Está en la página 1de 6

1

Techniques for High Performance Analysis of Transient Stability

Lei Wang, Fellow IEEE

Abstract Transient stability analysis is one of the basic an important control center application [4].
analyses in the planning, design, and operation of power systems. An on-line DSA system usually includes several analysis
In recent years, one form of this analysis, referred to as on-line modules for assessment of different types of power system
Dynamic Security Assessment (DSA), has become increasingly
security problems. One of such modules is TSA (Transient
popular as a real-time control center application to monitor,
alert, and enhance system security by performing stability Stability Analysis) for examining the transient performance of
analysis using real-time captured system conditions. However, a power system. The solver in TSA needs to deal with the
traditional transient stability analysis requires intensive following functional requirements:
numerical computations and thus usually takes long time to
Contingency screening to identify a small set of
complete. This often makes it a challenge to complete on-line
DSA in required cycling time. Improving the speed of transient potentially critical contingencies from a comprehensive
stability analysis has been the focus of extensive research and list for further detailed analysis. This function is
development for many years. This paper describes a number of primarily for speeding up the computations of TSA.
techniques and methods to improve the speed of transient Detailed contingency analysis with time-domain
stability analysis. Most of these methods have been practically simulations for the set of critical contingencies. This
used with good success to various applications, such as on-line
uses a numerical integration technique to solve the
DSA.
detailed non-linear differential equations describing the
Index Terms Transient stability analysis, on-line dynamic full system dynamics. This includes numerically
security assessment, real-time control center applications. intensive computations and is also the core function of
TSA.
Power transfer limit determination to show actual system
I. INTRODUCTION operation constraints and security margins. This
involves repetitive time-domain simulations so
numerically intensive computations are required. Refer
Secure and reliable operation is a fundamental requirement to Section II for further discussions on this function.
for a power system. In the current era of smart grid Remedial action recommendations if the system will be
development around the world, visibility, controllability, and unstable for a contingency or the security margin is
optimization for system operation is the focus of next insufficient for an operation condition. Such
generation control center applications [1] which target the recommendations are often obtained and/or verified with
high degree of uncertainty with regard to the scenarios and time-domain simulations and therefore this function is
contingencies that need to be examined so as to ensure system also subject to the same computation bottleneck as the
security. This leads to the requirements to evaluate system detailed contingency analysis.
security, calculate system operating limits and interconnection
As shown above, the core function in a TSA solver is the
reliability operating limits [2] based on near real-time
time-domain simulations which can take long time to
conditions. On-line dynamic security assessment (DSA) [3] is
complete for large system models. For this reason, the
an approach that offers promising solution to these challenges.
computation speed of TSA has been the bottleneck for
On-line DSA models the network topology and operating
applications of on-line TSA, particularly for large systems. It
conditions seen by operators in control centers, and it
is recognized that information technology (in both computer
calculates the system security and operational limits using this
hardware and application software) has had tremendous
data. Generally, on-line DSA results more accurately
progresses in the last decades and this has made it possible to
characterize the existing system conditions than the off-line
simulate larger system models, to perform longer simulations,
planning case and thus they allow the operator to evaluate the
and to process more contingencies in TSA. For example, a
impact of certain operating decisions with changing operating
simulation speed improvement of 20% to 30% for TSA is
conditions. Consequently, DSA conducted in a near real-time
observed by using the latest compiler technology.
setting greatly enhances the operational decision making
Nonetheless, to meet the ever increasing demand from on-line
capability of an operator and significantly reduces the risk of
TSA, special techniques are required and this has led to
cascading blackouts by evaluating limits based on real-time
significant research and development which have resulted in
system conditions and topology. As a result of these
some effective and practical methods.
advantages, on-line DSA has become increasingly popular as
This paper describes some of the techniques and methods
that have been developed and adopted for on-line TSA to
The author is with Powertech Labs Inc., Surrey, BC, Canada improve its speed performance. These are presented in two
(lei.wang@powertechlabs.com).

