Laser gas analysis device

Abstract

The invention provides a laser gas analysis device which comprises a signal acquisition module used for emitting a laser beam, converting a reflected laser beam signal of gas to be analyzed into an initial electric signal and sending the initial electric signal to a signal denoising module. And the signal denoising module is used for carrying out denoising processing on the initial electric signal to obtain a reconstructed light beam signal. And the gas monitoring module is used for inputting the reconstructed light beam signal into a gas component identification model for component identification, and then inputting the reconstructed light beam signal and the reconstructed light beam signal into a gas concentration identification model together to obtain a concentration value of a corresponding gas component. According to the invention, the dynamically changing gas is collected, the collected gas signal is firstly de-noised, and then the gas components and the concentrations thereof are distinguished, so that the different gas components and the corresponding concentrations thereof are accurately distinguished, and an important monitoring and analysis means is provided for the fields of medical care, public safety and the like.

Description

Technical Field

[0001] The invention relates to the field of laser measurement, and in particular to a laser gas analysis device. Background Art

[0002] With the rapid development of global industry, a large amount of gas is discharged into the atmosphere, posing a serious threat to the environment and human health. Gas detection has always been a very important task. It is closely related to human life, environmental changes and industrial production, especially the detection of toxic and harmful gases. Tunable Diode Laser Absorption Spectroscopy (TDLAS) technology plays an important role in the field of gas detection. Compared with traditional methods such as electrochemistry, catalytic combustion and semiconductor methods, it has the advantages of non-contact measurement, high resolution and high sensitivity. However, due to the detection system itself, environmental factors and other noises, the measurement accuracy of the system will be affected, which in turn affects the identification of the gas composition and its concentration. In the actual environment, the gas in dynamic change poses a challenge to the TDLAS technology to accurately identify the gas concentration and composition. Summary of the invention

[0003] In view of this, an object of the present invention is to provide a laser gas analysis device, which can accurately identify different gas components and their corresponding concentrations by first denoising the collected gas signal and then identifying the gas components and their concentrations.

[0004] In order to achieve the above-mentioned object of the invention, the present invention provides a laser gas analysis device, the device comprising:

[0005] The signal acquisition module is used to emit a laser beam, measure the reflected laser beam signal generated after the gas to be analyzed passes through the laser beam, and convert it into an initial electrical signal and send it to the signal denoising module;

[0006] A signal denoising module is used to reconstruct the initial electrical signal through variable mode decomposition to obtain a reconstructed electrical signal, use singular spectrum analysis to decompose the reconstructed electrical signal into a number of subsequences, and use a heuristic algorithm to find the optimal mode number to denoise the subsequences to obtain a reconstructed beam signal;

[0007] The gas monitoring module is used to input the reconstructed beam signal into the gas component recognition model to convert it into a two-dimensional image to identify the gas component, and then input it together with the reconstructed beam signal into the gas concentration recognition model to obtain the concentration value of the corresponding gas component.

[0008] Furthermore, the signal acquisition module includes a light emitting unit and a light receiving unit, the light emitting unit includes a laser for emitting a laser beam, and the light receiving unit includes a detector for receiving the light signal absorbed by the gas and converting the light signal into an initial electrical signal through photoelectric conversion.

[0009] Furthermore, the initial electrical signal is reconstructed by variable mode decomposition to obtain a reconstructed electrical signal, specifically including:

[0010] S11. Decompose the electrical signal x(t) into K frequency components u k (t), expressed as:

[0011]

[0012] Where k = 1, 2, 3…K, the frequency component u k (t) is defined as:

[0013] u k (t) = A k (t)cos(φ k (t))

[0014] Among them, A k (t) is u k The amplitude of (t), φ k (t) is u k (t), then each frequency component has a center frequency ω k :

[0015] ω k =φ′ k (t)

[0016] Construct a cost function consistent with the original frequency:

[0017]

[0018] Among them, {u k} is a set of frequency components, {ω k} is the corresponding center frequency set, is the derivative with respect to time, δ(t) is the instantaneous pulse of the electrical signal, j is the imaginary unit, and ★ is the convolution operation;

[0019] S12. Introduce the Lagrange multiplier λ(t) into the cost function to reconstruct the original electrical signal:

[0020]

[0021] Among them, a is the penalty factor;

[0022] S13, update the frequency component u through an iterative algorithm k(t), center frequency ω k And the Lagrange multiplier λ(t):

[0023]

[0024] Where T is the step size parameter for updating the Lagrange multiplier, and the iterative process is repeated until the frequency component u k (t) and center frequency ω k When the convergence condition is met, the reconstructed electrical signal is output.

