Professional Documents
Culture Documents
A New Multi-Criteria Fuzzy Logic Transformer Inrush Restraint Algorithm
A New Multi-Criteria Fuzzy Logic Transformer Inrush Restraint Algorithm
Keywords: magnetizing inrush, differential protection, multi- First, a reliable set of criteria signals that can be used to solve
criteria protection, fuzzy logic, transient analysis. the problem of magnetising inrush and internal fault
discrimination is proposed. Except criteria signals the
Abstract operation thresholds are also suggested. Recommended
criteria are aggregated with use of fuzzy logic methods in
This paper presents a new multi-criteria stabilization multi-criteria system. Thorough statistical analyses of
algorithm of transformer differential protection. Proposed performance of the differential protection with application of
scheme bases on new criteria signals and appropriate the new stabilisation algorithm have been carried out, with
operation thresholds. New algorithm employs fuzzy reasoning use of signals generated in prepared digital transformer model
technique for better discrimination of inrush conditions. The (ATP-EMTP) as well as with signals received from field
developed stabilization algorithm has been tested with ATP- measurements for various transformer types.
EMTP generated signals, proving to be reliable and much
more sensitive than standard stabilization algorithms with 2 Criteria signals
crisp settings. In order to select optimal set of criteria signals for
stabilization of the power transformer differential protection
1 Introduction thorough statistical analyses were performed. The analyses
encompassed behaviour of various criteria signals under wide
The differential protection has been successfully used for
scope of power transformer operating conditions. For this
decades to protect power transformers against faults.
purpose EMTP model of power transformer with fragment of
Nevertheless, to assure proper operation the differential
power system was prepared. Using this model over 80
principle has to be supported by additional stabilization
thousand of various cases of internal and external faults as
algorithms which are aimed at avoiding unwanted tripping
well as transformer energisation were generated. Simulating
during transformer magnetizing inrush. Numerous single or
internal faults also turn-to-turn shorts (especially these with
compound criteria are usually used or proposed in the
low number of turns involved) were taken into account. As
literature to discriminate inrush conditions [1, 2, 3, 5, 8],
far as energisation cases are concerned various configurations
among which the second harmonic ratio restraint [4] is most
of transformer operation schemes (loaded, unloaded, supplied
commonly applied. Unfortunately, numerous transformer
from both sides and so on) were studied. This base of signals
protection operation records show that the second harmonic
was used to determine criteria signals, their combinations
restraint as well as other methods may not always be
(mutual relationships) and threshold values. The authors
effective. Magnetic cores of modern transformers are made
analysed courses of changes of various criteria signals within
from amorphous materials, what may be a reason of low level
first 0.5s after inception of disturbance. From this part of
of second harmonic ratio generated during energisation, being
research the following conclusions could be drawn:
insufficient for effective protection stabilization [7].
it is hardly likely to find one universal criterion signal
Difficulties may also arise for the cases of loaded transformer
which could be used to realize (in an unaided way) the power
energisation (e.g. after fault clearing) and ultrasaturation
transformer stabilization task,
conditions [9], when the level of stabilization signal is very
to ensure better and more reliable operation of new
low. Thus, in order to improve protection operation,
stabilization algorithm appropriate combination of various
especially for such difficult cases, new protection stabilization
criteria values should be used,
criteria are required. The criteria signals should be selected to
from the wide scope of criteria signals which had been
meet the following requirements:
analysed an optimum set of signals was found, i.e:
immunity to magnetizing inrush conditions regardless of
ª Kd1h= Id1h /In ± ratio of magnitudes of fundamental
the second harmonic ratio;
harmonic of the differential current and transformer rated
fast operation under internal faults even for severe faults
current;
when CT saturation occurs;
ª Kr2h = Id2h /Id1h ± ratio of magnitudes of second harmonic
sensitivity for low current internal faults (e.g. single turn-
and fundamental harmonic of the differential current;
to-turn faults).
ª KDCoff = IrDCoff /Id1h ± ratio of DC component (reconstructed
This paper presents original stabilization algorithm for
value) and fundamental harmonic of the differential
transformer differential protection under inrush conditions.
current; The DC components are calculated using the algorithm based
ª KDCon = IrDCon /Id1h ± ratio of DC component (measured on full cycle averaging of the current, which corresponds to
value) and fundamental harmonic of the differential signal filtering with 0-order Walsh filter [6]. The measured
current; value of DC component, IrDCon, is determined using values of
ª D1d ± non-saturation interval distortion coefficient of the estimated time constant and initial value of DC component,
differential current [1, 8]; both measured at given time instant. The reconstructed DC
ª D2d ± non-saturation interval distortion coefficient of the component, IrDCoff, is based on actual values of time constant
differential current [1, 8]. and initial value of the DC component only until first
To estimate considered criteria signals the algorithms that are potentially correct estimates of the initial value of the DC
characterized by good dynamics and that at the same time component and time constant are known (about 20 ms after
ensure high ability to attenuate possible distortions were beginning of disturbance). From this moment consecutive
chosen. values of the reconstructed DC component are determined
In order to measure the fundamental Id1h and second harmonic employing first correct values of the DC component and its
Id2h components of differential currents traditional full cycle decaying time constant. Difference between these two
Fourier filters (a pair of sine and cosine filters) were methods of DC component estimation is illustrated in Fig. 1.
employed [6, 7]. Due to CT saturation measured DC component (solid curve)
a)
drops close to zero about 70 ms after fault inception, as
5 shown in Fig. 2b. It means that information about presence of
3 DC component in primary current is lost. On the contrary,
[A]
-1
-3 would not occur.
