International Journal 19 (2016) 2035–2042 2037

Start

Read the system data

Initialize the particles in the population

randomly within the search space

Evaluate the fitness of each particle using equation (8)

Choose the best as Centre

Set counter = 1

Sprinkle the particles from the

centre point using equation (9)

Evaluate the fitness of each particle based on objective. The particle which gives the
minimum error is choosen as Rbest

Generate the trial vectors around Gbest using equation (11).

No
If fitness of trial vector is better than the Gbest fitness, Then replace Gbest by the trial
vector. Repeat this step for Nt trials

Calculate the new centre

location from equation (11)

Yes
If iteration counter = max
No

Print the results Increment the iteration

counter by 1

End

Fig. 1. Flowchart for fuel cost parameter estimation using MRMO.

and the size of initial population is m  n matrix. Here m is the V ij ¼ r and ð0; 1Þ  V maxðjÞ
number of particles and n is the number of control variables. The
initial population is represented as X maxðjÞ  X minðjÞ
V max ðjÞ ¼ j ¼ 1; 2; . . . n ð7Þ
2 3 K
X 1;1 X 1;2  X 1;n
6 .. .. 7 where K is an integer which is chosen judiciously. Xmax(j) and Xmin(j)
6X .  . 7
6 2;1 7 are the maximum and minimum limits of jth control variable
X ij ¼ 6 7 ð6Þ
6 .. .. .. 7 respectively. The velocity vector has an inertia weight (W) associ-
4 .  . . 5
ated with it, which is chosen as constant in the proposed method.
X m;1 X m;2    X m;n
3.2.3. Step – 3: locate the center point
where i = 1, 2, 3. . .m and j = 1, 2,. . .n.
The fitness of each particle in the population vector is estimated
using the following equation
3.2.2. Step – 2: generate the initial velocity vector
The velocity vector Vij is randomly generated using the follow- rCi ¼ F i actual F i estimated i ¼ 1; 2; . . . . . . . . . m ð8Þ
ing equation
2038 M. Vanithasri et al. / Engineering Science and Technology, an International Journal 19 (2016) 2035–2042

4.1. Case study – I

r – error vector
F i actual - actual fuel cost of im generating unit ($/h) The data collected for the three units viz. coal, oil and gas [16]
F i estimated – estimated fuel cost of im generating unit ($/h) are utilized to find the parameters of fuel cost functions by first
order model given in Eq. (2). The coefficients and estimated values
The population vector which gives the minimum error is chosen of fuel cost function by RMO and MRMO, along with the error are
as the initial center point (Cp) and it is the Gbest vector in the initial given in Table 1. The fuel cost function generated by RMO and
population. MRMO are compared with the other techniques such as ABC, PSO
and LES [1,16] and are given in Table 1. Even though the error
3.2.4. Step – 4: Find the radial movement of the particle for 10 MW is high (more than 25) for all the fuels, the error is found
The radial movement of the particles from the center point Cp is to be zero for 20 MW and 40 MW for all the fuels using MRMO.
calculated by Hence, the total error gets reduced. The total errors are found to
be 32.62, 34.55 and 34.4 for coal, oil and gas respectively using
U ij ¼ W  V ij þ centerðjÞ i ¼ 1; 2; . . . m; j ¼ 1; 2; . . . n ð9Þ
the MRMO method. These total errors that’s obtained are less when
W is the inertia weight which is calculated by compared to RMO and other reported methods such as ABC, PSO
  and LES for all the fuels. The total error obtained by MRMO, when
W max  W min
W ¼ W max   Iteration ð10Þ compared to RMO, is lesser by 0.38, 0.149, 0.534 for coal, oil, gas
Iterationmax
respectively. It shows that the modification in the RMO algorithm
where Wmax = 1 and Wmin = 0. improves the accuracy of the result by decreasing the error. Simi-
Here, the particles are sprinkled. If any control variable in the larly the error values are lower by 0.012, 0.082, 0.0290 for coal,
population vector Uij violates their maximum or minimum limits oil, gas respectively when compared to ABC. But when compared
then that variable is fixed at their violated limit. to other methods, the total error is found to be, more than 4 for
PSO and more than 6 for LES for all the three units. It shows that
3.2.5. Step – 5: evaluate the fitness of the particles the MRMO estimated values are closer to the actual values than
The fitness of each vector in the population Uij is evaluated. The the available methods in the literature. The convergence is shown
particle which gives the minimum error is selected as radial best in Fig. 2. From the figure it is inferred that the proposed technique
(Rbest) vector. converges to a better solution within the first 20 iterations.

