A method and system for measuring methane concentration based on direct absorption spectroscopy

Abstract

The present application relates to the technical field of gas measurement, and discloses a methane concentration measurement method and measurement system based on direct absorption spectrum, which comprises the following steps: collecting N groups of methane samples with different concentrations to obtain corresponding real methane transmission spectrum data; adding Gaussian noise to the simulated methane transmission spectrum to obtain simulated methane transmission spectrum data; improving a deep residual contraction network model to obtain an improved deep residual contraction network model, and constructing a methane concentration measurement model; training the methane concentration measurement model by using training set data and verification set data to obtain a trained methane concentration measurement model; obtaining actual methane transmission spectrum data in a target scene, and inputting the actual methane transmission spectrum data into the trained methane concentration measurement model to obtain methane concentration data corresponding to the target scene; and the present application solves the problems of existing methane concentration measurement methods based on direct absorption spectrum, such as large noise interference and low measurement accuracy.

Description

Technical Field

[0001] This invention relates to the field of gas measurement technology, and in particular to a method and system for measuring methane concentration based on direct absorption spectroscopy. Background Technology

[0002] Methane gas is the second largest greenhouse gas after carbon dioxide, significantly impacting global climate and the environment. Colorless, odorless, and highly flammable and explosive, it is a major component of natural gas and widely used as a fuel in industry and society. Currently, the continued use and emission of methane gas has become a significant factor contributing to global warming and climate change. Furthermore, in the industrial sector, methane explosions in mines and leaks in natural gas systems also pose threats to human safety. Therefore, conducting highly sensitive, accurate, and robust methane concentration research and monitoring is crucial for both climate change research and industry. Optical sensors based on absorption spectroscopy are widely used in methane detection due to their advantages such as high sensitivity, non-contact operation, and fast response speed.

[0003] Direct absorption spectroscopy (DAS) is a gas detection technique based on tunable diode laser absorption spectroscopy. Due to its simple system setup, high resolution, real-time performance, and calibration-free operation, it is widely used in environmental monitoring and industrial production. However, the detection sensitivity and stability of DAS are susceptible to various types of noise interference, primarily categorized into two types: electronic noise and optical noise. Electronic noise typically originates from electronic components within the system and usually manifests as high-frequency white noise. Optical noise mainly arises from interference and multi-surface reflection effects in the optical path, exhibiting as low-frequency noise. Currently, common noise suppression methods are software-based, including Savitzky-Golay filtering, wavelet thresholding, and Kalman filtering. However, these methods require selecting appropriate parameters and can only handle specific types of noise. Therefore, existing methods for measuring methane concentration based on DAS suffer from significant noise interference and relatively low measurement accuracy. Summary of the Invention

[0004] This invention provides a method and system for measuring methane concentration based on direct absorption spectroscopy, which solves the problems of large noise interference and low measurement accuracy in existing methane concentration measurement methods based on direct absorption spectroscopy.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] In a first aspect, the present invention provides a method for measuring methane concentration based on direct absorption spectroscopy, comprising:

[0007] N groups of methane samples with different concentrations were collected to obtain the corresponding real methane absorption spectra, and the baseline of the real methane absorption spectra was removed to obtain the corresponding real methane transmission spectrum data, where N is a positive integer greater than 1;

[0008] M simulated methane transmission spectra with different concentrations were generated by simulation, and Gaussian noise was added to the simulated methane transmission spectra to obtain simulated methane transmission spectrum data, where M is a positive integer greater than 1.

[0009] A deep residual shrinkage network model was constructed, and the residual shrinkage building units in the deep residual shrinkage network model were improved to obtain an improved deep residual shrinkage network model. A methane concentration measurement model was then constructed based on the improved deep residual shrinkage network model.

[0010] The simulated methane transmission spectrum data is used as the training set data, and the real methane transmission spectrum data is used as the validation set data. The methane concentration measurement model is trained using the training set data and the validation set data to obtain the trained methane concentration measurement model.

