Investigation On Pattern Recognition Based On Random Forest Classifier For Phase-Sensitive OTDR Sensing System

There is an increasing interest in the research of phase-sensitive optical time domain reflectometer(?-OTDR)distributed optical fiber disturbance sensing system due to its high stability and precision.In practical applications,environmental and human interferences may lead to false alarm,resulting in high nuisance alarm rate(NAR).To reduce the nuisance alarm rate of ?-OTDR distributed optical fiber disturbance sensing system,an effective event identification method based on a random forest(RF)classifier is proposed in this dissertation.The main achievements of this dissertation are as follows:(1)The signals of ?-OTDR distributed optical fiber disturbance sensing system are analyzed,and the disturbance signals in four cases(watering,knocking,pressing and no disturbance)are extracted.The time-domain features of the disturbance signals are analyzed.Features are normalized to avoid over-fitting problem,which laid the foundation for the follow-up event identification.(2)The event identification method for ?-OTDR distributed optical fiber disturbance sensing system based on decision tree is proposed,implemented and tested.Experimental results show that the proposed method based on decision can effectively identify the different disturbance events.The experimental identification rates for watering,knocking,pressing and no-disturbance events reach 90.19%,92.06%,89.95%,and 97.37%,respectively,and the average identification rate is 92.36%.But the stability and robustness of the algorithm is poor.(3)The event identification method for ?-OTDR distributed optical fiber disturbance sensing system based on random forest(RF)is proposed,implemented and tested.The results show that the proposed method based on RF can effectively distinguish the four kinds of disturbance events,the identification rates has been significantly improved compared to the event identification method based on decision tree.The experimental identification rates for four kinds of disturbance events reach 93.79%,97.36%,97.06%,and 98.12%,respectively,and the average identification rate is over 96%.