Scientific Bulletin of the Electrical Engineering Faculty – Year 10 No. 2 (13)
ISSN 1843-6188
PREVENTION AND ELIMINATION OF POWER SYSTEM EMERGENCY
STATES BY MEANS OF NEW PREDICTION AND CONTROL METHODS
N. VOROPAI, V. KURBATSKY, D. PANASETSKY, N.TOMIN
Melentiev Energy Systems Institute, Lermontov Str., 130, Irkutsk, Russia
E-mail: [email protected]
Most of the existing emergency control strategies use
two types of control: control by the displacement of the
controlled parameter and disturbance-stimulated control.
The efficiency of these control principles has been
proved by longstanding operating practice. Nevertheless,
during the last decades large interconnected power systems throughout the world were frequently subjected to
widespread blackouts which interrupted millions of consumers and cost billions of dollars. The major reasons of
these blackouts are rapid development and complication
of operations of power systems caused by power consumption and network development growth as well as
transition to a market economy.
New power grid operating conditions require improvement and development of the existing control principles.
Abstract: General description of disadvantages of the existent
control systems is given in the article. These disadvantages may
have been one of the causes of the blackouts that took place all
over the world over the last several decades. Describing the
disadvantages authors suggest a possible ways of developing
and improving of the existent control systems.
Keywords: power system stability and control, prediction
methods, Kohonen maps
1. INTRODUCTION
Wide range of the modern power system control problems can be roughly divided on two parts: operational
management problems and emergency control problems.
Operational management is performed by the dispatchers
of transmission system operators (TSO). One of the main
goals of operational management is a control of normal
and post-emergency states for the purpose of emergency
state prevention. Emergency state elimination is performed automatically by means of emergency control
systems (relay protection, different local and centralized
emergency control devices, etc.). The main purpose of
emergency control is to provide a transition to a postemergency state (Figure.1).
2. THE PROPOSED APPROACH OUTLINES
Operational management deals with emergency state
prediction concept (probability of emergency state occurrence). Prediction functions can be realized by means
of different advice-giver software. But anyway, the final
decision (control action) is realized by a system operator
who for a variety of reasons is not able to realize rapid
and economically ineffective control actions that would
prevent an accident development. As it was mentioned
above most of the existing emergency control strategies
use two types of control: control by the displacement of
the controlled parameter and disturbance-stimulated control. In such a manner, emergency control schemes do
not predict the possible development of the normal,
emergency or postemergency states but operate only
when the disturbance has occurred (Figure 2).
Figure 1. Operational Management and
Emergency Control principles
Figure 2. Realization of control actions.
The current situation
The study was supported by the Grants of Leading Scientific of
RF#4633.2010.8 and Russian Foundation of Basic Researches #09-0891330 and Federal Agency for Science and Innovations within Federal
Program “R&D in Priory Areas of Russia’s Science and Technological
Complex Development for 2007-2012”
So, it is possible to suppose that there are two main disadvantages of the existing control ideology. First disad-
110
ISSN 1843-6188
Scientific Bulletin of the Electrical Engineering Faculty – Year 10 No. 2 (13)
vantage is the absence of fast control actions realization
in operational management. And the second one is the
absence of prediction procedures in emergency control.
Existing practice showed that most of blackouts carried
according to similar scenarios. If protection system
works correctly, a power system has sufficient stability
to withstand the first heavy disturbance in extra high
voltage transmission system. The post-disturbance phase
represents a deceptively calm period that lasts from several minutes to several hours with a normal level of frequency and then the system collapse that lasts seconds.
One of the main causes of such behavior is the absence
of the local control devices coordination which can be a
cause of different negative phenomena such as voltage
collapse and cascade line tripping. In such a way, besides the above listed disadvantages it is important to
pay attention to coordination of different local control
devices in a normal and a postemergency state.
The paper proposes to complement the existing control
strategies with prediction control strategy which implies
identification of the emergency until the emergency has
occurred. This needs to reveal key factors which have a
profound effect on a probability of the following negative
accident development. To decrease the probability of the
following negative accident development an emergency
control system based on the proposed strategy has to
coordinate the work of different control devices in both
normal and postemergency operating (Figure.3).
a self-organizing Kohonen maps [8] that will be able to
perform monitoring and prediction of emergency state
[6, 9]. A modified 10-bus test system taken from [1, 7]
was used as a case study (Figure 4).