978-1-4673-2729-9/12/$31.00 2012 IEEE


2

categories: ported for on-line TSA. Computation speed is a serious


Advanced analysis methods concern when dealing with such large system models.
Engineering-inspired methods Effective and fast computation techniques must be used to
achieve reasonable performance.
It is clear that to describe all methods ever developed for fast
TSA is an impossible task. This paper describes only those Model complexity
that have been widely used or have shown reasonable Not only power system model sizes are increased, but also
potential in practical applications. New technologies, such as the models are becoming more complex, thus requiring more
cloud computations, that could have significant impact on computational power to process. A typical example is the
TSA performance but are still in development stage, are not gradual adoption of dynamic load models in TSA which
considered in this paper. significantly increases the number of algebraic and differential
equations to be solved. Increased applications of controls
based on power electronics (such as those found in HVDC,
II. NEED AND CHALLENGES FACTS, wind turbines, etc.) also complicate the system
FOR FAST TRANSIENT STABILITY ANALYSIS models. Another dimension of model complexity is due to the
special need in on-line TSA to consider models usually not
included in off-line studies, for example, relay and protection
The pursuit for fast TSA started ever since this is models. Inclusion of these types of models not only makes
recognized as one of the critical problems to be studied for TSA slower, but also presents challenges to some of the fast
power system planning and operation. The computation speed TSA analysis techniques.
for TSA is of more concern for on-line applications as one
computation cycle in on-line TSA usually needs to be Overall transient performance
completed within specified time period, say within 15 to 30 Traditionally, transient stability has been defined as
minutes (some on-line applications, for example power market maintaining synchronism among all generators. This is no
settlement and special protection system lookup table update, longer the only objective for TSA now. Due to various
require even faster computation cycle). This calls for special reasons (for example, integration of renewable generation, use
algorithms and techniques for on-line TSA for the improved of advanced controls, inclusion of detailed models, etc.),
computation efficiency. On the other hand, the TSA problem assessment of other forms of transient performance in a power
has also evolved as power systems continue to grow, and has system becomes necessary for TSA:
become more and more challenge to deal with. Some of the
Transient voltage performance
issues that must be considered when developing fast TSA
Low frequency oscillations
methods include,
Transient frequency performance
System model size
Analysis of such problems has special requirements for
The sizes of power system models for TSA have been TSA. One of these, for example, is the extended simulation
steadily increasing as a result of system expansions. Figure 1 length which is now commonly set to 15-20 seconds. This
shows the sizes of the full eastern US/Canada interconnected demands more efficient computation algorithms.
system planning models from 2007 to 2012. It is seen that in Compliancy with security criteria
five years, the model size is increased from 48,036 buses to
To avoid widespread blackouts such as that occurred in
63,757 buses, or 32.7%.
Northeast US/Canada on August 14, 2003, regulators are
Buses Generators enforcing more strict security criteria for both system planning
70000 62156 63757 and operation, for example, the latest NERC TPL standards
59612
60000
48036
53667 56124
[6]. Compliancy to these standards requires extensive
50000
computation simulations for a large set of contingencies and
40000
30000
fast TSA will certainly help in speeding up the process.
20000
10000
7330 7454 7798 8059 8282 8328 Transfer limit determination
0
2007 2008 2009 2010 2011 2012 As mentioned earlier, TSA does not stop at performing
Year contingency analysis for the real-time system condition.
Often, the final results of TSA are presented as transfer limits
Figure 1: Sizes of the full eastern interconnection models (or stability limits) at key interfaces of the system [7]. This
requires repetitive simulations for all critical contingencies at
The model sizes for on-line TSA have also been different system conditions with different power transfer
consistently on the rise. For example, performing on-line TSA levels. This task is very computationally intensive, and for
using real-time system models of 13,500 buses has been transfer limit analysis in on-line TSA, special fast computation
reported [5], and 40,000-bus real-time models are being algorithm must be used in order to meet the required
3

