Line-by-line calculation and subtraction method for water vapor spectral line overlapping interference in laser methane detection

Abstract

The application relates to the technical field of laser spectrum detection, and particularly discloses a line-by-line calculation and deduction method for water vapor spectrum line overlapping interference in laser methane detection, which comprises the following steps: target mixed absorption spectrum of a to-be-detected gas and a reference signal are synchronously collected through a measurement channel and a reference channel of a double-channel detection unit, and an environmental parameter is acquired through an environmental parameter sensor; the target mixed absorption spectrum and the environmental parameter are input into a preset machine learning model, and an existing interference gas and a corresponding concentration estimation value thereof are determined in combination with a physical feature library. Through fusion of the preset machine learning model and the physical feature library, in combination with a multi-level priority dynamic line-by-line fitting deduction mechanism and a closed-loop verification iteration strategy, the application can adaptively and high-precisely strip the spectrum line overlapping influence of multiple interference gases in a complex dynamic environment.

Description

Technical Field

[0001] This invention relates to the field of laser spectral detection technology, specifically to a line-by-line calculation and subtraction method for water vapor spectral line overlap interference in laser methane detection. Background Technology

[0002] Laser absorption spectroscopy, with its high sensitivity and fast response speed, has been widely used in methane concentration detection in industrial safety. However, in complex actual industrial environments, such as underground coal mines, chemical industrial parks, and underground pipe corridors, the air is usually rich in high concentrations of water vapor and volatile organic compounds such as acetic acid, ethanol, and propane. The absorption lines of these interfering gases often severely overlap with the characteristic absorption lines of methane, making it difficult for conventional detection equipment to extract pure methane absorption signals, which can easily lead to false concentration reports.

[0003] Existing spectral interference subtraction techniques mostly rely on static feature libraries calibrated before shipment and fixed linear subtraction algorithms. However, this approach often faces the following problems:

[0004] The environmental parameters (temperature, pressure, humidity) in industrial sites fluctuate drastically and dynamically, which can cause nonlinear drift, broadening and deformation of the absorption spectra of multiple interfering gases.

[0005] Using static and fixed conventional algorithms to process dynamically and nonlinearly changing overlapping spectral lines inevitably leads to over-subtraction or under-subtraction, resulting in a significant decrease in the measurement accuracy of the system in complex environments, making it difficult to meet the stringent requirements of high-precision real-time detection in industrial settings. Summary of the Invention

[0006] This invention aims to at least partially solve one of the technical problems in related technologies. Therefore, the objective of this invention is to propose a line-by-line calculation and subtraction method for water vapor spectral line overlap interference in laser methane detection, thereby improving the measurement accuracy of laser methane detection equipment in harsh environments.

[0007] To achieve the above objectives, a first aspect of the present invention proposes a line-by-line calculation and subtraction method for water vapor spectral line overlap interference in laser methane detection, applied in a laser methane detection system including a dual-channel detection unit and an environmental parameter sensor, comprising the following steps:

[0008] The target mixed absorption spectrum and reference signal of the gas to be tested are simultaneously acquired through the measurement channel and reference channel of the dual-channel detection unit, and environmental parameters are obtained through the environmental parameter sensor.

[0009] The target mixed absorption spectrum and the environmental parameters are input into a preset machine learning model, and the interfering gases and their corresponding concentration estimates are determined by combining the physical feature library.

[0010] Based on the concentration estimate and the preset interference weight calculation rules, the interference weight coefficient of each interfering gas corresponding spectral line is calculated, and the corresponding spectral lines are divided into multiple priorities according to the interference weight coefficient.

[0011] According to the preset order and fitting algorithm corresponding to the priority, the line-by-line absorption contribution of each interfering gas in the target mixed absorption spectrum is deducted in turn to obtain the residual spectrum after deducting interference.

[0012] The residual spectrum is verified. If the verification is successful, the actual methane concentration is calculated based on the residual spectrum.

[0013] If the residual spectrum fails the verification process, the decoupling model parameters used to calculate the absorption contribution are adjusted in reverse, and the line-by-line absorption contribution subtraction step is re-executed based on the adjusted decoupling model parameters to obtain a verified residual spectrum.

[0014] To achieve the above objectives, a second aspect of the present invention provides a line-by-line calculation and subtraction system for water vapor spectral line overlap interference in laser methane detection, applied in a laser methane detection device including a dual-channel detection unit and an environmental parameter sensor. The system includes:

[0015] The data collaborative acquisition module is used to simultaneously acquire the target mixed absorption spectrum and reference signal of the gas to be tested through the measurement channel and reference channel of the dual-channel detection unit, and to acquire environmental parameters through the environmental parameter sensor.

[0016] The interference feature identification module is used to input the target mixed absorption spectrum and the environmental parameters into a preset machine learning model, and combine the physical feature library to determine the existing interfering gases and their corresponding concentration estimates.

[0017] The weight and priority division module is used to calculate the interference weight coefficient of each interfering gas corresponding spectral line based on the concentration estimate and the preset interference weight calculation rules, and divide the corresponding spectral line into multiple priorities according to the interference weight coefficient.

[0018] The line-by-line fitting subtraction module is used to subtract the line-by-line absorption contribution of each interfering gas in the target mixed absorption spectrum according to the preset order and fitting algorithm corresponding to the priority, so as to obtain the residual spectrum after subtracting interference.

[0019] The closed-loop verification and calculation module is used to verify the residual spectrum. If the verification is successful, the actual methane concentration is calculated based on the residual spectrum.

[0020] The closed-loop verification and calculation module is further configured to, when the residual spectrum fails the verification process, reversely adjust the decoupling model parameters used to calculate the absorption contribution, and trigger the line-by-line fitting subtraction module to re-execute the line-by-line absorption contribution subtraction step based on the adjusted decoupling model parameters, so as to obtain the verified residual spectrum.

[0021] To achieve the above objectives, a third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the above-described method for calculating and subtracting the overlapping interference of water vapor spectral lines in laser methane detection.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] The line-by-line calculation and subtraction method for water vapor spectral line overlap interference in laser methane detection, as described in this invention, overcomes the limitations of traditional static subtraction models and linear filtering algorithms. This invention integrates a pre-set machine learning model and a physical feature library, combined with a multi-level priority dynamic line-by-line fitting subtraction mechanism and a closed-loop verification iteration strategy. This enables adaptive and high-precision removal of the spectral line overlap effects of multiple interfering gases under complex dynamic environments. Simultaneously, this scheme introduces an adaptive blocking reconstruction mechanism for transient pulses and a nonlinear dynamic compensation mechanism for extreme optical path attenuation. This successfully resolves data distortion and missed detection conflicts caused by linear signal processing algorithms when facing extreme electromagnetic shocks and severe physical obstructions, greatly improving the environmental adaptability, measurement accuracy, and long-term operational reliability of laser methane detection equipment in harsh, high-interference industrial scenarios. Attached Figure Description

[0024] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0025] Figure 1 This is a schematic flowchart of the method for calculating and subtracting overlapping water vapor spectral line interference in laser methane detection provided by the present invention.

[0026] Figure 2 This is a comparison of the original mixed absorption spectrum and the spectral waveforms before and after progressive noise reduction in the line-by-line calculation and subtraction method for water vapor spectral line overlap interference in laser methane detection provided by this invention.

[0027] Figure 3 This is a schematic diagram of the decoupling and fitting of overlapping absorption spectra of water vapor and methane based on the Voigt line type in the line-by-line calculation and subtraction method for water vapor spectral line overlap interference in laser methane detection provided by the present invention.

[0028] Figure 4 This is a verification diagram of the spectral morphology and residual spectrum similarity before and after the subtraction of the absorption contribution of multiple interfering gases in the line-by-line calculation and subtraction method of water vapor spectral line overlap interference in laser methane detection provided by the present invention.

[0029] Figure 5 This is an adaptive tracking curve of short-time sudden energy change and dynamic threshold under transient high-energy pulse interference in the line-by-line calculation and subtraction method for water vapor spectral line overlap interference in laser methane detection provided by the present invention.

[0030] Figure 6 This is a comparison of the polluted spectral waveforms and the smooth reconstructed data segment morphology before and after the blocking command is triggered in the line-by-line calculation and subtraction method for water vapor spectral line overlap interference in laser methane detection provided by this invention.

[0031] Figure 7 This is a diagram showing the mapping relationship between the light intensity attenuation ratio and the nonlinear dynamic compensation coefficient in the line-by-line calculation and subtraction method for water vapor spectral line overlap interference in laser methane detection provided by this invention;

[0032] Figure 8 This is a schematic diagram illustrating the implementation of the line-by-line calculation and subtraction system for water vapor spectral line overlap interference in laser methane detection provided by the present invention;

[0033] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0034] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0035] The following describes, with reference to the accompanying drawings, a method, system, and electronic equipment for line-by-line calculation and subtraction of water vapor spectral line overlap interference in laser methane detection according to embodiments of the present invention.

[0036] Example 1:

[0037] This embodiment provides a line-by-line calculation method for subtracting overlapping interference from water vapor spectral lines in laser methane detection. This method is mainly applied in laser methane detection systems that include dual-channel detection units and environmental parameter sensors. By constructing a multi-dimensional signal acquisition and progressive data processing architecture, this method can accurately remove overlapping interference from water vapor and other volatile organic compounds from the extremely complex absorption spectra of mixed gases.

[0038] like Figure 1As shown, the method in this embodiment specifically includes the following steps:

[0039] Step 1: Coordinated acquisition of multidimensional spectral signals and environmental parameters.

[0040] In the initial stage of operation, the laser methane detection system in this embodiment first performs a collaborative acquisition of multi-dimensional data. The system simultaneously acquires the target mixed absorption spectrum and reference signal of the gas to be tested through the measurement and reference channels of the dual-channel detection unit, and simultaneously obtains the environmental parameters within the current detection physical space through the environmental parameter sensor.