3.2.6. Step – 6: search around Gbest 4.2. Case study – II

The convergence of radial movement optimization is improved
by generating number of trial vectors around the Gbest using The data used in case – I is used to find the parameters of sec-
ond order model given in Eq. (2). The coefficients and estimated
V ij ¼ Gbest þ unifrndð0; 1Þ  ðU r1j  U r2j Þi ¼ 1; 2; . . . m; j ¼ 1; 2; . . . n
values of fuel cost by RMO and MRMO along with the error are
ð11Þ given in Table 2. The estimated fuel cost values by RMO and MRMO
are compared with the estimated values by other techniques such
where r1 and r2 are randomly selected vector which are mutually
as teaching learning based optimization (TLBO) [22], PSO, ABC and
different from each other. The fitness of each vector in Vij is evalu-
LES [1,16] and are presented in Table 2. The error for 20 MW is
ated and are compared with Gbest. If the fitness of trial vector is bet-
more than 8 for all the fuels, but the error is found to be nearly zero
ter than Gbest, trial vector replaced the Gbest vector. This procedure is
for 10 MW and 30 MW for coal and oil as fuel. The error is found to
repeated for Nt times for each vector in every iteration.
be zero for 10 MW, 30 MW and 50 MW for gas using MRMO.
Hence, the total error gets reduced while using MRMO. The total
3.2.7. Step – 7: determine the new center location
errors are found to be 9.7601, 9.9774 and 9.7500 for coal, oil and
The new center location (center 1) is updated using the follow-
gas respectively using the MRMO method. The total errors are
ing equation
found to be less when compared to RMO and other reported meth-
CenterKþ1 ¼ CenterK þ ðC 1 ½Gbest  CenterK Þ þ ðC 2 ½Rbest  CenterK Þ ods such as TLBO, ABC, PSO and LES for all the fuels. The total error
ð12Þ is reduced by 0.0339, 0.0705, 0.0155 using MRMO for coal, oil, gas
respectively when compared to RMO. It shows that the modifica-
where K is the iteration number. In this study, the values of C1 and C2 tion in the RMO algorithm improves the accuracy of the result by
are choosen as 0.8 and 0.9 respectively and the population size is 50. decreasing the error. Similarly, the error values are gets lower by
The fitness of radial best vector (Rbest) is compared with the fit- 0.0069, 0.0806, 0.0090 for coal, oil, gas respectively when com-
ness of global best (Gbest) vector. If Rbest is better than Gbest, Gbest is pared to TLBO. But when compared to other methods the total
replaced by Rbest. error is found to be, more than 0.5/1.8/3 (coal/oil/gas) using PSO
The iteration is incremented by 1. The steps 2–7 is repeated for and more than 4.5 using LES for all the three units. It shows that
predefined iterations. the MRMO estimated values are closer to the actual values than
the available methods in the literature. The convergence is shown
4. Results and discussion in Fig. 3. From the figure it is inferred that the proposed technique
converges to a better solution within the first 20 iterations. It is
RMO and modified RMO techniques are employed to test differ- also found that the error is less when compared to first order
ent types of thermal generating units for estimating the fuel cost model employed in case-I.
curve function. The algorithm is applied in MATLAB 2011 and the
simulation results are presented in Tables 1–4. Simulation is per- 4.3. Case study – III
formed in an Intel core i3 PU with 2 GB RAM. The data are collected
from the literature [15,16,22]. The values are for coal, oil and gas The same data is used in case – III to find the parameters of third
with power ranging from 10 to 50 MW. The RMO and MRMO are order model given in Eq. (3). The coefficients and estimated values
applied to four cases described in Eqs. (2)–(5) and the results are of fuel cost by RMO and MRMO along with the error are given in
analyzed. Table 3. The fuel cost calculated by RMO and MRMO are compared
 M. Vanithasri et al. / Engineering Science and Technology, an International Journal 19 (2016) 2035–2042 2039

Table 1
RMO based coefficients for I order model and its comparison.

