A multi-source plume three-dimensional reconstruction and background decoupling method and system

Abstract

This invention provides a method and system for three-dimensional reconstruction and background decoupling of multi-source plumes, comprising: acquiring a two-dimensional column concentration matrix generated by imaging DOAS scanning; searching for primary / secondary peaks and estimating background noise based on the two-dimensional column concentration matrix; constructing a forward physical model based on the background noise; obtaining the optimal parameter set from the forward physical model through inversion, establishing a highly extended two-dimensional virtual mesh, and calculating and completing the upper-level plume concentration distribution missing due to hardware field-of-view limitations. This invention utilizes inversion to obtain the optimal parameter set (such as rise rate, diffusion coefficient, etc.) and establishes a highly extended virtual network for field-of-view extrapolation reconstruction. This method can mathematically complete the missing upper-level concentration distribution based on limited low-level observation data, significantly improving the accuracy and completeness of pollution flux calculation.

Description

Technical Field

[0001] This invention relates to the field of atmospheric environment monitoring technology, specifically to a method for inverting industrial emission gas concentrations, separating multiple sources, and reconstructing diffusion patterns based on Imaging Differential Absorption Spectroscopy (IDAS) technology. Background Technology

[0002] Imaging Differential Absorption Spectroscopy (DOAS) technology acquires the concentration distribution of a two-dimensional column of gas within a region through scanning, and is currently an important means of monitoring industrial pollution emissions. However, the following technical bottlenecks exist in practical applications.

[0003] Background interference is difficult to remove: The actual atmospheric environment contains complex light scattering and aerosol interference, resulting in background noise in the spectrum that is difficult to determine. Existing methods typically involve manually selecting a "clean area" to remove the background, which is highly subjective and inefficient, and produces significant errors on hazy days or under uneven lighting conditions.

[0004] Multiple overlapping sources cannot be distinguished: In chemical industrial parks, multiple chimneys (emission sources) are often close to each other. Traditional single Gaussian plume models can only handle a single source. When multiple plumes overlap in space, existing algorithms cannot decouple the emission intensity and diffusion parameters of each source.

[0005] Limited field of view due to hardware limitations: Due to the field of view (FOV) of the spectrometer or the mechanical limitations of the scanning pan-tilt unit, the imaging map often only covers the lower and middle part of the chimney, while the upper-level plume (which is also a key area for pollutant diffusion) is often outside the field of view. Existing technologies cannot use low-level data to extrapolate upper-level morphology, resulting in a significant underestimation of the final flux calculation. Summary of the Invention

[0006] The purpose of this invention is to solve the problem of difficulty in monitoring emission sources due to background interference, multiple source overlap, and limited field of view.

[0007] The present invention solves the above-mentioned technical problems through the following technical means:

[0008] A method for three-dimensional reconstruction and background decoupling of multi-source plumes includes: acquiring a two-dimensional column concentration matrix generated by imaging DOAS scanning; searching for primary / secondary peaks and estimating background noise for the two-dimensional column concentration matrix; constructing a forward physical model based on the background noise; obtaining the optimal parameter set from the forward physical model through inversion, establishing a highly extended two-dimensional virtual mesh, substituting the optimal parameter set into the forward physical model, and calculating the upper-level plume concentration distribution in the two-dimensional virtual mesh.

[0009] Furthermore, the process of searching for major / minor peaks and estimating background noise for the two-dimensional column concentration matrix is ​​as follows: S2.1 Data Preprocessing and Noise Suppression: Preprocessing the acquired measured discrete two-dimensional column concentration matrix data... D obs For quality control, outliers are first removed; then, a two-dimensional Gaussian low-pass filter is applied to smooth the matrix, resulting in a preprocessed concentration matrix. D curr ; S2.2 Environmental Background Baseline Statistics: Selection of Concentration Matrix D curr The edge region was used as a non-plume reference area. Statistical analysis was performed on the data from this reference area to construct a concentration-frequency histogram. Based on this histogram, the median or a specific quantile was selected as the background noise level. C bkg_set The initial estimate was obtained, and the standard deviation of the background noise in the reference area was calculated. As the noise floor level; S2.3 Iterative Main Source Search and Signal-to-Noise Ratio Determination: First, the concentration matrix... D curr Initialize to the current residual matrix D res Then, an iterative search loop is entered until there are no more significant emission sources in the current residual matrix, thus obtaining a list of candidate sources; S2.4 Dynamic Gaussian Mask Generation and Residual Update: For the first candidate source in the list... k One potential emission source, in order to ( x k , z k Construct a suppression function centered on ) M k ( x, z ):

[0010] in , The spatial diffusion scale parameter is used; the residual matrix is ​​updated using this mask: ; in, The horizontal and vertical coordinates represent the suppression intensity coefficient. The preset experience diffusion width; S2.5 Parameter Space Sort and Vector Assembly: After the iteration, all emission sources in the candidate source list are sorted from left to right according to their horizontal physical coordinate X, and the sorted source location parameters ( X s1 , Z s1 ),( X s2, Z s2 ...and map them together with preset physical empirical constants, and combine them with background noise. C bkg_set ， Finally, the initial parameter vector required for the multidimensional nonlinear optimization algorithm is assembled. P 0.

