Vous êtes sur la page 1sur 5

176 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 16, NO.

2, APRIL 2001

A Fundamental Approach to Transformer Thermal


ModelingPart II: Field Verification
Glenn Swift, Life Senior Member, IEEE, Tom S. Molinski, Member, IEEE, Ren Bray, and
Robert Menzies, Senior Member, IEEE

AbstractThis paper has two main objectives. One is to show


that the top oil rise thermal model proposed in the companion
paper is valid, for a large power transformer in service. The second
is to show that there is a convenient way of estimating the param-
eters without removing the transformer from service. A Manitoba
Hydro 250 MVA OFAF transformer was chosen and instrumented
with data-gathering equipment. Two twenty-four hour test runs
were performed, one in February of 1999 and the other in July of
1999. The most basic parameter to be determined was the rated top
oil rise but also found were the top oil time constant and the non-
linearity exponent, commonly given the symbol . The results are
very positive.
Index TermsPower transformer protection, temperature,
thermal factors.

I. INTRODUCTION

I N THE companion paper [1] a new formulation of the gov-


erning differential equation was derived. It is important to
test the validity of the approach, preferably with a large power
transformer as the subject. To this end, special instrumentation
was installed on a 150/200/250 MVA OFAF transformer on the
Manitoba Hydro system. The transformer is shown in Fig. 1.
Some thermal parameters were known for this transformer;
others were not. Factory test records were available and detailed Fig. 1. 250 MVA transformer used for this study.
in some respects but, for example, the top oil time constant had
not been measured during the factory tests in 1980. If planned or The ambient temperature information comes from
unplanned overloading is anticipated for a transformer, the top a 420 mA sensor mounted in the switchyard in a
oil time constant is useful information, and is further incentive meteorologically-approved manner. See Fig. 2. It is con-
for in-the-field testing, in this case. nected to the relay by means of a shielded cable.
In this paper, a new method of determining thermal param- The top oil temperature information comes from a Qual-
eters for a transformer in the field and without artificially ap- itrol installation that replaces the normal liquid temperature
plying a step load change (since this is often not possible) is meter. This device also supplies a 420 mA signal to the relay.
presented. The study is limited to top oil temperature; hot spot In other installations, a simpler top oil sensor can be used, if
temperature is not considered here. features such as LCD read-out and staged output alarms are not
required.
II. INSTRUMENTATION One of the functions of the relay is the capability to record
trend information: i.e., variables at a slow sampling rate (e.g.,
The input variables are load current and ambient temperature.
once per three minutes), over long periods (e.g., thirty days). For
The output variable that is the object of the model is the top oil
this study, the data were downloaded via modem and telephone
temperature. A commercial T-PRO relay/recorder manufac-
link to a remote location and into a spread-sheet for analysis.
tured by the firm employing one of the authors has the capability
The instrumentation arrangement is shown in Fig. 3.
of receiving not only load current (from the utilitys CTs) but
also ambient temperature and top oil temperature.
III. FIELD TESTS
Manuscript received December 8, 1999. The monitoring system was installed in mid-1998 and has
G. Swift is with APT Power Technologies, Winnipeg MB, Canada been logging ambient temperature, load current and top oil tem-
T. S. Molinski and R. Bray are with Manitoba Hydro, Winnipeg MB, Canada perature continuously, at three-minute intervals, for long-term
R. Menzies is with the Department of Electrical and Computer Engineering,
University of Manitoba Winnipeg MB, Canada. study purposes. Three transformers operate in parallel with the
Publisher Item Identifier S 0885-8977(01)01542-4. subject transformer, all at about fifty percent peak load. In order
08858977/01$10.00 2001 IEEE
SWIFT et al.: A FUNDAMENTAL APPROACH TO TRANSFORMER THERMAL MODELINGPART II: FIELD VERIFICATION 177

