Beruflich Dokumente
Kultur Dokumente
For
POWER GENERATION OPERATION
AND CONTROL
Allen J. Wood
Bruce F. Wollenberg
Gerald B. Shebl
August 2013
Allen J. Wood
Bruce f. Wollenberg
Gerald B. Shebl
Problem 3.7
NOTE: For instruction on gradient
methods see instructions ahead
of problem 3.10
Problem 3.10 instructions
Problem 3.10 requires that the student know something about the Gradient and Newton methods
of solution. These were purposely eliminated from the third edition of the text so the next pages
are included here. If you wish to use problem 3.10 then use these pages from the second edition
to teach the Gradient and Newton methods.
Problem 3.10 instructions
Gradient Search
This method works on the principle that the minimum of a function, f( x ), can be found by a
series of steps that always take us in a downward direction. From any starting point, x0 , we may
find the direction of steepest descent by noting that the gradient of f, i.e.,
f
x
1
f=
f
xn
always points in the direction of maximum ascent. Therefore, if we want to move in the direction
of maximum descent, we negate the gradient. Then we should go from x0 to x1 using:
x1 x0 f
Where is a scalar to allow us to guarantee that the process converges. The best value of
must be determined by experiment.
To solve the economic dispatch problem which involves minimizing the objective function and
keeping the equality constraint, we must apply the gradient technique directly to the Lagrange
function itself.
Problem 3.10 instructions
d
P F1 ( P1 )
1 dP1
d
P F2 ( P2 )
= 2 dP2
d
P F3 ( P3 )
3 dP3
N
Pload i 1 Pi
The problem with this formulation is the lack of a guarantee that the new points generated each
step will lie on the surface . We shall see that this can be overcome by a simple variation of
the gradient method.
The economic dispatch algorithm requires a starting value and starting values for P1 ,
P2 , and P3 . The gradient for L is calculated as above and the new values of , P1 , P2 , and P3 ,
etc., are found from:
x1 x 0 (L)
EXAMPLE 1
Given the generator cost functions found in Example 3A in the text, solve for the economic
dispatch of generation with a total load of 800 MW.
Using 100 and starting from P10 300 MW , P20 200 MW , and P30 300 MW , we
set the initial value of equal to the average of the incremental costs of the generators at their
starting generation values. That is:
1 3 d
0 Fi ( Pi 0 )
3 i 1 dPi
This value is 9.4484.
The progress of the gradient search is shown in Table 3.2. The table shows that the
Problem 3.10 instructions
iterations have led to no solution at all. Attempts to use this formulation will result in difficulty
as the gradient cannot guarantee that the adjustment to the generators will result in a schedule
that meets the correct total load of 800 MW.
Economic Dispatch by Gradient Method
A simple variation of this technique is to realize that one of the generators is always a
dependent variable and remove it from the problem. In this case, we pick P3 and use the
following:
P3 800 P1 P2
Note that this function stands by itself as a function of two variables with no load-generation
balance constraint (and no ). The cost can be minimized by a gradient method and in this case
the gradient is:
d dF1 dF3
dP Cost dP dP
1
Cost 1 1
d dF2 dF3
dP Cost
dP d P
2 2 2
Note that this gradient goes to the zero vector when the incremental cost at generator 3 is equal
to that at generators 1 and 2. The gradient steps are performed in the same manner as
previously, where:
x1 x 0 Cost
and
P
x 1
P
2
Each time a gradient step is made, the generation at generator 3 is set to 800 minus the sum of
the generation at generators 1 and 2. This method is often called the reduced gradient because
of the smaller number of variables.
Problem 3.10 instructions
EXAMPLE 2
Reworking example 1 with the reduced gradient we obtain the results shown in table below. This
solution is much more stable and is converging on the optimum solution.
