Factory exhaust emission intelligent inversion method and system based on satellite remote sensing image

Abstract

The invention discloses a factory exhaust emission intelligent inversion method and system based on a satellite remote sensing image, and the method comprises the steps: generating a target enhanced image and a fusion weight field through the radiation calibration, BRDF correction and local weight adaptive fusion by employing a multi-source satellite remote sensing image and meteorological and topographic data; generating a deviation-corrected concentration distribution field through forward physical simulation and a residual error correction network in combination with ground monitoring data; performing multi-scale decomposition and consistency analysis on the distribution field, generating a parameter adjustment atlas and triggering model updating; performing three-dimensional diffusion simulation based on the updated model and a real-time wind field, combining recursive least square assimilation to obtain a prediction correction vector set and prediction concentration distribution, and feeding back and adjusting a consistency verification strategy; and carrying out bounded bilinear combination on the parameter adjustment atlas and the prediction correction vector set to generate an observation weighting field, updating the fusion weighting field under freezing / unfreezing conditions, and forming a double-closed-loop interaction mechanism of a model updating path and a data weighting path.

Description

Technical Field

[0001] This invention relates to the field of airport airspace management technology, specifically to a method and system for intelligent inversion of factory exhaust emissions based on satellite remote sensing imagery. Background Technology

[0002] Currently, atmospheric pollutant concentration predictions mostly employ forward physics simulations based on numerical meteorological models and chemical transport models, supplemented by satellite remote sensing inversion data or ground monitoring data for correction. Some methods utilize data assimilation techniques to optimize simulation results and improve prediction accuracy.

[0003] For example, Chinese Patent Publication No. CN108519340A discloses an algorithm for inverting the spatiotemporal distribution of pollutant gas column concentrations using satellite remote sensing. This algorithm employs Differential Absorption Spectroscopy (DOAS) analysis to invert pollutant gas column concentrations, emphasizing spectral preprocessing and physical parameter correction. However, it still has the following shortcomings:

[0004] 1. Multi-source observation data vary greatly in terms of temporal resolution, spatial resolution, and observation perspective, and lack a unified spatiotemporal consistency processing mechanism;

[0005] 2. The physical model lacks the ability to dynamically adjust parameters based on real-time observations during operation, making it difficult to adapt to changes in meteorological conditions and pollution source characteristics;

[0006] 3. The data assimilation process is mostly a one-way path, which cannot form a double closed-loop structure for the collaborative optimization of model parameters and data weights, making it difficult to achieve a balance between prediction accuracy, stability and adaptability.

[0007] In view of this, the present invention provides a method and system for intelligent inversion of factory exhaust emissions based on satellite remote sensing imagery to solve the above problems. Summary of the Invention

[0008] The purpose of this invention is to provide a method and system for intelligent inversion of factory exhaust emissions based on satellite remote sensing imagery, addressing the lack of effective fusion and real-time adaptation of multi-source remote sensing and ground monitoring data in terms of spatiotemporal consistency, improved concentration prediction accuracy, and dynamic dual-closed-loop model-data collaborative updating.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] In a first aspect, the present invention provides an intelligent inversion method for factory exhaust gas emissions based on satellite remote sensing imagery, comprising the following steps:

[0011] S101: Based on terrain data, a three-dimensional coordinate system is constructed. Multi-source satellite remote sensing images and meteorological data are combined using a spatiotemporal reflectance fusion model based on radiometric calibration normalization, bidirectional reflectance distribution function correction, and local weight adaptation to generate target enhanced images and fusion weight fields corresponding to the spatial location of the images.

[0012] S102: The target enhancement image is used to generate a first concentration distribution field through a forward physical simulation model, and the first concentration distribution field is jointly fitted with the spatially registered ground monitoring data through a residual correction network to generate a bias-corrected second concentration distribution field.

[0013] S103: Multiscale decomposition and reconstruction of the second concentration distribution field yields the cross-scale residual tensor; the consistency index is calculated to generate the parameter adjustment map; and a forward physics simulation model that satisfies the update conditions is output.

