Research On Oil Pipeline Leak Location And Identification Based On Optical Fiber Sensor

In recent years,pipelines have been one of the most common systems for storing and transporting liquids.The Bragg fibre optic grating sensor is a low-cost,interference-resistant and low-energy sensor.Because the pipeline is affected by external conditions,leakage events often occur,resulting in economic losses and waste of manpower.Therefore,it is of great significance to study the monitoring system of the vibration signal spectrum that appears in the pipeline wall by using new sensors for leakage effects.To solve the problems of difficult extraction of physical features,low recognition accuracy,difficult system installation and excessive external interference in oil pipeline positioning systems,this thesis proposes a Bragg fiber grating sensor based on the combination of variable modal decomposition of the cylindrical spiral positioning method,first,the multi-point leak generated signal Hilbert transform to find the resolved signal,the signal is shifted to the baseband,the use of Gaussian smoothing to estimate the Secondly,this paper combines the idea of immune algorithm to improve the particle swarm algorithm to solve the optimal Lagrangian factor,determine the modal components,extract the signal frequency response characteristics and then find out the arrival time point of each leakage signal,establish the mathematical positioning model of the spiral curve and the cylindrical spiral model;finally,the three-dimensional location of the leakage is determined.To address the problem of leak type identification,this thesis proposes an improved particle swarm variational modal decomposition method combined with random forest for leak type identification.Firstly,a linear decreasing weight method is proposed,where the inertia weights are linearly decreasing from large to small,with adaptive step size,to solve for the optimal parameters of the variational modal decomposition;secondly,a machine learning method is used to establish the input matrix with energy,cross entropy and wave peak as features;then,a multi-learning machine classification is used to train single-point leaks and multi-point leaks respectively to identify different types of leaks;finally,with reference to the 40 m pipe network data set to test the leak identification accuracy as well as the algorithm time cost.The experiments show that,with 6500 data sets of DN200 pipeline inspection,the optimisation method of this thesis has an error of 2.3914 cm in locating 60 cm pipes;the single-point leak aperture of 1 mm and hydraulic pressure of 5 bar has an accuracy of86% in identifying small leaks;the random forest regression algorithm proposed in this thesis has an accuracy of 82% in identifying multi-point leaks.The improved particle swarm algorithm has certain advantages in terms of average standard deviation and time index comparison,its speed of finding the best and strong adaptability of parameters.