Beruflich Dokumente
Kultur Dokumente
ODPF
TP
SE
ONR
VVO
CVP
SC
AR SM
AMR: Automatic Meter Reading
AR: Adaptive Relaying
CA: Contingencies Analysis
CVP: Capacitor/Voltage regulator Placement
CVR: Conservative Voltage Regulation
DG: Distributed Generation
EDA: Environmental Data Acquisition
MDI: Meter Data Integration
ODPF: Optimal Distribution Power Flow
ONR: Network Reconfiguration
OTS: Operator Training Simulator
RCS: Remote Controlled Switches
SC: Short Circuit analysis
SCs: Switching Capacitors
SDF: Supply and Demand Forecast
SE: State Estimator
SF: Storage Facilities
SM: Switch Management
SMS: Storage Management System
TP: Topology Processor
ULTC: UnderLoad Tap Changer
VVO: Voltage Var Optimization
SDF
CA
SMS
CVR
EDA
OTS
Offline
Real time
concentrator
AMR
AMR
AMR
concentrator
SERVER
MDI
GISINTERFACE
Environmental
Monitoring
Stations
CONTROLCENTER
Signals from RTUs
AMI
DISTRIBUTION
NETWORK
RCS
SCs
ULTC
DG
DG
SF
3
described in [4], whereas bold and dotted contours identify the
elements necessary for developing ONR tools.
The main idea is to reconfigure the network during the day-
ahead or the extended real time (about 15-30 minutes [12])
framework of system operation. The starting configuration
must be identified through a topology processor (TP) and a
state estimator (SE). TP is an on-line module that collects the
status of each switch and gives back an actual snapshot of the
grid. The state estimator module calculates voltages, currents
and angles basing on pseudo-measurements from the field.
ONR evaluates the necessary switching maneuvers for
implementing the best grid configuration in terms of loss
reduction or that minimizes control effort during active power
rescheduling.
A. Formulation of the ONR problem
The ONR problem can be formulated as a
) , , ( C min
obj
,
u x V
u x
(1)
subject to equality and inequality constraints
0 = ) , , ( f u x V (2)
0 s ) , , ( g u x V (3)
with
x
O e x and
u
O e u .
In (1) function C
obj
is the objective function to be
minimized; V, x and u represent respectively the set of node
voltages, the set of continuous variables (for example
generated active and reactive power) and the set of discrete
control variables (i.e the status open/closed of each switch).
Eqn. (2) takes into account the non-linear load flow equations,
whereas in (3) all technical and operational requirements are
taken into account (line and transformer thermal limits,
minimum and maximum voltages, etc.) adding non linearity to
the problem. Since the ONR might contain both integer and
real variables, the resulting overall problem is usually a mixed
integer non linear optimization problem (MINLP).
The overall MINLP can be solved by means of several
techniques that might involve relaxation of integer variables
[13] or problem decomposition. Given the good performance
of SA in solving non linear integer problems, a common
approach consists in decomposing the problem by means of
Bender decomposition [14] or formulating it as a two stage
optimization [15]-[16].
In this paper, for the sake of preserving modularity of DMS
tools, the optimization algorithm adopts SA for the sole
solution of a integer non linear problem whereas all
operational constraints and continuous variables are taken into
account by means of an optimization code based on non linear
programming techniques (basically a TDOPF). This means
that to each configuration selected by the SA code, a single
feasible solution coming from the optimization tool is
associated. This methodology allows to embed network
constraints in the OPF code and not in the SA. This might
result in a higher computational burden but it avoids the
definition of time varying weights or constraint relaxation
rules in the first steps of SA as in [17].
B. Objective functions
Objective functions vary according to the specific operative
problem that must be solved. In the case of minimization of
system losses the objective function can be formulated as:
2
1
0
|
|
|
|
.
|
\
|
=
=
nbus
i
Li
obj
P
Losses Total
C o (4)
Other formulations can be aimed at the minimization of
control effort during power system rescheduling.
