chapter 7 stochastic economic emission dispatch-modeled ...shodhganga.inflibnet.ac.in › bitstream...
TRANSCRIPT
90
CHAPTER 7
STOCHASTIC ECONOMIC EMISSION
DISPATCH-MODELED USING WEIGHTING METHOD
7.1 INTRODUCTION
Nearly 70% of electric power produced in the world is by means of
thermal plants. Thermal power stations are the major causes of atmospheric
pollution because of the high concentration of pollutants they cause. The main
draw back of generating electricity from fossil fuel releases several
contaminants such as Sulphur Oxides, Nitrogen Oxides, and Carbon-di-Oxide
into the atmosphere and generate particulates. The contaminants cause
atmospheric pollution. The pollution minimization has attracted a lot of
attention due to the public demand for clean air.
In recent years rigid environmental regulation forced the utility
planners to consider emission control as an important objective. In this work
Economic Emission Dispatch (EED) is considered as a stochastic problem.
The cost and emission coefficients are considered as random variables and a
stochastic model is developed. Due to consideration of the problem as
stochastic there are three objectives to be minimized, cost, emission, variance
of power from its expected value. A multi-objective problem is formulated
considering all the three objectives.
91
7.2 PROBLEM FORMULATION
The first objective function to be minimized is the total operating
cost for thermal generating units in the system. A quadratic operating cost
curve is assumed.
N
21 i i i i i
i 1
F a p b p c
(7.1)
where F1 is the cost function to be minimized
ai,bi,ci are the cost coefficients of the ith generator
N is the total number of generators
Pi is the power output of ith generator.
7.2.1 Stochastic Model
A Stochastic model of function F1 is formulated by considering cost
coefficients and load demand as random variables. By taking expectation the
stochastic model can be converted into its deterministic equivalent. The
random variables are assumed to be normally distributed and statistically
dependent on each other. As the random variables are statically dependent
both variance and covariance of generated power exists. The expected value
of operating cost is obtained by expanding the operating cost function using
Taylor’s series, about mean. The expected cost is:
N
21 i i i i i
i 1
E(F ) E a P b P c
(7.2)
N
21 i i i i i
i 1
F [(E(a P ) E(b P ) E(c )]
(7.3)
N
21 i i i i i
i 1
F [(E(a )E(P ) E(b )E(P ) E(c )]
(7.4)
92
N 2
1 i i ii i ii i i i ii 1
F (a [P var(P )] 2P cov(a ,P )) b (P cov(b ,P )) c
(7.5)
where i i ia ,b ,c are the expected cost coefficient of the ith generator
iP is the expected value of power generated by ith generator
1F is expected cost function to be minimized
ivar(P ) is variance of power Pi
i icov(a ,P ) is the covariance of random variable ai and Pi
i icov(b ,P ) is the covariance of random variable bi and Pi.
The expected operating cost function is represented as:
N 2
1 ii i i i i i i i, ii ii 1
F a p b p c a var(P ) 2P cov(a ,p ) cov(b p ) (7.6)
By substituting for variance and covariance by its coefficient of variation and
correlation coefficient the equation (7.6) can be rewritten as:
i i i i i i i i i
N 221 i ii i iP a P a p b P b P
i 1
F (1 C 2R C C )a P (1 R C C )b P c
(7.7)
where a b Pi i iC ,C ,C is coefficient of variation of random variables ai, bi, Pi
a Pi iR is correlation coefficient of random variable ai and Pi
b Pi iR is correlation coefficient of random variable bi and Pi.
7.2.2 Expected NOx Emission
The amount of NOx emission is given as a function of generator
power output Pi which is quadratic.
93
N
22 i i i i i
i 1
F d p e p f
(7.8)
where F2 is the emission function to be minimized
di, ei, fi are the emission coefficients of the ith generator
N is the total number of generators
Pi is the generator power output of ith generator.
A stochastic model of function F2 is formulated by considering
emission coefficients and load demand as random variables.
. N 2
2 i i i i ii 1
E(F ) E d P e P f
(7.9)
N 2
2 i i i i ii 1
E(F ) E(d P ) E(e P ) E(f ) (7.10)
N 2
2 i ii i i i ii 1
F [d P d var(P ) 2P cov(d ,P )
i ii i ie P cov(e ,P ) f ] (7.11)
where ii id ,e ,f are the expected emission coefficient of the ith generator
iP is the expected value of power generated by ith generator
2F is emission function to be minimized
ivar(P ) is variance of power Pi
i icov(d ,P ) is the covariance of random variable di and Pi
i icov(e ,P ) is the covariance of random variable ei and Pi .
