Sunteți pe pagina 1din 7
A COMPARATIVE ANALYSIS OF PROTECTION PRINCIPLES FOR MULTI-CRITERIA POWER TRANSFORMER RELAYING Bogdan Kasztenny — Engeniusz Rosotowski ‘The Technical University of Wroclaw Wroclaw, POLAND Key words: Protection, ‘Transformer differential relaying, Decision making Abstract - The paper presents a survey and com- parative analysis of protection criteria for po- wer transformers. A dozen of criteria is eva- luated in terms of quality and speed of provided recognition as well as numerical complexity. This estimation based on statistical behaviour of the criteria, is given in a quantitative way - the above qualities are displayed by numbers in a formal scale. Also, the overall efficiency of each criterion is approximated as the ratio of its, recognition quality and numerical complexity. Based on the results of analysis, a new dynamic *p out of q" decision making principle is presen- ted for transformer relaying. 1. INTRODUCTION The differential principle has been recog- nized as the basis for the protection of power transformers. This criterion, although superior to other protection laws, in the case of a power transformer shows certain limitations: The dif- ferential current may be caused by certain non- internal-fault disturbances. The conventional approach to mitigate that problem is to apply percentage differential characteristic along with 2nd and 5th harmonic restraints. The advanced digital versions of such classical scheme employ Fourier methods [1], Kalman filtering techni- que [2] and optimal state observers [3] in the domain of measurement. Also, a number of new protection criteria have been elaborated [4] in- cluding A-differential principle [5], direct wave shape identification [6], protective algorithms based upon electro-magnetic equations of a protected transformer [7], and" adaptive. ap- proaches [8], to name only a few recognition techniques introduced in the last decades. However, unlike the measuring algorithms for power transformer relaying [1], the protec- tion criteria themselves (for example, 2nd har- monic restraint as a protection law regardless the estimators of the involved harmonics) have not been compared and evaluated in terms of their numerical complexity, reliability, quality and speed of provided recognition. On the other hand, a trend for designing a multi-criteria relay for power transformers is observed [9,10]. The multi-criteria approach or other adaptive actions, however, call for know- ing not only the reliability of each criterion used, but the dependence of its recognition power on other factors (including time) as well. Murai Mohan Saha Birger Hillstrom ABB Network Partner AB Vaster’s, SWEDEN In order to meet these needs, this paper has been motivated by three targets: First, it delivers a survey and comparative analysis of selected protection criteria for po- wer transformers. Twelve criteria are presented (Sec. 2) and evaluated (Sec. 3 and 4). Second, this paper introduces the formal platform for comparison of various protection criteria (Sec. 3 and 4). This comparison is based upon statistical behaviour of studied criteria stimulated using a large number of testing cases. The criteria are evaluated in terms of speed of provided recognition, quality of recog- nition (as a function of time), and computation- al complexity. This estimation is given in a quantitative Way - the above qualities are dis- played by numbers in a formal scale. Also, the overall efficiency of each criterion is’ ap- proximated as the ratio of its recognition power and numerical complexity. Third, this paper presents the application for the results of analysis (Sec. 5). Based upon evaluated recognition qualities of the presented criteria, a kind of dynamic "p out of q" rule (any p out of total number of q criteria must be satis- fied to make a decision) has been developed for a multi-criteria power transformer relay. 2. PROTECTION CRITERIA FOR POWER TRANSFORMERS 2.1, Modes of operation of a power transformer {as a protected plant In this paper, the following modes of opera- tion of a power transformer have been distin- guished from the point of view of protective relaying: (a) inrush condition, (b) stationary over excitation of a transformer core, (c) external short-circuit combined with satur- ation of Current Transformers (CTs), (d) external short-circuit or high load current without saturation of CTs, but mixed with mismatched ratios of a protected trans- former and CTs, (e) internal fault, (f) normal operation. Certainly, the goal for any protective relay for a power transformer is to recognize on-line these five cases. Usually, it is assumed that the protected transformer leaves the mode of normal opera- tion (case (f)) when its relay activates. The in- 12th