performance. computation speed of simulations while still producing


accurate results. Many parallel computation schemes have
been proposed for TSA, but unfortunately few have actually
III. ADVANCED ANALYSIS METHODS been used. The main difficulty lies in the fact that an effective
parallel code requires implementation of special algorithms
and such algorithms in most cases do not run efficiently in the
During the past decades, a number of advanced analysis non-parallel mode which is the base for any commercial
methods have been developed for TSA (some were especially software. It is obviously not practical for software developers
made for on-line TSA) with mixed success in applications. to maintain two sets of code.
Here a summary of such methods is given.
Distributed computation
Variable step-size integration methods
Distributed computation can be considered as a crude form
Variable step size integration methods are well understood of parallel computation. Instead of achieving parallelism for a
and they have been implemented in computer code for TSA single simulation, distributed computation parallelizes
[8, 9]. This type of methods is best suited for long-term simulations for multiple contingencies. Since TSA usually
simulations in which integration step size can be made large includes assessment of many contingencies, this technique can
after the initial fast transients are passed. They have also been effectively accomplish significant speed improvement.
successfully used in on-line TSA for which computation speed Distributed computation can be illustrated using the scheme
up to 10 times faster than the fixed step simulations has been in Figure 2. In this scheme, a Client computer is networked
reported [9]. with many Server computers. A client management tool runs
Direct methods in Client, which coordinates the tasks of distributing
contingencies to Servers and collecting results back from
Direct methods refer to a class of methods that assess the
Servers. In each Server, a simulation engine runs which is
transient stability of a power system, and also give a measure
controlled by the Client. The application of Client and Server
of degree of stability (stability margin), based on partial
can be very flexible; they can run in different physical servers
responses of the system obtained from time-domain
or they can all reside in a single multiple-core server to take
simulations. Two main types of such methods have been
full power of such a server. Thus, for a 24-core server, 1 client
developed:
and 23 servers can run simultaneously to allow simulations of
The transient energy function (TEF) methods such as the 23 contingencies in parallel.
BCU method [10]. Such methods are very fast but may
not be able to definitely identify stable and unstable Client

cases. Its main application has been in contingency


screening for which a large amount of contingencies can
be filtered to result in a small set for further detailed
analysis.
EEAC (or SIME) [11]. This method can give very
accurate conclusion on system stability from the time- ......
domain simulation trajectory. However, since full
Server 1 Server 2 Server n
simulations are used, it does not have computation speed
advantage. Figure 2: Distributed computation scheme
It should be pointed out that direct methods not only can
perform fast stability analysis, they also give valuable The advantage of the distributed computation technique is
information on stability margin and sensitivities, which are that very little code changes are necessary for the simulation
very useful in on-line TSA in which identification of system engine. Only additional tools need to be developed to manage
stability for contingencies processed must be done the distribution of contingencies and data communication
automatically. Such information is also useful in deriving between the Client and Server. This makes it possible to use
remedial actions for unstable system conditions to prevent the the same commercial code in both distributed and non-
system from losing stability. distributed mode with minimum amount of work.
Parallel computation The efficiency of distributed computation can be very high.
Parallel computation refers to performing a single Figure 3 shows its performance for a practical case. The
simulation in multiple CPU, CPU cores, or special processors power system model has 6,620 buses and 1,978 generators.
so as to speed up the computation. Parallelism may be 400 contingencies are processed, each of which includes a 10-
achieved for part of the computation [12] or for the entire second simulation. The distributed computation is tested with
simulation [13, 14]. Parallel computation is very attractive in up to 10 Servers. In the figure, the time it takes to run all
speeding up TSA as technically it can improve significantly contingencies in the distributed mode on the specified number
4

of Servers is shown as the percentage of the time taken to run dynamic reduction recently performed for BC Hydro is shown
all contingencies using the non-distributed version of the code here. A full WECC planning model with 18,022 buses and
on the same computer. Thus, it is understandable that with 1,922 generators was used as the base model. The objective of
only one Server, the performance (at 122.5%) of the the reduction is to create a reduced order model that has about
distributed mode is slower than the non-distributed mode due 7,000 buses which would include the full BC Hydro system as
to the additional data communication requirements. However, the study area. A coherency-based reduction was performed
with 10 Servers, the performance reaches an impressive using the EPRI DYNRED program and the resulting model
10.7%, implying that the overall computation speed is reduced has 7,210 buses (40.0% of the base model) and 1,040
almost linearly with the number of Servers used. generators (54.1% of the base model). The time that it takes to
130.0%
perform a simulation with the reduced model is 52.2% of the
120.0% time required for the full model. Figure 4 shows the accuracy
110.0%
100.0% of the reduced model for a critical contingency, by comparing
two models which are dispatched in the same way to increase
90.0%
Speed (%)