[0041] The dual-channel detection unit uses a beam splitter to divide the probe laser emitted by the tunable semiconductor laser into a probe beam and a reference beam. The specific process of acquiring the target mixed absorption spectrum and reference signal of the gas to be tested includes: acquiring the target mixed absorption spectrum through the measurement channel in a preset characteristic absorption band; acquiring the interference-free reference signal through a fixed band of the reference channel; and acquiring the characteristic spectral line signal of the target interfering gas through an extended band of the reference channel.

[0042] Here, the preset characteristic absorption bands mainly cover the fundamental or overtone absorption regions of methane gas. An open or long-range multiple-reflection gas absorption cell is installed within the measurement channel. After the probe beam passes through this absorption cell, it is received by a photodetector, thereby obtaining the target mixed absorption spectrum including methane, water vapor, and other potential background interfering gases. A sealed reference gas cell is installed within the reference channel, filled with a specific concentration of pure target gas and reference gas. The fixed band of the reference channel provides an absolute wavelength reference and a non-absorbed light intensity benchmark, i.e., the reference signal, unaffected by external environmental interference. The extended band is configured to cover the characteristic spectral lines of typical industrial target interfering gases such as water vapor, acetic acid, and ethanol, providing a priori physical reference for subsequent interference feature identification.

[0043] Meanwhile, the environmental parameter sensors include at least a high-precision temperature sensor, an absolute pressure sensor, and a humidity sensor, used to acquire the environmental parameters of the physical environment in which the target gas exists in real time. The aforementioned spectral data and environmental parameters are timestamped together, forming a synchronized data array that provides raw data support for subsequent algorithm analysis.

[0044] Step 2: Progressive noise reduction processing of spectral signals.

[0045] After acquiring the target mixed absorption spectrum and reference signal of the gas to be tested, the original spectral signal often has a low signal-to-noise ratio due to various noise sources in industrial environments, such as electromagnetic interference, power supply ripple, and mechanical vibration of the optical path. Therefore, this embodiment also includes a step of progressive noise reduction processing on the acquired signal data, which is specifically completed in sequence through the following four sub-steps:

[0046] In the first sub-step, digital filtering rules are used to filter out power frequency interference and harmonic components from the signal data. Specifically, the system constructs a finite-length unit impulse response digital filter. Let the original discrete signal sequence acquired as the first sequence; its calculation formula can be expressed as:

[0047] First intermediate sequence: ;

[0048] in, This indicates that the first intermediate sequence output after digital filtering is at the [number]th [position]. The values ​​at each sampling point; This indicates the order of a finite-length unit impulse response filter; This represents the first pre-designed bandstop filter. Each filter coefficient; This indicates that the first sequence of the original acquisition is in the... The values ​​at each sampling point. The filter coefficients are configured with specific zeros for 50 Hz and its integer multiples of harmonics, thereby suppressing power frequency interference in the frequency domain.

[0049] In the second sub-step, the DC bias component in the signal data is eliminated using an arithmetic averaging algorithm. DC bias is often caused by the dark current of the photodetector and the zero-point drift of the amplifier circuit; its calculation formula can be expressed as:

[0050] Second intermediate sequence: ;

[0051] in, This indicates that the second intermediate sequence output after eliminating the DC bias component is in the... The values ​​at each sampling point; This indicates that the first intermediate sequence of the input is at the 1st position. The values ​​at each sampling point; Indicates the total number of sampling points included in a single detection cycle; Indicates the first intermediate sequence at the 1st digit. The values ​​at each sampling point. This step shifts the baseline of the signal to near zero level.

[0052] In the third sub-step, a cumulative moving average algorithm is used to extract and eliminate low-frequency trend interference in the signal data. Because the slow change in ambient temperature causes low-frequency fluctuations in the laser output power, a sliding time window is applied for smoothing. The calculation formula can be expressed as:

[0053] Third intermediate sequence: ;

[0054] in, This indicates that the third intermediate sequence output after eliminating low-frequency trend interference is in the 1st... The values ​​at each sampling point; This indicates that the second intermediate sequence of the input is in the... The values ​​at each sampling point; This indicates the length of the sliding window used in the moving average algorithm. Indicates the second intermediate sequence in the th order. The values ​​at each sampling point.

[0055] In the fourth sub-step, baseline drift is corrected by combining wavelet transform rules, and light intensity fluctuations are corrected by an adaptive filtering algorithm.

[0056] For example, the least mean square algorithm is used for adaptive filtering, and its weight update equation is defined as:

[0057] Adaptive weight vector ;

[0058] in, Indicates the first The adaptive filter weight coefficient vector at the next iteration; Indicates the first The adaptive filter weight coefficient vector at the next iteration; The step size factor represents the value that controls the convergence speed and stability of the algorithm; Indicates the first The error signal value at the next iteration is obtained by subtracting the current filter output value from the expected signal value. Indicates the first The reference signal vector formed by the third intermediate sequence input in the next iteration.

[0059] The output sequence after adaptive filtering is the target mixed absorption spectrum data after noise reduction, which provides a high-fidelity data foundation for subsequent feature extraction.

[0060] like Figure 2 The figure shows a comparison of the original mixed absorption spectrum and the spectral waveforms before and after progressive noise reduction processing. The horizontal axis represents wavelength in nanometers, and the vertical axis represents absorption intensity in volts.

[0061] The figure shows two different colored curves. The red curve represents the unprocessed raw mixed absorption spectrum directly acquired by the dual-channel detection unit. The curve shows a clear upward bias and has dense sawtooth fluctuations and wave-like low-frequency baseline drift. This objectively reflects the 50 Hz power frequency interference, DC bias component, and low-frequency trend interference caused by laser temperature drift in the industrial environment.

[0062] The blue smooth curve represents the target mixed absorption spectrum after progressive noise reduction processing steps. It can be seen that after layers of cleaning using digital filtering rules, arithmetic average algorithm, cumulative moving average algorithm, and wavelet transform combined with adaptive filtering algorithm, the high-frequency fluctuations and low-frequency baseline drift that originally covered the red curve were effectively suppressed. The blue curve restored a relatively symmetrical absorption peak with a relatively flat baseline near the wavelength of 1653 nanometers. Its absorption intensity peak value stabilized at about 0.5 volts, and the baseline smoothly fell back to a level close to zero volts.

[0063] The waveform's transformation from a red, wavy state to a smoother blue state demonstrates that the multi-level progressive noise reduction architecture can effectively cope with electromagnetic interference and light intensity fluctuations under complex working conditions. This provides a high-fidelity spectral data foundation for subsequent preset machine learning models to perform feature identification of multiple interfering gases and to calculate and subtract line-by-line absorption contributions.

[0064] Step 3: Identification of interference gas features based on a preset machine learning model.

[0065] After obtaining the target mixed absorption spectrum after noise reduction, this embodiment inputs the target mixed absorption spectrum and the environmental parameters into a preset machine learning model, and combines it with a physical feature library to determine the interfering gases and their corresponding concentration estimates. This step is a core prerequisite for achieving complex spectral decoupling, and specifically includes the following operation sequence:

[0066] First, a spectral line morphology feature vector is extracted from the denoised target mixed absorption spectrum. This spectral line morphology feature vector contains multiple dimensions of geometric and physical feature parameters. Specifically, the feature extraction module precisely locates the center wavelength positions of all existing absorption peaks by performing first and second derivative operations on the spectral data. Subsequently, for each identified absorption peak, its feature parameters are calculated, including but not limited to: the absolute peak height, full width at half maximum (FWHM), integral area, and left-right asymmetry of the spectral line. These values ​​are arranged in a fixed dimensional order to form a multidimensional spectral line morphology feature vector.

[0067] Subsequently, the spectral line morphology feature vector and the environmental parameters are input into the preset machine learning model, which outputs the presence identifier of the interfering gas, the concentration estimate, and the cross-over degree of the corresponding spectral lines. Optionally, the preset machine learning model adopts a joint model of multivariate nonlinear regression and classification based on a deep random forest architecture. This model has undergone extensive supervised training before leaving the factory using massive amounts of standard sample spectral data containing water vapor, methane, hydrogen sulfide, and other hydrocarbons under different temperatures, pressures, and humidity conditions.

[0068] Multiple decision tree nodes within the model perform multi-level nonlinear logic discrimination on the input feature vector and environmental parameters, ultimately providing prediction results at the output layer. The presence of interfering gases is indicated by a Boolean value, signifying the presence of a specific interfering gas. The concentration prediction is an approximate value of the gas volume concentration rapidly inferred by the model based on peak height and integral area. The cross-over overlap is a continuous value from 0 to 1 output by the model after evaluating the wavelength distance and linewidth broadening effect range between the currently identified interfering spectral lines and the methane central absorption frequency, reflecting the degree of physical interpenetration between the spectral lines in the frequency domain.

[0069] Step 4: Calculate interference weights and dynamically divide priority deduction.

[0070] After obtaining the basic parameters for model prediction, this embodiment introduces a priority-based multi-level processing mechanism because the interference mechanisms and severity of different interfering gases on methane detection vary significantly. The specific process is as follows: based on the concentration prediction value and the preset interference weight calculation rules, the interference weight coefficient of each interfering gas's corresponding spectral line is calculated, and the corresponding spectral lines are divided into multiple priority levels according to the interference weight coefficient. This step specifically includes:

[0071] For any identified interfering gas spectral line, the system obtains its first overlap with the methane spectral line, the normalized value of the concentration estimate, and the second overlap with other interfering gas spectral lines; wherein the first overlap and the second overlap are both calculated based on the integral overlap area of ​​the spectral line shape function.

