Pengaruh Masukan Dan Fungsi Aktivasi Terhadap Kecepatan Pelatihan Jaringan Syaraf Tiruan (JST) Modular Sebagai Klasifikasi Dan Estimasi Lokasi Gangguan Pada Saluran Distribusi Bawah Tanah PT. Pertamina RU II Dumai

Abstract

This paper describes a method of modular artificial neural network (ANN) for identifying and locating short circuit faults in underground power distribution line at PT. Pertamina RU II Dumai. The underground power distribution line is modeled using ATP-EMTP software. ANN-based fault classifier made from ANN-based fault detection in different phases and ground which work in parallel. ANN- based fault locator made form ANN which is estimated locations of fault occurs for each different types of fault. The ANN-based fault classifier of fault was performed variety of the given input and The ANN-based fault locator was performed variety of the activation function type which is deployed in different layer. The study result shows 4 different sampling of voltage value as ANN-based fault detection input in different phase fault. In the another result shows 4 different sampling of zero sequence current value as ANN-based fault detection input in ground fault. The last result shows the activation function of ANN-based fault locator is sigmoid bipolar – sigmoid biner – linear in the different ANN-based fault locator. All the previous result was aggregated to enhance the training phase in order to speed up in achieving MSE target of 1E-5.
Keywords: modular artificial neural network (ANN), fault classifier, fault locator, short circuit faults , underground power distribution line