A CO based on TDLAS technology 2 Gas concentration field detection and reconstruction method

Abstract

The present invention belongs to the technical field of optoelectronic detection and gas detection, and specifically relates to a method for detecting and reconstructing the CO2 gas concentration field based on the TDLAS technology, which solves the technical problems of low spectral data dimension, low precision, response lag and low reconstruction resolution of existing related systems. It includes constructing a spatio-temporal discrete sequence of the concentration field, constructing a multi-modal three-dimensional CO2 gas concentration field simulation data set, initializing the hyperparameters of the U-SRResNet network model, training and optimizing the hyperparameters of the U-SRResNet network model, and inputting the original absorbance data of the collected CO2 gas concentration field into the trained U-SRResNet network model; outputting the reconstruction result of the CO2 gas concentration field amplified by N times through the U-SRResNet network model. The present invention meets the requirements of fast and efficient TDLAS detection and reconstruction, and provides technical support for the refined distribution test of the CO2 gas concentration field.

Description

Technical Field

[0001] The present invention relates to the technical field of photoelectric detection and gas detection, and is specifically a CO 2 Gas concentration field detection and reconstruction method. Background Art

[0002] The increase in greenhouse gas concentrations in the atmosphere is the main cause of global warming. Since the 1960s, the World Meteorological Organization and China's Ministry of Ecology and Environment and other departments and institutions have begun to monitor greenhouse gas concentrations in the atmosphere, and gradually formed a monitoring network at different scales, including global, regional, national, and urban. 2 High-precision monitoring of greenhouse gas concentrations such as CO2 emissions will help evaluate and verify the scientific nature of greenhouse gas accounting methods and emission factors, and support the establishment of a greenhouse gas accounting system that conforms to China's actual conditions.

[0003] Laser spectroscopy technology has the characteristics of high sensitivity, good selectivity, and real-time detection. Among them, Fourier transform infrared spectroscopy, non-dispersive infrared spectroscopy, differential absorption spectroscopy and other technologies have been widely used in the field of gas concentration detection. 2 Research and commercial instruments for gas concentration monitoring tend to focus on large-scale space environment monitoring and are not suitable for multi-dimensional analysis of gas concentration in limited space. The laser spectroscopy technology used is mostly based on broadband light sources, and its spectral resolution is relatively low, which limits the detection limit and sensitivity of gas concentration. In addition, due to the slow response speed, it is impossible to achieve pre-emptive predictive diagnosis of long-term and short-term dynamic adjustment of the measured scene, and the reconstruction resolution is low.

[0004] Based on the above, in order to meet the statistical needs of greenhouse gas emissions accounting for large emission sources, a high-sensitivity, high-precision, multi-dimensional gas concentration rapid detection system is urgently needed to achieve real-time and continuous monitoring of carbon emission gas concentrations. Summary of the invention

[0005] In order to overcome the technical defects of low spectral data dimension, low precision, delayed response and low reconstruction resolution of the existing gas concentration rapid detection system, the present invention provides a CO 2 Gas concentration field detection and reconstruction method.

[0006] The present invention provides a CO based on TDLAS technology 2 The gas concentration field detection and reconstruction method includes the following steps:

[0007] S1. Constructing the spatial and temporal discrete sequence of concentration field;

[0008] First, build CO 2A gas three-dimensional concentration field detection system, comprising a main control host computer, a data acquisition module, a signal generation module, a laser driver, a first adjustable semiconductor laser, a second adjustable semiconductor laser, a first laser collimator, a second laser collimator, a first beam expander, a second beam expander, a first photoelectric array detection module and a second photoelectric array detection module;

[0009] Secondly, a U-shaped residual 3D super-resolution imaging network model, namely the U-SRResNet network model, is constructed by combining a U-shaped deep convolutional neural network and a residual network. The U-SRResNet network model is used to extract deep features from limited spectral data.