[0014] S104: Input the updated forward physics simulation model and real-time wind field data into the three-dimensional diffusion simulation model to generate the predicted concentration distribution and perform recursive least squares assimilation with the new remote sensing observations to obtain the prediction correction vector set.

[0015] S105: The parameter adjustment map and the prediction correction vector set are combined in a bounded bilinear manner to generate the observation weighted field. The fusion weighted field is updated under preset conditions to form a double closed-loop structure of model update path and data weighting path.

[0016] As a preferred embodiment of the first aspect of the present invention, the application logic of the spatiotemporal reflectivity fusion model is as follows:

[0017] Radiometric calibration and normalization involves converting the radiometric values ​​of pixels in multi-source satellite remote sensing images to unify the original digital values ​​into surface reflectance, thereby obtaining a radiometrically calibrated image set to eliminate response differences between different sensors.

[0018] BRDF correction utilizes topographic data and meteorological conditions to calculate the observation geometry information of pixels, including solar azimuth, sensor azimuth, and zenith angle. Based on land cover type parameters, it corrects for angle effects on the radiometrically calibrated image set, generating a spectrally consistent image set to ensure the consistency of multi-source images in spectral space.

[0019] Local weight adaptive calculation: In the three-dimensional coordinate system, the residual variance of the spectral consistency image set in the spectral space is calculated in units of the pixel neighborhood; according to the inverse variance weighting principle, the fusion weight of each image in the neighborhood is dynamically allocated to obtain the local optimal fusion coefficient set;

[0020] Spatiotemporal reflectance fusion and weight field generation: The spectral consistency image set and the local optimal fusion coefficient set are input into the spatiotemporal reflectance fusion algorithm. The complementarity of high spatial resolution data and high temporal resolution data is used to generate target enhancement images. During the fusion process, the multi-source data weights of each pixel are recorded to form a fusion weight field that corresponds one-to-one with the pixel position.

[0021] As a preferred embodiment of the first aspect of the present invention, the application logic of the forward physics simulation model is as follows:

[0022] Using the factory emission source location, emission rate estimate, and meteorological data as simulation variables, and combined with the terrain constraints in the three-dimensional coordinate system, numerical simulation calculations are performed on the transport and diffusion process of pollutants in the atmosphere. The calculated pollutant concentrations corresponding to the pixel locations are stored in a regular grid form to form the first concentration distribution field matching the target enhancement image.

[0023] As a preferred embodiment of the first aspect of the present invention, the step of jointly fitting the first concentration distribution field with spatially registered ground monitoring data through a residual correction network includes:

[0024] The actual geographic coordinates of the ground monitoring stations are mapped to the grid coordinates of the first concentration distribution field. The discrete monitoring data are then projected onto a spatial grid consistent with the first concentration distribution field using bilinear interpolation or neighbor interpolation methods to obtain a registered monitoring data field.

[0025] Calculate the grid difference between the first concentration distribution field and the registered monitoring data field, and form a residual field based on the pixel position to characterize the systematic deviation and error characteristics between the physical simulation results and the actual observations;

[0026] The first concentration distribution field and the residual field are used as input factors of the residual correction network. The perturbation factors that the forward physics simulation model failed to capture are fitted and learned by the convolution feature extraction and mapping mechanism, and the deviation correction field is output.

[0027] The first concentration distribution field is superimposed on the deviation correction field pixel by pixel to obtain the second concentration distribution field.

[0028] As a preferred embodiment of the first aspect of the present invention, the logic for obtaining the cross-scale residual tensor is as follows:

[0029] The second concentration distribution field is decomposed into multiple scale sub-bands with different spatial resolutions and frequency components using a multi-scale decomposition algorithm; each scale sub-band corresponds to the characteristic manifestation of pollutant concentration changes at a specific spatial scale.

