Wavelet Transform and Artificial Neural Network

In transmission line when the current does not flow from transformers secondary side after flowing from its primary side, due to this over current in transformer occurs. This fault is known as an inrush current in the transformer. Inrush current is the transient maximum, current drawn by an electrical device when first turned on. Alternating current electric motor and transformer may draw inrush current several times. Its normal full load current first energizes for two cycles of the input waveform.

There are different methods to solve this fault. First method is Differential Transformation, in this method when currents of primary side and secondary side of transformer are equal. In such case, relay connected between them will detect the fault and trip the circuit. Another method is Fourier series transform. In this method when fault occurs then frequency of transmission line increases than the rated frequency. The drawback of this method is, it only detects the fault of frequency 50Hz and 0-5sec time period, it does not give the exact time of fault. To overcome this drawback Short Time Fourier Transform (STFT) is introduced. It gives accurate time of fault by dividing the overall time with ?. STFT also has some limitations that once you choose a particular size for the time window, that window is same for all frequencies. This work is easily done by Wavelet with more precision. Wavelet Transform is used to detect the inrush current conditions. Artificial Neural Network (ANN) is used to classify the inrush current conditions. The simulation process is done by MATLAB. Jazebi.et.al. [1] proposes an approach of magnetizing inrush current using Gaussian Mixture Models (GMM). The simulation is done by PSCAD/EMTDC software for various faults and switching conditions on a power transformer. 500 MVA, 400/230 kV, the three-phase power transformer is used in the simulation system. Mother wavelet type and decomposition level are used in detecting and localizing different kinds of fault transients. The sampling frequency and system basic frequency is 10 KHz and 50 Hz. The window size of WT is 50 samples per window for GMM. In power system, GMM is proved as simple identification criteria, best suited for protection, fast performance, and furthermore investments. In [2], presents the classification of transient phenomena in distribution systems. Wavelet transform algorithm based scheme is used the classification of many types of transients common in distribution systems. Simulation is done on ATP-EMTP is used for inrush current, load switching, capacitor switching and single phase to ground fault in primary 20 kV radial distribution feeder.

A.R.Sedighi and M.R.Haghifam [3] presents an efficient method for detection of inrush current in distribution transformer based on wavelet transform. Electro-Magnetic Transient Program (EMTP) is used for the simulation of Inrush current and other events for feature extraction and discrimination. 20kv distribution feeder is used and 20kHz sampling frequency in a single phase to ground fault and inrush current.

Ashrafian.et.al [4] describes the application of discrete S-transform differential protection of power transformer. Discrete S-transform is used for discrimination of inrush current and internal fault. A 13.5MVA, 132/33kv, 3phase transformer is used in simulation power system which has 980 and 424 turns of the primary and secondary winding. The transmission line is divided into two identical p-sections in the model. MATLAB and EMTP program is used for implementation.

In [5], the method is for discrimination of inrush current and the internal fault is proposed in power transformer. The method is based on Empirical Wavelet Transform (EWT) and Support Vector Machine (SVM). Matlab/Simulink is used for simulation. By taking the ratio of second harmonics to the fundamental of the current waveform, it distinguishes both types of current. It consists of two classes of data for validation such inrush and internal fault current waveforms. It has two transformers T1 and T2 which is connected through the transmission line.

