Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Identification of Synchronous Machines Parameters

using Genetic Algorithm and Load Rejection Test


Paulo V. V. Silva, Edson C. Bortoni J. Johnny Rocha E.
Center of Excellence on Energy Efficiency Global Technology Center Latin America
Itajubá Federal University GE Renewble Energy - Hydro
Itajubá, Brazil Taubaté, Brazil
paulovictor@unifei.edu.br johnny.rocha@ge.com

Abstract—This paper shows the use of Genetic Algorithm for the However, with the quadrature axis synchronous reactance
determination of synchronous machines parameters using load (Xq) in hands, one can easily calculate the reactive power
rejection test data. The conditions to a quadrature rejection are necessary to run the test, for a given active power. In this
obtained through machine modeling, instead of trial and error. work, this reactance is obtained beforehand through a
The methodology is applied using simulated data from real constant excitation test [6].
machines.
The Genetic Algorithm (GA) uses the equations that
Index Terms—Genetic Algorithm, Parameter Identification, model armature voltage behavior during a load rejection with,
Synchronous Machines. initially, random parameters and optimize it aiming the
results from the simulated test. The only input for this method
I. INTRODUCTION is the armature effective current before load rejection and the
instant that the rejection occurs.
The determination of synchronous machines parameters
allows the study of generation systems, which are steady state Currently methods to extract parameters from load
and transient analysis. Good parameter estimations directly rejection test data are through graphic inspection [5] and/or
assist one to obtain the required dynamic models for those using numerical calculations as exponential linear
studies. regressions. These two methods are the same in their essence
and will be used to evaluate results obtained through GA in
Although tests as sudden short-circuit are harmful to the this work.
machines, they are commonly applied due its high
recommendation in the main standards [1-2]. There are II. ANALYSIS OF THE LOAD REJECTION
methods that offer lower risk levels and are described in the The load rejection test is applied to the machine operating
main standards but not many applications of them have been as generator, connected to a load through a circuit breaker.
reported, those are the frequency response method [3] and Then the breaker is opened and armature three-phase voltage
voltage recovery test [4]. is recorded along the transient process.
Load rejection test [5] is a form of voltage recovery which A. Direct Axis Rejection
allows the determination of parameters from direct and
quadrature axes by rejecting two specific loads depending on Meeting the conditions described for a direct axis
the interesting axis. rejection, the armature current (Ia) should flow only in this
axis, as shown in the phasor diagram in Figure 1.
The load condition for direct axis is achieved by having
the machine at as closest as possible to zero active power, and Equation (1) represents the transient sinusoidal voltage
either drawing or supplying a considerable amount of reactive [7], and by taking its envelope it is possible to derivate all
power. It is also desired to keep the machine under-excited direct axis parameters. Trivially, the envelope is depict in (2),
during this test to ensure the lower voltage on the results. and its waveform shown in Figure 2.
௧ ௧
Ideally, the operation point for quadrature axis is where ܸሺ‫ݐ‬ሻ ൌ ቊͳ െ ‫ܫ‬଴ ቈܺௗ െ ሺܺௗ െ ܺௗᇱ ሻ݁
ି ᇲ
்೏బ
െ ሺܺௗᇱ െ ܺௗᇱᇱ ሻ݁
ି ᇲᇲ
்೏బ
቉ቋ •‹ሺ߱‫ݐ‬ሻ (1)
armature current flows only through this axis, which happens
when the absolute value power factor angle (ij) is equal to the ௧
ି ᇲ

ି ᇲᇲ
power angle (į). This condition can be found by successive ܸሺ‫ݐ‬ሻ ൌ ͳ െ ‫ܫ‬଴ ቈܺௗ െ ሺܺௗ െ ܺௗᇱ ሻ݁ ்೏బ െ ሺܺௗᇱ െ ܺௗᇱᇱ ሻ݁ ்೏బ ቉ (2)
load rejections with different power factors, monitoring the
field current deviation. The desired point occurs when there is
no noticeable transient deviation in field current.

‹,(((
Figure 1. Phasor diagram for d-axis rejection test. Figure 3. Phasor diagram for q-axis rejection test.

Figure 2. Terminal voltage variation after d-axis rejection [5]. Figure 4. Terminal voltage variation after q-axis rejection [5].