Power Systems Computation Conference Dresden, August 19 - 23, 1996 107 stantaneous overcurrent principle is commonly employed to pick-up the relay. In this paper, the relay activates if: liapn (=) >Ho qa) where: inp differential current related to the phase ph (ph = R, S or T); The differential and ‘through currents are formed from the pha- se currents from both sides of a transfor- mer according to the art of differential relaying (1]-[11]: the differential current - as the difference between the currents from both sides, while the through current - as the average. Ho an appropriate threshold (Ho = 0.02 pu in this paper), n discrete time index (the activation instant is marked by n=0). According to (1), a non-zero differential cur- rent in any phase activates the relay in all three phases at the same time. When activated, the relay issues the tripping command if, based on information contained in the relaying signals, it is capable of rejecting the non-internal-fault hypotheses (a)-(d). This approach is convergent with the common prac- tice of protective relaying for power trans- formers - instead of confirming internal fault, the relay rules-out the remained suppositions. 2.2. The protection crite: Twelve criteria (C1 .. C12) gathered in four groups (CASE a .. CASE d) have been analyzed. Each criterion is defined as a certain relation between specified measured quantities and ap- propriate thresholds (#1 .. #12) Case (a) The magnetizing inrush may be ruled-out if: Criterion C1: The value of the differential current is higher than the highest expected in- rush current level (instantaneous overcurrent principle). Certainly, this criterion is not able to ex- clude inrush conditions during internal faults with low values of the differential current (turn-to-turn faults as an example). Neverthe- less, for some population of internal faults it provides quite fast and reliable recognition. ‘The open question is whether to use the ab- solute value of a present sample of the differen- tial current or its fundamental component amplitude or even some combination of them. In this paper, the amplitude is used as the cri- teria signal: Taxco) > Hi @ where: Is; amplitude of the i-th harmonic of the differential current, In this paper, the protective relay is assumed to be a complex of three identical sub-relays, each for one phase. Thus, the condition (2) is to be checked in all three phases (for simplicity of notation the phase index is omitted). For two of. the presented criteria (C2 and C11), however, the checking is done in all three phases at once. Criterion C2: The waveshapes of the dif- ferential currents in all three phases do not show certain fragments (lasting no less than 1/6 of a cycle) where the levels of both the current and its derivative are close to zero (direct waveshape identification) [9], During heavy internal faults accompanied by CTs saturation, the differential current may also show such’ periods. However, these inter- vals, when observed in all three phases, are shifted in time; while during inrush conditions they are exactly synchronized. In this paper, the criterion C2 is formalized in two alternative ways: © C2a (both the current and its derivative ta- ken into account): First, the auxiuliary signal Sa) is formed as: z ING x IN6-1 Se =D _Y liapna-wl+G2 Dl apnea K=0 @) where: N number of samples per cycle, [] rounding operator, i*s _ rescalled first derivative of the differential current computed as: aa) = 2 (ia iawn) (4) G2 an auxiliary setting Finally, C2a rejects the inrush suposition if. wi Senn) > Ha © © Cap (only the current level taken into ac- Sa, Min Max Uaph(n—k~m) > Hw 6) Paes seee eT A 2 Criterion C3: The second harmonic in the differential current is below some 10-15% of its fundamental (2nd harmonic restraint): Ts240) < H3 Ian) a) When using a criterion of this type (har- monic restraint) one must ensure the ap- propriate relation between time responses of estimators of the involved harmonics. In the case of C3, to avoid false trippings, the es- timator of [a2 must be a little bit faster than the estimator of Jas; otherwise an extra delay must be introduced. ‘CASE (b) ‘The stationary over excitation of a transformer core may be ruled-out if: Criterion C4: The level of the differential current is higher than in cases of transformer 108 12th Power Systems Computation Conference Dresden, August 19 - 23, 1996 over excitation (overcurrent principle): Tana) > He (8) The overcurrent criterion is repeated here. But it is to be noted, that Cs is dedicated to sta- tionary over excitation, while C1 recognizes in- rush conditions. The setting H4 is usually much lower than Hj. Thus, for some cases of internal faults, C4 is able to exclude over excitation, while C1 is not able to rule-out