You are on page 1of 7

2013 IEEE.

Personal use of this material is permitted. Permission from IEEE must be


obtained for all other users, including reprinting/ republishing this
material for advertising or promotional purposes, creating new
collective works for resale or redistribution to servers or lists, or reuse
of any copyrighted components of this work in other works.


Digital Object Identifier: tba




1
DC Optimal Power Flow Including HVDC Grids
Roger Wiget, Student Member, IEEE, Gran Andersson, Fellow, IEEE,
AbstractMany simulations in power systems analysis contain
an AC grid which is solved using linearized power ow methods.
This paper extends the known DC optimal power ow method
for AC grids for use in combined AC and HVDC grids. Future
studies would now incorporate in their simulations combined AC
and HVDC grids. This paper describes the additions made to the
algorithm and the inuence of the soft penalties. The algorithm
is tested in the IEEE 14 bus test case with a 5 node DC overlay
grid. The accuracy compared to the full power ow model is in
a reasonable range. The calculation performance is signicantly
increased.
Index TermsHVDC transmission, multi-terminal HVDC, op-
timal power ow.
I. INTRODUCTION
T
RANSMISSION grids have to be reinforced to meet
future energy demands. There are proposals to solve
this issue with voltage-source converter (VSC) Multi-terminal
HVDC (MTDC) grids [1], [2]. The introduction of a HVDC
grid instead of new AC lines has several advantages, mainly
the reduction of the overall costs, due to the lower losses and
the higher controllability [3].
Steady-state power ow can been calculated with different
methods. In order to reduce the costs, or losses, in a grid,
optimal power ow (OPF) was introduced. The main obstacle
for the accurate full OPF calculation is the high computational
effort, which is needed to solve this nonlinear, not convex
problem. That is why many linearized, so called DC OPF
methods, where introduced to speed up the calculations [4],
[5]. It should be noted that with DC power ows methods
we denote the linearized methods in order to solve the OPF
problem. The DC grid is not incorporated in these methods.
This may lead to some confusion. Including the HVDC grid
in these methods is the contribution of this paper.
The additional benets of the DC calculation methods
are the reliable and unique solutions with relatively simple
methods. The injections and the resulting ows are optimized
efciently, particularly in the demanding area of contingency
analysis [6]. The accuracy of the known linearized methods
is still in a reasonable range for most applications. A detailed
study of the errors is done in [7].
Active and reactive power ows can be calculated with the
full OPF. The reactive power production has strong effect only
on a local basis, therefore the active power ows are of primary
interest for studies in topics such as large scale energy markets.
The method presented considers only the active power.
R. Wiget and G. Andersson are with the Power Systems Laboratory, ETH
Zurich (Swiss Federal Institute of Technology), Switzerland. Corresponding
authors email address wiget@eeh.ee.ethz.ch. The work proposed in this paper
has been carried out within the HVDC networks research project, which is
nancially supported by ABB, Alstom Grid, Siemens, and the Swiss Federal
Ofce of Energy.
Manuscript received March 28, 2013
The production costs of the generators in the AC grid are
usually calculated with a quadratic cost function to calculate
social welfare. Linear cost are used for electricity market
bidding. Both are possible to use in this paper. One of these
cost models is included in most OPF as an objective function
and the linearized steady-state ow equations are embedded as
constraints. This results in a convex optimization. Additionally
the voltage angle differences are penalized with a soft penalty
factor. It allows to estimate the size and pattern of the error
and improves the numerical stability and convergence [8]. The
equations and constraints for the converter stations and a DC
grid are added to use the algorithm for combined AC and DC
grids.
This paper rst gives a brief theoretical background about
the linearized ow in the DC grid, followed by the detailed
formulation of the combined AC and DC grid OPF. A study
case in the IEEE 14 bus test case compares the presented
algorithm with a full OPF for combined AC and HVDC grids,
which was presented in [9]. Finally conclusions are provided.
II. OPF FORMULATION
The well known DC power ow for AC grid is expanded
to MTDC grids. The AC formulation can be found in detail
in [8] and is not repeated here.
A. DC calculation of active power ows in the HVDC grid
The main idea is to linearize the quadratic power ow
equations for the DC grid. The exact steady-state power ow
in a HVDC line from node k to m is represented in (1), where
U
k
is the voltage at node k in pu. R
km
is the resistance of
the HVDC line between the these nodes.
P
km
=
U
k
(U
k
U
m
)
R
km
(1)
If it is assumed that the DC voltages are rather close to the
nominal voltage, dened as 1 pu, (1) can be approximated.
P
km