Hypothesizing that, at a certain moment of the day, too much
power is produced causing the violation of one or more
security limits and that the generated output must be
rescheduled, the objective function will be formulated as in
the following expression :
=
|
|
.
|
\
|
=
ngen
i
sched Gi
Gi sched Gi
i , obj
P
P P
C
1
2
0
o (5)
where, for the i-th generator, P
Gi
is the generated power after
rescheduling and P
Gi sched
is the amount of power that was
scheduled for production. Clearly the solution of this problem
assumes the availability of dispatchable generating power
resources. Further operative constraints in redispatching can
be taken into account as hard limits.
Other possible operative problems might be solved
adopting different formulations of the objective function and
assuming different set of control variables. Other possible
formulations can be found in [2]-[4].
C. Penalty functions
Penalty functions are introduced in order to take into
account inequality constraints in (3). Typically inequality
constraints are referred to thermal limits of branches and to
acceptable voltage profiles, but can clearly take into account
several other operative and technical constraints.
In this paper three penalty functions (C
p1
, C
p2
and C
p3
) have
been formulated. These functions are referred respectively to
line, transformer and voltages constraints, and were
formulated according to the following formulations:
=
|
|
.
|
\
|
=
nlines
i i max,
i max, i
i , p
I
I I
C
1
2
1 1
o (6)
with
i max, i i ,
I I if < = 0
1
o ;
=
|
|
.
|
\
|
=
ntrasf
j j max,
j max, j
j , p
S
S S
C
1
2
2 2
o (7)
with
j max, j j ,
S S if < = 0
2
o ;
=
|
|
.
|
\
|
=
nbus
k k lim,
k lim, k
k , p
V
V V
C
1
2
3 3
o (8)
with
s s =
< =
> =
k max, k k min, k ,
k min, k k min, k lim,
k max, k k max, k lim,
V V V
V V V V
V V V V
if 0
if
if
3
o
.
4
In (6)-(8), I, S and V represent respectively line current,
transformer apparent power and node voltage, whereas
coefficients are relative weights for penalty functions.
Suffixes max e min refer to maximum and minimum technical
constraints.
Under these assumptions the ONR problem can be
formulated as
) , , ( C min
,
u x V
u x
(9)
subject to equality constraints
0 = ) , , ( f u x V (10)
with
3 2 1 p p p obj
C C C C C + + + = (11)
and
x
O e x and
u
O e u .
D. Solving algorithm
The ONR algorithm proposed in this paper was formulated
in such way that the modularity of ADMS tools is ensured.
Basically the algorithm is structured with a SA algorithm that
searches and selects radial network configurations and a slave
code that performs a mere Distribution Three-Phase Load
Flow (DTLF) or solves a more complex optimization problem
(for example a TDOPF).
The ONR algorithm was implemented in a
Matlab/OpenDSS environment, exploiting the data exchange
COM (Component Object Model) interface that is available in
the OpenDSS package [18]. The ONR algorithm, based on
simulated annealing meta-heuristic method, is implemented in
a Matlab code. The algorithm looks for new configurations
and communicates necessary topological changes to the
OpenDSS grid model. Each state contains the status of about
100 controllable disconnector switches that interconnect
laterals and backbones of the eleven urban feeders.
If ONR does not involve the optimization of other variables
than the status of disconnectors, the slave code is constituted
by a simple DTLF. In this case, the OpenDSS simulation
engine performs a distribution load flow for each new state,
evaluate losses and calculates voltages and power flows.
When other control variables have to be optimized the
slave code will perform an optimization of all control
variables (i.e. active and reactive resources, loads, switching
capacities, under-load tap changers, etc.) minimizing the
chosen objective function. All inequality constraints are taken
into account by means of suitable penalty functions that are
minimized concurrently to the objective function.
The optimization block is developed in a Matlab/OpenDSS
environment and is based on the work presented in [2]-[4]. An
important feature of such optimization code is that it can treat
concurrently both single- and three-phase system
representation. Each system element, represented as a single
object with its single- or three-phase model, can be controlled
by the optimizing code, overcoming the limitations of the
nodal approach.
