You are on page 1of 23

Unit Commitment

Lectures prepared by

Prof. S. Shahnawaz Ahmed


For subsequent materials assistance taken mainly from :

[1]. John J. Grainger, William D. Stevenson, Jr. "Power System Analysis",


McGraw-Hill, 1994; Ch. 13.
What is UC and what are its relations with other control functions
(e.g. ED and AGC)?
Unit Commitment (UC)
Unit commitment is a systematic procedure (algorithm) used by the Load
Dispatch Centre to decide in advance which of the available generators
should be started up or shut down in such a manner that the total fuel
cost to meet the forecasted demand over a time period (usually 24 hours)
is minimized subject to various operating and security constraints.

Economic Dispatch (ED)


Economic dispatch function determines the best allocation of the power
to be generated among the already decided or committed (i.e. on-line)
units so that the cost to meet the actual demand only in a discrete
interval (an hour or less) is minimized within the constraints imposed.

AGC (Automatic Generation Control)


AGC is a closed-loop control system that implements the UC and ED
decisions (obtained from the Load Dispatch Centre server) on the
generation units under each control area and then does small
adjustments to those to maintain the power system frequency and power
exchange with a neighbouring area/system at the scheduled value.
Why is Unit Commitment needed? (Source: BPDB and PGCB
Example with respect to BPS websites/annual reports)
•Variable demand (load)
•Generation units have diverse
characteristics regarding
type of fuel and technology
fuel cost (zero for hydro, wind,
solar and high for thermal units),
input-output characteristics
minimum and maximum
output,
ramp rate (MW/minute)
minimum up and down time,
 start up and shut down costs,
start up time,
availability (e.g. intermittent
nature of renewable based units,
forced outages of thermal units,
scheduled maintenance of any
unit etc.),
emission characteristics (e.g.
zero emission solar, wind to high
emission coal fired units),
etc.
•A small percentage of saving effected
by a UC method in the fuel cost per
hour turns out to be substantial over
one year allowing new facilities
acquisition or reduction in electricity
tariff.
Typical summer day

Comparison of two typical days’ fuel cost for grid power in


Even just 1% saving means
about 72 lac taka per day
less on fuel

Generation Scheduling

Bangladesh
method used in BPS is not
rigorous mathematics
Typical winter day based rather based on
simply fuel ranking i.e.
gives priority to plants
respectively using gas, coal
and then oil unless voltage
support is needed
somewhere.

Source: PGCB Website


Typical input-output (IO)characteristics of thermal generation units (ST, GT, RE)

Convex, smooth and quadratic (parabolic) ST: Steam turbine; fuel: coal, gas, oil (HFO: Heavy fuel oil or furnace oil)
function
H= (a’/2)P2
+b’P + c’ Mbtu/h GT: Gas turbine; fuel: gas, oil (High Speed Diesel: HSD)
F = Hx(R $/Mbtu) RE: Reciprocating engine; fuel: gas, oil (HFO/ HSD)
=(a/2)P2 +bP + c $/h The a,b,c constants of IO curves depend upon fuel type and the unit’s design

Thermal efficiency η = 1 kWh/No. of kCal per kWh


=860/HR
when HR is expressed in kCal/kWh;
1 kWh =860 kCal Stepped for piecewise linear IO curve
1 kCal ≈ 4 Btu

Linear for convex IO curve

Source: A. J. Wood , B. F. Wollenberg and Gerald B. Sheble., “Power Generation, Operation and Control,” John Wiley, NY,
2014
Methods of Unit Commitment (mainly for thermal units)
•Merit order method
(simplest; units are loaded in ascending order of their full load average production cost
($/kWh) calculated from respective IO curves only at the rated full load but unable to
consider any constraints excepting capacity constraints of the units.)

•Dynamic programming method


(simpler requiring enumeration of the feasible combinations; considers most of the
constraints but dimensionality increases with the number of units)

•Lagrange relaxation method


(rigorous mathematics based, able to consider almost all the constraints, more accurate than
dynamic programming but may suffer from convergence problem)

•Evolutionary methods (such as genetic algorithm, particle swarm optimization, simulated


annealing etc.)
(these involve simpler computations but less accurate than Lagrange Relaxation method,
unable to consider transmission constraints and often lead to local minimum point instead of
a global minimum point).
The saving in fuel cost varies from method to method depending upon the consideration
of constraints and modeling complexities. However, in general more the constraints are
considered less will be the saving.
Dynamic Programming (DP) method
•For simplicity and ease of understanding the method will be illustrated using a
small system and disregarding transmission loss so that generators to be
committed should provide a total output equal to the demand in each interval .
•If K number of units then possible number of
UC is done using a forecasted load curve
combinations in each interval (stage) is is 2K -1
(all off i.e. 0 status is not acceptable).
• However, feasible number of combinations
are only a few because most of the
combinations may not provide a total output at
least equal to the demand in a stage due to
capacity limitations of the units.
•If other constraints are applied the number of
feasible combinations for enumeration will
reduce substantially.
•Table below shows the theoretical
(k) combinations out of 4 units (K=4) without any
mention of their capacities.
where,
Computation of Pi(k)involves economic
dispatch among the generators of xi
combination just as a potential combination in
stage k.
However, the ED will be done again using the
finally decided combination in each stage after
the UC has been completed for all the stages.

