Application of probabilistic neural network for differential relaying of power transformer M. Tripathy, R.P. Maheshwari and H.K. Verma Abstract: Investigations towards the applicability of probabilistic neural networks (PNNs) as core classifiers to discriminate between magnetising inrush and internal fault of power transformer are made. An algorithm has been developed around the theme of conventional differential protection of transformer. It makes use of the ratio of the voltage-to-frequency and the amplitude of differential current for the detection of the operating condition of the transformer. The PNN has a significant advantage in terms of a much faster learning capability because it is constructed with a single pass of exemplar pattern set and without any iteration for weight adaptation. For the evaluation of the developed algorithm, transformer modelling and simulation of fault are carried out in power system computer-aided designing PSCAD/EMTDC. The operating condition detection algorithm is implemented in MATLAB. 1 Introduction Power transformers are very expensive and vital components of electric power systems. The continuous monitoring of power transformer can provide early warning of electrical failure and could prevent catastrophic losses. It minimises the damages and provides uninterrupted power supply. Accordingly, high expectations are imposed on power trans- former protective relays. Expectations from protective relays include dependability (no missing operations), security (no false tripping) and speed of operation (short fault clearing time). The differential relaying principle is used for the protection of medium and large power transformers. This superior approach compares the currents at all terminals of the protected transformer by computing and monitoring a differential (unbalance) current. The value of differential current greater than the no-load value indicates an internal fault. The magnetising inrush occurs during the energisation of power transformer, which results in a high current of the order of 10 of the full-load current [1]. This high current may cause the relay to operate. To avoid the maloperation of the relay, discrimination between magnetising inrush current and the fault current is required. Literature review shows that generally two approaches are applied to discrimi- nate between magnetising inrush current and the faults current. These are harmonic restraint concept and waveform identification concept. The literature reveals that the first method, based on harmonic restraint, has been used exten- sively. In 1983, Phadke and Thorp [2] proposed flux- restrained current differential relay for power transformer protection. In 1990, Verma and Kakoti [3] presented an algorithm on the basis of harmonic restraint, using discrete Hartley transform. Different advanced digital filtering algor- ithms such as Kalman filtering [4], Fourier-based method [5] and so on are also used in harmonic restraint differential pro- tection schemes. The harmonic restraint-based method fails to prevent false tripping of relays because high second harmonic components are generated during internal faults and low second harmonic components are generated during magnetising inrush having modern core material of power transformer [6–9]. Therefore the detection of second/fifth harmonic is not a sufficient index to determine (or discrimi- nate) between the inrush and fault condition of power trans- former. The second method consists of distinguishing inrush and over excitation currents from the internal fault currents on the basis of waveform identification. In 1986, Verma and Basha [10] reported microprocessor-based waveform differential relaying scheme. More recently, with the developments in artificial neural network (ANN) and fuzzy approaches, their application for the protection of power transformer is also getting momentum. In 2000, Ma and Shi [11] proposed an algorithm on the basis of hidden Markov model to discriminate between fault and mag- netising inrush currents. The development of ANN enhances the scope of waveform identification approach. The ANN approach is faster, robust and easier to implement than the conventional wave form approach. In 1994, Perez et al. [9] proposed an algorithm on the basis of multilayer feed forward neural network (MFFNN) to discriminate between magnetising inrush and internal fault condition. In 1995, Bastard et al. [12] presented multilayer perceptron for power transformer differential relaying. In 1997, Pihler et al. [8] reported an improved operation of power transformer using ANN. Similarly, in 2003, Moravej [13] reported ANN-based harmonic restraint differential protection for power transformer. Most of the authors used MFFNN with back propagation learning technique. From the literature review, it is evident that feedforward back propagation neural network has a good generalisation ability and learning stability [14]. In the literature, another ANN model called radial basis function neural network (RBFNN) has been reported to protect power transformer [15]. MFFNN and RBFNN have two major drawbacks. 1. Both neural network models should be trained before their use or application. 2. The learning and testing are time-consuming processes. In this paper, another type of ANN model called probabil- istic neural network (PNN) is used to protect power # The Institution of Engineering and Technology 2007 doi:10.1049/iet-gtd:20050273 Paper first received 23rd June 2005 and in revised form 10th January 2006 The authors are with the Department of Electrical Engineering, Indian Institute of Technology Roorkee, Roorkee 247 667, Uttarakhand, India E-mail: rudrafee@iitr.ernet.in IET Gener. Transm. Distrib., 2007, 1, (2), pp. 218–222 218