A method for processing saturated spectrum of tunable diode absorption spectrum wide range detection

Abstract

The application discloses a kind of tunable diode absorption spectrum wide-range detection saturated spectrum processing method, it is related to spectral gas detection technical field.The tunable diode absorption spectrum wide-range detection saturated spectrum processing method, by the original absorption spectrum signal is carried out denoising and baseline correction, obtain effective spectrum signal, and extract multiple spectral feature parameters to calculate saturated quantization value, according to the quantization value, using spectral reconstruction algorithm reconstructs complete absorption spectral line profile, and the weight of absorption spectrum and wavelength modulation spectrum is calculated by S function, in combination with ambient temperature, pressure and compensation model, preliminary concentration value is corrected, and accurate final gas concentration is output.The application is by constructing saturated quantization evaluation system, stratified adaptive spectral reconstruction strategy, dual-mode detection fusion mechanism and working condition compensation model, without additional hardware, realize wide-range high-precision gas detection, eliminate environmental interference and saturation influence.

Description

Technical Field

[0001] This invention relates to the field of spectroscopic gas detection technology, specifically to a saturated spectral processing method for wide-range detection of tunable diode absorption spectra. Background Technology

[0002] Tunable diode laser absorption spectroscopy has become one of the mainstream technologies for gas concentration detection due to its advantages such as non-invasive measurement, fast response speed, and high selectivity. Its core principle is to utilize the wavelength tuning characteristics of a diode laser to scan the characteristic absorption lines of the target gas, and then measure the attenuation of the transmitted light intensity, combined with the Lambert-Beer law, to invert the gas concentration.

[0003] In practical applications, TDLAS systems often face a technical bottleneck where it is difficult to balance measurement range and detection accuracy. The core reason lies in the gas absorption saturation phenomenon. When the target gas concentration in the detection environment is too high, the absorption optical path is too long, or the intensity of spectral line transitions is too large, the transmitted light intensity will decrease sharply, resulting in a "flat-top" distortion in the peak region of the absorption spectrum, i.e., absorption saturation. At this time, traditional spectral processing methods have two major drawbacks: First, the concentration inversion algorithm based on full-spectrum profile fitting fails due to the lack of spectral feature information, leading to a significant decrease in measurement accuracy in the high-concentration range; second, the methods used to avoid saturation, such as shortening the optical path and diluting the sample, severely compress the system's measurement range and accuracy, failing to meet the wide range of application requirements from ppm-level trace detection to percentage-level constant detection.

[0004] Existing solutions for absorption saturation have significant limitations: First, some techniques reconstruct saturated spectra by fitting the two wings of spectral lines within a fixed range, but this approach fails to consider the differences in spectral broadening caused by environmental pressure and temperature variations, leading to unstable reconstruction accuracy under complex conditions. Second, multi-wavelength switching detection schemes adapt to different concentration ranges by selecting absorption lines of varying intensities, but require multiple laser sources and optical path switching mechanisms, significantly increasing system complexity and hardware costs, while also exhibiting discontinuous concentration switching. Third, traditional range switching techniques rely on mechanical adjustments to the optical path or manual sample dilution, resulting in response delays and preventing real-time online detection. Furthermore, existing technologies lack quantitative standards for judging saturation levels, leading to haphazard switching of spectral processing strategies and further impacting measurement stability and accuracy.

[0005] Therefore, developing a processing method that can adaptively identify the degree of gas absorption saturation, accurately reconstruct the saturation spectrum, and achieve high-precision detection of the TDLAS system across the entire range has become an urgent technical problem to be solved in this field. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a saturated spectral processing method for wide-range detection of tunable diode absorption spectra, which solves the problems of limited range and insufficient detection accuracy of existing TDLAS technology in gas absorption saturation scenarios.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a saturated spectral processing method for wide-range detection of tunable diode absorption spectra, comprising the following steps: denoising and baseline correction of the original absorption spectral signal to obtain an effective spectral signal; extracting multiple spectral feature parameters from the effective spectral signal and calculating saturation quantization values; fitting the effective spectral signal to reconstruct the complete absorption spectral profile based on the saturation quantization values ​​using a spectral reconstruction algorithm; calculating the weight of the direct absorption spectral detection result using an S-shaped function based on the saturation quantization values; calculating the weight of the wavelength modulation spectral detection result based on the saturation quantization values; obtaining the concentration values ​​obtained by the direct absorption spectral method and the concentration values ​​obtained by the wavelength modulation spectral method, and performing a weighted summation based on the corresponding weights to obtain a preliminary concentration value; obtaining real-time ambient temperature and pressure, and compensating the preliminary concentration value using a pre-trained compensation model based on the complete absorption spectral profile to output the final gas concentration.

