A dual-channel differential calibration method for resisting optical window pollution decay of gas detection

Abstract

The present application relates to laser gas measurement calibration technical field, specifically disclose a kind of gas detection anti optical window pollution decay dual-channel differential calibration method, the method comprises: constructing dual-channel optical measurement system, through main measurement channel acquisition through the main measurement optical signal of measured medium cavity, and through reference channel enclosed in clean gas cavity acquisition only affected by optical window pollution influence pure pollution reference optical signal;The main measurement optical signal and pure pollution reference optical signal both synchronously include the direct transmission light signal of first preset angle and the forward scattering light signal of second preset angle interval.This application is by constructing reference channel isolated measured medium and main measurement channel crossing measured medium, in actual industrial online monitoring process simultaneously extracting the characteristics of direct transmission light and forward scattering light, realizes the physical isolation and synchronous acquisition of pure pollution interference signal and main measurement mixed signal from bottom architecture.

Description

Technical Field

[0001] This invention relates to the field of laser gas measurement and calibration technology, specifically to a dual-channel differential calibration method for gas detection that resists optical window contamination and decay. Background Technology

[0002] In industrial process control and environmental monitoring, optical measurement systems are widely used for online analysis of the characteristics of the measured medium. These systems typically rely on optical windows for physical isolation from the complex measured fluid. However, under harsh operating conditions, particulate matter, oil, or salt spray in the fluid inevitably adhere to the surface of the optical window, causing contamination. Over time, this accumulated window contamination not only blocks light transmission but also causes scattering and refraction of the light signal, resulting in severe distortion of the original light signal received by the detector. This directly compromises the accuracy and reliability of the entire measurement system.

[0003] Specifically, traditional conventional algorithm compensation schemes typically treat pollution as a simple linear overall attenuation of light intensity, using only fixed coefficients for static correction. However, when pollution accumulates to a certain level in industrial settings and causes strong light scattering, the light signal will exhibit complex baseline dynamic drift and asymmetric peak distortion.

[0004] At this point, to accurately decouple this nonlinear distortion, the existing single-channel measurement architecture or static compensation model must interrupt the measurement process and introduce standard gas or perform complex offline recalibration manually; however, if online continuous measurement is maintained, it is difficult to effectively remove these complex nonlinear scattering interferences, resulting in severe distortion of instrument output or even direct failure. Summary of the Invention

[0005] 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 dual-channel differential calibration method for gas detection that resists optical window contamination and decay, thereby ensuring high accuracy in online monitoring under long-term, harsh industrial environments.

[0006] To achieve the above objectives, a first aspect of the present invention provides a dual-channel differential calibration method for gas detection to resist optical window contamination and decay, comprising the following steps:

[0007] A dual-channel optical measurement system is constructed. The main measurement channel acquires the main measurement optical signal passing through the cavity of the measured medium, and the reference channel, which is enclosed in a clean gas cavity, acquires the pure contamination reference optical signal that is only affected by the contamination of the optical window. Both the main measurement optical signal and the pure contamination reference optical signal synchronously contain the direct transmission light signal at a first preset angle and the forward scattered light signal in a second preset angle range.

[0008] Anti-interference features are extracted from the main measurement optical signal and the pure contamination reference optical signal, the real-time scattering feature vector is calculated, and the current contamination stage and contamination type of the optical window are determined based on the pre-established three-dimensional correlation benchmark model that includes the window transmittance attenuation, scattering feature distribution and signal distortion mapping relationship.

[0009] When it is determined that it is in the linear attenuation stage, a differential and ratio dual-dimensional calibration model is constructed based on the characteristic peak intensity of the direct transmitted light of the main measurement channel and the reference channel, and linear amplitude attenuation compensation is performed on the main measurement light signal.

[0010] When it is determined that it is in the nonlinear scattering stage, the real-time scattering feature vector is input into the three-dimensional associated reference model, and a pure pollution distortion substrate is generated by decoupling. The main measurement optical signal is then compensated for scattering distortion in all dimensions, including baseline drift and peak distortion, and the true feature signal of the measured medium is reconstructed.

[0011] A two-dimensional validity check is performed on the compensated real feature signal. If the check passes, the final measurement result is output. If the check fails, the parameter adaptive iteration process of the three-dimensional associated benchmark model is triggered, and the compensation and check are re-executed after the parameters are updated.

[0012] To achieve the above objectives, a second aspect of the present invention provides a dual-channel differential calibration system for gas detection resistant to optical window contamination and decay, the system comprising:

[0013] The dual-channel optical signal acquisition module is used to acquire the main measurement optical signal passing through the cavity of the measured medium through the main measurement channel, and to acquire the pure contamination reference optical signal that is only affected by the contamination of the optical window through the reference channel enclosed in the clean gas cavity; the main measurement optical signal and the pure contamination reference optical signal both synchronously contain the direct transmission light signal at a first preset angle and the forward scattered light signal in a second preset angle range.

[0014] The feature extraction and stage determination module is used to extract anti-interference features from the acquired main measurement optical signal and the pure pollution reference optical signal, calculate the real-time scattering feature vector, and determine the current pollution stage and pollution type of the optical window based on a pre-established three-dimensional correlation benchmark model that includes the window transmittance attenuation, scattering feature distribution and signal distortion mapping relationship.

[0015] The linear attenuation compensation module is used to perform linear amplitude attenuation compensation on the main measurement optical signal when the feature extraction and stage determination module determines that it is in the linear attenuation stage, based on the characteristic peak intensity of the direct transmitted light of the main measurement channel and the reference channel, to construct a differential and ratio dual-dimensional calibration model.

[0016] The nonlinear distortion compensation module is used to input the real-time scattering feature vector into the three-dimensional associated reference model when the feature extraction and stage determination module determines that it is in the nonlinear scattering stage, decouple and generate a pure pollution distortion substrate, perform full-dimensional compensation for scattering distortion including baseline drift and peak distortion on the main measurement optical signal, and reconstruct the true feature signal of the measured medium.

[0017] The verification and adaptive iteration module is used to perform a two-dimensional validity verification on the compensated real feature signal. If the verification passes, the final measurement result is output. If the verification fails, the parameter adaptive iteration process of the three-dimensional associated benchmark model is triggered, and the corresponding compensation module is controlled to re-execute the compensation and verification after the parameters are updated.

[0018] 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 dual-channel differential calibration method for gas detection to resist optical window contamination decay.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0020] This invention constructs a reference channel isolating the measured medium and a main measurement channel traversing the measured medium. During actual industrial online monitoring, it simultaneously extracts the characteristics of direct transmitted light and forward scattered light, achieving physical isolation and synchronous acquisition of pure contamination interference signals and the mixed main measurement signal from the underlying architecture. Based on this, the solution utilizes a three-dimensional correlation benchmark model to analyze the contamination evolution state of the on-site optical window in real time. This not only enables accurate ratio and differential compensation for linear attenuation in the early stages of contamination but also dynamically decouples the peak distortion and baseline drift caused by pure contamination when complex nonlinear scattering occurs under harsh operating conditions, thereby reconstructing the true measured medium signal in reverse.

[0021] This phased, multi-dimensional compensation mechanism based on real-time scattering characteristics, combined with an adaptive iterative closed loop that is automatically triggered when the verification fails, enables the optical measurement system to intelligently and continuously resist the dynamic decay of the optical window without frequently interrupting the production process for manual cleaning or offline calibration. This effectively ensures high accuracy and maintenance-free operation of online monitoring in harsh industrial environments over long periods. Attached Figure Description

[0022] 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:

[0023] Figure 1This is a flowchart illustrating the dual-channel differential calibration method for gas detection to resist optical window contamination and decay provided by the present invention.

[0024] Figure 2 This is a comparison data diagram of optical signal preprocessing based on adaptive weighted least squares method and interpolation reconstruction in the dual-channel differential calibration method for gas detection to resist optical window contamination and decay provided by the present invention;

[0025] Figure 3 This is a fitted surface plot of the three-dimensional correlation benchmark model for window contamination decay in the dual-channel differential calibration method for gas detection anti-optical window contamination decay provided by the present invention;

[0026] Figure 4 This is a schematic diagram of the decoupling and reconstruction of the true features of the nonlinear strong scattering stage distortion signal in the dual-channel differential calibration method for gas detection against optical window contamination and decay provided by the present invention.

[0027] Figure 5 This is a time-series diagram of the dynamic variance threshold adaptive tracking based on feature blank background noise perception in the dual-channel differential calibration method for gas detection to resist optical window contamination and decay provided by the present invention.

[0028] Figure 6 This invention provides a dual-channel differential calibration method for gas detection to resist optical window contamination and decay, which includes a two-dimensional pixel light intensity matrix and spatial feature gradient thermogram of the detector under island-like local contamination.

[0029] Figure 7 This is a schematic diagram of the dynamic spatial weight mask generation and weighted mapping matrix for resisting local contamination interference in the dual-channel differential calibration method for gas detection against optical window contamination decay provided by the present invention.

[0030] Figure 8 This is a schematic diagram illustrating the implementation of the dual-channel differential calibration system for gas detection that resists optical window contamination and decay provided by the present invention;

[0031] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0032] 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.

[0033] The following describes, with reference to the accompanying drawings, a dual-channel differential calibration method, system, and electronic device for gas detection resistant to optical window contamination and decay according to embodiments of the present invention.

[0034] Example 1:

[0035] In various industrial settings, such as continuous emission monitoring systems (CEMS) in coal-fired power plants, online gas component analyzers in chemical reactors, and material spectral detection equipment in high-dust environments, optical analysis instruments must be exposed for extended periods to environments containing high concentrations of particulate matter, corrosive gases, high humidity, and complex oil mist aerosols.

[0036] To address the aforementioned issues, this embodiment provides a dual-channel differential calibration method for gas detection to resist optical window contamination and decay. This method aims to solve the technical problem of signal distortion, loss of accuracy, and damage to online continuity caused by window contamination in optical measurement equipment under harsh industrial conditions.

[0037] This embodiment first constructs a dual-channel optical measurement system through a specific hardware layout. The specific structural configuration of this dual-channel optical measurement system includes: the main measurement channel and the reference channel adopt a parallel coaxial optical path design, and a beam splitter is used to split the beams from the same source light source. Here, the dual-channel optical measurement system refers to a detection system integrated within the same device housing, sharing the underlying optical emission source and data processing core module. The same source light source here refers to a single light-emitting element that outputs a broadband spectrum to ensure that the two beams subsequently separated have a high degree of consistency in initial spectral energy distribution, wavelength drift characteristics, and initial phase. The beam splitter is positioned after the light-emitting element and is used to separate the beams emitted from the same source light source into two parallel detection beams according to a fixed energy ratio.

[0038] Specifically, such as Figure 1 As shown, a dual-channel differential calibration method for gas detection to resist optical window contamination decay includes the following steps:

[0039] For example, a main measurement optical signal passing through the cavity of the measured medium is acquired through a main measurement channel, while a pure contamination reference optical signal, affected only by optical window contamination, is acquired through a reference channel enclosed in a clean gas cavity. Here, the main measurement channel is the detection path in the optical system that directly contacts the substance to be measured in the industrial environment. The cavity of the measured medium refers to the space inside an industrial pipeline, flue, or reaction vessel filled with the fluid to be measured and contaminants. The main measurement optical signal is a mixed signal that, after the light beam penetrates the cavity of the measured medium, carries both the optical absorption characteristics of the measured medium and the optical attenuation and scattering information caused by contaminants on the surface of the optical window.

