Method for correcting laser gas concentration sensor

Abstract

The invention discloses a method for correcting a laser gas concentration sensor, which is used for measuring gas concentration by using laser with different wavelengths and correcting by combining environmental parameters. The method comprises the following steps: extracting laser absorption characteristics by using a first feedforward neural network, extracting environmental matrix characteristics by using a convolutional neural network, fusing the characteristics through an attention mechanism, and predicting the gas concentration through a second feedforward neural network. The loss function not only considers the difference between the predicted value and the true value, but also comprises the physical constraint between the light intensity value and the theoretical light intensity value. According to the method, the measurement accuracy and the model robustness under different environmental conditions are improved, and the influence of measurement errors and environmental fluctuation is reduced through the combination of multi-wavelength measurement and physical knowledge.

Description

Technical Field

[0001] The invention relates to a laser gas concentration sensor correction method, belonging to the technical field of gas concentration detection. Background Art

[0002] The principle of the laser gas concentration sensor is: the laser gas concentration sensor passes through the gas to be measured through the laser light intensity determined by the light intensity value, and then measures the laser light intensity after passing through the gas to determine the concentration of the gas. This sensor uses the difference in the absorption of laser light in gases of different concentrations to obtain the concentration value of the gas. It has the characteristics of fast response and high precision, and is widely used in industrial monitoring and environmental protection.

[0003] However, the problem with existing laser gas concentration sensors is that they only consider the measurement results of gas in a specific environment, such as temperature, humidity and pressure. Once the environmental parameters change, the measurement results will also be erroneous. Therefore, the existing technology has the problem of limited measurement environment and small scope of use. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a laser gas concentration sensor correction method to overcome the deficiencies of the prior art.

[0005] The technical solution of the present invention is: a laser gas concentration sensor correction method, the method comprising:

[0006] Gas concentration measurements using spectra at different wavelengths;

[0007] Extracting laser absorption vector features through a first feedforward neural network, wherein the laser absorption vector is a vector composed of light intensity values ​​of lasers of different wavelengths after passing through the gas;

[0008] Extracting a feature map of an environmental matrix by a convolutional neural network, wherein the environmental matrix is ​​a matrix intercepted by a sliding window at the end of a temperature time series, a pressure time series, and a humidity time series;

[0009] The weights of the laser absorption vector features and the feature map of the environment matrix are assigned through the attention mechanism and the weighted sum is performed to obtain the fusion feature;

[0010] The fused features are input into the second feedforward neural network to obtain the predicted value of gas concentration.

[0011] Furthermore, the loss function of the second feedforward neural network is:

[0012]

[0013] Among them, a(λ j ,T,P,H)=S(T)·g(λ j,P)·f(H), S(T) represents the absorption cross section of the gas at temperature T; g(λ j ,P) is the absorption linear function, which means that at wavelength λ j The distribution of the absorption intensity at λ0 relative to the central wavelength of the absorption spectrum; f(H) is the humidity correction function; f(H)=1+β·H;

[0014] represents the predicted gas concentration output by the second feedforward neural network during the i-th training, c i represents the actual value of gas concentration in the training set, M represents the size of the training set, N represents the total number of lasers used to measure gas concentration, and λ j represents the wavelength of the jth laser used to measure gas concentration, μ represents the weight coefficient, and I j0 represents the incident light intensity of the jth laser used to measure the gas concentration, λ0 represents the central wavelength of the gas molecule absorption spectrum, γ P represents the half-height width of the spectral line broadening Lorentzian distribution caused by collisions between gas molecules, λ′ represents the integral variable, c represents the speed of light, k B represents the Boltzmann constant, T represents the Kelvin temperature of the gas, ω represents the wavelength-related index, m represents the molecular mass of the gas, β represents the humidity-related constant, H represents the relative humidity, and d represents the path length of the laser through the gas.

