Study On Leak Detection And Identification Method Of Natural Gas Pipeline Based On Negative Pressure Wave

It is of vital importance to ensure the safety in the production of chemical industry.As an important factor affecting the severity of accident consequences,the accurate and rapid detection and judgment of the equipment status when a fault occurs can effectively reduce the resulting accident consequences through finding the fault in time during the process of the equipment operation process.At present,the fault diagnosis technology is mainly used to monitor the running status of equipment in real time to realize the accurate detection of faults.In particular,the data-based intelligent fault diagnosis technology has become a research hotspot for chemical process.In the natural gas chemical production and transportation system,the leakage of natural gas pipelines occurring from time to time is extremely easy to cause serious accidents along with the increase of the service life and of the length of pipelines.It is of great significance to monitor the running state of the pipeline in real time,and detect and identify the occurrence of leakage events in time and accurately.One of the most active hot spots is to detect pipeline leakage by means of the artificial intelligence technology.The following research progress has been made in this thesis:(1)It is impossible to achieve the repeated data under the same conditions due to the remarked differences of various factors(e.g.,technology,operating environment and parameter,etc.)in the actual natural gas pipeline leakage events.Such facts would yield some considerable adverse results(i.e.,result differences,small samples,ect.)to the research of data analysis and leakage detection algorithms.For addressing these problems,this thesis aims to design the scheme,establish the platform and develop the system software for leakage detection of natural gas pipeline,which can generate the operation data of specific operation parameters according to the needs,and reproduce the leakage event data.(2)The collected data are not perfect because of various of reasons.For example,the abnormal or missing values are collected since the data acquisition equipment may work abnormally;the useful signals may be disturbed by noises since the negative pressure wave signal is also vulnerable to interference from the pipeline and the outside world.In this thesis,the Prophet method is used to repair the abnormal data,and the differences of denoising effect are compared and analyzed between the separately improved wavelet threshold algorithm and the combined denoising effect with the variational mode decomposition(VMD)algorithm.The proposed VMD-Wavelet algorithm improves the accuracy of extracting the effective modal components and the stability of the denoising effect.The proposed algorithm is of excellent performance in denoising the negative pressure wave signal being subject to noises,which lays the foundation for subsequently realizing the accurate detection of the pipeline leakage event.(3)There exists a challenging problem that is to extract the feature in the process of pipeline leakage detection technology.To be specific,the different levels of pipeline leak may lead to different signal characteristics due to the complexity of pipeline service environment and operating conditions.Although the existing feature extraction methods have certain advantages in describing time-varying signals,it is far from enough to analyze the leakage situation only using the statistical features,which could degrade the performance of pipeline fault detection algorithm.In recent years,the deep learning technique has attracted widespread concern and applied in the field of fault detection due to its powerful feature representation and classification capabilities,which has greatly improved the detection accuracy.This thesis proposes a more consummate feature extraction algorithm based on the deep learning.The pipeline health status is first described from four perspectives of statistical analysis,signal processing,signal local feature,and depth mining,and then extracts the features of different fields,respectively.The existing redundant features could increase the computational complexity of the model after directly inputting the features of different dimensions into the model,thus affecting the convergence speed and recognition accuracy of the model.For address this problem,in this thesis,the feature alignment method is then used to unify the dimensions of features in different fields,and then the Marine Predators Algorithm(MPA)is used to select the optimal features and produces the perfect feature set,for the purposes of alleviating the problem that a single feature cannot fully describe the fault data and improving the performance of the diagnostic model.(4)The traditional fuzzy c-means(FCM)algorithm require high complexity of clustering samples and high quality of initializing the cluster center.For dealing with this problem,the acquired feature data set is jointly optimized via the MPA and improved particle swarm optimization(PSO)algorithm.The weight gradient descent method and quantizer is introduced such that the convergence speed is improved on the basis of ensuring the convergence of the algorithm,and the possibility of falling into the local optimal value is reduced in the process of convergence.Meanwhile,the clustering effect and clustering stability of traditional FCM algorithm are improved.In the end,the thesis realizes the experimental verification of fault type identification of pipeline leakage detection data.The experimental results show that the MPA-QPSO-FCM algorithm proposed in this thesis can accurately identify the pipeline operation status with the average accuracy 91.53% and the average F1 value 91.25%.The proposed algorithm is more suitable for leakage detection of natural gas pipeline due to the better coordination and the trade-off between accuracy and recall rate.In summary,this thesis is concerned with the related problems of natural gas pipeline data preprocessing,feature data set and pipeline operation status recognition by combining theoretical research and experimental verification in the context of natural gas pipeline leakage detection.A kind of fast and accurate intelligent diagnosis methods is proposed by means of building a natural gas pipeline leakage detection experimental platform,which has important practical significance to reduce the harm of leakage accidents.