You are on page 1of 5

On-Line Dynamic Security Assessment

Using Intelligent Systems

Kip Morison

encountered, a very large set of system conditions

must be analyzed. In addition, despite best efforts, a
specific real-time operating condition encountered
may not have been captured by forecast. This is
particularly true when the system is in a critical state
such as may be encountered prior to, or during, a
cascading outage sequence.

Abstract: Power system security assessment has traditionally

been conducted using analytical techniques guided by human
control. The conditions to be studied are forecast and
deterministic studies conducted to establish the security and
determine the need for remedial actions; humans normally
play a significant role in the formulation of the input to the
studies and also in the interpretation of the output.
For security assessment in operations, on-line security
assessment in which a snapshot of the system is captured and
analytical engines are used to assess security in near-real-time
has eliminated some of the uncertainty previously introduced
by forecasting. However, it is still desirable to develop new
technologies using artificial intelligence and intelligent
control systems which can have the benefits of analytical
analysis as well as exhibit human-like adaptability and
decision making capabilities to assist in evaluating system
security and providing control decisions.

A key element of DSA is the assessment of stability;

the ability of an electrical power system, for a given
operating condition, to regain a state of operating
equilibrium after being subjected to a physical
disturbance, with most system variables bounded so
that practically the entire system remains intact[1].
evaluation of stability requires rigorous analysis,
including the assessment of power system dynamic
performance. This requires significant computation
effort in order to analyze all the forecast conditions
and possible contingencies.

If the off-line approach is unable to provide an

adequate assessment, the operator is required to judge
the system condition and take appropriate actions. In
the deregulated environment, system condition may
vary widely and this, combined with outages which
may exist, may make it very difficult for a human
operator to act with sufficient confidence.

Power system dynamic security assessment (DSA) is the
process of determining the degree of risk in a power
systems ability to survive imminent disturbances
(contingencies) without interruption to customer service
[1]. A power systems security depends on the system
operating condition as well as the contingent probability of
disturbances, and therefore, DSA must consider these
factors. Power system DSA has traditionally been
conducted off-line using a variety of analytical techniques
and the results used by human operators to guide real-time
operation. To achieve this, the conditions to be studied are
forecast and deterministic studies conducted to establish
the security and determine the need for remedial actions;
humans normally play a significant role in the formulation
of the input to the studies and also in the interpretation of
the output. This approach has a number of fundamental
shortcomings which, in todays evolving power systems,
often render it inadequate.

The method relies on forecasted conditions; todays

systems are often highly unpredictable and forecast
conditions may contain significant errors. To predict
all possible operating condition which could be

G.K. Morison is with Powertech Labs Inc., Surrey, BC, Canada

1-4244-0493-2/06/$20.00 2006 IEEE.

To deal with these issues, on-line DSA has emerged to

reduce, or eliminate entirely, the need for off-line analysis.
In this approach, a snapshot of the real-time system is
captured and, using a variety of methods, the security is
assessed for a large set of probable contingencies and
transactions. While many analytical methods have been
proposed as suitable for on-line DSA, because of the
complex non-linear nature of power systems, only a
handful have proven to be practical and have reached the
stage of actual on-line implementation. Most power
system operators have recognized the need for, and value
of, on-line DSA systems and are implementing them as
add-ons to existing energy management systems (EMS) or
as a requirement in the procurement of upgraded or new
EMS systems. As a result, numerous on-line systems are
being placed in-service world-wide, examples include
[2,3]. Despite the sophistication of state-of-the-art on-line
DSA systems there remains a tremendous opportunity for
the integration of intelligent systems (IS) with these
applications. It is anticipated that the use of IS can greatly
improve the efficiency, reliability, and robustness of online systems.


The conceptual requirements for an on-line DSA system
are described in some detail in [4]. Although not yet
completely fulfilling this comprehensive requirement,
state-of-the-art on-line DSA implementations currently
incorporate many of the components described this
conceptual design. Figure 1 shows the structure of many
leading-edge on-line DSA systems.