Units P (MW) Coefficients Estimated Factual Festimated by Error

by MRMO (GJ/h) MRMO (GJ/h)
MRMO RMO ABC [1] PSO [11] LES [11]
Unit 1 Coal 10 a01 45.200000 176.620000 150.800000 25.820000 25.902000 25.808000 14.715000 11.684000
20 a11 10.560000 256.400000 256.400000 0.000000 0.013000 0.012000 7.403000 10.236000
30 361.500000 362.000000 0.500000 0.515000 0.512000 4.202000 6.836000
40 467.600000 467.600000 0.000000 0.013000 0.012000 0.000000 2.436000
50 579.500000 573.200000 6.300000 6.557000 6.288000 10.002000 7.764000
Total error 32.620000 33.000000 32.632000 36.322000 38.956000

Unit 2 Oil 10 a02 47.600000 184.750000 157.900000 26.850000 26.822000 26.788000 13.049000 12.280000
20 a12 11.030000 268.200000 268.200000 0.000000 0.078000 0.072000 9.200000 10.580000
30 377.700000 378.500000 0.800000 0.895000 0.882000 5.400000 7.390000
40 488.800000 488.800000 0.000000 0.094000 0.092000 0.000000 2.600000
50 606.000000 599.100000 6.899990 6.810000 6.798000 11.501000 8.290000
Total error 34.549990 34.699000 34.632000 39.151000 41.140000

Unit 3 Gas 10 a03 48.400000 187.200000 160.600000 26.600000 26.953000 26.591000 13.398000 12.200000
20 a13 11.220000 272.800000 272.800000 0.000000 0.022000 0.019000 8.801000 10.500000
30 384.300000 385.000000 0.700000 0.775000 0.729000 5.101000 7.300000
40 497.200000 497.200000 0.000000 0.042000 0.039000 0.000000 2.700000
50 616.500000 609.400000 7.100000 7.142000 7.051000 11.501000 8.300000
Total error 34.400000 34.934000 34.429000 38.801000 41.000000

Table 2
RMO based coefficients for II order model and its comparison.

Units P Coefficients Estimated by Factual Festimated by Error

(MW) MRMO (GJ/h) MRMO (GJ/h)
MRMO RMO TLBO [14] ABC [1] PSO [11] LES [11]
Unit 1 Coal 10 a01 96.599840 176.620000 176.619900 0.000110 0.008000 0.002000 0.001000 0.262000 2.368000
20 a11 7.588005 256.400000 264.919900 8.519940 8.529000 8.522000 8.513000 8.365000 5.568000
30 a21 0.041400 361.500000 361.500000 0.000010 0.009000 0.003000 0.013000 0.000000 2.496000
40 461.600000 466.360000 1.239960 1.240000 1.237000 1.259000 1.038000 2.240000
50 579.500000 579.500100 0.000090 0.008000 0.002000 0.025000 0.452000 1.536000
Total error 9.760110 9.794000 9.767000 9.810000 10.117000 14.208000

Unit 2 Oil 10 a02 101.531250 184.750000 184.750000 0.000050 0.002300 0.029000 0.015000 1.150000 2.442000
20 a12 7.880000 268.200000 276.806100 8.606050 8.621000 8.617000 8.574000 7.200000 5.662000
30 a22 0.044187 377.700000 377.699500 0.000450 0.003100 0.003000 0.047000 1.300000 2.442000
40 488.800000 487.430400 1.369550 1.395000 1.384000 1.428000 2.200000 2.266000
50 606.000000 605.998700 0.001250 0.026500 0.025000 0.069000 0.000000 1.690000
Total error 9.977350 10.047900 10.058000 10.133000 11.850000 14.464000