[0015] Furthermore, the method of allocating weights of the laser absorption vector feature and the feature map of the environment matrix through the attention mechanism and performing weighted summation to obtain the fusion feature is specifically as follows:

[0016] The laser absorption vector feature is expanded into a tensor of the same dimension as the feature map of the environment matrix through a broadcasting mechanism;

[0017] Calculate the value tensor V corresponding to the tensor of the feature map of each environment matrix 2j =W j v X 2j , W j v Represents X 2j The learning matrix corresponding to the value tensor;

[0018] Calculate the attention weight corresponding to each value tensor,

[0019]

[0020] Calculate fusion features,

[0021]

[0022] Among them, Q1i =W i q X 1i , K 1j =W j k X 2j , X 1i represents the i-th tensor obtained by broadcasting the laser absorption vector feature, W i q Represents X 1i The corresponding query tensor learning matrix, X 2j The jth tensor of the feature map representing the environment matrix, W j k Represents X 2j The corresponding key tensor learning matrix; A(Q 1i ,K 2j ) represents the attention weight of the j-th value tensor corresponding to the tensor of the feature map of the environment matrix, and dx represents K 2j The dimension of , F represents the fused features, M represents the training set size, and N represents the total number of lasers used to measure the gas concentration.

[0023] Furthermore, the method further comprises: expanding the standard temperature, standard pressure and standard humidity into a time series with the same value at each moment, and obtaining a standard environment matrix by intercepting the end of the time series through a sliding window;

[0024] The standard environment matrix is ​​input into the convolutional neural network, and the different standard residual light intensity values ​​within the measurement range are input into the first feedforward neural network to obtain the predicted value. The gas concentration value on the standard absorption curve is taken as the true value, and the neural network is corrected by the following loss function:

[0025]

[0026] The standard absorption curve is a curve obtained by changing the gas concentration within the set measurement range and measuring the laser absorption value of each wavelength under standard temperature, pressure and humidity, and obtaining the laser of each wavelength according to the gas concentration value of each wavelength within the set measurement range and the corresponding laser absorption value. j ,T,P,H)=S(T)·g(λ j ,P)·f(H), S(T) represents the absorption cross section of the gas at temperature T; g(λ j ,P) is the absorption linear function, which means that at wavelength λ j The distribution of the absorption intensity at λ0 relative to the central wavelength of the absorption spectrum; f(H) is the humidity correction function;

[0027] represents the predicted gas concentration output by the second feedforward neural network during the i-th training, c i represents the actual value of gas concentration in the training set, M represents the size of the training set, N represents the total number of lasers used to measure gas concentration, and λ j represents the wavelength of the jth laser used to measure gas concentration, μ represents the weight coefficient, and I j0 represents the incident light intensity of the jth laser used to measure the gas concentration, λ0 represents the central wavelength of the gas molecule absorption spectrum, γ P represents the half-height width of the spectral line broadening Lorentzian distribution caused by collisions between gas molecules, λ′ represents the integral variable, c represents the speed of light, k B represents the Boltzmann constant, T represents the Kelvin temperature of the gas, ω represents the wavelength-related index, m represents the molecular mass of the gas, β represents the humidity-related constant, H represents the relative humidity, and d represents the path length of the laser through the gas.

[0028] Furthermore, the loss function of the second feedforward neural network also includes wavelength weight correction, and the corrected loss function of the second feedforward neural network is:

[0029]

[0030] Among them, ω j represents the weight of the jth laser used to measure gas concentration, is the detection noise variance of the laser intensity of the jth laser of the laser receiver at the standard temperature, T0 is the standard temperature, T is the actual temperature, and α is the noise gain coefficient related to temperature.

[0031] The beneficial effects of the present invention are: compared with the prior art,

[0032] 1) The present invention extracts the laser absorption vector feature through the first feedforward neural network, and extracts the environmental matrix feature map through the convolutional neural network. The environmental matrix feature map is a matrix intercepted at the end of the temperature time series, pressure time series and dust concentration time series through a sliding window, so that the feature map extracted by the convolutional neural network from the environmental matrix can have time characteristics. For environmental variables such as temperature, pressure and humidity, not only the current value will affect the current measurement result, but also the historical value in a short time will affect the current measurement result. The short-term impact can be extracted through the feature map, and then the laser absorption vector feature and the feature map of the environmental matrix are fused through the attention mechanism, and the fused multimodal features are nonlinearly mapped and linearly transformed through the second feedforward neural network, and finally the predicted value of the gas concentration is output, so that the model can better focus on those features that are important for gas concentration prediction, effectively combine spectral information with environmental factors, and improve the accuracy of prediction. Therefore, the present invention can take the influence of environmental factors into account in the measurement of gas concentration under different environments, and automatically correct the gas concentration results through the environmental parameters obtained by the neural network. Compared with the prior art that only determines the gas concentration by laser attenuation intensity, the scope of application is wider, and the measurement results are more accurate in non-standard environments;