80.0%
70.0%
60.0%
the BC Hydro export to US from 233 MW to 2,957 MW
50.0% (close to the rated maximum export capability). It is seen that
40.0%
30.0% the reduced model has acceptable performance at such an
20.0%
10.0%
extreme system condition.
0.0%
1 2 3 4 5 6 7 8 9 10 Use of the dynamically reduced models for on-line TSA
No. of servers
may be required for other reasons. Since on-line TSA is
Figure 3: Performance of distributed computations performed for real-time system conditions, the system
conditions external to the study area may not be available
from SCADA. In such a case, the external system will need to
Distributed computation technique has been successfully
be represented by an equivalent. Time-saving for TSA will be
used in practical on-line TSA systems, for example the system
the by-product with this approach.
reported in [5].

IV. ENGINEERING-INSPIRED METHODS

In addition to advanced computation algorithms, another


class of methods to improve TSA speed is originated from
engineering practices. The following describes some of them.
Early termination method
This is an intuitive approach in which conventional
simulation is performed up to a time point and a check is then
made periodically afterwards at a specified interval, based on
a combination of stability indices obtained from the
simulation. Very stable or unstable contingencies are Figure 4: Comparison of full and reduced models
terminated as soon as the contingency status can be reliably
determined. For contingencies whose stability status cannot be Model tuning
reliably determined prior to the completion of the simulations, Some of the large interconnected power system models
full simulations will be performed. Using this method, PJM include dynamic model parameters that demand small
reported [5] that simulations of 60% to 90% of contingencies integration step size in order to maintain numerical stability
in its on-line TSA system can be terminated before reaching when using the designated simulation tool. For example, the
the end of the simulations, thus achieving significant saving in current US/Canada eastern interconnection models require an
terms of computation time. integration step size of cycles (or smaller). The simulation
can obviously be made faster if a larger step size can be used,
Dynamic equivalencing
while the result accuracy and numerical stability are
An obvious method to reduce TSA computation time for a maintained. This can be achieved by tuning the critical
large system model is to use a reduced order equivalent parameters to eliminate the source of the numerical stability
model. The key to this method is to create a high-fidelity problem.
reduced order model to replace the full model. There is rich An example is presented here to show how effective such a
literature on this subject [15] and commercial tools are method could be. A full US/Canada eastern interconnection
available to accomplish such a task. model is considered, which has 45,000 buses and 210,000
As an example to illustrate the application of this method, a parameters in its dynamic models. By changing the values of
5

only 140 parameters ranging from synchronous machine as references when selecting an appropriate method in an on-
reactances and time constants to exciter/AVR time constants, line TSA project.
the integration step size can be safely increased from cycles
to cycles resulting in a simulation time reduction of 38%.
Figure 5 shows the comparison of a typical simulation VI. ACKNOWLEDGEMENT
performed with two step sizes ( and cycles). Only one
curve can be seen in the figure since the results are practically
the same for both cases. The author sincerely thanks his colleagues Mr. Fred Howell,
Dr. Xi Lin, and Dr. Carl Wang for providing valuable
information used in this paper.