test period, as described below. By this means the loading was


increased to a much higher value, though currents over one per
unit were not possible.
Field testing was done on two occasions: cold weather, Feb-
ruary 2021, 1999, and hot weather, July 34, 1999. In both
cases, the load was raised to as high a level as possible, by
switching out three transformers in parallel with the test trans-
former. Also in both cases, the loading was not high enough
throughout the test to keep the fans and pumps on, so these were
turned on manually since the objective was to find the model ap-
plicable during overloads, when of course, the fans and pumps
would be on. The tap position was also recorded, at regular in-
tervals, so that a correction factor could be applied to the results.
The purpose of the two tests, rather than just one, was to
see how well parameters determined under one condition (hot
weather) would apply under a quite different condition (cold
weather). The July test is taken as the standard since the same
loading level would result in higher temperatures than for the
February test, and therefore be closer to an overload condition.
The procedure for each of the two tests was as follows:
Fig. 2. Ambient temperature sensor (inside circular housing). 1) Select a week-end acceptable to the utility operators, and
in accordance with load-flow studies.
2) Start the 12-hour run-in part of the test at noon on
Saturday, by switching out a planned number of parallel
transformers, and manually turning on the fans and
pumps.
3) Record data for 36 hours, of which 12 hours would be the
run-in period, and the next 24 hours would be the data
used for analysis, i.e., for determination of the thermal
parameters required. Thus for each variable, the analysis
data would comprise 480 points, taken at 3 minutes per
point.
4) Download data for analysis.
The July test data is shown in Fig. 4. The data is shown for
two days. It is used as follows:
First 6 hours: pre-test period data, pumps and fans off, all
transformers on line (normal).
Next 12 hours: run-in period data, pumps and fans on, three
parallel transformers out.
Next 24 hours: actual test period data, ditto, pumps and
fans on, three parallel transformers out.
Last 6 hours: post-test period data, pumps and fans back to
automatic, all transformers on line.

IV. DATA ANALYSIS AND RESULTS


A. Differential Equation of the Top Oil Temperature Model
From the companion paper [1] for this study, the differential
equation for which parameters are to be estimated is as follows:

(1)

where
is time, the independent variable,
Fig. 3. Instrumentation. is the load current in per unit, a forcing function
variable,
to increase the loading to near-rated or above, all three of the is the ambient temperature, a second forcing func-
parallel transformers were removed from service during each tion variable,
178 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 16, NO. 2, APRIL 2001

Fig. 4. July 34 recorded test data. Measured load currentlight line; measured top oil temperatureupper heavy line; measured ambient temperaturebottom
heavy line.

is the top oil temperature, the solution variable, Initial values for the sought parameters are guessed at and a
is the ratio of copper loss to iron loss at rated load univariate search technique [3] is used. Each parameter is varied
( ), one at a time until the objective function is minimized in that
is the rated top oil temperature rise above ambient, direction. Then a new search is carried out in the nest parameter
is the top oil time constant, and direction starting from the previous minimum. This is repeated
is an exponent due to the nonlinear heat transfer over all three parameter directions until no further minimization
relationship. of the objective function takes place.
The corresponding difference equation, also from the com- The starting value for , that is , is set equal to
panion paper, is the measured value of at the start of the 24-hour test period.
Incidentally, it turns out that the results are quite insensitive to
this value.
When (3) cannot further be minimized, the optimum values
(2) of the parameters , , and have been found and can be
where used for any ambient temperature or loading condition, provided
is the small difference operator. that the pumps and fans are on, and that the tap position is not
If no parameters are known, then the last four items of the too far off (discussed below).
foregoing list of eight above, are to be found. For this study,
one of the parameters was known accurately a priori: the load- C. Results for July Test
loss-to-no-load-loss ratio from factory tests. The optimization procedure yielded the following values:
Thus the three parameters to be determined are C
which can be compared with the factory test value, days hours
which was not measured at the factory, and .
which might be presumed unity, but not assumed so Fig. 5 shows the top oil temperature measured and calculated
here, as a matter of interest. curves. They are so close that it is difficult to distinguish them
at this scale resolution.
B. Parameter Estimation Method It was surprising that was 0.9 rather than 1.0, which is the
The parameter estimation procedure is based on a nonlinear usual assumption for directed-flow oil cooling, as was the case
integral-squared error objective function [2] defined by here. The optimization minimum occurred for ,
for both tests, so there is little doubt about the validity. The im-
(3) plication is that the cooling is not as effective as one might as-
sume, for this transformer. The cooling is, however, adequate,
where since the top oil rise agrees well with the factory value, as men-
is a vector whose elements are the three parameters tioned below.
listed above,
is the top oil temperature at each time step as calcu- D. Application to the Conditions of the February Test
lated from (2), The very good fit of the previous section is all very well, but
is the measured top oil temperature at each time there needs to be a check under other conditions. Ideally, these
step, and would be overload conditions, but that was not easily possible
is the index for each 3-minute time step over the because of system loading conditions at the time of the tests. In-
24-hour period. stead, a comparison is made under drastically different ambient
SWIFT et al.: A FUNDAMENTAL APPROACH TO TRANSFORMER THERMAL MODELINGPART II: FIELD VERIFICATION 179

