A New Method of Power Swing Blocking for Distance Protection Using ANN

Transcription

1 ew Method of Swing locking for Distance Protection Using H. Khorashadi Zadeh, Department of Engineering University of irjand irjand, Iran S. Hosseini aveh,.t. Hosseini, S.H. bdol Khalegh Zadeh, Kerman Regional Electrical ompany Kerman, Iran, STRT This paper presents new methods to detect power swing correctly and accurately. ny sudden change in the configuration or the loading of an electrical network causes power swing between the load concentrations of the network. In order to prevent the distance protection from tripping during such conditions, a power swing blocking devices cannot cope with very fast power swings. ew power swing blocking method that was presented in this paper operates by artificial neural network. This method has the ability to immediately clear the block when a fault occurs within the relay trip zone. In this paper an -based approach is used and a very accurate swing blocking unit is designed. This principle can detect different power swings and block distance protection during them. It is also shown that it is possible to clear blocking if faults occur during a power swing.. ITRODUTIO The principle relation of protective relays to the problem of power stability is their vital role in clearing faults as rapidly as possible in order to maintain stability. The opening of a sound or healthy line, during swings from which the system can recover, is nearly always undesirable. It is apparent that relays should be required to distinguish normal conditions, short circuits, swing of large amplitude and out of step operation. They should also take the desired action at the right time in each of these circumstances []. When a power swing occurs a change appears in the relative voltage phase angle between an electrical machine or group of machines and the power system voltage source. t a relay location the measurable electrical quantities of voltage V, current, apparent impedance Z, active power P and reactive power Q will then vary as a function of this angle. When a fault occurs, however, these quantities change suddenly. Such behaviors are used in distance protection as criteria to block or unblock the tripping [-4]. Protection relaying is just as much a candidate for the application of pattern recognition. The majority of power system protection techniques are involved in defining the system state through identifying the pattern of the associated voltage and current waveforms measured at the relay location. This means that the development of adaptive protection can be essentially treated as a problem of pattern recognition/classification. rtificial eural etworks (s are powerful in pattern recognition and classification. They possess excellent features such as generalization capability, noise immunity, robustness and fault tolerance. onsequently, the decision made by an -based relay will not be seriously affected by variations in system parameters. -based techniques have been used in power system protection and encouraging results are obtained [5-0]. In this paper an -based approach is used and a very accurate swing blocking unit is designed. This principle can detect different power swings and block distance protection during them. It is also shown that it is possible to clear blocking if faults occur during a power swing. The proposed methods are both reliable and accurate, and are also demonstrated to work when a single pole is open for fault clearing on a transmission line. In this method, artificial neural network as pattern classifier has been used for detecting power swings and faults. n improved performance with the use of an artificial neural network approach is experienced once the block set can operate correctly. lso by simulating it in a distance relay has tested this method.. POWER SYSTEM MODEL The training data set of an should contain the necessary information to generalize the problem. Using an electro-magnetic transient program EMTD [] a two machine three phase 30 kv power system has been simulated for the study of transmission line distance measurement problem. The one-line diagram of the studied system is shown in ig. and its parameters are shown in Table. G R D I ig.. Simulated power system model Table. Simulated power system parameters Positive & negative sequence line 0.07+j 0.46 impedances ( /km Zero sequence line impedance ( /km j.066 Sources impedances ( 4.7 Sources X/R ratio 00 Sources Zo/Z ratio 0.5