(U
k
U
m
)
R
km
(2)
The resulting voltages U
k
and U
m
give only the deviation
from the nominal voltage at the reference bus.
With this approximation the linearization is already done.
The losses in the DC grid are neglected, similar to the AC
grid. The power balance equations for the AC (3) and DC (4)
grid are as follows:
P
Lk
= P
Gk

m
P
km
P
T k
(3)
P
Lk
=
M

m
P
km
+P
T k
(4)
2
P
Lk
, P
Gk
, and P
T k
are the load, generation and the AC/DC
terminal power at node k. N are the total amount of AC
branches. M the similar value for the DC branches.
Additionally some parameters have to stay within certain
limits. The active power generation at each bus is limited
between the maximal and minimal generation.
P
min
Gk
P
Gk
P
max
Gk
(5)
(6)
The power ow through the terminals is also limited. It is
assumed that in each direction the same maximum power can
be used.
P
max
T k
P
T k
P
max
T k
(7)
(8)
The limits in the branches, regardless of whether AC or DC,
are also xed. Similar to the converter stations, the ow can
either be positive or negative up to the same limit.
P
max
km
P
km
P
max
km
(9)
(10)
Equation 9 is valid for all DC and AC lines.
B. Objective function
The objective function (11) chosen in this paper represents
quadratic costs for the generators. In addition the voltage angle
differences for the AC grid and the DC voltage differences
in the DC grid are penalized to improve stability and allow
sensitivity experiments to get information about the AC-DC
approximation errors [8]. The soft penalty weights
AC
and

DC
are chosen rather small, such that the generator cost
strongly overweight the second part of the objective function.
I

i=1
[c
lini
P
Gi
+c
quad i
P
2
Gi
]+

AC
_

km
[
k

m
]
2
_
+
DC
_

km
[U
DCk
U
DCm
]
2
_
(11)
The coefcients for the linear costs are c
lini
and c
quad i
are
the coefcients for the quadratic costs for all I generators. The
AC voltage angle differences and DC voltage differences are
considered over all lines in the AC and DC grid.
This objective function is only one possibility, several other
options are possible to choose depending on the purpose of
the simulation. E.g. if c
quad i
= 0 and both
i
= 0, the whole
algorithm is linear.
C. Decision variable vector x
The vector x for the combined OPF contains different sets
of values. All active power generations P
G
are represented
in the state vector. Additionally the AC voltage angles are
contained in x. To control the power ow between the AC and
the DC grid also the power transfers through the terminals P
T
are in the state vector. Finally the voltages in the DC grid U
DC
are added to x. To calculate the ows in both grids only the
AC voltage angles differences and DC voltage differences are
of interest, therefore in each grid a slack bus can be chosen. At
these buses the angle , respectively voltage U
DC
are dened
as reference values and can be removed from x.
D. Nomenclature
The symbols used for the formulation of the algorithm are
summarized in table I.
TABLE I
NOMENCLATURE
I Total amount of generators
K Total amount of AC nodes
L Total amount of terminal stations
M Total amount of DC nodes
N Total amount of AC branches
O Total amount of DC branches
Z I+K+L+M-2 Length of the state vector x
Y 2(I+N+L+O) Amount of inequality constraints
E. Matrix form
The optimization problem can be formulated in matrix form
as follows:
min f(x) = min
1
2
x
T
Gx +a x (12)
with respect to
x = [P
G1
. . . P
GI