Figure 1 : Components of State-of-the-Art On-Line DSA

The primary data source remains the SCADA system, from

which data is input to the EMS state-estimator to produce a
base model for use in DSA. Although the computations
required in DSA are indeed challenging, the model
assimilation component of the DSA system is,
interestingly, perhaps the weakest, and therefore most
challenging element. There are three reasons for this,

Since a powerflow model forms the basis for most

DSA analyses, the basic powerflow model produced
by a state-estimator must be well conditioned and
robust. Inaccurate (or insufficient) SCADA data, or
poorly tested state-estimator algorithms, often lead to
a poor powerflow solution.

The geographic region of the power system that is

observable using the SCADA system is limited, and,
therefore, some assumptions must generally be made
regarding the external system (those portions of the
power system beyond the immediate study area).
Currently, the most common approach is to develop
external equivalents which can be pasted onto the
powerflow model produced by the state-estimator in
order to form a complete powerflow model of
sufficient size and adequate performance for studies.

The powerflow model must be combined with

dynamic data to produce a useable simulation model
for dynamic analysis such as time-domain or
eigenvalue analysis. This is often a challenge

because (a) the bus numbers used in the models

produced by the state-estimator may vary with each
snapshot taken from the SCADA, and (b) the
powerflow/dynamic data set must be consistent (for
example, generator outputs in the powerflow must be
within limits specified in the dynamic data) so the
simulation will initialize properly.
Many on-line DSA systems are required to assess voltage
security as well as transient security. While the former
may generally be accomplished using powerflow methods
(such as PV analysis), the latter requires dynamic
simulation. To reduce a very large number of possible
contingencies down to a manageable number of potentially
critical contingencies, some specialized screening
methods may be used. These may be based on direct
methods or simplified analysis methods. However,
comprehensive assessment of transient security requires
full time-domain simulations in order to faithfully capture
the dynamic response of the overall power system
including the response of generator, controls, loads,
protections, and transmission devices. Since the cycle
time (period at which the DSA must be run) is typically
quite short (15 minutes or less), for large power systems in
which many contingencies must be analyzed, significant
computer power may be required. Using distributed
computing, commercial DSA tools are completely scalable
and additional CPUs can be added to achieve the desired
performance. A typical architecture which permits this for
an on-line implementation using distributed computation is
shown in Figure 2.

Figure 2: Typical Architecture for On-Line DSA


A. Advantages of Intelligent Systems
There are significant opportunities for the introduction of
intelligent system into the type of on-line DSA system
described above which we may refer to as deterministic
systems since they rely largely on enumerated analytical
solutions. Intelligent systems are seen to have three
features which can bring benefits to the real-time


Intelligent systems can be very fast. Although

distributed computation is now commonplace, full
simulation methods require minutes of time to reach a
conclusion. For large power systems in which many
contingencies must be assessed, even with multipleCPU computing, this time may be of concern and for
on-line analysis, time is critical; particularly if a
system is entering an insecure state and decisions must
be made quickly. Once developed, IS technologies
such as neural networks (NN) or decision trees (DT)
can provide solutions very quickly.
Intelligent systems are learning systems.
Deterministic systems will conduct the same
computations every cycle even if some of the
calculations could be deemed inconsequential or if
conditions arise rendering the computations less
accurate. IS systems have the ability to establish if a
system condition has been seen previously and predict
the solution accordingly. Similarly, if properly
designed, an IS can adapt to new conditions by
learning from situations previously seen.
Intelligent system can provide a high degree of
discovery. Discovery refers to the ability to uncover
salient, but previously unknown, characteristics of, or
relationships in, a system. For example, through the
development of NN or DT, the system parameters
which are crucial to security can be established, much
in the same way as an operator would discover such
traits over extended periods of time by observing the
system behavior.