In Fig. 2 a schematic representation of the proposed
algorithm is given. First of all, it must be remembered that in
this approach only radial configuration can be accepted. In
fact, AMET system is always operated with a multiradial
configuration and any new network state must be
characterized by this same property. Since radial solutions are
very few within the entire search space the state boundary was
restricted to those configurations that can be reached by
closing and opening two elements of the grid per time. This
approach allows not only to take into account just radial
solutions, but also to obtain states that can be reached by the
first state through a limited number of switching maneuvers,
keeping a radial configuration after each maneuver.
In order to find only radial solutions a simple strategy, very
close to the one proposed in [15]-[17], is followed. Starting
from a radial known configuration (for example the actual
configuration produced by the TP) an open disconnector is
randomly selected for being closed. Since we have added a
closed branch to a connected radial graph the resulting
configuration is meshed. At this point, if one disconnector is
opened and the system remains connected the resulting
configuration is radial. The algorithm searches among all open
switches for those ones that can be opened without losing the
connectivity of the network, and randomly selects one of
them, producing a new radial system configuration.
select or detect initial
configuration
k=0 u=u
0
evaluate objective function C
through DTLF or TDOPF
k=k+1
close a random switch
search all switches that can be opened; open one
randomly and set new configuration u=u
k
is u
k
a new
configuration
?
evaluate objective function C
k
through DTLF or TDOPF
C
k
< C
k-1
?
yes
no
yes
accept new configuration u
k
and reduce temperature T
no select a random number R and
set probability P(T)
R< P(T)?
yes
T< T
min
?
yes
STOP
no
discard new configuration and
restore old one
u=u
k-1
no
Fig.2: Flow chart of the proposed ONR algorithm
Once a new configuration is selected for study, the
algorithm optimizes/evaluates objective and penalty functions
for that configuration by a simple DTLF or through an
optimization of all available control variables. In order to save
computation time, such optimization is performed only if this
configuration has not been analyzed before. The objective
function evaluated for this configuration is compared to the
one evaluated at the previous step, and the new configuration
is chosen or discarded according to SA working principles.
The algorithm stops after the temperature is below a certain
5
threshold or having reached a maximum number of iterations
or whenever a certain objective is reached (for example C is
below a certain tolerance value).
III. TEST RESULTS
Test results have been obtained carrying out simulations on
a realistic sized representation of the AMET distribution
system. The AMET system is interconnected with the
subtrasmission system through a single HV/MV substation
and supplies energy by means of two 150kV/20kV
transformers equipped with controllable tap changers and
eleven 20kV feeders. The two transformers have respectively
a capacity of 30 and 25 MVA, with a base load of 20 MW and
peaks of about 35 MW. A simplified scheme of the
distribution substation is given in Fig. 3.
Fig.3: Simplified scheme of the AMET distribution substation
The system model, comprising all HV and MV elements,
consists of 930 buses, 1000 distribution lines (cabled and
overhead), 100 controllable switches, 500 load buses.
Test results were obtained considering the whole
representation of the system comprising an urban area, served
by the 30 MVA transformer and the main four feeders, and the
rural network supplied by a 25 MVA transformer.
An important feature of this network, makings ONR
applications particularly appealing, is that the network is
characterized by a tangled architecture that allows several
reclosures on elements belonging to different feeders. Such
tangled structure is a legacy of an inhomogeneous
urbanization process that led to an incremental development
of the distribution network. The distributor in fact had to
connect, one after the other, blocks or buildings as soon as the
they were completed and new connections to the grid were
requested. The result is that very often laterals of a feeder can
be directly connected to laterals of different feeders. Moreover
each backbone has at least one reclosure switch at the end of
the line.
A. Case A
The first case was aimed at finding the optimal network
configuration for losses reduction. The adopted objective
function is the one formulated in (4). The initial configuration
is characterized by a value of losses of about 1100 kW.
In figs. 4 and 5, the overall convergence behavior is shown.