2
. . .
K
P
T1
. . . P
TL
U
DC2
. . . U
DCM
]
T
[Z1]
(13)
subject to
C
eq
x = b
eq
(14)
C
iq
x b
iq
(15)
The matrices used are constructed as explained in the fol-
lowing paragraphs. The quadratic costs matrix Gis constructed
out of QC and the reduced version of W
AC
and W
DC
.
The generators cost coefcients for the quadratic costs are
composed to a diagonal matrix QC with the size of I I.
QC = diag[2c
quad 1
2c
quad 2
. . . 2c
quad I
] (16)
W
DC
is the voltage difference weight matrix. The diagonal
elements W
ii
stand for the amount of DC lines connected to
node i. The off diagonal elements W
ij
indicate whenever a
line from i to j with a -1.
W
DC
= 2
DC
_

k=1
I
k1
I
12
. . . I
1M
I
21

k=2
I
k2
. . . I
2M
.
.
.
.
.
.
.
.
.
.
.
.
I
M1
I
M2
. . .

k=M
I
kM
_

_
[MM]
(17)
3
where I
km
is a function dened as:
I
km
=
_
1 if there is a branch from k to m or m to k
0 otherwise
The voltage angle difference weight matrix for the AC grid
can be constructed similarly. W
AC
is then a K K matrix.
Node 1 is chosen as the slack bus in the DC grid , similarly
AC node 1 is dened as the slack bus in the AC grid. Since
only the relative angle of the AC buses and the relative voltage
of the DC buses are of interest, the voltage angle at the AC
slack bus can be dened as zero, equally the voltage difference
at the DC slack bus is zero. Therefore the matrices W
AC
and
W
DC
can be reduced. The rst row and column of each of
them can be removed to get W
AC
r
and W
DC
r
.
The block diagonal matrix G is then composed out of QC,
W
AC
r
and W
DC
r
.
G = diag[QC, W
AC
r
, 0
[LL]
, W
DC
r
]
[ZZ]
(18)
The zero-columns with width of L correspond to the power
ow through the terminals. The zeros mean that no costs are
assigned to the terminal ows.
In vector a the linear cost coefcients c
lini
are included.
No other linear costs are considered in the proposed objective
function.
a = [c
lin1
c
lin2
. . . c
linI
0
[1 (Z I)]
]
[1Z]
(19)
F. Constraints
The equality constraints are given by the power balances
at each node. The power ow through the lines and terminals
is limited by the inequality constraints. The rst matrix to
calculate is H, it allocates the different generators to the nodes.
H =
_

_
H
11
. . . H
1I
H
21
. . . H
2I
.
.
.
.
.
.
.
.
.
H
K1
. . . H
KI
_

_
[KI]
(20)
H
pq
=
_
1 if generator q is connected at node p
0 otherwise
The next matrix to calculate is the admittance matrix B
DC
.
The rst row of the admittance matrix can be removed due to
the xation of the voltage at bus 1. Therefore
B
DC
r
=
_

_
B
DC
21

k=2
B
DC
k2
. . . B
DC
2K
B
DC
31
B
DC
32
. . . B
DC
3K
.
.
.
.
.
.
.
.
.
.
.
.
B
DC
M1
B
DC
M2
. . .

k=M
B
DC
kM
_

_
[(M1)M]
(21)
B
DC
km
=
_
1
R
km
[pu]
if branch km or mk exist
0 otherwise
where R
km
is the line resistance and R
km
= R
mk
.
The same matrix for the AC grid is B
AC
r
. Here B
AC
km
is
dened as follows:
B
AC
km
=
_
1
X
km
[pu]
if branch km or mk exists
0 otherwise
where X
km
is the line reactance and X
km
= X
mk
.
The line adjacency matrix A denes the branches. For each
branch the starting bus is marked positive conductivity and the
ending bus with the negative. Again the rst column can be
neglected.
A
DC
r
=
_

_
J
21
. . . J
M1
J
22
. . . J
M2
.
.
.
.
.
.
.
.
.
J
2O
. . . J
MO
_

_
[OM]
(22)
J
pq
=
_

_
B
DC
p
if line p starts at node q
B
DC
p
if line p ends at node q
0 otherwise
The same procedure done for the AC grid results in the A
AC
matrix, with the size of N K. The connection matrix T
between the AC nodes and the DC nodes is composed as
following:
T =
_