[0030] The residual components at each scale are obtained by performing difference calculations between each scale sub-band and the previous time or historical baseline concentration distribution field at the same scale.

[0031] The residual components at each scale are stacked according to their spatial location to form a cross-scale residual tensor containing scale, location, and amplitude information, which is used to characterize the deviation distribution pattern of the concentration field at different scales.

[0032] As a preferred embodiment of the first aspect of the present invention, the logic for obtaining the parameter adjustment spectrum is as follows:

[0033] The global consistency index is calculated based on the statistical characteristics of the cross-scale residual tensor and labeled as the consistency index result. The consistency index is used to quantify the overall matching degree between the current model prediction result and historical trends or actual monitoring data at multiple scales, thereby enhancing all aspects of the target image.

[0034] Based on the cross-scale residual tensor and consistency index results, the parameter adjustment amount corresponding to each spatial location is calculated. The parameter adjustment amount includes the emission source strength correction coefficient, the diffusion coefficient correction factor and / or the wind field correction vector.

[0035] The parameter adjustments are combined into a parameter adjustment map corresponding to a three-dimensional coordinate space grid. This parameter adjustment map is used to update the forward physics simulation model; specifically:

[0036] When the consistency index is higher than the preset update trigger threshold, the parameter adjustment map is invoked to locally or globally update the key parameters of the forward physics simulation model; otherwise, the current model parameters are kept unchanged to prevent frequent fluctuations in the prediction results.

[0037] Output the updated forward physics simulation model parameters, or maintain the existing parameter settings in the frozen state, while archiving the consistency index and parameter adjustment map.

[0038] As a preferred embodiment of the first aspect of the present invention, the application logic of the three-dimensional diffusion simulation model is as follows:

[0039] A three-dimensional diffusion simulation model based on the finite volume method or a high-precision convection-diffusion equation solver is used to combine the real-time wind field with the updated physical parameters to predict the diffusion concentration distribution of pollutants in the horizontal and vertical directions, thus obtaining the predicted concentration distribution field.

[0040] Using the predicted concentration distribution field as the prior field and the newly acquired remote sensing observation data as the observation field, a mapping matrix between the state vector and the observation vector is constructed.

[0041] The recursive least squares algorithm is used to update the state estimate time by time, and the observation confidence weight is introduced to reduce the impact of outliers. The output prediction correction vector set represents the amount of correction to be applied to each grid point. The prediction correction vector set is weighted and superimposed on the prediction concentration distribution field to obtain the fused prediction result, which is marked as the third concentration distribution field, and serves as the final concentration field after assimilation.

[0042] As a preferred technical solution of the first aspect of the present invention, it further includes, in step 104, if the final concentration field after multiple assimilations is not improved or the improvement is less than a set threshold, the prediction correction vector set is frozen to avoid invalid correction; when freezing, the prediction trend features are extracted and sent back to S103 to adjust the consistency index calculation strategy and update the triggering conditions, and while outputting the third concentration distribution field, the frozen state mark and the prediction trend features are marked.

[0043] As a preferred technical solution of the first aspect of the present invention, the parameter adjustment spectrum and the prediction correction vector set are calculated according to the bounded bilinear combination formula:

[0044] W(x,y,z)=α·P(x,y,z)+β·C(x,y,z);

[0045] Where: P(x,y,z) is the parameter adjustment spectrum, C(x,y,z) is the prediction correction vector set, W(x,y,z) is the observation weighted field; α and β are non-negative weight coefficients, and satisfy α+β≤1.

[0046] In a second aspect, the present invention provides an intelligent inversion system for factory exhaust emissions based on satellite remote sensing images. Based on the implementation of the first aspect, it includes a three-dimensional construction module, a physical simulation correction module, a cross-scale analysis module, a prediction assimilation module, and a weighted closed-loop update module. Data transmission between the modules is carried out via wired and / or wireless means.

