Characteristics of the Generating Units

B T G J/h

Input
Fuel Electric Power

(Input) (Output) MW
Pmin Pmax
Optimization Techniques and  Thermal generating units
Output
Economic Dispatch  Consider the running costs only
 Input / Output curve
 Fuel vs. electric power
 Fuel consumption measured by its energy content
 Upper and lower limit on output of the generating unit
3

Economic Dispatch Problem Cost Curve

 Multiply fuel input by fuel cost
 No-load cost
L  Cost of keeping the unit running if it could produce zero MW
A B C

$/h
 Several generating units serving the load
 What share of the load should each generating unit
produce?

Cost
 Consider the limits of the generating units
 Ignore the limits of the network
No-load cost
MW
Pmin Pmax
2 Output 4

1 2
Incremental Cost Curve Objective
 Incremental cost curve Cost [$/h]  Most engineering activities have an objective:
 Achieve the best possible design
 Achieve the most economical operating conditions
 FuelCost vs Power
 Power  This objective is usually quantifiable
∆F
 Examples:
 Derivative of the cost curve ∆P
MW  minimize cost of building a transformer
 In $/MWh  minimize cost of supplying power
 Cost of the next MWh Incremental Cost  minimize losses in a power system
[$/MWh]  maximize profit from a bidding strategy

MW
5 7

Mathematical formulation Decision Variables

 Objective function  The value of the objective is a function of some
C  C A (PA )  C B (PB )  CC (PC ) decision variables:
 Constraints A B C
L
F  f  x1 , x2 , x3 , .. xn 
 Load / Generation balance:
 Examples of decision variables:
 Unit Constraints:  Dimensions of the transformer
L  PA  PB  PC  Output of generating units, position of taps
 Parameters of bids for selling electrical energy
PAmin  PA  PAmax
PBmin  PB  PBmax
PCmin  PC  PCmax
This is an optimization problem

6 8

3 4
Optimization Problem Minimization and Maximization
 What value should the decision variables take so f(x*)
f(x)
that
F  f  x1 , x2 , x3 , .. xn 
is minimum or maximum?
x* x

-f(x)
-f(x*)

If x = x* maximizes f(x) then it minimizes - f(x)

9 11

Example: function of one variable Minimization and Maximization

 maximizing f(x) is thus the same thing as minimizing
f(x*) f(x)
g(x) = -f(x)

 Minimization and maximization problems are thus

interchangeable

x* x  Depending on the problem, the optimum is either a

maximum or a minimum

f(x) is maximum for x = x*

10 12

5 6
Necessary Condition for Optimality Example
f(x)
f(x*) f(x)

df df
0 0
dx dx

x* x
x
If x  x maximises f ( x ) then:
*

df df
f ( x )  f ( x * ) for x  x *   0 for x  x * For what values of x is 0 ?
dx dx
df In other words, for what values of x is the necessary
f ( x )  f ( x * ) for x  x *   0 for x  x *
13 dx condition
15
for optimality satisfied?

Necessary Condition for Optimality Example

f(x) df
0 f(x)
dx

x* x
A B C D x

df  A, B, C, D are stationary points

If x  x * maximises f ( x ) then  0 for x  x *  A and D are maxima
dx
 B is a minimum
 C is an inflexion point
14 16

7 8
How can we distinguish minima and maxima? How can we distinguish minima and maxima?

f(x) f(x)

A B C D x A B C D x

d2 f d2 f
For x  A and x  D, we have: 2
0 For x  C , we have: 0
dx dx 2
The objective function is concave around a maximum The objective function is flat around an inflexion point
17 19

How can we distinguish minima and maxima? Necessary and Sufficient Conditions of Optimality
 Necessary condition:
df
f(x) 0
 Sufficient condition: dx
 For a maximum:
d2 f
0
 For a minimum: dx 2
d2 f
0
A B C D x dx 2

d2 f
For x  B we have: 0
dx 2
The objective function is convex around a minimum
18 20

9 10
Isn’t all this obvious? Example

 Can’t we tell all this by looking at the objective

function?
 Yes, for a simple, one-dimensional case when we know
the shape of the objective function
 For complex, multi-dimensional cases (i.e. with many
decision variables) we can’t visualize the shape of the
objective function
 We must then rely on mathematical techniques

