2010 Sixth International Conference on Natural Computation (ICNC 2010)

Application of Artificial Intelligence

Technologies for Monitoring Large Power
Interconnections
V. G. Kurbatsky, member IEEE and N.V. Tomin, member IEEE

information on the state variables synchronized by using the

Abstract— Sophisticated operation conditions for large satellite systems GLONASS or the Global Positioning System
interconnected power systems (IPSs) need a powerful instrument (GPS) in time makes it possible to create an up-to-date system
to study dynamic characteristics of electric power systems of EPS operation monitoring [7]. One of them is a technology
(EPSs) in real time for different system states. It is a system of for vector measurement of state variables, i.e. the technology
operating condition monitoring that enhances control efficiency of Wide-Area Measurement & Control Systems
of normal and emergency conditions in the current market
(WAMS/WACS) [8, 9]. It takes control of EPS state
environment. Effective organization of the system of IPS
operation monitoring is possible only by a extensive involvement
synchronously and with a high degree of accuracy. The basic
of new tools for the analysis and calculations of operating measuring equipment in these systems is a device for phase
conditions, and first of all technologies of artificial intelligence. measurement – Phasor Measurement Units (PMU) of
The paper presents an approach to the super short-term complex electrical values.
forecasting of state variables on the basis of neural network Note that availability of such systems allows:
technologies and algorithms of nonlinear optimization that is • Monitoring of synchronous generators operation;
realized in the ANAPRO software. • On-line analysis of low frequency power oscillations;
• Power system state estimation on the basis of
Index Terms — monitoring, forecasting, artificial intelligence
methods, state variables
SCADA and PMU measurements;
• On-line adjustment of control actions of emergency
I. INRODUCTION control systems;
• Estimation of the extent to which the current state
evelopment of systems and devices for monitoring the
D state of energy and electrical equipment (devices and
systems of diagnostics) and also monitoring the EPS
variables are close to their maximum;
• On-line estimation of emergency severity;
• Monitoring of load node stability (dynamic
operation conditions seems to be highly important because of
properties);
radically changed development trends and complicated
operating conditions of large-scale IPSs. • Identification of parameters of power system
It should be underlined that in these conditions a system of mathematical models;
operation monitoring is a powerful tool to study dynamic • Confirmation of transient modeling results;
characteristics of EPSs in real time for different system states. • Automatic control of power system voltage level.
The system enhances control efficiency of normal and
emergency conditions in the current market environment (Fig. II. THE REVIEW OF PAPERS ON MONITORING OF ELECTRIC
1). POWER SYSTEMS
At present the use of effective mathematical methods of
control theory and advanced information technologies is Now WAMS/WACS technologies have found wide
among the most important trends in development of the application in power pool system of the various countries [15].
control system of EPS operating conditions. EIMV (Electric Power Research Institute of Milan Vidmar)
The achieved level of technological development in the and ELES is presented in [16], which on the basis of DACF
current measurement devices that are based on the advanced (Day Ahead Congestion Forecast) models and load-flow
information technologies of transmission and processing of calculation results, provides a graphical presentation of the
UCTE voltage angles. Voltage angles, calculated with a load
flow program, are appointed a geographical position based on
V. G. Kurbatsky is with the Department of Electric Power Systems, Energy the known position of the busbar in question. A two
Systems Institute, Irkutsk, Russia (e-mail: kurbatsky@isem.sei.irk.ru).
N. V. Tomin is with the Department of Electric Power Systems, Energy dimensional map of Europe is than constructed, showing the
Systems Institute, Irkutsk, Russia (e-mail: tomin@isem.sei.irk.ru). voltage angles in various colors. A map like that gives an easy
This study is supported by the grant 4633.2010.8 of the Leading Scientific to comprehend view of me system stress which is higher in
School of the Russian Federation and the grant #09-08-91330 of the Russian areas with high voltage phase angle gradients - fig. 2.
Foundation of Basic Researches.