0 0.1 0.2 t [s] 0.3 0.4 0.5 The last criterion signal, non-saturation interval distortion
b) coefficient of the differential current belongs to the family of
3
measured
reconstructed
direct waveshape analysis methods [1, 8]. This method is
based on an assumption that current waveform in saturation
stage (under inrush condition) becomes similar to sinusoid
2
IdDC [A]
8
saturation interval distortion coefficient during healthy
-0.6 4 transformer energisation with CT saturation is presented. It is
seen that distortion coefficient, D1d, does not reach 0 (its
-1 0
1 2 3 4 5 6 7 8 9 0 0.1 0.2 0.3 0.4 0.5 value exceeds 1 during all simulation) supporting the
t [ms] t [s] hypothesis of magnetizing inrush of the transformer.
Fig. 2. Extraction of the non-saturation interval, The usage of each criterion value was justified, in the context
approximation distortion coefficient calculation. of transformer inrush identification, by the simulative
analysis. Additionally, it must be emphasised that to enhance
simultaneously security, dependability and operation speed of 3.2 Fuzzification block
the differential protection selected signals should be properly
combined. To describe uncertainty of the measured criteria values all
input signals (1)-(8) are fuzzified at the beginning of
3 Multi-criteria fuzzy logic transformer inrush inference process. In fuzzification process the singleton
method was used. As a result of fuzzification process the
stabilization algorithm criterion signal FIX(n) is converted into logic signals
The new proposed stabilization algorithm is a part of power L(FIX(n)), M(FIX(n)) and H(FIX(n)), which then support
transformer differential protection presented in Fig. 3. certain hypotheses. To realize fuzzification process fuzzy sets
Decision-making process starts from estimation of criteria for each input signals have to be defined. To describe
signals described in former section. Obviously, each criterion particular value of input signal following shapes of
value is estimated for all phases. In presented scheme it was membership functions are used:
assumed that the path responsible for external fault and x Low value L(FIX(n)) ± L-function;
overexcitation exclusion (blocks 1, 2 and 3 in Fig. 3) operates x Mean value M(FIX(n)) ± trapezoidal function;
correctly. Thus proposed algorithm has was designed only for x High value H(FIX(n)) ± J-function.
distinguishing transformer inrush from internal fault cases. In All membership functions of each fuzzy input are based on
the next sections description of individual blocks of the analyses discussed in Section 2 as well as WKH GHVLJQHU¶V
proposed stabilization scheme (block 4) is presented. experience. Membership functions of logic signals (fuzzy
settings) for considered input signals are presented in Fig. 4.
CT CT
CB1 CB2 1 1
MEAN LOW HIGH
VT i1 i2 VT
P(FI1)
P(FI2)
u1 u2
Estimation of criteria signals LOW HIGH
Input
criteria Tripping
signals
2. External signal
1. Relay 3. Overexcitation 0 0
fault ruled 0.02 0.05 FI1 7.5 8.0 0.05 FI2 0.1
activated ? ruled out ?
out ? 1 1
Magnetizing LOW HIGH LOW HIGH
4. Multicriteria fuzzy logic inrush ruled out ?
transformer inrush AND P(FI4)
P(FI3)
stabilization algorithm
P(FI6)
All criteria signals were defined in Section 2 but since one of LOW HIGH
them is used in different rule bases, therefore for the sake of
clarity of description of designed fuzzy system the input
0 0
signals (FIX, where X is a number of variable) are recalled 0.9 1.0 FI5 3.0 3.5 0.4 FI6 0.5
here again, given by the following equations: 1 1
LOW HIGH LOW HIGH
FI1 = Kd1h (1)
FI2 = Kd2h (2)
P(FI7)
P(FI8)
1
PI (FO) PN (FO) PF (FO)
Membership functions of sets
PN (FO) - Normal operation
¦P
k 1
B ( FOk )
PI (FO) - Transformer inrush
where: M is an overall number of points considered in
PF (FO) - Internal fault
P(FO)
id mean and maximal operation times which are lower for new
0.2 1. harm
a)
0.1 algorithm than for traditional method (regardless of restraint
0
-0.1
threshold), as shown in Tab. 1.
-0.2
0.1 a) b)
id
0.05 1. harm 45
25
Phase A
Fig. 8. Results of statistical analysis of performance of the
d) 0.5
considered stabilisation schemes for all generated internal
fault cases (over 40000) presented as an occurrence frequency
0
7 of the transformer tripping against operation time for:
6 Phase A a) traditional restraint algorithm set high; b) proposed
CO1
e) 5 algorithm.