[0047] The 3D construction module is used to construct a 3D coordinate system based on terrain data. It generates target enhanced images and fusion weight fields corresponding to the spatial location of the images by using a spatiotemporal reflectance fusion model based on radiometric calibration normalization, bidirectional reflectance distribution function correction and local weight adaptation from multi-source satellite remote sensing images and meteorological data.

[0048] The physical simulation correction module is used to generate a first concentration distribution field from the target enhancement image through a forward physical simulation model, and then use a residual correction network to jointly fit the first concentration distribution field with spatially registered ground monitoring data to generate a bias-corrected second concentration distribution field.

[0049] The cross-scale analysis module is used to decompose and reconstruct the cross-scale residual tensor of the second concentration distribution field at multiple scales, calculate the consistency index to generate parameter adjustment maps, and output a forward physics simulation model that meets the update conditions.

[0050] The prediction assimilation module inputs the updated forward physics simulation model and real-time wind field data into the three-dimensional diffusion simulation model to generate a predicted concentration distribution and performs recursive least squares assimilation with the new remote sensing observations to obtain a prediction correction vector set.

[0051] The weighted closed-loop update module generates an observation weighted field by combining the parameter adjustment map and the prediction correction vector set in a bounded bilinear combination. Under preset conditions, it updates the fused weighted field, forming a double closed-loop structure of the model update path and the data weighting path.

[0052] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0053] This invention automatically identifies and corrects the location and number of bad pixels in the detector through residual fitting, and updates key parameters such as slit function and spectral drift in real time, effectively reducing the error of instrument drift in the inversion results. By utilizing multi-gas spectral de-overlap and atmospheric Ring effect elimination techniques, the accuracy and stability of pollutant column concentration inversion are significantly improved. The wavelength-dependent atmospheric quality factor lookup table constructed by combining my country's atmospheric profile, surface emissivity, and detector attitude not only enhances the algorithm's adaptability to multiple platforms such as satellite, ground-based, and airborne, but also improves its environmental adaptability and result reliability under complex observation conditions, thereby ensuring the application value of the inversion data in scientific research and environmental monitoring. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0055] Figure 1 This is a flowchart of the intelligent inversion method for factory exhaust emissions according to the present invention.

[0056] Figure 2 This is a flowchart of the forward physics simulation and residual correction process of the present invention;

[0057] Figure 3 This is a flowchart of the dual closed-loop structure update mechanism method of the present invention;

[0058] Figure 4 This is a framework diagram of the intelligent inversion system for factory exhaust emissions according to the present invention. Detailed Implementation

[0059] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art. The drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0060] Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more exemplary embodiments. Numerous specific details are provided in the following description to give a full understanding of exemplary embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure may be practiced with one or more specific details omitted, or methods, components, steps, etc. In other instances, well-known structures, methods, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0061] Example 1

[0062] like Figure 1 As shown, this invention provides an intelligent inversion method for factory exhaust gas emissions based on satellite remote sensing imagery, comprising the following steps:

[0063] S101: Based on terrain data, a three-dimensional coordinate system is constructed. Multi-source satellite remote sensing images and meteorological data are combined using a spatiotemporal reflectance fusion model based on radiometric calibration normalization, bidirectional reflectance distribution function correction, and local weight adaptation to generate target enhanced images and fusion weight fields corresponding to the spatial location of the images.

[0064] It should be noted that data is collected before step S101, including multi-source satellite remote sensing imagery, meteorological data, and topographic data; among which:

[0065] Topographic data, based on a digital elevation model (DEM) or digital surface model (DSM), generates a three-dimensional geographic coordinate system for the target area, serving as the spatial reference for fusion calculations. Multi-source satellite remote sensing imagery includes multi-temporal multispectral / hyperspectral data with different spatial and temporal resolutions, which has undergone radiometric calibration, normalization, and geometric correction. Meteorological data includes temporal elements such as wind speed, wind direction, temperature, humidity, and atmospheric pressure for the target area, and is registered to a three-dimensional coordinate system.