978-1-4244-5961-2/10/$26.00 ©2010 IEEE 1360

Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on July 29,2020 at 09:49:20 UTC from IEEE Xplore. Restrictions apply.
 Current power system state

Threat of emergency Post-emergency

Normal condition Emergency condition
situation condition

Monitoring of post-
Monitoring of normal and pre-emergency Recording of
conditions emergency condition emergency condition

• Identification of • Realization of control

• System state estimation
emergency condition actions to eliminate
• Prediction of parameters of the forthcoming (localize) emergency
condition • Creation of archives situation
of emergency
• Estimation of weak points
conditions • Determination of
• Estimation of reserves in the forthcoming indices and criteria
condition • Choice of control for system transition
actions to eliminate from post-emergency
• Determination of indices and criteria for system
(localize) to normal condition
transition from normal to pre-emergency
condition
emergency
situation

Control in normal Control in pre- Control in emergency Control in post-

condition emergency condition condition emergency condition
regime

Fig. 1. Correspondence of system states and problems of monitoring of operating conditions and their control

The Fig. 2 presents a static system state as expected day

ahead. This kind of a graphical presentation can be useful also
for presenting post-fault analysis (if system snapshots are used
instead of forecasts).
After the UCTE wide event in 2006 [1], WAMS proved as
one of the most useful means in post-mortem determination of
course of events. Fig. 3 shows a 3D view of the voltage
behavior in the Slovenian grid at that time.
In the paper [17] presents the concepts of one of the
effective tools for disturbance analysis used in Saudi
Electricity Company in the Western Region (SEC-WR) named
as Dynamic System Monitoring (DSM).
The DSM units are linked by telephone lines and modems
to the Master Station PC computer and each disturbance event
can be retrieved and analyzed by Data Analysis and Display
Software working under Windows 2000, NT or XP. Three
actual disturbances in the Western Region are described and
analyzed in details using the records of the concerned DSM
units and cross checked with the records of some numerical
Fig. 2. 2D view of the voltage phase angles across the UCTE, calculated from a protection relays to ensure the correct operation and the
DACF model
technical recommendations are listed and implemented in
most cases to avoid repeating of the same event in future. The

1361
Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on July 29,2020 at 09:49:20 UTC from IEEE Xplore. Restrictions apply.
paper presents also good and practical new method for the • Models applying an apparatus of artificial neural
evaluation of the dynamic behaviour of the generating units networks (ANN) [4, 5]
(MW/Hz) using DSM. • Genetic algorithms [3]
• Hybrid and fuzzy systems [2].
The Kohonen maps applied for visualization and analysis
of schemes and operating conditions in EPS may be given as
an example of practical use of the artificial intelligence
technologies. Use a self-organizing Kohonen network in the
structure of the main system of data processing of the WAMS
system is offered in [10] (Fig. 4).
The Kohonen network in this case is capable of analyzing
schemes and operating conditions "on-line" and "off-line". For
the given ANN the problem is solved by division of the
studied situations into clusters with close typical change of
processes in EPS. Hence, the whole variety of schemes and
operating conditions can be visualized effectively and
dangerous contingencies in the studied EPSs can be detected
timely.
The experimental studies have shown [12] that the
Kohonen maps applied allow an adequate detection of
inadmissible combinations of the scheme and operating
parameters and a correct choice of necessary control actions.
Fig. 3. A 3-dimensionaI view of voltage magnitude (upper chart) and voltage
phasors (lower chart) during the 2006 UCTE wide event III. SUPERSHORT-TERM FORECASTING OF STATE VARIABLES
ON THE BASIS OF ARTIFICIAL NEURAL NETWORKS
Effective organization of the system of EPS operating
In addition, a key monitoring problem is the problem of
condition monitoring is possible by an extensive involvement
forecasting state variables of EPS. In this case within the
of new tools for the analysis and calculations of operating
system state estimation [13, 14] it is supposed to forecast all
conditions, primarily technologies of artificial intelligence
the state variables for a very short time span. The lead time
[18].
can be from several seconds (an interval of obtaining
It should be underlined that the methods of artificial
information on EPS state variables in a unit time) to one
intelligence have nothing to do with algorithmic calculations
minute. Information on the system state is obtained by using
and therefore, do not require complex computational
the SCADA system and from PMU recorders.
mathematical models to be constructed for an object. The
Normally modern monitoring systems employ for these
artificial intelligence methods reproduce (copy) individual
purposes traditional forecasting models such as ARIMA and
functions of the creative activity of human brain, which
Kalman filter. These make it possible to obtain rather good
makes it possible to find optimal decisions in a large set of
forecasts provided the dynamics of random variations of state
possible states in the shortest time.
variables is represented as a stationary Gaussian process.
The following are primarily the main technologies of
artificial intelligence:

Learning vector Kohonen maps Current estimation

vector

Learning Request

“Off-line” “On-line”
mode mode

Fig. 4. System of analysis of schemes and operating conditions on the basis of the Kohonen maps

1362
Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on July 29,2020 at 09:49:20 UTC from IEEE Xplore. Restrictions apply.
 In the cases where state variables are rather variable and error [5]. This allows the balance between the reliability of
nonstationary it is more sensible to use neural network learning and the quality of model to be reached, thus making
models. In the course of numerous studies and calculations the the neural network forecasting model generalize better.
traditional forecasting methods have proved to be
insufficiently accurate in modeling state variables of power R(w)=EW(W)+λEc(W) (1)
systems as compared to the neural network methods.
In spite of this fact it is necessary to emphasize that the EW(W) - standard performance measure that depends on both
advantage of the neural network forecast as compared, for the network itself ( model) and the input data; Ec(W)-
example to ARIMA and Kalman filter is most pronounced in complexity penalty depends solely on the ANN itself and is
the cases where the trend of analyzed time series differs from determined on the basis of preliminary data on the model
the linear one and the initial data contain a considerable value structure; λ – regularization parameter.
of irregular component [1]. In the cases, however, where time The ANN types considered in the studies were: radial basis
series of the process contains a linear trend the traditional function (RBF) and generalized regression neural network
approaches practically do not differ from the neural network (GRNN) [4, 5]. They make it possible to obtain rather accurate
ones in terms of forecasting accuracy. results in supershort-term forecasting.
The authors propose the approach to supershort-term In the case of complex computational forecasting problem
forecasting of state variables on the basis of neural network (for example when learning sample contains many additional
technologies and nonlinear optimization algorithms. The factors) the principle of committee machine (CM) is used [4]
approach is implemented in the ANAPRO software [11, 12]. (Fig.4) as a basis for the proposed approach. The CM
The use of nonlinear optimization algorithms in the represents a neural network system consisting of combination
approach , namely, the methods of simulated annealing (SA) of neural networks-experts which allow one to find a general
and neuro-genetic input selection (NGIS) [4,5], provides the solution, y(n), which has priority over each solution of an
procedure of choosing the best forecasting model for each individual expert, (n).
individual sampling. For example in the process of learning
sampling analysis based on the NGIS algorithm individual
input data can be rejected as less informative. This method Input Output
signal Expert 1 signal
represents optimization on the basis of random search x(n) y(n)
techniques and combines the capabilities of GA and
Expert 2
PNN/GRNN networks to automatically determine optimal
Full adder
combinations of input variables (Fig.3).

Expert k

Fig.4. Committee machine for forecasting problem

The calculation practice shows that for different

combinations of learning and test samples the SA procedure
should be started 3-6 times to form the neural networks -
experts.

