UNIT COMMITMENT

Ahmad Firyal Adila, S.ST., M.T.

Bab 5. UNIT
COMMITMENT
5.1 Introduction
5.1.1 Constraints in Unit Commitment
5.1.2 Spinning Reserve
5.1.3 Thermal Unit Constraints
5.1.4 Other Constraints
5.1.4.1 Hydro-Constraints
5.1.4.2 Must Run
5.1.4.3 Fuel Constraints

5.2 Unit Commitment Solution Methods

5.2.1 Priority-List Methods
5.2.2 Dynamic-Programming Solution
5.2.2.1 Introduction
5.2.2.2 Forward DP Approach
5.2.3.1 Adjusting L
5.2.3 Lagrange Relaxation Solution

Problems
Discussion
[2]
5.1
INTRODUCTION
5.1 INTRODUCTION
 5.1.1 Constraints in Unit Commitment
 5.1.2 Spinning Reserve
 5.1.3 Thermal Unit Constraints
 5.1.4 Other Constraints
 5.1.4.1 Hydro-Constraints
 5.1.4.2 Must Run
 5.1.4.3 Fuel Constraints

[4]
5.1 INTRODUCTION

Committing a generating unit

Un-equal distribution of industrial load
Problem of unit commitment in electrical power
systems

[5]
Example 5A

with fuel costs:

Fuelcost 1 = 1.1 P/MBtu
Fuel cost 2 = 1.0 P/MBtu
Fuel cost 3 = 1.2 v/MBtu
Example 5A
if supply a load of 550 MW, what unit or combination of units should
be used to supply this load most economically?
Example 5A
Example 5B
Example 5B
✓ Example 5A and 5B have only obeyed one simple
constraint: Enough units will be committed to supply
the load.
✓ There are other constraints and other phenomena must be
taken into account in order to claim an optimum solution.
✓ These constraints will be discussed in the next section,
followed by a description of some of the presently used
methods of solution.
 5.1.1 Constraints in Unit Commitment

✓ Many constraints can be placed on the unit

commitment problem.
✓ The list presented here is by no means exhaustive.
✓ Each individual power system, power pool, reliability
council, and so forth, may impose different rules on
the scheduling of units, depending on the generation
makeup, load-curve characteristics, and such.

[11]
 5.1.2 Spinning Reserve
1. Spinning reserve is the term used to describe the
total amount of generation available from all units
synchronized (i.e., spinning) on the system, minus the
present load and losses being supplied.
2. Spinning reserve must be carried so that the loss of one
or more units does not cause too far a drop in system
frequency
3. if one unit is lost, there must be ample reserve on
the other units to make up for the loss in a specified
time period.

[12]
 5.1.3 Thermal Unit Constraints
•Thermal Unit Constraint:
❖Minimum Up Time
❖Minimum down time
❖Crew constraint
❖start-up cost and shut-down cost

[13]
 5.1.4 Other Constraints
5.1.4.1 Hydro-Constraints

Unit commitment if separated from the scheduling of hydro-

units as a separate hydrothermal scheduling or coordination
problem may not result in an optimal solution

[14]
 5.1.4 Other Constraints
5.1.4.2 Must Run

Some units are given a must-run status during certain times of

the year for reason of voltage support on the transmission network
or for such purposes as supply of steam for uses outside the
steam plant itself.

[15]
 5.1.4 Other Constraints
5.1.4.3 Fuel Constraints

We will treat the “fuel scheduling” problem briefly in

Chapter 6. A system in which some units have limited
fuel, or else have constraints that require them to burn a
specified amount of fuel in a given time, presents a most
challenging unit commitment problem.