[0066] Specifically, the application logic of the spatiotemporal reflectivity fusion model is as follows:

[0067] S1011: Radiometric calibration normalization, which converts the radiometric values ​​of pixels in multi-source satellite remote sensing images to unify the original digital values ​​(DN) into surface reflectance, and obtains a radiometric calibration image set to eliminate response differences between different sensors.

[0068] S1012: BRDF correction utilizes topographic data and meteorological conditions to calculate the observation geometry information of pixels, including solar azimuth, sensor azimuth, and zenith angle; based on land cover type parameters, it corrects the angle effect of the radiometric image set to generate a spectrally consistent image set to ensure the consistency of multi-source images in spectral space.

[0069] S1013: Local weight adaptive calculation. In the three-dimensional coordinate system, the residual variance of the spectral consistency image set in the spectral space is calculated in units of the pixel neighborhood. According to the inverse variance weighting principle, the fusion weight of each image in the neighborhood is dynamically allocated to obtain the local optimal fusion coefficient set.

[0070] S1014: Spatiotemporal reflectance fusion and weight field generation. The spectral consistency image set and the local optimal fusion coefficient set are input into the spatiotemporal reflectance fusion algorithm. The complementarity of high spatial resolution data and high temporal resolution data is used to generate target enhancement images. During the fusion process, the multi-source data weights of each pixel are recorded to form a fusion weight field that corresponds one-to-one with the pixel position.

[0071] S102: The target enhancement image is used to generate a first concentration distribution field through a forward physical simulation model, and the first concentration distribution field is jointly fitted with the spatially registered ground monitoring data through a residual correction network to generate a bias-corrected second concentration distribution field.

[0072] Specifically, the application logic of the forward physics simulation model is as follows:

[0073] Using the factory emission source location, emission rate estimate, and meteorological data as simulation variables, and combined with the terrain constraints in the three-dimensional coordinate system, numerical simulation calculations are performed on the transport and diffusion process of pollutants in the atmosphere based on the plume diffusion equation. The calculated pollutant concentrations corresponding to the pixel locations are stored in the form of a regular grid to form the first concentration distribution field matching the target enhancement image.

[0074] It should be noted that, in the process of exhaust gas emission concentration inversion, although forward physics simulation based solely on plume diffusion equations can reflect the overall trend of pollutant diffusion, it is prone to introducing systematic biases because initial parameters such as emission source strength and emission height are usually estimated. Meteorological field and topographic information have limitations in spatiotemporal resolution and are difficult to characterize small-scale disturbances. In addition, traditional physical models make simplistic assumptions in the expression of nonlinear processes such as turbulent transport, instantaneous emission fluctuations and chemical reactions, resulting in significant differences between simulation results and actual observations.

[0075] Therefore, a residual correction network needs to be introduced for the first concentration distribution field. This network is then jointly fitted with spatially registered ground monitoring data to learn and correct systematic biases in the physical model and nonlinear disturbances that are difficult to model explicitly. On the one hand, the global prediction capability provided by the physical model ensures that concentration distribution inference is still possible in areas without monitoring points. On the other hand, the high precision of the observation data effectively injects local real information into the global prediction, achieving a dual guarantee of accuracy and coverage. By using ground monitoring data from different regions, model adaptation can be completed quickly, avoiding the high-cost process of rebuilding a complex physical model.

[0076] To further explain, such as Figure 2 As shown, the application logic of the residual correction network is as follows:

[0077] The actual geographic coordinates of the ground monitoring stations are mapped to the grid coordinates of the first concentration distribution field. The discrete monitoring data are then projected onto a spatial grid consistent with the first concentration distribution field using bilinear interpolation or neighbor interpolation methods to obtain a registered monitoring data field.

[0078] The grid difference between the first concentration distribution field and the registered monitoring data field is calculated. The grid difference is used to form a residual field through the pixel position. The residual field is used to characterize the systematic deviation and error characteristics between the physical simulation results and the actual observation.

