A REVIEW OF DIFFERENT TECHNIQUES OF POWER TRANSFORMERPROTECTIONNeha Sahu1,Vijay Kumar Sahu2, Prakash C. Patil3*1PG Scholar, Electrical Department, NITRaipurRaipur (C.G.), India, E-mail: [email protected] AssistantProfessor, Electrical Department, SVIT, Chincholi,Nashik (M.S.), India, E-mail: [email protected].
in PG Scholar, Electrical Department, Central IndiaInstitute of TechnologyIndore (M.P.), India, E-mail: [email protected] Abstract: The power transformer is an electrical equipmentthat needs continuous monitoring and fast protection since it is very expensiveand an essential element for a power system to perform effectively. The mostcommon protection technique used is the percentage differential logic, whichprovides discrimination between different operating conditions and internalfault. Unfortunately, there are some operating conditions of power transformersthat can affect the protection behavior and the power system stability.
This paper presents an overview of existing protectionalgorithms used for protection of power transformer. It includes a discussionof new techniques which have been introduced using intelligent hybrid systemsand those being proposed to protect the power transformer. Keywords: Power Transformer, differential protection schemes,CT saturation, Artificial Neural Network (ANN). I. INTRODUCTIONThe power transformer is a piece of electricalequipment that needs continuous monitoring and fast protection since it is veryexpensive and an essential element for a power system to perform effectively.Power transformer internal faults may cause extensive damage and/or powersystem instability. Thus, different transformer protection schemes are used toavoid interruptions of the power supply and catastrophic losses. The mostcommon protection technique is the percentage differential logic, whichprovides discrimination between an internal fault and an external fault or anormal operating condition.
However, a simple detection of a differential currentis not sufficient to distinguish internal faults from other situations thatalso produce such a current. Some of these situations appear during transformerenergization (inrush currents), current transformer (CT) saturation, amongothers, which can result in an incorrect trip. The correct and fastdiscrimination of internal faults from the other situations mentioned is one ofthe challenges for modern protection of power transformers. Concerning theidentification of internal faults as opposed to inrush currents, the approachtraditionally used is the aforementioned differential logic together withharmonic restraint.
In this method, transformer inrush current due toenergization is recognized on the basis of second and other harmonic componentsobtained by filters.New algorithms that have been developed fordifferentiating between internal fault current and other situations that alsoproduce such current should be known for proper maintenance of powertransformer as it is an essential element for power system. Figure.1.Differential scheme used for the protection of large power transformers II. DIFFERENTIAL PROTECTIONThe diagram illustrating the differential logic usedfor the protection of large power transformers is shown in Fig. 1.
Theillustration also shows the connection of CTs coupled to the primary andsecondary branches. N1:N2 is the turn ratio between the primary and the secondarywindings of the transformer, and 1: n1 and 1:n2 are the turn ratios between thebranches and the CTs, selected to make N1n1=N2n2. Under normal conditions andexternal faults for a single-phase transformer, the currents i1S and i2S(secondary currents of CTs) are equal. However, in the case of internal faults,the difference between these currents becomes significant, causing thedifferential relay to trip.The differential currentid = i1S – i2S (1)gives a sensitive measure of the fault current.
Considering the restraint current ir = (i1S +i2S)/2,the relay will operate when id ? K.ir (2)where K is the slope of the differentialcharacteristic.As mentioned before, certain phenomena can cause asubstantial differential current to flow when there is not any fault, and thesefalse differential currents are generally sufficient to cause tripping.
However, in these situations, the differential protection should not disconnectthe system because an internal fault is not present.Magnetizing currents appear during transformerenergization due to its core magnetization and saturation. The slope of themagnetization characteristic in the saturated area determines its magnitude. Inmodern transformers, large inrush currents can be reached. In transformerenergization, as the secondary winding is opened, the differential current canreach sufficiently high values, causing a false relay operation. Some otherphenomena that cause false differential currents are magnetizing inrushcurrents during an external fault removal, transformer over excitation, as wellas CT saturation.
III. CT SATURATIONCTs are employed to provide a reduction of the primarycurrent as well as to supply galvanic insulation between the electric networkand equipment connected to the CT secondary, including protective relays.Therefore, CTs are made to support fault currents and other phenomena for a fewseconds, which can reach values of up to 50 times the magnitude of the loadcurrent. The current signals supplied on the secondary of a CT should be exactreproductions of the corresponding current signals on its primary. Althoughmodern devices perform satisfactorily well in this condition for most cases,the protection design needs to take into account the possible errors eventuallyintroduced by CTs, so that the relay performance in the presence of theseerrors can be enhanced. The CT performance under load current is not such aconcern compared to the fault situation in which the relay should operate.