III. EXPERIMENTAL C ALCULATIONS

The calculations show [12] that the suggested neural
network approach within the ANAPRO software makes it
Fig.3. Algorithm of neuro-genetic selection possible to much more efficiently solve this problem as
compared to the other traditional forecasting methods. This is
The PNN/GRNN networks allow the best results to be related to the use of nonlinear optimization algorithms as a
“remembered”, which improves the previous results. Owing to basis for the approach. The algorithms are used to analyze the
the radial layers with Gaussian function in the structure of time series and networks on the basis of Gaussian radial
PNN-algorithm, bad data in the input sampling can be reduced functions, in particular, the GRNN structure. The GRNN
to minimum. networks are learned practically in a flash, which is extremely
The SA technique makes it possible to analyze the important for on-line forecasts, and are robust to the presence
properties of the initial sampling and organize a competition- of bad data.
based system between different neural network forecasting The experimental calculation results obtained while
models when in the process of nonlinear optimization the best secondly forecasting active power flow in the 500 kV
forecasting model is selected. This competition procedure is intersystem transmission line that
based on the criterion of minimizing the total risk of ANN

1363
Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on July 29,2020 at 09:49:20 UTC from IEEE Xplore. Restrictions apply.
 40060

36050
Электрическая нагрузка, МВт

40
320
Power flow, MVt

30
280

24020

20010

1400
04.02.08 05.02.08 06.02.08 07.02.08 08.02.08 09.02.08 10.02.08 11.02.08 12.02.08 13.02.08

Fig.7. Change in the power flow in (he 500 kV intersystem transmission line

The experimental calculation results obtained while TABLE 1. TELEMETRY FORECASTING ERRORS ON THE BASIS OF
minutely forecasting active power flow in the 500 kV ARIMA AND ANN
intersystem transmission line that connects two large-scale
interconnected power systems (Fig.7) are presented in Fig.8 MAPE for interval, %
Model
and Table 1. 20:01 20:02 20:03
The calculation data show that the neural network model
GRNN 17,7 25,1 21,3
allows one to decrease the secondly forecast error almost
twice as compared to the forecast made by the ARIMA model. ARIMA 39,1 62,1 56,2

310

300

280

260

240

220

200
fact power flow
forecasting on the basis of GRNN
180
forecasting on the basis of ARIMA