[0033] 2) The present invention utilizes the different sensitivity of lasers of different wavelengths to the absorption characteristics of gases, and performs measurements through multiple wavelengths. On the one hand, it can integrate multiple spectral information and reduce measurement errors. The measurement results of different wavelengths verify and complement each other, thereby improving the accuracy of the overall measurement. On the other hand, spectra of different wavelengths exhibit different absorption characteristics under different conditions. By combining measurements of multiple wavelengths, it can better adapt to environmental changes and maintain the stability of measurement results;

[0034] 3) The loss function of the present invention not only considers the difference between the predicted value and the true value of the gas concentration, but also considers the difference between the light intensity value after the laser passes through the gas and the theoretical light intensity value as a physical constraint, so that the model not only relies on data, but also can utilize existing physical knowledge, effectively reduce the interference of noise on the measurement results, and enhance the robustness of the model in the face of uncertainty.

[0035] The model is able to identify and correct errors due to environmental fluctuations, thus providing a more stable output;

[0036] 4) The present invention uses the attention mechanism to perform weighted fusion of the laser absorption vector feature and the feature map of the environment matrix. First, the laser absorption vector is expanded to the same dimension as the feature map of the environment matrix through the broadcast mechanism, and then the value tensor and attention weight of each feature are calculated, and finally the fusion feature is obtained by weighted summation. The weights of different features can be dynamically assigned, so as to more accurately reflect the complex relationship between gas concentration and environmental factors, and improve the accuracy of measurement and the robustness of the model;

[0037] 5) The present invention expands the standard temperature, pressure and humidity into a time series, and uses a sliding window to intercept its end to form a standard environment matrix, which is then input into a convolutional neural network, and the light intensity values ​​under different standard conditions are input into a feedforward neural network for prediction. Finally, the predicted value is compared with the actual gas concentration value under the standard absorption curve through a loss function to correct the parameters of the neural network. The advantage is that the model can be calibrated using accurate measurement data under standard environmental conditions, thereby improving the generalization ability and prediction accuracy of the model under different environmental conditions, while effectively considering the influence of factors such as temperature, pressure and humidity on gas absorption characteristics;

[0038] 6) The present invention introduces wavelength weight correction into the loss function of the second feedforward neural network. The wavelength weight is adjusted according to the noise variance of the laser receiver at the standard temperature and the difference between the actual temperature and the standard temperature, thereby giving different weights to lasers of different wavelengths. The advantage is that it can effectively reduce the measurement error caused by temperature changes, improve the gas concentration prediction accuracy under non-standard temperature conditions, and enhance the robustness of the model in the face of temperature fluctuations. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0040] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0041] Implementation Example 1:

[0042] A laser gas concentration sensor correction method, the method comprising:

[0043] Gas concentration measurements using spectra at different wavelengths;

[0044] Extracting laser absorption vector features through a first feedforward neural network, wherein the laser absorption vector is a vector composed of light intensity values ​​of lasers of different wavelengths after passing through the gas;

[0045] Extracting a feature map of an environmental matrix by a convolutional neural network, wherein the environmental matrix is ​​a matrix intercepted by a sliding window at the end of a temperature time series, a pressure time series, and a humidity time series;

[0046] The weights of the laser absorption vector features and the feature map of the environment matrix are assigned through the attention mechanism and the weighted sum is performed to obtain the fusion feature;

[0047] The fused features are input into the second feedforward neural network to obtain the predicted value of gas concentration.

[0048] The gas concentration is measured by multi-wavelength spectroscopy. The first feedforward neural network is used to extract the features of the laser absorption vector, and the convolutional neural network is used to extract the feature map of the environmental matrix constructed based on the temperature, pressure and humidity time series. Then, the laser absorption features and environmental features are weighted fused using the attention mechanism to obtain the fused features. Finally, the fused features are input into the second feedforward neural network to generate the predicted value of the gas concentration. This method improves the accuracy and adaptability of gas concentration measurement by combining environmental factors and multi-wavelength spectral information.