Unit 3 Gas 10 a03 101.812500 187.200000 187.200000 0.000000 0.000100 0.000000 0.599000 1.420000 2.376000
20 a13 8.100000 272.800000 281.362500 8.562500 8.572600 8.560000 8.560000 6.321000 5.568000
30 a23 0.043875 384.300000 384.300000 0.000000 0.000400 0.000000 0.001000 2.278000 2.568000
40 497.200000 496.012500 1.187500 1.181200 1.179000 1.178000 2.716000 2.284000
50 616.500000 616.500000 0.000000 0.011200 0.019000 0.023000 0.007000 1.420000
Total error 9.750000 9.765500 9.759000 10.361000 12.742000 14.216000

with other methods such as TLBO, PSO, ABC and LES [1,16,22] and available methods in the literature. The convergence is shown in
are presented in Table 3. The error for 20 MW is more than 4 for all Fig. 4. From the figure it is inferred that the proposed technique
the fuels, but the error is found to be zero for other power outputs converges to a better solution within the first 30 iterations. It is
for all the fuels using MRMO. Hence, the total error gets reduced by also found that the error is less when compared to first and second
MRMO. The total errors are 4.8533, 4.825005 and 4.916667 for order model employed in case-I and II. Hence, the third model is
coal, oil and gas respectively using the MRMO method. The total suggested for evaluating input output parameters of thermal units.
errors are found to be less when compared to RMO and other The results show that the total error is comparatively small for
reported methods viz. TLBO, PSO, ABC and LES. The total error for MRMO when compared to RMO and other methods like TLBO,
MRMO is lowered by 0.30768, 0.398554, 0.324533 for coal, oil, ABC, PSO, LES.
gas respectively when compared to RMO. It shows that the modi-
fication in the RMO algorithm improves the accuracy of the fuel 4.4. Case study – IV
cost by decreasing the error. In similar ranges, the total error val-
ues are gets reduced for coal, oil, gas respectively using TLBO and The data from 13 units are collected from the literature [22].
ABC. But when compared to other methods the total error is found The third order equation applied to the data and the coefficients
to be, more than 4 (coal), 0.4 (oil) and 0.8 (gas) using PSO and more are calculated by MRMO technique. The valve point loading is
than 6 using LES for all the three units. It shows that the MRMO included in the estimation. The results obtained from the simula-
estimated values are much closer to the actual values than the tion studies are given in Table 4. The estimated fuel cost by teach-
2040 M. Vanithasri et al. / Engineering Science and Technology, an International Journal 19 (2016) 2035–2042

Table 3
RMO based coefficients for III order model and its comparison.

Units P Coefficients Estimated Factual Festimated by Error

(MW) by MRMO (GJ/h) MRMO (GJ/h)
MRMO RMO TLBO [14] ABC [1] PSO [11] LES [11]
1 Coal 10 a01 127.066665 176.620000 176.620000 0.000000 0.012500 0.014499 0.004800 0.186000 0.393000
20 a11 3.118667 256.400000 256.400000 0.000000 0.279680 0.842196 0.734200 4.157000 1.874000
30 a21 0.199933 361.500000 356.646700 4.853300 4.862200 4.244310 4.406800 0.451000 1.779000
40 a31 0.001627 461.600000 467.600000 0.000000 0.006100 0.074983 0.107800 3.846000 8.368000
50 579.500000 579.500000 0.000000 0.000500 7.35E05 0.168800 0.000000 2.915000
Total error 4.853300 5.160980 5.176061 5.422400 8.641000 10.329000

2 Oil 10 a02 132.499980 184.750000 184.750000 0.000000 0.013400 0.01623 0.010900 0.674000 0.449000
20 a12 3.332500 268.200000 268.200000 0.000010 0.342550 0.97343 0.963100 0.000000 1.362000
30 a22 0.205875 377.700000 372.875000 4.824994 4.861520 4.18122 4.192900 4.690000 3.477000
40 a32 0.001663 488.800000 488.800000 0.000001 0.005460 0.00226 0.028900 0.063000 0.716000
50 606.000000 606.000000 0.000000 0.000620 0.02388 0.044900 0.119000 5.055000
Total error 4.825005 5.223550 5.19702 5.240700 5.547000 11.059000

