Optimal Power Flow (DC and AC OPF)

Optimal Power Flow (DC-OPF and AC-OPF)
DTU Summer School 2017
Spyros Chatzivasileiadis
What is optimal power flow?
2 DTU Electrical Engineering Optimal Power Flow (DC-OPF and AC-OPF) Jun 12, 2017
Optimal Power Flow (OPF)
• In its most realistic form, the OPF is a non-linear, non-convex problem,
which includes both binary and continuous variables.
• Goal: minimize an objective function, respecting all physical, operational,
and technical constraints, such as:
• Ohm’s law and Kirchhoff laws
• Operational limits of generators
• Loading limits of transmission lines
• Voltage levels
• and many, many others
• Disclaimer:
• Realistic OPF implementations include thousands of variables and
constraints
• Here we focus on the most “fundamental” formulations of OPF
• These can be extended with several additional constraints to
accurately model the problem and/or system at hand
Use of OPF in the industry
• RTE, France: they invented the OPF! (Carpentier,
1962)
• Their focus is mostly on the optimization of the
system operation and not so much on markets
• They are working on the application of convex
relaxations of OPF on their system (see Dan
Molzahn’s talk!)
• CAISO, California, USA

• Electricity markets: An OPF runs every day
(Day-Ahead), every hour, every 15 minutes, and
every 5 minutes.
• Depending on the problem they run a DC-OPF
with unit commitment, a standard DC-OPF, or
just Economic Dispatch
Use of OPF in the industry
• PJM, East Coast, USA

• Security-Constrained Economic Dispatch every 5 minutes
• final test phase of optimal voltage control every 5 minutes (they use
AC-OPF as well)
• EUPHEMIA
• one common algorithm to calculate electricity prices across Europe, and
allocate cross border capacity on a day-ahead basis
• 19 European countries, over 150 million EUR in matched trades daily
• PLEXOS
• June 2000: PLEXOS was first-to-market with electric power market
simulation based entirely on mathematical programming
• Features: generation capacity expansion planning, transmission expansion
planning, hydro-thermal coordination, ancillary services
Summarizing
• DC-OPF
• market clearing uses DC-OPF (at the moment)
• convex
• can solve fast; can be applied in very large problems
• AC-OPF
• primarily used for optimization of operation and control actions
• non-convex (in its original form)
• continuous efforts to decrease computation time and increase system
size
• Trends
• Incorporating uncertainty
• Decomposition (and other) methods to solve very large problems
• Guarantees for a global minimum (convex relaxations)
• Market design
• Coupled energy networks and markets, e.g. gas and electricity
Outline
• Economic Dispatch
• DC-OPF
• AC-OPF
Outline
• used in power exchanges, e.g. EPEX, etc.
• Supply must meet demand
• Generator limits
Outline
• DC-OPF
• extends Economic Dispatch
• considers the power flows! (in a linearized form)
• includes the power flow limits of the lines
• only active power; no losses
Outline
• DC-OPF
• extends Economic Dispatch
• considers the power flows! (in a linearized form)
• includes the power flow limits of the lines
• only active power; no losses
• AC-OPF
• full AC power flow equations
• active and reactive power flow
• current, voltage
• losses
Economic Dispatch
∑
min c i P Gi
i
subject to:
PGmin ≤ PGi ≤ PGmax

i
∑ i
PGi = PD
i
Economic Dispatch
∑
min c i P Gi
i
subject to:

i
∑ i
PGi = PD
i
• The Economic Dispatch does not consider any network flows or network
constraints!
Economic Dispatch
∑
min c i P Gi
i
subject to:

i
∑ i
PGi = PD
i
• The Economic Dispatch does not consider any network flows or network
constraints!
• We assume a copperplate network, i.e. a lossless and unrestricted flow of