94
i i i i i i i i i
N 222 i i ii iP d P d P e P e P
i 1
F (1 C 2R C C )d P (1 R C C )e P f
(7.12)
where d e Pi i iC ,C ,C is coefficient of variation of random variables di, ei, Pi
d Pi iR is correlation coefficient of random variable di and Pi
e Pi iR is correlation coefficient of random variable ei and Pi.
Covariance of bivariate random variables is considered positive or
negative. Covariance is represented by correlation coefficient; it is varied
from –1.0 to 1.0. One pair of random variable is considered at a time while
the rest of random variables are considered independent of each other
(uncorrelated).
7.2.3 Expected power deviation
Generator outputs Pi are treated as random variables. There is
variation in random variable from their expected value which results in
surplus or deficit power. Expected deviations are proportional to the
expectation of the square of the unsatisfied load demand.
2N N
3 D Li ii 1 i 1
F var P E P (P P )
(7.13)
where DP is the expected power demand
LP is expected transmission loss.
Substituting power balance equation (7.24) in (7.13) we get
N N
23 ii
i 1 i 1
F E[ P P ]
(7.14)
95
2N
3 iii 1
F E (P P )
(7.15)
N N N
23 i i ji i j
i 1 i 1 j 1i j
F E (P P ) 2 (P P )(P P )
(7.16)
As random variables are assumed as normally distributed and
statistically dependent expected deviations are given by equation (7.18)
N N N
3 i i ji 1 i 1 j 1
i j
F var(P ) 2cov(P ,P )
(7.17)
i i j i j
N N N223 i i jP P P P P
i 1 i 1 j 1j i
F C P R C C P P
(7.18)
where P Pi jR is correlation coefficient of random variable Pi and Pj
7.2.4 Expected Transmission Loss
The transmission power loss expressed through the simplified well
known loss formula expression as a quadratic function of the power
generation are given by equation (7.19) as:
N N
L i ij ji 1 j 1
P PB P
(7.19)
where PL is the transmission loss
Bij is the loss coefficient.
96
Taking expected value of transmission loss
N N
L i ij ji 1 j 1
E[P ] E PB P
(7.20)
N N N
2L i ii i ij j
i 1 i 1 j 1j i
P E P B E PB P
(7.21)
The expected transmission loss is represented as equation (7.22)
N N N N N N2
i ii ijL ii i i ij j i ji 1 i 1 i 1 j 1 i 1 j 1
j i j i
P P B B var P P B P B cov(P ,P )
(7.22)
where LP is the expected transmission loss
ijB is the expected loss coefficient.
Substituting for variance and covariance equation (7.22) can be
written as equation (7.23).
i P i ji j
N N N22L ii i i ij jP p P P
i 1 i 1 j 1j i
P (1 C )B P (1 R C C )P B P
(7.23)
7.2.5 Equality and Inequality Constraints
Real power balance is the equality constraint to be satisfied is given
by equation (7.24).
N
i D Li 1
P P P
(7.24)
97
Expected power generation limits are the inequality constraints to be satisfied.
min maxi i iP P P (7.25)
where miniP and max
iP are expected lower and upper limits of ith generator
power output.
The multi-objective problem is formulated as follows:
i i i i i i i i i
i i i i i i i i i
i i j i j
N 221 i ii i iP a P a P b P b P
i 1N 22
2 i i ii iP d P d P e P e Pi 1N N N22
3 i i jP P P P Pi 1 i 1 j i
j i
Minimize
F (1 C 2R C C )a P (1 R C C )b P c
F (1 C 2R C C )d P (1 R C C )e P f
F C P R C C P P
N
i D Li 1min maxi i i
3
k kk 1
Subject to
P P P
P P P (i 1.......N)
w 1 w 0
To generate the non-inferior solution of multi objective optimization problem,
PSO is used. wK is the levels of the weighting coefficients. The values of
weighting coefficients vary from 0 to 1 for each objective. The weight
w1, w2, w3 are varied in the range 0 to 1 in such a way that their sum is 1.0.