[0011] Furthermore, for any spatial location ( x , z The forward physical model is as follows:

[0012] in, B kg This refers to the global environmental background concentration. Y To identify the number of emission sources; Indicates the parameter set The decision i The root plume is a Gaussian diffusion model of a single plume, including source strength Amp and location ( X s , Z s ), Slope parameter i Rise i ), sinusoidal oscillation term ( W a , W f , W p ) and diffusion coefficient ( D k , D b ). For the first i The set of parameter vectors to be inverted corresponding to each emission source is given by the following formula:

[0013] No. i The Gaussian diffusion model distribution formula for a single root plume is:

[0014] in, Indicates source strength, Indicates the source location. Indicates the lateral and longitudinal diffusion scales. This represents the lifting parameter, used to describe the upward trend of the plume with altitude.

[0015] Furthermore, the optimal parameter set is obtained by inversion of the aforementioned forward physical model. Popt The specific constraint boundary conditions are: the boundary conditions of the global environmental background concentration term. B kg The lower limit is set to 0, and the upper limit is set to the measured discrete two-dimensional column concentration matrix data. D obs The maximum value; geometric location parameter boundaries, the spatial source location parameters of each emission source are limited to the actual physical space range corresponding to the image; the parameter vector to be inverted. Horizontal and vertical spatial diffusion scale parameters and All values ​​are restricted to positive values; under the above physical boundary constraints, a nonlinear least squares algorithm is used to obtain the measured discrete two-dimensional column concentration matrix data. D obs With the objective of minimizing the sum of squared residuals of the model's predicted data, all unknown parameters in the forward physical model are iteratively updated until the algorithm converges, ultimately yielding the optimal parameter set P. opt .

[0016] Furthermore, the specific steps of establishing a highly expanded two-dimensional virtual mesh and calculating and completing the upper-level plume concentration distribution missing due to hardware field-of-view limitations are as follows: Define a two-dimensional virtual mesh: Create a new two-dimensional virtual mesh ( X new , Z new Its vertical height H visual Expand the setting upwards; Forward computation of the model: Optimal parameter set P opt Substituting into the forward physical model, forward numerical extrapolation is performed on a two-dimensional virtual mesh to calculate the main concentration distribution of the plume across the entire field of view, thus obtaining reconstructed data; Concentration texture enhancement: Random Gaussian noise conforming to a statistical distribution is superimposed on the reconstructed data to simulate the texture features of the real atmosphere; Visualization output: Output a concentration distribution map containing the complete upper-level morphology, and plot the center diffusion trajectory lines of each source.

[0017] This invention also provides a multi-source plume 3D reconstruction and background decoupling system, comprising: Data acquisition and spatial mapping module: acquires the two-dimensional column concentration matrix generated by DOAS imaging scan; Multi-source emission characteristic adaptive initialization module: For a two-dimensional column concentration matrix, it searches for primary / secondary peaks and estimates background noise. Physics module construction module: Based on background noise, construct a forward physics model; Multi-parameter joint iterative inversion module: Obtains the optimal parameter set by inversion from the forward physical model; Virtual field of view expansion and shape reconstruction module: Establish a highly expanded two-dimensional virtual mesh, substitute the optimal parameter set into the forward physical model, and calculate the upper-level plume concentration distribution in the two-dimensional virtual mesh.

[0018] Furthermore, the process of searching for major / minor peaks and estimating background noise for the two-dimensional column concentration matrix is ​​as follows: S2.1 Data Preprocessing and Noise Suppression: Preprocessing the acquired measured discrete two-dimensional column concentration matrix data... D obs For quality control, outliers are first removed; then, a two-dimensional Gaussian low-pass filter is applied to smooth the matrix, resulting in a preprocessed concentration matrix. D curr ; S2.2 Environmental Background Baseline Statistics: Selection of Concentration Matrix D curr The edge region was used as a non-plume reference area. Statistical analysis was performed on the data from this reference area to construct a concentration-frequency histogram. Based on this histogram, the median or a specific quantile was selected as the background noise level. C bkg_set The initial estimate was obtained, and the standard deviation of the background noise in the reference area was calculated. As the noise floor level; S2.3 Iterative Main Source Search and Signal-to-Noise Ratio Determination: First, the concentration matrix... D curr Initialize to the current residual matrix D res Then, an iterative search loop is entered until there are no more significant emission sources in the current residual matrix, thus obtaining a list of candidate sources; S2.4 Dynamic Gaussian Mask Generation and Residual Update: For the first candidate source in the list... k One potential emission source, in order to ( x k , z k Construct a suppression function centered on ) M k ( x, z ):

[0019] in , The spatial diffusion scale parameter is used; the residual matrix is ​​updated using this mask: ; in, The horizontal and vertical coordinates represent the suppression intensity coefficient. The preset experience diffusion width; S2.5 Parameter Space Sort and Vector Assembly: After the iteration, all emission sources in the candidate source list are sorted from left to right according to their horizontal physical coordinate X, and the sorted source location parameters ( X s1 , Z s1 ),( X s2 , Z s2 ...and map them together with preset physical empirical constants, and combine them with background noise. C bkg_set ， Finally, the initial parameter vector required for the multidimensional nonlinear optimization algorithm is assembled. P 0.