electricity from A to B.
Can we solve the economic dispatch problem without using an
optimization solver?
Can we solve the economic dispatch problem without using an
optimization solver?
Yes! With the help of the merit order curve.
The Merit-Order Curve
price
cG4
max
A = PG1
cG3
max
B = A + PG2
max
C = B + PG3
cG2
max
D = C + PG4
cG1
0 A B C D power
price
cG4
max
A = PG1
cG3
max
B = A + PG2
max
C = B + PG3
cG2
max
D = C + PG4
cG1
0 A B C D power
PD
• cG3 is the system
marginal price
price
• Gens G1 and G2
cG4
are fully
cG3 dispatched
• Gen G4 is not
dispatched at all
cG2
• Gen G3 is partially
cG1 dispatched
• G3 is the
0 A B C D power
PD “marginal
generator”
The Merit-Order Curve: An Example
Merit-Order of the German conventional generation in 2008. Source:

Forschungsstelle für Energiewirtschaft e. V.
Merit-Order Curve, Marginal Generators,
and Line Congestion
• Although G3 has
price enough capacity,
it cannot produce
cG4 enough to cover
cG3
the demand due
to line congestion
• Instead G4, a
cG2 more expensive
gen that does not
cG1
contribute to the
line congestion,
0 A B C D power must produce the
PD
missing power
• In a DC-OPF context, there is no longer a single system marginal price

(we will observe different nodal prices Optimal
14 DTU Electrical Engineering
in different nodes)
Power Flow (DC-OPF and AC-OPF) Jun 12, 2017
DC-OPF
∑
min c i P Gi
i
subject to:
PGmin
i
≤ PGi ≤ PGmax
i
DC-OPF
∑
min c i P Gi
i
subject to:
PGmin
i
≤ PGi ≤ PGmax
i
B · θ = PG − PD
1
(θi − θj ) ≤ Pij,max
xij
DC-OPF
∑
min c i P Gi
i
subject to:
PGmin
i
≤ PGi ≤ PGmax
i
B · θ = PG − PD
1
(θi − θj ) ≤ Pij,max
xij
• The DC-OPF with the standard power flow equations contains both the
power generation PG and the voltage angles θ in the vector of the
optimization variables.
Exercise
cG1 = 60 $/MWh, cG2 = 120 $/MWh
Pload = 150 MW
1 2 max = 100 MW, P max = 200 MW

PG1 G2
X12 = 0.1 pu, X13 = 0.3 pu, X23 = 0.1

pu, BaseMVA = 100 MVA
max = 40 M W (line limit)
P13
3
1 What are the optimization variables? Form
the optimization vector
2 Formulate the objective function
3 Formulate the constraints
DC-OPF in Matlab
How would you transfer your

problem formulation to Matlab?
Discussion Points
• sin δ ≈ δ
• δ is in rad!
Discussion Points
• sin δ ≈ δ
• δ is in rad!
•B·θ =P
• B is in p.u.
• θ is in rad, ⇒ dimensionless
• P must be in p.u.
Discussion Points
• sin δ ≈ δ
• δ is in rad!
•B·θ =P
• B is in p.u.
• Bus Admittance Matrix B in DC-OPF

• bij = x1ij ⇒ positive
• all off-diagonal elements are non-positive (zero or negative)
• all diagonal elements are positive
• AC-OPF: This differs from the case where zij = rij + jxij . In that
case, it is yij = gij + jbij with bij is negative.
Discussion Points
• sin δ ≈ δ
• δ is in rad!
•B·θ =P
• B is in p.u.
• Bus Admittance Matrix B in DC-OPF

• bij = x1ij ⇒ positive
• all off-diagonal elements are non-positive (zero or negative)
• all diagonal elements are positive
• AC-OPF: This differs from the case where zij = rij + jxij . In that
case, it is yij = gij + jbij with bij is negative.
• If the DC-OPF does not converge, check that the admittance matrix B is
correct!
4-slide “break”
DC-OPF: linear program = convex
AC-OPF: non-linear non-convex problem in its original form

⇒ recent efforts to convexify the problem
Why?
Convex vs. Non-convex Problem
Convex Problem Non-convex problem