This approach yields meaningful result to the decision maker when solved
many times for different values of wk, (k = 1,2,3). This provides K number of
non-inferior solution which are pareto optimal non dominating. To identify
one optimal solution out of K solution fuzzy membership satisfaction index kD is used.
(7.26)
98
7.3 PSO ALGORITHM FOR ECONOMIC EMISSION
DISPATCH
The following steps are involved in solving the deterministic and
stochastic model of economic and emission dispatch problem.
Step 1: Read total number of thermal units, cost coefficients,
emission coefficients, B coefficients, coefficient of variation of each plant,
correlation coefficients of random variables, maximum number of iterations,
population size, acceleration constants c1 and c2, inertia weight wmin and wmax
Step 2: Confine the search space. Specify the lower and upper limits
of each decision variable.
Step 3: Initialize the individual of population. The velocity and
position of each particle should be initialized with in the feasible decision
variable space Xi=[X1, X2, X3……XN].
Step 4: Feed or generate the weight (wk=1,2…n) where n is the
number of objectives. Here n=3. Sum of w1, w2, w3 must be equal to one.
Step 5: For each individual Xi of the population the transmission
loss PLi is calculated using B-coefficients.
Step 6: Evaluate the fitness of each individual Xi in terms of pareto
dominance.
Step 7: Record the non dominated solutions found sofar and save
them in archive.
Step 8: Initialize the memory of each individual where the
personnel best position ( t )best idp is stored.
99
Step 9: Find the best particle out of the population and store its
position as (t )best dg .
Step 10: Set iteration count t=1
Step 11: Update the velocity of each particle Xi using the
equation (4.1).
Step 12: Update the position of each particle Xi using the
equation (4.2).
Step 13: Check whether the new particles are with in the feasible
region. If any element violates inequality constraints then the position of the
individual is fixed to its minimum/maximum limits according to
equation (4.5).
Step 14: Check the power balance constraints. Any violations are
penalized by adding penalty.
Step 15: Calculate the fitness function of the new individual.
Step 16: Update the archive which stores the non dominated
solution.
Step 17: Update memory of each particle using the equation (4.6).
Compare new particles fitness (t 1)idX with particles ( t )
best idp . If the current value
is better then the previous value then set (t 1)best idp to the new value and its
location equal to current location in d dimensional space. Compare new
particle’s fitness with the population’s overall best (t )best dg particle fitness. If the
fitness value of the new particle is better, reset (t 1)best dg to current particle array
index and value.
100
Step 18: Iteration = Iteration +1
Step 19: The algorithm repeats Step 11 to Step 18 until a sufficient
good fitness or a maximum number of iterations/epochs are reached. Once
terminated, the algorithm outputs the points of (t)best dg and f( (t)
best dg ) as its
solution.
7.4 TEST SYSTEM AND RESULTS
The validity of the proposed method is illustrated on a six generator
sample system Dhillon et. al (1993) and the result obtained from the proposed
algorithm is compared with that of Newton-Raphson iterative method. For
deterministic case coefficient of variation and correlation coefficient is zero.
The coefficient of variation and correlation coefficient assumed are:
Cai = Cbi = Cdi = Cei = Cpi = 0.1 (i = 1...6).
Two cases are considered:
Case 1: With dependent variables: In this case all the random
variables are considered dependent on each other
a P b P c P d P P Pji i i i i i i i iR R R R R 1.0 (i=1,2….6,j=1,2,….6)
Case 2: With independent variables: In this case all the random
variables are considered independent of each other
a P b P c P d P P Pji i i i i i i i iR R R R R 0.0 (i =1,2…6, j = 1,2…6).
The best solution obtained for expected demand of 500 MW, 700
MW, 900 MW for both the cases is shown in Table 7.4.
101
Table 7.1 gives the fuel cost coefficient and the capacity limits of
the thermal generating units. Table 7.2 gives the emission coefficients for the
thermal generating units. Table 7.3 gives the loss coefficients of the generator.