[0010] For different areas to be tested, the detection data of discrete slices of the concentration field in the area to be tested is used as the input of the U-SRResNet network model, where the first photoelectric array detection module and the second photoelectric array detection module are arranged to form the area to be tested, and the area to be tested is divided into grid slices with M×N resolution, and M×N is defined to have a total of I grids. One of the detection lasers is analyzed, l j,i represents the path that the j-th laser beam passes through in the i-th grid; L represents the path length; A i Represents the laser path l in the i-th grid j,i The integrated absorbance; α i represents the absorbance in the i-th grid; P is the ambient pressure of the area to be measured; assuming that the temperature distribution of the gas to be measured in each grid is uniform, when the laser beam passes through the area to be measured, the i-th grid is about the path l of the laser. j,i The integrated absorbance A i It is expressed as:

[0011] A i = ∫ 0 L P ⋅ S [ T ( L )] ⋅ C ( L ) dl = ∫ o L α i ( L ) dL ,

[0012] Where C(L) is the molar concentration of the absorption component corresponding to path l, S [ T ( L )] Indicates the intensity of the gas absorption spectrum at temperature T;

[0013] After discrete gridding, the integrated absorbance A on all paths in the test area i The sum of the contributions of ,but It is expressed as:

[0014] ;

[0015] In order to facilitate computer solution, the integral absorbance calculation formula of the test area after discrete gridding is converted into a matrix form to represent the light absorption coefficient on all paths in the entire test area, that is:

[0016] ;

[0017] Where F is the matrix coefficient, which is obtained by CO in step S1 2 Determined by the gas three-dimensional concentration field detection system;

[0018] S2. Building a multimodal CO 2 Gas three-dimensional concentration field simulation data set;

[0019] According to the different gas concentration distribution states in the tested space, CO 2 Gas concentration diffusion distribution mode, establish a certain amount of samples, each sample contains a high-resolution concentration distribution C HR and the low-resolution absorbance matrix A data , the samples are divided into training set, validation set and test set, which together constitute the multimodal CO 2 Gas three-dimensional concentration field simulation data set;

[0020] S3, hyperparameter initialization of U-SRResNet network model;

[0021] S4. Training and optimization of hyperparameters of U-SRResNet network model;

[0022] S5, the CO 2 CO collected by the three-dimensional gas concentration field detection system 2 The original absorbance data of the gas concentration field is input into the trained U-SRResNet network model;

[0023] S6, the final output is amplified N times by the U-SRResNet network model 2 Gas concentration field reconstruction results.

[0024] Preferably, in step S1, CO 2The main control host computer of the gas three-dimensional concentration field detection system is connected to the signal generating module, the main control host computer is used to control the signal generating module to generate a sawtooth wave signal, the signal generating module is connected to the laser driver, the signal generating module controls the laser driver to output a corresponding signal, the laser driver is respectively connected to the first adjustable semiconductor laser and the second adjustable semiconductor laser, the first adjustable semiconductor laser is connected to the first laser collimator and the first beam expander in turn, the second adjustable semiconductor laser is connected to the second laser collimator and the second beam expander in turn, the laser driver controls the temperature and current to make the first adjustable semiconductor laser and the second adjustable semiconductor laser output laser light of a specific wavelength The first laser collimator and the first beam expander are used to ensure that the laser beam of the first coordinated semiconductor laser passes through the gas flow field to be measured and is incident on the first photoelectric array detection module. The second laser collimator and the second beam expander are used to ensure that the laser beam of the second coordinated semiconductor laser passes through the gas flow field to be measured and is incident on the second photoelectric array detection module. The first photoelectric array detection module and the second photoelectric array detection module are respectively located on two mutually perpendicular planes and are both connected to the data acquisition module. The data acquisition module is connected to the main control host computer. The first photoelectric array detection module and the second photoelectric array detection module are used to convert optical signals into electrical signals. The data acquisition module is used to collect electrical signals and calculate CO 2 Gas concentration information, and CO 2 The gas concentration information is sent back to the main control computer, and the main control computer receives the CO 2 Gas concentration information calculation and reconstruction of CO 2 Gas concentration field.