[0011] Obtain the actual methane transmission spectrum data corresponding to the methane sample in the target scene, and input the actual methane transmission spectrum data into the trained methane concentration measurement model to obtain the methane concentration data corresponding to the target scene.

[0012] Optionally, the step of collecting N groups of methane samples with different concentrations to obtain the corresponding true methane absorption spectra includes:

[0013] A direct absorption spectroscopy experimental setup was constructed. N groups of methane samples with different concentrations were prepared using a methane gas mixer in the direct absorption spectroscopy experimental setup. The methane absorption spectrum corresponding to each group of methane samples was collected as the true methane absorption spectrum, with the preset concentration of the methane samples prepared by the methane gas mixer as the label. Here, N is a positive integer greater than 1.

[0014] Optionally, generating M simulated methane transmission spectra at different concentrations through simulation includes:

[0015] The methane transmission spectrum under experimental conditions was simulated using a spectral database, generating M methane transmission spectrum simulation data of different concentrations. These simulation data were then used as the simulated methane transmission spectrum, where M is a positive integer greater than 1.

[0016] Optionally, adding Gaussian noise to the simulated methane transmission spectrum to obtain simulated methane transmission spectrum data includes:

[0017] Noise parameters were obtained by performing noise analysis on the direct absorption spectrum of methane under real-world conditions, and the mean, variance, and standard deviation of the noise parameters were calculated.

[0018] Using the mean and variance as the first labels, Gaussian white noise corresponding to the first label is added to the simulated methane transmission spectrum. At the same time, using the standard deviation as the second label, Gaussian noise corresponding to the second label is added to the simulated methane transmission spectrum. The simulated methane transmission spectrum with added Gaussian white noise and high-speed noise is used as simulated methane transmission spectrum data.

[0019] Optionally, the improvement of the residual shrinkage building blocks in the deep residual shrinkage network model to obtain the improved deep residual shrinkage network model includes:

[0020] A new threshold function is constructed, and this new threshold function replaces the original soft threshold function in the residual shrinkage building unit of the deep residual shrinkage network model, resulting in a new residual shrinkage building unit. The new threshold function satisfies the following relationship:

[0021] ;

[0022] In the formula, Represents the features of the input. Indicates the characteristics of the output. It is a symbolic formula function. This represents the threshold obtained through the above network threshold calculation. Represents an arbitrary constant;

[0023] The deep residual shrinkage network model containing new residual shrinkage building blocks is used as the improved deep residual shrinkage network model.

[0024] Optionally, the construction of the methane concentration measurement model based on the improved deep residual shrinkage network model includes:

[0025] A gas spectral filter was constructed using an improved deep residual shrinkage network model and a multilayer perceptron, and a gas concentration predictor was constructed using three fully connected layers.

[0026] A methane concentration measurement model is constructed using the gas spectral filter and the gas concentration predictor.

[0027] Optionally, the training process for the methane concentration measurement model includes:

[0028] S1, Freeze gas concentration predictor, uses training set data to train gas spectral filter to obtain filter model of methane transmission spectrum;

[0029] S2. Load the pre-trained weights of the gas spectral filter and train both the gas spectral filter and the gas concentration predictor simultaneously using the training dataset.

[0030] Secondly, embodiments of this application provide a methane concentration measurement system based on direct absorption spectroscopy, including a processor and a memory;

[0031] Memory, used to store computer programs;

[0032] When a processor executes a program stored in memory, it implements any of the steps of the method described in the first aspect.