inrush. How- ever, in multi-criteria approach, the inrush hypothesis may be excluded based on some other criteria enabling the tripping command to be sent. In this way, by multi-stage analysis of the differential current, one earns opportunity to improve the relay, Criterion Cs: The integral of the terminal voltage amplitude, V, for half a cycle, which reflects the flux in a transformer core, is below the saturation level (simplified flux based restraint) [10,11] INzI-1 TY Ven) < Hs (9) x0 where: 7} sampling period. Criterion Ce: The level of the Sth harmonic in the differential current is below some 30% of its fundamental (Sth harmonic restraint): Tasca) < He Taxa (10) Case (©) ‘The external short circuit mixed with saturation of CTs may be excluded if: Criterion C7: The large value of the through current did not exist during the cycle before the large value of the differential current was detected (sequence of events) [10] ‘The criterion is based on the observation that CTs usually transform accurately for at least 1/4 of a cycle after the fault inception, before they get saturated. Thus, during an exter- nal short circuit, the high level of the through current is detected before the differential cur- rent becomes high due to saturation of CTs. In this paper, the A-differential approach [5] is used when formalizing the criterion C7. This approach, by subtracting the appropriate pre- fault values, consideres only the fractions of dif- ferential and through currents caused by a fault. Thus, C7 is written as follows: r-sampleat which: Iria) > G7 (1) 5 - sample at which: Max(a)—Tatp|>H7 [R1¢@)FRp| (12) where: Ini amplitude of the fundamental frequen- cy component of the through current, Toxp, nip pre-fault amplitudes of the differen- 12th Power Systems Computation Conference Dresden, August 19 - 23, 1996 tial and through currents, respectively, a an auxiliary setting. The criterion C7 is satisfied if: eee (13) Certainly, when the marker s is not activated the criterion remains blocked. Criterion Cs: The level of the 2nd harmonic in the differential current is below some 20% of the fundamental component [10]. Tan) < He Taxa) (14) Again, the 2nd harmonic restraint was pre- viously used to rule-out the inrush hypothesis (C3). The goal for Cs, however, is different: it excludes the hypothesis of CTs'saturation, and it is founded on the fact that CTs, when saturated, produce significant amount of 2nd harmonic. Certainly, the settings of C3 and Cg are different. Criterion Cy: The differential current is greater than the greatest current expected during external short-circuit under CTs satura- tion (overcurrent principle) Toa) > Ho (1s) Case (a) ‘The external short-circuit or high load current without CTs saturation may be ruled-out if: Criterion Cio: The differential current is much greater than the through current (biased differential characteristic). In this paper, to gain sensitivity, the A-dif- ferential rule is applied [5] Ma1(ny—Jaip| >H10 [JRa¢)—ZRip | (16) Criterion Cir: The relations between amplitudes of the differential and through cur- ents are different in all three phases of the relay (asymmetry checking) [10]. The criterion actually detects asymmetry of the currents (an internal three-phase symmetri- cal fault in a power transformer is practically impossible). The auxiliary signals Wpn(a) for all the phases are first computed as: ay ‘The criterion is satisfied if: [Wacay—Ws¢ay| + |Ws¢ay— Wray + (18) [Wra)— Wray] > Har Criterion Ciz: The differential current is greater than the greatest expected value of the current caused by a near external fault and largest possible mismatch of the transformer and CTs ratios: 109 Toy) > Ha (19) 3. OPTIMAL SETTING OF THE CRITERIA 1. Testing cases In order to evaluate and compare the studied criteria, a data base containing more than 150 important cases of transformer operation has been collected. This data base consists of: (a) transformer energizing cases (16%), (b) stationary over excitation cases including overvoltages and frequency reduction (7%), (©) external faults with and without saturation of CTs (24%), (4) internal faults including terminal, inter-win- ding and turn-to-turn faults (53%). Digital ATP-based model of a three-phase and two-winding 5.86MW, Yd-connected, five- leg core type power transformer provides the input for the differential relay. The most impor- tant factors taken into account by the model in- clude [12]: - representation of both saturation and hysteresis loop of a transformer iron core, - feasibility for inputting a residual flux, - repre- sentation of main CTs in terms of their possible saturation, - representation of relay input ci cuits with relay CTs and anti-aliasing analog fil- ters, - feasibility for modelling