[0008] Further, the specific steps for denoising and baseline correction of the original absorption spectrum signal to obtain the effective spectrum signal are as follows: the original absorption spectrum signal is decomposed using wavelet transform, and the high-frequency coefficients obtained by decomposition are thresholded to remove noise; the thresholded coefficients are reconstructed using wavelet to obtain the denoised spectrum signal; the baseline trend of the denoised spectrum signal is fitted using a polynomial fitting algorithm; the baseline trend is subtracted from the denoised spectrum signal to obtain the effective spectrum signal.

[0009] Furthermore, the specific steps for extracting multiple spectral feature parameters from the effective spectral signal and calculating the saturation quantization value are as follows: extract the normalized peak intensity, half-maximum width ratio, peak-to-valley ratio, and second derivative range from the effective spectral signal; input the normalized peak intensity, half-maximum width ratio, peak-to-valley ratio, and second derivative range into a preset weighted summation model; calculate and output the saturation quantization value through the weighted summation model.

[0010] Furthermore, the specific steps for inputting the normalized peak intensity, half-peak width ratio, peak-to-valley ratio, and second derivative range into the preset weighted summation model are as follows: obtain the weight coefficients of the normalized peak intensity, half-peak width ratio, peak-to-valley ratio, and second derivative range; calculate the product of the normalized peak intensity, half-peak width ratio, peak-to-valley ratio, and second derivative range with their corresponding weight coefficients, and sum them to obtain the saturated quantization value.

[0011] Furthermore, based on the saturation quantization value, the specific steps for fitting the effective spectral signal and reconstructing the complete absorption spectral profile using a spectral reconstruction algorithm are as follows: determine the saturation level based on the numerical range of the saturation quantization value; select the corresponding spectral reconstruction algorithm from the preset algorithm library based on the saturation level; use the selected spectral reconstruction algorithm to perform curve fitting on the effective spectral signal; and output the complete absorption spectral profile obtained from the fitting.

[0012] Furthermore, the specific steps for selecting the corresponding spectral reconstruction algorithm from the preset algorithm library according to the saturation level are as follows: when the saturation level is slightly saturated, the fitting algorithm based on the improved Voigt contour model is selected; when the saturation level is moderately saturated, the two-wing segmented fitting and splicing algorithm is selected; when the saturation level is heavily saturated, the sub-interval iterative fitting and fusion algorithm is selected.

[0013] Furthermore, when the saturation level is moderately saturated, the specific steps of the two-wing segmented fitting and splicing algorithm are as follows: Divide the left-wing fitting interval and the right-wing fitting interval with the absorption peak of the effective spectral signal as the center; use a Gaussian function to fit within the left-wing fitting interval to obtain the left-wing fitting curve; use a Lorentz function to fit within the right-wing fitting interval to obtain the right-wing fitting curve; smoothly connect the left-wing and right-wing fitting curves at the boundary to form a complete absorption spectral profile.

[0014] Furthermore, the selected spectral reconstruction algorithm is used to perform curve fitting on the effective spectral signal, and the fitted complete absorption spectral profile is output. Specifically, the curve fitting is iteratively performed with the goal of satisfying the preset accuracy condition for the goodness of fit between the complete absorption spectral profile and the standard spectral line. After one curve fitting, the goodness of fit evaluation index between the current fitted spectral line and the standard spectral line is calculated. If the goodness of fit evaluation index satisfies the preset accuracy condition, the current fitted spectral line is output as the complete absorption spectral profile. If the goodness of fit evaluation index does not satisfy the preset accuracy condition, the parameters of the spectral reconstruction algorithm are adjusted, and the curve fitting of the effective spectral signal is re-executed based on the adjusted parameters.

[0015] Furthermore, based on the complete absorption spectral profile, the specific steps for compensating the initial concentration value using a pre-trained compensation model are as follows: extract profile features from the complete absorption spectral profile; input the profile features, real-time ambient temperature, and real-time ambient pressure into the pre-trained compensation model; in the pre-trained compensation model, correct the initial concentration value based on the profile features, real-time ambient temperature, and real-time ambient pressure; and output the corrected final gas concentration.

[0016] Furthermore, the specific steps for extracting profile features from the complete absorption spectral profile are as follows: calculate the integrated absorbance of the complete absorption spectral profile; determine the peak position and peak intensity of the complete absorption spectral profile; calculate the full width at half maximum (FWHM) of the complete absorption spectral profile; and use the integrated absorbance, peak intensity, and FWHM as profile features.

[0017] The present invention has the following beneficial effects:

[0018] (1) The saturation spectrum processing method for wide-range detection of tunable diode absorption spectrum realizes accurate and objective determination of the degree of gas absorption saturation, and solves the drawback of traditional methods that rely on subjective judgment of saturation level by spectral line morphology. For different saturation levels of light, medium and heavy, the corresponding hierarchical adaptive spectral reconstruction strategy is matched, which can specifically make up for the problem of missing spectral feature information under different saturation levels. Whether it is a high concentration saturation or a low concentration unsaturation scenario, the integrity and accuracy of the reconstructed spectral lines can be guaranteed, and the key spectral features related to gas concentration can be fully preserved.