[0033] Beneficial effects:

[0034] The methane concentration measurement method based on direct absorption spectroscopy provided by this invention can directly process the spectrum to obtain concentration data, which not only simplifies data processing and improves the efficiency of training and deployment, but also enhances the model's generalization ability to new data. This entire model is based on the DRSN backbone network, making it suitable for handling situations with strong noise interference. This is thanks to the threshold function in DRSN, which enables adaptive denoising, thus better extracting information from noisy spectra. Comparative analysis shows that the proposed GSF has better denoising performance than other denoising methods. In terms of concentration prediction, this invention, through transfer learning, not only accelerates the training convergence speed but also improves the accuracy and reliability of retrieving methane concentration from absorption spectra. Attached Figure Description

[0035] Figure 1 This is a flowchart of a preferred embodiment of the methane concentration measurement method based on direct absorption spectroscopy of the present invention;

[0036] Figure 2 This is a schematic diagram of the structural framework of the residual shrinkage building unit in a preferred embodiment of the present invention;

[0037] Figure 3 This is a schematic diagram of the structure of a methane concentration measurement model according to a preferred embodiment of the present invention. Detailed Implementation

[0038] The technical solution of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "connected" or "linked" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship also changes accordingly.

[0040] Please see Figure 1 This application provides a method for measuring methane concentration based on direct absorption spectroscopy, comprising:

[0041] N groups of methane samples with different concentrations were collected to obtain the corresponding real methane absorption spectra, and the baseline of the real methane absorption spectra was removed to obtain the corresponding real methane transmission spectrum data, where N is a positive integer greater than 1;

[0042] M simulated methane transmission spectra with different concentrations were generated by simulation, and Gaussian noise was added to the simulated methane transmission spectra to obtain simulated methane transmission spectrum data, where M is a positive integer greater than 1.

[0043] A deep residual shrinkage network model was constructed, and the residual shrinkage building units in the deep residual shrinkage network model were improved to obtain an improved deep residual shrinkage network model. A methane concentration measurement model was then constructed based on the improved deep residual shrinkage network model.

[0044] The simulated methane transmission spectrum data is used as the training set data, and the real methane transmission spectrum data is used as the validation set data. The methane concentration measurement model is trained using the training set data and the validation set data to obtain the trained methane concentration measurement model.

[0045] Obtain the actual methane transmission spectrum data corresponding to the methane sample in the target scene, and input the actual methane transmission spectrum data into the trained methane concentration measurement model to obtain the methane concentration data corresponding to the target scene.

[0046] In the above embodiments, the method specifically includes: S1, acquiring the direct absorption spectrum (DAS) of methane; S2, establishing a simulation dataset based on the acquired DAS signal; S3, constructing an end-to-end methane concentration measurement model based on a deep residual shrinking network (DRSN); S4, training and optimizing the methane concentration measurement model; and S5, evaluating the performance of the model.

[0047] The establishment of the simulation dataset includes: analyzing real DAS signals to obtain noise parameters; performing simulations using the HITRAN spectral database to obtain noise-free spectral data and corresponding concentration levels in the training set; and adding multiple noise levels to the noise-free signals to construct the model training set.

[0048] The process of establishing an end-to-end methane concentration measurement model based on DRSN includes: improvements to the modules in DRSN; the basic structure and module groups of the model; and performance evaluation of the model, including: filtering performance analysis of the test set absorption spectral data; concentration prediction performance analysis of the test set absorption spectral data; and model stability analysis.

[0049] Compared to other methods, the end-to-end methane concentration measurement model proposed in this invention can directly process spectral data to obtain concentration data, which not only simplifies data processing and improves the efficiency of training and deployment, but also enhances the model's generalization ability to new data. This entire model is based on the DRSN backbone network, making it suitable for handling situations with strong noise interference. This is thanks to the threshold function in DRSN, which enables adaptive denoising, thus better extracting information from noisy spectra. Comparative analysis shows that the proposed GSF has better denoising performance than other denoising methods. In terms of concentration prediction, this invention, through transfer learning, not only accelerates the training convergence speed but also improves the accuracy and reliability of retrieving methane concentration from absorption spectra.