[0072] Subsequently, the system performs a weighted calculation step, calculating the weighted sum of the first overlap, the normalized value, and the second overlap to obtain the interference weight coefficient. The calculation process can be expressed as follows:

[0073] Interference weighting coefficient: ;

[0074] in, This represents the calculated interference weighting coefficient for a specific interfering gas spectral line currently being processed; This represents the first preset static weight allocation coefficient corresponding to the first degree of overlap; This represents the first degree of overlap between the specific interfering gas spectral line and the target methane characteristic spectral line; This represents the second preset static weighting coefficient corresponding to the normalized value of the concentration estimate; This represents the normalized value obtained by scaling the estimated concentration of the specific interfering gas to its maximum and minimum values. This represents the third preset static weight allocation coefficient corresponding to the second degree of overlap; This represents the second degree of overlap between the specific interfering gas spectral line and other non-methane interfering gas spectral lines in the mixed gas system. In this embodiment, , , The sum of the three values ​​is always equal to one.

[0075] After calculating the interference weight coefficient, the system determines the preset value range in which the interference weight coefficient is located and divides the corresponding spectral lines into high priority, medium priority or low priority respectively.

[0076] For example, a first threshold boundary and a second threshold boundary are pre-defined in the system memory. If the value is greater than or equal to the first threshold boundary, the interfering spectral line poses an extremely high threat to methane measurement and is classified as high priority; if Those less than the first threshold boundary and greater than or equal to the second threshold boundary are classified as medium priority; if If the concentration is below the second threshold boundary, it indicates that the interference is located in the spectral edge region or has an extremely low concentration, and it is classified as low priority. This dynamic prioritization mechanism can concentrate limited system computing resources on the most destructive interference sources.

[0077] Step 5: Priority-based multi-interference gas line fitting subtraction.

[0078] This embodiment, based on the preset order corresponding to the priority and the fitting algorithm, sequentially subtracts the line-by-line absorption contribution of each interfering gas in the target mixed absorption spectrum to obtain the residual spectrum after interference removal. The specific process is detailed below:

[0079] First, the order of removal is determined according to the descending order of the interference weight coefficients. This means that the system always prioritizes removing the interference components that have the most severe impact on the current spectral morphology. For spectral lines of different priorities, the system calls line model models with different computational complexities to achieve an optimal balance between accuracy and real-time performance.

[0080] For spectral lines classified as high priority, a joint iterative algorithm incorporating multiple environmental parameters and a hybrid line-type convolution algorithm are used for subtraction. Due to the severe overlap of high-priority spectral lines, their physical broadening mechanism includes both Doppler broadening dominated by the thermal motion of gas molecules and collisional broadening dominated by intermolecular collisions. Therefore, a Voigt line-type model capable of simultaneously describing both effects must be employed. The Voigt line-type is the mathematical convolution of the Gaussian and Lorentz line-types, and its formula can be defined as:

[0081] Mixed line type values:

[0082] ;

[0083] in, This indicates the position at which the frequency is calculated. The numerical value of the mixed linear absorption intensity at the location; This represents the overall integrated intensity of the absorption line, which depends on the estimated gas concentration and temperature parameters. This represents the currently calculated absolute optical frequency value; This represents the numerical value of the central optical frequency corresponding to the absorption of the molecular transition; Represents the continuous integration variable parameters used for the convolution integral; This represents the full width at half maximum (FWHM) value of the Gaussian line shape determined by the Doppler effect, and this value is directly related to the currently acquired temperature parameter. This represents the half-height and half-width values ​​of the Lorentz line shape, determined by the collision effect. These values ​​are directly related to the currently obtained pressure parameters, temperature parameters, and composition of the gas mixture.

[0084] During joint iteration, the system uses the nonlinear least squares method to input the collected actual environmental parameters as initial values ​​into the iterative equation, continuously approximating and decoupling the precise linear profile of the high-priority interfering gas under the current temperature and pressure. Then, it subtracts and strips the corresponding line-by-line absorption contribution vector from the target mixed absorption spectrum.

[0085] like Figure 3 The results demonstrate the performance of decoupling and fitting the overlapping absorption spectra of water vapor and methane based on the Voigt line. The horizontal axis represents wavelength in nanometers, and the vertical axis represents relative light intensity in units of absorption.

[0086] The figure contains three curves with clear physical meaning. The red solid line represents the mixed absorption peak with water vapor interference obtained through the detection channel. This waveform exhibits asymmetry and wing broadening near the wavelength of 1653 nm, which reflects the overlap between the high concentration of water vapor absorption lines and the characteristic methane lines.

[0087] The blue dashed line represents the independent absorption curve of water vapor calculated by the system using the nonlinear least squares method in combination and iteratively for high-priority water vapor interference. The dashed line depicts the Voigt mixed linear profile well under the combined effect of Doppler broadening of gas molecule thermal motion and broadening of intermolecular collisions. It extracts the absorption interference intensity with a peak value of 0.450 in the overlapping region.

[0088] The green solid line represents the residual profile obtained after subtracting the blue water vapor interference component from the red mixed spectrum. The waveform transformation from the asymmetric shape of the red solid line to the relatively symmetrical shape of the green solid line confirms that the joint iterative algorithm and the hybrid linear convolution algorithm can effectively decouple and weaken influential dynamic interference components from the aliased physical spectrum. This ensures that the extracted residual spectrum mainly retains the characteristic absorption information of methane gas, providing data support for the accuracy of the final methane concentration calculation.

[0089] For the spectral lines classified as medium priority, their overlap with the methane main peak is relatively light, and they are typically located in the flank region of the spectrum. To accelerate the calculation, the system employs a partially iterative algorithm with a single environmental parameter variable and a Gaussian approximate convolution algorithm for subtraction. At this point, ignoring complex collision broadening effects and assuming that Doppler broadening is dominant, the calculation formula simplifies to the standard Gaussian distribution formula:

[0090] Gaussian line type values: ;

[0091] Here, only the temperature variable from the environmental parameters is introduced. In the computational iteration, it can quickly fit and remove the middle priority interference.

[0092] For spectral lines classified as low priority, since their crosstalk to methane detection is negligible, a comprehensive fitting subtraction is performed by directly calling pre-calculated standard values ​​from the physical feature library. That is, the standard spectral line shape is obtained directly from the mixed spectrum based on the presence identifier and concentration estimate using a lookup table method, and then proportionally scaled before being directly subtracted, greatly saving the computational power of the system's underlying controller.

[0093] After stripping away all priority levels of interfering spectral lines layer by layer, the remaining spectral data array in system memory is the residual spectrum after interference removal. Ideally, this residual spectrum should contain only the characteristic absorption information of pure methane gas.

[0094] Step 6: Residual spectrum closed-loop verification and nonlinear iterative correction.

[0095] To effectively avoid undersubtraction or oversubtraction caused by prediction errors in the early stages of machine learning models or drastic changes in the physical environment, this embodiment forcibly introduces a strict multi-dimensional closed-loop verification mechanism at the end of the algorithm. The system performs a verification process on the residual spectrum, specifically including determining whether the following three multi-dimensional verification conditions are met simultaneously:

[0096] First, determine whether the cosine similarity between the residual spectrum and the standard pure methane spectrum is greater than or equal to a preset similarity threshold. The mathematical formula for this is:

[0097] Cosine similarity score:

[0098] ;

[0099] in, This represents the calculated cosine similarity score. This represents the total number of discrete sampling points contained in the residual spectrum and the standard pure methane spectrum; The residual spectrum indicates that the residual spectrum is in the first... The absorption intensity values ​​at each sampling point; This indicates that the standard pure methane spectral line pre-stored in the system memory under the same temperature and pressure conditions is at the [number]th [number]. The formula represents the absorption intensity values ​​at each sampling point. It quantifies the directional consistency of two spectral vectors in multidimensional space.

[0100] Second, determine whether the residual absorption intensity of each interfering gas at its characteristic wavelength in the residual spectrum is less than or equal to a preset absorption intensity threshold. That is, extract the residual spectrum value corresponding to the center wavelength position of interfering gases such as water vapor and acetic acid. If the value still shows a significant spike, it indicates that the corresponding interfering gas has not been completely subtracted.

[0101] Third, determine whether the deviation between the methane absorption coefficient after interference deduction and the theoretical absorption coefficient under the current environmental parameters is within the preset allowable deviation range. This step mainly verifies the law of conservation of physical energy to prevent the deduction algorithm from incorrectly removing the absorption energy that originally belonged to methane.

[0102] If all the above verification conditions are met, the residual spectrum is determined to have passed the verification process, and the algorithm channel is allowed to proceed. It should also be noted that if the residual spectrum fails the verification process, i.e., any one of the above three conditions is not met, the system triggers a feedback correction mechanism:

[0103] Reverse tuning of the decoupled model parameters used to calculate the absorption contribution, such as fine-tuning. The weighting parameters or the joint iterative compensation factor of the modified environmental parameters are used, and the above-described line-by-line absorption contribution subtraction step is re-executed based on the adjusted decoupled model parameters. This closed loop will continue to iterate until a validated residual spectrum is obtained, thereby fundamentally ensuring the high reliability of the measurement results against interference at the algorithm level.

[0104] like Figure 4 The figure shows the spectral morphology before and after subtracting the absorption contributions of multiple interfering gases, as well as the effect of residual spectral similarity verification. The horizontal axis represents the detection wavelength in nanometers, and the vertical axis represents the relative absorption intensity.

[0105] The solid red line in the figure represents the unmixed absorption spectrum before subtraction without algorithm stripping. This waveform shows a significant asymmetric wide bulge near the wavelength of 1653 nm, which objectively reflects the cross-over and signal interference caused by water vapor and volatile organic compounds on the methane main peak.