Fig. 5. Measured (jagged trace) and Model (smoother trace) top oil temperature.

Fig. 6. July TestFebruary Test comparison. Upper curves: July Measured (jagged) and Model (smooth). Lower curves: February Measured (jagged) and Model
(smooth) using July parameters.

conditions, which occur in February, because in the Winnipeg where


area there are extreme differences between summer and winter known losses for desired case, tertiary unloaded,
ambient temperatures. 711 kW (from test report),
If the same values of the three parameters are used with the known losses for base case, tertiary loaded,
February load current and ambient temperature input data, the 768 kW (from test report).
result is as shown in Fig. 6, where the curve for the July test is Therefore, the correction factor here is
repeated so as to make the comparison clear.
If the three parameters are optimized for the February test
alone, the results are:
C Since the factory test top oil rise for the case of the tertiary
days hours loaded was 37.1 C, it follows that the temperature to be used
. for comparison is 37.1 C C.
With these values, the measured and model curves for Feb- This is considered to be in relatively good agreement with the
ruary (lower curves of Fig. 6) become as indistinguishable as result for the July test, 33 C.
those for the July test (upper curves of Fig. 6), but of course a As mentioned before, there is no factory test data for the time
single set of the parameters must be used for all cases. Using constant, which is a good reason for conducting tests like the
only the July results, it can be seen in Fig. 6 that the worst one described here.
error in the February plot is about one degree, which is certainly One other factor for this transformer is the position of the
acceptable. on-load tap changer position. The range was neutral eight
positions. The factory tests were done for the neutral position.
E. Comparison with Factory Measurement
During the July tests the position was 1 and for the February
The factory measurement of the top oil rise was made with tests the position was 2. These correspond to increases in the
the tertiary winding loaded whereas the installed transformer losses of about 2.5% and 5% respectively, which affects the ap-
tertiary is not loaded. A correction factor can be derived from parent error negligibly.
the companion paper [1, (6)]: Incidentally, the losses for this transformer are relatively
flat for positive-going tap positions, but rise sharply for
negative-going tap positions. At a tap position of 8 the
180 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 16, NO. 2, APRIL 2001