[0020] Furthermore, for any spatial location (x, z), the forward physical model is:

[0021] in, B kg This refers to the global environmental background concentration. Y To identify the number of emission sources; Indicates the parameter set The decision i The root plume is a Gaussian diffusion model of a single plume, including source strength Amp and location ( X s , Z s ), Slope parameter i Rise i ), sinusoidal oscillation term ( W a , W f , W p ) and diffusion coefficient ( D k , D b ). For the first i The set of parameter vectors to be inverted corresponding to each emission source is given by the following formula:

[0022] No. i The Gaussian diffusion model distribution formula for a single root plume is:

[0023] in, Indicates source strength, Indicates the source location. Indicates the lateral and longitudinal diffusion scales. This represents the lifting parameter, used to describe the upward trend of the plume with altitude.

[0024] Furthermore, the optimal parameter set is obtained by inversion of the aforementioned forward physical model. P opt The specific constraint boundary conditions are: the boundary conditions of the global environmental background concentration term. B kg The lower limit is set to 0, and the upper limit is set to the measured discrete two-dimensional column concentration matrix data. D obs The maximum value; geometric location parameter boundaries, the spatial source location parameters of each emission source are limited to the actual physical space range corresponding to the image; the parameter vector to be inverted. Horizontal and vertical spatial diffusion scale parameters and All values ​​are restricted to positive values; under the above physical boundary constraints, a nonlinear least squares algorithm is used to obtain the measured discrete two-dimensional column concentration matrix data. D obs With the objective of minimizing the sum of squared residuals of the model's predicted data, all unknown parameters in the forward physical model are iteratively updated until the algorithm converges, ultimately yielding the optimal parameter set P. opt .

[0025] Furthermore, the specific steps of establishing a highly expanded two-dimensional virtual mesh and calculating and completing the upper-level plume concentration distribution missing due to hardware field-of-view limitations are as follows: Define a two-dimensional virtual mesh: Create a new two-dimensional virtual mesh ( X new , Z new Its vertical height H visual Expand the setting upwards; Forward computation of the model: Optimal parameters P opt Substituting into the forward physical model, forward numerical extrapolation is performed on a two-dimensional virtual mesh to calculate the main concentration distribution of the plume across the entire field of view, thus obtaining reconstructed data; Concentration texture enhancement: Random Gaussian noise conforming to a statistical distribution is superimposed on the reconstructed data to simulate the texture features of the real atmosphere; Visualization output: Output a concentration distribution map containing the complete upper-level morphology, and plot the center diffusion trajectory lines of each source.

[0026] The advantages of this invention are: 1. This invention overcomes hardware field-of-view limitations, achieving complete reconstruction of the entire plume morphology. Existing imaging DOAS devices, limited by the field of view (FOV) or mechanical scanning range, often fail to cover upper-level plumes, leading to underestimation of emission fluxes. This invention utilizes inversion to obtain the optimal parameter set (such as rise rate and diffusion coefficient), establishing a highly extended virtual network for field-of-view extrapolation reconstruction. This method can mathematically complete the missing upper-level concentration distribution based on limited low-level observation data, significantly improving the accuracy and completeness of pollution flux calculations.

[0027] 2. Solved the problem of decoupling multi-source overlapping signals in complex backgrounds. Addressing the issue of multiple adjacent chimney emission sources spatially overlapping in chemical industrial parks, which traditional single Gaussian models cannot handle, this invention constructs a multi-source joint model incorporating background factors. Through a nonlinear least squares algorithm, the geometric and diffusion parameters of multiple emission sources can be simultaneously inverted, thereby effectively separating overlapping plume signals and achieving accurate quantification of the intensity of each independent emission source.

[0028] 3. Adaptive removal of environmental background noise is achieved, enhancing anti-interference capabilities. Unlike traditional methods that rely on manual selection of "clean areas" for background subtraction due to subjectivity and inefficiency, this invention incorporates the surrounding background value as a parameter to be optimized into the inversion model for synchronous iterative solution. Combined with an adaptive initialization strategy based on edge data statistics, this invention can automatically remove background interference such as aerosol scattering under complex meteorological conditions such as haze and uneven lighting, significantly improving the signal-to-noise ratio of monitoring data.

[0029] 4. Improved convergence stability and accuracy of the nonlinear inversion algorithm. To address the sensitivity of nonlinear inversion to initial conditions, this invention employs an adaptive initialization strategy of "main peak search - dynamic masking - secondary peak search." In particular, the introduction of a dynamic Gaussian soft suppression mask, compared to hard truncation, eliminates the influence of the main peak while preserving the continuity of the matrix, effectively avoiding the misleading influence of artificial edges on subsequent searches, and ensuring that the multi-parameter joint iterative inversion can quickly converge to the global optimum. Attached Figure Description

[0030] Figure 1 This is a flowchart of the method according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the evolution of the multi-source iterative locking process based on dynamic trajectory masks in an embodiment of the present invention; Figure 3 This is a schematic diagram of the multi-source coupling model and spatial reconstruction principle under a limited field of view in an embodiment of the present invention; Figure 4 This is a comparison diagram of the field of view expansion and reconstruction effects in an embodiment of the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] This embodiment provides a method for three-dimensional reconstruction and background decoupling of multi-source plumes with limited field of view based on imaging DOAS. It constructs a multi-source joint forward physical model including background factors, utilizes a nonlinear least squares algorithm to simultaneously invert emission source parameters and environmental noise floor, and reconstructs the upper-level plume morphology outside the field of view based on the inverted parameters. The main scheme is as follows: Data acquisition: Obtain the two-dimensional concentration matrix generated by the imaging DOAS scan.

[0033] Adaptive initialization: The strategy of "main peak search - masking suppression - secondary peak search" is adopted to automatically lock the initial position of multiple emission sources and use edge data to predict background noise.