Cost f (x)
f (x) Cost
x x
One global minimum Several local minima
Several local minima: So what?
Example: Optimal Power Flow Problem
• Assume that the difference in the
cost function of a local minimum
versus a global minimum is 5%
f (x)
• The total electric energy cost in Cost
the US is ≈ 400 Billion$/year
• 5% amounts to 20 billion US$ in

economic losses per year
• Even 1% difference is huge
• Convex problems guarantee that x

we find a global minimum ⇒
convexify the OPF problem
Convexifying the Optimal Power Flow problem
(OPF)
f (x)
Cost
• Convex relaxations transform the
OPF to a convex Semi-Definite
Program (SDP)
Convex Relaxation
1 Javad Lavaei and Steven H Low. “Zero duality gap in optimal power flow problem”. In:
IEEE Transactions on Power Systems 27.1 (2012), pp. 92–107

(OPF)
f (x)
Cost f˜(x)
Program (SDP)
Convex Relaxation

(OPF)
f (x)
Cost f˜(x)
Program (SDP)
• Under certain conditions, the

obtained solution is the global
optimum to the original OPF
problem1 x
Convex Relaxation

Break is over... More in Dan Molzahn’s lecture tomorrow!
AC-OPF
• Minimize
• subject to:
AC-OPF
• Minimize
Costs, Line Losses, other?
• subject to:
AC-OPF
• Minimize
• subject to:
AC Power Flow equations

Line Flow Constraints
Generator Active Power Limits
Generator Reactive Power Limits
Voltage Magnitude Limits
(Voltage Angle limits to improve solvability)
(maybe other equipment constraints)
AC-OPF
• Minimize
• subject to:
AC Power Flow equations

Line Flow Constraints
Line Current Limits
Generator Active Power Limits
Apparent Power Flow limits
Generator Reactive Power Limits
Voltage Magnitude Limits Active Power Flow limits
(Voltage Angle limits to improve solvability)

(maybe other equipment constraints)
• Optimization vector: [P Q V θ]T

AC-OPF2
obj.function min cT PG
∗ ∗
AC flow SG − SL = diag(V )Y bus V
Line Current |Y line,i→j V | ≤ Iline,max
|Y line,j→i V | ≤ Iline,max
∗ ∗
or Apparent Flow |V i Y line,i→j,i-row V | ≤ Si→j,max
∗ ∗
|V j Y line,j→i,j-row V | ≤ Sj→i,max
Gen. Active Power 0 ≤ PG ≤ PG,max
Gen. Reactive Power − QG,max ≤ QG ≤ QG,max
Voltage Magnitude Vmin ≤ V ≤ Vmax
Voltage Magnitude Vmin ≤ V ≤ Vmax
Voltage Angle θmin ≤ θ ≤ θmax
2 All shown variables are vectors or matrices. The bar above a variable denotes complex
numbers. (·)∗ denotes the complex conjugate. To simplify notation, the bar denoting a complex
number is dropped in the following slides. Attention! The current flow constraints are defined as
vectors, i.e. for all lines. The apparent power line constraints are defined per line.
Current flow along a line
Vi Vj
It is:
1
Rij + jXij yij =
Rij + jXij
jBij jBij
Bij
2 2 ysh,i = j + other shunt
2
elements connected to that bus
π-model of the line
Vi Vj
It is:
1
Rij + jXij yij =
Rij + jXij
jBij jBij
Bij
2

[ ]
[ ] Vi
i→j: Ii→j = ysh,i Vi + yij (Vi − Vj ) ⇒ Ii→j = ysh,i + yij −yij
Vj
Vi Vj
It is:
1
Rij + jXij yij =
Rij + jXij
jBij jBij
Bij
2

[ ]
[ ] Vi
i→j: Ii→j = ysh,i Vi + yij (Vi − Vj ) ⇒ Ii→j = ysh,i + yij −yij
Vj
[ ]
[ ] Vi
j→i: Ij→i = ysh,j Vj + yij (Vj − Vi ) ⇒ Ij→i = −yij ysh,j + yij
Vj
Line Admittance Matrix Yline
• Yline is an L × N matrix, where L is the number of lines and N is the

number of nodes
• if row k corresponds to line i − j:

• Yline,ki = ysh,i + yij
• Yline,kj = −yij
1
• yij = is the admittance of line ij
Rij + jXij
• ysh,i is the shunt capacitance jBij /2 of the π-model of the line
• We must create two Yline matrices. One for i → j and one for j → i
Bus Admittance Matrix Ybus
Si = Vi Ii∗
∑
Ii = Iik , where k are all the buses connected to bus i
k
Example: Assume there is a line between nodes i − m, and i − n. It is:
Ii = Iim + Iin
i→m
= (ysh,i + yim )Vi − yim Vm + (ysh,i
i→n
+ yin )Vi − yin Vn
i→m
= (ysh,i i→n
+ yim + ysh,i + yin )Vi − yim Vm − yin Vn
Ii = [ysh,im + yim + ysh,in + yin −yim −yin ][Vi Vm Vn ]T

| {z } | {z } |{z}
Ybus,ii Ybus,im Ybus,in
Bus Admittance Matrix Ybus
• Ybus is an N × N matrix, where N is the number of nodes
∑
• diagonal elements: Ybus,ii = ysh,i + k yik , where k are all the buses
connected to bus i
• off-diagonal elements:
• Ybus,ij = −yij if nodes i and j are connected by a line3
• Ybus,ij = 0 if nodes i and j are not connected
1
• yij = is the admittance of line ij
Rij + jXij
• ysh,i are all shunt elements connected to bus i, including the shunt
capacitance of the π-model of the line
3 If there are more than one lines connecting the same nodes, then they must all be added to
Ybus,ij , Ybus,ii , Ybus,jj .

AC Power Flow Equations
Si = Vi Ii∗
∗
= Vi Ybus V∗
For all buses S = [S1 . . . SN ]T :

∗
Sgen − Sload = diag(V )Ybus V∗
From AC to DC Power Flow Equations
• The power flow along a line is:
Sij = Vi Iij∗ = Vi (ysh,i

∗
Vi∗ + yij∗ (Vi∗ − Vj∗ ))
• Assume a negligible shunt conductance: gsh,ij = 0 ⇒ ysh,i = jbsh,i .

• Given that R << X in transmission systems, for the DC power flow we
1
assume that zij = rij + jxij ≈ jxij . Then yij = −j .
xij
• Assume: Vi = Vi ∠0 and Vj = Vj ∠δ, with δ = θj − θi .
1
Iij∗ = − jbsh,i Vi + j (Vi − (Vj cos δ − jVj sin δ))
xij
1 1 1
= − jbsh,i Vi + j Vi − j Vj cos δ − Vj sin δ
xij xij xij
From AC to DC Power Flow Equations (cont.)
• Since Vi is a real number, it is:
1
Pij =ℜ{Sij } = Vi ℜ{Iij∗ } = − Vi Vj sin δ
xij
• With δ = θj − θi , it is:
1
Pij = Vi Vj sin(θi − θj )
xij
• We further make the assumptions that:
• Vi , Vj are constant and equal to 1 p.u.
• sin θ ≈ θ, θ must be in rad
Then
1
Pij = (θi − θj )
xij
Some additional points...
• Nodal prices
In a market context, the nodal prices are:
• the lagrangian multipliers of the equality constraints Bθ = P
• of a DC-OPF (at the moment)
• with objective function the minimization of costs
Some additional points...
• Nodal prices
In a market context, the nodal prices are:
• the lagrangian multipliers of the equality constraints Bθ = P
• of a DC-OPF (at the moment)
• with objective function the minimization of costs
• Power Transfer Distribution Factors (PTDFs)

