| In recent years,with the rapid development of society,distributed optical fiber sensing technology has been widely used in safety and health monitoring of bridges and tunnels,oil and gas pipelines,railway operation,and real-time intrusion monitoring of perimeter security due to its advantages of long distance continuous real-time detection.In the process of detecting and monitoring optical fibers,the continuous accumulation of data has become a huge amount of optical power data.In order to make full use of the massive historical optical power data,the big data technology can be applied to provide valuable information for fiber detection.This paper takes the actual project as the background and real-time detection as the purpose.Based on the optical fiber big data,it studies the real-time detection of optical fiber events and prediction of optical fiber loss.The main contents of this paper are as follows:(1)According to the relationship between big data and deep learning,the DBN model in deep learning is used to classify the optical power data.The optical power data are divided into four categories: normal,stress,breakpoint and disturbance.Different types of optical power data are randomly selected to form the training set and test set.The DBN model is trained by the training set.The trained DBN model is used to classify the test set.And the classification results are compared with the actual category labels to obtain the classification accuracy of the DBN model.(2)According to the characteristics of wavelet transform with singularity detection and non-stationary signal processing,wavelet transform is used to locate the optical fiber events.As for the stress and breakpoint,they are static events.The optical fiber data are processed by bior3.5 discrete wavelet to decompose the optical fiber data in six layers,reconstruct the approximate part of the sixth layer,conduct spatial domain difference on the reconstructed data,and set the threshold for positioning.For disturbance,which is a dynamic event,optical fiber data are processed by gaus1 continuous wavelet to obtain the wavelet coefficient,calculate the difference between adjacent extreme points,and set a threshold for positioning.Using different types of optical power data,the two localization algorithms are simulated and verified.(3)The overall flow of the optical fiber disturbance detection is given by combining classification and location,that is,the optical power data is acquired,and the data after resampling and normalization processing are classified through the DBN model.According to the classification results,different positioning algorithms are used for positioning,and the final output contains the detection results of event type and event location.In order to evaluate the real-time performance of the detection algorithm,the optical fiber detection is experimentally verified.(4)In order to improve the accuracy of fiber loss prediction,a linear optimal combination model is constructed for prediction.The gray model GM(1,1),ARIMA model and Elman neural network were combined with a certain proportion coefficient to obtain the combination model.The value of each proportion coefficient was obtained by genetic algorithm.The prediction experiments of each model were carried out using the test data of optical fiber loss in each quarter for 11 years in a relay section,and the prediction errors of each model were calculated respectively.The experimental results show that:(1)the classification accuracy of the DBN model is 99.55%,which can effectively classify the optical power data.(2)the sensing distance of the optical fiber system is 100 km,and the SR is 50 m,the positioning error of the positioning algorithm is acceptable for the long distance optical fiber.(3)the detection time of each group of optical power data is about 10 ms,which can achieve the purpose of real-time detection.(4)compared with each individual prediction model,the combined prediction model has smaller prediction error and higher prediction accuracy. |
PDF Full Text Request