[0025] Furthermore, singular spectrum analysis is used to decompose the reconstructed electrical signal into several subsequences in time series, including:

[0026] S21, the reconstructed electrical signal time series is represented as R = {r 1 ,r 2 ,…,r N}, and set the nesting dimension L for it, where N is the length of the time series and the nesting dimension Establish the reconstructed electrical signal time series trajectory matrix:

[0027]

[0028] Where X is the reconstructed electrical signal time series trajectory matrix, L represents the number of columns of the matrix, K represents the number of rows of the matrix, K = N-L+1;

[0029] S22, perform singular value decomposition on the reconstructed electrical signal time series trajectory matrix:

[0030]

[0031] Among them, X′ is the singular value decomposition matrix of X, G i is the i-th component matrix in singular spectrum analysis, λ i is a singular value, Z i is the left singular vector, G i are right singular vectors, To reconstruct the singular spectral density of the electrical signal time series, X T is the transposed matrix;

[0032] S23, using random numbers to transform the singular value decomposition matrix in step S22 to obtain a random decomposition matrix:

[0033] X ′ =X 1 ′+X 2 ′+…+X d '

[0034]

[0035] Among them, X i ′ is the eigenvector after singular value decomposition of the reconstructed electrical signal time series;

[0036] S24. Convert the random decomposition matrix into a time series format to obtain a plurality of subsequences.

[0037] Furthermore, the optimal modal number is found through a heuristic algorithm to denoise several subsequences, including:

[0038] S31, optimizing the number of modes and penalty factors of several subsequences through a heuristic algorithm to obtain the optimal number of modes that converges to the fitness function;

[0039] S32, decomposing a plurality of subsequences according to the optimal modal number to obtain K modal components;

[0040] S33, calculating the variance contribution rate of the modal components, removing the smallest modal component, and performing wavelet threshold denoising on the retained modal components;

[0041] S34, adding the modal components that are retained and subjected to wavelet denoising to obtain a reconstructed beam signal.

[0042] Furthermore, in step S31, the number of modes and penalty factors of several subsequences are optimized by a heuristic algorithm, specifically including:

[0043] S41, initializing the number of modes and penalty factors of several subsequences;

[0044] S42, calculating K envelope entropies corresponding to the starting positions of each modal component, and taking the minimum envelope entropy as the fitness value;

[0045] S43, performing global search and iteration according to the set fitness function through a heuristic algorithm, when the number of iterations reaches the set maximum number of iterations, the fitness value converges to the optimal mode number.

[0046] Furthermore, the reconstructed beam signal is input into the gas composition recognition model to be converted into a two-dimensional image, specifically including:

[0047] S51, decomposing the reconstructed beam signal into a low-frequency approximate signal and a high-frequency detail signal;

[0048] S52, obtaining low-frequency coefficients and high-frequency coefficients from the low-frequency approximate signal and the high-frequency detail signal through wavelet transform, and extracting low-frequency coefficients and high-frequency coefficients at equal intervals respectively;

[0049] S53, filling the low-frequency coefficients and high-frequency coefficients extracted in step S52 into a pixel matrix to form a two-dimensional image;

[0050] S54, repeat steps S51 to S53 until all reconstructed beam signals are converted into two-dimensional images.

[0051] Furthermore, the gas composition is identified based on the two-dimensional image, specifically including:

[0052] S61, decomposing the two-dimensional image into a plurality of pixel blocks, each pixel block containing part of the global information;

[0053] S62, after the pixel block is passed through the convolution layer and the residual module, the high-level features contained therein are gradually extracted;

[0054] S63. Integrate and decode the features extracted in step S62 through a fully connected module, and output the identified gas components.

