EE 722: Restructured Power Systems

Economic Dispatch And Pricing

Anupama Kowli

Indian Institute of Technology Bombay, India

Outline of Today’s Lecture

Recap of constrained optimization

Economic dispatch examples

Implications for pricing electricity

2 / 16
Constrained Optimization

Consider general problem:

min f (x) such that

x
gi (x) = 0 i = 1, . . . , I ↔ λi
hj (x) ≤ 0 j = 1, . . . , J ↔ µj

Associated Lagrangian function:

X X
L(x, λ, µ) = f (x) + λi gi (x) + µj hj (x)
i j

KKT conditions:
∂L ∂L ∂L
=0 =0 ≤0
∂x ∗ ∂λ ∗ ∂µ ∗
µ∗j hj (x∗ ) = 0 and µ∗j ≥ 0 ∀j

3 / 16
Equal Lambda Criterion
Consider a system having M generators: Ci (PGi ) represents the
cost of the ith generator for producing PGi units of power

If the generators are expected to supply a demand of PL units, the

system operator can decide generator loading by solving a cost
minimization problem

Ignoring capacity constraints

P
min i Ci (PGi )
{PGi }
X
s.t. PGi = PL
i

Formulating the Lagrangian and solving the optimality conditions

indicates that all generators are loaded upto the point of equal
marginal costs

Note that this problem ignores capacity limits

4 / 16
Proof Of Equal Lambda Criterion
X X
Problem: min Ci (PGi ) s.t. PGi = PL
{PGi }
i i
!
X X
Lagrangian: L = Ci (PGi ) + λ PL − PGi
i i

Optimality conditions give rise to following conditions:

∂Ci (PGi∗ )
∇PGi L = 0 : = λ∗
∂PGi
X
∗
∇λ L = 0 : PGi = PL
i
That is, the incremental costs of all units are equal to λ∗
∗ )
∂Ci (PGi
Iterative solution: guess some λ, find PGi such that = λ,
∂PGi
check for mismatch in total generation and demand, update λ if
necessary and repeat
5 / 16
Example: Three Generators (Wood & Wollenberg)
2
Cost structure: Ci (PGi ) = αi PGi + βi PGi + γi

Equal-lambda criterion: 2αPGi ∗ + β = λ∗

Alternative: s7olution can be found by solving N + 1 equations

(including supply = demand)

Numerical example: demand PL = 850 and costs are as shown

Unit αi βi γi
G1 (coal) 0.00156 7.92 561
G2 (oil) 0.00194 7.85 310
G3 (oil) 0.00482 7.97 78

Solution: λ∗ = 9.148 and

∗
PG1 = 393.2 MW
∗
PG2 = 334.6 MW
∗
PG3 = 122.2 MW
6 / 16
Modified Lambda Criterion

The costs Ci (PGi ) are usually assumed to be monotonically

non-decreasing function; however, these are usually determined
min ≤ P max
over an interval PGi Gi ≤ PGi

Modified criterion for dispatch is as follows:

∂Ci (PGi∗ )
= λ∗ min
PGi max
< PGi < PGi
∂PGi
∂Ci (PGi∗ )
≤ λ∗ max
PGi = PGi
∂PGi
∂Ci (PGi∗ )
≥ λ∗ min
PGi = PGi
∂PGi

Can this criterion be derived? Use KKT conditions – Think of λ∗

as price!

7 / 16
Dispatch With Generator Limits

Consider system having M generators with each generator i’s

min ≤ P max
output subject to capacity limits: PGi Gi ≤ PGi

Ci (PGi ) represents the cost of the ith generator for producing PGi
units of power (assume Ci (·) monotonically non-decreasing
function)

If the total demand in the system is PL units, the generator loading

can be determined by solving a cost minimization problem
P
min i Ci (PGi )
{PGi }
X
s.t. PGi = PL
i
min ≤ P max
PGi Gi ≤ PGi ∀i

8 / 16
KKT Conditions For Constrained Dispatch
Using dual variables λ, µ− +
i and µi for the equality constraint, lower
bounds and upper bounds on generation, the associated Lagrangian
function is !
− +