[0049] Since the measurement in the real environment is inevitably affected by noise and uncertainty, when the model is learned through the training set, the noise will affect the generalization ability of the model. In this embodiment, further, the loss function of the second feedforward neural network is:

[0050]

[0051] Among them, a(λ j ,T,P,H)=S(T)·g(λ j ,P)·f(H), S(T) represents the absorption cross section of the gas at temperature T; g(λ j ,P) is the absorption linear function, which means that at wavelength λ j The distribution of the absorption intensity at λ0 relative to the central wavelength of the absorption spectrum; f(H) is the humidity correction function;

[0052] represents the predicted gas concentration output by the second feedforward neural network during the i-th training, c i represents the actual value of gas concentration in the training set, M represents the size of the training set, N represents the total number of lasers used to measure gas concentration, and λ j represents the wavelength of the jth laser used to measure gas concentration, μ represents the weight coefficient, and I j0 represents the incident light intensity of the jth laser used to measure the gas concentration, λ0 represents the central wavelength of the gas molecule absorption spectrum, and λ Prepresents the half-width of the Lorentzian distribution of spectral line broadening caused by collisions between gas molecules, λ ′ represents the integral variable, c represents the speed of light, k B represents the Boltzmann constant, T represents the Kelvin temperature of the gas, ω represents the wavelength-related index, m represents the molecular mass of the gas, β represents the humidity-related constant, H represents the relative humidity, and d represents the path length of the laser through the gas.

[0053] In order to obtain the optimal μ value, this patent proposes the following method for determining the value of μ, as follows:

[0054] S01, select an initial μ value range;

[0055] S02, perform equidistant segmentation within the initial μ value range;

[0056] S03, performing model training on the μ value of each segmentation point within the initial μ value range;

[0057] S04. Cross-validate the model and calculate the loss value of each μ value on the validation set;

[0058] S05, fitting the loss value with μ as an independent variable to obtain a fitting function;

[0059] S06, using genetic algorithm to search for saddle points of fitting function;

[0060] S07, if no saddle point is found within the initial μ value range, jump to step S08, otherwise the model is trained with μ corresponding to the found saddle point and the process ends;

[0061] S08, expanding the value range of μ and performing equidistant cutting, the cutting spacing is the same as step S02;

[0062] S09. Jump to step S03.

[0063] Through the above method, the value of μ can be determined without any prior knowledge, which has the advantage of strong adaptability.

[0064] Furthermore, the method of allocating weights of the laser absorption vector feature and the feature map of the environment matrix through the attention mechanism and performing weighted summation to obtain the fusion feature is specifically as follows:

[0065] The laser absorption vector feature is expanded into a tensor of the same dimension as the feature map of the environment matrix through a broadcasting mechanism;

[0066] Calculate the value tensor V corresponding to the tensor of the feature map of each environment matrix 2j =W j v X 2j , Wj v Represents X 2j The learning matrix corresponding to the value tensor;

[0067] Calculate the attention weight corresponding to each value tensor,

[0068]

[0069] Calculate fusion features,

[0070]

[0071] Among them, Q 1i =W i q X 1i , K 1j =W j k X 2j , X 1i represents the i-th tensor obtained by broadcasting the laser absorption vector feature, W i q Represents X 1i The corresponding query tensor learning matrix, X 2j The jth tensor of the feature map representing the environment matrix, W j k Represents X 2j The corresponding key tensor learning matrix; A(Q 1i ,K 2j ) represents the attention weight of the j-th value tensor corresponding to the tensor of the feature map of the environment matrix, d k K 2k The dimension of , F represents the fused features, M represents the training set size, and N represents the total number of lasers used to measure the gas concentration.

[0072] The present invention expands the dimension of the laser absorption vector through a broadcast mechanism to match it with the feature map of the environmental matrix, and then calculates the attention weight to give higher weights to the features that are more important for gas concentration prediction, and finally obtains more accurate fusion features for gas concentration prediction. This method effectively addresses the impact of changes in environmental parameters (such as temperature, pressure, and humidity) on the accuracy of gas concentration measurement, and improves the adaptability and robustness of the sensor in different environments.