B. The Role of Intelligent Systems

The role of intelligent systems is seen as one that is
complementary to the so-called deterministic DSA
approach. The concept is that an IS will supervise the
DSA and perform the following functions,

Establish if the system security can be established

without full simulations
determine what full simulations may be required
assist in the interpretation of the simulation results and
provide insight into system critical parameters
learn from the results of simulations
assist in the specification of remedial actions

Based on the above functions, the IS provides supervision

to the existing on-line DSA capabilities, much in the same
way that a human operator could if sufficiently
experienced and fast.

Data Builder

Intelligent System Builder

Scenario Builder
Create Scenarios to be
simulated for inclusion in

DB Interrogator
Analyze DB and select data to be
used in Building Intelligent System

Engineered Attribute Selector

Select attributes to be retained
in simulation output

Reduced Database of
scenarios and their
associated stability

Run full time-domain

simulations using existing DSA

IS Function Builder
Build Intelligent System

Powertechs DSA

Full Database of
scenarios and their
associated stability

Commercial Data
Mining Software

Decision Trees

Figure 3: Structure of the POSSIT Environment


A. On-Line Transient Stability Assessment
One example of an IS technology developed for on-line
use is the POSSIT project [5]. Completed in 2004, a
software environment was developed which can use
intelligent systems together with existing dynamic security
assessment software to provide very fast on-line transient
stability assessment of power system transient stability
limits. The approach uses Decision Trees (DT) and
Regression Trees (RT) as the primary component in the IS.
This approach was chosen due to the maturity of DT and
RT tools and their natural and simple fit to the power
system problem. The concept is to develop DT onto
which any given system condition (i.e snapshot in the form
of a solved powerflow from the state estimator) can be
dropped and the outcome of the tree used to establish the
security of the system. The POSSIT environment allows
the trees to be generated quickly and updated as new
unseen system conditions occur. In this sense the
system is learning system. Also, as noted earlier, the
branches of the trees provide valuable information
regarding the critical system parameters for which the
system stability is most sensitive. The structure of the IS is
shown in Figure 3.
To initially develop the decision trees, a very large Data
Base (DB) is created, using the Scenario Generator tool,
which consists of a large number of objects, each
representing a power system pre-contingency state
(effectively, each object is a set of key attributes from a
powerflow solution) together with the results (stable or
unstable) of a transient stability simulation for a
contingency as shown in Figure 4.

correlation methods in the DB Interrogator. The reduced

database is then used as input to DT Builder tools which
can build decision trees (DT). The POSSIT design is
intended to be general such that different decision tree
building tools could be used.
Once the trees are developed, they are placed in the on-line
architecture shown in Figure 4. When a snapshot is taken
from the real-time system, the full on-line DSA engine is
started as usual. In parallel, the DB is searched using a
nearest neighbor (KNN) routine to ensure that the snapshot
is within the scope of the DB; if it is not, the DT path is
not used and conventional DSA is used. The snapshot is
then dropped on the decision tree and the stability
condition is provided virtually instantaneously. As the full
DSA solution completes for all contingencies, the DB is
appended, and new DT can be built at anytime. This
process is critical to ensure the system learns over time.

Figure 5: Database Structure

The stability condition reported in the DB is established

using a full time-domain simulation. Prior to running the
time-domain simulations, the Engineered Attribute
Selector may be used to specify which output quantities
(such as bus voltages or generator output powers) should
be eliminated or retained. This assists in reducing the
output database size.
The large database can be interrogated and reduced to
eliminate redundant data using special clustering and

It is clear that this IS should perform very well when

system conditions are within, or close to, the expected
bounds of operation. The advantage is that a security
indication is provide very quickly. If a condition arises
that is well beyond criteria (i.e loss of multiple right-ofways), it may be difficult for the IS to accurately predict
the system stability due to lack of prior knowledge unless
the IS was trained using such conditions. In such an event,
the conventional DSA engine provides the accurate result.
Work is currently underway to integrate the POSSIT
technology into commercial DSA products.
It is important to appreciate that although the approach