[0050] In this embodiment, the real methane absorption spectrum is the absorption spectrum data obtained based on real methane samples, and is therefore the real methane absorption spectrum. The simulated methane transmission spectrum is the transmission spectrum data obtained through database simulation, and is therefore the simulated methane transmission spectrum. The simulated methane transmission spectrum data is obtained by adding Gaussian noise to the simulation data to simulate the real transmission spectrum data, and is therefore the simulated methane transmission spectrum data. The actual methane transmission spectrum data is the data obtained from methane samples in the actual measurement scenario after the methane concentration measurement model has been trained and applied to the actual measurement scenario, and is therefore the actual methane transmission spectrum data.

[0051] Optionally, the step of collecting N groups of methane samples with different concentrations to obtain the corresponding true methane absorption spectra includes:

[0052] A direct absorption spectroscopy experimental setup was constructed. N groups of methane samples with different concentrations were prepared using a methane gas mixer in the direct absorption spectroscopy experimental setup. The methane absorption spectrum corresponding to each group of methane samples was collected as the true methane absorption spectrum, with the preset concentration of the methane samples prepared by the methane gas mixer as the label. Here, N is a positive integer greater than 1.

[0053] In the above embodiments, using a constructed DAS experimental setup at room temperature (25°C) and atmospheric pressure (1 atm), 40 sets of methane concentrations ranging from 0 to 100 ppm were configured using a methane gas mixer within the setup. Using the preset concentration of the gas mixer as a label, 40 sets of methane absorption spectra were collected. Baseline removal was performed on the collected spectra to obtain the corresponding transmission spectrum data. These 40 sets of data were used as the test set for the model. Using a neural network model to predict concentrations from spectral data requires training with a large amount of data. Training with experimental data is extremely resource-intensive; therefore, a simulation dataset needs to be constructed.

[0054] The specific data in the embodiments are only examples and are not limited. Any data size that conforms to the requirements of this method flow is applicable to the steps of this method.

[0055] Optionally, generating M simulated methane transmission spectra at different concentrations through simulation includes:

[0056] The methane transmission spectrum under experimental conditions was simulated using a spectral database, generating M methane transmission spectrum simulation data of different concentrations. These simulation data were then used as the simulated methane transmission spectrum, where M is a positive integer greater than 1.

[0057] Optionally, adding Gaussian noise to the simulated methane transmission spectrum to obtain simulated methane transmission spectrum data includes:

[0058] Noise parameters were obtained by performing noise analysis on the direct absorption spectrum of methane under real-world conditions, and the mean, variance, and standard deviation of the noise parameters were calculated.

[0059] Using the mean and variance as the first labels, Gaussian white noise corresponding to the first label is added to the simulated methane transmission spectrum. At the same time, using the standard deviation as the second label, Gaussian noise corresponding to the second label is added to the simulated methane transmission spectrum. The simulated methane transmission spectrum with added Gaussian white noise and high-speed noise is used as simulated methane transmission spectrum data.

[0060] In the above embodiments, the noise parameters (mean and variance) of DAS spectra under real experimental conditions were analyzed. Methane transmission spectra under experimental conditions were simulated using the HITRAN database. First, 2000 simulated methane transmission spectra with concentrations ranging from 0 to 100 ppm were generated, each consisting of 1130 sampling points. Gaussian white noise with a mean of 0 and a variance of 0.0028 was added to each simulated spectrum to simulate the absorption spectra collected under experimental conditions. Furthermore, to enhance the model's generalization ability, Gaussian noise with standard deviations of 0.0008, 0.0018, 0.0038, and 0.0048 was added to the methane transmission spectra at each concentration level for data augmentation. Therefore, a total of 10,000 sets of methane transmission spectra were generated as a simulated dataset. Simultaneously, the concentration values ​​corresponding to each transmission spectrum constituted a concentration set, serving as the training labels for the model.

[0061] The specific data in the embodiments are only examples and are not limited. Any data size that conforms to the requirements of this method flow is applicable to the steps of this method.