[0106] The blue dashed lines represent the standard pure methane characteristic absorption lines pre-existing in the system's physical characteristic library, exhibiting a standard symmetrical peak shape as a theoretical benchmark anchor point.

[0107] The green solid line represents the residual spectral lines output after processing by a multi-level priority dynamic line-by-line fitting subtraction mechanism.

[0108] As can be observed from the waveform transformation, the interference envelope in the red curve is effectively pruned. The green residual spectral line maintains an overall consistency with the blue standard pure methane characteristic absorption line in terms of macroscopic morphology, peak extrema, and baseline broadening, with a calculated cosine similarity value of 0.990. Simultaneously, at the microscopic level, the green curve retains the true instrument random noise floor and the small fitting residual at the original water vapor absorption peak position. This consistency in macroscopic morphology and the realistic fluctuations in microscopic details indicate that the line-by-line calculation subtraction method can effectively isolate the overlapping effects of multiple interfering gases from a complex dynamic environment, contributing to improved signal quality input to the calibration function and the reliability of concentration measurement results.

[0109] Step 7: Final calculation of methane concentration.

[0110] After obtaining a rigorously verified pure residual spectrum, the system performs a step to calculate the actual methane concentration based on the residual spectrum. This process is based on the Beer-Lambert laws of physics, and the specific steps include:

[0111] The effective absorption signal quantity characterizing methane in the residual spectrum and the stable fluctuation quantity of the reference channel in a standard environment are obtained. The effective absorption signal quantity is usually taken as the peak height or integral area of ​​the main absorption peak of methane in the residual spectrum. Then, the first ratio of the effective absorption signal quantity to the stable fluctuation quantity is calculated. Finally, the first ratio is input into a pre-set methane concentration calibration function to calculate the actual value of the methane concentration.

[0112] The calibration function is a nonlinear polynomial fitting equation, defined as:

[0113] Actual value of methane concentration ;

[0114] in, This represents the actual value of the final calculated methane gas volume concentration. This represents the numerical value of the effective absorption signal characteristic of methane extracted from the residual spectrum; This represents the reference value of the stable fluctuation of the non-absorbed light intensity of the reference channel under a standard interference-free environment; , , These represent the nonlinear calibration coefficients determined by fitting multiple concentration standard gases during the system's factory calibration phase, corresponding to the quadratic term coefficient, the linear term coefficient, and the constant term bias value, respectively.

[0115] At this point, the system has completed all data processing tasks for a full detection cycle and outputs high-precision target concentration data.

[0116] For example, suppose this laser methane detection system is deployed at the working face of a deep coal mine. The environmental parameters collected by the environmental parameter sensors are: absolute temperature parameter... (Kelvin), absolute pressure parameters (Pascal) The tested environment contained extremely high concentrations of water vapor interference and trace amounts of acetic acid interference gas.

[0117] In step three above, the denoised target mixed absorption spectrum is input into a preset machine learning model. The model identifies the presence of water vapor as true and outputs the estimated water vapor concentration as a volume percentage. The presence marker of acetic acid was identified as true, and the estimated concentration of acetic acid was output as follows: ppm. Meanwhile, the model output water vapor spectral lines are at a center wavelength of... Overlap characteristic parameters of nanomethane spectral lines.

[0118] In step four, regarding the interference term "water vapor," the system obtains its relevant parameters and substitutes them into the formula to calculate the interference weighting coefficient: ;

[0119] The system presets static weight allocation coefficients: Let the first preset static weight allocation coefficient be... The second preset static weight allocation coefficient The third preset static weight allocation coefficient .

[0120] Model output parameters: the first overlap value of the characteristic spectral lines of water vapor and methane. (Severe overlap); Normalized values ​​of water vapor concentration estimates after scaling by maximum and minimum values. (Approaching full-scale saturation humidity); the second degree of overlap between water vapor and trace amounts of acetic acid in the mixed system. .

[0121] Substitute the specific numbers mentioned above directly into the formula to derive the result:

[0122] ;

[0123] ;

[0124] The first threshold boundary set internally by the system is Because of the calculated interference weighting coefficient of water vapor. Greater than The system classifies the water vapor interference spectral line as high priority.

[0125] Similarly, regarding the interference term of acetic acid, due to its extremely low concentration and deviation from the absorption peak... Nanometers are far away; assuming they are calculated using the same formula... Less than the second threshold boundary (set to) Therefore, it is classified as a low priority.

[0126] Moving to step five, the system first handles high-priority water vapor interference. A hybrid linear convolution algorithm (Voygett linear) is invoked for decoupling. At this specific temperature... Under Kelvin and standard atmospheric pressure, nonlinear least squares iterative fitting is performed based on the underlying theoretical formulas of Lorentz broadening and Doppler broadening.

[0127] Assuming that through iteration, the exact spectral frequency position can be calculated. wavenumber (corresponding to approximately) At the water vapor interference peak near the nanometer level, the mixed linear absorption intensity value is... (Relative light intensity absorption unit). The system then immediately... The numerical distribution vector is subtracted from the target mixed absorption spectrum data array. For low-priority acetic acid interference, feature library constants, such as the overall absorption background value, are directly used. Perform a simple subtraction.

[0128] In step six, the system performs a similarity check on the residual spectrum after subtracting water vapor and acetic acid. Assuming the extracted residual spectrum... Feature vector of discrete sampling points Compared with the pre-stored standard methane feature vector Substitute the values ​​into the cosine similarity formula to perform inner product and norm operations.

[0129] Set the sum of the vector inner product values ​​to be substituted. ;

[0130] residual spectral vector magnitude ;

[0131] Standard spectral vector magnitude .

[0132] Substituting the specific numbers mentioned above directly into the cosine similarity formula for derivation:

[0133] .

[0134] The system's preset similarity threshold is 0.985. Since the calculated value of 0.990 is greater than 0.985, and assuming that other absorption intensity thresholds and deviation verifications are met, the residual spectrum successfully passed the verification process and proceeded to the final concentration calculation.

[0135] Finally, proceed to step seven. Extract the validated residual spectrum and obtain the numerical value of the effective absorption signal quantity characterizing methane. (Volts); Obtain the reference channel's calibrated stable fluctuation baseline value. (volt).

[0136] Calculate the first ratio as follows: ;

[0137] Substitute this first ratio into the pre-set polynomial equation for the methane concentration calibration function. Let the factory calibration coefficients be: quadratic term coefficients. coefficient of the first term constant term .

[0138] Derivation by substituting numerical values ​​into the formula:

[0139] ;

[0140] ;

[0141] .

[0142] The final system output shows that the actual volume concentration of methane in the current environment is 3.16892%.

[0143] The above analysis and derivation show that, without employing the multi-dimensional priority partitioning and hybrid linear convolution subtraction mechanism of this invention, the significant absorption interference contributed by water vapor near 1653 nm (such as the assumed intensity of 0.450 mentioned above) will be directly mixed into the methane signal, leading to subsequent calculations... This can cause an abnormal amplification of the effective signal, and the final output methane concentration may be falsely inflated by more than 10%, leading to serious concentration false alarms.

[0144] The technical solution of this application, through the progressive stripping of the underlying algorithm and rigorous verification loop, can still output a highly accurate measurement result of 3.16892% under extreme water vapor interference conditions, and has outstanding application value.

[0145] Example 2:

[0146] Based on Example 1, this embodiment focuses on an extreme physical noise phenomenon commonly found in complex industrial applications: transient high-energy pulse interference caused by the start-up and shutdown of large electromechanical equipment, the engagement of relays with high current, or the instantaneous release of strong electromagnetic radiation in space.

[0147] Conventional linear low-pass filtering or moving average algorithms are ineffective at filtering transient pulses with extremely fast rise times and high energy amplitudes. Instead, they diffuse and smear the pulse's energy across adjacent time sampling points, artificially creating deceptive absorption peaks with wide amplitudes on the spectral waveform. To effectively suppress this vulnerability at the underlying data link level and ensure the purity of the feature recognition model's input, this embodiment introduces an independent hardware-level and algorithm-level collaborative defense mechanism at the front end of the basic noise reduction architecture.

[0148] Specifically, according to the aforementioned method framework, this embodiment includes an adaptive blocking and reconstruction step for transient pulse interference before filtering out power frequency interference and harmonic components from the signal data using digital filtering rules. This step, as an independent pre-processing data cleaning daemon thread, runs in a field-programmable gate array or high-speed digital signal processor, and performs rigorous abrupt energy checks on each frame of spectral data flowing into the system. Specifically:

[0149] The system first constructs a time sliding window containing a preset number of data points, and then calculates the short-time abrupt energy value of the target mixed absorption spectrum within the current time sliding window.

[0150] Since the original analog-to-digital converter outputs a continuous discrete-time sequence, the system allocates a first-in-first-out circular buffer in memory as the physical carrier for this time sliding window. Let the discrete-time sequence of the input target mixed absorption spectrum be... , representing the raw, unsmoothed mixed spectral discrete sequence; the capacity length of the time sliding window is set to . The length value is determined based on the Nyquist relationship between the system sampling rate and the typical industrial pulse width; the data point step size for each forward slide of the time window is set to... .

[0151] To accurately capture the highly distinctive transient abrupt changes in pulse signals, this embodiment abandons the conventional absolute amplitude energy calculation method and instead adopts a short-time abrupt change energy calculation model based on first-order difference. Its complete calculation formula can be expressed as:

[0152] Short-term mutation energy values: ;

[0153] in, Indicates the current number The short-time mutation energy value is calculated using a time sliding window; This indicates the preset capacity length value of the time sliding window, corresponding to the aforementioned preset number of data points; This indicates that the original sequence of the target mixed absorption spectrum is in the current... The first sliding window The amplitude value of the sampling point at each index position; This indicates that the original sequence of the target mixed absorption spectrum is in the current... The amplitude value of the sampling point at the previous index position adjacent to it within each sliding window; This represents the discrete time step order variable representing the sliding window at the current time. This indicates the step size of the time-based sliding window during each update.