[0034] Joint Inversion Model: Construction The forward physics model will incorporate environmental background values. B kg As parameters to be optimized, they are solved iteratively in sync with the geometric and diffusion parameters of all plume models.

[0035] Field of view extrapolation reconstruction: The optimal parameter set is obtained by inversion, a highly extended two-dimensional virtual grid is established, and the high-altitude plume concentration distribution that is missing due to hardware field of view limitations is calculated and completed.

[0036] like Figure 1 As shown, the specific process is as follows: Step S1: Data Loading and Physical Space Mapping The DOAS imaging system was used to perform a two-dimensional scan of the area under test to obtain measured discrete two-dimensional column concentration matrix data. D obs Based on the scanning distance (e.g., 600 m) and vertical scanning height (e.g., 200 m), the original physical mesh is established. X, Z ).

[0037] The specific physical space mapping process is as follows: First, the straight-line distance L of the emission source obtained by the rangefinder (e.g., 600m) is combined with the horizontal field of view of the imaging spectrometer. and vertical field of view Calculate the scale of each pixel in the actual space. Let the number of pixels in the horizontal and vertical directions of the imaging matrix be... and The spatial resolution corresponding to a single pixel is shown in equations (1)-(2).

[0038] (1) (2) in, and These represent the physical scales in the horizontal and vertical directions, respectively. Furthermore, the image pixel coordinates (i, j) are mapped to the actual physical space coordinates (x, z), and the transformation relationship is shown in equations (3) and (4): (3) (4) Through the above mapping relationship, the transformation from discrete pixel space to continuous physical space is realized, thereby constructing a two-dimensional virtual mesh (X, Z) with actual spatial scale, which provides a foundation for subsequent plume modeling and parameter inversion.

[0039] Step S2: Multi-source adaptive initialization based on iterative residual analysis and dynamic mask To automatically extract accurate initial values ​​for multiple potential sources in complex industrial environments with large data volumes and high noise levels, and to avoid nonlinear optimization algorithms getting trapped in local optima, this step preprocesses the data based on statistical probability and a dynamic weight decay mechanism. Specifically, it reads the measured discrete two-dimensional column concentration matrix data in the discrete imaging network coordinate system constructed in step S1. D obs (With a size of M×N), the following denoising and multi-source adaptive iterative process is executed sequentially: S2.1 Data Preprocessing and Noise Suppression: The measured discrete two-dimensional column concentration matrix data obtained in step S1 is processed... D obs Quality control is performed. First, the 3-Sigma criterion is used to identify and remove outliers generated by the imaging sensor. Then, a two-dimensional Gaussian low-pass filter is applied to smooth the matrix to suppress high-frequency random shot noise and improve the signal-to-noise ratio in the low-concentration plume region. The calculation form is shown in equations (5)-(6). (5) (6) in," " indicates convolution operation, For Gaussian kernel scaling parameters, These are the horizontal and vertical coordinates. S2.2 Environmental Background Baseline Statistical Selection: Preprocessed Concentration Matrix Obtained in Step S2.1 D currThe edge region (e.g., the first 5 columns of data at physical coordinate X=0) is used as the non-plume reference region. Statistical analysis is performed on the data in this reference region to construct a concentration-frequency histogram. This process is similar to the 'grayscale histogram' statistical method in digital image processing: first, the distribution range of concentration values ​​within the reference region is determined and divided into K equally spaced concentration intervals (bins); then, all pixels within the reference region are traversed, and the number of pixels falling into each concentration interval is counted, thereby generating a frequency histogram reflecting the distribution characteristics of background noise. Based on this histogram, the median or a specific quantile (e.g., the 40%-60% quantile) is selected as the environmental background noise level. C bkg_set The initial estimate was obtained, and the standard deviation of the background noise in the reference area was calculated. As the noise floor level.

[0040] S2.3 Iterative Main Source Search and Signal-to-Noise Ratio Determination. First, the preprocessed concentration matrix obtained in step S2.1 is... D curr Initialize to the current "residual matrix" D res (Right now D res = D curr Then, an iterative search loop is entered until no significant emission sources remain in the current residual matrix. Global extremum search: in the current residual matrix D res Search for the global maximum point and record its coordinates. x k , z k and peak strength I peak .

[0041] Let the estimated environmental background value be The standard deviation of background noise is Then the signal-to-noise ratio of the current candidate peak is defined as shown in equation (7).

[0042] (7) Set a threshold. The determination criteria are shown in equation (8).

[0043] (8) If the above conditions are not met, it is considered that there is no significant emission source in the current residual matrix. The iterative search is terminated. A signal-to-noise ratio (SNR) threshold check is performed, and a SNR threshold is set. (For example Determine if the current peak intensity meets the requirements. If the peak intensity does not meet the requirement, it is considered that the valid signal has been fully extracted or the remaining signal is overwhelmed by noise, and the iteration loop terminates. If the peak intensity meets the requirement, then ( x k , z k ) was determined to be the first k The central location of each potential emission source is identified and stored in the candidate source list. The evolution of this multi-source iterative search and locking process is illustrated in the accompanying diagram. Figure 2 .

[0044] S2.4 Dynamic Gaussian Mask Generation and Residual Update. To continue identifying weak sources masked by strong sources after extraction, this step does not use a hard truncation method of directly setting the mask to zero, but instead constructs a dynamic Gaussian suppression mask. The specific iterative locking and matrix update process is as follows: Figure 2 As shown.

