Disturbance event identification and localization based on temporal convolutional network (TCN) and sliding window algorithm

In this paper, a disturbance event identification and localization method, based on Temporal Convolutional Network (TCN) and sliding window algorithm, is proposed. By constructing a 25km long ϕ-OTDR experimental system and using the sliding window algorithm to locate disturbance locations, the TCN extracts feature and identifies disturbance events. The proposed method can effectively identify five types of disturbance events, including four types of real disturbance events, (watering, climbing, knocking and pressing), and one type of false disturbance event. The average identification accuracy for the five types of disturbance events exceeds 96%, with zero deviation between the actual and predicted locations. The high identification accuracy and localization precision highlight the value of this method in ϕ-OTDR system applications.