[0040] To achieve environmental isolation, the optical propagation path of the reference channel is entirely enclosed within a mechanically sealed structure, known as the clean gas chamber. This clean gas chamber is continuously circulated with high-purity nitrogen or zero gas to maintain positive pressure, ensuring that particulate matter and aerosols within the measured medium chamber cannot enter the reference channel. However, the optical window of the reference channel is physically adjacent to the optical window of the main measurement channel and is exposed to the boundary environment of the measured medium chamber. Therefore, contaminants will adhere to the outer surface of the optical window of the reference channel at the same deposition rate and morphology. The pure contamination reference light signal here refers to the signal that, after the probe beam penetrates this closed path, does not carry any absorption information from the measured medium but only reflects the optical characteristics caused by contaminants deposited on the outer surface of the optical window.

[0041] It is important to note that, regardless of whether it is the main measurement channel or the reference channel, the photon propagation path of the light beam changes when it penetrates the contaminated optical window. Therefore, both the main measurement light signal and the pure contaminated reference light signal simultaneously include a direct transmission light signal at a first preset angle and a forward-scattered light signal within a second preset angle range. The first preset angle refers to the direction where the angle between the light propagation direction and the detector center normal is 0 degrees. The light signal in this direction represents the set of photons that have not been deflected and have directly penetrated the optical window and the measured medium, i.e., the direct transmission light signal. The second preset angle range refers to the spatial conical surface range within which the angle between the light propagation direction and the detector center normal is greater than or equal to 2 degrees and less than or equal to 5 degrees. The light signal within this range represents the set of photons that have been deflected at a small angle due to Mie scattering or Rayleigh scattering when the light passes through particles attached to the surface of the optical window, i.e., the forward-scattered light signal.

[0042] In actual operation, the direct transmitted light signal mainly reflects the overall energy attenuation of the optical system and the macroscopic absorption characteristics of the measured medium, while the forward scattered light signal is rich in microscopic physical information such as the particle size distribution, morphology and aggregation state of pollutants.

[0043] Specifically, to acquire light signals from different angles simultaneously and separately, the system employs an area array photodetector to simultaneously acquire two signals. The photosensitive area of ​​the area array photodetector is divided into a direct light acquisition area at the center and an annular scattered light acquisition area on the periphery. Here, an area array photodetector refers to an image sensor device composed of multiple photosensitive pixels with independent photoelectric conversion capabilities arranged in a two-dimensional matrix. The photosensitive area refers to the physical region of the entire two-dimensional pixel matrix capable of generating an effective charge response to incident photons. The direct light acquisition area refers to a set of pixels located at the geometric center of this photosensitive area, specifically used to capture photons at a first preset angle. The annular scattered light acquisition area refers to a set of pixels arranged in concentric rings around the direct light acquisition area, specifically used to capture photons within a second preset angle range.

[0044] The direct light acquisition area, constrained by an aperture stop, collects only the direct transmitted light signal at the first preset angle. The annular scattered light acquisition area, constrained by an annular stop and an angle constraint stop, collects only the forward scattered light signal within the second preset angle range. The aperture stop is positioned in the front optical path of the area array photodetector, with a centrally located, fixed-diameter, transparent circular aperture to physically block all stray light deviating from the zero-degree propagation direction from entering the direct light acquisition area. The annular stop is a mechanical light-blocking plate with an opaque center and annular slits at the edges. The angle constraint stop is an optical component composed of light-blocking blades with specific tilt angles. Through the series connection of the annular stop and the angle constraint stop, not only is the central direct light physically shielded, but also large-angle scattered light with deflection angles exceeding the second preset angle range is filtered out, thus ensuring that the light signal received by the annular scattered light acquisition area is strictly limited to the forward scattering characteristic range.

[0045] For example, anti-interference feature extraction is performed on the acquired main measurement optical signal and the pure pollution reference optical signal to calculate the real-time scattering feature vector. Here, anti-interference feature extraction refers to the calculation process of extracting core parameters with physical significance characterizing the pollution state from the raw photoelectric conversion data mixed with industrial site environmental background noise, detector thermal noise, and high-frequency electromagnetic interference, using statistical and digital signal processing techniques to remove useless information.

[0046] To achieve this goal, the process of extracting anti-interference features from the acquired main measurement optical signal and the pure contaminated reference optical signal, and calculating the real-time scattering feature vector, specifically includes the following steps: synchronously acquiring optical signals for multiple consecutive cycles at a set sampling frequency; calculating the variance of single-cycle data at each sampling point relative to the average value of multiple cycles using the point-by-point variance statistical method; eliminating abnormal data groups with variances greater than a preset variance threshold; and extracting the average signal of the valid data groups. Here, the set sampling frequency refers to the number of times the system's analog-to-digital converter performs digital quantization of the photoelectric signal per second, while the cycle refers to a complete working time slice of the light source modulation. The synchronous acquisition requires the system's internal hardware trigger bus to strictly ensure that the main measurement channel and the reference channel complete photocharge integration and reading within the same microsecond-level time window. The point-by-point variance statistical method is an algorithm that evaluates the degree of data dispersion in both time series and spectral series dimensions.

[0047] The system first caches continuously collected data from multiple periods into a three-dimensional array and calculates the mean of the batch of data at each specific wavelength sampling point. Then, for each independent period's data sequence, the system calculates the square of the difference between the intensity value and the global mean for each sampling point, and the arithmetic mean is obtained to obtain the variance. The preset variance threshold is a boundary value line for evaluating the reasonableness of data fluctuations, pre-set by the system based on the dark noise level during factory calibration and the random jitter amplitude of the normal industrial environment. When the variance of a single period's data exceeds this preset variance threshold, it indicates that the data for that period has been subjected to sudden interference such as instantaneous obstruction by large particles in a pipeline, strong transient turbulence in the fluid, or mechanical vibration caused by the start-up and shutdown of large electromechanical equipment. In this case, the period is defined as an abnormal data group.

[0048] In the data removal stage, the system removes all sequences identified as abnormal data groups from the cache array, retaining the remaining data as valid data groups. The system then algebraically accumulates all valid data groups at each sampling point and takes the average, ultimately outputting a smooth average signal. This anti-interference feature extraction method effectively resists local data mutations caused by unsteady extreme interference in industrial environments from the underlying data structure.

[0049] Further, after extracting the average signal of the effective data set, the system performs subsequent digital signal smoothing and reconstruction. Specifically, a baseline correction algorithm is used to perform baseline correction on the average signal, and a data interpolation algorithm is used to reconstruct high-density points in the characteristic peak region. The baseline correction algorithm here refers to a method used to identify and subtract low-frequency fluctuations caused by the broadband emission spectrum profile of the light source itself, changes in detector dark current due to slow ambient temperature drift, and inherent non-uniformity of transmittance of the optical system lenses. Its purpose is to flatten the basic horizontal plane of the superimposed absorption peaks to near zero. The data interpolation algorithm refers to a technique that uses the mathematical variation trend of adjacent known data points between discrete wavelength sampling points to calculate and generate additional virtual data points, thereby improving the data resolution of a specific absorption spectral region. The high-density point reconstruction of the characteristic peak region aims to restore the fine peak profile lost due to the physical pixel spacing of the detector, thereby reducing subsequent fitting errors.

[0050] After high-density data reconstruction, a curve fitting algorithm is used to fit the reconstructed signal, extracting the center wavelength, peak intensity, and half-maximum width (WHM) of the characteristic peaks. This WHM is then combined with the direct transmitted light intensity and forward scattered light intensity to calculate the real-time scattering feature vector. The curve fitting algorithm is an optimization process that finds a specific mathematical distribution function that approximates the reconstructed discrete signal data points in terms of graphical form. The center wavelength refers to the abscissa of the spectral wavelength corresponding to the absorption or transmission peak determined by the fitting function; the peak intensity refers to the vertical height of the spectral intensity from the baseline horizontal plane to the peak apex; the WHM is the wavelength width spanned by the curve on both sides of the characteristic peak at half the height of the characteristic peak intensity; the direct transmitted light intensity is the energy value obtained by summing the integral regions of the direct transmitted light signal in the average signal; and the forward scattered light intensity is the energy value obtained by summing the integral regions of the forward scattered light signal in the average signal. The real-time scattering feature vector is a one-dimensional mathematical array composed of multiple independent scattering feature variables, used to comprehensively and quantitatively describe the optical scattering effect characteristics of contaminants on the optical window surface in the current sampling batch.

[0051] Taking methane gas measurement as an example, such as Figure 2 This figure shows a comparison of the waveform transformation results of optical signals after baseline correction and high-density reconstruction during the preprocessing stage. The horizontal axis represents the absorption wavelength of methane gas, measured in nanometers, with the data acquisition range shown between 1653 and 1654 nanometers. The vertical axis represents the light intensity, indicating the signal energy amplitude, measured in arbitrary units.

[0052] The light gray solid line in the figure corresponds to the original noisy signal in the legend. The curve shows a characteristic peak profile with specific absorption properties at a wavelength of 1653.7 nm, but the curve is superimposed with significant high-frequency random fluctuation noise, and the overall background shift trend increases with increasing wavelength.

[0053] The black dashed line corresponds to the fitted drift baseline in the legend. The system uses an adaptive iterative weighted least squares method to evaluate the original data and generates this baseline evaluation curve that reflects the low-frequency broadband offset pattern.

[0054] The solid black dots correspond to the baseline correction signal in the legend. The value is obtained by subtracting the fitted drift baseline point by point from the original noisy signal. This transformation eliminates the interference of the underlying background fluctuations, making the signal strength at both ends of the waveform even and dropping to a baseline close to zero.

[0055] The thick black solid line corresponds to the interpolation reconstruction curve in the legend. The system performs cubic spline interpolation at three times the data density for the relatively discrete black solid dots. After high-density interpolation, the data points that were originally sparse due to the physical resolution of the area array photodetector are expanded and connected into a continuous and smooth waveform.

[0056] The preprocessing operations represented by this series of graphs and curves of different colors effectively smoothed out noise abrupt changes and reconstructed the fine contours of the characteristic peak regions, laying a high-resolution data foundation for the subsequent extraction of characteristic parameters such as center wavelength and peak intensity by the dual-channel measurement system resistant to optical window contamination and decay.

[0057] It is also important to note that before the system is put into operation in the industrial field, in order to achieve dynamic optical decay analysis based on real-time scattering feature vectors, this embodiment pre-defines and constructs a core computational model within the system. Specifically, based on a pre-established three-dimensional correlation benchmark model that includes the mapping relationship between window transmittance attenuation, scattering feature distribution, and signal distortion, the current contamination stage and type of the optical window are determined. Here, the three-dimensional correlation benchmark model refers to a multi-dimensional mathematical surface model established with the optical window transmission attenuation coefficient, the spatial distribution of surface contaminant scattering, and the morphological distortion parameter of the final signal at the spectral receiver as the three main coordinate dimensions, used to describe the mapping rules between the physical phenomena of contamination and changes in photoelectric signals.