[0019] (2) The saturated spectrum processing method for wide-range detection of tunable diode absorption spectrum constructs a dual-mode weighted fusion mechanism of direct absorption spectrum and wavelength modulation spectrum. Based on the saturation quantization value, the weight allocation of the two detection modes is dynamically adjusted to achieve a smooth transition of detection modes in different concentration ranges. It effectively avoids the dual shortcomings of insufficient accuracy at low-concentration trace detection and poor stability at high-concentration saturated detection by a single detection mode. This method does not require additional hardware such as laser source and optical path switching mechanism. It achieves wide-range gas detection from ppm-level trace to percentage-level constant through algorithm optimization alone. It can maintain stable detection accuracy in the full concentration coverage range and meet the wide range application requirements in actual detection.

[0020] (3) The saturated spectrum processing method for wide-range detection of tunable diode absorption spectrum introduces real-time ambient temperature and pressure monitoring data, and combines a pre-trained neural network compensation model to correct the detection results under working conditions. This can effectively offset the interference of temperature and pressure changes on spectral features and concentration inversion, and greatly improve the detection stability of the detection system in complex industrial environments.

[0021] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0022] Figure 1 This is a flowchart of a saturated spectrum processing method for wide-range detection of tunable diode absorption spectra according to the present invention.

[0023] Figure 2This is a flowchart illustrating the specific steps in a saturated spectral processing method for wide-range detection of tunable diode absorption spectra according to the present invention. The method involves selecting a corresponding spectral reconstruction algorithm based on the saturation quantization value to fit the effective spectral signal and reconstruct the complete absorption spectral profile.

[0024] Figure 3 This is a flowchart illustrating the specific steps of a saturated spectrum processing method for wide-range detection of tunable diode absorption spectra in this invention, which combines the complete absorption spectral profile with a pre-trained compensation model to compensate for the initial concentration value. Detailed Implementation

[0025] Please see Figure 1 This invention provides a technical solution: a saturated spectral processing method for wide-range detection of tunable diode absorption spectra, comprising the following steps: denoising and baseline correction of the original absorption spectral signal to obtain an effective spectral signal; extracting multiple spectral feature parameters from the effective spectral signal and calculating saturation quantization values; fitting the effective spectral signal to reconstruct the complete absorption spectral profile based on the saturation quantization values ​​using a spectral reconstruction algorithm; calculating the weight of the direct absorption spectral detection result using an S-shaped function based on the saturation quantization values; calculating the weight of the wavelength modulation spectral detection result based on the saturation quantization values; obtaining the concentration values ​​obtained by the direct absorption spectral method and the concentration values ​​obtained by the wavelength modulation spectral method, and performing a weighted summation based on the corresponding weights to obtain a preliminary concentration value; obtaining real-time ambient temperature and pressure, and compensating the preliminary concentration value using a pre-trained compensation model based on the complete absorption spectral profile to output the final gas concentration.

[0026] The sigmoid function is used to determine the saturation quantization value. Calculate the weight of the direct absorption spectroscopy detection results The formula is:

[0027] ;

[0028] in, To adjust the slope (in this embodiment, the value is 200), The threshold value for saturation quantization (taken as 0.5 in this embodiment);

[0029] Weighting of wavelength modulation spectroscopy detection results for: .

[0030] Specifically, the steps for denoising and baseline correction of the original absorption spectrum signal to obtain the effective spectral signal are as follows:

[0031] The original absorption spectrum signal is decomposed using wavelet transform, and the high-frequency coefficients obtained from the decomposition are thresholded to remove noise. Specifically:

[0032] Selecting the db4 wavelet basis function, the original absorption spectrum signal is processed. Perform a 3-level wavelet decomposition to obtain the detail coefficients of each level. and approximation coefficients ,in =1,2,3;

[0033] For detail factor An adaptive thresholding method is used for threshold processing. ,in For the first The noise standard deviation of the layer detail factor (through 3) The criterion is estimated, and the calculation formula is as follows: =0.6745×median(| |), For the first Layer detail factor, median(| | is the median of the absolute values ​​of the detail coefficients. The signal length;

[0034] The threshold function uses a soft threshold function, that is:

[0035] ;

[0036] Wavelet reconstruction is performed on the coefficients after thresholding to obtain the denoised spectral signal, as follows:

[0037] Using inverse wavelet transform, the processed detail coefficients and approximation coefficients Reconstruct the signal to obtain the denoised signal. ;

[0038] The baseline trend of the denoised spectral signal is fitted using a polynomial fitting algorithm, specifically as follows:

[0039] The noise-reduced signal Represented as wavelength The function, denoted as The baseline is fitted using a 5th-order polynomial, i.e.:

[0040] ;

[0041] Solving the coefficients using the least squares method ;

[0042] The fitted baseline was smoothed using a sliding window with a size of 15 sampling points.