Figure 4 shows how the overall objective function is
minimized along the iterative method. It also suggests how the
method is able to move out of suboptimal areas and search for
a global minimum. At the last iterations the overall amount of
losses is reduced to about 690 kW.
In Table I the value of each minimized function, at the
beginning and at the end of the optimization, is shown. At the
last iteration all penalty functions are null, indicating how to
method is able to minimize losses and remove security
violations at the same time.
0 20 40 60 80 100
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
iteration #
C
[
p
.
u
.
]
Fig.4: Case A, function C along the SA algorithm
0 20 40 60 80 100
650
700
750
800
850
900
950
1000
1050
1100
iteration #
a
c
t
i
v
e
p
o
w
e
r
l
o
s
s
e
s
[
k
W
]
Fig.5: Case A, overall losses along the SA algorithm
TABLE I
CASE A. OBJECTIVE AND PENALTY FUNCTIONS AT FIRST AND LAST ITERATION
Iter. # C C
obj
C
p1
C
p2
C
p3
1 1.749 0.992 0.756 0.000 0.001
100 0.413 0.413 0.000 0.000 0.000
B. Case B
The second test case was carried out introducing in the
network a significant amount of power generators (about 13
generators, producing about 32 MW). In this case the huge
amount of generated power gives rise to congestions on
several distribution lines (see Fig. 6).
By applying the proposed methodology and employing the
objective function formulated in (5), an optimal configuration
is found. As shown in Fig. 7, the methodology converges very
quickly to a solution where all constraints are respected and
rescheduling can be avoided. After reconfiguration the current
flowing in each distribution line is below its ampacity (Fig. 8).
6
0 100 200 300 400 500 600 700 800 900
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
# line
I
/
I
m
a
x
[
p
.
u
.
]
Fig.6: Case B, line current vs. ampacity (before ONR)
0 5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
iteration #
C
[
p
.
u
.
]
Fig.4: Case B, function C along the SA algorithm
0 100 200 300 400 500 600 700 800 900
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
# line
I
/
I
m
a
x
[
p
.
u
.
]
Fig.8: Case B, line current vs. ampacity (after ONR)
C. Case C
A third test case was performed considering a significant
increase of power that has to be generated. All other
hypotheses are analogous to case B. The amount of power to
be generated (about 60 MW) is so massive that, even with the
best system configuration, it will be affected by a generation
rescheduling. The ONR scheme is aimed at the minimization
of the controlling effort and, therefore, at ensuring that the
maximum amount of power is generated.
Figure 9 shows how the method converges towards an
optimal solution in about 150 iterations. Figure 10 shows how
the total generated power is maximized along the process. At
the final iteration the system configuration allows the
production of about 47.9 MW, whereas at the first iteration
the total power would have to be cut at 30 MW.
0 50 100 150
0
0.5
1
1.5
2
2.5
iteration #
C
[
p
.
u
.
]
Fig.9: Test case C, function C along the SA algorithm
0 50 100 150
20
25
30
35
40
45
50
iteration #
t
o
t
a
l
g
e
n
e
r
a
t
e
d
p
o
w
e
r
[
M
W
]
Fig.10: Test case C, function C along the SA algorithm
In both cases the system is not able to produce the entire
quantity (60 MW) and generation has to be curtailed through
generated active power control. In this case, thanks to system
reconfiguration, DSO would be able to reduce the amount of
power to be curtailed from 30 MW to 12.1 MW.
In Tables II and III, it is shown how the DTOPF code finds
the minimum acceptable curtailment in the presence of all
above quoted inequality constraints. In both cases (first and
last iteration of SA) the DTOPF converges rapidly in about 10
iterations.