[0058] For example, the construction process of the three-dimensional associated benchmark model includes: first, defining the real-time scattering feature vector, which includes the proportion of scattered light intensity, the variance of the scattering angle distribution used to describe the spatial distribution characteristics of scattered light, and the spectral distribution coefficient of scattered light intensity used to describe the wavelength dependence of scattered light. The proportion of scattered light intensity is the ratio of the sum of forward scattered light intensity and the sum of direct transmitted light intensity and forward scattered light intensity, used to characterize the overall proportion of incident photons lost by random scattering from the pollutant surface. The variance of the scattering angle distribution is a statistical parameter calculated by analyzing the signal intensity gradient differences of pixel rings of different radii within the ring-shaped scattered light acquisition area. Its function is to reflect the average particle size of the attached pollutants; generally, large pollutant particles correspond to a smaller forward scattering angle distribution, resulting in a smaller variance of the scattering angle distribution; while fine aerosol particles cause the light to diverge at a wider angle, making it larger. The spectral distribution coefficient of scattered light intensity is calculated using the ratio of the intensity of the forward scattered light signal in the short-wavelength band to the intensity in the long-wavelength band, used to characterize the selective scattering tendency of pollutants to light of different wavelengths. This parameter is closely related to the chemical composition and refractive index of the pollutants.

[0059] When establishing a three-dimensional correlation benchmark model, the distorted master measurement optical signal needs to be decomposed into multiple independent components. This is because optical decay caused by complex pollution in industrial environments is not a single mechanism, but rather the result of a combination of factors. Specifically, the master measurement optical signal is characterized as: the product of the amplitude attenuation coefficient and the true characteristic signal of the measured medium, plus the sum of the peak distortion component, the baseline drift component, and the random noise component.

[0060] The true characteristic signal here refers to the spectral signal that should have been received by the area array photodetector under the assumption that the system's optical window is in a contamination-free and ideally clean state. It contains accurate concentration or composition information of the measured medium. The amplitude attenuation coefficient is a dimensionless real factor representing the proportion of global macroscopic energy loss in the detector beam caused by absolute blockage absorption or back reflection of contaminants. The peak distortion component refers to the asymmetric peak broadening and deformation waveform sequence superimposed on the true characteristic peak due to non-uniform phase shift and diffraction interference in the spatial distribution of light reaching the detector at different wavelengths caused by inhomogeneous island-like contamination deposition or strong multiple scattering effects on the optical window surface. The baseline drift component refers to a slowly fluctuating broadband background shift sequence caused by the increase in diffuse reflection background energy due to strong scattering of contaminants, across the entire wavelength range. The random noise component here refers to various transient high-frequency irregular white noise fluctuations superimposed on the optical signal.

[0061] For example, the above signal decomposition formula is shown below:

[0062] ;

[0063] In the formula: An array used to characterize the intensity sequence distribution of the original main measurement optical signal, which contains various contamination interferences, actually collected by the area array photodetector in the main measurement channel; Used to characterize the real-time amplitude attenuation coefficient caused by the combined effects of decreased transmittance due to optical window contamination and forward scattering; A sequence of true characteristic signals used to characterize the measured medium under ideal optical transmission conditions, unaffected by window contamination; Used to characterize the real-time peak distortion component sequence caused by the intensity-wavelength dependence of scattered light within a specific absorption spectrum. A sequence of real-time global low-frequency baseline drift components used to characterize the target monitoring band range; Used to characterize high-frequency transient random noise components in a system that cannot be eliminated by a fixed benchmark model.

[0064] It is important to note that the amplitude attenuation coefficient, peak distortion component, and baseline drift component in the above formula are not fixed constants, but parameters that change dynamically in real time as the contamination of the optical window intensifies. Specifically, through pre-calibration experiments, function mapping expressions between the amplitude attenuation coefficient, peak distortion component, and baseline drift component and the real-time scattering feature vector are fitted and constructed, thus solidifying the three-dimensional correlation benchmark model. These pre-calibration experiments are performed in a controlled environment chamber before the equipment leaves the factory. Test personnel manually spray standard dust, atomized silicone oil, or salt water droplets of different particle sizes onto the surface of the optical window to simulate the entire contamination cycle from light to extremely heavy levels. While recording the known physical parameters of the contamination under each contamination state, the system simultaneously records the collected real-time scattering feature vector and the corresponding actual data on signal attenuation and distortion. Subsequently, a multivariate nonlinear regression algorithm is used to process a large number of experimental data samples, ultimately obtaining a set of multivariate equations reflecting the changes of each distortion component with the three parameters of the scattering feature vector, which is the function mapping expression. By writing this set of verified function mapping expressions as read-only parameters into the device's underlying firmware, the solidification of the three-dimensional associative benchmark model is completed.

[0065] like Figure 3 This diagram illustrates the spatial data distribution and evolution of the fitted surface plot of a three-dimensional correlation benchmark model for window contamination decay. The first horizontal axis at the bottom of the diagram represents the percentage of scattered light intensity, measured as a percentage, ranging from 0 to 35. The second horizontal axis represents the variance of the scattering angle distribution, measured in arbitrary units, ranging from 0 to 15. The vertical axis represents the amplitude attenuation coefficient, a dimensionless real factor whose value gradually decreases from 1 to 0. The three-dimensional surface and accompanying color mapping visually reflect the nonlinear attenuation trajectory of the received signal energy under different stages of contamination in the optical measurement system.

[0066] When the proportion of scattered light intensity is in the range of 0 to 5, the curved surface presents a warm yellow tone representing high light transmittance, and the overall downward slope is relatively gentle. This spatial region corresponds to the linear attenuation stage described in the specific implementation. At this time, the amplitude attenuation coefficient is maintained above 0.7, indicating that there are only sparse fine particles attached to the surface of the optical window, which only produce macroscopic proportional energy blocking of the light signal.

[0067] As the proportion of scattered light intensity continues to increase and enters the range of 5 to 20, the color of the surface gradually transitions to green and light blue. Simultaneously, the spatial gradient of the surface undergoes a significant deflection and exhibits a steep downward trend; this region corresponds to the weak nonlinear scattering stage. Within this stage, as pollutants locally aggregate and form a microscopic optical lens effect, the amplitude attenuation coefficient rapidly drops from 0.7 to around 0.3, indicating that multiple scattering of light leads to a sharp decrease in the energy received by the system, accompanied by waveform broadening.

[0068] When the proportion of scattered light intensity exceeds 20% and the variance of the scattering angle distribution is also high, the curved surface extends to the deep blue, gently sloping valley region, corresponding to the stage of severe nonlinear strong scattering. At this point, the amplitude attenuation coefficient has fallen below 0.3, confirming that the optical signal has been severely refracted, scattered, and lost due to the thick contamination layer.

[0069] This smooth and continuous curved surface shape verifies the objective rationality of the multivariate regression mapping relationship established based on a large number of full-contamination cycle calibration experiments, and provides a reliable three-dimensional mathematical model support for the dual-channel optical system to dynamically decouple the pure contamination distorted substrate and reversely reconstruct the real signal of the measured medium under complex working conditions.

[0070] After the system is put into actual measurement, this embodiment evaluates the contamination status of the optical window in real time based on the aforementioned pre-established model. Specifically, the contamination stage of the optical window is determined based on the pre-established three-dimensional correlation benchmark model, and the quantification rule is as follows: calculate the real-time window transmittance and the proportion of scattered light intensity. The real-time window transmittance is an index parameter that measures the global light transmission capability of the current optical window to the probe beam. The real-time window transmittance is obtained by the ratio of the real-time direct transmitted light intensity of the reference channel to the initial clean reference light intensity. Since the reference channel is not affected by the measured medium, its light intensity attenuation is entirely attributed to the contaminants attached to its outer surface. The initial clean reference light intensity refers to the absolute value of the direct transmitted light intensity of the reference channel stored in the system during the power-on initialization calibration phase when the system is first installed and the optical window is thoroughly wiped clean. The proportion of scattered light intensity is obtained by dividing the forward scattered light intensity of the reference channel by the sum of the direct transmitted light intensity and the preset zero-prevention constant, in order to avoid the calculation overflow problem under extreme black screen obstruction. This parameter is not affected by the natural decay of the overall light intensity of the light source over its lifespan, and has excellent anti-common-mode drift characteristics. It can independently characterize the intensity of light scattering caused by particles on the window surface.

[0071] Based on the calculated real-time parameters, the system executes a three-level judgment logic for the contamination stage. For example, when the real-time window transmittance is greater than or equal to a first transmittance threshold and the proportion of scattered light intensity is less than or equal to a first scattering proportion threshold, the system determines it is in the linear decay stage. The first transmittance threshold is the upper limit safety boundary parameter that the system allows for slight light transmission loss at the optical window. The first scattering proportion threshold is the upper limit boundary parameter that the system defines the optical window surface as having only sparse and fine dust adhesion, without forming an optical rough surface that causes large-scale complex deflection of light. The linear decay stage represents the initial contamination state after the instrument has just been put into operation or recently cleaned and maintained. At this time, the contaminants are relatively uniformly and sparsely distributed, and the impact on the light signal is only a proportional decrease in overall energy, without causing distortion of the spectral waveform.

[0072] Furthermore, with the continuation of industrial production, the thickness and density of pollutants gradually increase. When the real-time window transmittance is less than a first transmittance threshold but greater than or equal to a second transmittance threshold, and the proportion of scattered light intensity is greater than a first scattering proportion threshold but less than or equal to a second scattering proportion threshold, it is determined to be in the weak nonlinear scattering stage. The second transmittance threshold refers to the judgment boundary when the light transmittance of the optical window decreases to a moderate level. Here, the second scattering proportion threshold refers to the judgment boundary when pollutants have begun to locally aggregate and form a more obvious microscopic optical lens effect. Among them, the first transmittance threshold is greater than the second transmittance threshold, and the second scattering proportion threshold is greater than the first scattering proportion threshold. The weak nonlinear scattering stage means that a continuous pollutant film or a dense particle accumulation layer has formed on the surface of the optical window. At this time, in addition to the overall energy attenuation, the scattering direction of light begins to become disordered, which not only leads to a further reduction in the received light intensity of the system, but also causes the signal substrate to drift. However, the core contour of the characteristic peak has not yet undergone severe asymmetric deformation.

[0073] Optionally, if the operating conditions continue to deteriorate and timely maintenance is not provided, when the real-time window transmittance is less than a second transmittance threshold and the proportion of scattered light intensity is greater than a second scattering proportion threshold, it is determined to be in a severely nonlinear strong scattering stage. In this stage, the optical window surface may be completely covered by a thick layer of tar, sticky dust, or irregular crystalline salt blocks. Light undergoes intense multiple scattering and refraction loss when penetrating the contamination layer, resulting in not only extremely weak light intensity but also severe deviations in the propagation paths of different wavelengths. At this time, the signal will experience severe peak distortion, and the true absorption characteristics are almost completely submerged by environmental background noise and scattered waveforms. The weak nonlinear scattering stage and the severely nonlinear strong scattering stage are collectively referred to as the nonlinear scattering stage.

[0074] It should be noted that when the data combination of real-time window transmittance and scattered light intensity ratio does not fall within the preset range of the above three stages, the system determines it as an abnormal working condition boundary state. At this time, the highest level of compensation strategy for the severe nonlinear strong scattering stage is executed by default, or the abnormal state review process is directly triggered.

[0075] To ensure the accuracy of output data throughout the entire pollution cycle, the system implements differentiated compensation strategies based on the different pollution stages determined. For example, when the system is determined to be in the linear attenuation stage, since the signal distortion is of a single type at this time, there is no need to initiate complex full-dimensional decoupling. The system simply constructs a differential and ratio dual-dimensional calibration model based on the characteristic peak intensity of the direct transmitted light from the main measurement channel and the reference channel, and performs linear amplitude attenuation compensation on the main measurement optical signal.