X X
L PGi , λ, µi , µi = Ci (PGi ) + λ PL − PGi
i i
X X
µ− min
µ+ max


+ i PGi − PGi + i (PGi − PGi )
i i

∗ )
∂Ci (PGi
∇PGi L = 0 : = λ∗ + µ−∗ +∗
i − µi
∂PGi
X
∗
∇λ L = 0 : PGi = PL
i
∗ min
∇µ− L ≤ 0 : PGi ≥ PGi
i
∗ max
∇µ+ L ≤ 0 : PGi ≤ PGi
i

µ−∗ +∗ −∗ min ∗
= 0 , µ+∗ ∗ max


i , µi ≥ 0, µi PGi − PGi i (PGi − PGi ) = 0
9 / 16
Generator At Limits
∗ > P min , then complementarity slackness
Please note that if PGi Gi
−∗ ∗ < P max =⇒ µ+∗ = 0
forces µi = 0; likewise, PGi Gi i

At the optimum, marginal costs satisfy the following condition:

∗ )
∂Ci (PGi
= λ∗ + µ−∗i − µi
+∗
∂PGi

If a generator is at its maximum limit, then µ−∗

i = 0 and
∗ )
∂Ci (PGi
≤ λ∗ ∵ µ+∗
i ≥0
∂PGi

If a generator is at its minimum limit, then µ+∗

i = 0 and
∗
∂Ci (PGi )
≥ λ∗ ∵ µ−∗
i ≥0
∂PGi
This is the basis of the modified Lambda criterion proposed in
previous lecture
10 / 16
Example: Three Generators with Limits

Quadratic costs (similar to previous example) and demand

PL = 850

Generation cost and other data are as shown

Unit αi βi min
γi PGi max
PGi
G1 (coal) 0.00128 6.48 459 150 600
G2 (oil) 0.00194 7.85 310 100 400
G3 (oil) 0.00482 7.97 78 50 200

Solution: λ∗ = 8.576, and

∗ = 600 MW, P ∗ = 187.1 MW, P ∗ = 62.9 MW
PG1 G2 G3

Remark: If we proceed with solution as before, we notice that G1’s

output exceeds 600 MW; solution is obtained by forcing G1 at its
upper limit and using equal lambda dispatch for G2 and G3 to meet
the remaining load (check out Wood and Wollenberg for details)
11 / 16
Lambda Iteration Algorithm

1 Pick some initial value of λ0 , set k = 0

2 k to
For the given λk , calculate the outputs of the generators PGi
satisfy the modified Lambda criterion
If the given λk forces certain generators at their limits, those
generators are removed from the λ-iteration
The equal λ-criterion is applied only to the generators that are
within their limits
k − P | ≤ ; if yes, STOP
P
3 Check if | PGi L

4 Else, λk+1 = λk + αk , and repeat step 2 with k = k + 1

The algorithm is useful in case the cost functions do not have

well-defined derivatives or the result in system of equations which are
not linear

12 / 16
Example: Two Generator Case

Setting: no congestion, inelastic demand, no generator limits and

quadratic costs
C1 (x1 ) = a1 + b1 (x1 )2 , C2 (x2 ) = a2 + b2 (x2 )2

If the load is L, then equal Lambda criterion dictates

dC1
λ= = 2b1 x1
x1
dC2
= = 2b2 x2
x2
b2
=⇒ 2b1 x1 = 2b2 (L − x1 ) ∴ x1 = L
b1 + b2
What is the price?
13 / 16
Example: Alternative Approach

14 / 16
Example: Dispatch with Linear Costs

Load cheapest unit first – merit order dispatch

15 / 16
References

https://towardsdatascience.com/
optimization-stories-kkt-conditions-f86aea4fb6c2

https://ocw.mit.edu/courses/mechanical-engineering/
2-854-introduction-to-manufacturing-systems-fall-2016/
lecture-notes/MIT2_854F16_KktExample.pdf

16 / 16