[0154] This formula can amplify the energy weight of abrupt changes by calculating the sum of squares of the differences between adjacent sampling points, while maintaining extremely low sensitivity to the slowly changing shape of real gas absorption peaks, thus achieving accurate stripping and quantization of high-frequency transient pulses in the time domain.

[0155] Next, after obtaining the energy state of the current sliding window, the system determines whether the short-term sudden energy value is greater than the dynamic energy threshold, wherein the dynamic energy threshold is calculated based on the background energy benchmark value within a historical interference-free period.

[0156] It is also important to note that the background optical white noise in industrial environments slowly and globally shifts due to laser operating temperature drift, detector dark current thermal effects, and the accumulation of dust in the optical path. Using a fixed energy threshold can easily lead to frequent false triggers when background noise increases, or to missed detections of weak pulses when background noise decreases. Therefore, the system establishes a dedicated background evaluation queue in memory to store historical, interference-free, short-term abrupt energy change values.

[0157] Specifically, the system calculates the expected value and standard deviation of all samples in the queue in real time, and the derived calculation formula can be expressed as follows:

[0158] Dynamic energy threshold: ;

[0159] in, This indicates the dynamic energy threshold result applicable to the current detection cycle; This represents the arithmetic mean of all historical, undisturbed, short-term mutation energy samples in the current background evaluation queue. This represents the overall standard deviation of all historical, undisturbed, short-term mutation energy samples in the current background evaluation queue. This represents the pre-configured sensitivity tolerance factor. The sensitivity tolerance factor is usually set to a real number greater than two and less than five, used to control the tolerance boundary of the system to sudden signals.

[0160] This formula allows the dynamic energy threshold to be adaptively tightened or relaxed based on the stability of the on-site electromagnetic environment. The system uses the short-term abrupt energy values ​​calculated in the previous steps. The dynamic energy threshold derived from the current update Logical operations for comparing numerical values.

[0161] like Figure 5 The figure shows the short-time sudden energy change and dynamic threshold adaptive tracking curves under transient high-energy pulse interference. The horizontal axis represents the discrete time sliding window sequence, and the vertical axis represents the energy amplitude calculated by the system in real time.

[0162] The solid red line in the figure represents the short-term mutation energy curve calculated in real time, while the dashed blue line represents the dynamic energy threshold curve calculated based on the background energy baseline value within a historical period without interference.

[0163] From the waveform transformation of the graph, it can be observed that during the stable phase of the time sliding window sequence from 0 to 240, the red solid line exhibits a slight fluctuation around the value of 0.5, which reflects the background optical white noise of a typical industrial environment. At the same time, the blue dashed line undergoes a smooth adaptive evolution based on the slight drift of the red solid line and remains at a position of around 1.5.

[0164] When the time window sequence advances to around 250, it simulates the transient high-energy pulse interference caused by the start-up and shutdown of electromechanical equipment. At this time, the red short-term abrupt change energy curve undergoes a significant change, and its energy amplitude rapidly climbs to around 8.2, surpassing the blue dynamic energy threshold curve.

[0165] At the intersection of the red solid line and the blue dashed line, the system identifies a sudden pulse in the spectral data flowing into the current time window and generates a blocking command to freeze the weight update process of the subsequent adaptive filtering algorithm. This judgment and dynamic tracking logic demonstrate that the pre-processing data cleaning daemon has a good ability to capture physical noise, and the dynamic threshold built using historical benchmarks helps to avoid the false triggering defects that are easily caused by fixed thresholds.

[0166] Next, following the conditional judgment logic, if the short-term mutation energy value is greater than the dynamic energy threshold, the system generates a blocking instruction to freeze the subsequent weight update process of the adaptive filtering algorithm.

[0167] For example, when the inequality result is true, it indicates that the spectral data currently flowing into the time window has been severely contaminated by a high-energy electromagnetic pulse. The system's underlying controller will immediately trigger a high-priority hardware interrupt signal or set a specific out-of-bounds flag register in the software state machine. This hardware interrupt signal or out-of-bounds flag register is collectively referred to as the blocking instruction.

[0168] Because the technical solution of this application uses an adaptive filtering algorithm based on criteria such as minimum mean square error in subsequent steps, the core of this type of adaptive algorithm lies in relying on the error gradient to iteratively find the optimal weight coefficients. Once contaminated data containing huge abrupt errors flows into the adaptive filter, the calculated error gradient will instantly tend to infinity, causing the filter weight matrix to diverge and collapse, requiring an extremely long iterative convergence period to recover.

[0169] To completely block this erroneous learning process, the system responds to the blocking command by directly suspending the step size update function of the adaptive filtering algorithm at the memory addressing level. Upon receiving the blocking command, the frozen state variable is forcibly assigned a value of 0; during normal, interference-free periods, the frozen state variable remains a constant of 1. In the weight update formula of the adaptive filter, the step size factor is multiplied and bound to this frozen state variable, ensuring that the weight vector remains at the static historical best value of the previous normal sampling moment throughout the entire time span of the abrupt pulse, thereby achieving absolute isolation and protection of the core filtering model.

[0170] In order to ensure the continuity of the morphological data used for feature extraction in the time domain, the system responds to the blocking command by extracting the pre-stored clean background waveform data segment, and uses the clean background waveform data segment to replace the target mixed absorption spectrum within the current time sliding window with a waveform of the same frequency and bit width to obtain a smooth reconstructed data segment.

[0171] Specifically, in a laser methane detection system, the wavelength of the semiconductor laser is tuned and scanned by injecting periodic triangular or sawtooth wave currents. Each phase point within a tuning cycle corresponds precisely to a specific output wavelength. Therefore, the system utilizes this precise spatiotemporal mapping relationship to pre-store an array of complete scan cycles as a template in non-volatile memory. This template array consists of pure background waveform data segments extracted and persistently stored during the equipment's factory calibration phase or in an extremely clean environment with a long field history.

[0172] When the blocking command takes effect, the system first extracts the start and end phase indices of the time window where the pulse interference is occurring within the entire laser scanning modulation cycle. Then, it precisely extracts a subset of data corresponding to the same start and end phase indices from the pre-stored clean background waveform data segment. However, if the extracted data subset is directly and abruptly replaced with the original contaminated data, this non-smooth splicing replacement will create artificial level step breaks at the left and right boundaries of the time window due to the potential slow drift in the baseline absolute level before and after the contamination. These level step breaks will generate a large amount of high-frequency secondary noise in the frequency domain, still interfering with subsequent systems.

[0173] To achieve truly seamless morphological fusion, this embodiment constructs a set of morphological reconstruction formulas based on linear decay cross-gradient rules to calculate and generate smooth reconstruction data segments. The calculation formula can be expressed as:

[0174] Smoothly reconstruct data segments: ;

[0175] in, This indicates that the smoothed reconstructed data segment obtained after waveform replacement and boundary smoothing calculation is the [number]th [unit] segment within the current local time sliding window. Smooth reconstruction value for each relative index position; This indicates that the subset of reference data extracted from the pre-stored clean background waveform data segment according to the same phase index is in the 1st... The pure value at each relative index position; This represents the stable level value of the target mixed absorption spectrum at the previous normal sampling moment immediately preceding the blocking command, and is used as the replacement reference anchor point; Indicates local index The driving linear decay hybrid splicing coefficients. The value range of the hybrid splicing coefficients is strictly limited to between zero and one, when the index... When located at the starting boundary of the sliding window, its value approaches 1; when the index... As it moves toward the center of the sliding window, its value decreases linearly to 0.

[0176] Through this controlled proportional mixing operation, the waveform replacement process with the same frequency and bit width using the pure background waveform data segment can present an extremely smooth transition envelope, effectively suppressing the cutting and destruction of the macroscopic morphology of the absorption spectrum by transient high-energy pulses.

[0177] like Figure 6 The figure shows a comparison between the contaminated spectral waveforms and the smoothed reconstructed data segment shapes before and after the blocking command is triggered. The horizontal axis represents the sampling point order within the local time sliding window, and the vertical axis represents the absorption intensity amplitude, which characterizes the energy level.

[0178] The red curve in the figure represents the contaminated spectral waveform flowing into the system before the blocking command is triggered. It can be observed that in the local range of sampling points 100 to 150, the waveform is impacted by a transient high-energy pulse, and its absorption intensity amplitude suddenly increases to about 2.5. Furthermore, a significant level step discontinuity is generated after the pulse ends. If such abrupt peaks and non-smooth discontinuous contaminated data directly enters the filtering pipeline, it can easily affect the update of adaptive weights. The green curve represents the smooth reconstructed data segment obtained after the system responds to the blocking command and replaces it with a pre-stored clean background waveform.

[0179] By comparing the waveform transformations of the two, it can be seen that the green curve, within the same pulse range, completes the waveform replacement of the same frequency and bit width based on the linear attenuation cross-gradient rule. At the starting boundary of sampling point order 100 and the ending boundary of sampling point order 150, the green curve shows a relatively smooth envelope transition, and its absorption intensity amplitude is stably maintained at the normal background baseline level of about 0.5.

[0180] This morphological reconstruction process from the red distorted waveform to the green smooth waveform effectively suppresses the cutting effect of high-frequency pulse spikes and level faults on the macroscopic spectral morphology, indicating that the pre-adaptive blocking and reconstruction mechanism can maintain the physical continuity of signal energy in the time domain.

[0181] Finally, after completing the pollution isolation and waveform repair within the local window, the system uses the smoothed reconstructed data segment as input for subsequent noise reduction processing, and unfreezes the weight update process after the data processing of the current time sliding window is completed.