21 23

Example Feasible Set

 The values that the decision variables can take are usually
limited
 Examples:
 Physical dimensions of a transformer must be positive
 Active power output of a generator may be limited to a certain
range (e.g. 200 MW to 500 MW)
 Reactive power output of a generator may be limited to a
certain range (e.g. -100 MVAr to 150 MVAr)

22 24

11 12
Feasible Set Two-Dimensional Case
f(x1,x2)

f(x)

xMIN A D xMAX x
x1*

x1
Feasible Set x2*
The values of the objective function outside f(x1,x2) is minimum for x1*, x2*
the feasible set do not matter
25 x2 27

Interior and Boundary Solutions Necessary Conditions for Optimality

f(x)
f(x1,x2)

f ( x 1 ,x 2 )
0
x 1
x * ,x *
1 2
f ( x 1 ,x 2 )
0
x 2
x * ,x *
1 2
x1*
xMIN A B D E xMAX x
 A and D are interior maxima x1
 B and E are interior minima x2*
 XMIN is a boundary minimum Do not satisfy the
 XMAX is a boundary maximum Optimality conditions!
26 x2 28

13 14
 Multi-Dimensional Case Sufficient Conditions for Optimality
f(x1,x2)
At a maximum or minimum value of f  x1 , x2 , x3 , .. xn 
Saddle point
we must have: f
0
x 1
f
0
x 2

f
0
x
 n x1

A point where these conditions are satisfied is called a stationary point

29 x2 31

Sufficient Conditions for Optimality Sufficient Conditions for Optimality

f(x1,x2) minimum maximum
Calculate the Hessian matrix at the stationary point:

  2 f 2 f  2 f 
 x 2 
x 1 x 2 x 1 x n 
 1

  2 f 2 f  2 f 
  
 x 2 x 1 x 22 x 2 x n 
     
  2 f 2 f  2 f 
  
x1 x x x n x 2 
 n 1 x n2 

x2 30 32

15 16
 Sufficient Conditions for Optimality Contours
 Calculate the eigenvalues of the Hessian matrix at the A contour is the locus of all the point that give the same value
to the objective function
stationary point
 If all the eigenvalues are greater or equal to zero:
x2
 The matrix is positive semi-definite
 The stationary point is a minimum
 If all the eigenvalues are less or equal to zero:
 The matrix is negative semi-definite
 The stationary point is a maximum
 If some or the eigenvalues are positive and other are negative:
 The stationary point is a saddle point

x1
33 35 Minimum or maximum

Contours Example 1
f(x1,x2)
Minimise C  x 12  4 x 22  2 x 1 x 2

F2
Necessary conditions for optimality:
F1 C
 2 x1  2 x 2  0
x 1 x 1  0 is a stationary
 point
C x 2  0
 2 x 1  8 x 2  0
x 2

x1

F1 F2

x2 34 36

17 18
Example 1 Example 2
Sufficient conditions for optimality: Minimize C  x12  3x22  2x1 x2

  2 C  2 C  Necessary conditions for optimality:

 x 2 x 1 x 2   2 2 
Hessian Matrix:  2
1
   C
  2x1  2x2  0
  C  2 C  2 8  x1 x 1  0 is a stationary
   point
x 2 x 1 x 2 
2
C
 2x1  6x2  0 x 2  0
x2
must be positive definite (i.e. all eigenvalues must be positive)

2 2
 0   2  10   12  0
2 8
10  52 The stationary point
 = 0 is a minimum
37
2 39

Example 1 Example 2
Sufficient conditions for optimality:

x2   2 C  2 C 
 x 2 x 1 x 2  2 2 
Hessian Matrix:  2   
1

  C  2 C   2 6 
C=9  
C=4 x 2 x 1 x 2 
2

C=1
2 2
 0  2  4   8  0
x1
2 6
4  80
 = 0 The stationary point
2
Minimum: C=0 is a saddle point
4 80
or  = 0
2
38 40

19 20
Example 2 Number of Constraints
x2
 N decision variables
C=0

C=9
 M equality constraints
 If M > N, the problems is over-constrained
C=4
C=1
 There is usually no solution

C=-9 C=-4 C=-1 C=-4 C=-9

 If M = N, the problem is determined
 There may be a solution
x1
C=1  If M < N, the problem is under-constrained
C=4  There is usually room for optimization
C=9