Figure 4. 10-Bus test system
The purpose of the intelligent model is a prediction of
the voltage instability occurrence in the test system. Undoubtedly, different types of instability may take place in
postemergency state, but it is voltage instability in combination with line tripping that caused most of blackouts
in different parts of the world.
The test system under consideration includes the following set of the decentralized control devices:
 Turbine Governor (TG) and Automatic Voltage
Regulator (AVR) at Gen 2 and Gen 3. Gen 1 is
modeled as an infinite power source.
 Gen 3 is equipped with Over Excitation Limiter
(OXL); the maximum excitation current of Gen 3
is 12 pu.
 Bus 7 load is modeled as an equivalent 3130 MW
induction motor. Bus 10 load is modeled as 50%
constant impedance and 50% constant current for
both active and reactive components.
 Bus 9 – Bus 10 transformer was equipped with
continuous ULTC device, other transformers have
a fixed tap ratio.
More details about the test system model can be found in
[1, 7]. Necessary power flows and time domain simulations were carried in Matlab/PSAT environment [2].
The following sequence of disturbances is examined:
 2 seconds after the simulation starts. Loss of
100% of the capacitor banks, connected to Bus 7.
 20 seconds after the simulation starts. Loss of
20% of the capacitors banks, connected to Bus 8.
Voltage reductions at different buses of the test system
are shown in Figure 5. The change of Gen 3 rotor current
during the simulation is given in Figure 6.
Figure 3. Realization of control actions.
The proposed approach
The paper also gives a brief description of new possible
approaches to decentralized adaptive coordination of
emergency control based on new technical capabilities.
The proposed principles, on the one hand, provide preventive measures and hence decrease the probability of
the following negative accident development, and, on the
other hand, they must complement and do not contradict
to the existing emergency control ideology.
3. CASE STUDY. DEVELOPMENT OF AN
INTELLIGENT MODEL FOR VOLTAGE
INSTABILITY PREDICTION
In this section a simple network model is used to illustrate the possibility of making an “intelligent”1 model
that will be able to detect the proximity of an emergency
state. The main idea of the intelligent approach here
[9, 13] is a creation of a neural network model, based on
Figure 5. Changes in substation voltage level
The term “intelligent” is used in reference to the approaches, methods, systems and complexes using artificial intelligence technologies
1
111
Scientific Bulletin of the Electrical Engineering Faculty – Year 10 No. 2 (13)
After the first disturbance, rotor current of Gen 3 comes
closer to the thermal limit, the system remains stable.
After the second disturbance, rotor current of Gen 3 becomes higher than the thermal limit. The second disturbance leaded to the fast grows of the reactive power deficiency and after 15 seconds to the voltage collapse of the
whole system.
ISSN 1843-6188
[Tap ratio of the ULTC at Bus 10]
Normal state
[Rotor angle of Gen 2 and Gen 3]
[Field winding voltage of the Gen 2]
Heavy state #1
[AVR reference voltage of Gen 3]
[Bus 1 – Bus 10 voltage]
Heavy state #2
[Voltage signal of the OXL,
installed at Gen 3]
Kohonen map
[Reactive power, produced
by the capacitor banks]
Emergency state #2
Figure 7. Overall schematic of the proposed intelligent
model for voltage instability prediction.
When making the time domain simulation 36 parameters
were used as input parameters for Kohonen network
model (Figure 8).
Figure 6. Gen 3 Rotor current change
3.1 Detection of the critical situation by means of Kohonen maps
The following set of the parameters, obtained during the
time domain simulation, was used to form the intelligent
model:
 Tap ratio of the ULTC at Bus 9 - Bus 10 transformer;
 Rotor angles of Gen 2 and Gen 3;
 Field winding voltage of Gen 2 and Gen 3;
 Voltage signal of OXL, installed at Gen 3;
 Bus 1 – Bus 10 voltages;
 AVR reference voltage of Gen 3;
 Current signal of OXL, installed at Gen 3;
 Active and reactive power and current, produced by Gen 1, Gen 2 and Gen 3.