[0045] First, such as Figure 2 As shown in the top figure, read the current two-dimensional concentration matrix containing the multi-source interleaved superposition distribution or the residual matrix of the current iteration loop; Secondly, such as Figure 2 As shown in the middle figure, for the identified first... k Each source, with its central coordinates ( x k , z k A dynamic Gaussian suppression mask is constructed with as the center. M k ( x, z The calculation formula is shown in equation (9).

[0046] (9) in , This represents the spatial diffusion scale parameter.

[0047] Finally, as Figure 2 As shown in the base diagram, an inhibition intensity coefficient is introduced. The preset empirical diffusion width is used. The residual matrix is ​​updated using this mask, as shown in equation (10).

[0048] (10) This update method achieves adaptive suppression of the main peak region through continuous weight decay. Figure 2 The comparison of the bottom residual intensity shows that while eliminating interference from the identified main source signal (such as P1), the continuity characteristics of the physical space are fully preserved. Through this soft suppression operation, the influence of the identified main peak is eliminated, while the continuity characteristics of the matrix are preserved, avoiding the misleading influence of artificial edges caused by hard truncation on subsequent searches.

[0049] S2.5 Parameter Space Sorting and Vector Assembly. After iteration, all sources in the candidate source list are sorted from left to right according to their horizontal physical coordinate X to ensure the uniqueness of their physical meaning. The sorted source location parameters ( X s1 , Z s1 ), ( X s2 , Z s2 ...and map them together with preset physical empirical constants (such as initial diffusion coefficient, lift rate, and initial value of sinusoidal oscillation amplitude), and combine them with the background noise obtained in step S2.2. C bkg_set Finally, the initial parameter vector required for the multidimensional nonlinear optimization algorithm is assembled. P 0. The adaptive initialization step is now complete. The output parameter vector... P 0 will be directly used as the iterative starting point for the nonlinear least squares inversion in step S4 to avoid the algorithm getting trapped in local optima; while the estimated background noise in step S2.2 C bkg_set This information is then used as prior information to construct the environmental background estimate C in the forward physical model of step S2.3. bkg The constraint boundary was determined. This step enabled the extraction of data from measured discrete two-dimensional column concentration matrix data. D obs Effective connection to the parameter space of the forward physical model.

[0050] Step S3: Construction of a forward model of multi-source plumes including background terms To describe the superimposed distribution characteristics of multiple emission plumes in the observation space, a forward physical model incorporating environmental background data is constructed. For any spatial location ( x , z The forward physical model for calculating the concentration at point () is shown in equation (11): (11) in, B kg The global environmental background concentration term is used as a parameter to be inverted and participates in the optimization. Y To identify the number of emission sources. (Right now P i ) indicates the parameter set The decision i The root plume is a Gaussian diffusion model of a single plume, including source strength Amp and location ( X s , Z s), Slope, Rise, Sine oscillation term ( W a , W f , W p ) and diffusion coefficient ( D k , D b ). Let be the set of parameter vectors to be inverted corresponding to the i-th emission source, and its formula is shown in equation (12).

[0051] (12) No. i The Gaussian diffusion model distribution of a single root plume is shown in equation (13).

[0052] (13) in, This represents the source strength of the i-th emission source. Indicates the source location of the i-th emission source. Let represent the spatial diffusion scale parameters of the i-th plume in the horizontal and vertical directions, respectively. This represents the vertical lift correction function for the i-th plume, coupled with the height z. It is mainly used to quantitatively describe the spatial morphological evolution trend of the plume as it rises with vertical height under the combined effects of thermal and dynamic lift. To introduce dynamic sinusoidal oscillation parameters to characterize the lateral oscillation of the plume caused by wind disturbance, the model introduces a sinusoidal oscillation term function, as shown in Equation (14). This is to simulate the serpentine oscillation of the plume caused by wind turbulence and improve the goodness of fit.

[0053] (14) In the sinusoidal oscillation term shown in equation (14), This represents the initial value of the sinusoidal oscillation amplitude of the plume axis of the i-th emission source; This represents the spatial frequency of the oscillation as it varies with the vertical height z-axis. This represents the initial phase of the i-th plume at its source. By constraining this function, the plume oscillation caused by wind turbulence can be quantitatively characterized.

[0054] The constraint condition for the nonlinear joint inversion with multi-parameter constraints is to minimize the measured discrete two-dimensional column concentration matrix data. D obs The sum of squared residuals predicted by the model. Its formula is shown in Equation (15).

[0055] (15) In the nonlinear joint inversion objective function shown in equation (15), the symbol D obs The only specific reference is to the imaging DOAS system in the real field of view. H means Discrete two-dimensional column concentration matrix data obtained from actual internal measurements; and This represents the theoretical prediction matrix calculated iteratively from the current parameter vector to be optimized. The inversion objective is to minimize the measured data. D obs The sum of squared residuals between the data and the data predicted by the positive model.

[0056] Step S4: Setting parameter boundary conditions for nonlinear inversion under multi-parameter constraints. Environmental background values. B kg The lower limit is 0, and the upper limit is the maximum value of the observed data. The source location parameter is limited to the physical range of the image.

[0057] The diffusion parameter is constrained to be positive. Using a nonlinear least squares algorithm, with the objective of minimizing the sum of squared residuals between observed data and model predictions, all parameters are iteratively updated until convergence, yielding the optimal parameter set P. opt .