B. Quadrature Axis Rejection The envelope in (5), different from (2), has also the power
In this work, to get the conditions for a quadrature angle (į) in it, which is a known value. The power angle then
rejection, the reactive power is calculated by (3) at a given serve as result confirmation since it is also an output of the
active power and known Xq (e.g. using constant excitation test GA.
[6]). The load point defined by P and Q, from (3), forms the III. PARAMETERS IDENTIFICATION
phasor diagram in Figure 3, which illustrates how the armature
current flows only in quadrature axis direction. The synchronous machines parameters are identified as
consequence of the genetic algorithm optimizing the
ଶ theoretical model (2) or (5) with the data obtained from the
ܸଶ ܸଶ actual load rejection.
ܳ ൌ ඨ൥ቆ ቇ െ ܲଶ ൩ െ (3)
ʹܺ௤ ʹܺ௤
The algorithm follows the flowchart in Figure 5.
The equation that governs this condition is in (4) and, as A. Start
for direct axis, its envelope contains all information necessary
for quadrature parameters identification. The calculated The program starts with the individuals configuration,
envelope equation depicted in (5) and the waveform is shown where each individual is a vector containing all parameters of
in Figure 4. one function, which allows us to say that each individual is a
function itself. The configuration is made by setting the range

ି ᇲᇲ of each parameter value when generated, keeping them
ܸሺ‫ݐ‬ሻ ൌ •‹ሺ߱‫ ݐ‬െ ߜሻ െ ‫ܫ‬଴ ቈܺ௤ െ ൫ܺ௤ െ ܺ௤ᇱᇱ ൯݁ ఛ೜బ
቉ …‘•ሺ߱‫ݐ‬ሻ (4)
according to factual values.

௧ ଶ ௧
B. Initial Population
ି ି
ܸሺ‫ݐ‬ሻ ൌ ඩ‫ܫ‬଴ଶ ൥ܺ௤ െ ൫ܺ௤ െ ܺ௤ᇱᇱ ൯݁
ᇲᇲ
ఛ೜బ
൩ ൅ ʹ •‹ሺߜሻ ൥ܺ௤ െ ൫ܺ௤ െ ܺ௤ᇱᇱ ൯݁
ᇲᇲ
ఛ೜బ
൩൅ͳ (5) The population size is defined by the user, and defines
how many individuals (vectors) there are per epoch. This is
the stage where each individual has its parameters generated,
respecting the ranges previously set.

‹,(((
Figure 6. Simulation Setup.

Figure 5. Flowchart with the GA stages. Figure 7. Simulated terminal voltage for direct axis rejection.

C. Selection
The most important stage, selection is responsible to
evaluate the best individuals, those closest to the actual load
rejection curve. In this work, the evaluation function used is
the Mean Squared Error (MSE) in (6), where the MSE for the
ith individual is the sum of squared errors between each kth
voltage samples from the load rejection (ܸ௟௥ ௞ ) and the
individual function (ܸ௜ ௞ ) by the total number of samples (N).

σே
௞ୀଵ൫ܸ௟௥ ௞ െ ܸ௜ ௞ ൯ (6)
‫ܧܵܯ‬௜ ൌ
ܰ
So the best individual is that with the lower MSE.
Figure 8. Simulated terminal voltage for quadrature axis rejection.
D. Crossing and New Population
Crossing can be done in many ways, but always giving
some sort of advantage to those individuals best evaluated in E. Stopping Criteria
Selection. Crossing is done between the best individual and all
the others. Stage to check if one of the stopping criteria is met: if yes,
the program stops and the results are shown; if not, the flow
The offspring constitute the new population, and each new follows to the check for catastrophe. The criterias used in this
individual parameters are calculated by a weighted average work are either the tolerance for the MSE or the number of
from its parents parameters, it is composed of 80% from the epochs, both chosen by the user.
best parent and 20% from the other.