[0043] The effective spectral signal is obtained by subtracting the baseline trend from the denoised spectral signal. Specifically, it is as follows:

[0044] Subtract the fitted baseline from the denoised signal, i.e.:

[0045] ,in, The baseline function obtained by fitting is the wavelength. The single-valued function, representing the baseline trend value of the spectral signal at the corresponding wavelength, is obtained by fitting a 5th-order polynomial and smoothing it with a sliding window.

[0046] In this embodiment, when processing the spectral signal of methane gas, after this step, the noise standard deviation of the original spectrum is reduced from 0.025 to 0.003, the baseline drift is controlled within 0.005, and the absorption spectral features that were originally masked by noise and baseline shift are clearly presented. This provides a high-quality and effective spectral signal for the smooth progress of subsequent saturation assessment and spectral reconstruction, avoiding subsequent processing deviations caused by signal quality issues.

[0047] In this implementation scheme, a suitable wavelet basis function is selected for hierarchical wavelet decomposition. Combined with an adaptive thresholding method that estimates the noise standard deviation based on reasonable criteria, and a soft thresholding function for high-frequency coefficient processing, system noise and environmental interference can be accurately separated and removed. At the same time, key information related to gas absorption in the spectrum is completely preserved, avoiding the loss of useful signals. A polynomial fitting baseline is used with sliding window smoothing to accurately capture the slow trend of baseline change, effectively offsetting the baseline shift caused by laser intensity fluctuations and optical path attenuation, making the baseline more stable. The entire process, through the synergistic effect of denoising and baseline correction, provides a clean and reliable effective spectral signal for subsequent steps such as saturation assessment and spectral reconstruction.

[0048] Specifically, the steps for extracting multiple spectral feature parameters from the effective spectral signal and calculating the saturation quantization value are as follows:

[0049] The normalized peak intensity, half-maximum width at half maximum (FWHM), peak-to-valley ratio, and second derivative range are extracted from the effective spectral signal, specifically as follows:

[0050] Let the effective spectral signal be The wavelength corresponding to the peak of the absorption spectral line is Then the normalized peak intensity for:

[0051] ;

[0052] Full width at half maximum (FWHM) for:

[0053] Measured full width at half maximum The full width at half maximum (FWHM) of the absorption spectrum of this gas under standard conditions (296 K, 1 atm) The ratio, that is:

[0054] ;

[0055] Peak-to-valley ratio For: the absorption peak value and the two adjacent troughs (located respectively in and The difference between the average values ​​of () divided by the peak value, i.e.:

[0056] ;

[0057] Second derivative range The second derivative is obtained by performing a second-order numerical differentiation on the effective spectral signal. Calculate its interval The difference between the maximum and minimum values ​​within the range, i.e.:

[0058] ,in For It is a small range centered on the peak, usually taken as half the half-peak width;

[0059] Input the normalized peak intensity, half-peak width ratio, peak-to-valley ratio, and second derivative range into the preset weighted summation model;

[0060] The saturation quantification value is calculated by a weighted summation model. It is a quantification index obtained by weighted summation of normalized peak intensity, half-maximum width ratio, peak-to-valley ratio, and second derivative range after weighting by the analytic hierarchy process. It is used to quantitatively characterize the saturation degree of the absorption spectrum of the target gas. Its value is positively correlated with the spectral saturation degree. The larger the value, the higher the spectral saturation degree.

[0061] The specific steps for inputting the normalized peak intensity, half-maximum width ratio, peak-to-valley ratio, and second-derivative range into the preset weighted summation model are as follows:

[0062] The weighting coefficients for the normalized peak intensity, half-maximum width ratio, peak-to-valley ratio, and second derivative range are obtained as follows:

[0063] The weights of each indicator are determined using the analytic hierarchy process (AHP), and the normalized peak intensity is set. The weight is Full width at half maximum (FWHM) The weight is Peak-to-valley ratio The weight is range of second derivative The weight is And satisfy In this embodiment, we take ; ; ; ;

[0064] Calculate the products of the normalized peak intensity, half-maximum width at half maximum (FWHM), peak-to-valley ratio, and second-derivative range with their corresponding weighting coefficients, and sum them to obtain the saturated quantization value, specifically:

[0065] Saturation quantization value The calculation formula is:

[0066] .

[0067] In this embodiment, parameters were extracted and calculated from the spectra of methane at different concentrations. The saturation quantization value for the 10ppm methane spectrum was 0.12, for the 25% VOL moderately saturated spectrum it was 0.63, and for the 50% VOL heavily saturated spectrum it was 0.88. Each value accurately corresponds to the actual saturation state of the spectrum. This quantization value can directly provide a clear basis for the selection of subsequent reconstruction algorithms, avoiding the misjudgment problem that easily occurs in existing methods that rely on subjective judgment of saturation degree based on spectral morphology.