 Active and reactive power and current consumed by the loads at Bus 7 and Bus 10;
 Reactive power, produced by the capacitor
banks;
All the parameters are directly or indirectly point to the
possibility of the voltage instability beginning. Such a
big number of parameters were chosen to analyze all the
interrelations between voltage instability and operating
conditions. Later on the set of parameters have to become smaller.
Data array, containing the information about the temporal variation of the represented parameters, was divided on clusters. The clusters contain data with the joint
properties and represent different operating modes of the
test system: from normal to emergency. Kohonen neural
network model was trained to identify different states of
the test system and predict the occurrence of the emergency state (Figure 7).
Figure 8. Architecture of Kohonen neural network
Data array, containing the information about the temporal
variation of the represented parameters, was divided on
six clusters. A 3x2 network topology map was formed
(Figure 9).
(0,0) Heavy state #1
(1,0) Heavy state #2
(2,0) Emergency state #2
(0,1) Normal state
(1,1) Heavy state #3
(2,1) Emergency state #1
Figure 9. Kohonen topological map for monitoring of the
test system
The network topological map was identified, i.e. each
cluster was subsumed under one of the categories, which
correspond to the following operating states of the test
system:
 Normal state.
 Heavy state №1, this state occurred after the
first disturbance.
 Heavy state №2.
 Heavy state №3.
 Emergency state №1, this state occurred after
the second disturbance.
112
Scientific Bulletin of the Electrical Engineering Faculty – Year 10 No. 2 (13)
ISSN 1843-6188

Emergency state №2, blackout of the test system.
Preliminary test results showed that the proposed intelligent model proactively responds to the change of the test
system state. After the first disturbance the “Heavy State
№1” cluster was activated (Figure 10, a). Starting from
time t=3.8 seconds, the activation of the “Heavy State
№1” cluster increased considerably (Figure 10,b).
After the second disturbance at time t=20.4 seconds the
“Emergency state №1” cluster was activated (Figure 11).
(0,0) Heavy state #1
(1,0) Heavy state #2
(2,0) Emergency state #2
(0,1) Normal state
(1,1) Heavy state #3
(2,1) Emergency state #1
The proposed Kohonen neural network model is an example of the intelligent model that will be able to perform monitoring and prediction of emergency state. Such
model can be used as a key element of a control system,
based on prediction control strategy (Figure 3). On the
one hand, such model makes it possible to increase the
speed of response of a system operator - nowadays system operators need to spend a lot of time to process a
mass of information during postdisturbance period. On
the other hand, such model can be used as a triggering
device for automatic emergency control system with
prediction.
The further work aims to make a detail study of the operating parameters. The set of the parameters that closely
coupled with the probability of the emergency state occurrence needs to be defined. As distinct from the other
types of instability, voltage instability is a local problem.
The prime cause of the voltage instability is a lack of
reactive power in a subsystem. It was mentioned in papers [3, 4] that considerable reduction of transmission
voltage levels and increase of excitation currents on
some reactive power sources up to thermal limit can be
used to indicate the proximity to voltage collapse. The
case study analysis confirmed this conclusion (Figure 5
and Figure 6).
a) time t=0.1 sec.
(0,0) Heavy state #1
(1,0) Heavy state #2
(2,0) Emergency state #2
(0,1) Normal state
(1,1) Heavy state #3
(2,1) Emergency state #1
4. NEW CONTROL METHODS.
When the approach of the emergency state has been detected, a protection system must be triggered. What kind
of control actions should the system use to eliminate the
emergency state appearance?
4.1 Voltage instability control.
In case of voltage instability the main purpose of the
protection system is to control the capacity of available
reactive power resources. Power industry has already
used the philosophy of load shedding by selecting nonessential load to prevent frequency reduction. The analysis of recent blackouts showed that the rapid load shedding is usually the only way to prevent the collapse of
the whole system [10]. On the one hand, load shedding
should be as fast as possible, on the other hand, it should
be optimal. The optimal load shedding scheme can be
realized by using different optimization procedures, but
it is hard to solve optimization problem for any possible
situation in advance, because the number of situations is
too big. This means that some optimization computations
should be made during the postdisturbance period. Inspite of the fact that there is a number of optimization
techniques that can be used to calculate emergency control actions quickly, the amount of input data required to
solve the problem is usually too big. The state estimation
alone can take from tens of seconds to minutes. However, load shedding under postdisturbance conditions has to
work faster. Hence, load shedding procedure has to use
less complex methods to control postdisturbance phenomenon. The following simple countermeasures to control postdisturbance phenomenon were proposed in [5]:
 Countermeasure 1. Fast tap changing on transmission substation transformers.
b) time t=3.8 sec.