TABLE II
CASE C. MINIMIZATION OF CONTROL EFFORT (INITIAL CONFIGURATION)
Iter. # C C
obj
C
p1
C
p2
C
p3
0 54.908 1.243 53.654 0.000 0.011
1 19.947 0.532 19.415 0.000 0.000
2 19.174 0.530 18.643 0.000 0.000
3 6.649 0.501 6.148 0.000 0.000
4 2.951 0.465 2.486 0.000 0.000
5 2.296 0.446 1.850 0.000 0.000
6 2.216 0.434 1.782 0.000 0.000
7 2.196 0.423 1.773 0.000 0.000
8 2.190 0.404 1.786 0.000 0.000
9 2.180 0.410 1.770 0.000 0.000
10 2.177 0.408 1.769 0.000 0.000
7
TABLE III
CASE C. MINIMIZATION OF CONTROL EFFORT (OPTIMAL CONFIGURATION)
Iter. # C C
obj
C
p1
C
p2
C
p3
0 51.790 1.243 50.530 0.000 0.017
1 81.051 0.000 81.051 0.000 0.000
2 77.026 0.000 77.026 0.000 0.000
3 8.827 0.069 8.758 0.000 0.000
4 0.912 0.134 0.778 0.000 0.000
5 0.398 0.150 0.248 0.000 0.000
6 0.274 0.153 0.121 0.000 0.000
7 0.182 0.155 0.027 0.000 0.000
8 0.161 0.155 0.006 0.000 0.000
9 0.162 0.157 0.005 0.000 0.000
D. CPU timings
All simulations were carried out on an ordinary desktop
PC, HP Compaq 8000 Elite CMT PC, with Intel Core 2 Quad
CPU Q 9650 3.00 GHz and 4.00 GB RAM. The total elapsed
time is show for each case in Table IV.
All timings appears to be compatible with extended real-
time system operation and a 15-30 minutes control
framework. The computation effort for Case C is about 27
minutes, but it can be drastically reduced through some code
optimization and executing the simulation on a more powerful
machine. Furthermore, in the presence of severe time
requirements, it is always possible to accept suboptimal
solutions that are available to the SA algorithm after few tens
of iterations.
TABLE IV
CPU EXECUTION TIME
Case Time [s]
A 295
B 386
C 1650
IV. CONCLUSIONS
The paper presented an Advanced Distribution
Management System scheme that integrates Optimal Network
Reconfiguration (ONR) within its monitoring and control
functions. ONR problem was solved through a simulated
annealing based algorithm that allows an easy integration with
other system analysis and optimization tools.
Test results obtained by implementing such algorithm on a
detailed representation of a medium-sized urban distribution
network showed how the methodology can be applied in the
extended real-time operative framework of such system. The
methodology is able to ensure the selection of an optimal
configuration minimizing system losses and respecting all
technical and operative constraints. Moreover, when
associated to other DMS tools such as a TDOPF, the
methodology permits to minimize further the control effort
requested for removing security violations.
The main foreseeable bottleneck relies in the on-line
implementation of system state estimators that will have to
exploit data received from the field from smart meters and
distributed generators, and build the system model to be
adopted in the optimization code.
V. ACKNOWLEDGMENT
The present study was funded under the grant PST #44
Smart-Grids: Advanced Technologies for utilities and
energy, granted by the Regione Puglia as Strategic Project in
the Framework Program Agreement on the scientific research
sector in the Apulian region.
The authors would like to thank P.Eng Walter Leggieri and
all personnel at AMET for the help provided during the
modelling of the system representation. The authors would
also thank Antonio Ripa for his help in the preparation of
numerical results.
VI. REFERENCES
[1] J. Fan and S. Borlase, The Evolution of Distribution, IEEE Power &
Energy, Vol. 7, pages 63-68, No. 2, Mar/Apr 2009.
[2] S. Bruno, S. Lamonaca, G. Rotondo, U. Stecchi, M. La Scala,
Unbalanced Three-phase Optimal Power Flow for Smart Grids, IEEE
Transactions on Industrial Electronics, Vol. 58, No. 10, pp. 4504-4513,
October 2011.
[3] M. Bronzini, S. Bruno, M. La Scala, R. Sbrizzai, Coordination of
Active and Reactive Distributed Resources in a Smart Grid, PowerTech
2011, Trondheim (Norway), 19-23 June, 2011.