Reduced Gradient Results ( 10)
Iteration P1 P2 P3 Ptotal Cost
1 300 200 300 800 7938.0
2 320.04 222.36 257.59 800 7858.1
3 335.38 239.76 224.85 800 7810.4
4 347.08 253.33 199.58 800 7781.9
5 355.97 263.94 180.07 800 7764.9
10 380.00 304.43 115.56 800 7739.2
then
g1 g1 g1
x x2 x3
1
g ( x)
g 2
x
1
which is the familiar Jacobian matrix. The adjustment at each step is then:
x [ g ( x)]1 g( x)
Problem 3.10 instructions
and is as it was defined before. The Jacobian matrix now becomes one made up of second
derivatives and is called the Hessian matrix:
d2 d2
dx 2
1 dx1dx2
d2
x x dx2 d x1
d2
d dx1
Generally, Newtons method will solve for the correction that is much closer to the minimum
generation cost in one step than would the gradient method.
EXAMPLE 3
In this example we shall use Newtons method to solve the same economic dispatch as used in
Examples 3E and 3F.
The gradient is the same as in Example 3E, the Hessian matrix is:
d 2 F1
dP 2 0 0 1
1
d 2 F2
0 2
0 1
[H ] dP2
2
d F3
0 0 1
dP32
1 1 1 0
In this example, we shall simply set the initial equal to 0, and the initial generation values will
be the same as in Example 3E as well. The gradient of the Lagrange function is:
Problem 3.10 instructions
88572
86260
108620
0
The Hessian matrix is:
00031 0 0 1
0 00039 0 1
[H ]
0 0 00096 1
1 1 1 0
Solving for the correction to the x vector and making the correction, we obtain
P1 3696871
P 3156965
x 2
P3 1146164
90749
and a total generation cost of 7738.8. Note that no further steps are necessary as the Newtons
method has solved in one step. When the system of equations making up the generation cost
functions are quadratic, and no generation limits are reached, the Newtons method will solve in
one step.
We have introduced the gradient, reduced gradient and Newtons method here mainly as
a way to show the variations of solution of the generation economic dispatch problem. For many
applications, the lambda search technique is the preferred choice. However, in later chapters,
when we introduce the optimal power flow, the gradient and Newton formulations become
necessary.
Problem 3.11
continued
Problem 3.13 continued
Problem 3.13 a
1600
Unit 1
1400 Unit 2
Unit 3
1200
fuel rate [BTU/h]
1000
800
600
400
200
20 30 40 50 60 70 80 90 100
P [MW]
Problem 3.13 b
22
20
incremental heat rate [BTU/MWh]
18
16
14
Unit 1
12 Unit 2
Unit 3
10
6
20 30 40 50 60 70 80 90 100
P [MW]
Problem 3.13 c
5500
5000
4500
4000
fuel rate [BTU/h]
3500
3000
2500
2000
1500
1000
0 50 100 150 200 250 300
P [MW]
prob3_13
dynamic programming
demand: 100.000000 cost: 1663.596000 unit1: 40.00 unit2: 40.00 unit3: 20.00
demand: 140.000000 cost: 2306.379660 unit1: 60.00 unit2: 52.00 unit3: 28.00
demand: 180.000000 cost: 3101.291160 unit1: 76.00 unit2: 67.00 unit3: 37.00
demand: 220.000000 cost: 3990.412980 unit1: 82.00 unit2: 72.00 unit3: 66.00
demand: 260.000000 cost: 4975.289625 unit1: 97.00 unit2: 85.00 unit3: 78.00
lamda iteration
demand: 100.000000 cost: 1663.789824 unit1: 39.29 unit2: 40.71 unit3: 20.00
demand: 140.000000 cost: 2306.503107 unit1: 60.85 unit2: 51.78 unit3: 27.36
demand: 180.000000 cost: 3101.336569 unit1: 76.45 unit2: 66.80 unit3: 36.75
demand: 220.000000 cost: 3990.448568 unit1: 81.89 unit2: 71.70 unit3: 66.41
demand: 260.000000 cost: 4975.325797 unit1: 96.98 unit2: 85.29 unit3: 77.73
Problem 3.14
3.1
Problem 3.15
Problem 3.15, continued
Problem 3.15, continued
Problem 3.16
Problem
Problem 3.17 3.17, part a