[0079] The first concentration distribution field and the residual field are used as inputs and fed into a residual correction network based on a deep neural network for fitting and learning. This network learns complex perturbation factors (such as small-scale turbulence, instantaneous emission fluctuations, etc.) that cannot be fully captured by the physical model through convolutional feature extraction and mapping, and outputs the deviation correction field.

[0080] The first concentration distribution field is superimposed on the bias correction field pixel by pixel to obtain the second concentration distribution field. This distribution field retains the interpretable structure of the physical model and introduces data-driven correction, thus outperforming simple physical simulation or statistical fitting methods in overall accuracy.

[0081] S103: Perform multi-scale decomposition and reconstruction on the second concentration distribution field to generate a cross-scale residual tensor. Calculate the consistency index based on the residual tensor and form a parameter adjustment map. When the consistency index meets the update trigger condition, call the parameter adjustment map to update the forward physics simulation model. When the consistency index meets the freezing condition and the continuous freezing time meets the preset time window, stop updating and maintain the current model parameters.

[0082] It should be noted that existing concentration inversion methods usually keep the model parameters fixed after they are set, and cannot be dynamically corrected according to monitoring data. This leads to a rapid decline in the model prediction accuracy when meteorological conditions or emission source status change. Therefore, a consistency index evaluation mechanism based on multi-scale decomposition is introduced, which can comprehensively capture the changing trend of the concentration field at different spatial scales and dynamically update the model in a targeted manner through parameter adjustment maps.

[0083] Specifically, the logic for obtaining the cross-scale residual tensor is as follows:

[0084] A multi-scale decomposition algorithm (such as discrete wavelet transform, multi-resolution analysis, or Laplace pyramid decomposition) is used to decompose the second concentration distribution field into multiple scale sub-bands with different spatial resolutions and frequency components; each scale sub-band corresponds to the characteristic manifestation of pollutant concentration changes at a specific spatial scale.

[0085] The residual components at each scale are obtained by performing difference calculations between each scale sub-band and the previous time or historical baseline concentration distribution field at the same scale.

[0086] The residual components at each scale are stacked according to their spatial location to form a cross-scale residual tensor containing scale, location, and amplitude information, which is used to characterize the deviation distribution pattern of the concentration field at different scales. This cross-scale residual tensor not only preserves the spatial distribution pattern of multi-scale concentration changes but also characterizes the deviation features at different scales, serving as the core basis for subsequent consistency assessment and parameter correction.

[0087] Further explanation of the logic for obtaining the parameter adjustment spectrum:

[0088] Based on the statistical characteristics of the cross-scale residual tensor (such as mean squared residual, standard deviation, structural similarity index SSIM, etc.), the global consistency index is calculated. The consistency index is used to measure the degree of matching between the current model output and the actual monitoring trend at multiple scales. The value range is [0,1], and the higher the value, the better the consistency. It is used to quantify the overall degree of matching between the current model prediction results and historical trends or actual monitoring data at multiple scales.

[0089] Based on the results of cross-scale residual tensors and consistency indices, the parameter adjustment amounts (such as emission source strength correction coefficient, diffusion coefficient correction factor, wind field correction vector, etc.) corresponding to each spatial location are calculated; the parameter adjustment amounts are combined into a parameter adjustment map corresponding to the three-dimensional coordinate space grid; the forward physical simulation model is updated based on the parameter adjustment map.

[0090] When the consistency index is higher than the preset update trigger threshold, the parameter adjustment map is invoked to update the key parameters of the forward physics simulation model (including emission source strength, diffusion coefficient, wind field distribution, etc.) locally or globally.

[0091] In other cases, stop model updates and maintain the current parameter settings to prevent frequent parameter fluctuations from causing prediction instability.

[0092] The updated forward physics simulation model parameters are output, and the consistency index and parameter adjustment map are archived for use in the subsequent three-dimensional diffusion simulation and data assimilation process (S104).

