Application Research Of One-dimensional Convolutional Neural Network In Identification Of Leakage Aperture Of Natural Gas Pipeline

In recent years,the micro-aperture leakage of natural gas pipeline has become an important factor threatening the operation of the pipeline,if the size of the leakage aperture can be accurately identified,it will help the safety operation department to accurately estimate the degree of damage of the leak,and make corresponding measures to reduce the loss,so the research on the identification of natural gas pipeline leakage aperture is of great significance.In this paper,the leakage aperture of the pipeline is taken as the research object,and a method for identifying the leakage aperture of natural gas pipeline sits based on one-dimensional convolution neural network is proposed.Firstly,it discusses the research background of natural gas pipeline leakage,and explains the significance of the research on the pore of pipeline leakage,then lists the current situation of research on natural gas pipeline leakage detection at home and abroad,and analyzes the challenges and opportunities of leak detection.Further,it expounds the structure and characteristics of convolution neural network,and the construction and training process of deep convolutional neural network is analyzed in detail.Finally,the strategy of preventing overfitting in the training process of deep neural network is summarized.Aiming at the problems of the large redundancy of raw data,feature extraction and excessive reliance on prior knowledge and the high real-time requirements of the system,a leakage aperture recognition model combining compression sensing with one-dimensional convolution neural network is proposed.Firstly,the random Gaussian matrix is used to compress and process the original leakage signal,and most of leakage information is obtained with less compressed sensing domain data.Furthermore,a deep one-dimensional convolutional neural network is constructed,and the compressed acquisition data is fed into the network to realize adaptive feature extraction with high-accuracy leakage aperture identification.At last,the paper also analyzes the impact of the main parameters.The experimental results show that the proposed method can quickly and accurately realize the leakage aperture identification of natural gas pipelines,the overall performance is superior to traditional classification methods.Aiming at the problem that the environment along the pipeline is complex and the signal-to-noise ratio of the leakage signal is low,which leads to the low accuracy of the traditional leak aperture recognition method,a multi-scale convolutional neural network is proposed to study the leakage aperture of natural gas pipeline.First,the global characteristics of the original leakage signal are extracted by using the large convolution core,and then the proposed multi-scale block is used to extract the upper layer output of the convolution neural network for multi-scale feature in parallel.and the generalization ability of the model is improved by using BN processing and Dropout processing.At last,the paper also analyzes the impact of the main parameters.The experimental results show that the proposed method can accurately identify the leakage aperture of natural gas pipeline,and has a better robustness in the low signal-to-noise ratio environment.