Abnaki.et.al. [6] gives a method for identification of magnetizing inrush current from the internal fault in power transformer protection. Symmetrical components are used in this technique. The simulation takes place in all the cases such as normal condition, inrush condition, internal fault condition, external fault condition and over flux condition using PSCAD/EMTDC. The ratings of simulated power transformer system used are 30 MVA in rate 33kV/11kV. With the proposed model, probably in all cases, the simulation result is obtained. Ozgonenel.et.al. [7] introduces a modern approach for the power transformer protection. WT is used for the extraction of inrush current and internal fault in power transformer. In phase to ground fault and phase to phase fault, WT helps to analyze the current signals with their discontinuities. Coiflet 6 wavelet functions are used for the study of discontinuity of current signals. While fault conditions and inrush currents, Coif 6 is selected as it gives less error in reconstructions and gives more accurate results. The given model is simulated using ATP-EMTP at 50Hz fundamental frequency and 200Hz sampling frequency. Omar A.S. Youssef presents an advance scheme for the discrimination of faults in power system and inrush currents [8]. Using EMTP, a transformer is connected 132/11kv to power system. 11/132kv transformer with both sides star connected with grounded neutral. The transmission line is two 132kv at 50km sections is used. The data window required for proposed algorithm is less than half frequency cycle. The results obtained for the technique is accurate, fast and reliable.

Distinguish between inrush current and internal faults in indirect symmetrical phase shift transformer (ISPST) demonstrated in Bhasekar.et.al. [9]. Using Parseval’s theorem, wavelet energy is used for the extraction of different current signals from different operating conditions. WT is used to convert time domain into frequency domain. The software PSCAD/ EMTDC is used from which the data is generated. Using DB7 mother wavelet, WT decomposes from level 1 to level 7. D1 to D7 is used for the discrimination of internal fault from inrush current. The theory of wavelet transform is explained in Introduction to Wavelets by Amara Grap [10] and ANN is explained in The Ann Book by R.M. Hristev [11].

This paper presents results detection of inrush current using wavelet transform and artificial neural network. This helps to distinguish the inrush current and internal fault current. The data is generated from Db4 which is used as mother wavelet with level 5.

Inrush current is the maximum instantaneous input current given by an electrical device when it is switched on. This current arises due to high starting current. To charge the capacitor, inductor, and transformer, a high current is produced at the time of switch on. Its value depends on the core material, residual flux and instant of energization.

In power transformer, inrush current other than energization also takes after the clearance of external fault until voltage recovery. Inrush current also contains even and odd harmonics. It also has DC offset.

Inrush current can be high as 20times the normal current value it can only last for about 10ms. It requires about 30 to 40 cycles for the current to settle down to its normal current value.

WAVELET TRANSFORM (WT): Wavelets are mathematical functions that cut up data into different frequency components, and then study each component with a resolution matched to its scale. They have advantages over traditional Fourier methods in analyzing physical situations where the signal contains discontinuities and sharp spikes. The drawback of STFT is that once a particular size is selected for the time window, then time window remains same for all frequencies. To accurately analyze signals that have abrupt changes, a new class of functions that are well localized in time and frequency is used. Wavelets were developed independently in the fields of mathematics, quantum physics, electrical engineering, and seismic geology has discussed [10]. WT are classified as Continuous Wavelet Transform (CWT) and Discrete Wavelet Transform (DWT). WT breaks the signals into various frequencies which are used for the detection of inrush current, fault current and normal current. The detection of inrush current is implemented at DWT.

ARTIFICIAL NEURAL NETWORK (ANN): The basic building block of Artificial Neural Network (ANN) is the neuron. A neuron is processing units which have some inputs and only one output. The ANN is built by putting the neurons in layers and connecting the output of the neurons from one layer to the inputs of the neurons from the next layer has discussed [11].

An ANN is configured for a specific application, such as pattern recognition or data classification, through a learning process. Learning in biological systems is done by adjustments to the synaptic connections that exist between the neurons. WT breaks the signal into small contents, depending upon the content of each frequency signal is classified by ANN.

The generalized model is as shown in the figure. It consists of a generator, transformer, transmission line and loads having different combinations. Other parameters like the circuit breaker and different fault resistances were applied. The simulation was done with different combinations on MATLAB using Simulink. The data is generated from db4 with level 5 as shown in the table. An efficient technique is used for the detection of inrush current using wavelet transform and artificial neural network. The proposed technique is based on the decomposition of three-phase currents using WT with db4 as mother wavelet. ANN is used for the discrimination of inrush and fault current.