[0073] In order to improve the prediction accuracy of the model, the method further includes: expanding the standard temperature, standard pressure and standard humidity into a time series with the same value at each moment, and intercepting the end of the time series through a sliding window to obtain a standard environment matrix;

[0074] The standard environment matrix is ​​input into the convolutional neural network, and the different standard residual light intensity values ​​within the measurement range are input into the first feedforward neural network to obtain the predicted value. The gas concentration value on the standard absorption curve is taken as the true value, and the neural network is corrected by the following loss function:

[0075]

[0076] The standard absorption curve is a curve obtained by changing the gas concentration within the set measurement range and measuring the laser absorption value of each wavelength under standard temperature, pressure and humidity, and obtaining the laser of each wavelength according to the gas concentration value of each wavelength within the set measurement range and the corresponding laser absorption value. j ,T,P,H)=S(T)·g(λ j ,P)·f(H), S(T) represents the absorption cross section of the gas at temperature T; g(λ j ,P) is the absorption linear function, which means that at wavelength λ j The distribution of the absorption intensity at λ0 relative to the central wavelength of the absorption spectrum; f(H) is the humidity correction function;

[0077] represents the predicted gas concentration output by the second feedforward neural network during the i-th training, c i represents the actual value of gas concentration in the training set, M represents the size of the training set, N represents the total number of lasers used to measure gas concentration, and λ j represents the wavelength of the jth laser used to measure gas concentration, μ represents the weight coefficient, and I j0 represents the incident light intensity of the jth laser used to measure the gas concentration, λ0 represents the central wavelength of the gas molecule absorption spectrum, γ P represents the half-width of the Lorentzian distribution of spectral line broadening caused by collisions between gas molecules, λ ′ represents the integral variable, c represents the speed of light, k B represents the Boltzmann constant, T represents the Kelvin temperature of the gas, ω represents the wavelength-related index, m represents the molecular mass of the gas, β represents the humidity-related constant, H represents the relative humidity, and d represents the path length of the laser through the gas.

[0078] First, the standard temperature, pressure and humidity are expanded into a time series, and the standard environment matrix is ​​obtained through a sliding window. This standard environment matrix is ​​input into the convolutional neural network, and the light intensity value under standard conditions is input into the feedforward neural network to obtain the predicted value. By comparing these predicted values ​​with the true value of the gas concentration under the standard absorption curve, the neural network is corrected using the loss function.

[0079] Furthermore, the loss function of the second feedforward neural network also includes wavelength weight correction, and the corrected loss function of the second feedforward neural network is:

[0080]

[0081] Among them, ω j represents the weight of the jth laser used to measure gas concentration, is the detection noise variance of the laser intensity of the jth laser of the laser receiver at the standard temperature, T0 is the standard temperature, T is the actual temperature, and α is the noise gain coefficient related to temperature.

[0082] The modified loss function includes the wavelength weight correction part, where the weight ω j The weight ω is adjusted according to the laser detection noise variance and temperature variation. j It is calculated based on the noise level of the laser under different environmental conditions and is used to weight the error term in the loss function to better adapt to the measurement errors of lasers of different wavelengths.

[0083] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A laser gas concentration sensor correction method, characterized in that: The method comprises: Gas concentration measurements using spectra at different wavelengths; Extracting laser absorption vector features through a first feedforward neural network, wherein the laser absorption vector is a vector composed of light intensity values ​​of lasers of different wavelengths after passing through the gas; Extracting a feature map of an environmental matrix by a convolutional neural network, wherein the environmental matrix is ​​a matrix intercepted by a sliding window at the end of a temperature time series, a pressure time series, and a humidity time series; The weights of the laser absorption vector features and the feature map of the environment matrix are assigned through the attention mechanism and the weighted sum is performed to obtain the fusion feature; The fused features are input into the second feedforward neural network to obtain the predicted value of gas concentration.

2. The laser gas concentration sensor correction method according to claim 1, characterized in that: The loss function of the second feedforward neural network is: Among them, a(λ j ,T,P,H)=S(T)·g(λ j ,P)·f(H), S(T) represents the absorption cross section of the gas at temperature T; g(λ j ,P) is the absorption linear function, which means that at wavelength λ j The distribution of the absorption intensity at λ0 relative to the central wavelength of the absorption spectrum; f(H) is the humidity correction function; f(H)=1+β·H; represents the predicted gas concentration output by the second feedforward neural network during the i-th training, c i represents the actual value of gas concentration in the training set, M represents the size of the training set, N represents the total number of lasers used to measure gas concentration, and λ j represents the wavelength of the jth laser used to measure gas concentration, μ represents the weight coefficient, and I j0 represents the incident light intensity of the jth laser used to measure the gas concentration, λ0 represents the central wavelength of the gas molecule absorption spectrum, and λ P represents the half-height width of the spectral line broadening Lorentzian distribution caused by collisions between gas molecules, λ′ represents the integral variable, c represents the speed of light, k B represents the Boltzmann constant, T represents the Kelvin temperature of the gas, ω represents the wavelength-related index, m represents the molecular mass of the gas, β represents the humidity-related constant, H represents the relative humidity, and d represents the path length of the laser through the gas.