Table 7.1 Expected fuel cost coefficients and capacity limits
Generator No. ia ib ic maxiP (MW)
miniP (MW)
1 0.15274 38.53973 756.79886 10 125
2 0.10587 46.15916 451.32513 10 150
3 0.02803 40.39655 1049.99770 35 225
4 0.03546 38.30553 1243.53110 35 210
5 0.02111 36.32728 1658.56960 130 325
6 0.01799 38.27041 1356.65920 125 315
Table 7.2 Expected NOx emission coefficients
Generator No. id ie if
1 0.00419 0.32767 13.85932
2 0.00419 0.32767 13.85932
3 0.00683 0.54551 40.2669
4 0.00683 0.54551 40.2669
5 0.00461 0.51116 42.89553
6 0.00461 0.51116 42.89553
102
Table 7.3 Expected loss coefficients
0.002022 -0.000286 -0.000534 -0.000565 -0.000454 0.000103
-0.000286 0.003243 0.000016 -0.000307 -0.000422 -0.000147
-0.000533 0.000016 0.002085 0.000831 0.000023 -0.000270
-0.000565 -0.000307 0.000831 0.001129 0.000113 -0.000295
-0.000454 -0.000422 0.000023 0.000113 0.000460 -0.000153
0.000103 -0.000147 -0.000270 -0.000295 -0.000153 0.000898
Table 7.4 Comparison of best optimal solution
Load MW
PSO Newton Raphson Cost
(Rs/hr) NOx
(kg/hr) Risk
(MW)2 Cost
(Rs/hr) NOx
(kg/hr) Risk
(MW)2
Case 1
500 28525.49 304.10 2688.05 28550.15 312.51 2674.56
700 38951.57 506.6 5420.01 39070.74 528.44 5401.18
900 50634.68 798.38 9225.77 50807.24 864.06 9110.65
Case 2
500 28430.29 286.68 566.01 28476.63 287.48 558.75
700 38910.66 492.80 1117.33 39010.74 493.97 1114.28
900 50620.7 798.11 1862.71 50854.86 800.62 1861.07
103
Table 7.5 Expected optimal generation schedule using PSO
Load (MW)
1P (MW)
2P (MW)
3P (MW)
4P (MW)
5P (MW)
6P (MW)
LP (MW)
Case 1 500.0 57.34 37.99 41.03 74.67 179.2 128.31 18.5 700.0 88.72 58.51 66.55 110.00 240.28 172.39 36.48 900.0 124.67 92.89 82.07 153.85 286.94 220.09 60.8
Case 2 500.0 63.82 44.63 35.00 86.43 158.42 129.40 17.6 700.0 88.72 58.51 66.55 110.0 240.28 172.39 36.48 900.0 123.51 89.49 88.49 151.09 288.76 220.65 63.08
Table 7.6 Expected optimal generation schedules using NR method
Load (MW)
1P (MW)
2P (MW)
3P (MW)
4P (MW)
5P (MW)
6P (MW)
LP (MW)
Case 1 500.0 59.87 39.65 35.00 72.39 185.24 125.00 17.16 700.0 85.92 60.96 53.90 107.12 250.50 176.50 34.92 900.0 122.00 86.52 59.94 140.95 325.00 220.06 54.49
Case 2 500.0 59.67 41.41 51.87 83.26 157.83 126.75 20.79 700.0 89.09 65.47 69.71 116.07 223.59 175.70 39.66 900.0 124.61 92.96 87.09 152.74 286.84 220.77 65.03
Table 7.4 gives the comparison of best optimal results obtained
using PSO and NR method. Table 7.5 gives the expected optimal generation
schedule using PSO. Table 7.6 gives the expected optimal generation
schedule obtained using NR method.
104
Table 7.7 gives the deterministic results obtained using PSO. The
corresponding generation schedule is given in Table 7.8.
Table 7.7 Deterministic results using PSO
Load (MW) Cost (Rs/hr) NOx (kg/hr) Transmission loss (MW)
500 28289.99 286.14 17.67
700 38744.2 491.94 36.1
900 50510.3 790.2 64.0
Table 7.8 Deterministic generation schedule using PSO
Load (MW)
P1 (MW)
P2 (MW)
P3 (MW)
P4 (MW)
P5 (MW)
P6 (MW)
PL (MW)
500.0 61.68 38.79 39.11 77.41 172.65 127.67 17.67
700.0 87.2 61.23 68.98 108.0 240.22 171.42 36.1 900.0 121.1 89.19 89.68 151.87 289.09 222.41 64.0
Table 7.9 shows the Minimum, maximum values of each objective
for Case1. Table 7.10 shows the weight vector for the best optimal solution in
PSO and the corresponding kD . min
1F is obtained by giving full weightage to 1F
and neglecting other objectives. This is minimum cost dispatch. min2F is
obtained by giving full weightage to 2F . This is minimum emission dispatch. min3F is obtained by giving full weigtage to 3F . This is minimum risk dispatch.