turn-to-turn in- ternal faults. During the simulation, certain random vari- ables have been distributed uniformly pro- viding diversity of the studied cases. These vari- ables include: - residual magnetism, - voltage angle at the disturbance beginning, - fault loca- tion and resistance, - amount of short-circuited turns, - type of fault, - pre-fault transformer burden, - saturation level of CTs, - mismatch of transformer and CTs ratios and power system configuration. Like the simulation phase, the digital measuring unit is assumed to be common for all the analyzed criteria, and it is based upon Finite Impulse Response (FIR) full-cycle or- thogonal filters designed using the least-square method with perfect separation between the harmonics [13]. The sampling rate is assumed 1kHz (N = 20 samples per cycle) 3.2. Optimal setting To use a criterion, one needs to draw a set- | 1 @ oo 460 TIME {me} Fig. Distribution of the criteria signal 142 /1A1 2nd harmonic restraint-C3) for inrush conditions (a) and internal faults (6). ting for it. Statistical analysis may provide some recommendations for the sought settings: Using the established data base, distributions of the criteria signals for the key hypotheses (inrush versus internal fault, as an example) have been found. Fig.1 shows time distributions of the ratio of 2nd and Ist harmonics amplitudes of the differential current for inrush conditions (a) and internal faults (b). The figures are ob- tained by plotting the signal [42/41 on the same plane for all the coliected cases (for in- rush cases in Fig.la and for internal faults in Fig.1b). Comparing the figures and observing the overlapping region between the distribu- tions, one may evaluate the 2nd harmonic restraint (C3) as the protection criterion and choose an appropriate threshold for it In this paper, however, more formal techni- que has been used. Each criterion is set to mini- mize the performance index J: J=CP+(01-OPa tA tip (20) where: A. weights enabling to change the inclination of a criterion under setting between blocking (C=1, A low) and tripping (C=0, A high) tendencies, of course: 05 C1; in this paper C = 0.5 and.A = 1 [%/ms}, 1, P2 percentages of false trippings and missing operations caused by a studied criterion, respectively, trip average tripping time for internal faults. The optimization process for the criterion Ci is illustrated in Fig.2: Certainly, amplitude of the differential current being the criteria sig- nal for C12, shows some overlapping between internal faults and external short-circuits mixed with ratio mismatch (it should be remembered, that this criterion is established to distinguish these two cases). It is so, because certain population of internal faults is accompanied by lower differential current than some external faults mixed with ratio mismatch. Increasing the threshold Hyz, one reduces the percentage of unwanted operations of C12; for Hiz higher than 3.36 [A] the criterion is completely stable. On the other hand, rising this threshold, one gains the percentage of missing operations of C12, P2. ‘The average tripping time is more or less con- Tat (b) ° 20 40 60 1E [ms] 00 110 12th Power Systems Computation Conference Dresden, August 19 - 23, 1996 PERFORMANCE INDEX. is fas 3 33 4 a 3 3 ‘THRESHOLD [A] UNWANTED TAIFPINGS, MISOPERATIONS [) Fig.2. Results of optimization of the criterion C12 (over- current principle). The optimal setting H12 = 3.36 [A] gives the performance index J = 16.82. stant and oscillates around 6ms regardless the setting. As a result of that, the performance index, J, reaches its minimum (16.8) for Hiz 3.36 [A]. For this value, Ciz is completely stable (P} = 0%) and able to operate for 79% of, all the internal faults (P2 = 21%) with average tripping time of 5.82ms. Certainly, the optimal settings differ be- tween individual transformers and power system configurations. In this study, however, the criteria are set in order to make their evalua- tion immune to their settings. 3.3, Results of optimization C3 (2nd harmonic restraint) is the best criterion in the group recognizing inrush condi- tions. C2 (direct wave shape identification) has also low values of decision errors Py and P2, but it needs at least one cycle to operate. Cy (over- current principle), in turn, is fast, but it does not operate for 39.5% of all the internal faults. Co (Sth harmonic restraint) is the best protection principle provided for excluding over excitation conditions. The criterion of the sequence of events (C7) is the most valuable criterion when ruling-out the hypothesis of ex- ternal short-circuit under CTs saturation; while A-differential principle (Cio) is the best when considering external faults without CTs satura- tion, but mixed with ratio mismatch, Let us observe the overcurrent criterion (Ci, C4, Cy and Ci2). When used to exclude inrush conditions (C1, setting at 11.64 [A]), it operates for 60.5% of all the internal faults (mostly for short-cireuits at the transformer terminals); while when used to exclude external fault com- bined with ratio mismatch (Ci2, setting at 3.36 {A}, it_is active for 79% of all the internal faults. This is the reason for using - in the multi-criteria approach - the same criterion with several different settings. 