[16]
5.2
UNIT COMMITMENT
SOLUTION METHODS
 5.2 UNIT COMMITMENT SOLUTION
METHODE

 5.2.1 Priority-List Methods

 5.2.2 Dynamic-Programming Solution
 5.2.2.1 Introduction
 5.2.2.2 Forward DP Approach
 5.2.3.1 Adjusting L
 5.2.3 Lagrange Relaxation Solution

[18]
 UNIT COMMITMENT PROBLEM

It is not economical to run all the units available

all the time
Optimum allocation (commitment) of
generators (units) at each generating station at
various load levels
To determine the units of a plant that should
operate for a particular load– problem of UC
There should be least operating cost
This problem is important for thermal plants

[19]
The Unit Commitment Problem
 If a unit is on, we designate this with 1 and respectively, the off
unit is 0
 So, somehow we decide that for the next hour we will have "0
1 1 0 1" if we have five units
 Based on that, we solve the economic dispatch problem for unit
2, 3 and 5
 We start turning on U2, U3, U5
 When the next hour comes, we have them up and running
The Unit Commitment Problem
Tradeoffs

- A unit that is on line may be expensive and running at

minimum capacity
- Shutting the unit down may save money
BUT
- Will incur startup cost when needed again
- May not be possible to start by the time it is needed
The Unit Commitment Problem
Unit Characteristics Startup/Shutdown

Nuclear – shut down only for refueling

Hydro – “zero” resource cost
Large coal ( 250 MW+) Very long start time (Days)
Gas (< 200 MW or so) 8-24 Hours start time
Combustion turbines As low as 5 Min
The Unit Commitment Problem
Unit Characteristics Startup/Shutdown Costs

Fuel to get system to correct temperature/pressure

Crew time
Electricity for auxiliary system
…
The Unit Commitment Problem
 The question is, _how_ do we come up with this unit
commitment "0 1 1 0 1" ?
 One very simplistic way: if we have very few units, go over
all combinations from hour to hour
 For each combination at a given hour, solve the economic
dispatch
 For each hour, pick the combination giving the lowest cost!
 The Unit Commitment Problem
Lets postulate the following situation:
A loading pattern must be established for M periods
There are N units to commit
Any one unit or a combination of units can supply the loads.
The total number of combinations to try each hour is
C (N, 1) + C (N, 2) + …+ C (N, N-1) + C (N, N) = 2N–1
C (N, j) is the combination of N items taken j at a time.
Maximum number of possible combinations is (2N-1) M

[25]
The Unit Commitment Problem
Discrete Optimization Problem
3 units -- 23=8 possibilities each hour
24 hours -- 824=4.722E+21 possibility( many
unrealistic)
Large Scale Problem

Discrete Optimization
-- Convert to continuous
-- add constraints to force towards discrete
-- may find approx. answer; may miss answer
-- Develop a search technique
Priority-List Methods
Priority-List Methods
 Priority-List Methods

Sort based cost least

 Priority-List Methods

So, we got combination cheapest based load

demand
Dynamic Programming
 Backward DP Approach:
The solution starts at the last interval and proceeds back the initial point
Fcost(K, I) = Min [Pcost (K, I) + Scost(I, K: J,K+1) + Fcost(K+1,J)]
where
Fcost (K, I) = minimum total fuel cost
Pcost (K, I) = minimum generation cost
Scost (I, K: J, K+1) = incremental start-up cost.
{J} = set of feasible states in interval K+1.

[32]
[33]
 Forward DP Approach
The initial conditions are easily specified
Previous history of the unit can be computed at each stage
Fcost (K, I) = Min [Pcost (K, I) + Scost (K-1, L: K, I) +
Fcost (K-1, L)]
where
Fcost (K, I) =least total cost to arrive at state (K, I)
Pcost (K, I) = production cost for state (K, I).
Scost (K-1, L: K, I) = transition cost for state (K-1, L)
to state (K, I)
where state (K, I) is the Ith combination in hour K.