Figure 4: Integration of IS into on-line DSA System

described above was used for the assessment of transient

stability, it can be readily extended to other applications
provided the problem can be formulated to conform with
the database structure shown in Figure 4. That is, each
condition analyzed (object) can be defined in terms of a
set of initial conditions (attributes) and a result
(outcome). If the outcome is a discrete binary result
(such as stable/unstable) then decision trees are suitable. If
the outcome is continuous (such as a stability limit or
voltage) then regression trees must be used.
The POSSIT approach has also proven to be very useful in
the off-line mode such as encountered in system planning.
In this application the main value of the method is in the
level of discovery it provides in analyzing complex power
systems. It is often difficult to establish what factors most
significantly influence a systems stability; this is
particularly true for large complex systems. Using the
process described above, a large number of simulation s
can be loaded into a database from which a decision tree
can be developed. The branches of the tree provide good
insight into the key system parameters which most heavily
influence the system security. These may include such
attributes as line loadings, generator outputs or angles, bus
voltages, or load magnitude.

An overview of state-of-the-art of on-line DSA has been
presented and the opportunities for application of
intelligent systems have been discussed. Intelligent
systems hold promise to improve DSA speed, provide
adaptive learning capabilities and offer the ability to
identify key system parameters. To date, efforts have been
made to use intelligent systems in a supervisory rile in the
on-line assessment of transient stability and these concepts
are being extended for voltage stability.

Definitions and Classification of Power System

Stability IEEE/CIGRE Joint Task Force on Stability
Terms and Definitions, IEEE Transaction s on Power
Systems Vol. 19, No. 2., May 2004.


Rene Avila-Rosales, A. Sadjadpour, M. Gibescu K.

Morison, H. Hamadani, L. Wang, ERCOTs
Implementation of On-Line Security Assessment, a
paper presented at the Panel Session of IEEE/PES 2003
General Meeting, July 13-17, 2003, Toronto, Canada.


K. Morison et al, Critical Requirements for Successful

On-line Security Assessment, Panel paper at 2004 IEEE
PSC&E Conference, New York, NY, Oct. 2004


K. Morison, L. Wang. P. Kundur, Power System

Security Assessment, , IEEE Power & Energy
Magazine, September/October 2004


J.A. Huang, A. Valette, M. Beaudoin, K. Morison, A.

Moshref, M. Provencher, J. Sun, An Intelligent System
for Advanced Dynamic Security Assessment, PowerCon
2002, Kunming, China, October 2002.

B. On-Line Voltage Security Assessment

Voltage security is perhaps the most significant concern
for system operators today and the number of installations
of on-line voltage security assessment (VSA) has grown
dramatically in recent years. Since powerflow-based
methods can often be used for VSA, computation speed is
less of a concern than for transient security assessment in
which time-domain simulation are needed. Therefore,
even for very large systems, state-of-the-art on-line VSA
systems can readily provide a variety of information
including proximity to instability (margin), critical
contingencies, and remedial measures. However, some
information, such as weak regions in a system may be best
assessed using intelligent technologies. To this end a
project has recently been initiated by EPRI to develop
methods for the Identification of Voltage Control Areas
and Determination of Reactive Reserves. The proposed
approach is to develop a derivative of the method used in
the POSSIT project. One main difference in the
formulation is that instead of a single outcome
(stable/unstable) for each system condition studied, the
outcome would consist of a number of buses representing
the weak area (which may be computed using modal
analysis) together with the required reactive reserves. The
project is expected to be completed mid-2006.

Kip Morison received his BaSc and MaSc degrees in
Electrical Engineering from the University of Toronto in
1980 and 1985 respectively. From 1981 to 1993 he
worked in the Analytical Methods and Specialized System
Studies Department at Ontario Hydro in Toronto, Canada.
In 1993 he joined Powertech Labs in Vancouver, where he
is now Director of the Power System Studies Group, which
provides international consulting services and commercial
power system software. His interests include power
system stability and control and on-line dynamic security
assessment and he has authored numerous publications on
the subjects. He is a registered professional engineer in the
Provinces of Ontario, Alberta, and British Columbia and is
a member of the IEEE.