Beruflich Dokumente
Kultur Dokumente
I. INTRODUCTION
0885-8950 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
2
state-of-the-art, this paper proposes an integrated heuristic optimization framework for simultaneous determination of optimal
allocation and sizing of dynamic Var sources (ODynVarS).
Considering multiple pre-selected contingencies in the decision making, mean-variance mapping optimization (MVMO)
is used as the only searching algorithm in conjunction with
intervention schemes for fitness evaluation and knowledge
management to achieve high computational efficiency. Dynamic solver and optimization technique are independent and
only loosely coupled, which makes possible the individual
inclusion of high-performance computing paradigms while
principally relying on off-the-shelf software.
The paper is structured as follows. Section II discusses preliminary considerations to be followed by the detailed presentation of the ODynVarS approach in Section III. A test case study
is performed in Section IV, and finally conclusions are drawn
with an outlook for future research in Section V.
II. PRELIMINARY CONSIDERATIONS
ODynVarS focuses on short-term dynamic performance enhancement of power systems subject to credible, severe disturbances through the addition of one or more dynamic Var
source(s). The actual decision about which type to incorporate
(e.g., STATCOM vs. SVC) is not a limiting factor since the approach is broad enough to consider any fast-acting device influencing the transient voltage recovery without loss of generality.
It is treated as an optimization task with the goal of achieving a
substantiated techno-economic tradeoff.
A. Assessment of Short-Term Dynamic Performance
Post-disturbance voltage requirements are commonly formulated by means of performance criteria in grid codes [9]. As
shown in Fig. 1, generic lower and upper reference boundaries
are defined, to where the associated bus voltage trajectory is restricted:
(1)
The lower part is described by expression (1) while
corresponds to the upper part.
and
denote the instant of fault occurrence and dynamic
simulation time span. Parameter
is used to calibrate the
decaying of the boundaries towards the steady-state value
. In this research,
p.u. and
p.u. are
used except in Section III-B, where
is varied within the
range
. As illustrated in Fig. 1, the area outside
the boundaries, defined as trajectory violation integral (TVI),
characterizes the individual voltage violations with respect to
magnitude and time duration:
(2)
(3)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
WILDENHUES et al.: OPTIMAL ALLOCATION AND SIZING OF DYNAMIC VAR SOURCES USING HEURISTIC OPTIMIZATION
a natural characterization of the limiting boundaries. This inherently enables accurate quantification of the degree of severity
compared to composite approaches, where multiple counting of
components cannot be excluded and only sustained violations
are processed [3], [11].
Equation (5) is used for calculation of the individual contingency severity index (CSI). Moreover, other definitions, e.g., to
emphasize the relevance of load buses, may be introduced:
(5)
Contingencies are defined to be single-element three-phase
short-circuits in the middle of transmission lines. The contingencies are ranked according to CSI in a descending order. From
this ranking, the input set for consideration in ODynVarS is then
established after selecting the first five elements. Note that alternative contingency selection procedures based on CSI can be
used instead.
C. Candidate Locations Specification
Determination of candidate locations for potential dynamic
Var source installation constitutes another essential preliminary
step. In connection with steady-state frameworks, this can be
accomplished through (mature) sensitivity analysis techniques,
e.g., based on a reduced power flow Q-U-Jacobian matrix [13].
Nevertheless, a more comprehensive approach should be used
when considering the nonlinear power system short-term dynamics in response to large disturbances.
A basic approach would be to perform the selection by measuring the marginal impact of a control variable (corresponding
to a dynamic Var source addition) on relevant system quantities
(e.g., bus voltages) in the time domain. This is referred to in literature as trajectory sensitivity analysis [14]. Generally, the determination of trajectory sensitivities entails a significant computational burden, especially for large-scale systems. Numerical
approximations have hence been adopted to perform the selection process more efficiently [3]. It has not been proved to date
whether these measures can provide a reliable indication of sensitivities when considering abrupt variations along the voltage
trajectory, such as those involved in cases where transient instability could fortuitously play a complementary role under
stressed operating conditions.