2 Table. Training patterns data generation ondition Slow and fast power swing without and with different faults, different faults without power swing (Single phase to ground, phase to phase to ground, phase to phase, three Phase open single phase and closed other phases ault Location 5,50,70,75,79,80,8,85,90,95 (km Load Different values between ngle (deg ( Inception Different values between ngle (deg (0-360 Training patterns were generated by simulating different types of condition on the power systems includes power swing without and with fault and different types of faults without power swing. ault type, fault location, fault inception time and load angle were changed to obtain training patterns covering a wide range of different power system conditions. lso condition that a single pole is open has been applied to generate pattern training. ombination of different conditions considered for training pattern data generation is shown in Table. The disturbance is created by a three-phase fault at location on D line. Such disturbances on the system will cause the generator to accelerate or decelerate resulting in a power swing. 3. THE EURL ETWORK SED POWER SWIG LOKIG DESIG neural network based algorithm is proposed as an unit that detect power swing and faults. The block diagram of the proposed power swing blocking approach is shown in ig.. System swing are characterized by slow Variation of currents, voltage and impedances. Therefore, changing rates of the samples of three phase currents and measured phase to phase impedances are taken as input quantities of that amount to 7 variables. Z and Z correspond to measured impedance and Impedance, respectively. i = i i ( n ( i = i i ( n ( i = i i ( n (3 Z = Z Z ( n (4 Z = Z Z ( n (5 Z = Z Z ( n (6 Z = Z Z ( n (7 In (-7 n is the sample number and is the number of samples per cycle. Z and Z are calculated as following: v v Z = (8 i i v v Z = (9 i i The proposed neural network has two nodes in output layer. irst output becomes when power swings occur and second output is active when fault occur. 3.. PREPROESSIG The process of generating input patterns from the recorded voltages and currents is depicted in ig. 3. Samples of three phase voltages and currents at the relay location were obtained from the EMTD. These sample, were processed by th order low pass anti alising s and were resampled at khz. The anti alising s had a cut-off frequency of 400 Hz. -sample IR digital flitter then removed the dc, component. Elimination of the dc component enhanced the training capabilities of the discriminator. Simulation by EMTD nti alising Samoling by khz ι Voltages urrents Pre_processing ι ι Ζ ΑΒ Ζ c ( Ζ ( ΖΒ ig.. Structure of proposed principle Swing ault Samples of each of the phase currents and measured impedances are compared with the samples of the same phase current taken and measured quantity in two cycles before. or reliability and fault tolerance the measured impedances are compared with the measured quantity in one cycle before. Patterns inputs are made based on the combination of the current samples and measured impedances using (-7. In these equations i, i and i corresponds to phase, and and Z, Z, Dc mponent DT Elimination ig. 3. Pre-processing Scalling Patterns of have generated by preprocessed samples of voltage and current and using equation (-7. inally inputs were scaled to have a maximum value of + and a minimum value of. This was archived by using a scaling factor equal to the peak value of the normal rated voltage for the samples of voltage and the most great of current for the samples of current. The base value of impedance equals to ratio of voltage base to the current base value. 3.. TRIIG Multilayer feedforward networks were chosen to process the sampled input data. umber of the networks to detect the power swing and faults was decided by experimentation, which involved training and testing different number of networks. etworks include 7 inputs and two outputs. irst outputs neural network is related to power swing and second outputs neural network is related to faults.