This stationary point is a saddle point C=0

41 43

Optimization with Equality Constraints Example 1

 There are usually restrictions on the values that the Minimise f  x 1 , x 2   0.25 x 12  x 22
decision variables can take x2
Subject to   x 1 , x 2   5  x 1  x 2  0

Minimise
  x1 , x 2   5  x1  x 2  0
f  x1 , x2 ,.. , xn  Objective function

subject to:
 1  x1 , x2 ,.. , xn   0 Minimum
 Equality constraints

 m  x1 , x2 ,.. , xn   0
x1
42 44 f  x 1 , x 2   0.25 x 12  x 22

21 22
Example 2: Economic Dispatch Solution by substitution

x1 x2
 Difficult
L
G1 G2  Usually impossible when constraints are non-
linear
C 1  a1  b1 x 12 Cost of running unit 1
 Provides little or no insight into solution
C 2  a2  b2 x 2
2 Cost of running unit 2

C  C 1  C 2  a1  a 2  b1 x 12  b 2 x 22 Total cost  Solution using Lagrange multipliers

Optimization problem:

Minimise C  a 1  a 2  b1 x 12  b 2 x 22

45
Subject to: x 1  x 2  L 47

Solution by substitution Gradient

Minimise C  a 1  a 2  b1 x 12  b 2 x 22
Consider a function f (x1 , x2 ,.. , xn )
Subject to: x 1  x 2  L
 f 
 x 2  L  x1  x1 
 
 C  a1  a 2  b1 x 12  b 2  L  x 1 
2
 f 
Unconstrained minimization The gradient of f is the vector f   x2 

dC   
 2 b1 x 1  2 b 2  L  x 1   0 
dx 1 f 
 
 xn 
b2 L  b1 L  
 x1   x 2  
b1  b 2  b1  b 2 
d 2C
46  2b1  2 b 2  0  minimum 48
dx 12

23 24
Properties of the Gradient Example 4
 Each component of the gradient vector f ( x , y )  ax  by f  f3
indicates the rate of change of the function in f 
that direction x  a  f  f2 y
f       f
f
 The gradient indicates the direction in which a   b 
y 
function of several variables increases most
rapidly f  f1 f
 The magnitude and direction of the gradient f
usually depend on the point considered
 At each point, the gradient is perpendicular to
the contour of the function x

49 51

Example 3 Lagrange multipliers

f ( x , y )  ax  by
2 2
Minimise f  x 1 , x 2   0.25 x 12  x 22 subject to   x 1 , x 2   5  x 1  x 2  0

f 
x  2 ax    x1 ,x 2   5  x1  x 2
f      
f
  2 by  y
y 
f  0.25 x 12  x 22  6
B f  0.25 x 12  x 22  5

C
A

D
50 52

25 26
 Lagrange multipliers Lagrange multipliers
The solution must be on the constraint
 f  To reduce the value of f, we must move
 x  in a direction opposite to the gradient
f   1 
 f    x1 , x 2 
f   f
x 2 
f  x1 , x 2   6 f  x1 , x 2   6
A
f  x1 ,x 2   5 f f  x1 ,x 2   5
?

f f

53 55

Lagrange multipliers Lagrange multipliers

• We stop when the gradient of the function
   is perpendicular to the constraint because
moving further would increase the value
 x  of the function
   1 
  x1 , x 2       x1 , x 2 
 
x 2  f
f  x1 , x 2   6 f  x1 , x 2   6
A
f  x1 ,x 2   5 f  x1 ,x 2   5 f

 
C
f
 
At the optimum, the gradient of the
function is parallel to the gradient B
of the constraint
54
 56 

27 28
Lagrange multipliers Example
At the optimum, we must have: f  Minimise f  x 1 , x 2   0.25 x 12  x 22 subject to   x 1 , x 2   5  x 1  x 2  0
Which can be expressed as: f     0
f    x 1 , x 2 ,    0.25 x 12  x 22    5  x 1  x 2 
In terms of the co-ordinates:
 0
x 1 x 1
f    x 1 , x 2 ,  
 0  0.5 x 1    0
x 2 x 2 x 1
The constraint must also be satisfied:   x1 , x 2   0   x 1 , x 2 ,  
 2x2   0
x 2
 is called the Lagrange multiplier
  x 1 , x 2 ,  
 5  x1  x 2  0

