The document proposes a wavelet and artificial neural network (ANN) based method to detect inrush currents and internal faults in power transformers. It begins with an introduction to transformer faults and differential protection challenges in discriminating faults from inrush currents. It then describes analyzing primary currents using discrete wavelet transforms and using the decomposition coefficients as inputs to an ANN to classify situations as faults or inrush/healthy. Simulation results on single and three-phase transformers demonstrate the approach's ability to accurately discriminate between inrush and fault scenarios. The document concludes the wavelet-ANN technique is efficient for this application and also discusses using ANNs for power system load flow analysis.
The document proposes a wavelet and artificial neural network (ANN) based method to detect inrush currents and internal faults in power transformers. It begins with an introduction to transformer faults and differential protection challenges in discriminating faults from inrush currents. It then describes analyzing primary currents using discrete wavelet transforms and using the decomposition coefficients as inputs to an ANN to classify situations as faults or inrush/healthy. Simulation results on single and three-phase transformers demonstrate the approach's ability to accurately discriminate between inrush and fault scenarios. The document concludes the wavelet-ANN technique is efficient for this application and also discusses using ANNs for power system load flow analysis.
The document proposes a wavelet and artificial neural network (ANN) based method to detect inrush currents and internal faults in power transformers. It begins with an introduction to transformer faults and differential protection challenges in discriminating faults from inrush currents. It then describes analyzing primary currents using discrete wavelet transforms and using the decomposition coefficients as inputs to an ANN to classify situations as faults or inrush/healthy. Simulation results on single and three-phase transformers demonstrate the approach's ability to accurately discriminate between inrush and fault scenarios. The document concludes the wavelet-ANN technique is efficient for this application and also discusses using ANNs for power system load flow analysis.
Dr. C. Venkatesh Prof. and HoD Dept. of EEE SR Engineering College Warangal, AP Introduction Transformer Faults Differential Protection Inrush Current Fault Current and Inrush Current Detection Wavelet Transforms Artificial Neural Networks Conclusion Presentation Sequence Power transformers are one of the most important components of power system network. Their protection is of utmost importance. Any maloperation can lead to heavy power losses and monetary losses. Faults in Power Transformer INTERNAL FAULTS Terminal to ground fault Turn to turn fault Core insulation failure Phase to ground fault Phase to phase fault Turn to turn fault Core insulation failure EXTERNAL FAULTS Open circuit fault External system short circuits Line-to-ground fault (LG) Line-to-line fault (LL) Line-to-line to ground fault (LLG) 3-phase fault (LLL) 3-phase to ground fault (LLLG) DIFFERENTIAL PROTECTION The electric protective relaying of power transformers is based on percentage differential relaying technique. These compare currents from all the terminals to a predetermined threshold . In case of an internal fault the circuit breaker isolates the transformer from the system INRUSH CURRENT When the transformer is suddenly connected to an AC voltagesource, theremaybeasubstantial increaseinthe primarywindingcurrent,knownasinrushcurrent. Thephaserelationisthatfluxandcurrentareinphaseand lagvoltageby 90degrees. Voltage at zero position Voltage at its positive peak Continuous Operation of Power Transformer Cold Start of Power Transformer Inrush Current and Resulting Differential Current Transformer Equivalent Circuit Differential protection discriminates between internal and external fault currents Butfailstodiscriminatebetweenfaultandinrushcurrent. The magnitude of magnetization inrush current is similar to internal fault currentandinrushcurrent. Inrushisatransientphenomenonanddoesntrequirerelayoperation. Relaytobeoperatedonlyforinternal faultsandnotforinrush Manyrelaysareofslowactingtypesothattheyarenotoperatedforinrush. Amethod to discriminate between inrush and fault current, to prevent any maloperationoftherelayisrequired. EARLIER ALGORITHMS USED FOR DISCRIMINATION AND THEIR DISADVANTAGES The first method was by using the magnitude of second harmoniccomponent. However during extensive fault conditions the second harmonicoffaultcurrentisgreaterthanthatofinrush. Thentherewasthealgorithmusing wavelettransformswith fuzzysystems However themajor drawbackof thiswastheneedof fuzzy lawswhichrequireextensivetrainingpatterns Thentherewasthealgorithmusingwavelet transformusing feedforwardneural networks Thencamemathematical modelslikesupport vectormachine andGaussianmixturemodel These suffered the drawback of requirement of large data