[0055] Furthermore, the corresponding gas component concentration value is obtained through the gas concentration identification model, specifically including:

[0056] S71, inputting the reconstructed beam signal into the gas concentration recognition model in chronological order to form an input matrix;

[0057] S72, performing linear transformation of the linear layer, nonlinear transformation of the sigmoid function and normalization processing of the softmax function on the input matrix through the weight distribution module to obtain a weight matrix;

[0058] S73, multiplying the input matrix by the weight matrix to obtain a signal matrix with weights;

[0059] S74, input the signal matrix into the feature extraction module, perform feature extraction through the self-attention mechanism, connect the extracted features with the adder to form a residual connection, and obtain a feature matrix;

[0060] S75. Expand the feature matrix and input it into the fully connected module for decoding to obtain the gas concentration value.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] The present invention provides a laser gas analysis device, which converts the reflected laser beam signal of the gas to be analyzed into an initial electrical signal through a signal acquisition module, and sends it to a signal denoising module for denoising. The signal denoising module first reconstructs the initial electrical signal through variable mode decomposition technology to obtain a reconstructed electrical signal with clear resolution and high precision. Secondly, the reconstructed electrical signal is decomposed into a time series and decomposed into several subsequences through singular spectrum analysis, which provides a data basis for the next step of denoising the decomposed sequence in combination with a heuristic algorithm, can minimize the amount of calculation, and is not prone to signal distortion during the denoising process, thereby maintaining the integrity of the signal.

[0063] The optimal number of modes is found through a heuristic algorithm, so that the signal can be denoised more accurately, the quality of the denoised signal can be improved, and a high-precision reconstructed beam signal can be provided for the final step of the gas monitoring module to identify the gas components and their corresponding concentrations. The gas monitoring module inputs the reconstructed beam signal into the gas component identification model for component identification, and then inputs it together with the reconstructed beam signal into the gas concentration identification model to obtain the concentration value of the corresponding gas component. The present invention collects dynamically changing gases, denoises the collected gas signals, and then identifies the gas components and their concentrations, thereby achieving accurate identification of different gas components and their corresponding concentrations, providing important monitoring and analysis methods for fields such as health care and public safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the embodiments of the present application 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, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0065] Figure 1 A schematic diagram of the overall structure of a laser gas analysis device provided by an embodiment of the present invention;

[0066] Figure 2 A schematic diagram of the working process of a signal denoising module of a laser gas analysis device provided by an embodiment of the present invention;

[0067] Figure 3 A schematic diagram of a workflow for reconstructing an initial electrical signal in a signal denoising module provided in an embodiment of the present invention;

[0068] Figure 4 A schematic diagram of the working process of a gas monitoring module of a laser gas analysis device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0069] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0070] Reference Figure 1 , this embodiment provides a laser gas analysis device, the device comprising:

[0071] The signal acquisition module is used to emit a laser beam, measure the reflected laser beam signal generated after the gas to be analyzed passes through the laser beam, and convert it into an initial electrical signal and send it to the signal denoising module.

[0072] The signal acquisition module includes a light emitting unit and a light receiving unit. The light emitting unit includes a laser for emitting a laser beam, and the light receiving unit includes a detector for receiving the light signal absorbed by the gas and converting the light signal into an initial electrical signal through photoelectric conversion.

[0073] The key components of the signal acquisition module in this embodiment include a light emitting unit and a light receiving unit. The core of the light emitting unit is a high-performance laser that can emit a laser beam of a specific wavelength to penetrate the gas to be measured. In this embodiment, a wavelength-adjustable semiconductor laser is selected as the light source of the light emitting unit. The laser is driven by a control system, and the control system can accurately control the wavelength and power of the laser to adapt to different measurement requirements and environmental conditions. The control system is a prior art and will not be explained in detail in this embodiment. The wavelength range of the laser beam emitted by the laser covers the main absorption spectrum of the gas to be measured, ensuring that the laser beam can effectively interact with the gas molecules. The light receiving unit includes a highly sensitive detector that has a high responsiveness to absorption light signals of a specific wavelength and can receive the laser beam after gas absorption to ensure accurate signal acquisition. The received light signal is then converted into an electrical signal through a photoelectric converter, providing a data basis for the subsequent accurate measurement of gas components and their corresponding concentrations.

