Single and composite disturbance event recognition based on the DBN-GRU network in φ-OTDR.

Abstract

An end-to-end deep learning model based on the deep belief network (DBN) and gated recurrent unit (GRU) is proposed to recognize the single disturbance events and composite disturbance events in the phase-sensitive optical time-domain reflectometer (φ-OTDR). Making use of the DBN to fit the original data, five kinds of single disturbance events can be effectively recognized with the GRU network as the classifier. An average recognition accuracy of 96.72% with a short recognition time of 0.079s can be achieved for single disturbance events. Moreover, the proposed method is also applied for recognizing composite disturbance events. Four kinds of composite disturbance events can be recognized with an average recognition accuracy as high as 90.94%, and the corresponding recognition time is only 0.084s. Up until now, there have been fewer reports about the recognition of composite disturbance events in φ-OTDR systems. High recognition accuracy and short recognition time make the model based on DBN-GRU more capable in a high sensitivity, real-time φ-OTDR system.