[0076] Specifically, this involves extracting the characteristic peak intensities of the direct transmitted light from the main measurement channel and the direct transmitted light from the reference channel. Subsequently, a differential signal is constructed by calculating the difference between the characteristic peak intensities of the direct transmitted light from the main measurement channel and the direct transmitted light from the reference channel. Since both channels use the same source light source and the detector temperature environment is consistent, the calculation of the differential signal can directly offset the transient high-frequency jitter of the light source power, the common-mode drift of the photodetector's thermal noise floor, and the interference errors caused by rapid changes in the system ambient temperature, thus extracting a pure measurement difference trend.

[0077] Next, a ratio signal is constructed by comparing the characteristic peak intensity of the direct transmitted light from the main measurement channel with the characteristic peak intensity of the direct transmitted light from the reference channel. Based on this, since the light intensity of the reference channel characterizes the pure contamination attenuation, a real-time window decay factor is obtained by mapping this ratio relationship using a preset lookup table. This real-time window decay factor is a correction coefficient that dynamically characterizes the proportion of the light transmittance of the main measurement channel's optical window relative to the equivalent remaining capacity in an absolutely clean state within the current second. Finally, the linear amplitude attenuation compensation of the main measurement light signal is completed by dividing the characteristic peak intensity of the direct transmitted light from the main measurement channel by the real-time window decay factor. Through this division operation, the signal intensity that has dropped due to contamination is directly and proportionally restored to the true energy level on the vertical axis.

[0078] For example, the decay factor for linear recovery calculated using the algorithm formula based on signal ratio inversion is specifically expressed as follows:

[0079] ;

[0080] In the formula: The dynamic compensation correction magnification factor is used to characterize the loss of light transmittance of the optical window at the current acquisition time, obtained by comparing dual-channel optical information. Used to characterize the absolute value of the energy integral of the characteristic peak of the direct transmitted light of the measured medium, which is acquired by the main measurement channel and processed after baseline correction. It is used to characterize the absolute value of the energy integral of the characteristic peak of reference direct transmitted light under the influence of pure contamination, which is synchronously acquired by a reference channel in a closed clean environment within the same sampling period.

[0081] It is important to note that as industrial field measurements progress, when the feature extraction and stage determination module determines that the system is in the nonlinear scattering stage, the aforementioned simple linear amplitude amplification will produce a severe overcompensation distortion problem, that is, it will simultaneously amplify the peak distortion and erroneous substrate caused by strong scattering. Therefore, the system will automatically switch processing paths.

[0082] Specifically, the real-time scattering feature vector is input into a three-dimensional correlated reference model to decouple and generate a pure contamination distortion substrate. Full-dimensional compensation for scattering distortion, including baseline drift and peak shape distortion, is then performed on the main measured optical signal. The specific process of decoupling and generating the pure contamination distortion substrate is as follows: the real-time scattering feature vector is input into the function mapping expressions in the three-dimensional correlated reference model. This is because, at this point, it is necessary to consider not only the overall transmittance but also the spatial and spectral distortion effects caused by the variance of the scattering angle distribution and the spectral distribution coefficient of the scattered light intensity. The real-time amplitude attenuation coefficient, real-time peak shape distortion component, and real-time baseline drift component caused solely by optical window contamination at the current moment are calculated. Here, the real-time amplitude attenuation coefficient, real-time peak shape distortion component, and real-time baseline drift component are mathematical stripping array sets specific to the current contamination type.

[0083] Subsequently, a complex signal reconstruction operation is performed. The reconstructed signal yields the true characteristic signal of the medium under test. Specifically, the true characteristic signal of the medium under test is reconstructed as follows: the main measurement optical signal minus the real-time baseline drift component, the real-time peak distortion component, and the random noise component, and the resulting residual is divided by the real-time amplitude attenuation coefficient.

[0084] This reconstruction process strictly follows the inverse operational logic from nonlinear shape correction to linear intensity normalization. First, at the waveform morphology level, from the original spectrum containing contamination interference, the undulating base and tail distortion imposed by multiple scattering of contaminants are removed point by point. Then, after stripping away the spectral footprint and random high-frequency jitter unique to the contaminants, the system obtains a clean residual signal with a pure shape but weak overall intensity due to overall obstruction by the contaminant layer. Finally, this clean signal array is uniformly divided by the real-time amplitude attenuation coefficient for overall amplification at the macroscopic energy level, ultimately recovering the true characteristic signal of the measured medium, hidden behind the heavily contaminated layer, with the correct wavelength position and true absorption depth.

[0085] For example, the following algorithm formula for nonlinear distortion reconstruction is provided:

[0086] ;

[0087] In the formula, This array of spectral response curves, representing the true physical concentration of the measured medium, is used to characterize the system after performing a full-process algorithm stripping operation during the nonlinear scattering stage.

[0088] Taking methane gas measurement as an example, such as Figure 4 This paper presents a detailed breakdown of the process of decoupling and reconstructing the true features of the distorted signal in the stage of severe nonlinear strong scattering in an optical measurement system. The horizontal axis in the figure uniformly represents the absorption wavelength of methane gas, with the measurement unit being nanometers and the data range between 1653 and 1654 nanometers. The vertical axis represents light intensity, with arbitrary units.

[0089] The gray curve presented in the first part is the original distorted signal. Under the influence of complex multiple scattering and local clustering contamination, the overall energy background of the signal is significantly raised to the range of 0.4 to 0.9, and the characteristic peak near 1653.7 nm shows obvious asymmetric broadening and high-frequency jitter.

[0090] The blue curve in the second part shows the baseline drift component extracted by the system algorithm. The curve shows a gradual upward trend with increasing wavelength, quantifying the diffuse broadband background shift caused by strong scattering.

[0091] The magenta curve in the third part represents the peak distortion component obtained by decoupling. It exhibits a significant one-sided asymmetric protrusion near 1653.8 nm, reflecting the spectral diffraction and phase shift effects caused by the contamination film.

[0092] Part Four presents the reconstruction verification results, where the black dashed line represents the uncontaminated standard signal under ideal conditions, and the red solid line represents the reconstructed true feature signal after the system performs full-dimensional compensation.

[0093] The system subtracts the baseline drift component, peak distortion component and random noise from the original distorted signal one by one, and divides the resulting residual by an amplitude attenuation coefficient of less than 0.3. The red reconstruction curve and the black standard curve are highly coincident in terms of center wavelength and peak intensity.

[0094] This series of top-down waveform transformations and multi-curve comparison data objectively verifies that the multi-stage, full-dimensional compensation logic can reversely restore the true absorption characteristics of the measured medium under extremely severe pollution shielding and scattering interference, thereby ensuring the measurement continuity and reliability of industrial online analytical instruments throughout the entire pollution cycle.

[0095] Specifically, to avoid false data output due to model overcompensation, a two-dimensional validity check is performed on the compensated true feature signal. If the check passes, the final measurement result is output. This two-dimensional validity check includes a first-dimensional scattering feature consistency check and a second-dimensional standard spectral line residual check. The purpose of the first-dimensional scattering feature consistency check is to evaluate the self-consistency of the reconstruction algorithm from the perspective of data residual patterns.

[0096] The specific steps are as follows: The reconstructed true feature signal is subtracted from the original master measurement optical signal to generate a residual signal. Theoretically, if the reconstructed signal is completely accurate and reliable, then this residual portion should be substantially equivalent to the distortion imposed on the signal by contaminants. Therefore, the correlation coefficient between this residual signal and the pure contamination distortion substrate is calculated. If the correlation coefficient is greater than or equal to the first correlation coefficient threshold, the consistency check is considered passed. Here, the correlation coefficient is the Pearson index used in statistics to measure the degree of linear dependence between two variables. The first correlation coefficient threshold refers to a high-standard passing boundary for the lower limit of the allowable algorithm fitting deviation. If the correlation coefficient does not reach this boundary, it indicates that a calculation deviation has occurred in the decoupling process, and the generated true feature signal may be at risk of distortion.

[0097] The purpose of the second-dimensional standard spectral line residual verification is to introduce a known signal with a precise physical traceability benchmark for hardware-level verification. The specific steps are as follows: The calibration module built into the system outputs a standard characteristic spectral line. This calibration module is a device independent of the measurement light source system, capable of generating a spectral line with an extremely narrow bandwidth and a highly constant physical wavelength. The optical signal of this spectral line is acquired and a nonlinear compensation process is performed. During this process, the calibration light also passes through a contaminated optical window and is received by the detector. Subsequently, the relative residual between the compensated standard spectral line characteristic parameters and the inherent standard value is calculated. If the relative residual is less than or equal to the second relative residual threshold, the residual verification is considered successful. The relative residual is the absolute percentage deviation between the compensated output value and the known standard theoretical value, and the second relative residual threshold is the maximum allowable measurement uncertainty index boundary for the system at the factory.

[0098] For example, if the verification fails, the parameter adaptive iteration process of the three-dimensional associated benchmark model is triggered, and compensation and verification are re-executed after parameter updates. If the number of consecutive verification failures reaches a preset maximum iteration threshold, the iteration process is terminated, a hardware fault alarm is triggered, and a measurement invalidation flag is output. The adaptive iteration process is triggered when any item in the two-dimensional validity verification fails, or when a change in the pollution stage is determined. A stage change means that the physical pollution mechanism may have abruptly changed, such as a dry dust zone suddenly entering a humid condensation zone, and the original model parameter weights may no longer be applicable. In this case, using the standard characteristic spectrum output by the calibration module as the iteration benchmark, an online parameter optimization algorithm is used to perform online iterative correction of the parameters of the function mapping expression in the three-dimensional associated benchmark model.

[0099] The online parameter optimization algorithm here refers to a dynamic self-adjusting model update process in which the system, in a background computing thread that continuously operates within the actual industrial measurement process, uses gradient descent or Newton's iteration mechanism to find the optimal solution to a system of multivariate equations. The online iterative correction of parameters aims to minimize the relative residual of the standard characteristic spectral lines after compensation. After each iteration, the updated model parameters are fixed for use in the next compensation calculation. By continuously updating the model coefficients based on the latest on-site conditions, the system maintains its self-evolutionary vitality in long-term, complex, and harsh environments.

[0100] For example, the core formula of the objective function for iterative optimization is provided below:

[0101] ;

[0102] In the formula: The output value of the overall system calibration cost function is used to characterize the minimum limit value that the online parameter optimization algorithm needs to pursue in each iteration cycle. This is used to characterize the simulated values ​​of band characteristics obtained after performing mathematical reconstruction compensation on the characteristic signals excited by the calibration module under the current round of tentative model parameter mapping; This is used to characterize the absolute true peak intensity and wavelength positioning physical invariant set inherent to the calibration module, traceable to national metrological standards.

[0103] Optionally, to achieve proactive maintenance management of factory equipment and reduce the risk of unplanned downtime, the method further includes predictive maintenance steps for optical windows based on a time-series prediction network model. Specifically, the system requires long-term data tracking: real-time storage of historical operational sequence data containing the real-time window transmittance and the proportion of scattered light intensity throughout the optical window's lifespan.