[34]
[35]
 EXAMPLE OF DP
The problem is to find out
the minimum cost from A to
N
At the terminal of each stage
there is a set of choices of
nodes {Xi} to be chosen
The symbol Va (Xi, Xi+1)
represents the cost of
traversing stage a (=1…V)

[36]
 fI(X1) : Minimum cost for the 1st stage
is obvious :
fI(B) : VI(A, B) = 5.
fI(C) : VI(A, C) = 2.
fI(D) : VI(A, D) = 3.
fII(E)= min [fI(X1) + VII(X1, E)]
{X1}
= min [5+11, 2+8, 3+ ] =10
X1 =B =C =D
fII(F) = min [, 6, 9] = 6, X1 = C
fII(G) = min [, 11, 9] = 9,X1 = D
[37]
 (X2) E F G
fII (X2) 10 6 9
Path X0X1 AC AC AD
Tracing back, the path of minimum cost is found as follows:
Stage {Xi} fi
1 B, C, D 5, 2, 3
2 E, F, G 10, 6, 9
3 H, I, J, K 13, 12, 11, 13
4 L, M 15, 18
5 N 19

[38]
Lagrange Relaxation
 Lagrange Relaxation (1)

 Min f = (0.25 x21+15)U1 + (0.255 x22+15)U2

 subject to:
 W = 5 – x1U1 - x2U2
 0 < x1 < 10
 0 < x2 < 10

 U may be only 0 or 1
 Lagrange Relaxation (2)

 L = (0.25 x21+15)U1 + (0.255 x22+15)U2 +

(5 – x1U1 - x2U2)

 Pick a value for  and keep it fixed

 Minimize for U1 and U2 separately
 0 = d/dx1(0.25x21 + 15 - x11)
 0 = d/dx2(0.255x22 + 15 - x21)
 Lagrange Relaxation (3)

 0 = d/dx1(0.25x21 + 15 - x11)
 if the value of x1 satisfying the above falls outside the 0 < x1 < 10,
we force x1 to the limit.
 If the term in the brackets is > 0, set U1 to 0, otherwise keep it 1
 0 = d/dx2(0.255x22 + 15 - x21)
 same as above
 Lagrange Relaxation (4)
 Now assume the variables x1, x2, U1, U2 fixed
 Try to maximize L by moving  around
 dL/d = (5 – x1U1 - x2U2)
  =  + dL/d ()
 if dL/d    = 
 if dL/d    = 
 After we found 2, repeat the whole process
starting at step 1
 Formulating the Lagrangean
 Rewrite the constrained optimization problem as an
unconstrained optimization problem !
 Then we can use the simple derivative (unconstrained
optimization) to solve
 The task is to interpret the results correctly

44
 Formulating the Lagrangean
 We are minimizing gradients of both multivariate equations
 CT & ΣPGi = PL
 For both equations to be at a minimum these gradients must be
linearly dependent vectors
 CT – λw = 0
 with w ≡ ΣPG – PL = 0
 The “Lagrangean multiplier”
 λ is defined to be the scaling variable that brings CT and w into
linear alignment

45
 Lagrangean Example
max g(x) = 5x12x2
s.t. h(x) = x1 + x2 = 6 or x1 + x2 – 6 = 0

Formulate L =
L = g(x) – λh(x)
Find ?
dL/dx1, dL/dx2, dL/dλ
x1 = 4, x2 = 2, λ = 80

46
Economic Dispatch & the Lagrangean
min CT = ΣCi(PGi)
s.t. Σ(PGi) = PL
PGi min <= PGi <= PGi max

Then L = ?

L = CT −  (PGi − PL )
47
EXAMPLE 5D
EXAMPLE 5D
 EXAMPLE 5D

Firsly, Lambda = 0
 EXAMPLE 5D

The result is to schedule unit 3 off during hours 1, 2, and 4 and

on during hour 3. Further, unit 3 is scheduled to be at its
maximum of 200 MW during hour 3.
EXAMPLE 5D
EXAMPLE 5D
 EXAMPLE 5D

After 10 iterations, q(A) = 19,485, J* = 20,017, and (J* -

q*)/q* = 0.027.
This latter value will not go to zero, nor will the solution settle
down to a final value; therefore, the algorithm must stop when (J* -
q*)/q* is sufficiently small (e.g., less than 0.05 in this case)