[0068] In this implementation scheme, by extracting multi-dimensional spectral feature parameters and scientifically allocating weights to calculate saturation quantization values, the accuracy of saturation state judgment is effectively improved. The selected feature parameters comprehensively capture saturation-related information of the absorption spectrum from different perspectives, such as peak intensity, spectral linewidth, peak-valley difference, and derivative change, avoiding the one-sidedness of single-parameter judgment and fully reflecting the actual situation of spectral saturation. The weights are determined by reasonable methods, which fit the degree of influence of each parameter on the saturation state, making the weighted calculation more based on evidence. The final saturation quantization value can clearly define different saturation levels, providing clear guidance for subsequent adaptation to the corresponding spectral reconstruction algorithm and getting rid of the subjective blindness of saturation judgment in traditional methods.

[0069] Specifically, such as Figure 2 As shown, based on the saturation quantization value, the specific steps for fitting the effective spectral signal using a spectral reconstruction algorithm to reconstruct the complete absorption spectral profile are as follows:

[0070] The degree of saturation is determined based on the numerical range of the saturation quantization value, specifically as follows:

[0071] Set two thresholds and ,satisfy For example, take =0.5, =0.8;

[0072] when At that time, it was slightly saturated;

[0073] when At that time, it was moderately saturated;

[0074] when At this time, it is considered severely saturated;

[0075] Based on the saturation level, select the corresponding spectral reconstruction algorithm from the preset algorithm library;

[0076] The selected spectral reconstruction algorithm is used to perform curve fitting on the effective spectral signal, specifically as follows:

[0077] Based on the selected algorithm, nonlinear least squares fitting is performed on the effective spectral signal;

[0078] The complete absorption spectral profile obtained from the output fitting is as follows:

[0079] The fitted spectral profile function is denoted as... Its domain covers the entire scanning wavelength range.

[0080] The specific steps for selecting the corresponding spectral reconstruction algorithm from the preset algorithm library based on the saturation level are as follows:

[0081] When the saturation level is slightly saturated, a fitting algorithm based on the improved Voigt contour model is selected, specifically as follows:

[0082] The improved Voigt contour model is used, and its expression is:

[0083] ;

[0084] in, For a mixed scaling factor, satisfy 0 < <1, The Gaussian function is expressed as:

[0085] ;

[0086] To introduce the Lorentz function with pressure broadening correction, its expression is:

[0087] ;

[0088] in, Expand the half-width for Gaussian. The Lorentz broadening factor is... This is the pressure correction factor obtained through experimental calibration. Due to environmental pressures;

[0089] When the saturation level is moderate saturation, the two-wing segmented fitting and splicing algorithm is selected, specifically as follows:

[0090] The effective spectral signal is centered at the absorption peak. Divided into left-wing sections by the boundary. and right wing section ,in Typically, 2 to 3 times the full width at half maximum (FWHM) is used.

[0091] The left wing was fitted with a Gaussian function, and the right wing with a Lorentz function. Then... Smooth splicing is performed at the point;

[0092] When the saturation level is severely saturated, the sub-interval iterative fitting and fusion algorithm is selected, specifically as follows:

[0093] The entire wavelength range is evenly divided into M sub-intervals, M=4~6, with each sub-interval having an overlapping portion;

[0094] The Levenberg-Marquardt algorithm is used for local fitting within each sub-interval, and then the fitting results of each sub-interval are fused.

[0095] When the saturation level is moderately saturated, the specific steps for selecting the two-wing segmented fitting and splicing algorithm are as follows:

[0096] Centered on the absorption peak of the effective spectral signal, the left and right fitting intervals are divided as follows:

[0097] Let the center wavelength of the absorption peak be The wavelengths corresponding to the absorption line peaks represent the wavelength values ​​corresponding to the positions of maximum absorption intensity in the target gas absorption line. The left wing interval is... The right-wing section is ,in This is the measured full width at half maximum (FWHM).

[0098] A Gaussian function is used to fit the left-wing fitting interval to obtain the left-wing fitting curve, which is as follows:

[0099] Use Gaussian function Least squares fitting was performed on the data points on the left wing to obtain the parameters. and , The amplitude characteristic parameters of the Gaussian function within the fitting interval of the left wing are given. The broadening characteristic parameter of the Gaussian function within the left-wing fitting interval is used to characterize the spectral fitting curve features within the left-wing fitting interval.

[0100] Within the right-wing fitting interval, the Lorentz function is used for fitting to obtain the right-wing fitting curve, which is as follows:

[0101] Using the Lorentz function Least squares fitting was performed on the data points on the right wing to obtain the parameters. and , Let be the amplitude characteristic parameters of the Lorentz function within the fitting interval of the right wing. The broadening characteristic parameter of the Lorentz function within the right-wing fitting interval is used to characterize the spectral fitting curve features within the right-wing fitting interval.