In light of this aspect, an alternative engineering-based procedure, which further utilizes the outcomes from the contingency
assessment described in Section II-B, is used for candidate location specification. Considering the possibility of a 3-phase disturbance occurring along system branches, and after selecting a
given number of credible contingencies, the buses for potential
dynamic Var sources addition are defined as being the buses or
nearby buses corresponding to the branches where the contingencies occur.
The subsequent task is to identify minimum and maximum
rating
at individual candidate buses as component search range for the optimization. Technically, it is reasonable to define
in terms of the local system strength,
which is obtained through
, where
results from a
short-circuit calculation [15]. Short-circuit ratio SCR is typically defined in planning studies to reflect the relationship between short-circuit power and plant size in order to analyze, for
instance, the response of wind power plants to voltage dips in
systems of varying strength [16]. This value has been set to 20
for this research. In contrast to
cannot be determined
in a similar manner. Instead, this limit may be predetermined
by manufacturer rating specifications.
is set equal to 25
Mvar. It can in conclusion be stated thus that while minimum
physical rating
used in the optimization as the lower search
boundary is fixed,
varies at individual candidate buses in
relation to the local system strength.
A functional relationship between investment cost and device rating is further needed to quantify the allocation of individual dynamic Var source(s). In this research, reference per
unit STATCOM investment costs, reflecting expenditures both
for equipment and infrastructure have been obtained from [5],
and extrapolated for different ranges comprising any conceivable Mvar ratings. The data was fitted to a polynomial cost function, which is used to evaluate the economic viability of a candidate solution. In the formulation of the ODynVarS problem, the
overall Var investment cost is determined using the reference
cost function,
, for every STATCOM installation having
rating among
candidate locations as
(6)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
4
unique feature of MVMO is its incorporation of a special mapping function for mutating genes in the offspring generation.
Its dynamically changing shape is based on mean and variance
from the set comprising the best solutions achieved so far,
which in turn are archived in a descending order of fitness.
The mapping function is defined within the interval
and guarantees generated offspring to be located strictly within
the search range, e.g., minimum and maximum device ratings.
For function evaluation, optimization variables are de-normalized to real-world quantities. Original MVMO is based on a
single-agent concept requiring only one evaluation per generation [17]. This is beneficial from the computational viewpoint
[19]. A generalized and more global variant is presented in [18].
It offers the possibility of further elaborating the computational
aspects towards large systems by exploiting natural concurrency
provided by a swarm of multiple agents. It is assumed that in the
given time-domain framework, certain inter-agent communication via message passing is acceptable while achieving significant speedup.
B. Multi-Contingency Framework Incorporating Intervention
Proper treatment of ODynVarS under consideration of multiple contingencies in a power system dynamic framework involves key aspects encouraging the inclusion of intervention
schemes into the optimization. It should be stressed that for candidate STATCOM allocation , created through MVMOs offspring generation process, the technical validation could entail
execution of
consecutive time domain simulations for determination of
, cf. (12). Ingenuous implementation would
imply a tremendous computational burden for large-scale power
systems. Thus, considering that a limited net amount of function evaluations is vital to avoid computational bottlenecks, the
twofold intervention scheme illustrated in Fig. 4 is incorporated
into the multi-contingency framework.
Of individual candidate dynamic Var allocation proposed
by MVMO, the procedure evaluates 1) the economic potential
(cf. Intervention I) prior to and 2) the technical feasibility (cf.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
WILDENHUES et al.: OPTIMAL ALLOCATION AND SIZING OF DYNAMIC VAR SOURCES USING HEURISTIC OPTIMIZATION
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
6
Fig. 6. IEEE New England 39-bus test system with candidate buses (bold bars)
and locations of the five most severe N-1 line contingencies (arrows).
Fig. 7. Mean of bus voltages (top) and ranked CSI (bottom) corresponding to
is equal to 0.95 p.u.
N-1 transmission line contingencies (numbered).
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
WILDENHUES et al.: OPTIMAL ALLOCATION AND SIZING OF DYNAMIC VAR SOURCES USING HEURISTIC OPTIMIZATION
Fig.
tions,
6
indicates
which
have
the
corresponding
maximum
possible
locarating
Mvar.
is set to 25
Mvar. In this research, any proposed STATCOM device
operates under a proportional continuous local voltage control
regime.