[0058] Step S5: Field of View Extrapolation Reconstruction and Statistical Feature Compensation Based on 2D Virtual Mesh. To overcome the spatial limitations of the physical hardware field of view and to complete the plume morphology in key upper-level regions lost due to measurement range truncation, this example employs a technique based on optimal physical parameter spatial extrapolation coupled with statistical residuals to reconstruct the overall field concentration. The specific reconstruction process and the principle of spatial mesh division are as follows: Figure 3 As shown, the implementation steps include the following four steps.

[0059] S5.1 Virtual Extended Network Spatial Definition: Establish a new two-dimensional virtual grid with a spatial range much larger than the original measurement network. X new , Z new ). Combination Figure 3 The reconstruction principle shown, due to hardware limitations of imaging DOAS instruments, has a true field of view vertical height of [missing information]. H means (For example, 200m), resulting in missing contamination data above this boundary. Therefore, this step will adjust the vertical height from the actual field of view. H means Extending vertically upwards to expand to the height of the field of view H visual (e.g., 400m), thereby constructing an extended grid space that can cover the entire thermal rise and spatial diffusion area.

[0060] S5.2 Forward derivation of model morphology: The optimal parameter set obtained from the inversion solution in step S4 P opt Substituting the source strength, spatial location, lift rate, diffusion coefficient, and sinusoidal oscillation of each independent source into the forward physical model described in step S3, in the following... Figure 3 The extended field of view height shown H visual Forward numerical extrapolation is performed within the range. At this point, the measured two-dimensional column concentration matrix from step S4 is then used... D obs The parameters such as the optimal source strength, spatial location, lift rate, and sinusoidal oscillation term obtained through joint iteration are seamlessly extended to the upper atmosphere, thereby mathematically supplementing the upper-altitude plume concentration distribution that is missing due to the hardware field of view truncation in the extrapolated region, and realizing full field-of-view reconstruction.

[0061] S5.3 Preservation of Statistical Characteristics and Compensation for Dynamic Perturbations: Since the concentration field extrapolated using the Gaussian diffusion forward model possesses ideal mathematical smoothness, this step compensates for high-frequency perturbations in the reconstructed field to eliminate its non-physical oversmoothing. Specifically, it extracts the real environmental background residuals and noise analysis characteristics (such as standard deviation) obtained in step S2.2. In the extrapolated reconstruction region of the virtual network, random Gaussian noise with the same statistical probability density distribution is superimposed to simulate the spatial correlation perturbations and high-frequency turbulence characteristics generated by the real atmosphere under instantaneous wind fields.

[0062] S5.4 Full-field shape and trajectory visualization output: Real field of view H means Concentration matrix measured in the interior and extended field of view H ext The extrapolated and reconstructed concentration matrix is ​​spatially assembled and smoothly connected, and the final output is as follows: Figure 3 Core Model F ( x,z As shown in the figure, a full-field two-dimensional concentration distribution map is generated, which includes automatic decoupling of environmental background and multi-source superposition distribution characteristics. At the same time, based on the inverted geometric center of each source, the center diffusion trajectory lines of independent emission sources (such as source 1 and source 2) are drawn in the reconstructed map, realizing intuitive and traceable visualization of multi-source overlapping pollution sources.

[0063] like Figure 4 As shown, Figure 4 The top-middle image shows the original input data (field of view limited to 200m). Figure 4 The lower figure is a reconstruction diagram based on the forward physics model prediction (with the field of view extended to 400m). As can be seen from the two figures, the diagram reconstructed by the forward physics model prediction of this application can clearly describe the central diffusion trajectory of each independent emission source, realizing intuitive and traceable visualization of multiple overlapping pollution sources.

[0064] Based on the method described in the above embodiments, this embodiment can achieve the following effects: This invention overcomes hardware field-of-view limitations, achieving complete reconstruction of the entire plume morphology. Existing imaging DOAS devices, limited by the field of view (FOV) or mechanical scanning range, often fail to cover upper-level plumes, leading to underestimation of emission fluxes. This invention utilizes inversion to obtain the optimal parameter set (such as rise rate and diffusion coefficient), establishing a highly extended virtual network for field-of-view extrapolation reconstruction. This method mathematically completes the missing upper-level concentration distribution based on limited low-level observation data, significantly improving the accuracy and completeness of pollution flux calculations.

[0065] This invention solves the problem of decoupling multi-source overlapping signals in complex backgrounds. Addressing the issue of multiple adjacent chimney emission sources spatially overlapping in chemical industrial parks, which traditional single Gaussian models cannot handle, this invention constructs a multi-source joint model incorporating background factors. Through a nonlinear least squares algorithm, the geometric and diffusion parameters of multiple emission sources can be simultaneously inverted, thereby effectively separating overlapping plume signals and achieving accurate quantification of the intensity of each independent emission source.

[0066] This invention achieves adaptive removal of environmental background noise, enhancing its anti-interference capability. Unlike traditional methods that rely on manual selection of "clean areas" for background subtraction due to subjectivity and inefficiency, this invention incorporates the surrounding background value as a parameter to be optimized into the inversion model for synchronous iterative solution. Combined with an adaptive initialization strategy based on edge data statistics, this invention can automatically remove background interference such as aerosol scattering under complex meteorological conditions such as haze and uneven lighting, significantly improving the signal-to-noise ratio of monitoring data.