[0074] Reference Figure 2 The signal denoising module is used to reconstruct the initial electrical signal through variable mode decomposition to obtain a reconstructed electrical signal, and use singular spectrum analysis to decompose the reconstructed electrical signal into several subsequences in time series. The optimal mode number is found through a heuristic algorithm to denoise several subsequences to obtain a reconstructed beam signal.

[0075] As a preferred embodiment, the signal denoising module first reconstructs the initial electrical signal through the variable mode decomposition technology to obtain a reconstructed electrical signal with clear resolution and high precision. The reconstructed electrical signal can improve the quality of the original signal, and can remove redundant or irrelevant information in the original signal, so that the subsequent time series decomposition is more efficient and reduces the consumption of computing resources, including but not limited to correcting possible distortion, compensating for signal attenuation, calibrating sensor deviation, etc., to ensure that the collected signal has a higher signal-to-noise ratio and more accurate amplitude when the time series is decomposed. Secondly, the reconstructed electrical signal is decomposed into a number of subsequences through singular spectrum analysis, which provides a data basis for the next step of combining the heuristic algorithm to denoise the decomposed sequence, which can minimize the amount of calculation, and is not easy to produce signal distortion during the denoising process, maintaining the integrity of the signal. Finally, the optimal number of modes can be determined by the heuristic algorithm, so that the signal can be denoised more accurately, the quality of the denoised signal can be improved, and the high-precision reconstructed beam signal can be provided for the last step of the gas monitoring module to identify the gas composition and its corresponding concentration.

[0076] Reference Figure 3 , reconstructing the initial electrical signal through variable mode decomposition to obtain a reconstructed electrical signal, specifically including:

[0077] S11. Decompose the electrical signal x(t) into K frequency components u k (t), expressed as:

[0078]

[0079] Where k = 1, 2, 3…K, the frequency component u k (t) is defined as:

[0080] u k (t) = A k (t)cos(φ k (t))

[0081] Among them, A k (t) is u k The amplitude of (t), φ k (t) is u k (t), then each frequency component has a center frequency ω k :

[0082] ω k =φ′ k (t)

[0083] Construct a cost function consistent with the original frequency:

[0084]

[0085] Among them, {u k} is a set of frequency components, {ω k} is the corresponding center frequency set, is the time derivative, δ(t) is the instantaneous pulse of the electrical signal, j is the imaginary unit, and ★ is the convolution operation.

[0086] S12. Introduce the Lagrange multiplier λ(t) into the cost function to reconstruct the original electrical signal:

[0087]

[0088] Among them, a is the penalty factor.

[0089] S13, update the frequency component u through an iterative algorithm k (t), center frequency ω k And the Lagrange multiplier λ(t):

[0090]

[0091] Where T is the step size parameter for updating the Lagrange multiplier, and the iterative process is repeated until the frequency component u k (t) and center frequency ω k When the convergence condition is met, the reconstructed electrical signal is output.

[0092] In this embodiment, variational mode decomposition (VMD) is used to decompose the signal into a series of intrinsic mode functions (IMFs), each of which has a fixed center frequency, which is adaptively determined during the decomposition process. Its purpose is to regard the decomposition of the signal as a variational problem, that is, an optimization problem, and use a variational method to minimize the bandwidth of each mode, thereby achieving effective decomposition of the signal and overcoming the limitations of traditional empirical mode decomposition in processing signals, such as modal aliasing and boundary effects. The cost function takes into account the bandwidth of each IMF and the fidelity between the reconstructed signal and the original signal. In order to constrain the difference between the reconstructed signal and the original signal, Lagrange multipliers are introduced to ensure that the decomposed IMFs can accurately reconstruct the original signal. Finally, an iterative algorithm is used to solve the optimization problem. In each iteration, the estimated values, center frequencies, and Lagrange multipliers of the IMFs are updated, and the iterative process is repeated until u k (t) and ω k When the convergence condition is met, a set of band-limited IMFs is finally output, that is, each IMF has its own center frequency, and these IMFs jointly reconstruct the original signal.