[0093] S104: Input the updated forward physics simulation model output and real-time wind field data into the three-dimensional diffusion simulation model to obtain the predicted concentration distribution field. Then, use the recursive least squares assimilation method to fuse the predicted concentration distribution field with the newly acquired remote sensing observation data to generate a prediction correction vector set and a third concentration distribution field. If the target performance index is not improved after multiple consecutive assimilations, freeze the prediction correction vector set and back the prediction trend characteristics to step S103 to adjust the consistency verification strategy.

[0094] It should be noted that existing pollutant diffusion prediction methods mostly employ single-batch assimilation or offline simulation, which cannot achieve real-time prediction-correction closed loops. Therefore, by using a recursive least squares assimilation method, the prediction results are corrected in real time with each batch of observation data, while retaining historical weight information to maintain convergence stability. At the same time, a freeze and backpropagation mechanism is introduced to avoid over-correction when the model error tends to stabilize, thereby reducing computational overhead and maintaining prediction stability.

[0095] Specifically, the application logic of the three-dimensional diffusion simulation model is as follows:

[0096] A three-dimensional diffusion simulation model based on the finite volume method or a high-precision convection-diffusion equation solver is used to combine the real-time wind field with the updated physical parameters to predict the diffusion concentration distribution of pollutants in the horizontal and vertical directions, thus obtaining the predicted concentration distribution field.

[0097] Using the predicted concentration distribution field as the prior field and the newly acquired remote sensing observation data as the observation field, a mapping matrix between the state vector and the observation vector is constructed.

[0098] The recursive least squares algorithm is used to update the state estimate time by time, and the observation confidence weight is introduced to reduce the impact of outliers; the output prediction correction vector set represents the amount of correction to be applied to each grid point.

[0099] The weighted summation of the predicted correction vector set is applied to the predicted concentration distribution field to obtain the fused prediction result, which is labeled as the third concentration distribution field and serves as the final concentration field after assimilation. This also includes freezing the predicted correction vector set if, after multiple consecutive assimilations, performance indicators (such as the rate of decrease in mean squared error and the rate of improvement in structural similarity) do not improve or the improvement is below a set threshold, to avoid ineffective corrections. During freezing, prediction trend features (such as the direction of local deviation changes and spatial distribution patterns) are extracted and fed back to S103 to adjust the consistency index calculation strategy and update the triggering conditions. Simultaneously with outputting the third concentration distribution field, the frozen state and prediction trend features are also labeled.

[0100] S105: The observation weighted field is calculated by combining the parameter adjustment map and the prediction correction vector set in a bounded bilinear combination. When the observation time window ends or the number of iterations reaches the upper limit, the observation weighted field is frozen and the third concentration distribution field is output as the result of this round of inversion. When the consistency index meets the update trigger condition again, the freeze is lifted and the fusion weighted field is updated using the observation weighted field, thus forming a double closed-loop interactive structure that includes the model update path and the data weighting path.

[0101] It should be noted that the dual-closed-loop structure can solve the problem of single-channel update in the model update and observation data fusion of existing technologies. By fusing information from the model update path (parameter adjustment map) and the data assimilation path (prediction correction vector set), a stable and adaptive observation weighted field is generated. The dynamics and stability are balanced through the freeze / thaw mechanism, thus forming a dual-closed-loop interactive structure.

[0102] Specifically, such as Figure 3 As shown, the logic for obtaining the double closed-loop structure of the model update path and the data weighting path is as follows:

[0103] The observation weighted field is generated by a bounded bilinear combination of the parameter adjustment map and the prediction correction vector set. The current consistency index and its changing trend are read, and the preset observation time window parameters and the upper limit of the number of iterations are loaded. Among them, the parameter adjustment map describes the local or global parameter correction of the forward physics simulation model at different spatial locations; the prediction correction vector set reflects the observation correction information obtained based on the recursive least squares assimilation method.

