ANN-Based Modeling of Directional Overcurrent Relay Characteristics Applied in Radial Distribution System with Distributed Generations

The time multiplier setting (TMS) and the relay pickup current are very important parameters in modeling of DOCR characteristics. These two parameters can be adjusted to move the time current characteristic (TCC) curve to the position suitable to the protection coordination. Distributed generations (DGs) can cause change of fault current in the system and this can affect the protection coordination. In this paper, artificial neural network (ANN) based on Levenberg-Marquardt algorithm is used to model the DOCR characteristics where the trained ANN model is able to compute the time multiplier setting, the pickup current as well as the trip time of each DOCR in terms of changes caused by DG connection. A long radial distribution feeder of 7 buses penetrated by DGs is designed and simulated in order to get the training data set. The training results are quite interesting and encouraging with a mean squared error (<;span style="text-transform: uppercase;">MSE<;/span>) equal 1.6766e-l2. At the end, a sample of ANN outputs is implemented in ETAP software for further verification of the developed model and it works perfectly.

I. Introduction

The overcurrent relay is widely used in protection of electrical distribution feeders which are naturally radial, i.e, there is a single power source supplying the feeder and the current flows only in one direction, FROM the source to the consumer's end [1], [2]. In this situation, the fault discrimination is achieved by using a combination of two overcurrent relays (OCRs) where one OCR close to the fault known as “primary relay” must act very fast. In case the primary relay fails to clear the fault, the other relay known as “Backup relay” must intervene to clear that fault.