A NEURAL NETWORK APPROA CH TO FA UL T DIA GNOSIS IN INDUSTRIAL POWER NETWORKS USING SEQUENTIAL CURRENTS

This paper presents a new neural network approach for an on-line fault diagnosis using the data obtained From digital fault recorders. This method applies forward multilayer perceptronasa learning system in which the pre-processing phase uses some concepts of digital signal processing such as Fourier transform. We begin our exploration with filtering the fundamental component of short circuit current. After that, a preprocessor unit is used to reduce the number of training patterns and also to estimate the location of short circuit fault. In the next step, we proceed with classification of the common four fault types: one-phase-to-ground, two-phase, two-phase-to-ground and three-phase-to-ground short circuit. In order to show the capability of the proposed method some simulations have been performed. The results are very encouraging indicating that the proposed neural network approach can be used for short circuit problems in real-size Industrial Power Networks.