[4] S. Bruno, M. La Scala, U. Stecchi, Monitoring and Control of a Smart
Distribution Network in Extended Real-Time DMS Framework, Cigr
International Symposium - The Electric Power System of the Future.
Integrating supergrids and microgrids, Bologna, Italy, September 13-15,
2011.
[5] G. Celli, F. Pilo, Optimal distributed generation allocation in MV
distribution networks, in proc. Power Industry Computer Applications,
2001 (PICA 2001) - Innovative Computing for Power - Electric Energy
Meets the Market. 22nd IEEE Power Engineering Society International
Conference on, pp. 81-86.
[6] G. Celli, E. Ghiani, S. Mocci, F. Pilo, A Multiobjective Evolutionary
Algorithm for the Sizing and Siting of Distributed Generation, IEEE
Transaction on Power Systems, Vol. 20, No. 2, pp. 750-757, May 2005.
[7] A. Silvestri, A. Berizzi, S. Buonanno, Distributed generation planning
using genetic algorithms, Conference on Electric Power Engineering,
PowerTech Budapest 99 International, Budapest (Hungary) 29 Aug -02
Sep 1999.
[8] J.M. Nahamn, D.M. Peri, Optimal Planning of Radial Distribution
Networks by Simulated Annealing Technique, IEEE Transactions On
Power Systems, vol. 23, no. 2, May 2008.
[9] M. Gandomkar, M. Vakilian, M. Ehsan, A combination of genetic
algorithm and simulated annealing for optimal dg allocation in
distribution networks, presented at Electrical and Computer
Engineering, 2005. Canadian Conference on, pp. 645 648, Saskatoon,
Saskatchewan Canada, 2005.
[10] Ma Zhigang, Study on Distribution Network Reconfiguration Based on
Genetic Simulated Annealing Algorithm, in Proc., Electricity
Distribution, 2008. CICED 2008. China International Conference on,
pp. 1-7.
[11] A. Vargas and M.E. Samper, Real-Time Monitoring and Economic
Dispatch of Smart Distribution Grids: High Performance Algorithms for
DMS Applications, under publication on IEEE Transaction on Smart
Grids, Digital Object Identifier 10.1109/TSG.2012.2187078.
[12] S. Paudyaly, C.A. Canizares, K. Bhattacharya, Three-phase distribution
OPF in smart grids: Optimality versus computational burden,
Innovative Smart Grid Technologies (ISGT Europe) 2011, Manchester,
December 5-7, 2011.
[13] F.V. Gomes, S. Carneiro, J.L.R. Pereira, M.P. Vinagre, P.A.N. Garcia,
Leandro Ramos de Araujo, A New Distribution System
Reconfiguration Approach Using Optimum Power Flow and Sensitivity
Analysis for Loss Reduction, IEEE Transactions on Power Systems,
Vol. 21, No. 4, pp. 1616-1623, 2006.
[14] H.M. Khodr, J. Martinez-Crespo, M.A. Matos, J. Pereira, Distribution
Systems Reconfiguration Based on OPF Using Benders
8
Decomposition, IEEE Transactions on Power Delivery, Vol. 24, No. 4,
pp. 2166-2176, 2009.
[15] H.-D. Chiang, R. Jean-Jumeau, Optimal network reconfigurations in
distribution systems. I. A new formulation and a solution methodology,
IEEE Transactions on Power Delivery, Vol. 5, No. 4, pp. 1902-1909,
1990.
[16] H.-D. Chiang, R. Jean-Jumeau, Optimal network reconfigurations in
distribution systems. II. Solution algorithms and numerical results,
IEEE Transactions on Power Delivery, Vol. 5, No. 3, pp. 1568-1574,
1990.
[17] Young-Jae Jeon, Jae-Chul Kim, Jin-O Kim, Joong-Rin Shin, K.Y. Lee,
An efficient simulated annealing algorithm for network reconfiguration
in large-scale distribution systems, IEEE Transactions Power Delivery,
Vol. 17. No. 4, pp. 1070-1078, 2002.
[18] OpenDSS, available at http://sourceforge.net/projects/electricdss/.