Research On 2D Localization Method Of DAS Vibration Sources Based On Ensemble Learning Model

Distributed Acoustic Sensing(DAS)systems are based on Phase-sensitive Optical Time-domain Reflectometry(?-OTDR)technology and are rapidly being used in longrange security monitoring due to their unique advantages.It has been rapidly applied in long-distance safety monitoring due to its unique advantages.In addition to the basic detection and identification functions,current research is increasingly focused on mining deeper information such as location to help make more accurate decisions.In the safety detection of long-distance buried fiber optic cable,due to the special nature of vibration signals received by long-distance DAS systems and the complexity of signal transmission under buried conditions,it is challenging to achieve two-dimensional location monitoring of vibration threat events in such environments and there are very few studies on it.This paper proposes a vibration source localization method based on the spatial energy attenuation characteristics based on the vibration signal detected by DAS to estimate the vertical offset distance and threat level of a specific vibration source relative to the buried fiber optic cable.Experimental tests by us have shown the limitations of the traditional source localization method when applied to long-distance monitoring of DAS and buried environments in urban and rural areas.Our field experimental tests show that the average recognition rate of striking signals using this method can reach 99%,which has better recognition effect and timeliness.The specific work is as follows.(1)The current status of research on safety monitoring and positioning technology of fiber optic cables is analyzed,as well as the current application of DAS in the field of safety monitoring.The development status of DAS vibration source positioning technology is discussed,and it is found that there is little research on two-dimensional positioning technology that identifies the vertical offset distance of vibration sources in the horizontal direction on the basis of longitudinal positioning along buried fiber optic cables.(2)In order to explore the methods applicable to the two-dimensional positioning of vibration sources in DAS,specific experimental tests were conducted on TDOA and MUSIC methods which are commonly used for sound source positioning,and their limitations and infeasibility in the application of DAS buried fiber optic cable safety monitoring and positioning were found.Based on this,we propose a two-dimensional localization idea and method based on feature mining of DAS vibration source signals under different vertical distance conditions and using machine learning models to identify the vertical offset distance.(3)We analyzed the multi-class features of DAS vibration signals are mined and their distinguishability.And then the extraction method of spatial energy decay features is proposed.An ensemble learning model based on two-stage Stacking is constructed which based on the recognition results of different pitches by a single model.It fuses different machine learning models to assist in mining deeper features distinguishable laws and performs more accurate recognition.(4)The feasibility of our method was verified by field experiments in different conditions.In the case of hierarchical classification of threat level,the average accuracy of identification of regular events such as striking can reach 99%,and the accuracy of identification of typical threat events such as mining is around 89%,and the test of realtime shows that the algorithm's running time can meet the actual security monitoring requirements.