Arc Fault Detection in DC Microgrid Using Deep Neural Network

Abstract

DC microgrid getting acceptance because of recent development in renewable energy technology. The solar PV is the major part of DC microgrid. Solar PV is always at risk of fire hazard due to arc fault and if the fault is not detected in time then it not only damages the microgrid but also causes a serious threat to the safety of the operator. Exiting pattern base fault recognition techniques do not perform properly due to the nonperiodic nature of arc fault. Also, switching noise signals from power electronics converters affect the detection techniques. This paper proposes a deep neural network based approach for the detection of the arc faults in DC microgrid. Multi-layer perception(MLP)/Dense neural networks and Convolution Neural Networks(CNN) have been employed for the detection of the arc fault. A detailed analysis for both dense and convolution network have been performed to validate the choice of the neural network. The MATLAB simulation is developed to test the proposed methodology. Both MLP and CNN percentage accuracy was comparable under arc fault detection. However, CNN performance was better under noisy environment.