Research on underground cable abnormal vibration identification technology based on multi-model fusion

Abstract

With the full implementation of cable cabling in the urban core, thousands of kilometers of underground cables have placed new demands on cable operation management. In finding faults, fault identification by trial delivery of faulty lines section by section is inefficient and difficult, and there is no effective technical means to locate faults quickly. To address these problems, this paper constructs a multi-model fusion strategy of statistical learning + deep learning to achieve effective improvement of algorithm fitting effect and generalization ability, which can learn various types of features as comprehensively as possible, achieve differential extraction of distributed fiber optic timing features through differential feature construction techniques, and obtain spatio-temporal information of vibration events along the fiber optic cable for training to effectively solve the above problems and achieve more reliable and accurate prediction.