[0104] Historical operational sequence data is a long-term database that continuously records the light transmission and scattering degradation trajectory of the optical window according to timestamps. This historical operational sequence data is input into a pre-trained time-series prediction network model, which outputs a time-series prediction curve of optical window degradation. The time-series prediction network model is a recurrent neural network computational architecture with memory feature extraction capabilities. It not only focuses on the absolute value of the current data but also captures the slope change pattern of the parameter's slow degradation over time. The time-series prediction curve is a parabolic or exponential curve inferred by the network model, depicting the decay trajectory of the optical window's transmittance index over the next tens of days or even months. Based on this time-series prediction curve, the remaining estimated time for the optical window to transition from a linear decay stage or a weak nonlinear scattering stage to a severely nonlinear strong scattering stage is calculated. When this remaining estimated time reaches a preset alarm time threshold, a predictive maintenance prompt containing instructions for window cleaning or replacement is generated. The remaining estimated time refers to the time margin before the prediction curve reaches the underlying functional failure threshold. The preset alarm time threshold here is a buffer safety constant set in advance by factory equipment engineers to allow sufficient time for the procurement of maintenance materials and personnel scheduling.

[0105] In summary, this embodiment introduces independent reference channels for acquiring direct and scattered light in the hardware architecture, changing the conventional setting of relying solely on empirical coefficients for fixed magnification compensation in existing technologies. In harsh industrial applications, existing technologies are often helpless against nonlinear optical signal distortion and baseline drift caused by large amounts of dust, leading to severe distortion of the measured spectrum and ultimately forcing a shutdown.

[0106] This technical solution achieves adaptive assessment of different contamination stages of the optical window by accurately calculating real-time scattering feature vectors and establishing a three-dimensional correlation benchmark model. For initial contamination, efficient differential and ratio linear calibration is implemented; for deep contamination causing severe distortion, a rigorous algorithmic mechanism is used to reconstruct, decouple, and nonlinearly strip the pure contamination substrate, and a built-in calibration module is introduced to perform dual-dimensional validity verification and closed-loop iterative adaptive optimization. The entire implementation scheme ensures that the system can continuously, stably, and accurately output the characteristic information of the measured medium in extremely harsh and dynamically changing measurement environments such as high dust, strong corrosion, and heavily polluting aerosols without frequent manual cleaning, significantly improving the long-term availability and industrial practical value of online analytical instruments.

[0107] Example 2:

[0108] In actual optical online monitoring operations, environmental background noise is not a constant. For example, in the exhaust pipes of desulfurization towers in coal-fired power plants or large chemical reactors, the flow velocity of the fluid medium changes drastically with the increase or decrease of production load, resulting in fluid turbulence of varying intensity within the pipes. Simultaneously, the periodic start-up and shutdown of large induced draft fans and compressors, as well as the operation of mechanical valves within the plant, all transmit vibrations to the mechanical structure of the optical measurement equipment through pipe flanges, causing optical vibrations of varying amplitudes. Traditional anti-interference processing often uses a statically preset variance threshold calibrated in the laboratory to remove outlier data. However, when faced with a surge in overall environmental noise due to high-load production, the fixed static threshold becomes overly stringent. This leads to a large number of periodic data points, which are only affected by normal environmental fluctuations, being misjudged as abnormal and widely rejected. Ultimately, a sufficient number of valid data sets cannot be extracted for averaging, forcing the measurement process to be interrupted. Conversely, when the production line is in a low-load, stable state, this static threshold is too broad and cannot effectively identify and eliminate the tiny distorted data caused by transient electromagnetic pulses.

[0109] To overcome the above problems, this embodiment proposes that the preset variance threshold is a dynamic variance threshold that is adaptively generated based on the actual working condition noise level. Its specific generation and anomaly removal steps form an adaptive environmental perception and data closed-loop screening mechanism.

[0110] For example, to accurately perceive the background noise of actual industrial conditions without introducing additional physical vibration sensors, this system fully reuses existing optical hardware links. Within the spectral signal range acquired by the area array photodetector, a wavelength range devoid of the optical absorption characteristics of the measured medium is pre-selected as a characteristic blank noise band. The spectral signal range acquired by the area array photodetector refers to the full-band physical range that the device's spectral system can cover and project onto the sensor's photosensitive surface. The optical absorption characteristics of the measured medium refer to the characteristic photon absorption cross-section generated at a specific wavelength position by target gas molecules or particles due to energy level transitions. Here, the characteristic blank noise band refers to a specific continuous pixel range artificially defined within the entire acquired spectrum, within which the absorption coefficient of the target measured substance approaches a minimum value and is unaffected by the cross-absorption of common coexisting interfering gases in industrial environments.

[0111] In practical applications, since there is no chemical absorption attenuation in this band, the high-frequency intensity fluctuations of the photoelectric signal in this band are purely caused by the micro-shift of the optical path due to mechanical structural oscillations, the vibration of particles attached to the surface of the optical window, and the optical refractive index disturbance caused by changes in fluid density inside the pipe. Using this as a characteristic blank noise base band is equivalent to building a highly sensitive optomechanical-fluidic composite noise probe into the optical system.

[0112] It is important to note that after determining the probe band, the system needs to perform quantitative analysis. After synchronously acquiring optical signals for multiple consecutive cycles, the standard deviation of light intensity within the characteristic blank noise band is extracted for each single-cycle data point. Multiple consecutive cycles refer to several independent emission and detection cycles completed by the light source modulator driving the light-emitting components within a set time window. Single-cycle data refers to the full-band signal array acquired by the area array photodetector within a single integration time. The standard deviation of light intensity is a physical quantity used in statistics to measure the dispersion of a set of data from its arithmetic mean. Calculating this standard deviation of light intensity is the process of quantifying the energy of pure environmental noise interference within each independent integration cycle.

[0113] For example, the system uses the following single-cycle discreteness algorithm formula:

[0114] ;

[0115] In the formula: Used to characterize the system in the first The standard deviation of light intensity per cycle is calculated for the characteristic blank noise band region in each independent sampling period. Used to characterize the total number of effective physical pixel sampling points on the area array photodetector included in the system when dividing the characteristic blank noise band; Used to characterize the in Within each sampling period, the first [number]th ... The real-time digital quantization value of photoelectric intensity output by a specific physical pixel sampling point; Used to characterize the in Within each sampling period, all contained within the characteristic blank noise band The arithmetic average level of the digital quantization values ​​of photoelectric intensity at each pixel sampling point.

[0116] Specifically, after obtaining the dispersion of each cycle, the system does not directly use the result of a single cycle as the final evaluation basis to prevent misjudgment caused by an extreme random electronic pulse noise. The system further calculates the average standard deviation of the multi-cycle data, which is defined as the real-time turbulence coefficient characterizing the current fluid turbulence and optical vibration state. Here, the average standard deviation of the multi-cycle data refers to averaging the series of single-cycle standard deviations obtained above over the time dimension. The fluid turbulence and optical vibration state is a physical summary of the macroscopic severity of the current industrial measurement site. The real-time turbulence coefficient is a dimensionless relative evaluation index. The larger the value of this coefficient, the more intense the airflow vortex in the measuring pipe at the current moment, the greater the mechanical vibration amplitude of the equipment base, that is, the stronger the comprehensive physical background noise faced by the current measurement batch.

[0117] For example, in order to output smooth and statistically significant environmental assessment indicators, the system calls the following time-series smoothing algorithm formula:

[0118] ;

[0119] In the formula: The real-time turbulence coefficient output is used to characterize the intensity of the external environment in which the measuring device is located after time series smoothing. This is used to characterize the total number of continuous optical signal cycles that the system initially preset and actually synchronously acquired in the current evaluation batch.

[0120] Optionally, after completing the real-time quantitative perception of the external environmental noise level, the system multiplies the pre-calibrated baseline variance threshold under steady-state conditions with the real-time turbulence coefficient to generate the dynamic variance threshold applicable to the current sampling batch. Here, the steady-state environment refers to the testing environment of the equipment before it leaves the factory, placed in an ideal laboratory calibration chamber with constant temperature and humidity, static fluid, and vibration-proof platform isolation. The baseline variance threshold is the equipment's inherent data tolerance limit value calibrated under the aforementioned steady-state environment, considering only the detector's inherent dark current noise and the internal thermal radiation jitter of the light source. The current sampling batch refers to the data processing process currently undergoing photoelectric conversion and before the final concentration result is output. The dynamic variance threshold refers to a dynamic judgment criterion that is freed from the constraints of fixed parameters and can expand or contract in real time according to the severity of the measurement environment.

[0121] For example, to achieve adaptive threshold mapping transformation, the system has a built-in threshold dynamic reconstruction algorithm formula as follows:

[0122] ;

[0123] In the formula: This is used to characterize the real-time dynamic variance threshold variable generated after the system's adaptive calculation and will be applied to the subsequent data outline elimination process. A steady-state environmental baseline variance threshold constant used to characterize the device's storage in the underlying non-volatile memory, representing the device's physical background tolerance; Quantitative measurements of turbulent systems used to characterize the severity of real-time feedback environments; An empirical scaling calibration factor is artificially set to characterize the upper and lower limit response slopes of the dynamic threshold in engineering implementation, in order to prevent the real-time turbulence coefficient from mapping to a threshold that exceeds the reasonable physical boundary under extreme conditions.

[0124] like Figure 5This figure demonstrates the time sequence distribution of the judgment effect of different threshold setting methods when an optical measurement system performs anomaly data removal in an unsteady industrial environment. The horizontal axis of the figure represents time, with the measurement unit being minutes, and the data spanning from 0 to 100 minutes; the vertical axis represents variance, with the measurement unit being arbitrary, and the values ​​ranging from 0 to 12.

[0125] The light gray solid line in the figure corresponds to the actual operating noise fluctuation curve in the legend. This curve reflects the real environmental interference energy perceived by the system through the characteristic blank noise floor band. During the stable production phase from 0 to 40 minutes, the noise floor variance remains at a low level of 2 to 3. However, during the period from 40 to 60 minutes, affected by drastic changes in production load or the start-up and shutdown of large fluid equipment, the noise floor variance increases significantly and exhibits drastic fluctuations with a peak value close to 7.

[0126] The red dashed line in the figure corresponds to the static preset variance threshold in the legend. This threshold is fixed at 4.5. During the stable phase, this fixed threshold is significantly higher than the actual noise floor, which may cause minor abnormal electromagnetic pulse data to be missed. In the middle phase when the noise floor surges, the actual noise floor fluctuations greatly exceed the fixed threshold of 4.5, causing the system to misjudge and remove a large number of valid data sets that are only affected by normal environmental fluctuations, thus leading to the risk of operation interruption due to insufficient data.

[0127] The blue solid line in the figure corresponds to the dynamic variance threshold curve in the legend. The system adaptively maps the real-time extracted turbulence coefficient with the benchmark threshold, so that the blue curve always closely follows the fluctuation trend of the gray noise curve. In the stable stage, it tightens to about 3.5 to improve data purity, and in the harsh working conditions, it adaptively relaxes to about 10, effectively accommodating the overall noise increase caused by drastic environmental changes.

[0128] This time series comparison intuitively illustrates that the dynamic threshold generation mechanism can effectively perceive the evolution of the severity of the external environment, thereby ensuring the rationality of data screening and the continuity of the measurement process for optical online monitoring equipment under complex working conditions.

[0129] It is important to note that after obtaining the judgment criteria with environmental adaptability, when the system uses the point-by-point variance statistical method, it replaces the preset variance threshold with the dynamic variance threshold to perform the operation of removing abnormal data groups with variances greater than the preset variance threshold, and counts the number of valid data groups remaining after removal in real time. The point-by-point variance statistical method, as described in the previous embodiment, is used to check the dispersion of all periodic data at each wavelength position. Abnormal data groups refer to those periodic data that are subjected to severe non-steady-state interference, such as instantaneous large-area obstruction by large fly ash particles, and whose fluctuation level even exceeds the dynamic variance threshold relaxed by the current real-time environment. The number of valid data groups refers to the total number of frames of high-quality spectral sequences whose fluctuations are within the reasonable acceptable range of the current environment after passing through this adaptive filter.