Whenfaults occur, the current values can reach high levels. They can also contain asignificant dc component as well as the remnant flux in the CT core. All thesefactors can lead to the saturation of the current transformer core and canproduce significant distortion in the secondary current. In this case, thesecondary current of a CT cannot represent its primary current exactly. Thus,relays that depend on this current to make their decision can easily operateincorrectly during this period, affecting the reliability of the protection.The possibility of CT saturation should then becarefully considered in a protection system design in terms of relayperformance.
Some methods are used to avoid it, but some of the solutions canaffect the cost of such a piece of equipment. IV. DISCRIMINATION OF INTERNALFAULTSMany algorithms have been developed to discriminateinternal faults. Few modern techniques that have been developed are discussed.A. Discrimination using IntelligentToolSegatto and Coury have used ANN for discrimination ofinternal faults with other operating conditions.
Training process was doneusing multilayer perceptron (MLP) for discriminating internal faults from othersituations described.In their method, the simulated cases were thesituations that involved a significant differential current. Steady state andcapacitor bank energization were not included. For the transformer of 25MVA atotal of 2556 cases (7668 patterns, considering a moving window of three steps)were generated.For training and validation processes, 50% of thecases in which the saturation phenomenon caused by the current transformers wasconsidered. The architecture used for the purpose is shown in Fig.2.
Thedifferential current per phase are the input signals and the output willindicate the fault situation, if the cases. Figure.2. ANNused for pattern recognition B.
Discrimination using Clarke’sTransformBarbosa et al, proposed an algorithm in which uncoupled signals was obtained byapplying Clarke’s Transform to the three-phase currents in the secondarywinding current of the CT in both transformer ends. The equations arerepresented as where ph is the phase of current reference and k is thesample number of the discrete signal. Clarke’s transform could be applied toboth phasors as well as the instantaneous values.The main concept of using Clarke’s Transform iscarried out in a pattern-recognition process to discriminate internal faults,sympathetic inrush, energization and over-excitation. The differential ?-?-?components of the current are used.
where I?(k), I?(k), I?(k), i?(k), i?(k) and i?(k) are?-?-? components of the primary and secondary currents from a transformer,respectively, and N is the number of signal samples in the observation window.The computed values of the differential ?-?-?components of the currents are approximately zero in the case of a normaloperation, while the range of each differential current value fluctuatesaccording to the specific situation. Therefore, the various phenomena of thetransformer could be discriminated.With input of the differential ?-?-? components of thecurrent, fuzzy system is used to determine the fault condition more accuratelythan conventional differential protection methods, which has predefined rulesto discriminate between steady state and fault conditions. C. Discrimination using Harmoniccomponent, GA and Flux-restaint differentail CurrentMagnetizing inrush current have high second harmoniccomponent and internal fault has low second harmonic component. Similarlyover-excitation current has high fifth harmonic component compare to internalfault.
Barbosa et al. proposed another algorithm in which thecurrent signals are the inputs to the Genetic Algorithm (GA) in order toextract the fundamental and harmonic components of both the primary and thesecondary currents from the protected transformer.The decision making is done by use of Fuzzy System.Rules are formed by using four variables. The first two are the harmoniccomponents i.e. second harmonic and fifth harmonic component, third theoperating current (iop=id/irt ) where id(=| ip+is|) isdifferential current, ip is primary current and is is secondary current andlastly the flux-restraint differential current (dFlx) where dFlx is the flux-restraintdifferential current; ?t is the sampling interval; i is the input current; v isthe primary voltage; Lp is the leakage inductance of the primary winding and kis the number of sample.
Using these four quantities 16 rules are formed todifferentiate between steady state and fault condition. D. Differentiating using FuzzyLogic-Based Relaying TechniqueAs described earlier, second harmonic component isused to differentiate between inrush (high second harmonic component) andinternal fault (low second harmonic component).
Recently in power system thefrequency environment has been made complicated and the quality of secondharmonic component in inrush state has been decreased because of theimprovement of core steel. Due to this reason, the traditional approaches willlikely to mal-operate in the case of magnetizing inrush with low secondharmonic component and internal fault with high second harmonic component.Shin et al. proposed a relaying algorithm a relayingalgorithm to enhance the fault detection sensitivities of conventionaltechniques by using a fuzzy logic approach. The algorithm consists offlux-differential current derivative curve, harmonic restraint and percentagedifferential characteristic curve.
