TDV-LSTM: A New Methodology for Series Arc Fault Detection in Low Power AC Systems

AC are faults, especially series arc, can produce very high temperatures and can easily ignite combustible materials, representing one of the most important causes of electrical fires. None of the detection methods presented in literature guarantees a perfect discrimination. The challenge for fault detection algorithms is to have good performance under different circuit configurations in which series arc faults are difficult to identify. In this paper, temporal domain visualization long short term memory network (TDV-LSTM) based methodology is proposed. The temporal domain loop current of the typical loads is framed and arranged into a special matrix image as input data and using LSTM network to identify the feature image. The general classification accuracy of every half-cycle signal in the 10 tested classes could reach 98.9% and could be higher if adjusting parameters perfectly. Experimental results clearly demonstrate benefits of TDV-LSTM and show its accuracy for the AC series arc fault identification.