[0130] For example, in order to perform the data screening and counting process, the system iterates through the array and executes the following logical judgment algorithm formula:

[0131] ;

[0132] In the formula: Used to characterize the target extraction band (including measurement bands containing absorption features), the first The cycle is in the first The local variance measure of each effective physical pixel relative to the multi-period average of that point; Used to characterize the The cycle is in the first The photoelectric intensity value at each target pixel is measured; Used to characterize the first The statistical mean is obtained by averaging the arithmetic mean of all collected periodic light intensity values ​​for each target pixel. This is used to characterize the dynamic discrimination guideline automatically generated by the system in conjunction with real-time operating conditions. Only when the variance of all target pixels within a period satisfies the above inequality is the period considered a valid data set.

[0133] Specifically, under extremely harsh operating conditions, even with relaxed thresholds, continuous high-frequency interference may still occur. Therefore, if the number of valid data sets is less than the preset minimum number of data sets, a closed-loop dynamic data acquisition mechanism is automatically triggered. The preset minimum number of data sets is the minimum number of valid spectral frames that the system must collect to ensure that subsequent averaging signal processing can effectively reduce white noise and thus achieve basic statistical confidence. The closed-loop dynamic data acquisition mechanism means that the internal hardware timing controller does not interrupt the current measurement task but automatically resends the integration sampling trigger command to the light source and detector, establishing a feedback control loop for cyclically supplementing data.

[0134] Once this mechanism is triggered, the system controls the dual-channel optical measurement system to continue performing single-cycle supplementary sampling, and uses the dynamic variance threshold in real time to verify the validity of the supplementary single-cycle data until the accumulated number of valid data sets reaches the preset minimum number of data sets. Then, the closed-loop dynamic supplementary sampling mechanism terminates, and the subsequent operation of extracting the average signal of the valid data sets is performed. Here, single-cycle supplementary sampling refers to an additional frame-by-frame independent detection action added outside of the originally planned acquisition batch. Validity verification means that after acquiring a new frame of data, the current dynamic variance threshold is immediately applied for filtering and judgment.

[0135] The system uses a memory accumulator to record the total number of all data frames that pass the screening. Terminating the closed-loop dynamic supplementary sampling mechanism here means that the system exits the loop instruction flow of supplementary sampling and returns control to the main data processing flow. By executing the subsequent operation of extracting the average signal of the effective data groups, the system uses the collected high-quality data matrix for subsequent baseline correction and concentration inversion calculations.

[0136] In summary, most existing conventional optical inspection instruments employ rigid static threshold filtering logic. This traditional approach, when faced with increasingly complex variable-load production conditions in industrial settings, leads to a significant amount of data being relentlessly blocked by the fixed threshold under harsh conditions and increased overall noise levels. This results in frequent system crashes and disconnections due to insufficient data. Conversely, under abnormally stable conditions, the lenient fixed threshold allows abnormal interference data, including minute electromagnetic interference, to slip into the averaging calculation, causing inexplicable jumps in the results.

[0137] The technical solution described in this embodiment utilizes the blank noise band outside the measurement spectrum as an environmental monitoring probe, constructing an adaptive adjustment chain from environmental perception to turbulence coefficient extraction, and then to dynamic threshold generation. This solution gives the optical measuring instrument an environmental "breathing" feel for on-site vibrations and airflow disturbances. In harsh environments, the threshold is automatically widened to maintain data continuity, while in stable environments, the threshold is automatically tightened to improve data purity. Simultaneously, the introduction of a closed-loop dynamic re-sampling mechanism significantly reduces the risk of data link breakage due to transient extreme conditions at the underlying logic level. This entire data preprocessing mechanism based on dynamic environmental evolution improves the online continuous operation capability and data output quality of optical instruments in various non-steady-state industrial environments without requiring any changes to physical hardware, possessing high industrial promotion value and practical application significance.

[0138] Example 3:

[0139] In practical industrial online spectral analysis applications, such as flue gas emission pipelines in waste-to-energy plants or pipelines transporting high-concentration chemical fluids, the contaminants adhering to the surface of the optical window are often not ideally uniformly distributed. Influenced by hydrodynamic boundary layer effects, local condensation, and the random collision and adhesion mechanism of large particles, island-like, locally uneven contamination deposition is highly likely to occur on the optical window surface. This asymmetric local contamination aggregation leads to severe local dark spots in the spatial distribution of the measurement spot passing through the optical window. In existing conventional signal processing techniques, macroscopic global energy integration is typically performed directly on the entire photosensitive area of ​​the photoelectric sensor to obtain the characteristic peak intensity. However, when there is spatial asymmetry in the local contamination distribution between the reference channel and the main measurement channel, direct global integration equates this extremely attenuated local dark spot error to a global decrease in transmittance. This results in a correction factor that deviates from the true situation during subsequent linear amplitude attenuation compensation, leading to systematic overcompensation or undercompensation distortion.

[0140] To address the aforementioned issues, this embodiment improves the steps for extracting the characteristic peak intensities of the direct transmitted light from the main measurement channel and the direct transmitted light from the reference channel. Specifically, dynamic spatial masking technology is used to remove local abnormal interference and reconstruct the effective signal.

[0141] For example, the two-dimensional pixel intensity distribution matrix of the direct light acquisition area of ​​the area array photodetector within the target characteristic peak wavelength range is obtained. The area array photodetector is a photoelectric conversion device with spatial resolution. The direct light acquisition area refers to the physical pixel array in the central region of the detector used to receive transmitted photons without significant angular deflection. The target characteristic peak wavelength range refers to the optical frequency band containing specific spectral absorption information pre-locked by the system based on the physicochemical properties of the measured medium. The two-dimensional pixel intensity distribution matrix refers to the discrete digital voltage signal array output by each photosensitive pixel within the direct light acquisition area based on the energy of the photons it receives within a specific sampling integration period. This matrix not only contains spectral intensity information that varies with wavelength but also retains the two-dimensional spatial physical position distribution shape of the light beam passing through the optical window. When subjected to island-like local contamination interference, significant data concave distortion will appear in the pixel region corresponding to specific spatial coordinates in this matrix.

[0142] It is important to note that in order to accurately locate these physically existing island-like contamination areas in digital space, the system needs to convert light intensity data into spatial gradient data. Two-dimensional spatial gradient calculations are performed on the two-dimensional pixel light intensity distribution matrices of the main measurement channel and the reference channel, respectively. Here, two-dimensional spatial gradient calculation refers to the mathematical process of using discrete differential operators to evaluate the rate of change of the numerical value of any data point in the pixel matrix relative to its surrounding neighboring data points in the spatial dimension. Through spatial gradient calculation, the system can effectively highlight high-frequency edge information in the light intensity distribution matrix. In areas with uniform light intensity on the optical window surface, the spatial transition is gentle, and the corresponding spatial gradient value is at a low level; however, in the edge areas of island-like contamination deposits, the light intensity drops sharply, and the corresponding spatial gradient value shows a rapid, pulse-like increase.

[0143] To achieve quantized extraction of two-dimensional spatial gradients, the system incorporates the following formula for calculating the discrete spatial gradient magnitude:

[0144] ;

[0145] In the formula: Used to characterize the extracted value located at the th matrix after two-dimensional spatial discrete differentiation operation. line, number The spatial gradient magnitude of a specific pixel in a column; Used to characterize the actual output of an area array photodetector within the target characteristic peak wavelength range of a specific channel, located at the [missing value]. line, number The original discrete pixel light intensity value at the column coordinate position; and These represent the light intensity values ​​of the two adjacent pixels of the specific pixel in the horizontal and lateral dimensions, respectively. and These represent the light intensity values ​​of the two adjacent pixels in the vertical dimension of the specific pixel.

[0146] like Figure 6 This diagram illustrates the two-dimensional pixel light intensity matrix and its corresponding spatial characteristic gradient heatmap received by an area array photodetector under island-like localized pollution conditions. The diagram contains two related sub-plots, with the horizontal axis representing the pixel column index and the vertical axis representing the pixel row index, both ranging from 1 to 60. The left sub-plot displays the two-dimensional pixel light intensity matrix, and the color bars on its right represent the light intensity values ​​in arbitrary units.

[0147] In the left image, the background area displays yellow and bright green hues representing high light intensity, with intensity values ​​generally between 80 and 100, reflecting the direct transmitted light energy normally received by the detector's central region. However, near coordinate indices 20 and 40, and 45 and 15, two distinct dark spots appear, exhibiting a deep blue hue, with their local intensity dropping sharply below 20. These dark spots objectively reflect uneven island-like contamination deposition on the optical window surface, where localized clusters of contaminants severely obstruct the transmission of normal light.

[0148] The right-hand subplot shows a spatial feature gradient heatmap generated after performing a two-dimensional spatial gradient calculation on the light intensity matrix on the left. The color bars represent gradient magnitudes, with units being arbitrary. In the right-hand plot, the gradient magnitude of the global background region corresponding to the smooth transition of light intensity on the left is dark blue, with values ​​close to 0. However, in the edge region corresponding to the dark spot on the left, due to the precipitous drop in light intensity, the spatial distribution difference is significantly amplified by the algorithm, presenting a bright yellow ring-shaped strong gradient band, with its local gradient magnitude suddenly increasing to over 15.

[0149] This feature extraction method, which incorporates two-dimensional machine vision, effectively highlights the edges of abnormal light intensity abrupt changes caused by local asymmetric contamination. Based on this, the system can accurately locate the physical boundaries of abnormal dark spot pixel clusters, thus providing an intuitive and objective data analysis basis for subsequent generation of dynamic spatial weight masks and implementation of targeted data weighting and masking.

[0150] Optionally, after extracting the spatial gradient features of all pixels, the system retrieves abnormal dark spot pixel clusters where the local light intensity attenuation gradient exceeds a preset uniformity threshold. The local light intensity attenuation gradient is the spatial gradient magnitude sequence of each pixel obtained from the aforementioned calculation. The preset uniformity threshold is a spatial smoothness measure that the system pre-sets based on the inherent luminous inhomogeneity of the optical system, the detector pixel crosstalk baseline, and the allowable spot smoothness under steady-state conditions. An abnormal dark spot pixel cluster refers to a set of pixels that are spatially continuous in a two-dimensional matrix coordinate system, and whose spatial gradient magnitudes, including those inside and at the edges, are generally greater than the preset uniformity threshold. This pixel cluster physically maps to a contaminated patch area on the surface of the optical window that experiences severe local aggregation, leading to a sharp decrease in transmittance.

[0151] Specifically, for the polluted interference areas located by the retrieval, the system generates a dynamic spatial weight mask based on the boundaries of the retrieved abnormal dark spot pixel clusters. This dynamic spatial weight mask is a two-dimensional multiplier factor matrix that perfectly corresponds to the two-dimensional pixel light intensity distribution matrix in terms of dimensionality. Its core function is to perform targeted spatial weight redistribution on the original light intensity data, thereby achieving logical shielding of harmful data and preservation of the value of valid data.