4, EVALUATION OF THE CRITERIA 4.1, Reliability For an optimally set criterion, its reliability, Q, (0 < Q <1) has been approximated as a func- tion of discrete time, n, as follows: 1 1 Qa) = (1-D) FD (1-Hea)) +Daq_d mem) 21) eases (roa Fi cases fom where: ‘ mer logic variable taking value 0 when criterion blocks, and 1 when it allows to trip, Fi part of the testing data base containing Mi appropriate non-internal-fault cases, part of the testing data base containing M Internal fault cases, D weighting factor (D = 0 - only blocking quality counts, D = J - only tripping inclination counts), of course 0-< D <1; in this paper D = 6.5. According to (21) each criterion is tested by evaluating its ability to distinguish between the class of disturbances Fy (inrush conditions for Ci, C2 and C3; over excitation for C4, Cs and Co; etc.) and F (internal faults). For the class Fi a criterion should block (u = 0) making the first sum in (21) large (close to 1). On the other hand, the tested criterion should operate for in- ternal faults, F, increasing the second sum in (21). Certainly, the first sum in (21) cor- esponds to the percentage of unwanted trip- pings, Pi, and reflects the stability of the criterion; while the second sum corresponds to the percentage of missing operations, P2, ref- lecting the sensitivity of the criterion. The higher is Q, the more reliable is the criterion: when it is perfect, Q equals 1 for all the time. Unlike the performance index, J, the recog- nition power, Q, is a function of time. It dis- plays the average statistical blocking and trip- ping inclination as a function of time, giving the Impression of the speed of recognition provided by the analyzed criterion. Fig.3 presents the recognition powers, Qs, versus time. From the figure one concludes: (a) All the considered criteria reach their maxi- mum recognition abilities in one cycle, (b) The overcurrent criterion (Ci, Ca, Co and C12) is initially weak, but gains after some 10ms after activation of the relay (caused by the dynamic of the measuring unit), (©) The direct wave shape identification criter- ion (C2) significantly gains after one cycle (the criterion waits for defined period of low current and its derivative - the reliable recognition is possible after one cycle), (d) The 2nd harmonic restraint against inrush conditions (C3) also gains after one cycle. ‘This change in Tecognition power is induced by transient overshoots of the ratio Isz/Ta1 during internal faults (Fig.1b) what reflects in delayed ‘rippings. Different nature of the transients of the ratio Ias/I, makes the Sth harmonic restraint (C6) much faster, (e) The criterion of the sequence of events (C7) does not change its power with time because it operates very fast or not at all. Similar ef- fect is observed for the criteria Cio (A-dif- 12th Power Systems Computation Conference Dresden, August 19 - 23, 1996 uu unity 2 20 Sw 0 60 timo iho iio 10 100 ow COMPUTATIONAL COMPLEXITY ah Fig.4. Performance index - computational complexity map of the evaluated protection criteria for a power transformer. 33. Recognition qualities of the evaluated criteria ver- sus time (C2a version of C2 used). The Q-axes cover the interval 0-1 ferential principle) and C1 (asymmetry checking), (Q) The criterion of 2nd harmonic, detecting saturation of CTs (Cs), is initially weak be- cause CTs transform accurately at the begin- ning of an external fault 4.2. Efficiency The numerical complexity of the criteria, has been evaluated as the total number of re- quired operations regardless their nature (ad tions, multiplications, comparisons, etc.) what is acceptable for modern microprocessors. Since several criteria share the same measur- ings, the conditional complexity, y, assuming the measurings to be computed separately, has been also approximated giving the impression of an extra numerical burden caused by adding a criterion to the relay algorithm. The overall efficiency of a criterion is defi- ned in this paper as: Efficiency rd (22) The lower are both the performance index and the computational complexity, the more ef- ficient is the criterion. Fig.4 displays the performance index - com- putational complexity map of the selected criteria for power transformer relaying; while Fig.5 presents their overall efficiency. crite: eso te Fig.5. Quality / complexity ratio of the considered This kind of information is recommended as a base for selecting criteria for a particular ap- plication under limited computational power of the hardware. However, one must remember that comparison is allowed only between cri- teria provided for the same target (for example, one can not compare 2nd and Sth harmonic restraints). 