The proposed fuzzy-based relaying algorithm preventstrip mal-operation of relay in the case of magnetizing inrush with low secondharmonic component and internal fault with high second harmonic component andthen show improved accuracy and robustness against the change in conditions inpower system. E. Different methods for CTsaturation CorrectionCT saturation is one of the phenomenon which causesfalse tripping of the differential protection. Before development of the CTsaturation correction, the algorithm were not able to cope-up with CTsaturation and leads to unnecessary tripping. It can be stated that the CTsaturation phenomenon may impair protection system reliability if appropriatealgorithms for saturation detection and/or correction are not applied toeliminate the problem.
· One ofthe methods of avoiding CT saturation is by increasing the size of the CT core.Another is using a core material that supports larger flux densities. Both canaffect the cost and case of transformer application.· If theRMS value supplied by CT is distorted by saturation, the RMS value sensed willbe much lower than the actual fault current.· One ofthe approaches is that if the CT saturation characteristic is known, anestimate of the input current when the CT saturates can be found, assuming azero-offset waveform and resistive burden.· Otherapproaches have used Kalman filters and iterative approaches. These methods arenot convenient for real-time applications.
Some of the methods developed using intelligentsystems are described below. a) RecurrentANNs Correcting The Distorted Signals Originated from Saturated CTs.Segatto and Coury used the Stuttgart Neural NetworkSimulator (SNNS) to train the different architectures in order to correct thedistorted signals. Due to its efficiency, simplicity and user-friendly graphicinterface the program was chosen.Using the SNNS simulator, the Elman training methodwas carried out. The training stage finished in 7000 cycles, where it reachedthe mean square error of 0.
0005. in this method, 1600 training patterns wereused, whereby 800 were utilized in the training and 800 in the validationprocess. The ANN input and output layers contain 24 neurons each. For such aresult, the best ANN topology obtained was 24+20+24, with 24 neurons in theinput layer, 20 neurons in the hidden layer, 24 neurons in the recurrent layerand 24 neurons in the recurrent layer and 24 neurons in the output layer, whichreconstructed the input signal.
In the result obtained, a total of 188 patterns (94for internal faults and 94 for inrush situations) were used for testing thearchitectures. These patterns were not used in the training stage. The ElmanANN obtained an accuracy rate of 98.
5%.Advantage of this algorithm is that no pre-processingof signals was required. b) CTSaturation Detection with Optimized Neural NetworkWaldemar and Daniel proposed CT saturation detectionwith genetically optimized neural network. Advantages of neural computingmethodologies over conventional approaches include faster computation, learningability, robustness and noise rejection.While preparing a useful and efficient ANN-basedclassification/recognition unit, one has to take into consideration the issuesof the ANN choice (structure type, number of layers and neurons activation functions,input signals) and its training (learning method, initial conditions). Choosingthe type of ANN structure and its further parameters are rather a matter of thedesigner experiences with ANN usage since, unfortunately there are no generalpractical rules that could be applied for that purpose. The heuristic waysequential trial-and-error attempts maybe followed, however this may notguarantee the optimal ANN structure to be found.An optimization approach based on genetic principleswas proposed and its efficiency was studied.
Figure.3. Flowchart ofthe GA procedure for the ANN topology optimization. In Fig.3, the block scheme of the genetic optimizationprocedure as applied for the ANN structure optimization is presented. In this method, optimization of the neural CTsaturation detector was performed out of a population of neural networksconsisting of 20 individuals. The ANNs are being trained with the first halfand tested with the second half. The Levenberg-Morquardt training algorithm wasadopted with the desired output of the ANN set to 1.
0 for the periods of linearCT operation and 0.0 when the CT was saturated.In order to drive the optimization process in bothefficiency and ANN size directions, the quality index (Qeff/size)was proposed. where and nANN is ANN size (total number ofneurons).The values of Qeff/size are inverselyproportional to the total number of neurons of ANNs being assessed, thus givinga chance of obtaining efficient yet reasonably small ANNs.
The best ANNs obtained for respective quality indicesare:· For Qeffindex- ANN having 14 neurons (13-1), classification efficiency equal0.978;· For Qeff/sizeindex- ANN having 6 neurons (3-2-1), classification efficiency 0.954. c) Correctionof CT distorted signal due to saturation caused by Faults using ANNsYu et al. proposed use of an ANN to correct CTsaturation caused by faults in power systems including DC offset and randomfault incidence angle. The ANN was trained to provide the inverse function ofCT.
ANN presented was trained only for resistive load.The algorithm for running the network was implementedon an Analog Devices ADSP-2101 digital signal processor. The calculating speedand accuracy proved to be satisfactory for real-time applications.Neural network structure is a feed-forward typenetwork with two hidden layers, shown in Fig.