[0093] Singular spectrum analysis is used to decompose the reconstructed electrical signal into several subsequences in time series, including:

[0094] S21, the reconstructed electrical signal time series is represented as R = {r 1,r 2 ,…,r N}, and set the nesting dimension L for it, where N is the length of the time series and the nesting dimension Establish the reconstructed electrical signal time series trajectory matrix:

[0095]

[0096] Wherein, X is the reconstructed electrical signal time series trajectory matrix, L represents the number of columns of the matrix, K represents the number of rows of the matrix, and K=N-L+1.

[0097] S22, perform singular value decomposition on the reconstructed electrical signal time series trajectory matrix:

[0098]

[0099] Among them, X′ is the singular value decomposition matrix of X, G i is the i-th component matrix in singular spectrum analysis, λ i is a singular value, Z i is the left singular vector, G i are right singular vectors, To reconstruct the singular spectral density of the electrical signal time series, X T is the transposed matrix.

[0100] S23, using random numbers to transform the singular value decomposition matrix in step S22 to obtain a random decomposition matrix:

[0101] X ′ =X 1 ′+X 2 ′+…+X d '

[0102]

[0103] Among them, X i ′ is the eigenvector after singular value decomposition of the reconstructed electrical signal time series.

[0104] S24. Convert the random decomposition matrix into a time series format to obtain a plurality of subsequences.

[0105] As a preferred embodiment, singular spectrum analysis (SSA) is a method for processing nonlinear time series data. By decomposing and reconstructing the trajectory matrix of the time series to be studied, different component sequences in the time series are extracted. In this embodiment, the singular spectrum analysis method is used to decompose and reconstruct the reconstructed electrical signal time series to obtain the characteristic components of its gas components and their corresponding concentrations. First, it is decomposed to establish the time series trajectory matrix of the reconstructed electrical signal. By mapping the one-dimensional time series data to a two-dimensional trajectory matrix, the data is structured so that the eigenvalues ​​and eigenvectors of the matrix can be obtained to identify the trend and periodicity in the time series, and these features can be used for further analysis, which provides convenience for subsequent processing and analysis. By converting the singular value decomposition matrix with random numbers, the time series data is randomized, which helps to eliminate certain possible patterns or deviations while maintaining the original statistical characteristics of the reconstructed electrical signal, and improve the flexibility and effectiveness of subsequent analysis. Finally, it is converted into several subsequences with a time series format to remove the influence of its nonlinear characteristics, provide high-quality data for subsequent denoising processing, and improve the accuracy of subsequent gas prediction.

[0106] The optimal modal number is found through a heuristic algorithm to denoise several subsequences, including:

[0107] S31. Optimize the number of modes and penalty factors of several subsequences through a heuristic algorithm to obtain the optimal number of modes that converges to the fitness function.

[0108] S32. Decompose several subsequences according to the optimal modal number to obtain K modal components.

[0109] S33, calculating the variance contribution rate of the modal components, removing the smallest modal component, and performing wavelet threshold denoising on the retained modal components.

[0110] S34, adding the modal components that are retained and subjected to wavelet denoising to obtain a reconstructed beam signal.

[0111] As a preferred embodiment, in order to achieve more accurate and precise denoising of several subsequences, the optimal number of modes is determined by a heuristic algorithm, which can narrow the search range of parameters, thereby simplifying the calculation process and reducing code redundancy, thereby achieving more accurate denoising of the signal, improving the quality of the denoised signal, and providing a high-precision reconstructed beam signal for the final step of the gas monitoring module to identify the gas components and their corresponding concentrations. First, the number of modes and penalty factors of several subsequences are optimized by a heuristic algorithm to achieve more accurate signal decomposition and then combine wavelet denoising technology to achieve a more efficient denoising effect. Secondly, the modes obtained based on the optimized parameter decomposition are screened by using the variance contribution rate, and the screened modal components are subjected to wavelet denoising, and finally the denoised reconstructed beam signal is obtained.

[0112] In step S31, the number of modes and penalty factors of several subsequences are optimized by a heuristic algorithm, specifically including:

[0113] S41, initializing the number of modes and penalty factors of several subsequences;

[0114] S42, calculating K envelope entropies corresponding to the starting positions of each modal component, and taking the minimum envelope entropy as the fitness value;

[0115] S43, performing global search and iteration according to the set fitness function through a heuristic algorithm, when the number of iterations reaches the set maximum number of iterations, the fitness value converges to the optimal mode number.