[0102] The left and right wing fitted curves are smoothly connected at the boundary to form a complete absorption spectral profile, as follows:

[0103] exist At any point, the forced function values ​​are equal and the first derivative is continuous, i.e., satisfying:

[0104] ;

[0105] ;

[0106] A smooth connection is achieved by adjusting the parameters, resulting in a complete spectral profile. .

[0107] In this embodiment, the reconstructed spectrum obtained by fitting and splicing the two-wing segmented methane spectrum at 25% VOL has a correlation coefficient of 0.996 with the standard methane spectrum. The reconstructed spectrum of heavily saturated methane at 50% VOL also has a correlation coefficient of 0.991 after sub-interval iterative fitting and fusion. After the spectrum of different saturation levels is reconstructed by the adaptation algorithm, the core concentration-related features are completely preserved, and there is no obvious loss of features.

[0108] In this implementation scheme, by adapting the spectral reconstruction algorithm to the saturation level, accurate restoration of spectra in different saturation states is achieved. For mild saturation, an improved model incorporating pressure correction is adopted, which can adapt to spectral differences caused by environmental changes and accurately capture key spectral features. For moderate saturation, the spectrum is split into two wing regions, and each is fitted with an adapted function to force continuous boundaries, ensuring smooth and complete spectral lines. For severe saturation, multiple sub-regions are overlapped, fitted, and then fused to compensate for feature loss caused by saturation. This hierarchical adaptation method avoids the limitations of a single algorithm, allowing spectra at different saturation levels to be reconstructed with high quality, fully preserving spectral line information related to gas concentration, providing a reliable basis for subsequent concentration calculations, and effectively solving the problem of poor performance of traditional reconstruction methods in complex saturation scenarios.

[0109] Specifically, the selected spectral reconstruction algorithm is used to perform curve fitting on the effective spectral signal, and the complete absorption spectral profile obtained from the fitting is output, as follows:

[0110] With the goal of ensuring that the goodness of fit between the complete absorption spectral profile and the standard spectral line meets the preset accuracy conditions, curve fitting is performed iteratively, specifically as follows:

[0111] Standard spectral lines are The goodness-of-fit evaluation index is the correlation coefficient. and root mean square error (RMSE);

[0112] This index is calculated after each fit, requiring... ≥0.99 and RMSE≤0.01;

[0113] If the conditions are not met, adjust the parameters of the fitting algorithm and refit until the conditions are met or the maximum number of iterations is reached.

[0114] After one curve fitting, the goodness-of-fit evaluation index between the current fitted spectrum and the standard spectrum is calculated, specifically: correlation coefficient. The calculation formula is:

[0115] ;

[0116] in, It is the average value of the standard spectral lines;

[0117] The root mean square error (RMSE) is:

[0118] ;

[0119] If the goodness-of-fit evaluation index meets the preset accuracy condition, the current fitted spectrum will be output as the complete absorption spectrum profile, specifically as follows:

[0120] when If the value is ≥0.99 and RMSE≤0.01, the fitting is considered successful, and the current fitted spectrum is output. ;

[0121] If the goodness-of-fit evaluation index does not meet the preset accuracy condition, the parameters of the spectral reconstruction algorithm are adjusted, and the curve fitting of the effective spectral signal is re-executed based on the adjusted parameters. Specifically:

[0122] For the Levenberg-Marquardt algorithm, adjusting the damping factor Alternatively, adjust the initial parameter values ​​and then refit until the accuracy condition is met or the maximum number of iterations is reached.

[0123] In this embodiment, when the spectrum of heavily saturated methane with 50% VOL was first fitted, the root mean square of the fitting residual was 0.013, which did not meet the preset accuracy requirements. After fine-tuning the damping factor of the LM algorithm and refitting, the root mean square of the residual decreased to 0.009 and the correlation coefficient increased to 0.991, which fully met the accuracy standard. This iterative optimization process can effectively correct the parameter deviation in the fitting process and ensure the effectiveness of the reconstructed spectrum.

[0124] In this implementation plan, by establishing clear goodness-of-fit evaluation criteria and an iterative optimization mechanism, the accuracy and completeness of the reconstructed spectral lines are effectively guaranteed. Using standard spectral lines as a reference, two key indicators are used to comprehensively evaluate the fitting effect, avoiding the one-sidedness of judging by a single indicator and making the judgment of fitting quality more based on evidence. When the fitting result does not meet the preset requirements, the algorithm parameters are adjusted and refitted. For specific algorithms, relevant parameters are precisely adjusted to gradually optimize the fitting effect until the accuracy standard is met or the iteration limit is reached. This closed-loop optimization process can promptly correct deviations in the fitting process and make up for problems caused by insufficient initial parameter settings or algorithm adaptation, so that the reconstructed spectral lines fit the real spectral characteristics to the greatest extent, providing high-quality data support for subsequent concentration inversion, and effectively solving the problems of unstable results and difficulty in guaranteeing accuracy in traditional fitting methods.