4
3 Operation Traditional Traditional Proposed
1 time set low set high algorithm
Trad. algorithm
Decision
f) set high
Mean [ms] 22.2 17.9 16.3
0 Maximal [ms] 134 111 66
1 Proposed Table 1: Results of statistical analysis of performance
Decision
algorithm
signal
g)
(operation time) of the considered stabilisation schemes for
all generated internal fault cases.
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Analysing cases of transformer energisation accompanying
t [s] winding faults the Authors concentrated mainly on situations
Fig. 7. Transformer energisation with ultrasaturation effect: when fault embraces low number of turns (even just one
a), b), c) instantaneous values of differential currents; d) turn). In such cases the differential protection with traditional
second harmonic ratio; e) crisp intermediate input value CO1; stabilization algorithm was completely ineffective, blocking
f) decision signals of differential protection with traditional tripping signal in all analysed cases. The new proposed
stabilization; g) decision signals of differential protection algorithm identified all considered cases and the fastest
with new proposed fuzzy logic stabilization algorithm. tripping took place after 22 ms, while the longest delay was
equal to 312 ms. In Fig. 9 signals observed during transformer
Due to this the second harmonic ratio is very low and for energisation with CT saturation and winding fault (concerned
some period drops close to zero (Fig. 7d). In such conditions small number of turns) initiated at time instant 200ms are
the traditional stabilization method fails (regardless of presented. One can see that once again traditional
thresholds) and transformer would be tripped (Fig. 7f). When stabilization was insufficient and differential protection was
new stabilization algorithm is applied one can see that blocked (Fig. 9f). Although fault current is hardly noticeable
intermediate input value CO1 (Fig. 7e) does not overreach 7.5 in differential current the new algorithm reacts correctly.
level what guaranties proper stabilization in this situation. Intermediate output CO2 exceeded threshold value 24ms after
Results of statistical analysis of considered stabilization fault inception (Fig. 9e) leading to transformer tripping (Fig.
algorithms performance for internal fault cases are presented 9g).
in Fig. 8. One can find there occurrence frequency of the
energisation fault
10 References
id /IN , L2 [pu] id /IN , L1 [pu]
id
5 1. harm
a) [1] D. Q. Bi, X. A. Zhang, H. H. Yang, G. W. Yu, X. H.
0 :DQJ:-:DQJ³&RUUHODWLRQDQDO\VLVRIZDYeforms
-5 in nonsaturation zone-based method to identify the
2 PDJQHWL]LQJ LQUXVK LQ WUDQVIRUPHU´ IEEE Transactions
id
0 1. harm on Power Delivery, Vol. 22, No. 3, pp. 1380-1385, (July
b)
-2 2007).
[2] Z. BR * :HOOHU 7 /RPDV ³$ QHZ WHFKQLTXH IRU
-4
2 WUDQVIRUPHU SURWHFWLRQ EDVHG RQ WUDQVLHQW GHWHFWLRQ´
Id2h /Id1h [pu] id /IN , L3 [pu]
set low
signal
f)
set high Signal Processing in Power System Protection and
Control", Springer Verlag, Series: Signals and
0 Communication Technology, London, (2011).
1 Proposed [7] + 8QJUDG : :LQNOHU $ :LV]QLHZVNL ³3URWHFWLRQ
Decision
g)
Dekker Inc., New York, (1995).
0
[8] = :DQJ < 0D < ;X / 0D ³$ 1HZ 3ULQFLSOH RI
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Inrush Current and Internal Fault Current of
t [s] Transformer based on Self-&RUUHFWLRQ )XQFWLRQ´ 7th
Fig. 9. Transformer energisation, then turn-to-turn internal International Power Engineering Conference, IPEC
fault: a), b), c) instantaneous values of differential currents; d) 2005, Vol. 2, pp. 614-617, (29 Nov.- 2 Dec. 2005).
second harmonic ratio; e) crisp intermediate input value; f) [9] A. Wiszniewski, W. Rebizant, D. Bejmert, L. Schiel,
decision signals of differential protection with traditional ³8OWUDVDWXUDWLRQ SKHQRPHQRQ LQ SRZHU WUDQVIRUPHUV-
stabilization; g) decision signals of differential protection myths and realit\´ IEEE Transactions on Power
with new proposed fuzzy logic stabilization algorithm. Delivery, vol. 23 nr 3 pp. 1327-1334, (2008).
5 Conclusions
The problems of effective power transformer inrush current
discrimination and at the same time reliable operation of
differential protection appear very essential especially in the
light of continuously increasing requirements of utilities to
prevent unwanted outages with the consequential costs. In the
paper new stabilization algorithm of the power transformer
differential protection is presented. Proposed stabilization
method employs several criteria signals combined with use of
fuzzy logic system. Stabilization scheme is described in detail
and validated using digital simulation (EMTP model).
The proposed multi-criteria fuzzy logic transformer inrush
restraint algorithm in comparison with traditional stabilization
method shows higher security and sensitivity as well as
higher immunity to CT saturation.