Leak Identification And Location Of Submarine Pipeline Based On Support Vector Machine

With people's exploration of marine resources,submarine oil and gas pipelines have become the lifeline of offshore oil and gas transportation,playing a vital role in the development of modern industry and national economy.However,because of the complicated environment the submarine pipelines lying in and various factors such as inevitable corrosion and man-made damage,pipelines leakage accidents frequently emerge in recent years.Therefore,research on safe and feasible pipeline leakage monitoring technology is of great practical significance.This paper,with the application of optical fiber sensing technology and the combination of support vector machine,classifies diverse working conditions in the course of pipeline operation,and further locates the leakage points.This paper mainly contains several following aspects:This paper summarizes the development history and current development status of submarine pipelines at home and abroad,and discloses the current safety issues of submarine oil and gas pipelines as well.Besides,existing detection methods as well as the merits and demerits of diverse oil and gas transportation pipelines are briefly introduced.The review,combined with the development status of pipeline leakage monitoring based on machine learning,puts forward the research direction and contents of the present paper,and further describes its research significance.In the present paper,an experimental device of gas pipeline leakage is built through the selection and calculation of main equipment,and its design principle and philosophy are indeed briefly described.Meanwhile,serving as a simulation platform for various working conditions,the experimental platform lays the foundation for later acquisition of experiment data and classification accuracy of multiple working conditions.The basic theory of the support vector machine is introduced in detail.At the same time,the categorization of abnormal working conditions of pipelines is realized through the processing of multiple-condition experimental data acquired from the experimental platform and the combination with multiple-classification learningmethods of support vector machine.During this course,the kernel parameters are optimized by 5-fold cross-validation to obtain the highest prediction accuracy.Through comparing the effectiveness of diverse kernel functions,the optimal parameter combination of each kernel function is obtained with the use of 5-fold cross-validation.Then elapsed time used to search the maximum accuracy is considered,and eventually the radical basis kernel function is selected as the kernel function of pipeline anomaly classification based on support vector machine(SVM).In consideration of the above,the parameters of the radical basis kernel function are optimized through the combination of diversified optimization algorithms,and similarities and differences of each optimization algorithm are compared,thus achieving the highest classification accuracy.On the basis of support vector regression theory,the support vector regression(SVR)is applied to locate the pipeline leakage.Besides,the anti-interference ability of the model to noise is observed through the addition of white noise to simulate the interference of normal operation.Meanwhile,the genetic optimization algorithm is applied to optimize the key parameters of the model,thus improving the accuracy of the model for leakage locating.In view of this,the model's ability to suppress noise and the number of sensors for the effect of noise suppression are investigated.The methods adopted in the present paper are conducive to the pre-monitoring and early post-detection of pipeline leakage,and also are of positive significance for the real-time monitoring and locating of pipeline leakage.