[0182] Optionally, the system's underlying data routing scheduler regains control of the data flow. The repaired time-series array, i.e., the smoothed reconstructed data segment, is seamlessly spliced ​​back onto the original data flow bus and delivered to subsequent noise reduction submodules that use digital filtering rules to filter out power frequency interference and eliminate DC bias components, according to normal pipeline rhythm. Due to the careful smoothing reconstruction, all subsequent low-frequency smoothing filters will not detect the extreme high-energy sudden impact that occurred previously, and the phase response and amplitude frequency response of the filters remain in the linear stable operating region.

[0183] Simultaneously, when the current time window experiencing a sudden disturbance moves out of the current evaluation field of view, and the short-term disturbance energy value of the next adjacent time window, after recalculation, falls back to the normal range, the system's main control logic cancels the aforementioned blocking command. The system resets the aforementioned frozen state variable to a constant of one, thereby unfreezing the weight update process of the adaptive filtering algorithm. The adaptive filter then restarts the error calculation and gradient descent optimization process from the historical breakpoint where it was last safely suspended.

[0184] In summary, the method in this embodiment not only completely resolves the technical pain points of traditional linear smoothing algorithms being powerless against nonlinear abrupt high-energy noise and contaminating the original data, but also ensures the physical continuity of time-domain energy through phase-synchronized waveform reconstruction. This provides a relatively pure and high-quality spectral data source for subsequent machine learning models to accurately identify the features of multiple interfering gases and reliably classify their priorities. As a result, it improves the anti-damage capability and detection accuracy stability of the laser methane detection equipment of this application under complex electromagnetic interference and harsh electromechanical operating conditions during long-term continuous operation.

[0185] Example 3:

[0186] This embodiment focuses on addressing a typical extreme physical interference phenomenon encountered by laser methane detection equipment during long-term operation in actual industrial settings.

[0187] In harsh environments such as underground coal mines, high-dust chemical plants, or underground pipe corridors with severe water vapor condensation, the outer surface of the equipment's permeable explosion-proof membrane or optical window often accumulates a large amount of coal dust particles, chemical dust, or dense condensation droplets over time. This physical deposit causes strong Mie scattering and non-selective broadband absorption of the detection laser penetrating the window, leading to extreme light intensity attenuation in the optical path system. When the energy of the beam reaching the photodetector is extremely weak, the signal-to-noise ratio of the detector's output electrical signal drops sharply, and the effective absorption signal of the target gas is compressed to near the detector's background noise level due to the lack of total optical energy. At this point, if the system still uses a static proportional calculation model based on ideal optical transmittance, the calculated concentration value will appear falsely low, leading to a serious risk of safety underreporting.

[0188] In order to establish a real-time monitoring and self-rescue mechanism for the health status of optical systems, this embodiment introduces a set of compensation logic based on full-band light intensity status assessment into the basic data calculation process.

[0189] This embodiment includes a nonlinear compensation step for the target optical path attenuation environment before the step of calculating the first ratio of the effective absorbed signal quantity to the stable fluctuation quantity. This nonlinear compensation step, as an independent physical dimension correction module, is connected in series after the residual spectrum closed-loop verification process and before the final concentration calculation process. The system main control chip quantifies and calibrates the degree of physical attenuation of the optical link in the current detection cycle by calling its internal light intensity state evaluation subroutine. Specifically:

[0190] In the first sub-step, the system's underlying driver obtains the real-time non-absorption reference light intensity value of the reference channel in the current detection cycle.

[0191] Specifically, in the dual-channel optical path architecture within the laser methane detection system, the detector in the reference channel is responsible not only for acquiring the characteristic absorption spectral lines used for wavelength locking but also for monitoring the absolute light intensity state of the laser's emission source. Within a complete wavelength scan modulation cycle of the semiconductor laser, the laser's output wavelength covers both the absorption peak band of the target gas and the baseline flat band, where there is no absorption by the target gas or background interfering gases. The system's data acquisition card continuously samples the voltage signal output by the photodetector in the reference channel within a specific time window of this baseline flat band at high frequency.

[0192] Subsequently, the microprocessor extracts the arithmetic mean of the voltage values ​​at all sampling points within the time window, and uses this arithmetic mean as an objective physical quantity characterizing the absolute transmission energy of the current physical optical path. This arithmetic mean is the real-time non-absorption reference light intensity value. By extracting the light intensity value in a band without gas absorption, the system can effectively eliminate the interference of gas concentration fluctuations on the light intensity evaluation results, thereby ensuring that the extracted energy value is only related to the physical transmittance of the optical lens and the background luminous power of the laser.

[0193] In the second sub-step, the system arithmetic unit executes a division instruction to calculate the light intensity attenuation ratio, which is the second ratio of the real-time non-absorption reference light intensity value to the preset clean initial reference light intensity value.

[0194] The preset clean initial reference light intensity value is the absolute light intensity reference data of the reference channel baseline flat band, obtained during the equipment's factory calibration phase in a standard dust-free, constant-vapor-free high and low temperature environment test chamber. This data is permanently stored in the non-volatile memory read-only sector of the system motherboard, serving as the absolute origin for evaluating the degree of optical health degradation throughout the equipment's life cycle. The system reads the value in this read-only sector and the real-time acquired value in the current cycle memory to construct an attenuation ratio calculation model, the complete formula of which can be expressed as:

[0195] Light intensity attenuation ratio ;

[0196] in, This represents the calculated light intensity attenuation ratio. This represents the real-time non-absorption reference light intensity value extracted by the system in the current detection cycle; This represents the clean initial reference light intensity value pre-stored in non-volatile memory.

[0197] The formula outputs a dimensionless percentage constant, objectively reflecting the proportion of residual energy of the laser probe beam after passing through the external physical environment. Due to the inevitable aging of optical components and the accumulation of dust in industrial environments, this light intensity attenuation ratio gradually decreases throughout the equipment's lifespan.

[0198] In the third sub-step, the system logic judgment unit determines whether the light intensity attenuation ratio is lower than the preset light intensity attenuation threshold.

[0199] For example, the optical attenuation process in industrial settings can be divided into a linear response region and a nonlinear collapse region. When light dust or trace amounts of moisture are present, the photodetector still operates within the linear range of its response curve. In this case, the conventional ratio cross-normalization algorithm can directly offset the overall decrease in light intensity. However, when the thickness of the deposit exceeds a certain physical limit, the number of photons reaching the detector decreases sharply. The quantization noise and thermal noise of the operational amplifier circuit and analog-to-digital converter within the system gradually become dominant, significantly reducing the applicability of the conventional linear normalization mathematical model.

[0200] To address this, system developers conducted numerous extreme environment resistance experiments to calibrate a critical percentage value characterizing the failure boundary of the linear model; this value is the preset light intensity attenuation threshold. The system's central processing unit uses logical comparison instructions to compare the currently calculated light intensity attenuation ratio with the preset light intensity attenuation threshold stored in a register. If the light intensity attenuation ratio is greater than or equal to the preset light intensity attenuation threshold, the system determines that the current optical environment is still within the safe linear control zone, and the algorithm proceeds according to the original normal process.

[0201] In the fourth sub-step, if the light intensity attenuation ratio is lower than the preset light intensity attenuation threshold, a dynamic compensation coefficient that has an inverse nonlinear mapping relationship with the light intensity attenuation ratio is calculated.

[0202] It is also important to note that once the condition of light intensity attenuation below the preset threshold is triggered, it indicates that the optical transmittance of the detection environment has decreased sharply. At this point, the absorption peaks in the spectral morphology not only shrink proportionally in amplitude, but also, due to the extreme deterioration of the signal-to-noise ratio, the proportion of the effective absorbed signal being swallowed up by the background noise nonlinearity increases dramatically. In order to digitally compensate for this implicit signal gain swallowed by noise at the mathematical level, the system abandons the simple linear multiplication amplification method and instead constructs a reverse nonlinear compensation mathematical model based on the exponential expansion mechanism. Its complete formula algorithm can be defined as:

[0203] Dynamic compensation coefficient: ;

[0204] in, This represents the calculated value of the dynamic compensation coefficient used for subsequent concentration amplification; This represents the preset nonlinear gain amplitude factor value obtained during the factory calibration stage, which is used to control the initial slope of the exponential expansion function. This represents the value of the preset exponential decay rate factor, which controls the compensation ratio to increase more rapidly as the light intensity deteriorates. This represents the preset light intensity attenuation threshold value stored in the register; This represents the light intensity attenuation ratio obtained from the aforementioned calculation; This represents the value of the basic compensation offset constant that ensures the continuity of compensation at the critical trigger point.

[0205] Through this inverse nonlinear mapping relationship, the lower the light intensity attenuation ratio, i.e., the more severe the optical environment, the larger the difference within the brackets, and the output of the exponential function will exhibit a geometrical growth trend. This nonlinear boosting mechanism can accurately offset the nonlinear drop in detector responsivity under extremely low light intensity, ensuring the physical scientific basis of the compensation force.

[0206] like Figure 7 This figure displays a data graph showing the mapping relationship between the light intensity attenuation ratio and the nonlinear dynamic compensation coefficient under extreme optical path attenuation conditions. The horizontal axis represents the light intensity attenuation ratio, which gradually decreases from one to zero from right to left to match the physical energy attenuation process. The vertical axis represents the dynamic compensation coefficient output by the system.

[0207] The solid blue dots in the figure represent discrete calibration data points obtained in the simulated environment attenuation experiment, the dashed green lines represent the conventional linear normalized compensation curve, and the solid red lines represent the constructed nonlinear exponential dynamic compensation fitting curve.