[0025] Preferably, step S4 comprises: 2 The gas concentration super-resolution reconstruction problem is formulated as a generation problem, that is, using low-resolution images to generate high-resolution images. 2 High-resolution concentration distribution C in the three-dimensional gas concentration field simulation data set HR Set as the identification label and convert the low-resolution absorbance matrix A data Set as input data for the U-SRResNet network model;

[0026] Among them, the objective function of super-resolution reconstruction is:

[0027] ,

[0028] In the formula, is the objective function with respect to the current model parameters The derivative of is the gradient of parameter θ; let , Represent the low-resolution and high-resolution absorbance two-dimensional projection distribution data, respectively, where That is the low-resolution absorbance matrix A data , That is the high-resolution concentration distribution C HR , I SR represents the CO after super-resolution reconstruction 2 Gas concentration absorbance projection matrix;

[0029] By changing the number of layers of the U-shaped deep convolutional neural network, the magnification N of the U-SRResNet network model is changed. A , N l is the number of symmetric grid layers;

[0030] ;

[0031] The U-shaped deep convolutional neural network is divided into an input layer, a hidden layer, and an output layer. The input layer is determined by the data sample and the input layer is a low-resolution absorbance matrix A. data , design hidden layers and output layers through super-resolution reconstruction multiples;

[0032] Adaptively adjust the learning rate, update the learning rate in exponential decay mode, and train the U-SRResNet network model for super-resolution reconstruction performance:

[0033] ,

[0034] Where lr(0) and lr(n) are the learning rates of the 0th and nth generations respectively. ; is the global step size;

[0035] The MSE loss is selected as the loss function to train and optimize the U-SRResNet network model. The MSE loss function is calculated. If it is greater than the target value, the next step is performed. If it is less than the target value, the formula (6) is returned to continue training the optimization model. The MSE loss function is:

[0036] I MSE SR = 1 N A 2 ∙W∙H ∑ q=1 N A 2 ∙W ∑ P=1 N A 2 ∙H [ I q,p HR - G θ g I q,p LR ] 2 ,

[0037] represents the mean square error loss function of the super-resolution reconstructed image; W and H represent the width and height of the low-resolution image respectively; Represents the pixel value of the high-resolution image, located in the qth row and pth column; Represents the pixel value of the low-resolution image, located in the qth row and pth column; The generated super-resolution image function is related to its parameters The gradient of

[0038] After the MSE loss calculation process is completed, the super-resolution reconstruction results are evaluated. The evaluation indicators include simulated CO 2 The reconstruction error, determination coefficient R², peak signal-to-noise ratio PSNR and structural similarity index SSIM of the gas concentration field; if the evaluation indicators all meet the preset threshold requirements, the parameter configuration of this round of training is considered to be valid, the corresponding model parameters are stored and fixed, and directly loaded for use; if the preset threshold requirements are not met, return to step S4.

[0039] Compared with the prior art, the technical solution provided by the present invention has the following technical effects: the method of the present invention constructs a CO 2 The gas three-dimensional concentration field detection system can accurately obtain photoelectric information such as laser transmission, molecular absorption energy transition, energy attenuation, etc. in complex gas flow fields, and realize the detection of complex flow field CO 2 Rapid multi-dimensional detection of gas concentration fields solves the problems of slow dynamic response and low dimensionality of spectral data in current detection methods. 2 The gas three-dimensional concentration field detection system is a detection imaging model based on conical optics. 2 The nonlinear tomography model of "gas concentration-spectral intensity" solves the problem that traditional atmospheric monitoring technology is difficult to meet the multi-dimensional detection needs of non-uniform flow fields, and realizes a more accurate mathematical description of the laser transmission process; the U-SRResNet network model constructed in the method of the present invention fully considers the format differences of limited detection projection data obtained in different test environments, sets a size self-correction channel, and can standardize the matrix parameters of the input network, improve the applicability of the model, enhance the anti-interference ability, meet the fast and efficient TDLAS detection reconstruction, solve the problem of low resolution of absorption spectrum detection reconstruction during the test process, and provide a reference for CO 2 Provide technical support for refined distribution testing of gas concentration fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 CO is described in an embodiment of the present invention 2 Schematic diagram of the structure of the gas three-dimensional concentration field detection system;

[0043] Figure 2 is the absorption spectrum of carbon dioxide gas molecules in a certain embodiment of the present invention;

[0044] Figure 3 A CO based on TDLAS technology described in an embodiment of the present invention 2 Flow chart of the gas concentration field detection and reconstruction method;

[0045] Figure 4 Schematic diagram of the structure of the U-SRResNet network model described in an embodiment of the present invention;

[0046] Figure 5 It is a physical model for the spatiotemporal reconstruction of the gas concentration field in a certain embodiment of the present invention;

[0047] Figure 6 This figure shows the influence of the kernel design of the U-SRResNet network model on the training time and the loss function value in a certain embodiment of the present invention.