Figure 10. Kohonen topological map state after the first
disturbance
In such a manner the proposed intelligent model makes it
possible to realize the test system monitoring and predict
emergency state before it has occurred.
(0,0) Heavy state #1
(1,0) Heavy state #2
(2,0) Emergency state #2
(0,1) Normal state
(1,1) Heavy state #3
(2,1) Emergency state #1
Figure 11. Kohonen topological map state after the second
disturbance. Time t=20.4 seconds
3.2 Analysis of the results
113
Scientific Bulletin of the Electrical Engineering Faculty – Year 10 No. 2 (13)
 Countermeasure 2. Raising terminal voltage on
selected synchronous condensers and hydro generators.
 Countermeasure 3. Fast tap changing on selected
generator transformers.
 Countermeasure 4. Strategic load shedding at selected transmission substations only if voltage
levels and reactive outputs do not meet the requirements, or some transmission lines are overloaded.
 Countermeasure 5. Rearranging generator MW
outputs. Connecting part of the disconnected load.
Countermeasures 1 – 3 have approximately the same
execution time and their main purposes are to impede the
sharp increase of series reactive power losses, to increase
transmission line charging and to inhibit tap changing on
subtransmission and distribution transformers. Load is
shed (Countermeasure 4) only after countermeasures 1 –
3. This will decrease the amount of the load to be shed.
Countermeasure 5 considers an optimization procedure.
The optimization procedure takes much more time in
comparison with countermeasures 1 – 4 and provides
postemergency operation optimization. Thereby, countermeasures 1 – 4 provide fast control of the postdisturbance phenomenon to avoid voltage collapse and Countermeasure 5 provides long-time-period postemergency
operation optimization. The proposed control principles
can be applied to various parts of the grid that work independently. Briefly, the control actions aim to control
the capacity of the available reactive power resources
and do not let reactive power demand of the affected
region increase beyond their sustainable capacity.
4.2 Cascade line tripping control.
Another cause that can lead to power system breakdown
is the cascade line tripping. It is necessary to consider
two types of overloading:
 Critical overloading. When the affected power
system experiences critical overload of a transmission line, fast load shedding is usually the only way to prevent the line tripping and the following collapse of the whole system [10]. In this case
there will be a short-term transient process, and
control actions to be performed should be fast but
not optimal.
 Noncritical overloading. If the transmission line
overload is not too high, we have time to perform
control actions without load shedding procedure
being triggered. These control actions can include
redistribution of generator active outputs and
power flow control by means of different FACTS
devices. In this case we have a long-term transient process that lasts from tens of seconds to several hours.
Analysis showed that it was noncritical overloading that
occurred before cascade line tripping during the recent
blackouts in different parts of the world. But usually the
time between the first heavy contingency and the voltage
collapse, caused by the following line tripping, is not
enough for centralized control of power flow. Therefore,
new fast algorithms based on distributed principles are
ISSN 1843-6188
needed to avoid line tripping during the postdisturbance
period. Nowadays FACTS technologies have new capabilities for fast power flow control, and the central task
now is development of new approaches where these capabilities can be efficiently applied. A new algorithm for
power flow control by means of FACTS devices has to
meet the following requirements:
 The new algorithm should use only local parameters of the load flow. This makes it possible to
apply the algorithm to various parts of the grid
independently.
 The algorithm has to work fast enough to provide
power flow control timely. To do this, it should
use some simple principles and work optimally.
 The algorithm should be capable of taking into
consideration different constraints, such as power
flow and operating constraints of the FACTS devices.
 The algorithm should be intelligent enough to detect the situation when the control capabilities of
the FACTS devices have been exhausted. This requirement guarantees that the other control actions
such as load shedding and rearranging generator
MW outputs will be triggered as soon as possible.