Fig. 8 shows the fitness convergence with investment costs
to be minimized in (7), and highlights coherent behavior for
10 independent optimization trials. A high level of reliability
can thus be guaranteed concerning the optimal dynamic Var
sources allocation task. Fig. 9 depicts the convergence of best
STATCOM allocation of the best trial. Random initialization of
involves following diverse search paths in the initial stage,
but after 550 function evaluations (FE), i.e., time domain simulations, the individual best solutions reach near-optimality in a
coherent manner. According to the evolution of the individual
components of , representing normalized ratings of the best installations attained so far, the effect of IMIS can clearly be identified. After locating the most promising candidate location(s)
by prior rejection one after another due to high cost participations, the strategy encourages the choice of a single large device
rating at bus 21.
Fig. 10 illustrates major implications of the expected computational time benefit accruing from the twofold intervention
scheme introduced in Section III-B. By the recurrent evaluation
of 1) the economic potential before and 2) the technical feasibility after entering the multi-contingency framework, ODynVarS obviously prevents the execution of a substantial amount
of FEs in time domain, and hence makes beneficial use of the
computational resources. In the initial stage, the search focuses
on exploration, and in this way, localizes the most promising
candidate locations. Several STATCOM devices are dispersed
over local problem areas of interest for short-term dynamic performance, which help to easily avoid violation of the mandated
criterion, that is: constraint (12) is satisfied. As the evolutionary
search progresses, the strategy increasingly intervenes to circumvent unnecessary FEs needed to validate numerous infeasible allocations, accompanied predominantly by few devices
seeking to touch the (lower) trajectory boundary by fast-acting
voltage support. Thus, the cumulative expected computational
time saved is shown in the lower part of Fig. 10. It highlights
the benefit that can be attributed to saved FEs for cost-attractive
allocations being 1) inferior, cf. Intervention I, and 2) superior
but infeasible due to any still critical contingency from the set,
cf. Intervention II. See Figs. 8 and 9 for further reference to this
study case.
Assuming identical load conditions, ODynVarS was tested
considering fault clearing time
ms, which implies
an even higher operational risk. Fig. 11 illustrates how the
combined activity of intervention and IMIS guides the search
towards identifying two devices for fulfillment of (12). Using
circles and crosses, the complexity and variability with which
different contingencies may restrict the evolutionary search
progress towards the optimal allocation, become visible.
C. Investment Cost Sensitivities
Fig. 12 visualizes the sensitivity of overall final STATCOM
investments with respect to the dynamic load percentage
and the strictness of the performance criterion described by ,
cf. (1). It was assumed
ms accounts for potential collateral loss of synchronism of various generators when there are
limited fast-acting remedial measures under stressed conditions,
i.e., heavily depressed transient voltages.
Minimum and maximum refer to the best and worst achieved
numerical outcomes of 5 independent optimization trials.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
8
an integrated mixed-integer search strategy and two intervention schemes, the search is guided towards techno-economic efficiency of dynamic Var investments while making rational use
of the computational resources. Numerical experiments highlight the efficiency of the approach.
The control scheme applied to individual dynamic Var
sources has a decisive impact on the numerical outcome as
sensitivity studies have revealed. For instance, specifying the
voltage control droop as an (additional) optimization variable
would be desirable from the generality point of view. This
however constitutes a highly complex and computationally expensive task which is further tightened by the need to consider
multiple dispatch scenarios in the decision making. Future
research could be directed towards the smart incorporation of
control parameters and system scenarios while additionally
taking into account risk-based aspects into both the assessment and enhancement of power system short-term dynamic
performance. This involves large computational effort but can
become reality by exploiting todays computing power and the
intrinsic parallelism provided by modern heuristic optimization
techniques and contingency analysis.
REFERENCES
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
WILDENHUES et al.: OPTIMAL ALLOCATION AND SIZING OF DYNAMIC VAR SOURCES USING HEURISTIC OPTIMIZATION
[16] T. Neumann and C. Feltes, Response of DFG-based wind farms operating on weak grids to voltage sags, in Proc. IEEE PES General
Meeting, Jul. 2011.
