Research On Recognition Method For Ф-OTDR Optical Fiber Pre-warning Pattern

The distributed fiber pre-warning system based on coherent Rayleigh scattering(Ф-OTDR) through detecting the light intensity of each part of optical fiber to achieves the purpose of detecting and positioning the external vibration signal. It is very important in optical fiber pre-warning system is the identification and classification of vibration signals. It will result in not only a waste of manpower and physical but alsoeven resources processing tardinessand life and propertysafety if there is a false alarm. So how to accurately identify the intrusion event types, timely alarm, reduce false alarm, avoid unnecessary waste of resources has been thekey problem in the study on optical fiber pre-warning system and has attracted a wide spread attention.According to the defect and deficiency of the Ф-OTDR fiber pre-warning system, this paper proposes to use a high precision and high efficiency pattern recognition methodbased on our group’s research, and mainly studies the signal characteristic from intrusion events outside the gas pipes.Through analyzing the principle of Ф-OTDR distributed optical fiber sensor, studying on the propagation process of the vibration signal, and establishing a disturbance signal propagation model and a signal conversion model of optical fiber pre-warning system, the built model was verified using field collected signal. The model can fully explain the process of the signal generation when the intrusion events affect the optical fiber pre-warning system. Because the scattering curves can’t characterize the signal characteristic, we can obtain the signal in time and space field through reconstruction. Through the simulation from the established model, we can analyze the characteristics from different intrusion events and factors of the differences between pattern recognition methods.According to the characteristics of the signals in time and space, we firstly locate the disturbance signals, and then extract time domain signal of the position. Because features extraction method contains the localization process which is not only a waste of time, but also affected by the initial phase of the prone positioning error, resulting in separability differences in feature vectors. So this paper puts forward pulse scanning morphological feature extractingmethod. We obtain the region segmentation of the signal and then extract the morphological featureof the intrusion events as the feature vectors and select the feature from calculation of scatter matrix to obtain the optimal feature vector.In order to meet the demand of Ф-OTDR fiber pre-warning system, aiming at the non-linear and small sample characteristics of those three typical instruction events, this paper proposed related vector machine classifier method to recognize various instruction events. A pair of multi-classification decisions was used to meet the needs of the relevance vector machine multi-classification, and cross validation method was used to test the performance of the classifier.The method presented in this paper is verified through three kinds of typical pipeline intrusion events, the results show that pulse scanning morphological feature extracting method can improve the accuracy and the efficiency of the recognition, The pattern recognition method proposed in this paper in Ф-OTDR fiber pre-warning system can meet the real-time monitoring requirement.