VII. REFERENCES

[1] Pei Zhang, Fangxing Li, and Navin Bhatt, Next-Generation


Monitoring, Analysis, and Control for the Future Smart Control
Center, IEEE Trans. on Smart Grid, Vol. 1, No. 2, Sept 2010,
pp. 186-192.
Figure 5: Comparison of simulations at different step sizes [2] Standard IRO-004-0 Reliability Coordination Operations
Planning - ftp://www.nerc.com/pub/sys/all_updl/standards/rs/
Lookup table approach IRO-004-0.pdf
An approach for determining stability limits in on-line TSA [3] Lei Wang and Kip Morison, Implementation of New Tools for
On-line Security Assessment Tools For Reducing The Risk Of
uses the so-called lookup tables, instead of performing
Blackouts, IEEE Power & Energy Magazine,
simulations using the real-time system conditions [16]. In this September/October, 2006.
approach, stability limits are computed using off-line [4] CIGRE Technical Brochure on Review of On-Line Dynamic
simulations for possible system operating conditions. These Security Assessment Tools and Techniques, CIGRE Working
results are entered in a database (referred to as the lookup Group C4.601, June 2007.
tables). In on-line mode, the system conditions available in the [5] Jianzhong Tong and Lei Wang, Design of a DSA Tool for Real
Time System Operations, Powercon 2006, Chongqing, China,
lookup tables are compared with the real-time system
October 22-26, 2006.
condition and the closest match is identified. The limits for [6] NERC, Reliability standards for transmission planning TPL-001
this condition in the lookup tables are then used as the limits to TPL-006, available from www.nerc.com, 2009.
for the on-line system condition. This approach surely is one [7] IEEE Special Publication, Techniques for Power System
of the fastest to determine stability limits in on-line mode as Stability Limit Search, IEEE Catalog Number 99TP138, 1999.
no simulations are performed for the real-time conditions. In [8] J.Y. Astic, M. Jerosolimski, and A. Bihain, The Mixed Adams-
BDF Variable Step Size Algorithm to Simulate Transient and
practical applications, however, it may be subject to two
Long Term Phenomena in Power Systems, IEEE Transactions
shortcomings: on Power Systems, Vol.9, N 2, May 1994.
There are possibilities that a real-time system condition [9] Neto, C.A.S., Quadros, M.A., Santos, M.G., and Jardim, J.,
cannot be matched with the conditions available in the Brazilian system operator online security assessment system, a
paper presented at a panel session at the IEEE PES General
lookup tables. In such a situation, stability limits will not
Meeting, Minneapolis, MN, July 2010.
be available which would restrict the real-time system [10] Chiang, H.D., Tong, J., and Tada, Y., On-line transient stability
operation. screening of 14,000-bus models using TEPCO-BCU:
Stability limits computed from off-line models are often Evaluations and methods, a paper presented at a panel session at
the IEEE PES General Meeting, Minneapolis, MN, July 2010,
more conservative than those computed directly from
pp. 18.
the real-time model. This would result in lower transfer [11] Xue, Y., Van Custem, T., and Ribbens-Pavella, M., Extended
capabilities. equal area criterion justifications, generalizations, applications,
IEEE Transactions on Power Systems, 4(1), 4452, February
1989.
V. CONCLUSIONS [12] M.A. Tomim, J.R. Mart, and L. Wang, Parallel solution of
large power system networks using the Multi-Area Thvenin
Equivalents (MATE) algorithm, International Journal of
This paper discusses methods and techniques for improving Electrical Power & Energy Systems, Volume 31, Issue 9 (2009),
computation speed of transient stability analysis with Pages 497-503.
[13] V. Jalili-Marandi and V. Dinavahi, SIMD-based large-scale
application to the on-line TSA. These methods are transient stability simulation on the graphics processing unit,
summarized with their application potential to on-line TSA. IEEE Trans. Power Syst., vol. 25, no. 3, pp. 1589-1599, Aug.
Application examples are presented for some of the methods 2010.
to show the benefits. These discussions are intended to serve
6

[14] Zhenyu Huang, Ross Guttromson, Jarek Nieplocha, and Rob


Pratt, Transforming Power Grid Operations, Scientific
Computing, 24(5): 22-27, April 2007.
[15] L. Wang, M. Klein, S. Yirga, P. Kundur, "Dynamic Reduction
of Large Power Systems For Stability Studies", IEEE Trans.
Vol. PWRS-12, No. 2, pp.889-895, May 1997.
[16] L. Loud, S. Guillon, G. Vanier, J.A. Huang, L. Riverin, D.
Lefebvre, and J-C. Rizzi, Hydro-Qubecs challenges and
experiences in on-line DSA applications, a paper presented at a
panel session at the IEEE PES general meeting, Minneapolis,
July 2010.

VIII. BIOGRAPHIES

Lei Wang is the Manager of Software Technologies at Powertech


Labs Inc., Surrey, BC, Canada. He is a Fellow of IEEE.

También podría gustarte