[0048] In the figure: 1. main control host computer; 2. data acquisition module; 3. first photoelectric array detection module; 4. second photoelectric array detection module; 5. first adjustable semiconductor laser; 6. second adjustable semiconductor laser; 7. laser driver; 8. signal generating module; 9. first laser collimator; 10. first beam expander; 11. second laser collimator; 12. second beam expander. DETAILED DESCRIPTION

[0049] In order to more clearly understand the above-mentioned objectives, features and advantages of the present invention, the scheme of the present invention will be further described below. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.

[0050] In the description, it should be noted that the terms "first" and "second" are only used for descriptive purposes and should not be understood as indicating or implying relative importance. It should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected" and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be internal communication between two components. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0051] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present invention, rather than all of the embodiments.

[0052] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0053] There are many gas-phase molecules in complex carbon emission scenarios. Based on the absorption spectrum characteristics of characteristic molecules in different bands, laser transmission, molecular absorption energy transition and energy attenuation and other spectral absorption mechanisms, a specific laser wavelength is selected as the effective working band for spectrum-concentration coupled forward quantization calculation according to the HITRAN database. According to the HITRAN database, when the temperature is 296K and the air pressure is standard atmospheric pressure, CO 2 At 6280cm -1 To 6375cm -1 Infrared absorption spectrum in the range Figure 2 As shown. It can be seen that CO 2 At 6310cm -1 and 6375cm -1 There are strong absorption lines, and the absorption of water vapor and oxygen is weak in this band, which can effectively avoid interference from high-concentration gas components in the environment, so two absorption spectral lines are selected within this band.

[0054] In one embodiment, Figure 3 As shown, a CO based on TDLAS technology is disclosed. 2 The gas concentration field detection and reconstruction method includes the following steps:

[0055] S1. Constructing the spatial and temporal discrete sequence of concentration field;

[0056] First, build CO 2 Gas three-dimensional concentration field detection system, such as Figure 1As shown, it includes a main control host computer 1, a data acquisition module 2, a signal generating module 8, a laser driver 7, a first adjustable semiconductor laser 5, a second adjustable semiconductor laser 6, a first laser collimator 9, a second laser collimator 11, a first beam expander 10, a second beam expander 12, a first photoelectric array detection module 3 and a second photoelectric array detection module 4. The main control host computer 1 is connected to the signal generating module 8 to control the signal generating module 8 to generate a sawtooth wave signal. The signal generating module 8 is connected to the laser driver 7. The signal generating module 8 controls the laser driver 7 to output a corresponding signal. The laser driver 7 is respectively connected to the first adjustable semiconductor laser 5 and the second adjustable semiconductor laser 6. The first adjustable semiconductor laser 5 is sequentially connected to the first laser collimator 9 and the first beam expander 10. The second adjustable semiconductor laser 6 is sequentially connected to the second laser collimator 11 and the second beam expander 12. The laser driver 7 controls the temperature and current to make the first and second coordinated semiconductor lasers 5 and 6 output laser beams of specific wavelengths. The first laser collimator 9 and the first beam expander 10 are used to ensure that the laser beam of the first coordinated semiconductor laser 5 passes through the gas flow field to be measured and is incident on the first photoelectric array detection module 3. The second laser collimator 11 and the second beam expander 12 are used to ensure that the laser beam of the second coordinated semiconductor laser 6 passes through the gas flow field to be measured and is incident on the second photoelectric array detection module 4. The first photoelectric array detection module 3 and the second photoelectric array detection module 4 are respectively located on two mutually perpendicular planes and are both connected to the data acquisition module 2. The data acquisition module 2 is connected to the main control host computer 1. The first photoelectric array detection module 3 and the second photoelectric array detection module 4 are used to convert optical signals into electrical signals. The data acquisition module 2 is used to collect electrical signals and calculate CO 2 Gas concentration information, and CO 2 The gas concentration information is sent back to the main control host computer 1, and the main control host computer 1 receives the CO 2 Gas concentration information calculation and reconstruction of CO 2 Gas concentration field;