load-loss-plus-core-loss is 30% higher than it is for the neutral ACKNOWLEDGMENT


position. This means that corrections should be made if it The authors would like to thank J. Kane, Network Studies
is known that the tap position will be well into the negative Engineer at Manitoba Hydro, for his system studies prior to the
region, for this transformer. tests. The authors would also like to thank J. Rempel, Protection
Maintenance Technical Officer and his colleagues at Manitoba
F. Practical Benefits Hydro, for their valuable support both during installation of the
Loss of Life as a Replacement Criterion: Manitoba Hydro is test apparatus and during the tests.
actively pursuing use of the latest recommended loss of life (of
the cellulose insulation) calculation procedures as a criterion for REFERENCES
transformer replacement [4]. According to the latest Guide [5], [1] G. Swift, T. Molinski, W. Lehn, and R. Bray, A fundamental approach
it is no longer recommended that typical values of the thermal to transformer thermal modelingPart I: Theory and equivalent cir-
parameters be used. This makes the on-line determination of cuit, IEEE Trans. on Power Delivery, vol. 16, no. 13, 2001.
[2] A. O. Converse, Optimization, NY: Holt, Rinehart and Winston, 1970,
parameters attractive, especially when some of the parameters p. 83.
are not known from factory tests. [3] G. S. G. Beveridge and R. S. Schecter, Optimization: Theory and Prac-
The Manitoba Hydro methodology is to predict the optimum tice, NY: McGraw-Hill, 1970, p. 363.
[4] T. J. Molinski and G. W. Swift, Reducing the life-cycle cost of power
replacement date for a particular transformer using predicted transformers, in CEPSI1 Proceedings, Kuala Lumpur, Malaysia, Oct.
load growth and historical ambient temperatures, with hourly 2125, 1996.
data that takes into account daily, weekly and annual patterns. [5] IEEE Guide for Loading Mineral-Oil-Immersed Transformers, IEEE
Standard C57.91, 1995.
Ideally, the thermal model for such calculations should change
as the cooling mode changes, as triggered by hot spot tempera-
ture sensor, that is, winding temperature indicator settings.
Glenn Swift (M56SM75LM96) is part of the research team at APT Power
Using this methodology along with such precautions as Technologies in Winnipeg. He received the B.Sc. and M.Sc. degrees from the
on-line dissolved gas analysis, transformers may be replaced University of Alberta, and the Ph.D. degree from the Illinois Institute of Tech-
safely at a date a few years beyond the date determined by nology. He is a Professor Emeritus and Senior Scholar in the Department of
Electrical Engineering at the University of Manitoba. His industrial employment
the present rather conservative approach. Many thousands of has included Westinghouse Canada, Federal Pioneer Electric, Ontario Hydro,
dollars are saved for each such delayed replacement, without Manitoba Hydro and Metropolitan-Vickers Electrical Co. Ltd. in the United
significant increase in risk [4]. Kingdom.
Failure or Degradation of the Cooling System: The method
of determination of thermal model parameters would be quite
practical for an on-line parameter determination calculation Tom S. Molinski (M93) received the B.Sc. and M. Eng. degrees in electrical
engineering (1975 and 1985, respectively) from the University of Manitoba. He
that automatically and continually estimates the key thermal was a member of the System Performance Department at Manitoba Hydro for
model parameters. There might be three models, one applicable twenty years, working mainly on the analysis of protective relaying systems. In
to each of the cooling system modes, for example: no forced 1995, he became the Supply-Side Enhancement Engineer with the same utility,
responsible for overseeing supply-side efficiency improvements, investigating
cooling, pumps and half fans, and pumps and all fans. A sudden distributed generation technologies, as well as procurement and inquiries re-
change in the value of one of the parameters, e.g., the rated garding nonutility generation for Manitoba Hydro. He is an Active Member of
top oil rise, could indicate that the fans or pumps have failed. the Winnipeg Chapter IEEE Power Engineering Society.
A gradual increase in the calculated rated top oil rise might
indicate that the radiators are becoming increasingly fouled in
some way. Ren Bray received the B.Sc. degree in electrical engineering (1988) from the
University of Manitoba. From 1988 to 1998, he was with Federal Pioneer Ltd.,
now Pauwels Canada Inc. where he was the Manager and Design Engineer
V. CONCLUSION with the Quality Assurance and Test Department. In 1998, he joined Manitoba
Hydro where he is responsible for apparatus procurement, including specifica-
A fundamental thermal model for top oil temperature deter- tion, tender evaluation and witness tests.
mination is shown to be valid for a 250 MVA transformer.
The parameters for the model can be found from a test that
does not involve removing the transformer from service, namely Robert Menzies (M67SM75) received the B. Eng. degree in electrical engi-
minimization of the integral-squared-error over one day of test neering from McMaster University and the Ph.D. degree from the University of
St. Andrews. Since 1967, he has been with the University of Manitoba where he
data. is presently Professor and Head of the Electrical and Computer Engineering De-
The parameters found under one condition of ambient tem- partment. He has worked with Westinghouse Canada on large induction motor
perature are found to be valid under a radically different ambient optimization, with Brown Boveri, Switzerland on inverter locomotive drives and
the control of static compensators, and with ABB Drives on large GTO inverter
temperature as well. drives. His research interests include the design of electric machines and the ap-
The agreement with factory tests is good. plication of power electronics to power systems.

Vous aimerez peut-être aussi