5. APPLICATION FOR THE RESULTS OF ANALYSIS The results of criteria evaluation are, in general, a base for selecting the criteria. This selection can be also made on-line, when cer- tain criterion is included or excluded from the algorithm, depending on other factors. In this paper, a new dynamic 'p out of q” decision making rule has been introduced. Let us consider the recognition qualities Q1, Q2 and Q3 (see Fig3) as measures of Teliabilities of C1, C2 and Cs (provided for iden- tification of inrush conditions). Assuming next, that a pair of criteria Cx - Cy is strong enough to exclude the inrush supposition at a particular sample n only if: Qx(n) + Qyeny > & (Q1(n)+Q2¢0)+Q3(0)) (23) where: « an arbitrary coefficient (0 $a <1), 112 12th Power Systems Computation Conference Dresden, August 19-23, 1996 C1 C23 230- o17 C4C5 Co ocn3 TRIP C7Ca Cy oen<13, 2815 wD Co Cn Cra o30 FFig.6. Decision making scheme of the relay employing the evaluated criteria and based on elaborated dynamic "2 out of 3° rule. The signals are: (a) - inrush ruled-out, (b) - over excitation ruled-out, external short-circuit with (©) and without (A) saturation of CTs ruled-out (a = 0.5), one gets a weighted "2 out of 3° rule. Certainly, in addition, it is assumed that: (1) when all of the criteria are satisfied inrush is ruled-out, and (2) no single criterion is able to do 0. By applying (23) one gets the following results: © the pair Ci - Cz is strong enough to exclude the inrush hypothesis between 2nd and 15th and after 30th sample since relay activation, © the pair Cz - Cs is sufficiently powerful till the 18th sample, © the pair Cy - C3, in turn, is satisfactory - in first 3 samples and after the sample no. 17. Fig.6. displays the logic circuit for the whole relay based on the proposed approach. Note, that the pairs Ca - Cs, C7 - Ce and Cri - Ciz are always powerful enough to be a base for trip- ping initiation; while the pair Cio - C12 - is not 6. CONCLUSIONS The paper gives a literature survey and com- arative analysis of selected protection criteria for power transformers. The criteria are formal- ized by forming the criteria conditions and set- ting the appropriate thresholds involved. Prior to comparison, the criteria have been optimized with the objective to minimize the defined per- formance index, which combines the percentage of recognition errors with the average tripping time. Such statistical picture of the studied criteria has been taken over a data base consist- ing of the selected key cases of operation of a protected power transformer. 12th Power Systems Computation Conference Dresden, August 19 - 23, 1996 ‘The criteria have been compared by evalua- tion of quality and speed of offered recognition as well as their efficiency defined as the ratio of recognition quality and computational com- plexity. The results of comparison serve as a base for selecting the set of criteria for a power transformer differential relay under limited computational power of the hardware. The analysis has been performed based upon the simulation of a particular transformer. The results, however, may be cautiously extended to other transformers owing to the differential relaying principle and when using the p.u. units. An application example for the analysis results has been given establishing a new, dynamic "p out of q* decision making principle for power system relaying. 7. REFERENCES on bower Delivery, Wol-5, No.3, pp.1299-1306, July (41 Migiat rlaying algorithms for the differential protec: ie (7) armer tael protection suitable Tor iigicd compater 9} jewski and B.Kasztenny, “A multicriteria dif Iranaformer ‘relay based om fanz) logic, IEEE. PES. Winter Meeting, paper” 95-WM-039-9 PWRD, New York 1998 [11] 1S.Thorp and A.G.Phadke, "A new computer based iu restrained current differential relay for power Itansformer protection”, IEEE Trans on Power Ap aratue and ‘Systems, Vol.PAS-102, No.I1, pp.3624- 629, November 1983 [12] B.Kasztenay, E.Rosolowski, M.M.Saha and. B.Hilst- fom, "A power transformer model for investigation of iltérenda protection schemes", Int. Conf. on Power System Transients, PST95, ppA36-i41, Lisbon, Sept 3 ioos 13] A.1.Degens, "Algorithm for a digital transformer dit 051 Pefami protection bared on a least-squares curve ft- ting Proceedings IEE, Vol138, Par ©, No.3 pp-135- toh May 1981 13

S-ar putea să vă placă și