‹,(((
F. Catastrophe and New Population
Catastrophe is a method that allows the GA to cover a
wider area of possible parameters combinations, which
prevents the algorithm of falling in local minima. It consists of
in every certain number of epochs all individuals but the best
are discarded, and new individuals are generated to replace
them. In this GA, a catastrophe happens at every 10 epochs,
due to the high possibility of local minima.
IV. SIMULATION ANALYSIS
The software SIMSEN is responsible for the simulation of
the circuit depict in Figure 6 [8]. For this circuit, the
synchronous machine (SM1) is set with data from four
different given machines (360 MVA, 263 MVA, 140 MVA
and 888.9 MVA), excited by a voltage source (VS1) and
connected to power line (VS) through two dummy power lines
(LN1 and LN2) and circuit breaker (CB1), which allows the
load rejection. The circuit breaker (CB2) is kept opened to this Figure 9. Comparison between the GA results and load rejection envelope
test. after one rum, for direct axis.

Figure 7 and 8 present the terminal voltages resulting of


the simulation of direct axis and quadrature axis rejections,
respectively, for the first machine. General Electric (GE)
provided the parameters based on real machines. It is
important to notice that the graphics in Figures 7 and 8
resemble the forms in Figures 2 and 4, respectively.
Confirming that the tests conditions are correct for each axis.
V. RESULTS
A bandwidth in the range of 10 to 20 times the rating
frequency of the machine under test is sufficient. Since all four
simulated are 50 Hz machines, a 1000 Hz bandwidth is used to
sample the data (20 times the rated frequency).
The duration of each test depends on the time constant in
study, so that the total time is enough to all transient effects
disappear. Usually 30 to 40 seconds for direct axis and 2 to 3
seconds to quadrature axis.
Figure 10. Comparison between the GA results and load rejection envelope
The GA ran for a population size of 100 individuals, after the second run, just for direct axis subtransient parameters
tolerance of 10-7 for the MSE and maximum of 500 epochs.
The parameters determined are displayed in parallel to the TABLE I. RESULTS FOR MACHINE 1 (360 MVA), TERMINATED BY
rated ones in Tables 1 to 4, along with the absolute error. NUMBER OF EPOCHS.

Sometimes, after one run, direct axis subtransient Parameter Rated


Determined Error (%)
parameters present large errors, what is expect due the short Numeric GA Numeric GA
period of time these parameters effect. To fix that, the Xd (p.u.) 1.1473 1.1423 1.1430 0.44 0.37
algorithm runs once more, varying only the subtransient X’d (p.u.) 0.3631 0.3679 0.3593 1.32 1.04
parameters, keeping the others resulted from the first run, T’d0 (s) 12.2018 12.1484 12.1295 0.44 0.59
aiming also a lower tolerance for the MSE. This method X”d (p.u.) 0.2871 0.2868 0.2818 0.10 1.86
guarantees the curve fitting for the first seconds of the
T”d0 (s) 0.1691 0.1702 0.1669 0.65 1.33
envelope as seen in Figures 9 and 10.
Xq (p.u.) 0.8088 0.7998 0.8022 1.11 0.81
Estimating the same parameters using exponential linear X”q (p.u.) 0.2879 0.2632 0.2786 8.58 3.22
regression (numeric method) inspired by P. de Mello [5] we T”q0 (s) 0.3173 0.3247 0.3116 2.33 1.79
can compare the results of the GA approach. Although, there į (°) -27.249 - -27.468 - 0.81
is no power angle estimation through numeric only method.