_
T
11
. . . T
1L
T
21
. . . T
2L
.
.
.
.
.
.
.
.
.
T
K1
. . . T
KL
_

_
[KL]
(23)
T
pq
=
_
1 if terminal q is connected at node p
0 otherwise
The same is done for the DC nodes.
S =
_

_
S
11
. . . S
1L
S
21
. . . S
2L
.
.
.
.
.
.
.
.
.
S
M1
. . . S
ML
_

_
[ML]
(24)
S
pq
=
_
1 if terminal q is connected at DC node p
0 otherwise
The equality constraint matrix C
eq
can be built out of the
prepared matrices.
C
eq
=
_
H B
AC
r
T 0
0 0 S B
DC
r
_
[(K+M)Z]
(25)
The rst row corresponds to the power balances at the AC
nodes, the second row is the respective equations in the DC
grid. The equality vector b
eq
consists out of the total load at
each node, for AC and DC nodes.
4
b
eq
= [P
L1
P
L2
. . . P
L(K+M)
]
T
[(K+M)1]
(26)
Loads can be connected directly to the DC grid, if no such
connections exist P
Li
= 0 for all i > K. The inequality
constraints are represented by the C
iq
matrix. I
I
and I
L
are
identity matrices with the size of I and L respectively.
C
iq
=
_

_
I
I
0 0 0
I
I
0 0 0
0 A
AC
r
0 0
0 A
AC
r
0 0
0 0 I
L
0
0 0 I
L
0
0 0 0 A
DC
r
0 0 0 A
DC
r
_

_
[Y Z]
(27)
The rst two rows correspond to the generators limits,
followed by two rows for the AC line limits. The transfer
capacities of the terminal are limited in row 5 and 6. The
last two rows limit the ow through the DC lines. The vector
b
iq
consists of 8 other vectors. There b
Up
gen
and b
Up
gen
stand for
the upper and lower limits of the generators. b
br
and b
DCbr
represent the line limits for the AC and DC lines, respectively.
The active power through each terminal is limited by the
values in b
ter
.
b
Up
gen
= [P
Up
G1
P
Up
G2
. . . P
Up
GI
]
T
[I1]
(28)
b
Lo
gen
= [P
Lo
G1
P
Lo
G2
. . . P
Lo
GI
]
T
[I1]
(29)
b
br
= [P
br1
P
br2
. . . P
brN
]
T
[N1]
(30)
b
ter
= [P
T1
P
T2
. . . P
T L
]
T
[L1]
(31)
b
DCbr
= [P
DCbr1
P
DCbr2
. . . P
DCbrO
]
T
[O1]
(32)
All this vectors together result in b
iq
.
b
iq
= [b
Up
gen
b
Lo
gen
b
br
b
br
b
ter
b
ter
b
DCbr
b
DCbr
]
T
[Y 1]
(33)
III. CASE STUDY
The comparison between the linearized and full AC OPF [9]
algorithms is carried out at the IEEE 14 bus test case, with a 5
node overlay DC grid as shown in Fig. 1. Node 25 is a DC bus
only, with no connection to the AC grid. All parameters for the
AC grid are from the MATPOWER test case14 [10]. All DC
lines have the same resistance of 2.78 , which represents a
100 km DC cable. Their capacity is limited to 50 MW, which
compares to the capacities of the rest of the grid.
A. Accurancy
Three different values for the soft penalties are used in the
linearized OPF. First
AC
and
DC
are set to zero, second
both have the same value of 0.05. The third simulation uses
weighted penalty factors. They are calculated iteratively, the
average penalty costs per line in the AC and DC grid are
4
5
6
7
8
9
10
11
12
13
14
1
3
2
V
S
C
V
S
C
V
S
C
V
S
C
25
AC Line
Generator
Voltage source
converter
DC Line
Transformer
VSC
Fig. 1. Combined AC and HVDC grid
calculated. Then
AC
is adjusted, until the average penalty
cost on the AC lines are equal to the value on the DC lines.
This is achieved for
AC
= 28 and
DC
= 0.05. For all
simulations the ows in the lines and terminals are compared
to each other.
The power injection to the DC grid is in all simulations
node 1, as shown in Fig. 2. This is reasonable since the
cheapest and highest generation capacity is located at this
node. All other terminals take power out of the DC grid. The
most load is connected to node 3, 4 and 9 are next in the
loading. To reduce the losses it makes sense to control the
DC grid in such a way that the output is as close as possible
to the highest loads. For the full OPF case, node 3 takes the
highest power out of the DC grid. In the linearized OPF it is
depending on the penalty terms, where the weighted values
ts best compared to the full OPF.
The ows in the DC lines are shown in Fig. 3. Except
the small ow on line 13-25, all branch ows have the same
direction for all simulations. Only two lines are loaded close
to their limit of 50 MW for the full OPF. In the linearized
method the line loadings are far below this level, except the
simulations with the weighted penalty terms. For the other two
linearized simulations the utilization of the DC grid is much
smaller than for the full OPF. Since losses are considered in the
full OPF, its preferable to transfer the power over the DC lines
instead of using the AC lines. Therefore an opposite situation
is observable in the AC grid.
Figure 4 shows the AC lines loadings, which are in general
more loaded in the linearized OPF. All major ows in the AC
lines are similar for the full and the linearized OPF. Again the
simulation with weighted penalty factors ts best to the full
OPF.
In the simulation without penalty terms the ow are not
penalized, therefore its hard to decide if the power should
ow through the AC or DC grid. If similar values for the
penalty terms are used, the ows are penalized, but not in
a similar amount. The voltage difference are usually bigger
than the voltage angle differences. Therefore the AC grid is
preferred choice in this simulation. With the weighted soft
penalty factors all ows are penalized in the same amount.
5
1-2 1-5 2-3 2-4 2-5 3-4 4-5 4-7 4-9 5-6 6-11 6-12 6-13 7-8 7-9 9-10 9-14 10-11 12-13 13-14
-50
0
50
100
AC line from k to m
P
o
w
e
r