The block diagram that schematically describes the algorithm and includes all the requirements mentioned above
is presented in Figure 12.
Figure 12. Block diagram of the algorithm
4.3 Control system implementation
How the proposed methods can be implemented to power system control? Overall schematic of the new control
system is represented in Figure 13.
114
ISSN 1843-6188
Scientific Bulletin of the Electrical Engineering Faculty – Year 10 No. 2 (13)
[3]
Carson W. Taylor, D. C. Erickson, Recording and
and analyzing the July 2 cascading outage,” Comput. Applicat. Power Syst., vol.10,no.1, pp.2630,Jan.1997.
[4] W. R. Lachs, A New Horizon for System Protection
Schemes, IEEE Trans. Power Syst., vol.18, no.1,
pp.334-338, Feb.2003.
[5] W. R. Lachs, Voltage Instability in Interconnected
Power Systems: a Simulation Approach, IEEE
Trans.
Power
Syst.,
vol.7,
no.2,pp.753761,May1992.
[6] Kurbatsky V.G., N.V.Tomin. Analysis of electric
power losses on the basis of advanced artificial intelligence algorithms. // Elektrichestvo. – № 4. –
2007 – P. 5-12.
Books:
[7] P. Kundur, Power System Stability and Control,
Mc-Grall Hill, New York, 1994.
[8] Kohonen T. Self-organizing maps. / T. Kohonen. –
Berlin ets: Spzinger, 1995. XV. 362 p.
[9] Haykin S. Neural networks. A comprehensive foundation. Second edition / S. Haykin. – Williams Publishing House, 2006. – 1104 p.
Conference or Symposium Proceedings:
[10] CIGRE Defense Plan Against Extreme Contingencies, CIGRE TaskForce C2.02.24, April2007.
[11] D. A. Panasetsky, N. I. Voropai, A Multi-Agent Approach to Coordination of Different Emergency
Control Devices Against Voltage Collapse, Proc. of
IEEE Bucharest PowerTech International Conference, Bucharest, Romania, 2009.
[12] D. A. Panasetsky, N. I. Voropai, A New Approach
to Coordination of FACTS Devices Based on a Sensitivity Analysis, Accepted for publication on International Conference on Power Systems Technology
POWERCON 2010, October 24-28, Hangzhou,
China, 2010.
[13] Kurbatsky V.G., Tomin N.V. Adaptive cluster analysis in electric network of power systems. / Proceedings of the International Conference “EEEIC2010”, Czech Republic, Prague , 2010, CD ROM,
Paper 82
Figure 13. Overall schematic of the control system
implementation
The main idea is that the new methods that deal with
voltage instability and cascade line tripping must complement and do not contradict to the existing ideology.
The new control system can be built by using distributed
intelligence principles. The distributed intelligence is
taken to mean the multi-agent systems.
A control system based on the multi-agent technique was
described in [11]. It coordinates different discrete and
continuous control devices during postdisturbance period
in order to prevent voltage collapse of the whole system.
The efficiency of the proposed technique has been
proved by numerical simulations.
Paper [12] describes a new approach to distributed coordination of TCSC FACTS devices which makes it possible to control power flow during the long-term transients. The proposed algorithm is based on the sensitivity
analysis and fulfills the requirements mentioned above
(Figure 12). It can be realized as a discrete-time controller and can be applied to different parts of a power system independently. This algorithm can be part of a multiagent automation described in [11].
5. CONCLUSION
General description of the disadvantages of the existent
control systems is given in the article. According to the
authors’ opinion, these disadvantages may have been one
of the causes of the blackouts that took place all over the
world over the last several decades. Describing the disadvantages authors suggest a possible ways of developing and improving of the existent control systems. The
article is a synopsis of the general ideas, which will be
detailed and improved in the future.
6. REFERENCES
Journals:
[1] Carson W. Taylor, Concepts of Undervoltage Load
Shedding for Voltage Stability, IEEE Transactions
on Power Delivery, Vol. 7, No. 2, April 1992.
[2] F. Milano, An Open Source Power System Analysis
Toolbox, IEEE Trans. Power Syst., vol.20, no.3,
pp. 1199 – 1206, Aug. 2005.
115