3 Gas 10 a03 132.333331 187.200000 187.200000 0.000000 0.015250 0.014059 0.016700 0.099000 0.396000
20 a13 3.625000 272.800000 272.800000 0.000001 0.302450 0.741380 1.832300 1.526000 1.888000
30 a23 0.202417 384.300000 379.383300 4.916666 4.920620 4.431299 3.738700 3.300000 1.848000
40 a33 0.001625 497.200000 497.200000 0.000000 0.002360 0.003978 0.029700 0.874000 3.296000
50 616.500000 616.500000 0.000000 0.000520 0.023343 0.159300 0.000000 2.720000
Total error 4.916667 5.241200 5.214059 5.776700 5.799000 10.148000

Table 4
RMO based coefficients for III order model with valve point loading.

Units P Coefficients Estimated Factual Festimated by Error Total Error

(MW) by MRMO (GJ/h) MRMO (GJ/h)
MRMO RMO TLBO [14] MRMO RMO TLBO [14]
1 0 550.000000 550.000000 0.000001 0.632545 0.645271 0.020445 1.458405 1.517337
100 a01 549.999999 1468.080000 1468.078700 0.001269 0.401265 0.393575
200 a11 8.100993 2378.620000 2378.627300 0.007281 0.185471 0.198763
300 a21 2.7988E04 3270.220000 3270.217900 0.002137 0.035214 0.039493
400 a31 4.0255 E8 4134.610000 4134.606500 0.003475 0.048756 0.050103
500 e1 299.650350 4967.810000 4967.812100 0.002134 0.056458 0.061449
600 f1 0.035001 5770.650000 5770.653300 0.003312 0.002154 0.004579
680 6491.900000 6491.900800 0.000838 0.096542 0.124104
2,3 0 309.000000 309.000000 0.000001 0.022361 0.208770 0.022328 0.591484 0.739317
50 a01 309.000000 888.050000 888.049600 0.000393 0.009658 0.012453
100 a11 8.100080 1299.000000 1298.995800 0.004185 0.099874 0.017767
150 a21 5.5973E4 1540.240000 1540.242500 0.002478 0.154242 0.181202
200 a31 8.1151E8 2122.970000 2122.968500 0.001488 0.050365 0.051457
250 e1 199.993356 2546.200000 2546.204400 0.004417 0.091112 0.092017
300 f1 0.042000 2798.310000 2798.315200 0.005163 0.083657 0.089686
360 3412.290000 3412.285200 0.004203 0.080215 0.085965
4,5 60 716.090000 716.090100 0.000075 0.154870 0.168557 0.009872 0.417817 0.462576
6,7 80 a01 239.984477 1022.810000 1022.810100 0.000065 0.081247 0.087316
8,9 100 a11 7.740589 1133.860000 1133.862700 0.002730 0.082541 0.084475
120 a21 0.003234 1305.040000 1305.036000 0.003995 0.022348 0.025655
140 a31 1.2897E7 1529.430000 1529.430600 0.000574 0.006915 0.006868
160 e1 150.001349 1364.320000 1564.319500 0.000514 0.009965 0.025963
180 f1 0.0630000 1882.390000 1882.390200 0.000200 0.059875 0.063674
200 2003.780000 2003.781700 0.001719 0.000056 6.74E05
10,11 40 474.560000 474.559900 0.000084 0.112540 0.192437 0.014200 0.862112 1.080832
50 a01 126.003276 637.590000 637.594900 0.004868 0.051245 0.050658
60 a11 8.599972 751.680000 751.679700 0.000269 0.032658 0.03058
70 a21 0.002839 800.230000 800.228700 0.001277 0.245982 0.266942
80 a31 2.4E7 853.960000 853.961400 0.001367 0.006985 0.007169
90 e1 100.001584 1010.330000 1010.330600 0.000574 0.000999 0.001784
100 f1 0.084000 1109.310000 1109.313400 0.003353 0.254720 0.374866
120 1241.600000 1241.602400 0.002446 0.156983 0.156396
12,13 55 607.630000 607.629700 0.000328 0.000865 0.005216 0.022363 0.783193 0.799582
65 a01 126.008147 771.530000 771.529900 0.000112 0.213251 0.200681
75 a11 8.599884 886.480000 886.477900 0.002106 0.196584 0.203910
85 a21 0.002839 935.890000 935.887900 0.002054 0.039826 0.042292
95 a31 2.4E7 990.500000 990.501000 0.001048 0.168754 0.171664
105 e1 100.03961 1147.730000 1147.739700 0.009704 0.032547 0.035750
115 f1 0.084002 1247.590000 1247.586900 0.003142 0.013254 0.013731
120 1272.620000 1272.616100 0.003868 0.118112 0.126337
 M. Vanithasri et al. / Engineering Science and Technology, an International Journal 19 (2016) 2035–2042 2041