o
w
[
M
W
]
Lin OPF,
AC
=
DC
= 0
Lin OPF,
AC
= 0.05,
DC
= 0.05
Lin OPF,
AC
= 28,
DC
= 0.05
Full AC OPF
Fig. 4. Flows in the AC grid
Therefore the algorithm tries to minimize the ow in the
grid, which is in line with a reduction of losses. That is
why this simulation is close to the full OPF. Nevertheless the
costs for the generators are several magnitudes higher then the
penalty factors. The costs are 7642.59 $/h and the weighted
simulations the penalties are 0.73$/h for the AC lines and
0.26$/h for the DC lines in total.
1 3 9 13
-50
0
50
100
Terminal Number, Connected to AC Bus
P
o
w
e
r

o
w
[
M
W
]
Lin OPF,
AC
=
DC
= 0
Lin OPF,
AC
= 0.05,
DC
= 0.05
Lin OPF,
AC
= 28,
DC
= 0.05
Full AC OPF
Fig. 2. Interaction between AC and HVDC grid
1-13 1-25 3-9 3-25 9-13 9-25 13-25
-40
-20
0
20
40
DC line from k to m
P
o
w
e
r

o
w
[
M
W
]
Fig. 3. Flows in the DC grid
B. Performance
The main advantage is the fast calculation performance.
The full OPF for the combined AC and DC grid, with its
nonlinear constraints and the quadratic objective function,
requires 2.57 seconds to nd a solution within the tolerances.
The problem solved on the same ofce pc (Q9950 2.83GHz,
8GB RAM) with the linearized OPF takes 14.09 milliseconds,
this is over 180 times faster. This could be a major advantage
especially for all simulations which use multiple runs such as
Monte Carlo simulations. If several ten thousand repetitions
are needed, for example to calculate varying renewable infeed.
It is expected that the time ratio for bigger systems could even
increase, due to the simplicity of the algorithm.
IV. CONCLUSION
The method presented in this paper allows the inclusion of
multi-terminal HVDC grids in several different studies. Espe-
cially the topic of energy markets and uctuating infeed could
prot and it is possible to include the future grid structure.
It can be simply implemented in most existing simulations.
The calculation performance gained compared to the full OPF
justies the simplications, whereas the simulation results are
never exact, but still in a reasonable range. Especially if the
weighted penalty factors are used. A future expansion of this
algorithm will include some security assessments, allowing an
OPF which fullls the N-1 security constraints.
ACKNOWLEDGMENT
The authors would like to thank Emil Iggland for linguistic
input and Spyros Chatzivasileiadis for the support.
REFERENCES
[1] D. Jovcic, D. van Hertem, K. Linden, J.-P. Taisne, and W. Grieshaber,
Feasibility of dc transmission networks, in Proc. IEEE ISGT Europe,
Manchester UK, Dec 2011, pp. 18.
[2] Friends of the Supergrid FOSG WG2. (2012, Mar) Roadmap
to the supergrid technologies. Final Report. [Online]. Available:
http://www.friendsofthesupergrid.eu
6
[3] H. Wang and M. Redfern, The advantages and disadvantages of using
HVDC to interconnect ac networks, in 45th International Universities
Power Engineering Conference (UPEC), 2010, pp. 1 5.
[4] W. Tinney and C. Hart, Power ow solution by newtons method,
Power Apparatus and Systems, IEEE Transactions on, vol. PAS-86,
no. 11, pp. 1449 1460, nov. 1967.
[5] B. Stott, Review of load-ow calculation methods, Proceedings of the
IEEE, vol. 62, no. 7, pp. 916 929, july 1974.
[6] B. Stott, J. Jardim, and O. Alsac, DC power ow revisited, IEEE
Trans. Power Systems, vol. 24, no. 3, pp. 1290 1300, aug. 2009.
[7] K. Purchala, L. Meeus, D. Van Dommelen, and R. Belmans, Usefulness
of DC power ow for active power ow analysis, in IEEE Power
Engineering Society General Meeting, 2005, pp. 454 459 Vol. 1.
[8] J. Sun and L. Tesfatsion, DC Optimal Power Flow Formulation and
Solution Using QuadProgJ, ISU Economics Working Paper, 2010.
[9] R. Wiget and G. Andersson, Optimal power ow for combined AC and
multi-terminal HVDC grids based on VSC converters, in IEEE Power
and Energy Society General Meeting, july 2012, pp. 1 8.
[10] R. Zimmerman, C. Murillo-Sandchez, and R. Thomas, MATPOWER:
Steady-State Operations, Planning, and Analysis Tools for Power Sys-
tems Research and Education, IEEE Trans. on Power Systems, vol. 26,
no. 1, pp. 12 19, feb. 2011.
Roger Wiget (S11) was born in Lucerne, Switzer-
land. He received a bachelor degree in electri-
cal engineering from the ETH Zurich, Switzerland
in 2009 and a masters degree in Energy Science
and Technology from the same institution in 2011.
He joined the Power Systems Laboratory of ETH
Zurich, Switzerland in 2011 where he is working
towards a PhD. His research is dedicated to HVDC
networks. He is a student member of the IEEE.
Gran Andersson (M86, SM91, F97) obtained
his M.S. (1975) and Ph.D. (1980) degrees from
the University of Lund, Sweden. In 1980 he joined
ASEAs, now ABBs, HVDC division in Ludvika,
Sweden, and in 1986 he was appointed full professor
in electric power systems at KTH (Royal Institute
of Technology), Stockholm, Sweden. Since 2000
he is full professor in electric power systems at
ETH Zurich (Swiss Federal Institute of Technology),
where he also heads the powers system laboratory.
His research interests include power systems dynam-
ics and control, power markets, and future energy systems. Gran Andersson
is a fellow of the Royal Swedish Academy of Sciences, and of the Royal
Swedish Academy of Engineering Sciences. He is Editor-in-Chief of IET
Proceedings Generation, Transmission and Distribution and the recipient of
the IEEE PES Outstanding Power Educator Award 2007 and of the George
Monteore International Award 2010.