[0067] This invention improves the convergence stability and accuracy of nonlinear inversion algorithms. To address the sensitivity of nonlinear inversion to initial conditions, it employs an adaptive initialization strategy of "main peak search - dynamic masking - secondary peak search." Specifically, a dynamic Gaussian soft suppression mask is introduced. Compared to hard truncation, this method eliminates the influence of the main peak while preserving the continuity of the matrix, effectively avoiding the misleading influence of artificial edges on subsequent searches, and ensuring that multi-parameter joint iterative inversion can quickly converge to the global optimum.

[0068] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for three-dimensional reconstruction and background decoupling of multi-source smoke plumes, characterized in that, include: Obtain the two-dimensional column concentration matrix generated by the DOAS imaging scan; for the two-dimensional column concentration matrix, search for the main / secondary peaks and estimate the background noise. Based on the background noise, a forward physical model is constructed; the optimal parameter set is obtained by inversion of the forward physical model, a highly extended two-dimensional virtual grid is established, the optimal parameter set is substituted into the forward physical model, and the upper-level plume concentration distribution is calculated in the two-dimensional virtual grid.

2. The method for three-dimensional reconstruction of multi-source plumes and background decoupling according to claim 1, characterized in that, The process of searching for major / minor peaks and estimating background noise for a two-dimensional column concentration matrix is ​​as follows: S2.1 Data Preprocessing and Noise Suppression: Preprocessing the obtained raw two-dimensional column concentration matrix... D raw For quality control, outliers are first removed; then, a two-dimensional Gaussian low-pass filter is applied to smooth the matrix, resulting in a preprocessed concentration matrix. D curr ; S2.2 Environmental Background Baseline Statistics: Selection of Concentration Matrix D curr The edge region was used as a non-plume reference area. Statistical analysis was performed on the data from this reference area to construct a concentration-frequency histogram. Based on this histogram, the median or a specific quantile was selected as the background noise level. C bkg_set The initial estimate was obtained, and the standard deviation of the background noise in the reference area was calculated. As the noise floor level; S2.3 Iterative Main Source Search and Signal-to-Noise Ratio Determination: First, the concentration matrix... D curr Initialize to the current residual matrix D res Then, an iterative search loop is entered until there are no more significant emission sources in the current residual matrix, thus obtaining a list of candidate sources; S2.4 Dynamic Gaussian Mask Generation and Residual Update: For the first candidate source in the list... k One potential emission source, in ( x k , z k Construct a suppression function centered on ) M k ( x, z ): in , The spatial diffusion scale parameter is used; the residual matrix is ​​updated using this mask: ； in, The horizontal and vertical coordinates represent the suppression intensity coefficient. The preset experience diffusion width; S2.5 Parameter Space Sort and Vector Assembly: After the iteration, all emission sources in the candidate source list are sorted from left to right according to their horizontal physical coordinate X, and the sorted source location parameters ( X s1 , Z s1 ),( X s2 , Z s2 ...and map them together with preset physical empirical constants, and combine them with background noise. C bkg_set ， Finally, the initial parameter vector required for the multidimensional nonlinear optimization algorithm is assembled. P 0.

3. The method for three-dimensional reconstruction of multi-source plumes and background decoupling according to claim 1 or 2, characterized in that, For any spatial location (x, z), the forward physical model is: in, B kg This refers to the global environmental background concentration. Y To identify the number of emission sources; Indicates the parameter set The decision i The root plume is a Gaussian diffusion model of a single root plume, including source strength Amp and location ( X s , Z s ), Slope, Rise, Sine oscillation term ( W a , W f , W p ) and diffusion coefficient ( D k , D b ); For the first i The set of parameter vectors to be inverted corresponding to each emission source is given by the following formula: No. i The Gaussian diffusion model distribution formula for a single root plume is: in, Indicates source strength, Indicates the source location. Indicates the lateral and longitudinal diffusion scales. This represents the lifting parameter, used to describe the upward trend of the plume with altitude.

4. The method for three-dimensional reconstruction of multi-source plumes and background decoupling according to claim 3, characterized in that, The process of obtaining the optimal parameter set by inversion from the aforementioned forward physics model is as follows: B kg The lower limit is 0, and the upper limit is set to the measured discrete two-dimensional column concentration matrix data. D obs The maximum value; the spatial source location parameters of each emission source are limited to the actual physical space range of the image; the diffusion parameter is limited to a positive value; a nonlinear least squares algorithm is used to obtain the measured discrete two-dimensional column concentration matrix data. D obs With the objective of minimizing the sum of squared residuals of the model's predicted data, all unknown parameters in the forward physical model are iteratively updated until the algorithm converges, ultimately yielding the optimal parameter set P. opt .

5. The method for three-dimensional reconstruction of multi-source plumes and background decoupling according to claim 4, characterized in that, The specific steps for establishing a highly expanded two-dimensional virtual mesh and calculating and completing the upper-level plume concentration distribution missing due to hardware field-of-view limitations are as follows: Define a two-dimensional virtual mesh: Create a new two-dimensional virtual mesh ( X new , Z new Its vertical height H visual Expand the setting upwards; Forward computation of the model: Optimal parameter set P opt Substituting into the forward physical model, forward numerical extrapolation is performed on a two-dimensional virtual mesh to calculate the main concentration distribution of the plume across the entire field of view, thus obtaining reconstructed data; Concentration texture enhancement: Random Gaussian noise conforming to a statistical distribution is superimposed on the reconstructed data to simulate the texture features of the real atmosphere; Visualization output: Output a concentration distribution map containing the complete upper-level morphology, and plot the center diffusion trajectory lines of each source.