[0125] Specifically, such as Figure 3 As shown, the specific steps for compensating the initial concentration value using a pre-trained compensation model, based on the complete absorption spectral profile, are as follows:

[0126] Extracting contour features from the complete absorption spectral profile, specifically:

[0127] Contour features include integrated absorbance Peak intensity Half-peak full width ;

[0128] Integral absorbance The following was obtained by integrating the absorption spectral profile over the scanning wavelength range:

[0129] ;

[0130] Peak intensity for The maximum value;

[0131] Half-peak full width The width is half the height of the peak.

[0132] The contour features, real-time ambient temperature, and real-time ambient pressure are input into the pre-trained compensation model, specifically as follows:

[0133] The pre-trained compensation model is an Elman neural network, which includes an input layer, hidden layers, an output layer, and a continuation layer. Its input layer consists of five neurons: integral absorbance. Peak intensity Half-peak width ,temperature ,pressure ;

[0134] The output layer consists of one neuron, which is the compensated concentration value;

[0135] The network structure is as follows: 5 nodes in the input layer, 12 nodes in the hidden layer, and 1 node in the output layer;

[0136] In the pre-trained compensation model, the initial concentration value is corrected based on contour features, real-time ambient temperature, and real-time ambient pressure. Specifically:

[0137] Let the initial concentration value be... The neural network outputs a correction factor. The corrected concentration is ;

[0138] Alternatively, the neural network can directly output the compensated concentration value. ;

[0139] The output is the corrected final gas concentration, which is specifically: the concentration value output by the compensation model is used as the final gas concentration.

[0140] The specific steps for extracting profile features from the complete absorption spectral profile are as follows:

[0141] The integral absorbance of the complete absorption spectral profile is calculated as follows:

[0142] Numerical integration methods, such as the trapezoidal rule, can be used.

[0143] ;

[0144] The peak positions and peak intensities of the complete absorption spectral profile are determined as follows:

[0145] Peak intensity Peak position In order to make The wavelength with the maximum value;

[0146] The full width at half maximum (FWHM) of the complete absorption spectral profile is calculated as follows:

[0147] Find the height at half the peak ;

[0148] turn up and Two intersecting wavelength points and Then half the peak full width ;

[0149] The integrated absorbance, peak intensity, and full width at half maximum (FWHM) are used as contour feature quantities, specifically:

[0150] These three quantities form the feature vector. .

[0151] In this embodiment, methane gas with a volume of 10% was detected at 70℃ / 1.2 atm. The initial concentration value obtained by dual-mode fusion had a certain deviation. After inputting the integrated absorbance, peak intensity, full width at half maximum (FWHM), and real-time temperature and pressure parameters of the reconstructed spectrum into the compensation model, the relative error between the output compensated concentration value and the standard concentration was only 1.7%. Under the low temperature and low pressure conditions of -40℃ / 0.8 atm, the relative error of the compensated detection was also only 1.9%, which can effectively offset the influence of environmental temperature and pressure changes on the detection results.

[0152] In this implementation plan, by comprehensively extracting key spectral features and compensating for them with environmental parameters, the stability and accuracy of concentration detection are effectively improved. The extracted contour features reflect the core information of the absorption spectrum from different dimensions. Combined with real-time temperature and pressure parameters, they comprehensively cover the key factors affecting concentration calculation, avoiding the one-sidedness of considering a single parameter. The pre-trained neural network structure is designed to meet actual needs, with reasonable configuration of nodes at each layer. It can accurately learn the correlation between feature parameters and concentration error under different operating conditions. By outputting correction factors or directly outputting the compensated concentration, it effectively offsets the interference caused by environmental changes.

[0153] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0154] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A saturated spectral processing method for wide-range detection of tunable diode absorption spectra, characterized in that, Includes the following steps: The original absorption spectrum signal is denoised and baseline corrected to obtain an effective spectral signal; Multiple spectral feature parameters are extracted from the effective spectral signal, and the saturation quantization value is calculated. Specifically: Extract normalized peak intensity, half-maximum width ratio, peak-to-valley ratio, and second derivative range from the effective spectral signal; The normalized peak intensity, half-maximum width ratio, peak-to-valley ratio, and second-derivative range are input into a pre-defined weighted summation model, specifically: Obtain the weighting coefficients for normalized peak intensity, half-maximum width ratio, peak-to-valley ratio, and second derivative range; Calculate the product of the normalized peak intensity, half-maximum width ratio, peak-to-valley ratio, and second derivative range with their corresponding weighting coefficients, and sum them to obtain the saturated quantization value. Based on the saturation quantization value, the corresponding spectral reconstruction algorithm is selected to fit the effective spectral signal in order to reconstruct the complete absorption spectral profile; Based on this saturation quantization value, the weights of the detection results of direct absorption spectrum and wavelength modulation spectrum are calculated respectively. The concentration values ​​obtained by direct absorption spectroscopy and wavelength modulation spectroscopy are obtained, and then weighted and summed according to their respective weights to obtain the preliminary concentration value. The system acquires real-time ambient temperature and pressure, combines them with the complete absorption spectral profile, and compensates for the initial concentration value using a pre-trained compensation model to output the final gas concentration.