Transition means off to on status or vice versa between two successive stages
and hence it includes start up (may be fixed or time dependent)
or shut down cost (usually fixed).
Cc MBtu

Ct :Mbtu/hr to maintain in banking mode


i.e. at operating temperature

Start-up cost when banking = C, x t x F + C,

F: $/MBtu

Source: A. J. Wood , B. F. Wollenberg and Gerald B. Sheble,“Power Generation, Operation


and Control” ,John Wiley, NY, 2014
Backward DP: A dynamic-programming
algorithm can be run backward in time
starting from the final stage (hour) to be
studied and back to the initial stage as shown
by the following recursive equation. The
minimum cumulative cost for last N-k+1 stages
is then
Fi*(k) = min {Pi*(k) + Ti*j (k) + Fj (k+1)}
{xj(k+1)}
This equation needs to be applied for
searching each possible i-th combination in
conjunction with each possible j-th
combination for stage (k+1) so that F(k)
becomes minimum up to stage k starting from
final stage N.

Forward DP: Conversely, the algorithm can be


run forward in time from the initial hour to the
final hour using the following recursive
equation. The minimum cumulative cost for
first k stages is then Fi*(k) = min {Pi*(k) + Ti*j (k) + Fj (k+1)}
{xj(k+1)}
Fi*(k) = min {Pi*(k) + Ti*j (k-1) + Fj (k-1)}
{xj(k-1)} Backward DP is just for computations only but physical
operation of the units takes place in the forward direction.
Both approaches will give the same results if
the initial and the final stages have same
loads.
The forward approach has distinct
advantages#. For example, if the start-up cost
of a unit is a function of the time it has been
off-line (i.e. its temperature), then a forward
dynamic-program approach is more suitable
since the previous history of the unit can be
computed at each stage. There are other
practical reasons for going forward. The initial
conditions are easily specified and the
computations can go forward in time as long as
required.

On the other hand backward DP is suitable


when the initial (k=1) and final stage (k=N)
both have the same demand level. Because
then Tij(N)=0 and Fj(N)= Pj(N).

#Source: A. J. Wood , B. F. Wollenberg and Gerald B. Sheble, “Power


Generation, Operation and Control” ,John Wiley, NY, 2014
Backward sweep
involves calculations
starting from the final
to the 1st stage
Stage k

Forward sweep is simply


tracing the minimum
cumulative cost path

Summary of calculations in stage k of Backward DP


Numerical example on the UC for the system with the given forecasted load curve
Can be broken down into two parts (Ex 13.8, 13.9).

Neglect transmission loss.


Fi = (ai/2)Pi2 +biPi + ci $/h

In practice such a restricted combination


is seldom imposed for any stage. Here it
has been assumed only to illustrate the
basics of DP without increasing
computational effort.
Ex. 13.8 Note on ED
ED among units within a plant for a load PL = Pg
(neglecting loss) in a stage occurs when each unit is run at
an output Pgi (subject to maximum and minimum
capacity) so that their individual incremental fuel cost
becomes equal to each other and to a common value
termed system λ.

Details computations have been shown


for combination x3 as an example in
next slide.
$/MWh

In this way we can proceed for other


combinations for the load level 1100
MW and other load levels and can
have the results as in Table 13.6
Only the feasible combinations with units 1 and 2 must run status have
been considered as per the condition in Ex. 13.8
Solution to Ex. 13.9
Using Table 13.6 results and start up and shut down costs where needed in
transition from one combination to another between two stages,
the grid is constructed starting from the final stage 6 to represent backward
DP based UC.
#The final solution is
the least cumulative
cost path (bold line)
traced in the forward
sweep i.e. traversing
from the initial
condition node of
stage 1 to the
destination node of
the final stage 6.
Details of each stage calculations 105604=58236+1500+45868
follow in the next slides.
Computational effort reduction for F
calculation by DP:
3+9+6+6+3=27 interstage transitions
instead of 3x9x6x6x3 = 2916 transitions. 45868
Another dimension is that in absence of
infeasible combinations the number of 45868
interstage transitions would have been
15x152x152x152x15 =2.563x109
namely x9

You might also like