[0057] Secondly, a U-shaped residual 3D super-resolution imaging network model, namely the U-SRResNet network model (U-Super resolution residual net, U-SRResNet), is constructed by combining a U-shaped deep convolutional neural network and a residual network. The U-SRResNet network model is used to extract deep features from limited spectral data; the U-SRResNet network model can improve CO 2 Reconstruction accuracy of gas concentration distribution; In a specific embodiment, the structure diagram of the U-SRResNet network model is as follows Figure 4 As shown, Figure 4In the figure, (a) is the four-fold amplification structure; (b) is the residual and doubling module; (c) is the replication and doubling module;

[0058] For different areas to be tested, the low-resolution absorbance matrix A data As the input of the U-SRResNet network model, the first photoelectric array detection module 3 and the second photoelectric array detection module 4 are arranged to form a test area, and the test area is divided into grid slices with M×N resolution. It is defined that there are I grids in M×N. One of the detection lasers is analyzed. j,i represents the path that the j-th laser beam passes through in the i-th grid; L represents the path length; A i Represents the laser path l in the i-th grid j,i The integrated absorbance; α i represents the absorbance in the i-th grid; P is the ambient pressure of the area to be measured; assuming that the temperature distribution of the gas to be measured in each grid is uniform, when the laser beam passes through the area to be measured, the i-th grid is about the path l of the laser. j,i The integrated absorbance A i It is expressed as:

[0059] A i = ∫ 0 L P ⋅ S [ T ( L )] ⋅ C ( L ) dl = ∫ o L α i ( L ) dL ,

[0060] Where C(L) is the molar concentration of the absorption component corresponding to path l, S [ T ( L )] Indicates the intensity of the gas absorption spectrum at temperature T;

[0061] After discrete gridding, the integrated absorbance A on all paths in the test area i The sum of the contributions of ,but It is expressed as:

[0062] ;

[0063] Its physical model is Figure 5 As shown, in order to facilitate computer solution, the integral absorbance calculation formula of the test area after discrete gridding is converted into a matrix form to represent the light absorption coefficient on all paths in the entire test area, that is:

[0064] ;

[0065] Where F is the matrix coefficient, which is obtained by CO in step S1 2 Determined by the gas three-dimensional concentration field detection system;

[0066] S2. Building a multimodal CO 2 Gas three-dimensional concentration field simulation data set;

[0067] According to the different gas concentration distribution states in the tested space, CO 2 Gas concentration diffusion distribution mode, establish a certain amount of samples, each sample contains a high-resolution concentration distribution C HR and the low-resolution absorbance matrix A data , the samples are divided into training set, validation set and test set, which together constitute the multimodal CO 2 Gas three-dimensional concentration field simulation data set; in a specific embodiment, the number of samples is set to 25,000, of which 15,000 are training sets, 5,000 are validation sets, and 5,000 are test sets;

[0068] S3, hyperparameter initialization of U-SRResNet network model;

[0069] S4. Training and optimization of hyperparameters of U-SRResNet network model;

[0070] CO 2 The gas concentration super-resolution reconstruction problem is formulated as a generation problem, that is, using low-resolution images to generate high-resolution images. 2 High-resolution concentration distribution C in the three-dimensional gas concentration field simulation data set HR Set as the identification label and convert the low-resolution absorbance matrix A data Set as input data for the U-SRResNet network model;

[0071] Among them, the objective function of super-resolution reconstruction is:

[0072] ,

[0073] In the formula, is the objective function with respect to the current model parameters The derivative of is the gradient of parameter θ; let , Represent the low-resolution and high-resolution absorbance two-dimensional projection distribution data, respectively, where That is the low-resolution absorbance matrix A data , That is the high-resolution concentration distribution C HR , I SR represents the CO after super-resolution reconstruction 2 Gas concentration absorbance projection matrix;

[0074] By changing the number of layers of the U-shaped deep convolutional neural network, the magnification N of the U-SRResNet network model is changed. A , N l is the number of symmetric grid layers;

[0075] ;

[0076] The U-shaped deep convolutional neural network is divided into an input layer, a hidden layer, and an output layer. The input layer is determined by the data sample and the input layer is a low-resolution absorbance matrix A. data , design the hidden layer and output layer by super-resolution reconstruction multiples; simulate the relationship between the training time and loss value of the second, third and fourth layers with different kernel designs as shown in Figure 6 As shown in the figure, the convergence behaviors of training loss and time cost of different kernels and layers are different. The optimal network layer and convolution kernel structure are determined according to actual needs to promote faster and more efficient feature extraction and training process;