Fig. 2. Convergence characteristics of MRMO for case study – I.

Fig. 4. Convergence characteristics of MRMO for case study – III.

iii) Case – III: the total error gets reduced from 0.3 to 0.8 by
MRMO when compared to RMO, TLBO and ABC. The error
is reduced by more than 4 when compared with PSO and
LES.
iv) Case – IV: Total error in MRMO is reduced by 0.407917–
1.4381 when compared to RMO

The comparison shows the superiority of MRMO over other

algorithms for estimating fuel cost coefficients of economic dis-
patch problem in a power system. Further the proposed MRMO
converges at earlier iteration. Hence MRMO based optimization is
a promising technique for solving complicated optimization prob-
lems in power systems.

Fig. 3. Convergence characteristics of MRMO for case study – II. References

[1] Y. Sonmez, Estimation of fuel cost curve parameters for thermal power plants
using the ABC algorithm, Turk. J. Electr. Eng. Comput. Sci. 21 (2013) 1827–
ing learning based optimization (TLBO) is given n Table 4 and are 1841.
[2] F.J. Taylor, C.H. Huang, Recursive estimation of incremental cost curves,
compared with the RMO and MRMO estimated values. The total
Comput. Electr. Eng. 4 (1977) 297–307.
error obtained in MRMO is reduced from 0.407917 to 1.4381 when [3] M.E. El-Hawary, S.Y. Mansour, Performance evaluation of parameter
compared to RMO. It shows that the modification in the RMO algo- estimation algorithms for economic operation of power systems, IEEE Trans.
rithm improves the accuracy of the result by decreasing the error. Power Ap. Syst. PAS-101 (1982) 574–582.
[4] R.R. Shoults, M.M. Mead, Optimal estimation of piece-wise linear incremental
The total error is found to be less (0.0099–0.224) when compared cost curves for EDC, IEEE Trans. Power Ap. Syst. PAS-103 (1984) 1432–1438.
to TLBO method. From the errors it is inferred that the calculated [5] H.Y.K. Chen, C.E. Postel, On-line parameter identification of input-output
fuel cost values by MRMO method is much closer to the actual val- curves for thermal units, IEEE Trans. Power Ap. Syst. PWRS-1 (1986) 221–224.
[6] S.A. Soliman, S.E.A. Emam, G.S. Christensen, Optimization of the optimal
ues when compared to RMO and TLBO. The convergence is shown coefficients of non-monotonically increasing incremental cost curves, Electr.
in Fig. 4. From the figure it is inferred that the proposed technique Power Syst. Res. 21 (1991) 99–106.
converged to a better solution within the first 50 iterations. [7] Z.X. Liang, J.D. Glover, Improved cost functions for economic dispatch
computations, IEEE Trans. Power Syst. 6 (1991) 821–829.
[8] S.A Soliman, A.M. Al-Kandari, Kalman filtering algorithm for on-line parameter
5. Conclusions identification of input-output curves for thermal units, in: 8th Mediterranean
Electrotechnical Conference, 1996, pp. 1588–1593.
[9] I.M. Ferreira, F.P. Maciel Barbosa, A square root filter algorithm for dynamic
In this work, a new stochastic optimization technique called, state estimation of electric power systems, in: 7th Mediterranean
modified RMO has been developed to estimate the parameters of Electrotechnical Conference, vol. 3, 1994, pp. 877–880.
thermal power plant fuel cost function. The main features of [10] Belkacem Mahdad, Kamel Srairi, Solving practical economic dispatch problems
using improved artificial bee colony method, Int. J. Intell. Syst. Appl. 7 (2014)
MRMO are the need for less memory storage size, few control 36–43.
parameters and the ability to find solutions in an efficient manner. [11] M. El-Shibini, Z.H. Osman, A novel technique to estimate the fuel cost
The performance of the MRMO is tested for four test cases and functions for economic operation of power systems, Int. J. Electr. Power 11
(1989) 109–114.
compared with RMO and other methods in recent literatures. The [12] H.T. Yang, P.C. Yang, C.L. Huang, Evolutionary programming based economic
results can be summarized as: dispatch for units with non-smooth fuel cost functions, IEEE Trans. Power Syst.
11 (1) (1996) 112–118.
[13] A.M. Al-Kandari, K.M. El-Naggar, A genetic-based algorithm for optimal
i) The total error gets reduced to 0.5 and 0.08 by MRMO when estimation of input–output curve parameters of thermal power plants,
compared to RMO and ABC respectively. The total error is Electr. Eng. 89 (2006) 585–590.
reduced by more than 4 and 6 when compared with PSO [14] J.B. Park, K.S. Lee, J.R. Shin, K.Y. Lee, A particle swarm optimization for
economic dispatch with non smooth cost functions, IEEE Trans. Power Syst. 20
and LES respectively.
(2005) 34–42.
ii) Case – II: the total error obtained by MRMO method is less [15] K.M. El-Naggar, M.R. AlRashidi, A.K. Al-Othman, Estimating the input-output
when compared to RMO and TLBO and more than 2 for parameters of thermal power plants using PSO, Energy Convers. Manage. 50
PSO and LES. The total error is reduced by more than 0.6 (2009) 1767–1772.
[16] M.R. Alrashidi, K.M. El-Naggar, A.K. Al-Othman, Particle swarm optimization
when compared to ABC. based approach for estimating the fuel-cost function parameters of thermal
2042 M. Vanithasri et al. / Engineering Science and Technology, an International Journal 19 (2016) 2035–2042