3 or all the networks, hyperbolic tangent function is used as the activation function of the hidden layer neurons. Sigmoid function is used as the output layer activation function []. Various networks considered were trained with Marquardt-Levenberg (ML training algorithms [3]. ML algorithm is an approximation to the ewton s method. ML algorithm is a nonlinear least square algorithm applied to the batch learning of multilayer perceptrons. oncerning the architecture, parameters such as the number, of inputs to the network, as well as the number of neurons in the input and hidden layer were decided empirically. This process involved experimtation with various network configurations. Through a series of trial and error and modifications, the architecture the best performance is a three layer perceptron, with 7 inputs, output and 9, 5 neurons in the hidden layers. Once trained, the networks performance was tested using test patterns that were different from the training patterns. validation data set consisting of about 80 different conditions was generated using the power system model shown in ig.. The validation set condition patterns were different than the condition patterns used to train the network. or different conditions of the validation set, fault type, fault location, fault inception time and load angle were changed to investigate the effects of these factors on the performance of the proposed algorithm. 4. IMPLEMETTIO and TEST for EW PRIIPLE In order to analysis of purposed principle, a distance relay has been implemented in computer by program and new method of power swing blocking has been applying in it. The block diagram of the distance relay approach is shown in ig. 4. Three phase voltage and current input signals were processed by nd-order lowpass utterworth s. The anti-aliasing s had a cut-off frequency of 400 Hz. In addition, -sample IR digital s were used to remove the dc component. ia ib ic va vb vc nti-aliasing & dc Measuring elements a-g b-g c-g a-b b-c c-a phase-selector power swing detector ig. 4. The distance relay structure Phasors of the voltage and current have been obtained by DT algorithm. These phasors are entering to 6 impedance measuring units. In additional these are l o g i c entering to power swing blocking and this unit issues order block to main unit logic by processing data inputs. Logic unit relay issues suitable order by results obtained from phase selection, power swing blocking and impedance measuring units. Measured currents at the relay location are subject to changes when a fault occurs on a transmission line. Phase selector principle may be based upon detecting these changes. The principle of variation of fault before and after the fault incidence is used a fast and reliable Phase selector module is designed to detect the fault and classify the fault type. Samples of each of the phase currents are compared with the samples of the same phase current taken half cycle and one cycle before. Phase selector module outputs are made based on the combination of the current samples using (0-. In these equations Sel, Sel and Sel corresponds to phase, and Phase selector outputs, respectively. If the fault classifier output signals exceed a pick-up threshold for three consecutive samples, a fault on the relevant phase is detected. The level above which the classifier operates could be set to prevent operation in the case of system load changes. The outputs of the three Phase selectors are ORed according to (3 to make the fault detection signal. Sel = i +. i ( n + i ( n (0 Sel = i +. i ( n + i ( n ( Sel = i +. i ( n + i ( n ( Det = Sel+ Sel+ Sel (3 In (0- n is the sample number and is the number of samples per cycle. further characteristic is also used to detect fault type changes. One cycle of data of each of the phase currents prior to the fault incidence is stored in the memory. ault current samples are compared to the prior to fault current samples based on the same principle used in (0-3 to identify the fault type changes. Samples of voltages and currents related to different condition are used as the distance relay inputs. The distance relay with proposed power swing blocking, operation for different power system conditions is presented in Table 3. s an example, test results for a single pole open, is presented in the first row and column of the Table 3. or this condition, the relative angle of the sending end source with respect to the angle of the receiving-end source was 5 deg. The fault inception angle with respect to phase voltage zero crossing was 5 deg. s shown relay has not been blocked and has not been operated. Relay operation for four other different conditions is shown in the last four columns of the table. or the faults that involve power swing and the faults, which don t involve power swing, the relay operation for different fault type and fault location is investigated. or all faults has been issued order trip by relay.