[0062] Optionally, the improvement of the residual shrinkage building blocks in the deep residual shrinkage network model to obtain the improved deep residual shrinkage network model includes:

[0063] A new threshold function is constructed, and this new threshold function replaces the original soft threshold function in the residual shrinkage building unit of the deep residual shrinkage network model, resulting in a new residual shrinkage building unit. The new threshold function satisfies the following relationship:

[0064] ;

[0065] In the formula, Represents the features of the input. Indicates the characteristics of the output. It is a symbolic formula function. This represents the threshold obtained through the above network threshold calculation. Represents an arbitrary constant;

[0066] The deep residual shrinkage network model containing new residual shrinkage building blocks is used as the improved deep residual shrinkage network model.

[0067] In the above embodiments, the DRSN is composed of multiple residual shrinkage building blocks (RSBUs) of different scales, thereby enabling the extraction of deep information block by block from noisy transmission spectra; the DRSN is composed of multiple RSBUs, and the structure of the RSBUs is as follows: Figure 2As shown; RSBU uses a feature map of size C×W×1 as the input to the unit; where C and W represent the number of channels and width of the feature map, respectively; the height of the feature map is always 1. In this embodiment, a one-dimensional transmission spectrum signal is used as the input data. The input data first enters two convolutional layers to obtain the corresponding feature map. This feature map is then sent to two paths: one is directly connected to the threshold function, denoted as X, and the other enters a special module, which can be considered the attention mechanism module of RSBU, and is also one of the advantages of DRSN. In the special module, there are also two paths: one is to first calculate the absolute value, and then obtain a feature through global average pooling (GAP), denoted as A; the other feature map after GAP is input into a small fully connected network, and the Sigmoid function is used as the last layer to normalize the output to between 0 and 1, obtaining a coefficient, denoted as . The final threshold can be expressed as: *A, this approach not only ensures that the threshold is positive and not too large, but also allows different samples to have different thresholds. This is why this special module can be regarded as a special attention mechanism; finally, the threshold... * Thresholding is performed on A and X; this invention uses a new threshold function to replace the original RSBU soft threshold function; RSBU contains a special module for estimating the threshold required to implement the new threshold function; its threshold calculation process is as follows: First, a feature map of size C×W×1 is used as the input of the unit; where C and W represent the number of channels and width of the feature map, respectively; the height of the feature map is always 1 because this paper uses a one-dimensional vibration signal as the input data; Second, the input feature map is trained with two convolutional layers (Conv); where BN represents batch normalization and ReLU represents one type of activation function; Third, in the special module, absolute value processing is performed first, and then global mean pooling (GAP) is used to obtain a one-dimensional vector with a length equal to the number of channels. Fourth, this one-dimensional vector After passing through two fully connected (FC) layers, the scaling parameters of the one-dimensional vector will be obtained. Fifth, the sigmoid function is used to scale the parameters. Mapping within the 0-1 range; sixth, the threshold of RSBU. Equals the scaling parameter and the one-dimensional vector The product of the thresholds; after the threshold calculation is completed, the new threshold function will be used to filter information in the multi-channel feature map.

[0068] The new threshold function overcomes the shortcomings of both soft and hard threshold functions. First, it possesses the same continuity as the soft threshold function, without the jump points observed in the hard threshold function. Second, while applying soft and hard threshold functions introduces a constant deviation or adjustment to the overall signal level, the new threshold function, with a suitable N value, exhibits a more consistent and stable performance. As an asymptote, it can eliminate the effects of constant deviation.

[0069] Optionally, the construction of the methane concentration measurement model based on the improved deep residual shrinkage network model includes:

[0070] A gas spectral filter was constructed using an improved deep residual shrinkage network model and a multilayer perceptron, and a gas concentration predictor was constructed using three fully connected layers.

[0071] A methane concentration measurement model is constructed using the gas spectral filter and the gas concentration predictor.

