Methane Gas Detection Based On Tunable Lasers And Deep Learning

The main component of natural gas is methane,which accounts for about 85%.In the process of natural gas transportation,there will be leakage in the transportation pipeline when working for a long time,when the methane concentration in the air reaches 5%,the oxygen concentration is greater than or equal to 12%,and an explosion will occur when an open flame is encountered.In order to prevent safety accidents,methane gas detection around the pipeline is required.The combination of traditional TDLAS technology and deep learning to detect methane gas with high precision is of great significance to prevent natural gas explosions in the chemical industry,safety gas explosions in coal mines,and large-scale natural gas leakage.The main research contents of this paper are as follows:.First,the TDLAS system for methane gas concentration data acquisition was built.Then,the second harmonic signal of methane gas concentration was collected by the system,and the noise in the system was analyzed.Secondly,in order to solve the problem of long time in the process of artificially screening the second harmonic signal of methane gas and fitting the optimal amplitude-concentration straight line in the traditional TDLAS technology,the ability to extract signal features is improved,and the detection accuracy of trace methane gas concentration is further improved.A trace methane gas concentration detection method based on wide convolution and wide convolution kernel one-dimensional convolutional neural network is proposed.In this method,the feature extraction of trace methane gas second harmonic signal is performed by using a wide convolutional layer and a wide convolutional kernel one-dimensional convolutional layer,so that the network can obtain a longer sequence in the methane gas concentration signal and the characteristic relationship between the sequence boundary information and the gas concentration after one convolution.The maximum pooling layer is used to retain the main characteristics of the signal,remove redundant information,reduce the amount of calculation,and simplify the network complexity.The test results show that this method improves the efficiency and accuracy of methane gas detection.Finally,aiming at the problem of fluctuation interference caused by human factors and noise on the amplitude of the second harmonic signal of methane gas concentration in traditional TDLAS technology,a deep residual shrinkage module is added on the basis of the fourth chapter model to reduce the number of convolutional layers and further reduce the number of model parameters,and a trace methane gas concentration detection method based on deep residual shrinkage wide convolutional one-dimensional convolutional neural network is proposed.In this method,the second harmonic signal data loaded into the model is enhanced for data,and then the one-dimensional convolutional layer is used to extract the noisy signal,soft thresholding is added to reduce the noise of the noisy signal,and finally the methane gas concentration is output by the fully connected layer.The experimental results show that this method can effectively reduce the influence of noise on the amplitude fluctuation of the second harmonic concentration signal of methane gas,reduce the interference of amplitude fluctuation on gas concentration detection,and improve the accuracy of methane gas concentration detection.