[17] I. Erlich, G. K. Venayagamoorthy, and W. Nakawiro, A mean-variance optimization algorithm, in Proc. IEEE World Congr. Computational Intelligence, Barcelona, Spain, 2010.
[18] J. L. Rueda and I. Erlich, Hybrid mean-variance mapping optimization
for solving the IEEE-CEC 2013 competition problems, in Proc. IEEE
World Congr. Evolutionary Computation, Cancun, Mexico, June 2013.
[19] I. Erlich, J. L. Rueda, and S. Wildenhues, Solving the IEEE-CEC
2014 expensive optimization test problems by using single-particle
MVMO, in Proc. IEEE World Congr. Evolutionary Computation,
Peking, China, Jul. 2014, competition winners.
[20] S. K. Khaitan and J. D. McCalley, High Performance Computing for
Power System Dynamic Simulation, in High Performance Computing
in Power and Energy Systems. New York, NY, USA: Springer, 2013.
[21] J. F. Benders, Partitioning procedure for solving mixed variables programming problems, Numer. Math., vol. 4, pp. 238252, 1962.
[22] A. M. Geoffrion, Generalized benders decomposition, J. Optimiz.
Theory Applicat., vol. 10, no. 4, pp. 237260, 1972.
[23] F. Capitanescu and L. Wehenkel, State-of-the-art, challenges, and future trends in security-constrained optimal power flow, Elect. Power
Syst. Res., vol. 81, pp. 17311741, May 2011.
[24] M. A. Pai, Energy Function Analysis for Power System Stability.
Norwell, MA, USA: Kluwer, 1989.
[25] IEEE Task Force on Load Representation, Standard load models for
power flow and dynamic performance simulation, IEEE Trans. Power
Syst., vol. 10, no. 3, pp. 13021313, Aug. 1995.
Sebastian Wildenhues (S14) was born in 1987. He received the B.Eng. degree
in electrical power engineering from the University of Applied Sciences Dortmund, Germany, in 2010 and the M.Sc. degree in electrical power engineering
from the University of Duisburg-Essen, Germany, in 2013.
He is currently with Fraunhofer Institute for Wind Energy and System Technology, Kassel, Germany. His research interests include the stability, control,
and simulation of power systems considering uncertainties.
Jose L. Rueda (M07SM12) was born in 1980. He received the Electrical Engineer diploma from the Escuela Politcnica Nacional, Quito, Ecuador, in 2004,
and the Ph.D. degree in electrical engineering from the Universidad Nacional
de San Juan, San Juan, Argentina, in 2009.
From September 2003 until February 2005, he worked in Ecuador, in the
fields of industrial control systems and electrical distribution networks operation and planning. From 2010 to 2014 he was working as a research associate at
the Institute of Electrical Power Systems, University of Duisburg-Essen, Germany. Currently, he is an Assistant Professor for Intelligent Electrical Power
Grids at the Department of Electrical Sustainable Energy, Technical University
Delft, The Netherlands. His research interests include power system stability
and control, system identification, power system planning, and probabilistic and
artificial intelligence methods.
Istvan Erlich (SM08) was born in 1953. He received the Dipl.-Ing. degree
in electrical engineering and the Ph.D. degree from the University of Dresden,
Dresden, Germany, in 1976 and 1983, respectively.
From 1979 to 1991, he was with the Department of Electrical Power Systems
of the University of Dresden. In the period of 1991 to 1998, he worked with the
consulting company EAB, Berlin, Germany, and the Fraunhofer Institute IITB
Dresden. During this time, he also had a teaching assignment at the University of
Dresden. Since 1998, he has been a Professor and head of the Institute of Electrical Power Systems at the University of Duisburg-Essen, Duisburg. Germany.
His major scientific interest comprises power system stability and control, modeling, and simulation of power system dynamics, including intelligent system
applications.
Dr. Erlich is a member of VDE and the chairman of the International Federation of Automatic Control (IFAC) Technical Committee on Power Plants and
Power Systems.