[0072] In the above embodiments, an end-to-end methane concentration measurement model is constructed based on an improved DRSN and a multilayer sensing mechanism. The model framework is as follows: Figure 3 As shown, the entire model consists of two modules: Gas Spectral Filter (GSF) and Gas Concentration Predictor (GCP). The GSF is composed of an improved DRSN and a multilayer perceptron, and the DRSN contains 6 RSBU sub-modules. The GCP is constructed by three fully connected layers. For the GSF, three fully connected layers are added after the GSF to upsample the deep features extracted by the GSF layer by layer, thereby obtaining a pure transmission spectrum.

[0073] Optionally, the training process for the methane concentration measurement model includes:

[0074] S1, Freeze gas concentration predictor, uses training set data to train gas spectral filter to obtain filter model of methane transmission spectrum;

[0075] S2. Load the pre-trained weights of the gas spectral filter and train both the gas spectral filter and the gas concentration predictor simultaneously using the training dataset.

[0076] In the above embodiment, the specific training parameters are as follows: Batchsize=16; Optimizer=Adam; Rate=10−4, learning rate decays by a factor of 10 every 100 epochs; Epochs=300. The model training steps are mainly divided into two parts: First, freeze the GCP part and pre-train the GSF to obtain the filtering model for methane transmission spectra; second, load the pre-trained weights of GSF and train both the GSF and GCP modules simultaneously. The model is validated by loading a test set. The denoising performance of the model is validated using the pre-trained weights of GSF obtained in the first step, and the concentration prediction performance of the model is validated using the training parameters of GSF and GCP obtained in the second step.

[0077] In addition, this embodiment also includes: performance evaluation of the trained methane concentration measurement model, the specific implementation steps of which include:

[0078] Filtering performance analysis was performed on the test set spectral data. The signal-to-noise ratio (SNR) of the noisy spectrum and the spectrum obtained by different filtering methods was calculated. Filtering methods such as SG filtering, wavelet thresholding, and Kalman filtering were selected.

[0079] Concentration prediction analysis was performed on the spectral data of the test set. The correlation coefficient between the predicted concentration and the tag concentration was calculated to evaluate the performance of the concentration prediction. The traditional Vogit nonlinear fitting algorithm was selected for comparison. The larger the correlation coefficient, the more accurate the measurement results of the method.

[0080] Stability analysis was performed on the model. To evaluate the impact of the end-to-end methane concentration measurement model on the stability of the DAS, two MFCs were used to configure a methane gas flow rate of 12 ppm into the absorption cell, and continuous methane measurements were performed for one hour. The half-peak width at half-width (HWHM) was calculated from the statistical histogram of the measured concentration values. The magnitude of the HWHM reflects the accuracy of the methane concentration measurement, and a smaller HWHM is better.

[0081] This application also provides a methane concentration measurement system based on direct absorption spectroscopy, including a processor and a memory;

[0082] Memory, used to store computer programs;

[0083] The processor, when executing a program stored in memory, implements any of the steps described in the method for measuring methane concentration based on direct absorption spectroscopy.

[0084] The methane concentration measurement system based on direct absorption spectroscopy described above can realize all embodiments of the methane concentration measurement method based on direct absorption spectroscopy described above, and can achieve the same beneficial effects. Here, it will not be elaborated further.