4 Table 3. swing blocking operation test results θ δ Single pole Low- ast- Swing ault (deg (deg open Swing Swing with ault 0 5 _ 0 km 0 km 0-5 _ 30 km 30 0 _ 50 km 0-0 _ 70 km 50-5 _ 75 km 0 0 _ 80 km 60-5 _ 90 km 30 5 _ 55 km 0-5 _ 65 km 5 0 _ 85 km 0-0 _ 45 km 30-5 _ 35 km 45 5 _ 90 km 30 0 _ 63 km 30km 50 km 70 km 75km 80 km 90 km 55 km 65 km 85 km 45 km 35 km 90 km 63 km *: not block, -: not trip, δ : load angle, θ : fault inception angle In additional for condition that has been occurred power swings without faults relay has not operated. s shown in Table 3, the relay performs quite accurate and reliably and the result of table shows performance of power swing blocking unit. 5. DDITIOL TESTS for PERORME STUDIES Proposed of the blocking method has been tested for several multi-machine systems. Diagram of one of them is shown in ig 5. onsidered relay is installed in line ST at busbar S. When fault appears at point and line TD is switched off the generator G swings via Line ST. ault has occurred in 0 ms and has cleared in 0 ms. Example of the plot for fault is shown in ig. 6. This fault is seen in the second zone of distance relay located in busbar S (Point. fter fault clearance (point the impedance locus moves in path to. ig. 7 shows output of the proposed method for this disturbance. It is shown that before clearance fault, second output of is active but after clearance fault and appear power swings, first output becomes. G G3 30 kv, 00km 45 MW 0kV/30kV 600 MW S Relay T 30/3 M 30kV/ 0kV 345 MW 3kV, 40km 30kV, 40 km T 30/3 30kV/0kV ig. 5. single line diagram of the six-bus model of a transmission network Output-ault Output- Swing X X X R R ig. 6. Impedance shown by distance relay for disturbance in line DT R E T G4 30 kv, 60 km D G 30kV/ 0kV 50MW t (ms ig. 7. Outputs of for disturbance in line DT Test results indicated that in general the proposed relay performs fast and reliable. More studies are being conducted in a wide range of system conditions and they show accurate the proposed method power swing blocking. 6. OLUSIO eural networks capabilities in pattern recognition and classification are used and a neural network-based power swing blocking is designed. Simulation studies are performed and the unit s performance with different system parameters and conditions is investigated. s it is shown by different examples in the paper the new principle can detect different power swings and block distance protection during them. It is also shown that it is possible to clear blocking if faults occur during a power swing. The proposed methods are both reliable and accurate, and are also demonstrated to work when a single pole is open for fault clearing on a transmission line.

5 REEREES [] E. W. Kimbark; System Stability, Vol., pril, 97, John Wiley &sons edition []. Mechraoui, D. W. P. Thomas, The Influence of Swing on Transmission Line Distance Protection, Measurement of and Voltage Phasor Oscillation at The Relaying Point UPE 93, Volume, September 93, pp [3]. Ilar, baden, Innovations In The Dejection of Swing In Electrical etwork, rown overi REV. -8, pp [4] Z. d. gad, G.. Wang, ew Swing lock In Distance Protection of EHV/UHV transmission line International onference In dvanced System ontrol, Operation and Management, ov 99, Hong Kong, pp [5] M. Sanaye-Pasand, H. Khorashadi-Zadeh and O. P. Malik High Speed ccurate Transmission Line Distance Protection Using s Proceedings of the IEEE/PES Summer Meeting, July 00 Vancouver, anada. [6] T. S. Sidhu, H. Singh, and M. S. Sachdave, Design, Implementation and Testing of rtificial eural etwork-based ault Direction Discriminator for Protection Transmission Lines, IEEE Trans. on Delivery, vol. 0, no., pr. 995, pp [7] T. Dalstein and. Kulicke, eural etwork pproach to ault lassification for High Speed Protective Relaying, IEEE Trans. on Delivery, vol. 0, no., pr. 995, pp [8] M. Kezonuic, Survey of eural et pplication To Protective Relaying and ault nalysis, Eng. Int. Sys. Vol. 5, no. 4, Dec. 997, pp [9] Y. H. Song,.T. Johns and Q.Y. Yuan, rtificial eural etwork-based Protection Scheme for ontrollable Series-compensated EHV Transmission Lines, IEE Proc. on Gener., Transm. and Distr., vol. 43, no. 6, ovember 996, pp [0] M. Sanaye-Psand and O.P. Malik, High Speed Transmission System Directional Protection Using an Elman etwork, IEEE Trans. on Delivery, vol. 3, no. 4, Oct. 998, pp [] PSD/EMTD User s Manual, Manitoba HVD Research enter, Winnipeg, Manitoba, anada. [] S. Haykin, eural etworks, IEEE Press, ew York, 994. [3] M. T. Hagan and M.. Menhaj, Training eedforward etworks with the Marquardt lgorithm, IEEE Trans. on eural etworks, Vol. 5, no. 6, 994, pp