• PTDFs are linear sensitivies that relate the line flows to the power
injections
• the DC-OPF can be formulated with respect to PTDFs
• PTDFs eliminate the need of θ as optimization variable
• In the zonal pricing in Europe PTDFs are used to model the flows
between the zones
AC-OPF
• Resources about AC-OPF from the US Federal Energy Regulatory

Commission (FERC)
https://www.ferc.gov/industries/electric/indus-act/
market-planning/opf-papers.asp
• Overview paper on Economic Dispatch and DC-OPF:
R.D. Christie, B. F. Wollenberg, I. Wangesteen, Transmission

Management in the Deregulated Environment, Proceedings of the IEEE,
vol. 88, no. 2, February 2000
Wrap-up
DC-OPF AC-OPF
• market clearing uses DC-OPF (at • primarily used for optimization of
the moment) operation and control actions
• convex • full AC power flow equations
• can solve fast; can be applied in but
very large problems
• non-convex (in its original form) →
but no guarantee that we find the
global optimum
• only active power flow
• computationally expensive and
• no losses and no voltage levels
possibly intractable for very large
systems
DC approximations more suitable efforts to decrease computation

for DTU
35 transmission systems
Electrical Engineering
time and increase system size
Optimal Power Flow (DC-OPF and AC-OPF) Jun 12, 2017
Thank you!
spchatz@elektro.dtu.dk

Optimal Power Flow (DC and AC OPF)

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Optimal Power Flow (DC and AC OPF)

Uploaded by

Copyright:

Available Formats

Optimal Power Flow (DC-OPF and AC-OPF)

DTU Summer School 2017

• CAISO, California, USA

• PJM, East Coast, USA

PGmin ≤ PGi ≤ PGmax

PGmin ≤ PGi ≤ PGmax

PGmin ≤ PGi ≤ PGmax

• We assume a copperplate network, i.e. a lossless and unrestricted flow of

Yes! With the help of the merit order curve.

Merit-Order of the German conventional generation in 2008. Source:

• In a DC-OPF context, there is no longer a single system marginal price

1 2 max = 100 MW, P max = 200 MW

X12 = 0.1 pu, X13 = 0.3 pu, X23 = 0.1

2 Formulate the objective function

3 Formulate the constraints

How would you transfer your

• Bus Admittance Matrix B in DC-OPF

• Bus Admittance Matrix B in DC-OPF

AC-OPF: non-linear non-convex problem in its original form

Convex Problem Non-convex problem

• 5% amounts to 20 billion US$ in

• Even 1% diﬀerence is huge

• Convex problems guarantee that x

IEEE Transactions on Power Systems 27.1 (2012), pp. 92–107

IEEE Transactions on Power Systems 27.1 (2012), pp. 92–107

• Under certain conditions, the

IEEE Transactions on Power Systems 27.1 (2012), pp. 92–107

Costs, Line Losses, other?

Costs, Line Losses, other?

AC Power Flow equations

Costs, Line Losses, other?

AC Power Flow equations

(Voltage Angle limits to improve solvability)

• Optimization vector: [P Q V θ]T

π-model of the line

π-model of the line

π-model of the line

• Yline is an L × N matrix, where L is the number of lines and N is the

• if row k corresponds to line i − j:

Example: Assume there is a line between nodes i − m, and i − n. It is:

Ii = [ysh,im + yim + ysh,in + yin −yim −yin ][Vi Vm Vn ]T

Ybus,ij , Ybus,ii , Ybus,jj .

For all buses S = [S1 . . . SN ]T :

Sij = Vi Iij∗ = Vi (ysh,i

• Assume a negligible shunt conductance: gsh,ij = 0 ⇒ ysh,i = jbsh,i .

• Assume: Vi = Vi ∠0 and Vj = Vj ∠δ, with δ = θj − θi .

• Power Transfer Distribution Factors (PTDFs)

• Resources about AC-OPF from the US Federal Energy Regulatory

• Overview paper on Economic Dispatch and DC-OPF:

R.D. Christie, B. F. Wollenberg, I. Wangesteen, Transmission

DC approximations more suitable eﬀorts to decrease computation

You might also like