[0085] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for measuring methane concentration based on direct absorption spectroscopy, characterized in that, include: N groups of methane samples with different concentrations were collected to obtain the corresponding real methane absorption spectra, and the baseline of the real methane absorption spectra was removed to obtain the corresponding real methane transmission spectrum data, where N is a positive integer greater than 1; M simulated methane transmission spectra with different concentrations were generated by simulation, and Gaussian noise was added to the simulated methane transmission spectra to obtain simulated methane transmission spectrum data, where M is a positive integer greater than 1. A deep residual shrinkage network model was constructed, and the residual shrinkage building units in the deep residual shrinkage network model were improved to obtain an improved deep residual shrinkage network model. A methane concentration measurement model was then constructed based on the improved deep residual shrinkage network model. The simulated methane transmission spectrum data is used as the training set data, and the real methane transmission spectrum data is used as the validation set data. The methane concentration measurement model is trained using the training set data and the validation set data to obtain the trained methane concentration measurement model. Obtain the actual methane transmission spectrum data corresponding to the methane sample in the target scene, and input the actual methane transmission spectrum data into the trained methane concentration measurement model to obtain the methane concentration data corresponding to the target scene; The step of adding Gaussian noise to the simulated methane transmission spectrum to obtain simulated methane transmission spectrum data includes: Noise parameters were obtained by performing noise analysis on the direct absorption spectrum of methane under real-world conditions, and the mean, variance, and standard deviation of the noise parameters were calculated. Using the mean and variance as the first labels, Gaussian white noise corresponding to the first label is added to the simulated methane transmission spectrum. At the same time, using the standard deviation as the second label, Gaussian noise corresponding to the second label is added to the simulated methane transmission spectrum. The simulated methane transmission spectrum with added Gaussian white noise and high-speed noise is used as simulated methane transmission spectrum data.

2. The method for measuring methane concentration based on direct absorption spectroscopy according to claim 1, characterized in that, The acquisition of N groups of methane samples with different concentrations to obtain the corresponding true methane absorption spectra includes: A direct absorption spectroscopy experimental setup was constructed. N groups of methane samples with different concentrations were prepared using a methane gas mixer in the direct absorption spectroscopy experimental setup. The methane absorption spectrum corresponding to each group of methane samples was collected as the true methane absorption spectrum, with the preset concentration of the methane samples prepared by the methane gas mixer as the label. Here, N is a positive integer greater than 1.

3. The method for measuring methane concentration based on direct absorption spectroscopy according to claim 1, characterized in that, The generation of M simulated methane transmission spectra at different concentrations through simulation includes: The methane transmission spectrum under experimental conditions was simulated using a spectral database, generating M methane transmission spectrum simulation data of different concentrations. These simulation data were then used as the simulated methane transmission spectrum, where M is a positive integer greater than 1.

4. The method for measuring methane concentration based on direct absorption spectroscopy according to claim 1, characterized in that, The improved deep residual shrinkage network model is obtained by improving the residual shrinkage building units in the deep residual shrinkage network model, including: A new threshold function is constructed, and this new threshold function replaces the original soft threshold function in the residual shrinkage building unit of the deep residual shrinkage network model, resulting in a new residual shrinkage building unit. The new threshold function satisfies the following relationship: ； In the formula, Represents the features of the input. Indicates the characteristics of the output. It is a symbolic formula function. This represents the threshold obtained through network thresholding. Represents an arbitrary constant; The deep residual shrinkage network model containing new residual shrinkage building blocks is used as the improved deep residual shrinkage network model.

5. The method for measuring methane concentration based on direct absorption spectroscopy according to claim 1, characterized in that, The methane concentration measurement model constructed based on the improved deep residual shrinkage network model includes: A gas spectral filter was constructed using an improved deep residual shrinkage network model and a multilayer perceptron, and a gas concentration predictor was constructed using three fully connected layers. A methane concentration measurement model is constructed using the gas spectral filter and the gas concentration predictor.

6. The method for measuring methane concentration based on direct absorption spectroscopy according to claim 5, characterized in that, The training process for the methane concentration measurement model includes: S1, Freeze gas concentration predictor, uses training set data to train gas spectral filter to obtain filter model of methane transmission spectrum; S2. Load the pre-trained weights of the gas spectral filter and train both the gas spectral filter and the gas concentration predictor simultaneously using the training dataset.

7. A methane concentration measurement system based on direct absorption spectroscopy, characterized in that, Including processor and memory; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the steps of the method described in any one of claims 1-6.

Images