[0152] The system assigns a first preset extraction weight to the pixels corresponding to the abnormal dark spot pixel clusters, and a second preset extraction weight to global background pixels whose local light intensity attenuation gradient is less than or equal to the preset uniformity threshold. The first preset extraction weight is less than the second preset extraction weight. The first preset extraction weight is an attenuation factor assigned to pixels severely obscured by local contamination; in practical applications, to minimize local distortion interference, the first preset extraction weight is usually set to 0 or an extremely small positive real number. The global background pixels whose local light intensity attenuation gradient is less than or equal to the preset uniformity threshold refer to the set of effective pixels that are not affected by island-like contamination and whose light intensity attenuation patterns truly represent the global uniform decay characteristics of the optical window. The second preset extraction weight is a retention factor assigned to these effective pixels, typically set to a value of 1 to maintain the original signal intensity properties.

[0153] For example, in order to transform the above logical rules into executable low-level machine code, the system has a built-in formula for generating conditional branch masks:

[0154] ;

[0155] In the formula: In the dynamic spatial weight mask matrix generated by the system to characterize the current contamination distribution state of a specific channel, the corresponding to the first... line, number Extract the multiplier variable from the mask at the column coordinate position; A preset uniformity threshold parameter used to characterize the system curing process and to distinguish between smooth optical backgrounds and local anomalous abrupt changes; The first preset extraction weight variable is used to characterize the system's pre-configured first preset extraction weight variable that is effective for abnormal dark spot pixel clusters; The second preset extraction weight variable is used to characterize the system's pre-configured second preset extraction weight variable that is effective for globally smooth background pixels.

[0156] like Figure 7 This figure illustrates the two-dimensional distribution of the dynamic spatial weight mask generation matrix, which resists local contamination interference. The horizontal axis represents the pixel column index, and the vertical axis represents the pixel row index; both range from 1 to 60, visually covering the effective direct light acquisition area of ​​the area array photodetector. The color bars on the right side of the figure represent the mask weights, a dimensionless multiplier factor with values ​​between 0 and 1.

[0157] The large bright yellow area in the image represents a mask weight value of 1. This area corresponds to a globally smooth and effective background pixel where the local light intensity attenuation gradient is less than or equal to a preset uniformity threshold. The system extracts and retains the original light intensity data of this part of the pixel.

[0158] In the vicinity of coordinate row and column indices 20 and 40, and 45 and 15, clearly defined dark blue patches appear. Their mask weights are precisely attenuated and set to 0. These dark areas accurately correspond to aberrant dark spot pixel clusters retrieved based on spatial gradients. By generating this weight matrix with two-dimensional spatial discrimination capabilities, the system can use element-wise matrix multiplication in subsequent weighted integration processing to shield and isolate the severely distorted signals corresponding to the dark blue areas, while retaining only the effective light intensity data from the bright yellow areas to participate in the reconstruction of feature peak intensity.

[0159] This dynamic weight allocation method based on spatial morphology eliminates the local interference of asymmetric island contamination on light intensity attenuation assessment in the underlying data processing stage, and improves the computational objectivity and reconstruction accuracy of the dual-channel differential calibration system when dealing with complex and non-uniform spatial contamination deposition.

[0160] It is important to note that after acquiring a dynamic mask that is strictly aligned with the current real-time contamination pattern, the system uses the dynamic spatial weight mask to perform weighted integration on the two-dimensional pixel intensity distribution matrices of the two channels, respectively, to reconstruct the characteristic peak intensities. This weighted integration process changes the traditional method of indiscriminately summing all pixels. Instead, it performs element-wise Hadamard product operations on the original two-dimensional pixel intensity distribution matrix and the generated dynamic spatial weight mask matrix, and then sums the product results in two-dimensional space. Reconstructing the characteristic peak intensities separately means that this spatially discriminative operation effectively removes and fills in the concave distortion signals caused by local asymmetric contamination in the main measurement channel and the reference channel, thereby obtaining an equivalent characteristic peak intensity algebraic value that truly reflects the global uniform decay law of each channel.

[0161] For example, in order to accurately perform the numerical reconstruction operation of the spatial dimension, the system invokes the following spatial weighted integral algorithm formula:

[0162] ;

[0163] In the formula: This is used to characterize the reconstructed characteristic peak intensity value of a specific channel (i.e., the main measurement channel or the reference channel) after dynamic mask weighting calculation, which eliminates the interference of local asymmetric spatial contamination. and These represent the maximum effective pixel index constants contained in the direct light acquisition area of ​​the area array photodetector in the matrix row and column dimensions, respectively.

[0164] Optionally, after completing the spatial reconstruction calculation of the characteristic peak intensities of each channel, the system outputs the reconstructed characteristic peak intensities and performs the operation of constructing the difference signal and ratio signal. Outputting the reconstructed characteristic peak intensities here signifies that the morphological correction process of the signal in the spatial dimension has ended, and the data has returned to the scalar calculation system of spectral intensity. Performing the operation of constructing the difference signal and ratio signal refers to substituting the newly obtained reconstructed characteristic peak intensities of the main measurement channel and the reference channel into the system's two-dimensional calibration model. The ratio signal constructed at this time, having already eliminated the asymmetric bias caused by unilateral local large particle occlusion at the underlying two-dimensional physical pixel level, will have a real-time window decay factor with extremely high physical reliability.

[0165] For example, in order to reflect the final application logic of this reconstructed data in the overall algorithm chain, the system performs the calculation using the following updated ratio calibration formula:

[0166] ;

[0167] In the formula: The ratio of highly reliable reconstructed features used to characterize the final adoption of the system and its ability to resist interference from local non-uniform contamination. Used to characterize the intensity values ​​of the reconstructed feature peaks of the main measurement channel after weighted integration processing through a dedicated dynamic mask; Used to characterize the intensity values ​​of the reconstructed feature peaks of the reference channel after weighted integration processing is performed on the reference channel through its dedicated dynamic mask.

[0168] In summary, analysis of existing technologies reveals that traditional online optical measurement instruments generally employ a simple one-dimensional spectral integration algorithm when extracting feature signals. This traditional method rests on an overly idealized premise, assuming that contamination of the optical window always manifests as a uniform film thickness. However, in harsh real-world industrial environments, this ideal model often fails due to particulate aggregation, localized tar dripping, or irregular distribution of condensation droplets. Once a large dark spot appears in a localized area of ​​the reference channel, the indiscriminate global integration of the traditional algorithm leads to a sharp reduction in the extracted reference light intensity, causing the system to miscalculate an excessively large compensation coefficient, amplifying normal fluctuations in the main measurement channel into erroneous concentration peaks.

[0169] The technical solution disclosed in this embodiment activates the two-dimensional spatial vision capability of the area array photodetector, introducing two-dimensional spatial gradient analysis and morphological masking techniques from the field of digital image processing into the extraction process of spectral feature intensity. This solution accurately identifies abnormal dark spot pixel clusters by real-time retrieval of local high-frequency gradients, and implements targeted data weighting and masking by generating a dynamic spatial weight mask. This mechanism eliminates the negative impact of asymmetric island contamination on light intensity attenuation assessment, ensuring that the mapped real-time window decay factor always accurately represents the globally uniform attenuation characteristics of the optical window, effectively improving the robustness and measurement accuracy of the dual-channel differential calibration system when dealing with complex and non-uniform spatial contamination morphologies.

[0170] Example 4:

[0171] In the fields of industrial process control and environmental monitoring, such as continuous emission monitoring systems (CEMS) for flue gas in coal-fired power plants or online analyzers for chemical fluids, the light-transmitting components of optical measurement equipment often come into contact with complex fluids, leading to the continuous accumulation of particulate matter, oil, or salt spray on the surface of the optical window. This physical contamination not only obstructs light transmission, causing energy attenuation, but also causes severe scattering and refraction of the light signal propagation path, resulting in complex baseline dynamic drift and asymmetric peak distortion in the original light signal received by the detector. Existing technologies struggle to accurately decouple this dynamic nonlinear optical distortion while ensuring continuous online measurement, often requiring interruptions to the production process to introduce standard gases for offline recalibration.

[0172] To solve the above problems, such as Figure 8 As shown, this embodiment provides a dual-channel differential calibration system for gas detection resistant to optical window contamination and decay. The system in this embodiment employs a physical device design combining hardware optical path physical isolation and a low-level computing architecture. The dual-channel differential calibration system resistant to optical window contamination and decay includes: a dual-channel optical signal acquisition module, a feature extraction and stage determination module, a linear attenuation compensation module, a nonlinear distortion compensation module, and a verification and adaptive iteration module.

[0173] A dual-channel optical signal acquisition module is used to acquire the main measurement optical signal passing through the cavity of the measured medium through the main measurement channel, and to acquire a pure contaminated reference optical signal, which is only affected by optical window contamination, through a reference channel enclosed in a clean gas cavity. In the actual physical device, this module includes a broadband continuous light source, a high-precision optical beam splitter, and a lens group. The optical probe of the main measurement channel extends directly into the interior of the industrial measurement pipe, while the optical components of the reference channel are sealed in an external metal cavity. This cavity is continuously circulated with high-purity zero gas to maintain positive pressure, ensuring that the measured fluid does not enter the reference optical path. Furthermore, both the main measurement optical signal and the pure contaminated reference optical signal simultaneously include a direct transmitted light signal at a first preset angle and a forward scattered light signal within a second preset angle range. At the hardware receiving end, the system is equipped with an area array photodetector, and a spatial filter component consisting of an aperture stop and a ring-shaped light shield is mounted in front of its optical path to ensure, through physical isolation, that the central pixel of the detector receives only zero-degree direct photons, and the peripheral pixels receive only forward scattered photons deflected at a specified angle.

[0174] The feature extraction and stage determination module is used to extract anti-interference features from the acquired main measurement optical signal and the pure pollution reference optical signal, calculate the real-time scattering feature vector, and determine the current pollution stage and type of the optical window based on a pre-established three-dimensional correlation benchmark model that includes the mapping relationship between window transmittance attenuation, scattering feature distribution, and signal distortion. In the actual control unit, this module is implemented by a digital signal processor (DSP) or field-programmable gate array (FPGA) mounted on the main board of the field transmitter. The main control chip reads the digital matrix after analog-to-digital conversion from the area array detector, performs digital filtering to reduce high-frequency electromagnetic noise in the environment, calculates scattering variables in the processor's cache, and calls the three-dimensional correlation benchmark model data table embedded in non-volatile memory, such as a Flash chip, to quantitatively determine the current physical decay state of the optical window.

[0175] The linear attenuation compensation module is used to perform linear amplitude attenuation compensation on the main measured optical signal when the feature extraction and stage determination module determines that it is in the linear attenuation stage. This is done by constructing a two-dimensional calibration model based on the difference and ratio of the characteristic peak intensities of the direct transmitted light from the main measurement channel and the reference channel. In the early stages of optical window contamination, the deposits mainly appear as sparse particles, and the optical interference characteristics are primarily macroscopic energy absorption and occlusion. At this time, the main control chip calls the linear attenuation operation instruction set, compares the voltage integral difference and ratio output from the two channels after photoelectric conversion, calculates the compensation correction factor representing the global uniform attenuation characteristics, and directly multiplies and amplifies the vertical axis amplitude of the main measured spectral data proportionally, thereby restoring an accurate initial light intensity scale on the instrument display.

[0176] The nonlinear distortion compensation module is used when the feature extraction and stage determination module determines that the system is in a nonlinear scattering stage. It inputs the real-time scattering feature vector into the three-dimensional associated reference model, decouples it to generate a pure pollution distortion substrate, and performs full-dimensional compensation for scattering distortion, including baseline drift and peak distortion, on the main measured optical signal to reconstruct the true characteristic signal of the measured medium. This module is activated when a sudden, severe pollution event occurs in an industrial setting, such as high-viscosity oil mist condensation leading to strong optical refraction and multiple scattering. At the underlying algorithm level, the processor decouples pixel-by-pixel the asymmetric peak broadening caused by severe scattering and the data floating on the broadband spectral substrate. It subtracts these nonlinear pollution superpositions point-by-point from the mixed signal output by the detector, thereby restoring the absorption spectrum waveform that characterizes the true physical concentration of the chemical substance being measured.