[0116] As a preferred embodiment, the value of the modal number directly determines the number of modal components after decomposition. If the modal number is set too large, it will cause under-decomposition and produce false modes; if the modal number is set too small, the implicit features of the time series cannot be fully extracted. The penalty factor affects the bandwidth of the modal component. If the modal number is set to a small value, modal aliasing will occur and affect feature extraction; if the modal number is set to a large value, although modal aliasing can be avoided, it will cause the problem of local information loss. Envelope entropy can be used as a fitness function to optimize the modal number and penalty factor. By minimizing the envelope entropy, the optimal modal number and penalty factor value can be found, thereby improving the decomposition effect of several subsequences.

[0117] Reference Figure 4 The gas monitoring module is used to input the reconstructed beam signal into the gas component recognition model to convert it into a two-dimensional image to identify the gas component, and then input it together with the reconstructed beam signal into the gas concentration recognition model to obtain the concentration value of the corresponding gas component.

[0118] As a preferred embodiment, in order to achieve accurate identification of gas, a gas component identification model and a gas concentration identification model are established respectively. After the reconstructed beam signal is converted into a two-dimensional image by the gas component identification model, it is input into a neural network to extract the features of the image, and finally the gas component identification result is output. The gas concentration identification model automatically assigns corresponding weights to different gas components, maps the weighted features to the sample label space, and finally decodes them to obtain the gas components contained in the final measured gas and their corresponding concentration values.

[0119] The reconstructed beam signal is input into the gas composition recognition model to convert it into a two-dimensional image, including:

[0120] S51, decomposing the reconstructed beam signal into a low-frequency approximate signal and a high-frequency detail signal.

[0121] S52, obtaining low-frequency coefficients and high-frequency coefficients from the low-frequency approximate signal and the high-frequency detail signal through wavelet transform, and extracting the low-frequency coefficients and high-frequency coefficients at equal intervals.

[0122] S53, filling the low-frequency coefficients and high-frequency coefficients extracted in step S52 into the pixel matrix to form a two-dimensional image.

[0123] S54, repeat steps S51 to S53 until all reconstructed beam signals are converted into two-dimensional images.

[0124] In this embodiment, by converting the reconstructed beam signal into a two-dimensional image, the spectral characteristics of the gas can be more intuitively displayed, which facilitates the gas component identification model to capture the characteristic absorption spectrum of the gas. By extracting low-frequency and high-frequency coefficients, the signal can be effectively compressed. The low-frequency coefficients usually contain the main information of the signal, while the high-frequency coefficients contain detailed information. By selectively retaining certain coefficients, the amount of data can be reduced while maintaining the important characteristics of the signal. After converting the reconstructed beam signal into a two-dimensional image, it is subjected to feature extraction through a neural network to identify each gas component.

[0125] Identify gas components based on two-dimensional images, including:

[0126] S61. Decompose the two-dimensional image into a number of pixel blocks, each pixel block containing part of the global information.

[0127] S62, after the pixel block is passed through the convolution layer and the residual module to gradually extract the high-level features it contains.

[0128] S63. Integrate and decode the features extracted in step S62 through a fully connected module, and output the identified gas components.

[0129] As a preferred embodiment, when a two-dimensional image is divided into several pixel blocks, each pixel block can represent or reflect the information and features of the entire image to a certain extent, so that when analyzing these pixel blocks, the understanding of the important features of the entire image will not be lost. Since each pixel block contains global information and has a global perspective, the accuracy of the analysis is improved. Even if part of the data is lost or damaged, the remaining data can still provide useful information about the original image. The convolution layer can gradually extract high-level features containing rich information, the residual module can obtain deeper features and solve the problem of network degradation, and the fully connected module can aggregate the features extracted by the network and decode and output the final gas component recognition results.

[0130] The corresponding gas component concentration value is obtained through the gas concentration identification model, including:

[0131] S71. Input the reconstructed light beam signal into the gas concentration recognition model in chronological order to form an input matrix.

[0132] S72, the input matrix is ​​subjected to linear transformation of the linear layer, nonlinear transformation of the sigmoid function and normalization processing of the softmax function through the weight distribution module to obtain a weight matrix.

[0133] S73. Multiply the input matrix by the weight matrix to obtain a signal matrix with weights.