160

Fig.8. Minute ahead secondly forecast of power flow

1364
Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on July 29,2020 at 09:49:20 UTC from IEEE Xplore. Restrictions apply.
 [9] Balance J. W. Monitoring Power System Dynamics using
As is seen from Fig.7 the time series of change in the Phasor Measurement Technology for Power System
power flow on this section of ESR is extremely variable and Dynamic Security Assessment / J. W. Balance, B.
non-stationary which does not allow the traditional forecasting Bhargava, G. D. Rodriguez. Bologna PowerTech
models to be used. Conference 2003 IEEE, 23-26 June 2003.
Figure 8 and Table 1 present the results of day ahead [10] Handshin E. Asset management and survivability of
minutely power flow forecasting on the basis of ARIMA and electric enrgy systems during large disturbances / E.
ANN models. As is seen the neural network forecast on the Handshin, E. Hauptmeter, D. Hause // Liberalization and
basis of the proposed approach allows one to decrease the modernization of power systems: risk assessment and
forecasting error by a factor of 2 as compared to the ARIMA optimization for asset management: The 3rd International
model. workshop. - Irkutsk Energy Systems Institute, 2006. - Pp.
156-168
V. CONCLUSION
[11] Kurbatsky V.G. Application of the ANAPRO software
for analysis and forecasting of state parameters and
The need for wider application of monitoring systems is process characteristics in electric power systems /
growing increasingly urgent in the light of the future V.G.Kurbatsky, N.V.Tomin. Proceedings of the 8th
possibility to interconnect the main power grids of Europe and Baikal All-Russian Conf. "Information and mathematical
Russia. The technological progress in the last decades has technologies in science and management." Part 1. -
shown that the use of perspective information technologies, Irkutsk: SEI SB RAS, 2008. - P.91-99.
first of all ANN, can provide reliable operation of future [12] Kurbatsky V.G., Tomin N.V. Use of the "ANAPRO"
interconnected power grids of EU and Russia, optimal use of software to analyze and forecast operating parameters
energy resources on vast territories and mutually beneficial and technological characteristics on the basis of macro
electricity trade between different regions in terms of market applications // Proceedings of the International
requirements. Conference "PowerTech'2009", Bucharest, 2009. CD
ROM, Paper 93.
VI. REFERENCES [13] A.Glazunova, I.Kolosok, E.Korkina. PMU placement on
Books: the basis of SCADA measurements for fast load flow
[1] Borovikov V.P. Forecasting in STATISTICA system in calculation in electric power systems // Proceedings of
Windows environment: Theory and intensive practice on the International Conference "PowerTech'2009",
computer: Tutorial/V.P.Borovikov, G.I.Ivchenko. - 2nd Bucharest, 2009. CD ROM, Paper 195.
edit., revised and enlarged. - M.: Finansy i statistika, [14] A.Gamm, Y.Grishin, A.Glazunova, I.Kolosok,
2006. - 368 p. E.Korkina. New EPS state estimation algorithms based
[2] Fuzzy sets in the control and artificial intelligence on the technique of test equations and PMU
models/ ed. by DA. Pospelov. - M.: Nauka, 1986. - 206 p. measurements// Proceedings of the International
[3] Goldberg D. Genetic Algorithms in Search, Optimization Conference "PowerTech'2007", Lausanne, 2007. CD
and Machine Learning / D.Goldberg. - Reading, MA: ROM.
Addison-Wesley, 1989 [4] Haykin S. Neural networks. A [15] Kulikov J.A. Use of technology of vector measurement of
comprehensive foundation. parametres in IPS for supply with information of
[4] Second edition / S. Haykin. - Williams Publishing House, operatively-dispatching management / JA.Kulikov//
2006.-1104 p. Energetik,2009,№l. With. 10-13
[5] Ossovsky S. Neural networks for information processing / [16]Kostevc J. WAMS development and practice / J. Kostevc,
S. Ossovsky. - transl. from Polish by I.D.Rudinsky. - M.: U. Gabrijel. Proceedings 4th International Conference
Finansy i statistika, 2004. - 344 p. "Liberalization and Modernization of Power Systems:
[6] V. Dale, Z., Krishan, and O. Paegle, Dynamic Coordinated Monitoring and Control towards Smart
Optimization of Power Networks Development, Zinatne, Grids", Irkutsk, 2009. CD ROM
Riga, 1990 (in Russian). [17] Allaf B. A. Applications of Dynamic System Monitoring
in Power System Analysis / B. A. Allaf, M. H. Shwehdi
Papers from Conference Proceedings (Published): // Proceedings of the International Conference
"PowerTech'2009", Bucharest, 2009. CD ROM, Paper
[7] Phadke G. Synchronized sampling and phasor 14.
measurements for relaying and control / B. Pickett et al. [18] Kurbatsky V.G. Artificial Intelligence Technologies for
G. Phadke. IEEE Transactions on Power Delivery, vol. 9, Monitoring Large Power Interconnections /
no. 1, January 1994, pp. 442-452. V.G.Kurbatsky, N.V.Tomin. Proceedings 4th
[8] Hemmingsson M. Wide Area System Monitoring and International Conference "Liberalization and
Control / D. Karlsson, M. Hemmingsson, S. Lindahl. Modernization of Power Systems: Coordinated
IEEE power & energy magazine, September/October Monitoring and Control towards Smart Grids", Irkutsk,
2004, pp. 68-76. 2009. CD ROM

1365
Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on July 29,2020 at 09:49:20 UTC from IEEE Xplore. Restrictions apply.