[0177] The verification and adaptive iteration module performs a two-dimensional validity verification on the compensated true feature signal. If the verification passes, the final measurement result is output; if the verification fails, the parameter adaptive iteration process of the three-dimensional associated benchmark model is triggered, and the corresponding compensation module is controlled to re-execute compensation and verification after the parameters are updated. At the system hardware level, this module is linked to a known fixed-wavelength calibration light-emitting component, such as a solid-state laser or feature-emitting lamp, deployed within the system. When the system determines that the confidence of the currently reconstructed spectral features is insufficient, the main control chip will drive the calibration component to emit light, injecting a calibration beam with an absolute physical reference into the currently contaminated optical system. The processor compares the reconstructed output deviation of the calibration signal and uses a gradient optimization algorithm to correct the model weight coefficient matrix in the flash memory chip online, completing the system's self-correction.

[0178] The system described in this embodiment achieves physical isolation and synchronous extraction of pure pollution interference signals and the main measurement mixed signals from the underlying hardware architecture. Relying on a multi-stage compensation module and closed-loop iterative mechanism integrated within the core processor, the system possesses the ability to assess the evolution of optical window pollution status online. This device can perform precise ratio compensation for linear attenuation in the early stages of pollution and dynamically decouple the distorted substrate when complex nonlinear scattering occurs under harsh operating conditions. This allows optical measuring instruments to continuously resist the dynamic decay of the optical window without frequent interruptions to the process for manual cleaning or instrument return for calibration, effectively ensuring the online monitoring accuracy and maintenance-free operation capability of industrial analytical instruments in long-term complex environments.

[0179] Example 5:

[0180] Corresponding to the above embodiments, the present invention also proposes an electronic device.

[0181] like Figure 9The 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.

[0182] 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.

[0183] 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.

[0184] The memory 103 stores a computer program corresponding to the dual-channel differential calibration method for gas detection against optical window contamination decay according to the above embodiments of the present invention. This computer program is 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.

[0185] 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.

[0186] 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 dual-channel differential calibration method for gas detection to resist optical window contamination and decay, characterized in that, Includes the following steps: A dual-channel optical measurement system is constructed. The main measurement channel acquires the main measurement optical signal passing through the cavity of the measured medium, and the reference channel, which is enclosed in a clean gas cavity, acquires the pure contamination reference optical signal that is only affected by the contamination of the optical window. Both the main measurement optical signal and the pure contamination reference optical signal synchronously contain the direct transmission light signal at a first preset angle and the forward scattered light signal in a second preset angle range. Anti-interference features are extracted from the main measurement optical signal and the pure pollution reference optical signal, and a real-time scattering feature vector is calculated. The real-time scattering feature vector includes the proportion of scattered light intensity, the variance of the scattering angle distribution used to describe the spatial distribution characteristics of scattered light, and the spectral distribution coefficient of scattered light intensity used to describe the wavelength dependence of scattered light. Based on a pre-established three-dimensional correlation benchmark model that includes the mapping relationship between window transmittance attenuation, scattering feature distribution and signal distortion, the pollution stage and pollution type of the current optical window are determined. When it is determined that it is in the linear attenuation stage, a differential and ratio dual-dimensional calibration model is constructed based on the characteristic peak intensity of the direct transmitted light of the main measurement channel and the reference channel, and linear amplitude attenuation compensation is performed on the main measurement light signal. When it is determined that it is in the nonlinear scattering stage, the real-time scattering feature vector is input into the three-dimensional associated reference model, and a pure pollution distortion substrate is generated by decoupling. The main measured optical signal is then compensated for scattering distortion in all dimensions, including baseline drift and peak distortion, and the true feature signal of the measured medium is reconstructed. A two-dimensional validity check is performed on the compensated real feature signal. If the check passes, the final measurement result is output. If the verification fails, the parameter adaptive iteration process of the three-dimensional associated benchmark model is triggered, and compensation and verification are re-executed after the parameters are updated.

2. The method according to claim 1, characterized in that, The specific structural configuration of the dual-channel optical measurement system includes: The main measurement channel and the reference channel adopt a parallel coaxial optical path design and achieve beam splitting of the same source light source through a beam splitter; The system uses an area array photodetector to simultaneously acquire two signals. The photosensitive area of ​​the area array photodetector is divided into a direct light acquisition area located at the center and an annular scattered light acquisition area located at the periphery. The direct light acquisition area is constrained by an aperture stop to acquire only the direct transmitted light signal at the first preset angle, and the annular scattered light acquisition area is constrained by an annular stop and an angle constraint stop to acquire only the forward scattered light signal in the second preset angle range.

3. The method according to claim 1, characterized in that, The process of extracting anti-interference features from the acquired main measurement optical signal and the pure contamination reference optical signal, and calculating the real-time scattering feature vector, specifically includes: The optical signal of multiple consecutive cycles is synchronously acquired at a set sampling frequency. The variance of the single-cycle data at each sampling point is calculated relative to the average value of multiple cycles using the point-by-point variance statistical method. Abnormal data groups with variances greater than a preset variance threshold are removed, and the average signal of the effective data groups is extracted. The average signal is baseline corrected using a baseline correction algorithm, and high-density points in the characteristic peak region are reconstructed using a data interpolation algorithm. The reconstructed signal is fitted using a curve fitting algorithm to extract the center wavelength, peak intensity, and half-peak width of the characteristic peaks. The real-time scattering feature vector is then calculated by combining the direct transmitted light intensity and the forward scattered light intensity.

4. The method according to claim 1, characterized in that, The method for determining the current contamination stage of the optical window based on the pre-established three-dimensional correlation benchmark model is as follows: Calculate the real-time window transmittance and the proportion of scattered light intensity; The real-time window transmittance is obtained by the ratio of the real-time direct transmitted light intensity of the reference channel to the initial clean reference light intensity, and the proportion of scattered light intensity is obtained by dividing the forward scattered light intensity of the reference channel by the sum of the direct transmitted light intensity and the preset zero constant. When the real-time window transmittance is greater than or equal to the first transmittance threshold and the scattered light intensity ratio is less than or equal to the first scattering ratio threshold, it is determined that it is in the linear decay stage. When the real-time window transmittance is less than the first transmittance threshold and greater than or equal to the second transmittance threshold, and the proportion of scattered light intensity is greater than the first scattering proportion threshold and less than or equal to the second scattering proportion threshold, it is determined that it is in the weak nonlinear scattering stage. When the real-time window transmittance is less than the second transmittance threshold and the proportion of scattered light intensity is greater than the second scattering proportion threshold, it is determined that the system is in a severe nonlinear strong scattering stage; wherein, the first transmittance threshold is greater than the second transmittance threshold, and the second scattering proportion threshold is greater than the first scattering proportion threshold. The weak nonlinear scattering stage and the heavily nonlinear strong scattering stage are collectively referred to as the nonlinear scattering stage.

5. The method according to claim 1, characterized in that, The construction process of the three-dimensional associated benchmark model includes: The main measurement optical signal containing distortion is decomposed into multiple independent components. The main measurement optical signal is characterized as: the product of the amplitude attenuation coefficient and the true characteristic signal of the measured medium, plus the sum of the peak distortion component, the baseline drift component and the random noise component. Through pre-calibration experiments, functional mapping expressions between the amplitude attenuation coefficient, peak distortion component, baseline drift component, and the real-time scattering feature vector are fitted and constructed, and solidified to form the three-dimensional correlation benchmark model.

6. The method according to claim 1, characterized in that, The method for constructing a two-dimensional calibration model based on the difference and ratio of the characteristic peak intensities of the direct transmitted light from the main measurement channel and the reference channel specifically includes: Extract the characteristic peak intensity of the direct transmitted light from the main measurement channel and the characteristic peak intensity of the direct transmitted light from the reference channel; Construct a differential signal by taking the difference between the characteristic peak intensity of the direct transmitted light from the main measurement channel and the characteristic peak intensity of the direct transmitted light from the reference channel. A ratio signal is constructed by comparing the characteristic peak intensity of the direct transmitted light from the main measurement channel with the characteristic peak intensity of the direct transmitted light from the reference channel, and then mapping the ratio to obtain the real-time window decay factor. The linear amplitude attenuation compensation of the main measurement light signal is completed by dividing the characteristic peak intensity of the direct transmitted light from the main measurement channel by the real-time window decay factor.

7. The method according to claim 5, characterized in that, The decoupling process generates a pure contaminated distortion substrate, and performs full-dimensional compensation for scattering distortion, including baseline drift and peak shape distortion, on the main measured optical signal. The specific process is as follows: The real-time scattering feature vector is input into the function mapping expressions in the three-dimensional correlation benchmark model to calculate the real-time amplitude attenuation coefficient, real-time peak distortion component, and real-time baseline drift component caused only by optical window contamination at the current moment. The true characteristic signal of the measured medium is reconstructed as follows: the main measurement optical signal minus the real-time baseline drift component, the real-time peak distortion component, and the random noise component, and the resulting residual difference is divided by the real-time amplitude attenuation coefficient.

8. The method according to claim 1, characterized in that, The step of performing a two-dimensional validity check on the compensated true feature signal specifically includes: First-dimensional scattering feature consistency verification: The real feature signal obtained after reconstruction is subtracted from the original main measurement optical signal to generate a residual signal. The correlation coefficient between the residual signal and the pure contaminated distorted substrate is calculated. If the correlation coefficient is greater than or equal to the first correlation coefficient threshold, the consistency verification is determined to be passed. The second dimension of standard spectral line residual verification: the calibration module built into the system outputs a standard characteristic spectral line, the optical signal of the spectral line is collected and a nonlinear compensation process is performed, the relative residual between the compensated standard spectral line characteristic parameters and the inherent standard value is calculated, and if the relative residual is less than or equal to the second relative residual threshold, the residual verification is deemed to have passed.

9. The method according to claim 8, characterized in that, The parameter adaptive iteration process that triggers the three-dimensional associated benchmark model is as follows: If any item in the two-dimensional validity check fails, or if the pollution stage is determined to have switched, an adaptive iteration process is triggered. Using the standard characteristic spectral lines output by the calibration module as the iterative benchmark, an online parameter optimization algorithm is used to perform online iterative correction on the parameters of the function mapping expression in the three-dimensional associated benchmark model; The online iterative correction of the parameters aims to minimize the relative residual of the standard feature spectral lines after compensation. After the iteration is completed, the updated model parameters are fixed for use in the next compensation calculation.

10. The method according to claim 4, characterized in that, The method also includes an optical window predictive maintenance step based on a time series prediction network model: The system stores historical operating sequence data containing the real-time window transmittance and the proportion of scattered light intensity during the optical window's lifespan; the historical operating sequence data is input into a pre-trained time series prediction network model, which outputs a time series prediction curve of optical window degradation. Based on the time-series prediction curve, the remaining estimated time for the optical window to transition from the linear decay stage or the weak nonlinear scattering stage to the severe nonlinear strong scattering stage is calculated. When the remaining estimated time reaches a preset alarm time threshold, a predictive maintenance prompt containing window cleaning or replacement instructions is generated.

Images