[0134] S74. Input the signal matrix into the feature extraction module, perform feature extraction through the self-attention mechanism, connect the extracted features with the adder to form a residual connection, and obtain the feature matrix.

[0135] S75. Expand the feature matrix and input it into the fully connected module for decoding to obtain the gas concentration value.

[0136] As a preferred embodiment, the reconstructed beam signal is organized into an input matrix in chronological order to provide continuous observation data for the model. Each row of the input matrix represents the beam signal data at a time point, including but not limited to spectral intensity, wavelength and other information. Different gases have different components and their corresponding concentrations. In order to improve the gas concentration estimation accuracy of the model, a weight distribution module is used to enable the model to automatically assign different weights to different gas components and their corresponding concentrations according to the spectral waveform of the gas, reflecting the importance of each input feature.

[0137] The input matrix is ​​preliminarily processed through the weight assignment module to extract useful features and assign weights to them. The linear transformation uses a fully connected layer to transform the input data and extract features; the Sigmoid activation function is applied to introduce nonlinearity so that the model can capture complex relationships; the Softmax function is used for normalization to convert the output into a probability distribution. The input matrix is ​​multiplied by the weight matrix to obtain the weighted sum of each input feature to reflect the contribution of different features to the final result.

[0138] The features of the sequence are extracted through the self-attention mechanism, so that the model can consider the information of the entire input sequence to capture long-term dependencies when processing data of one time step. The extracted features are added to the original input to form a residual connection to solve the gradient vanishing problem in the deep network and improve the learning ability and stability of the model. The feature matrix is ​​expanded into a one-dimensional vector and input into the fully connected module to map the extracted weighted features with the gas components identified by the gas composition recognition model, and decode to obtain the predicted value of the gas concentration.

[0139] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A laser gas analysis device, characterized in that: The device comprises: The signal acquisition module is used to emit a laser beam, measure the reflected laser beam signal generated after the gas to be analyzed passes through the laser beam, and convert it into an initial electrical signal and send it to the signal denoising module; A signal denoising module is used to reconstruct the initial electrical signal through variable mode decomposition to obtain a reconstructed electrical signal, use singular spectrum analysis to decompose the reconstructed electrical signal into a number of subsequences, and use a heuristic algorithm to find the optimal mode number to denoise the subsequences to obtain a reconstructed beam signal; The gas monitoring module is used to input the reconstructed beam signal into the gas component recognition model to convert it into a two-dimensional image to identify the gas component, and then input it together with the reconstructed beam signal into the gas concentration recognition model to obtain the concentration value of the corresponding gas component.

2. The laser gas analysis device according to claim 1, characterized in that: The signal acquisition module includes a light emitting unit and a light receiving unit. The light emitting unit includes a laser for emitting a laser beam, and the light receiving unit includes a detector for receiving the light signal absorbed by the gas and converting the light signal into an initial electrical signal through photoelectric conversion.

3. The laser gas analysis device according to claim 1, characterized in that: The initial electrical signal is reconstructed by variable mode decomposition to obtain a reconstructed electrical signal, specifically including: S11. Decompose the electrical signal x(t) into K frequency components u k (t), expressed as: Where k = 1, 2, 3…K, the frequency component u k (t) is defined as: u k (t)=A k (t)cos(φ k (t)) Among them, A k (t) is u k The amplitude of (t), φ k (t) is u k (t), then each frequency component has a center frequency ω k : oh k =φ′ k (t) Construct a cost function consistent with the original frequency: Among them, {u k } is a set of frequency components, {ω k } is the corresponding center frequency set, is the derivative with respect to time, δ(t) is the instantaneous pulse of the electrical signal, j is the imaginary unit, and * is the convolution operation; S12. Introduce the Lagrange multiplier λ(t) into the cost function to reconstruct the original electrical signal: Among them, a is the penalty factor; S13, update the frequency component u through an iterative algorithm k (t), center frequency ω k And the Lagrange multiplier λ(t): Where T is the step size parameter for updating the Lagrange multiplier, and the iterative process is repeated until the frequency component u k (t) and center frequency ω k When the convergence condition is met, the reconstructed electrical signal is output.