power plants with valve loading effects, Electr. Power Compon. Syst. 37 (2009) [21] M.R. AlRashidi, K.M. El-Naggar, M.F. AlHajri, Convex and non-convex heat
1219–1230. curve parameters estimation using Cuckoo search, Arab. J. Sci. Eng. 40 (2015)
[17] M. Zivic Djurovic, A. Milacic, M. Krsulja, A simplified model of quadratic cost 873–882.
function for thermal Generators, in: Proceedings of the 23rd International [22] S. Durai, S. Subramanian, S. Ganesan, Improved parameters for economic
DAAAM Symposium, vol. 23, 2012, pp. 25–28. dispatch problems by teaching learning optimization, Electr. Power Energy
[18] Hardiansyah, Solving economic dispatch problem with valve-point Syst. 67 (2015) 11–24.
effect using a modified ABC algorithm, Int. J. Electr. Comput. Eng. 3 (2013) [23] Ziane Ismail, Benhamida Farid, Solving the generation scheduling with cubic
377–385. fuel cost function using simulated annealing, Int. J. Energy Inf. Commun. 7
[19] M. Narayanan, S. Siva, D. Velmurugan, A direct search method to solve (2016) 1–8.
economic dispatch problem with valve-point effect, IOSR J. Electr. Electron. [24] A.Y. Abdelaziz, E.S. Ali, S.M. Abd Elazim, Flower pollination algorithm to solve
Eng. 9 (2014) 36–42. combined economic and emission dispatch problems, Eng. Sci. Technol. Int. J.
[20] M.R. AlRashidi, K.M. El-Naggar, M.F. AlHajri, Estimation of fuel cost function 19 (2016) 980–990.
characteristics using Cuckoo search, in: International Conference on Computer [25] Rasoul Rahmani, Rubiyah Yusof, A new simple, fast and efficient algorithm for
Science, Data Mining & Mechanical Engg., Bangkok, Thailand, April 20–21, global optimization over continuous search-space problems: radial movement
2015. optimization, Appl. Math. Comput. 248 (2014) 287–300.