2. The saturation spectrum processing method for wide-range detection of tunable diode absorption spectra according to claim 1, characterized in that, The specific steps for denoising and baseline correction of the original absorption spectrum signal to obtain the effective spectral signal are as follows: The original absorption spectrum signal is decomposed using wavelet transform, and the high-frequency coefficients obtained from the decomposition are thresholded to remove noise. Wavelet reconstruction is performed on the coefficients after thresholding to obtain the denoised spectral signal; The baseline trend of the denoised spectral signal is fitted using a polynomial fitting algorithm; The effective spectral signal is obtained by subtracting the baseline trend from the denoised spectral signal.

3. The saturation spectrum processing method for wide-range detection of tunable diode absorption spectra according to claim 1, characterized in that, Based on the saturation quantization value, the specific steps for selecting the corresponding spectral reconstruction algorithm to fit the effective spectral signal and reconstruct the complete absorption spectral profile are as follows: The saturation level is determined based on the numerical range of the saturation quantization value; Based on the saturation level, select the corresponding spectral reconstruction algorithm from the preset algorithm library; The selected spectral reconstruction algorithm is used to perform curve fitting on the effective spectral signal; Output the complete absorption spectral profile obtained from the fitting.

4. The saturation spectrum processing method for wide-range detection of tunable diode absorption spectra according to claim 3, characterized in that, The specific steps for selecting the corresponding spectral reconstruction algorithm from the preset algorithm library based on the saturation level are as follows: When the saturation level is slightly saturated, the fitting algorithm based on the improved Voigt contour model is selected; When the saturation level is moderately saturated, the two-wing segmented fitting and splicing algorithm is selected. When the saturation level is severely saturated, the sub-interval iterative fitting fusion algorithm is selected.

5. The saturation spectrum processing method for wide-range detection of tunable diode absorption spectra according to claim 4, characterized in that, When the saturation level is moderately saturated, the specific steps for selecting the two-wing segmented fitting and splicing algorithm are as follows: The left and right fitting intervals are divided with the absorption peak of the effective spectral signal as the center. A Gaussian function was used to fit the left-wing fitting interval to obtain the left-wing fitting curve. The Lorentz function was used to fit the right wing within the fitting interval to obtain the right wing fitting curve. The left and right fitted curves are smoothly connected at the boundary to form a complete absorption spectral profile.

6. The saturation spectrum processing method for wide-range detection of tunable diode absorption spectra according to claim 3, characterized in that, Using the selected spectral reconstruction algorithm, curve fitting is performed on the effective spectral signal, and the complete absorption spectral profile obtained from the fitting is output, specifically: With the goal of ensuring that the goodness of fit between the complete absorption spectral profile and the standard spectral line meets the preset accuracy conditions, curve fitting is performed iteratively. After one curve fitting, calculate the goodness-of-fit index between the current fitted spectrum and the standard spectrum; If the goodness-of-fit evaluation index meets the preset accuracy condition, the current fitted spectrum will be output as the complete absorption spectrum profile. If the goodness-of-fit evaluation index does not meet the preset accuracy conditions, the parameters of the spectral reconstruction algorithm are adjusted, and the curve fitting of the effective spectral signal is re-executed based on the adjusted parameters.

7. The saturation spectrum processing method for wide-range detection of tunable diode absorption spectra according to claim 1, characterized in that, The specific steps for compensating the initial concentration value using a pre-trained compensation model, based on the complete absorption spectral profile, are as follows: Extract profile features from the complete absorption spectral profile; The contour features, real-time ambient temperature, and real-time ambient pressure are input into the pre-trained compensation model. In the pre-trained compensation model, the initial concentration value is corrected based on the contour feature quantity, real-time ambient temperature, and real-time ambient pressure. Output the corrected final gas concentration.

8. The saturation spectrum processing method for wide-range detection of tunable diode absorption spectra according to claim 7, characterized in that, The specific steps for extracting profile features from the complete absorption spectral profile are as follows: Calculate the integrated absorbance of the complete absorption spectral profile; Determine the peak position and peak intensity of the complete absorption spectral profile; Calculate the full width at half maximum (FWHM) of the complete absorption spectral profile; Integral absorbance, peak intensity, and full width at half maximum (FWHM) are used as profile feature quantities.

Images