4. The laser gas analysis device according to claim 3, characterized in that: Singular spectrum analysis is used to decompose the reconstructed electrical signal into several subsequences in time series, including: S21, the reconstructed electrical signal time series is expressed as R = {r1, r2, ..., r N }, and set the nesting dimension L for it, where N is the length of the time series and the nesting dimension Establish the reconstructed electrical signal time series trajectory matrix: Where X is the reconstructed electrical signal time series trajectory matrix, L represents the number of columns of the matrix, K represents the number of rows of the matrix, K = N-L+1; S22, perform singular value decomposition on the reconstructed electrical signal time series trajectory matrix: Among them, X′ is the singular value decomposition matrix of X, G i is the i-th component matrix in singular spectrum analysis, λ i is a singular value, Z i is the left singular vector, G i are right singular vectors, To reconstruct the singular spectral density of the electrical signal time series, X T is the transposed matrix; S23, using random numbers to transform the singular value decomposition matrix in step S22 to obtain a random decomposition matrix: X ′ =X1′+X2′+…+X d ′ Among them, X i ′ is the eigenvector after singular value decomposition of the reconstructed electrical signal time series; S24. Convert the random decomposition matrix into a time series format to obtain a plurality of subsequences.

5. The laser gas analysis device according to claim 4, characterized in that: The optimal modal number is found through a heuristic algorithm to denoise several subsequences, including: S31, optimizing the number of modes and penalty factors of several subsequences through a heuristic algorithm to obtain the optimal number of modes that converges to the fitness function; S32, decomposing a plurality of subsequences according to the optimal modal number to obtain K modal components; S33, calculating the variance contribution rate of the modal components, removing the smallest modal component, and performing wavelet threshold denoising on the retained modal components; S34, adding the modal components that are retained and subjected to wavelet denoising to obtain a reconstructed beam signal.

6. The laser gas analysis device according to claim 5, characterized in that: In step S31, the number of modes and penalty factors of several subsequences are optimized by a heuristic algorithm, specifically including: S41, initializing the number of modes and penalty factors of several subsequences; S42, calculating K envelope entropies corresponding to the starting positions of each modal component, and taking the minimum envelope entropy as the fitness value; S43, performing global search and iteration according to the set fitness function through a heuristic algorithm, when the number of iterations reaches the set maximum number of iterations, the fitness value converges to the optimal mode number.

7. The laser gas analysis device according to claim 1, characterized in that: The reconstructed beam signal is input into the gas composition recognition model to convert it into a two-dimensional image, including: S51, decomposing the reconstructed beam signal into a low-frequency approximate signal and a high-frequency detail signal; S52, obtaining low-frequency coefficients and high-frequency coefficients from the low-frequency approximate signal and the high-frequency detail signal through wavelet transform, and extracting low-frequency coefficients and high-frequency coefficients at equal intervals respectively; S53, filling the low-frequency coefficients and high-frequency coefficients extracted in step S52 into a pixel matrix to form a two-dimensional image; S54, repeat steps S51 to S53 until all reconstructed beam signals are converted into two-dimensional images.

8. The laser gas analysis device according to claim 7, characterized in that: Identify gas components based on two-dimensional images, including: S61, decomposing the two-dimensional image into a plurality of pixel blocks, each pixel block containing part of the global information; S62, after the pixel block is passed through the convolution layer and the residual module, the high-level features contained therein are gradually extracted; S63. Integrate and decode the features extracted in step S62 through a fully connected module, and output the identified gas components.

9. The laser gas analysis device according to claim 1, characterized in that: The corresponding gas component concentration value is obtained through the gas concentration identification model, including: S71, inputting the reconstructed beam signal into the gas concentration recognition model in chronological order to form an input matrix; S72, performing linear transformation of the linear layer, nonlinear transformation of the sigmoid function, and normalization processing of the softmax function on the input matrix through a weight distribution module to obtain a weight matrix; S73, multiplying the input matrix by the weight matrix to obtain a signal matrix with weights; S74, input the signal matrix into the feature extraction module, perform feature extraction through the self-attention mechanism, connect the extracted features with the adder to form a residual connection, and obtain a feature matrix; S75. Expand the feature matrix and input it into the fully connected module for decoding to obtain the gas concentration value.

Images