6. A multi-source plume three-dimensional reconstruction and background decoupling system, characterized in that, include: Data acquisition and spatial mapping module: acquires the two-dimensional column concentration matrix generated by DOAS imaging scan; Multi-source emission characteristic adaptive initialization module: For a two-dimensional column concentration matrix, it searches for primary / secondary peaks and estimates background noise. Physics module construction module: Based on background noise, construct a forward physics model; Multi-parameter joint iterative inversion module: using measured discrete two-dimensional column concentration matrix data D obs With the goal of minimizing the sum of squared spatial residuals at the pixel level of the forward model prediction data, the optimal parameter set is obtained by performing nonlinear joint iterative inversion on the forward physical model. Virtual field of view expansion and shape reconstruction module: Establish a highly expanded two-dimensional virtual mesh, substitute the optimal parameter set into the forward physical model, and calculate the upper-level plume concentration distribution in the two-dimensional virtual mesh.

7. The multi-source plume three-dimensional reconstruction and background decoupling system according to claim 6, characterized in that, The process of searching for major / minor peaks and estimating background noise for a two-dimensional column concentration matrix is ​​as follows: S2.1 Data Preprocessing and Noise Suppression: Preprocessing the acquired measured discrete two-dimensional column concentration matrix data... D obs For quality control, outliers are first removed; then, a two-dimensional Gaussian low-pass filter is applied to smooth the matrix, resulting in a preprocessed concentration matrix. D curr ; S2.2 Environmental Background Baseline Statistics: Selection of Concentration Matrix D curr The edge region was used as a non-plume reference area. Statistical analysis was performed on the data from this reference area to construct a concentration-frequency histogram. Based on this histogram, the median or a specific quantile was selected as the background noise level. C bkg_set The initial estimate was obtained, and the standard deviation of the background noise in the reference area was calculated. As the noise floor level; S2.3 Iterative Main Source Search and Signal-to-Noise Ratio Determination: First, the concentration matrix... D curr Initialize to the current residual matrix D res Then, an iterative search loop is entered until there are no more significant emission sources in the current residual matrix, thus obtaining a list of candidate sources; S2.4 Dynamic Gaussian Mask Generation and Residual Update: For the first candidate source in the list... k One potential emission source, in ( x k , z k Construct a suppression function centered on ) M k ( x, z ): in , The spatial diffusion scale parameter is used; the residual matrix is ​​updated using this mask: ； in, The horizontal and vertical coordinates represent the suppression intensity coefficient. The preset experience diffusion width; S2.5 Parameter Space Sort and Vector Assembly: After the iteration, all emission sources in the candidate source list are sorted from left to right according to their horizontal physical coordinate X, and the sorted source location parameters ( X s1 , Z s1 ),( X s2 , Z s2 ...and map them together with preset physical empirical constants, and combine them with background noise. C bkg_set ， Finally, the initial parameter vector required for the multidimensional nonlinear optimization algorithm is assembled. P 0.

8. A multi-source plume three-dimensional reconstruction and background decoupling system according to claim 6 or 7, characterized in that, For any spatial location (x, z), the forward physical model is: in, B kg This refers to the global environmental background concentration. Y To identify the number of emission sources; Indicates the parameter set The decision i The root plume is a Gaussian diffusion model of a single root plume, including source strength Amp and location ( X s , Z s ), Slope, Rise, Sine oscillation term ( W a , W f , W p ) and diffusion coefficient ( D k , D b ); For the first i The set of parameter vectors to be inverted corresponding to each emission source is given by the following formula: No. i The Gaussian diffusion model distribution formula for a single root plume is: in, Indicates source strength, Indicates the source location. Indicates the lateral and longitudinal diffusion scales. This represents the lifting parameter, used to describe the upward trend of the plume with altitude.

9. A multi-source plume three-dimensional reconstruction and background decoupling system according to claim 8, characterized in that, The process of obtaining the optimal parameter set by inversion from the aforementioned forward physics model is as follows: B kg The lower bound is 0, and the upper bound is the maximum value of the observed data; the source location parameter is limited to the physical range of the image; the diffusion parameter is limited to positive values; using a nonlinear least squares algorithm, with the objective of minimizing the sum of squared residuals between the observed data and the model prediction data, all parameters are iteratively updated until convergence, yielding the optimal parameter set. P opt .

10. The multi-source plume three-dimensional reconstruction and background decoupling system according to claim 9, characterized in that, The specific steps for establishing a highly expanded two-dimensional virtual mesh and calculating and completing the upper-level plume concentration distribution missing due to hardware field-of-view limitations are as follows: Define a two-dimensional virtual mesh: Create a new two-dimensional virtual mesh ( X new , Z new Its vertical height H visual Expand the setting upwards; Forward computation of the model: Optimal parameters P opt Substituting into the forward physical model, forward numerical extrapolation is performed on a two-dimensional virtual mesh to calculate the main concentration distribution of the plume across the entire field of view, thus obtaining reconstructed data; Concentration texture enhancement: Random Gaussian noise conforming to a statistical distribution is superimposed on the reconstructed data to simulate the texture features of the real atmosphere; Visualization output: Output a concentration distribution map containing the complete upper-level morphology, and plot the center diffusion trajectory lines of each source.

Images