105
Table 7.9 Minimum and maximum values of objective function (Case 1)
Load (MW)
min1F
(Rs/hr)
max1F
(Rs/hr)
min2F
(kg/hr)
max2F
(kg/hr)
min3F
(MW)2
max3F
(MW)2
500 28172.25 29428.39 267.5 354.2 2576.2 2881.46
700 38775.48 39762.64 469.3 556.7 5317.6 5623.78
900 50470.65 51434.49 766.3 854.3 9122.7 9430.77
Table 7.10 Values of weight and kDµ (Case 1)
Load (MW) w1,w2,w3 kDµ
500 0.5,0.3,0.2 0.01621
700 0.4,0.3,0.3 0.01546
900 0.4,0.4,0.2 0.01557
By taking the weight w1,w2,w3 as 0.4,0.3,0.3 the percentage relative
deviation in 1F from their deterministic value with respect to
a P b P P Pi i i i i jR ,R ,R (i j) is calculated and shown in Figure 7.1. The percentage
deviation in cost increases as RPiPj is varied from positive to negative values.
There is a decrease in percentage deviation of cost when RaiPi, RbiPi is varied
from positive to negative value.
With the same weight percentage relative deviation in 2F from their
deterministic value with respect to d P e P P Pi i i i i jR ,R ,R (i j) is calculated and
shown in Figure 7.2. The percentage deviation in emission increases as RPiPj is
varied from positive to negative values. There is a decrease in percentage
deviation of emission when RdiPi, ReiPi is varied from positive to negative
value.
106
-0.5
-0.3
-0.1
0.1
0.3
0.5
0.7
0.9
1.1
1.3
-1 0 0 0.5 1
Correlation coefficient
Perc
enta
ge v
aria
tion
in e
xpec
ted
cost
variations in Rpipj
variation in Rbipi
variation in Raipi
Figure 7.1 Percentage variations in expected cost with respect to
correlation coefficients
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
-1 0 0 0.5 1Correlation coefficients
Perc
enta
ge v
aria
tion
emiss
ion
variation in Rpipj
variation in Rdipi
variation in Reipi
Figure 7.2 Percentage variations in expected emission with respect to
correlation coefficients
107
Figure 7.3 shows the variation of the cost with iteration. Figure 7.4
shows the variation of emission with iteration. The trade of curve between
cost and emission is shown in Figure 7.5.
With various combination of learning factors the convergence
characteristic of the swarm is studied. It is found that the learning factor of
c1 = 2, c2 = 2 gives good convergence. Figure 7.6 shows the convergence for
different sets of learning factor. Figure 7.7 shows the convergence
characteristic for different population size. The population size of 50 gives
good convergence.
Figure 7.3 Fuel cost characteristic of economic emission dispatch using
PSO
104
Cos
t
Iteration
108
Figure 7.4 Emission characteristic of economic emission dispatch using
PSO
Figure 7.5 Trade off curve between cost and emission using PSO
Emis
sion
Iteration
Emis
sion
Cost
109
Figure 7.6 Convergence characteristic for different learning factor
using PSO
Figure 7.7 Convergence characteristic of various population size for
economic emission dispatch
104
Iteration
Iteration
Fitn
ess
Fitn
ess
104
110
7.5 CONCLUSION
PSO is capable of reducing the cost and emission, but there is
increase in risk. Compared to the saving in cost and reduction in the emission
level the increase in risk is tolerable. So the proposed algorithm is efficient in
finding the best optimal solution. For the loads considered in this problem
stochastic cost on an average is 0.3% higher and emission is 0.865% higher
compared to deterministic results. Table 7.11 gives the average % variation of
stochastic results obtained using PSO with NR method. The execution time is
21 seconds in PIV 3GHz system.
Table 7.11 Average results for all the three loads using PSO
Details Case 1 Case 2
Cost reduction (Rs/h) 0.30% 0.25%
NOx reduction (kg/h) 4.71% 0.177%
Risk increase (MW2) 0.59% 0.606%