1572 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 18, NO. 4, OCTOBER 2003 Wavelet-Based ANN Approach for Transmission Line Protection Francisco Martín, Member, IEEE, and José A. Aguado, Member, IEEE Abstract—A new approach of digital relays for transmission line protection is presented. The proposed technique consists of a preprocessing module based on discrete wavelet transforms (DWTs) in combination with an artificial neural network (ANN) for detecting and classifying fault events. The DWT acts as an extractor of distinctive features in the input signals at the relay location. This information is then fed into an ANN for classifying fault conditions. A DWT with quasioptimal performance for the preprocessing stage is also presented. Index Terms—Artificial neural networks, discrete wavelet trans- form, transmission line protection. I. INTRODUCTION A LTHOUGH traditional digital protective relay algorithms have been proven to be greatly reliable in actual power systems, better performance is required in deregulated power systems due to increased standards of quality of service. Traditional digital protective relays present several drawbacks. For instance, they are usually based on algorithms that estimate the fundamental component of the current and voltage signals neglecting higher frequency transient components. Moreover, phasor estimation requires a sliding-window of a cycle that may cause a significant delay. Furthermore, accuracy is not assured. During the last decade, digital protective relaying of trans- mission lines has greatly benefited from the development of ar- tificial intelligence techniques [1], and more recently, from new signal processing techniques such as the discrete wavelet trans- form (DWT) [2]. As opposed to conventional techniques, the DWT takes advantage of the valuable information contained in the fast transient components of the voltage and current signals. A combination of these approaches for transformer protection has recently been proposed in [3]. In this letter, a preprocessing module based on DWT in combination with artificial neural net- works (ANNs) are used for the detection, analysis, and classifi- cation of faults events. II. SIMULATION OF FAULTED TRANSMISSION SYSTEM A system with two generators and three lines (distributed parameters model) has been simulated (see Fig. 1) using the ATP-EMTP software. The line protected is the central one (Line BC). The location of the relay is at bus B. Extensive series of simulation studies has been carried out to obtain fault transient signals for subsequence analysis. Simula- tions include ten different type of faults at 20-, 40-, 60-, 80-, and 90-km distances from the beginning of each line, several fault Manuscript received October 9, 2002. The authors are with the Department of Electrical Engineering, University of Malaga, Malaga E-29013, Spain. Digital Object Identifier 10.1109/TPWRD.2003.817523 Fig. 1. Simulated power system. resistances (0, 10, 20, 30 and 40 ), and different fault inception angles (0, 20, 40 and 60 ), and finally steady states. The fundamental frequency is 50 Hz and the sampling fre- quency is 1600 Hz. This corresponds to 32 samples per cycle. The proposed fault detection scheme is as follows: input sig- nals are preprocessed by a DWT extracting information from the transient signals simultaneously in both time and frequency domains. The output signal of the preprocessing module is then fed into an ANN that classifies the transient. The DWT con- siderably simplifies the input signal of the ANN; it reduces the volume of input data of the ANN without loss of information. This dramatically reduces the training stage in the ANN and in- creases the overall performance of the digital relay. A. DWT Selection Wavelet transforms are fast and efficient means of analyzing transient voltage and current signals. The wavelet transform not only decomposes a signal into frequency bands, but also, unlike the Fourier transform, provides a nonuniform division of the fre- quency domain (i.e., the wavelet transform uses short windows at high frequencies and long windows for low frequency com- ponents). Wavelet analysis deals with expansion of functions in terms of a set of basis functions (wavelets) which are generated from a mother wavelet by operations of dilatations and transla- tions [4]. A wavelet transform is defined by a sequence of functions (low-pass filter) and (high-pass filter). The scaling function and wavelet are defined by the difference equations where and . A sequence defines a wavelet transform. There are many types of wavelets such as Haar, Daubechies, Morlet, etc. For the relay to operate in real time, this work uses wavelets 0885-8977/03$17.00 © 2003 IEEE