[0208] The evolution of the curves reveals that when the light intensity attenuation ratio on the horizontal axis is between 1 and 0.3, the system's optical path is in the linear response region, with the red solid line and the green dashed line maintaining a smooth and close compensation level. When the light intensity attenuation ratio falls below the preset light intensity attenuation threshold marked with a vertical dividing line (0.3), it indicates a significant decrease in optical transmittance. At this point, the signal gain provided by the green conventional linear curve is relatively limited, while the red nonlinear fitting curve triggers the inverse mapping compensation mechanism, exhibiting an exponential expansion characteristic that pulls upwards to the left. When the attenuation ratio approaches 0.05, the red solid line fits the densely distributed blue experimental calibration data points well and outputs a dynamic compensation coefficient of approximately 15.

[0209] This curve transformation from a gradual transition to an exponential increase demonstrates that the method can offset the drop in photodetector responsivity under conditions of weak probe beam energy through a low-level nonlinear mathematical mapping model, which helps to restore the effective absorbed signal and improve the measurement stability of the equipment in complex operating environments.

[0210] In the fifth sub-step, the system calculation unit uses the product of the dynamic compensation coefficient and the first ratio to replace the original first ratio and input it into the methane concentration calibration function to calculate the corrected actual value of the methane concentration.

[0211] Specifically, the pre-processor has extracted the effective absorption signal representing methane gas from the residual spectrum and the stable fluctuation of the reference channel in a standard environment, and divided the two to obtain the first ratio representing the relative absorption intensity. Under normal operating conditions, this first ratio is directly input as an independent variable into the quadratic polynomial concentration calibration function. However, after triggering the extreme attenuation nonlinear compensation mechanism in this embodiment, the system reallocates registers in memory space and performs a floating-point multiplication operation on the dynamic compensation coefficient calculated in the fourth sub-step and the extracted first ratio. This product physically represents the virtual ideal absorption ratio after reconstruction with extreme optical attenuation resistance.

[0212] Subsequently, the system treats this product as a new independent variable and substitutes it into the system's methane concentration calibration function for evaluation. The complete modified concentration calculation formula can be expressed as:

[0213] ;

[0214] in, This represents the corrected actual value of the methane concentration output after reconstruction and calculation using a nonlinear compensation mechanism. This represents the curvature calibration constant of the quadratic curve in the calibration function; This indicates the value of the dynamic compensation coefficient calculated and issued in the current detection cycle; This represents the first ratio value extracted from the original spectral form before compensation; This represents the linear slope calibration constant in the calibration function; This represents the zero-point baseline bias constant in the calibration function.

[0215] By deeply integrating dynamic compensation coefficients into the characteristic variables of each order in the concentration calibration polynomial, the system can, in the final stage of the algorithm calculation, rely on closed-loop iteration of the pure data link to reconstruct and restore weak signal features that were originally obscured by physical blockage from heavy dust or severe water vapor condensation with high confidence. This nonlinear compensation mechanism effectively prevents the safety hazard of concentration monitoring values ​​collapsing to zero due to drastic deterioration of the environment outside the breathable membrane, and effectively reduces false alarm blind spots in industrial safety monitoring systems.

[0216] This significant expansion of the technical solution not only extends the operation cycle of laser methane detectors in harsh working environments without manual maintenance and cleaning, but also, in the face of the objective physical law of irreversible decay of physical optical components, finds the balance point of measurement accuracy through a rigorous nonlinear mathematical mapping algorithm, achieving a high standard of industrial practicality.

[0217] Example 4:

[0218] like Figure 8 As shown, this embodiment provides a line-by-line calculation and subtraction system for water vapor spectral line overlap interference in laser methane detection. This embodiment serves as a system-level mapping of the aforementioned method embodiments at the physical entity and logical architecture levels, focusing on the hardware support and underlying operating mechanisms of the functional modules in practical industrial applications.

[0219] In response to the technical problems mentioned in the background art, such as the drastic fluctuation of environmental parameters, severe interference from water vapor and volatile organic compounds in industrial sites, such as underground coal mines, chemical industrial parks, and underground pipe corridors, as well as the insufficient ability of traditional static subtraction models to cope with complex dynamic interference, the system disclosed in this embodiment provides an engineering solution with on-site deployability and high environmental adaptability through high-level hardware and software synergy.

[0220] Specifically, this embodiment discloses a line-by-line calculation and subtraction system for water vapor spectral line overlap interference in laser methane detection, which is applied in a laser methane detection device including a dual-channel detection unit and an environmental parameter sensor.

[0221] In practical industrial applications, this laser methane detection equipment is typically enclosed in an explosion-proof enclosure that meets protection requirements to accommodate hazardous work areas containing flammable gases. The dual-channel detection unit physically includes a tunable semiconductor laser, a beam splitter, an open or multiple-reflection measuring gas chamber, a closed reference gas chamber, and a corresponding photodetector. Environmental parameter sensors include a high-precision platinum resistance temperature sensor, a microelectromechanical system (MEMS) absolute pressure sensor, and an industrial-grade capacitive humidity sensor, deployed at or inside the measuring gas chamber inlet. The system is internally equipped with an embedded computing platform containing a central processing unit (CPU), a digital signal processor (DSP), or a field-programmable gate array (FPGA), as well as non-volatile memory and random access memory (RAM). All functional modules described below run on this embedded computing platform in the form of firmware code or hardware logic circuits.

[0222] The system specifically includes the following functional modules:

[0223] First, the data collaborative acquisition module is used to simultaneously acquire the target mixed absorption spectrum and reference signal of the gas to be tested through the measurement channel and reference channel of the dual-channel detection unit, and to acquire environmental parameters through the environmental parameter sensor.

[0224] In actual equipment operation, this module relies on multi-channel synchronous analog-to-digital converter (ADC) hardware. When the laser receives the periodic scanning current sent by the drive circuit, the ADC synchronously converts the weak analog voltage signal containing background interference output by the photodetector of the measurement channel, as well as the stable analog voltage signal output by the photodetector of the reference channel, into discrete digital quantities at a set high-frequency sampling rate. Simultaneously, the module reads the current values ​​of the temperature, pressure, and humidity sensors via a communication bus (such as I2C or SPI protocol). All acquired spectral data and environmental parameters are assigned a unified hardware timestamp in time sequence, packaged and stored in a designated data buffer of random access memory, forming a multi-dimensional spectral-environmental collaborative sample, providing a time-consistent underlying data source for subsequent decoupling algorithms.

[0225] Second, the interference feature identification module is used to input the target mixed absorption spectrum and the environmental parameters into a preset machine learning model, and combine it with a physical feature library to determine the existing interfering gases and their corresponding concentration estimates. In terms of hardware resource allocation, the physical feature library contains standard absorption cross-section data of common interfering gases such as water vapor, acetic acid, ethanol, and propane under different temperature and pressure conditions, which are pre-programmed into the device's non-volatile memory (such as a Flash chip). The preset machine learning model, as a lightweight multivariate nonlinear regression and classification algorithm, runs in a digital signal processor. This module extracts spectral features from the data buffer after basic noise reduction processing (such as filtering out transient high-energy pulse interference caused by the start-up and shutdown of large electromechanical equipment mentioned in the background technology), and combines this with the currently read actual operating parameters such as temperature and pressure to drive the machine learning model to perform forward inference calculations. The model output is used to identify the specific type of interfering gas intruding into the current sampling space and provides a rough concentration order of magnitude (i.e., a concentration estimate) for subsequent weight allocation.

[0226] Third, the weight and priority division module is used to calculate the interference weight coefficient of each interfering gas corresponding to the spectral line based on the concentration estimate and the preset interference weight calculation rules, and divide the corresponding spectral line into multiple priorities according to the interference weight coefficient.

[0227] In the actual system operation process, due to the limited computing resources of the embedded platform in the industrial field, this module acts as the core of system computing power scheduling. The processor's computing unit performs weighted matrix multiplication based on parameters such as the cross-over overlap of each interfering gas and the normalized value of the concentration estimate, calculating numerical interference weight coefficients. Subsequently, a logic comparator compares these coefficients with pre-existing multi-level thresholds in the system, allocating the water vapor or specific concentration organic spectral lines that have the greatest impact on methane measurement to the high-priority calculation queue, and allocating less influential edge interference to the medium-priority or low-priority calculation queues. This hierarchical mechanism ensures that computing resources are tilted towards the most critical interference sources within a limited processing cycle.

[0228] Fourth, the line-by-line fitting subtraction module is used to sequentially subtract the line-by-line absorption contribution of each interfering gas in the target mixed absorption spectrum according to the preset order and fitting algorithm corresponding to the priority, so as to obtain the residual spectrum after subtracting interference. This module is the core computing power consumption node of the entire system, and is usually executed by a microprocessor with a hardware floating-point operation acceleration unit.

[0229] According to the instructions of the priority queue, for interfering spectral lines in the high-priority queue, this module retrieves basic parameters from the physical feature library, combines them with the current temperature and pressure, and uses a joint iterative algorithm and mixed line shapes, such as the Voigt line shape, to perform high-precision nonlinear fitting and generate the interference. It then performs point-to-point matrix subtraction from the target mixed absorption spectrum data array. For low-priority interference, a lookup table method is used for rapid matching and subtraction. After this progressive digital stripping operation, the originally overlapping and mixed spectral data is pruned, outputting a residual spectral data block that theoretically retains only the characteristic absorption of methane.

[0230] Fifth, the closed-loop verification and calculation module is used to verify the residual spectrum. If the verification is successful, the actual methane concentration is calculated based on the residual spectrum. The closed-loop verification and calculation module is also used to reverse-adjust the decoupling model parameters used to calculate the absorption contribution when the residual spectrum fails the verification process, and to trigger the line-by-line fitting subtraction module to re-execute the line-by-line absorption contribution subtraction step based on the adjusted decoupling model parameters to obtain a verified residual spectrum.