57 59

Lagrangian function Example

To simplify the writing of the conditions for optimality,
it is useful to define the Lagrangian function:
  x 1 , x 2 ,  
 0.5 x 1    0  x1  2
  x 1 , x 2 ,    f  x 1 , x 2     x 1 , x 2  x 1
  x 1 , x 2 ,   1
The necessary conditions for optimality are then given  2x2   0  x2  
by the partial derivatives of the Lagrangian: x 2 2

  x 1 , x 2 ,   f    x 1 , x 2 ,   1
  0  5  x1  x 2  0  5  2    0
x 1 x 1 x 1   2

  x 1 , x 2 ,   f  2
  0
x 2 x 2 x 2
 x1  4
  x 1 , x 2 ,  
58    x1 ,x 2   0 60
 x2 1
 

29 30
Example Application to Economic Dispatch
Minimise f  x 1 , x 2   0.25 x 12  x 22
x1 x2
minimise f  x 1 , x 2   C 1  x 1   C 2  x 2 
Subject to   x 1 , x 2   5  x 1  x 2  0
x2 L
G1 G2
s.t .   x 1 , x 2   L  x 1  x 2  0

  x1 , x 2   5  x1  x 2  0
  x 1 , x 2 ,   C1  x 1   C 2  x 2     L  x 1  x 2 
 dC 1
  0
x 1 dx 1 dC 1 dC 2
f  x1 , x 2   5 Minimum
 
1  dC 2 dx 1 dx 2
  0
x 2 dx 2

x1  Equal incremental cost

4  L  x1  x 2  0 solution
61 63 

Important Note! Equal incremental cost solution

If the constraint is of the form: ax 1  bx 2  L
Cost curves:
C1 ( x 1 ) C2 ( x 2 )
It must be included in the Lagrangian as follows:

= f  x1 ,.. , xn     L  ax1  bx2  x1 x2

And not as follows: dC 1 dC 2

Incremental
cost curves: dx 1 dx 2
= f  x1 ,.. , xn     ax1  bx2 

62 64
x1 x2

31 32
Interpretation of this solution Physical interpretation
dC 1
: Cost of one more MW from unit 1
dC 1 dC 2 dx 1
dx 1 dx 2 dC 2
: Cost of one more MW from unit 2
 dx 2
dC 1 dC 2
Suppose that 
x1 dx 1 dx 2
x1* x2* x2
dC 1
- Decrease output of unit 1 by 1MW  decrease in cost =
- dx 1
L
+
dC 2
Increase output of unit 2 by 1MW  increase in cost =
If < 0, reduce λ dx 2
L  x1*  x2* If > 0, increase λ
dC 2 dC 1
65 Net change in cost =
67  0
dx 2 dx 1

Physical interpretation Physical interpretation

dC C
 lim
C( x ) dx x0 x It pays to increase the output of unit 2 and decrease the
output of unit 1 until we have:

C For x sufficiently small:

x dC dC 1 dC 2
C   x  
dx dx 1 dx 2
x If x  1 MW :
dC The Lagrange multiplier λ is thus the cost of one more MW
dC(x) C  at the optimal solution.
dx dx
This is a very important result with many applications in
The incremental cost is the cost of
economics.
one additional MW for one hour.
This cost depends on the output of
the generator.
66 x 68

33 34
Generalization
Minimise
f  x1 , x2 ,.. , xn 
subject to:
 1  x1 , x2 ,.. , xn   0

 m  x1 , x2 ,.. , xn   0
Lagrangian:

= f  x1 ,.. , xn   1 1  x1 ,.. , xn    m m  x1 ,.. , xn 

• One Lagrange multiplier for each constraint

• n + m variables: x1, …, xn and λ1, …, λm

69

Optimality conditions
= f  x1 ,.. , xn   1 1  x1 ,.. , xn    m m  x1 ,.. , xn 
 f  1  m
  1    m 0
x 1 x 1 x 1 x 1
 n equations

 f  1  m
  1    m 0
x n x n x n x n

  1  x 1 ,, x n   0
 1
m equations


  m  x 1 ,, x n   0 n + m equations in
 m
70

n + m variables

35