3. The laser gas concentration sensor correction method according to claim 1 or 2, characterized in that: The method of allocating weights of the laser absorption vector feature and the feature map of the environment matrix through the attention mechanism and performing weighted summation to obtain the fusion feature is specifically as follows: The laser absorption vector feature is expanded into a tensor of the same dimension as the feature map of the environment matrix through a broadcasting mechanism; Calculate the value tensor V corresponding to the tensor of the feature map of each environment matrix 2j =W j v X 2j , Wj v Represents X 2j The learning matrix corresponding to the value tensor; Calculate the attention weight corresponding to each value tensor, Calculate fusion features, Among them, Q 1i =W i q X 1i , K 1j =W j k X 2j , X 1i represents the i-th tensor obtained by broadcasting the laser absorption vector feature, W i q Represents X 1i The corresponding query tensor learning matrix, X 2j The jth tensor of the feature map representing the environment matrix, W j k Represents X 2j The corresponding key tensor learning matrix; A(Q 1i ,K 2j ) represents the attention weight of the j-th value tensor corresponding to the tensor of the feature map of the environment matrix, d k K 2j The dimension of , F represents the fused features, M represents the training set size, and N represents the total number of lasers used to measure the gas concentration.

4. The laser gas concentration sensor correction method according to claim 3, characterized in that: The method further comprises: The standard temperature, standard pressure and standard humidity are expanded into a time series with the same value at each moment, and the end of the time series is intercepted by a sliding window to obtain the standard environment matrix; The standard environment matrix is ​​input into the convolutional neural network, and the different standard residual light intensity values ​​within the measurement range are input into the first feedforward neural network to obtain the predicted value. The gas concentration value on the standard absorption curve is taken as the true value, and the neural network is corrected by the following loss function: The standard absorption curve is a curve obtained by changing the gas concentration within the set measurement range and measuring the laser absorption value of each wavelength under standard temperature, pressure and humidity, and obtaining the laser of each wavelength according to the gas concentration value of each wavelength within the set measurement range and the corresponding laser absorption value. j ,T,P,H)=S(T)·g(λ j ,P)·f(H), S(T) represents the absorption cross section of the gas at temperature T; g(λ j ,P) is the absorption linear function, which means that at wavelength λ j The distribution of the absorption intensity at λ0 relative to the central wavelength of the absorption spectrum; f(H) is the humidity correction function; f(H)=1+β·H; represents the predicted gas concentration output by the second feedforward neural network during the i-th training, c i represents the actual value of gas concentration in the training set, M represents the size of the training set, N represents the total number of lasers used to measure gas concentration, and λ j represents the wavelength of the jth laser used to measure gas concentration, μ represents the weight coefficient, and I j0 represents the incident light intensity of the jth laser used to measure the gas concentration, λ0 represents the central wavelength of the gas molecule absorption spectrum, γ P represents the half-height width of the spectral line broadening Lorentzian distribution caused by collisions between gas molecules, λ′ represents the integral variable, c represents the speed of light, k B represents the Boltzmann constant, T represents the Kelvin temperature of the gas, ω represents the wavelength-related index, m represents the molecular mass of the gas, β represents the humidity-related constant, H represents the relative humidity, and d represents the path length of the laser through the gas.

5. The laser gas concentration sensor correction method according to claim 2, characterized in that: The loss function of the second feedforward neural network also includes wavelength weight correction, and the corrected loss function of the second feedforward neural network is: Among them, ω j represents the weight of the jth laser used to measure gas concentration, is the detection noise variance of the laser intensity of the jth laser of the laser receiver at the standard temperature, T0 is the standard temperature, T is the actual temperature, and α is the noise gain coefficient related to temperature.

Images