[0077] Adaptively adjust the learning rate, update the learning rate in exponential decay mode, and train the U-SRResNet network model for super-resolution reconstruction performance:

[0078] ,

[0079] Where lr(0) and lr(n) are the learning rates of the 0th and nth generations respectively. ; is the global step size;

[0080] The MSE loss is selected as the loss function to train and optimize the U-SRResNet network model. The MSE loss function is calculated. If it is greater than the target value, the next step is performed. If it is less than the target value, the formula (6) is returned to continue training the optimization model. The MSE loss function is:

[0081] I MSE SR = 1 N A 2 ∙W∙H ∑ q=1 N A 2 ∙W ∑ P=1 N A 2 ∙H [ I q,p HR - G θ g I q,p LR ] 2 ,

[0082] represents the mean square error loss function of the super-resolution reconstructed image; W and H represent the width and height of the low-resolution image respectively; Represents the pixel value of the high-resolution image, located in the qth row and pth column; Represents the pixel value of the low-resolution image, located in the qth row and pth column; The generated super-resolution image function is related to its parameters The gradient of

[0083] After the MSE loss calculation process is completed, the super-resolution reconstruction results are evaluated. The evaluation indicators include simulated CO 2 The reconstruction error, determination coefficient R², peak signal-to-noise ratio PSNR and structural similarity index SSIM of the gas concentration field; if the evaluation indicators all meet the preset threshold requirements, the parameter configuration of this round of training is considered to be valid, and the corresponding model parameters are stored and fixed, and directly loaded for use; if the preset threshold requirements are not met, return to step S4;

[0084] S5, the CO 2 CO collected by the three-dimensional gas concentration field detection system 2 The original absorbance data of the gas concentration field is input into the trained U-SRResNet network model;

[0085] S6, the final output is amplified N times by the U-SRResNet network model 2 Gas concentration field reconstruction results.

[0086] Specifically, in step S1, the CO 2 In the gas three-dimensional concentration field detection system, the first adjustable semiconductor laser 5 and the second adjustable semiconductor laser 6 can also be replaced by other types of adjustable semiconductor lasers, the beam expander can be replaced by other optical components with similar functions, and the photoelectric array detection module can be replaced by a photosensitive diode array or a photosensitive array detector with similar functions.

[0087] The main control host computer 1 is used to control the CO signal generated by the signal generation module 8 and the data acquisition module 2 to feedback the CO signal. 2 Gas concentration information and calculation to reconstruct CO 2 The signal generating module 8 is used to generate a sawtooth wave signal to control the laser driver 7 to output a corresponding signal; the laser driver 7 controls the temperature and current to make the first adjustable semiconductor laser 5 and the second adjustable semiconductor laser 6 output a laser beam of a specific wavelength; the first laser collimator 9, the second laser collimator 11, the first beam expander 10 and the second beam expander 12 are used to ensure that the laser beam passes through the gas flow field to be measured according to the designed laser path; the data acquisition module 2 is used to collect the electrical signals of the first photoelectric array detection module 3 and the second photoelectric array detection module 4, and feed the electrical signals back to the main control host computer 1 and calculate the CO 2 Gas concentration information.

[0088] The main control host computer 1 controls the signal generation module 8 to generate a sawtooth wave signal, which is input into the laser driver 7. The laser driver 7 controls the temperature and current of the semiconductor laser to output the corresponding signal f(t) and the wavelength of the laser beam. The dual-path laser output after tuning is collimated by the corresponding laser collimator to reduce the defocusing phenomenon of the laser beam. The laser collimator can improve the focusing degree of the laser beam. The collimated laser is diverged by the beam expander in a cone beam shape. After the laser beam passes through the area to be measured, the photoelectric array detection module at the corresponding position receives the attenuated photoelectric signal to achieve CO 2 Dynamic detection of three-dimensional gas concentration information. Data acquisition and sorting are performed through the data acquisition module 2, and finally handed over to the main control host computer 1 for analysis and processing.

[0089] Based on the absorption spectrum characteristics of characteristic molecules in different bands, the present invention selects a specific laser wavelength as the effective working band for spectrum-concentration coupling forward quantization calculation according to the HITRAN database, and proposes a CO detection method that can couple the tunable semiconductor laser absorption spectroscopy technology with the photoelectric array detection module. 2 The gas concentration field detection system can accurately obtain the photoelectric information such as laser transmission, molecular absorption energy transition, energy attenuation, etc. in the complex gas flow field, establish a geometric cone beam optical projection detection model, and realize CO 2 Multi-dimensional detection of gas concentration distribution.