4. The input layer of the networkhas 32 input nodes. Two hidden layers, one with 10 nodes and another with 6nodes, were used. Each of the hidden nodes accumulates a sum of the samplespresented at the input layer multiplied by a weighing factor for eachconnection. The output node accumulates the sum of outputs of the 6 hiddennodes and processes the sum through a tan-sigmoidal function.
A feed-forwardtopology has the advantage of simplicity and inherent stability. To train the network, a range of fault cases weredefined. Variables for the test casesincluded fault impedance magnitude, X/R ratio, and closing angle. Combining allthree sets of parameters, fault impedance magnitude, X/R ratio and faultincidence angles, and produces around 40 sets of waveform data to train theneural network. Figure.4.The network structure The template is designed so that author affiliationsare not repeated each time for multiple authors of the same affiliation. Pleasekeep your affiliations as succinct as possible (for example, do notdifferentiate among departments of the same organization).
This template wasdesigned for three affiliations.F. ConclusionFor the protection of the power transformer differentmethods have discussed. Different methods for fault detection, CT saturationcorrection and basic differential protection has been explained.
The conclusioncomes here that all methods have individual importance but CT saturationcorrection by ANN is effective but the training of ANN is time consuming.REFERENCES 1 A. G. Phadke and J. S. Thorp, “A new computer-basedflux-restrained current-differential relay for power transformer protection,”IEEE Trans.
Power App. Syst., vol. PAS-102, no. 11, pp. 3624–3629, Nov. 1983.
2 D.C. Yu, Z. Wang, J.C. Cummins, H.-J.
Yoon and L.A.Kojovic, “Correction of Current Transformer Distorted Secondary Currents Due toSaturation Using Artificial Neural Networks” IEEE Trans. Power Del., vol. 18,no. 2, pp. 189-194, April 2001.
3 D.C. Yu, Z. Wang, J.C. Cummins, H.
-J. Yoon, L.A.Kojovic and D. Stone, “Neural Network for Current Transformer SaturationCorrection” IEEE Transmission and Distribution, vol.
1, pp. 441-446, 1999.4 M.-C. Shin, C.
-W. Park, and J.-H. Kim, “Fuzzy logicbased relaying for large power transformer protection,” IEEE Trans. Power Del.
,vol. 18, no. 3, pp.
718–724, Jul. 2003.5 E. Segatto and D. Coury, “A differential relay forpower transformers using intelligent tools,” IEEE Trans. Power Syst., vol. 21,no.
3, pp. 1154-1162, Aug. 2006. 6 D. Barbosa, D.V. Coury and M.
Oleskovicz, “PowerTransformer Differential Protection Based on Clarke’s Transform and FuzzySystems” IEEE Trans. Power Del., vol. 26, no. 2, pp. 1212-1220, April 2011. 7 D.
Barbosa, D.V. Coury and M. Oleskovicz, “Newapproach for power transformer protection based on intelligent hybrid systems”IET Generation, Transmission and Distribution, vol. 6, no. 10, pp. 1003-1018,2012. 8 A.
Zilouchian and M. Jamshidi, Eds., IntelligentControl Systems Using Soft Computing Methodologies. Boca Raton, FL: CRC, 2000.9 Z.Kovacic and S. Bogdan, Fuzzy Controller Design.
Boca Raton, FL: CRC., 2005. 10 Y. C Kang, J.
K. Park, S. H.
Kang, A. T. Johns and R.K. Agganval, “An Algorithm for compensating secondary currents of currenttransformer” IEEE Transactions on Power Delivery, Vol 12, No.
1, January 1997,pp. 116-126.11 Waldemar Rebizant and Daniel Bejmert,”Current-Transformer Saturation Detection With Genetically Optimized NeuralNetworks” IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 22, NO. 2, APRIL 2007, pp.820-827.12 M.
Kezunovit, C. W. Fromen and F.
Phillips,”EXPERIMENTAL EVALUATON OF EMTP-BASED CURRENT TRANSFORMER MODELS FOR PROTECTIVERELAY TRANSIENT STUDY” IEEE Transactions on Power Delivery, Vol. 9, No. 1,January 1994, pp. 405-413.13 Manoj Tripathy, Rudra Prakash Maheshwari, and H.
K.Verma, “Power Transformer Differential Protection Based On OptimalProbabilistic Neural Network Optimal Probabilistic Neural Network” IEEETRANSACTIONS ON POWER DELIVERY, VOL. 25, NO.
1, JANUARY 2010, pp. 102-112.