[0231] In engineering implementation, this module constitutes the system's self-diagnosis and self-correction defense line. The processor performs morphological verification on the residual spectrum generated after subtraction, such as calculating its cosine similarity to the standard pure methane spectrum. If the system's optical lenses are heavily covered by dust or experience severe condensation, resulting in non-selective broadband absorption and causing light intensity attenuation beyond the normal linear range, the previous subtraction fitting will be distorted or over-subtracted, and the residual spectrum will fail the similarity or residual intensity threshold verification. In this case, the system's main control state machine will not output an erroneous concentration alarm, but will instead generate an internal feedback interrupt signal. This signal triggers the system to modify the decoupling model parameters located in memory, such as adjusting the broadening coefficient compensation factor or calling the dynamic compensation gain coefficient for extremely low light intensity, and instructs the line-by-line fitting subtraction module to clear the current cache and re-perform the matrix subtraction operation using the updated parameters. This closed-loop iterative process continues until the quality of the generated residual spectrum meets stringent preset standards. Finally, the system uses the high-fidelity residual spectrum effective signal quantity, substitutes it into the concentration calibration function, calculates the highly accurate actual value of methane concentration, and sends the safety data to the host computer or distributed control system through digital-to-analog conversion or industrial fieldbus (such as RS485, CAN bus, etc. interface).

[0232] In summary, the system disclosed in this embodiment overcomes, to some extent, the poor environmental adaptability of traditional instruments that employ fixed filtering and static calibration by solidifying the highly complex line-by-line calculation subtraction algorithm into multi-level collaborative hardware logic and software modules. Relying on pipelined operation between modules and a strict closed-loop feedback mechanism, the system effectively mitigates the uncontrollable risks to signal processing caused by dynamic nonlinear interference and physical optical path attenuation, thereby improving the monitoring accuracy and false alarm stability of related equipment in complex, high-interference application scenarios.

[0233] Example 5:

[0234] Corresponding to the above embodiments, the present invention also proposes an electronic device.

[0235] like Figure 9 The diagram shows a structural schematic of an electronic device according to the present invention. The electronic device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the electronic device 100 may further include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one unit, and the structure of this electronic device 100 does not constitute a limitation on the embodiments of the present invention.

[0236] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0237] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0238] The memory 103 stores a computer program corresponding to the line-by-line calculation and subtraction method for water vapor spectral line overlap interference in laser methane detection according to the above embodiments of the present invention. This computer program is controlled and executed by the processor 101. The processor 101 executes the computer program stored in the memory 103 to implement the content shown in the aforementioned method embodiments.

[0239] Among them, electronic devices 100 include, but are not limited to: mobile terminals such as laptops and PADs (tablet computers) and fixed terminals such as desktop computers. Figure 9 The electronic device 100 shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0240] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for line-by-line calculation and subtraction of water vapor spectral line overlap interference in laser methane detection, characterized in that, The application in a laser methane detection system, which includes a dual-channel detection unit and an environmental parameter sensor, includes the following steps: The target mixed absorption spectrum and reference signal of the gas to be tested are simultaneously acquired through the measurement channel and reference channel of the dual-channel detection unit, and environmental parameters are obtained through the environmental parameter sensor. The target mixed absorption spectrum and the environmental parameters are input into a preset machine learning model, and the interfering gases and their corresponding concentration estimates are determined by combining the physical feature library. Based on the concentration estimate and the preset interference weight calculation rules, the interference weight coefficient of each interfering gas corresponding spectral line is calculated, and the corresponding spectral lines are divided into multiple priorities according to the interference weight coefficient. According to the preset order and fitting algorithm corresponding to the priority, the line-by-line absorption contribution of each interfering gas in the target mixed absorption spectrum is deducted in turn to obtain the residual spectrum after deducting interference. The residual spectrum is verified. If the verification is successful, the actual methane concentration is calculated based on the residual spectrum. If the residual spectrum fails the verification process, the decoupling model parameters used to calculate the absorption contribution are adjusted in reverse, and the line-by-line absorption contribution subtraction step is re-executed based on the adjusted decoupling model parameters to obtain a verified residual spectrum.

2. The method according to claim 1, characterized in that, The acquisition of the target mixed absorption spectrum and reference signal of the gas to be tested includes: The target mixed absorption spectrum is acquired through the measurement channel in a preset characteristic absorption band; The reference signal without interference is acquired through the fixed band of the reference channel, and the characteristic spectral line signal of the target interfering gas is acquired through the extended band of the reference channel.

3. The method according to claim 2, characterized in that, After acquiring the target mixed absorption spectrum and reference signal of the gas to be tested, the method further includes a step of progressive noise reduction processing on the acquired signal data: Digital filtering rules are used to filter out power frequency interference and harmonic components in signal data; The DC bias component in the signal data is eliminated by the arithmetic averaging algorithm. Low-frequency trend interference in signal data is extracted and eliminated using a cumulative moving average algorithm; Baseline drift is corrected by combining wavelet transform rules, and light intensity fluctuations are corrected by an adaptive filtering algorithm.

4. The method according to claim 3, characterized in that, Before using digital filtering rules to filter out power frequency interference and harmonic components in the signal data, an adaptive blocking and reconstruction step is also included: Construct a time sliding window containing a preset number of data points, and calculate the short-time abrupt energy value of the target mixed absorption spectrum within the current time sliding window; Determine whether the short-term mutation energy value is greater than the dynamic energy threshold, wherein the dynamic energy threshold is calculated based on the background energy benchmark value within a historical period without interference; If the short-term mutation energy value is greater than the dynamic energy threshold, a blocking instruction is generated to freeze the subsequent weight update process of the adaptive filtering algorithm. In response to the blocking command, a pre-stored clean background waveform data segment is extracted, and the target mixed absorption spectrum within the current time sliding window is replaced with a waveform of the same frequency and bit width using the clean background waveform data segment to obtain a smooth reconstructed data segment. The smoothed reconstructed data segment is used as input for subsequent noise reduction processing, and the weight update process is unfrozen after the data processing of the current time sliding window is completed.

5. The method according to claim 1, characterized in that, The step of inputting the target mixed absorption spectrum and the environmental parameters into a preset machine learning model, and combining it with a physical feature library to determine the interfering gases and their corresponding concentration estimates, includes: Extract spectral line morphology feature vectors from the target mixed absorption spectrum; The spectral morphology feature vector and the environmental parameters are input into the preset machine learning model, which outputs the presence identifier of the interfering gas, the concentration estimate, and the cross-over degree of the corresponding spectral lines.

6. The method according to claim 5, characterized in that, The calculation of interference weight coefficients for the corresponding spectral lines of each interfering gas, and the division of the corresponding spectral lines into multiple priorities according to the interference weight coefficients, includes: For any interfering gas spectral line, obtain its first degree of overlap with the methane spectral line, the normalized value of the concentration estimate, and the second degree of overlap with other interfering gas spectral lines. The interference weight coefficient is obtained by calculating the weighted sum of the first overlap, the normalized value, and the second overlap. Determine the preset value range in which the interference weight coefficient is located, and divide the corresponding spectral lines into high priority, medium priority, or low priority respectively.

7. The method according to claim 6, characterized in that, The step of deducting the line-by-line absorption contribution of each interfering gas in the target mixed absorption spectrum according to the preset order and fitting algorithm corresponding to the priority includes: The order of deduction is determined according to the descending order of the interference weight coefficients; For the spectral lines classified as high priority, a joint iterative algorithm incorporating multiple environmental parameter variables and a hybrid linear convolution algorithm are used for subtraction. For spectral lines classified as medium priority, a partial iterative algorithm with single environmental parameter variables and a Gaussian approximate convolution algorithm are used for subtraction. For spectral lines classified as low priority, the overall fitting subtraction is performed by directly calling the pre-calculated standard values ​​in the physical feature library.

8. The method according to claim 1, characterized in that, The verification process for the residual spectrum includes determining whether the following verification conditions are met simultaneously: The cosine similarity between the residual spectrum and the standard pure methane spectrum is greater than or equal to a preset similarity threshold. The residual absorption intensity of each interfering gas at its characteristic wavelength in the residual spectrum is less than or equal to a preset absorption intensity threshold. The deviation between the methane absorption coefficient after interference removal and the theoretical absorption coefficient under the current environmental parameters is within the preset allowable deviation range; If all the above verification conditions are met, the residual spectrum is determined to have passed the verification process.

9. The method according to claim 1, characterized in that, The calculation of the actual methane concentration based on the residual spectrum includes: Obtain the effective absorption signal quantity characterizing methane in the residual spectrum, and the stable fluctuation quantity of the reference channel in a standard environment; Calculate the first ratio of the effective absorbed signal quantity to the stable fluctuation quantity; The first ratio is input into a pre-set methane concentration calibration function to calculate the actual methane concentration value.

10. The method according to claim 9, characterized in that, Before the step of calculating the first ratio of the effective absorbed signal quantity to the stable fluctuation quantity, a nonlinear compensation step is also included: Obtain the real-time non-absorption reference light intensity value of the reference channel in the current detection cycle; Calculate the light intensity attenuation ratio, which is a second ratio of the real-time non-absorption reference light intensity value to the preset clean initial reference light intensity value; Determine whether the light intensity attenuation ratio is lower than a preset light intensity attenuation threshold; If the light intensity attenuation ratio is lower than the preset light intensity attenuation threshold, then a dynamic compensation coefficient that has an inverse nonlinear mapping relationship with the light intensity attenuation ratio is calculated. The product of the dynamic compensation coefficient and the first ratio is used to replace the original first ratio and input into the methane concentration calibration function to calculate the corrected actual value of the methane concentration.

Images