[0090] In view of the limited detection data in complex environments, a U-shaped residual three-dimensional super-resolution imaging network model is constructed based on convolutional neural networks. The influence of relevant parameters on the performance of the three-dimensional super-resolution imaging model is dynamically adjusted, and the imaging network is optimized. The residual network and U-shaped neural network model are used to extract deep features from limited spectral data, reconstruct concentration distribution, and solve CO 2 High-precision three-dimensional super-resolution rapid tomography technology of concentration field.

[0091] The above is only a specific implementation of the present invention, which enables those skilled in the art to understand or implement the present invention. Although detailed descriptions are given with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments, and they should all be covered by the protection scope of the claims.

Claims

1. A CO2 gas concentration field detection and reconstruction method based on TDLAS technology, characterized in that: The steps are: S1. Constructing the spatial and temporal discrete sequence of concentration field; First, a three-dimensional CO2 gas concentration field detection system is constructed, which includes a main control host computer (1), a data acquisition module (2), a signal generation module (8), a laser driver (7), a first tunable semiconductor laser (5), a second tunable semiconductor laser (6), a first laser collimator (9), a second laser collimator (11), a first beam expander (10), a second beam expander (12), a first photoelectric array detection module (3) and a second photoelectric array detection module (4); Secondly, a U-shaped residual 3D super-resolution imaging network model, namely the U-SRResNet network model, is constructed by combining a U-shaped deep convolutional neural network and a residual network. The U-SRResNet network model is used to extract deep features from limited spectral data. For different areas to be tested, the detection data of discrete slices of the concentration field in the area to be tested is used as the input of the U-SRResNet network model, wherein the first photoelectric array detection module (3) and the second photoelectric array detection module (4) are arranged to form the area to be tested, and the area to be tested is divided into grid slices with a resolution of M×N, and a total of I grids are defined in M×N. One of the detection lasers is analyzed, and l j,i represents the path that the j-th laser beam passes through in the i-th grid; L represents the path length; A i Represents the laser path l in the i-th grid j,i The integrated absorbance; α i represents the absorbance in the i-th grid; P is the ambient pressure of the area to be measured; assuming that the temperature distribution of the gas to be measured in each grid is uniform, when the laser beam passes through the area to be measured, the i-th grid is about the path l of the laser. j,i The integrated absorbance A i It is expressed as: ， Where C(L) is the molar concentration of the absorption component corresponding to path l, Indicates the intensity of the gas absorption spectrum at temperature T; After discrete gridding, the integrated absorbance A on all paths in the test area i The sum of the contributions of ,but It is expressed as: ； In order to facilitate computer solution, the integral absorbance calculation formula of the test area after discrete gridding is converted into a matrix form to represent the light absorption coefficient on all paths in the entire test area, that is: ； Wherein F is a matrix coefficient, which is determined by the CO2 gas three-dimensional concentration field detection system in step S1; S2, construct a multi-modal CO2 gas three-dimensional concentration field simulation data set; According to the different gas concentration distribution states in the measured space, the CO2 gas concentration diffusion distribution mode under various emission scenarios is simulated to establish a certain amount of samples, each of which contains a high-resolution concentration distribution C HR and the low-resolution absorbance matrix A data , the samples are divided into training set, validation set and test set, which together constitute a multi-modal CO2 gas three-dimensional concentration field simulation data set; S3, hyperparameter initialization of U-SRResNet network model; S4. Training and optimization of hyperparameters of U-SRResNet network model; S5, inputting the original absorbance data of the CO2 gas concentration field collected by the CO2 gas three-dimensional concentration field detection system into the trained U-SRResNet network model; S6. The final output is the reconstruction result of the CO2 gas concentration field amplified N times by the U-SRResNet network model.

