Deep neural network enabled fault detection in LVDC microgrid using empirical mode decomposition

An advanced method of fault recognition and location identification for a low voltage direct current (LVDC) microgrid is introduced in this paper. Sources are tied-up with power electronics converters to the microgrid. Components of converters are very much at risk of damage during a fault. It became necessary to isolate the healthy part fast. Again, magnitude of fault current increases at a very high rate during the fault, the entire system might get de-energized and making it difficult to identify the fault location. With a view to this, an empirical mode decomposition (EMD) based deep learning classifier is proposed to recognize and locate the fault in the LVDC microgrid. EMD based methods are suitable for a system where transient values of the fault current are an important feature to detect a fault. EMD decomposes the fault signal into different components to detect the temporal variation (transient caused in segment current because of the fault). Convolutional neural network (CNN) the architecture of deep neural network (DNN) is used to classify the signal into normal and faulty, as well as to recognize the fault location in the system. The suggested technique is evaluated through MATLAB/Simulink simulations. The data set needed for the classification of normal, fault, and abnormal condition signals are collected under different fault resistances, fault locations, and various system conditions. The outcome shows that the classifier has a high accuracy of 94.97 % and a low error rate compared to other classification models such as Gaussian mixture model (GMM) and support vector machine (SVM). The method is also validated on designed and developed laboratory ring-type LVDC microgrid hardware. Texas Instruments' TMS320F28069 digital signal processors (DSP) are employed for the execution of the proposed protection scheme. The hardware result also demonstrates the proposed methodology's superior performances and gives an accuracy of 90.60%.