‹,(((
TABLE II. RESULTS FOR MACHINE 2 (263 MVA), TERMINATED BY was used a personal computer with 2.40 GHz processor and 8
NUMBER OF EPOCHS.
GB of RAM.
Determined Error (%)
Parameter Rated It is important to point that numeric methods need the user
Numeric GA Numeric GA interference to choose points during linear fitting, while GA is
Xd (p.u.) 1,0331 1.0230 1,0296 0.98 0,34 fully automatic.
X’d (p.u.) 0,2607 0.2546 0,2541 2.38 2,54
T’d0 (s) 9,9797 9.5096 9,6852 4.71 2,95 VI. CONCLUSION
X”d (p.u.) 0,1930 0.1878 0,1902 2.69 1,46 The work showed another way to identify synchronous
T”d0 (s) 0,1120 0.1001 0,1242 10.63 10,85 machine parameters other than using algebraic or numerical
Xq (p.u.) 0,6811 0.6732 0,6741 1.16 1,03 methods. The Genetic Algorithm allows full automation in
X”q (p.u.) 0,1936 0.1843 0,1732 4.80 10,56 the identification process, while taking not much time and
T”q0 (s) 0,2143 0.2339 0,2168 9.15 1,18 computational effort.
į (°) -21,620 - -21,879 - 1,20 With the fitting of the complete envelope curve, this work
does not suffer much with error propagation as methods that
TABLE III. RESULTS FOR MACHINE 3 (140 MVA), TERMINATED BY
goes through exponential linear regression over transient and
NUMBER OF EPOCHS. subtransent periods.
The use of this method to get quadrature axis parameters,
Determined Error (%)
Parameter Rated mainly due the small errors in subtransient quantities, is a
Numeric GA Numeric GA
great value to the industry, as there is a lack in methods to get
Xd (p.u.) 1,0674 1.0602 1,0638 0.67 0,34
them with such precision.
X’d (p.u.) 0,3098 0.3052 0,3038 1.48 1,93
T’d0 (s) 8,5620 8.3136 8,3959 2.87 1,94 ACKNOWLEDGMENT
X”d (p.u.) 0,2352 0.2317 0,2327 1.49 1,06 This work was supported in part by the Global Technology
T”d0 (s) 0,0862 0.0841 0,0941 2.44 9,22 Center - Latin America, GE Renewable Energy - Hydro. The
Xq (p.u.) 0,6996 0.6898 0,6907 1.40 1,28 authors appreciate FAPEMIG, CNPq, and INERGE for their
X”q (p.u.) 0,2317 0.2050 0,1941 11.52 16,23 support to conduct research.
T”q0 (s) 0,1401 0.1582 0,1464 12.92 4,48
REFERENCES
į (°) -22,366 - -22,559 - 0,86
[1] IEEE Std. 115. IEEE Guide for Test Procedures for Synchronous Ma-
chines Part II—Test Procedures and Parameter Determination for Dy-
TABLE IV. RESULTS FOR MACHINE 4 (889 MVA), TERMINATED BY namic Analysis. IEEE Power and Energy Society, New York, 2009.
NUMBER OF EPOCHS.
[2] IEC 60034-4. Rotating electrical machines - Part 4: Methods for deter-
Determined Error (%) mining synchronous machine quantities from tests. Edition 3.0. Interna-
Parameter Rated tional Electrotechnical Commission, Genève, Switzerland, 2008.
Numeric GA Numeric GA [3] IEEE Guide: Test Procedures for Synchronous Machines, IEEE Std. 115
Xd (p.u.) 1,0481 1.0474 1,0444 0.07 0,35 A, 1983.
X’d (p.u.) 0,3462 0.3447 0,3408 0.43 1,56 [4] IEEE Std. 1812. IEEE Trial-Use Guide for Testing Permanent Magnet
Machines. IEEE Power and Energy Society and IEEE Industry Applica-
T’d0 (s) 10,3206 10.0613 10,2986 2.51 0,21 tions Society, New York, 2014.
X”d (p.u.) 0,2951 0.2955 0,2911 0.14 1,36 [5] F. P. de Mello and J. R. Ribeiro, “Derivation of synchronous machine
T”d0 (s) 0,1370 0.1215 0,1468 11.31 7,15 parameters from tests,” IEEE Trans. Power Apparat. Syst., vol. 96, pp.
1211–1218, July/Aug. 1977.
Xq (p.u.) 0,7270 0.7187 0,7195 1.14 1,04
[6] E.C. Bortoni, B.T. Araujo, J.A. Jardini, Estimation of Quadrature Axis
X”q (p.u.) 0,2959 0.2885 0,2823 2.50 4,61 Synchronous Reactance Using the Constant Excitation Test. IEEE Power
T”q0 (s) 0,2693 0.2902 0,2672 7.76 0,77 and Energy Technology Systems Journal, Vol. 3(2), June 2016, pp. 43-
50.
į (°) -23,5040 - -23,7337 - 0,98
[7] E. C. Bortoni and J. A. Jardini, “Identification of Synchronous Machine
Parameters Using Load Rejection Test Data”, IEEE Trans. On Energy
Conv., vol. 17, no. 2, June 2002.
The direct axis first run takes a maximum elapsed time of [8] EPFL, SIMSEN - Simulation Software for Power Networks. Online
13 seconds, while the direct axis subtransient optimization and available at http://simsen.epfl.ch/. École Polytechnique Fédérale de
the quadrature axis runs take a maximum of 9 seconds each. It Lausanne, Lausanne, Switzerland.

‹,(((

You might also like