2. According to claim 1, a CO2 gas concentration field detection and reconstruction method based on TDLAS technology is characterized in that: In step S1, a main control host computer (1) of the CO2 gas three-dimensional concentration field detection system is connected to a signal generating module (8), the main control host computer (1) is used to control the signal generating module (8) to generate a sawtooth wave signal, the signal generating module (8) is connected to a laser driver (7), the signal generating module (8) controls the laser driver (7) to output a corresponding signal, the laser driver (7) is respectively connected to a first tunable semiconductor laser (5) and a second tunable semiconductor laser (6), the first tunable semiconductor laser (5) is sequentially connected to a first laser collimator (9) and a first beam expander (10), the second tunable semiconductor laser (6) is sequentially connected to a second laser collimator (11) and a second beam expander (12), the laser driver (7) controls the temperature and current so that the first tunable semiconductor laser (5) and the second tunable semiconductor laser (6) output laser beams of specific wavelengths, the first laser collimator (9) and the first beam expander (10) ) is used to ensure that the laser beam of the first tunable semiconductor laser (5) passes through the gas flow field to be measured and is incident on the first photoelectric array detection module (3); the second laser collimator (11) and the second beam expander (12) are used to ensure that the laser beam of the second tunable semiconductor laser (6) passes through the gas flow field to be measured and is incident on the second photoelectric array detection module (4); the first photoelectric array detection module (3) and the second photoelectric array detection module (4) are respectively located on two mutually perpendicular planes and are both connected to the data acquisition module (2); the data acquisition module (2) is connected to the main control host computer (1); the first photoelectric array detection module (3) and the second photoelectric array detection module (4) are used to convert optical signals into electrical signals; the data acquisition module (2) is used to collect electrical signals and calculate CO2 gas concentration information, and transmit the CO2 gas concentration information back to the main control host computer (1); the main control host computer (1) calculates and reconstructs the CO2 gas concentration field according to the received CO2 gas concentration information.

3. According to claim 1, a CO2 gas concentration field detection and reconstruction method based on TDLAS technology is characterized in that: Step S4 includes: expressing the CO2 gas concentration super-resolution reconstruction problem as a generation problem, that is, using low-resolution images to generate high-resolution images, and converting the high-resolution concentration distribution C in the multimodal CO2 gas three-dimensional concentration field simulation data set into a high-resolution image. HR Set as the identification label and convert the low-resolution absorbance matrix A data Set as input data for the U-SRResNet network model; Among them, the objective function of super-resolution reconstruction is: ， In the formula, is the objective function with respect to the current model parameters The derivative of is the gradient of parameter θ; let , Represent the low-resolution and high-resolution absorbance two-dimensional projection distribution data, respectively, where That is the low-resolution absorbance matrix A data , That is the high-resolution concentration distribution C HR , I SR Represents the CO2 gas concentration absorbance projection matrix after super-resolution reconstruction; By changing the number of layers of the U-shaped deep convolutional neural network, the magnification N of the U-SRResNet network model is changed. A , N l is the number of symmetric grid layers; ； The U-shaped deep convolutional neural network is divided into an input layer, a hidden layer, and an output layer. The input layer is determined by the data sample and the input layer is a low-resolution absorbance matrix A. data , design hidden layers and output layers through super-resolution reconstruction multiples; Adaptively adjust the learning rate, update the learning rate in exponential decay mode, and train the U-SRResNet network model for super-resolution reconstruction performance: ， Where lr(0) and lr(n) are the learning rates of the 0th and nth generations respectively. ; is the global step size; The MSE loss is selected as the loss function to train and optimize the U-SRResNet network model. The MSE loss function is calculated. If it is greater than the target value, the next step is performed. If it is less than the target value, the formula (6) is returned to continue training the optimization model. The MSE loss function is: ， represents the mean square error loss function of the super-resolution reconstructed image; W and H represent the width and height of the low-resolution image respectively; Represents the pixel value of the high-resolution image, located in the qth row and pth column; Represents the pixel value of the low-resolution image, located in the qth row and pth column; The generated super-resolution image function is related to its parameters The gradient of After the MSE loss calculation process is completed, the obtained super-resolution reconstruction results are evaluated. The evaluation indicators include the reconstruction error of the simulated CO2 gas concentration field, the determination coefficient R², the peak signal-to-noise ratio PSNR and the structural similarity index SSIM; if the evaluation indicators all meet the preset threshold requirements, the parameter configuration of this round of training is considered to be valid, and the corresponding model parameters are stored and fixed, and directly loaded for use; if the preset threshold requirements are not met, return to step S4.

Images