windowforinputsandwerecomparativelyless effectivethan artificialneural networks. PROPOSED ALGORITHM Captureonecycleofprimarycurrent Obtaindifferential currentId=Ip-Is CalculateDWTofprimarycurrent Obtainstandarddeviationofdecomposedlevelsd1,d2,d3. These parameters are given to ANN as input data to discriminatethefaultsandinrushthatishealthycondition. If ANNoutput isdiscriminatedasfault, thenissuetripsignal otherwise proceed further i.e. monitor the differential current. SINGLE PHASE TRANSFORMER Single phase transformer simulation circuit INRUSH CURRENT ANALYSIS Inrush current at t=0.04sec Inrush current at t=0.045sec Inrush current at t=0.052ec. INTERNAL FAULT CURRENT ANALYSIS Primary fault current at t=0.04sec Primary fault current at t=0.045sec WAVELET ANALYSIS OF PRIMARY CURRENTS Coefficients obtained from the inrush current waveform time d1 d2 d3 0.04 1.055 5.926 41.54 0.041 0.8699 7.486 52.37 0.042 0.7731 9.461 62.06 0.043 0.7869 9.199 66.64 0.044 0.8488 10.55 80.81 0.045 0.8818 10.4 93.54 0.046 0.8047 9.529 76.87 0.047 0.7619 8.604 65.31 0.048 0.7993 8.743 51.83 0.049 0.9599 6.434 40.07 0.05 1.049 5.304 43.35 0.051 0.938 7.287 49.59 0.052 0.7419 9.636 60.09 0.053 0.7927 9.05 67.88 0.054 0.8511 10.36 80.79 0.055 0.8684 10.55 95.44 0.056 0.7953 9.52 67.84 0.057 0.7331 8.835 62.44 0.058 0.8514 8.836 46.06 0.059 0.989 6.193 41.54 0.06 1.159 5.963 43.49 standard deviation values of inrush current detail coefficients WAVELET ANALYSIS OF PRIMARY CURRENTS Coefficients obtained from internal fault current waveform (terminal to ground fault) time d1 d2 d3 0.04 83.95 199.7 2565 0.041 198.4 408.4 1932 0.042 478.8 902.5 5146 0.043 499 1631 7373 0.044 179.5 288.6 1634 0.045 161.8 220.4 1430 0.046 166.1 172.3 1628 0.047 484.2 2574 4137 0.048 175.8 740.7 2234 0.049 190.1 436.4 1787 0.05 15.21 46.98 77.02 0.051 160.6 573.5 1572 0.052 296.9 1262 2384 0.053 70.65 166.6 1270 0.054 43.58 142.6 945.9 0.055 54.41 196.5 1139 0.056 39.14 168.5 1167 0.057 112.9 216.3 1251 0.058 49.1 133.3 1179 0.059 135.9 401.7 730.1 0.06 70.45 131 916.6 standard deviation values of primary current for internal fault on primary time d1 d2 d3 0.04 579.8 1252 2735 0.041 376.3 1138 1780 0.042 291.9 928.6 527.4 0.043 282.9 962.6 1540 0.044 287.1 469.4 557.4 0.045 226.7 751.4 725.9 0.046 222.4 203.1 886.8 0.047 213.5 562.1 531.1 0.048 303.1 1283 1.36E+04 0.049 275.8 1628 1.53E+04 0.05 357.9 2481 1.66E+04 0.051 573.7 3468 2.01E+04 0.052 666.4 3586 7.65E+03 0.053 362.2 3677 1.28E+04 0.054 764.6 4033 1.16E+04 0.055 596.9 3103 1.21E+04 0.056 561.5 3189 2.25E+04 0.057 680.3 3602 1.87E+04 0.058 359.2 3653 1.60E+04 0.059 530 3756 2.13E+04 0.06 680.1 3892 3.98E+03 WAVELET ANALYSIS OF PRIMARY CURRENTS standard deviation values of primary current for internal fault on secondary Structure of ANN Data set One input layer with three input nodes, One hidden layer with four nodes and an output layer with two output nodes 45 values are given for training and 18 values are used to test the ANN ANN Training: Training output Testing output THREE PHASE TRANSFORMER INRUSH CURRENT ANALYSIS t = 0.04 s Inrush waveform of phase A Inrush waveform of phase B Inrush waveform of phase C INTERNAL FAULT CURRENT ANALYSIS ABG fault on primary side of transformer at t = 0.04 s Phase A current Phase B current Phase C current INTERNAL FAULT CURRENT ANALYSIS ABG fault on secondary side of transformer at t = 0.04 s Phase A current Phase B current Phase C current Structure of ANN Data set One input layer with three input neurons, One hidden layer with four neurons and an output layer with two output neurons. 63 values are given for training and 42 values are used to test the ANN ANN for each phase current analysis Training output Network A Network B Network C Test output Network A Network B Network C LOAD FLOW ANALYSIS OF A IEEE 14 BUS SYSTEM USING BACK PROPAGATION ALGORITHM Technique Employed: Artificial Neural Network. Algorithm: Back Propagation Algorithm. BUS NO.1 IS TAKEN AS SLACK BUS NO. OF PQ BUSES = 9 NO. OF PV BUSES = 4 At each bus Pgen, Qgen, Pload and Qload are given Pinj = Pgen Pload and Qinj = Qgen - Qload are calculated and are inputs HENCE, NO. OF INPUT NODES = 13*2 = 26. For pq buses we need to calculate |V| and <V. Assuming that Q constraints are specified for the PV buses, we need to calculate |V| and <V. HENCE, NO. OF OUTPUT NODES = 13*2 = 26. No. of hidden layer nodes = 26 No. of training patterns = train_p = 16 No. of testing patterns = test_p = 4 output: ETA= 0.50 Iterations= 4534 Error Rate= 0.000100 output: ETA= 0.500000 Iterations= 3714 Error Rate= 0.000100 ETA= 0.550000 Iterations= 3895 Error Rate= 0.000100 ETA= 0.600000 Iterations= 3427 Error Rate= 0.000100 ETA= 0.650000 Iterations= 4308 Error Rate= 0.000100 ETA= 0.700000 Iterations= 3449 Error Rate= 0.000100 ETA= 0.750000 Iterations= 3965 Error Rate= 0.000100 ETA= 0.800000 Iterations= 4420 Error Rate= 0.000100 CONCLUSION Efficient algorithm has been developed for di scrimi nation of inrush current and internal fault current in power transformers Wavelet-ANN detection technique demonstrated. ANN application for Load Flow Anal ysis discussed. Thankyou