[0104] The parameter adjustment map and the predicted correction vector set are calculated using the bounded bilinear combination formula:

[0105] W(x,y,z)=α·P(x,y,z)+β·C(x,y,z);

[0106] Wherein: P(x,y,z) is the parameter adjustment spectrum, C(x,y,z) is the prediction correction vector set, and W(x,y,z) is the observation weighted field; α and β are non-negative weight coefficients, and satisfy +≤1 to ensure that the combined result will not produce outliers outside the range; the result contains both model optimization information and observation correction information, which can directly reflect the comprehensive weight distribution of the model and observation in space.

[0107] During the iteration process, the observed weighted field is continuously updated and its convergence characteristics are evaluated;

[0108] The freeze operation is performed when any of the following conditions are met: the observation time window ends; the number of iterations reaches the upper limit; or the change of the weighted field in several consecutive iterations is lower than a set threshold.

[0109] After freezing, the observation weighted field remains in its current state and is used to generate the final concentration inversion results for this round.

[0110] Using the frozen observation weighted field, the third concentration distribution field is weighted and corrected to obtain the final concentration distribution as the result of this round of inversion. The final concentration distribution includes both the multi-scale consistency correction of the forward physics simulation model and the assimilation optimization of the observation data, ensuring the comprehensiveness and stability of the prediction results.

[0111] While frozen, the consistency index is continuously monitored; when the consistency index meets the update trigger condition again, the freeze is lifted, and the fusion weight field is optimized using the new observation weight field; at this time, the synchronous optimization of the model update path and the data weight path is realized, forming a double closed-loop structure of the model update path and the data weight path.

[0112] Example 2

[0113] like Figure 4 As shown in the figure, the parts not described in detail in this embodiment are as shown in Embodiment 1. This embodiment provides an intelligent inversion system for factory exhaust emissions based on satellite remote sensing images, including a three-dimensional construction module, a physical simulation correction module, a cross-scale analysis module, a prediction assimilation module, and a weighted closed-loop update module. Data is transmitted between the modules via wired and / or wireless means.

[0114] The 3D construction module is used to construct a 3D coordinate system based on terrain data. It generates target enhanced images and fusion weight fields corresponding to the spatial location of the images by using a spatiotemporal reflectance fusion model based on radiometric calibration normalization, bidirectional reflectance distribution function correction and local weight adaptation from multi-source satellite remote sensing images and meteorological data.

[0115] Specifically, the application logic of the spatiotemporal reflectivity fusion model is as follows:

[0116] Radiometric calibration and normalization involves converting the radiometric values ​​of pixels in multi-source satellite remote sensing images to unify the original digital values ​​into surface reflectance, thereby obtaining a radiometrically calibrated image set to eliminate response differences between different sensors.

[0117] BRDF correction utilizes topographic data and meteorological conditions to calculate the observation geometry information of pixels, including solar azimuth, sensor azimuth, and zenith angle. Based on land cover type parameters, it corrects for angle effects on the radiometrically calibrated image set, generating a spectrally consistent image set to ensure the consistency of multi-source images in spectral space.

[0118] Local weight adaptive calculation: In the three-dimensional coordinate system, the residual variance of the spectral consistency image set in the spectral space is calculated in units of the pixel neighborhood; according to the inverse variance weighting principle, the fusion weight of each image in the neighborhood is dynamically allocated to obtain the local optimal fusion coefficient set;

[0119] Spatiotemporal reflectance fusion and weight field generation: The spectral consistency image set and the local optimal fusion coefficient set are input into the spatiotemporal reflectance fusion algorithm. The complementarity of high spatial resolution data and high temporal resolution data is used to generate target enhancement images. During the fusion process, the multi-source data weights of each pixel are recorded to form a fusion weight field that corresponds one-to-one with the pixel position.

[0120] The physical simulation correction module is used to generate a first concentration distribution field from the target enhancement image through a forward physical simulation model, and then use a residual correction network to jointly fit the first concentration distribution field with spatially registered ground monitoring data to generate a bias-corrected second concentration distribution field.

[0121] Specifically, the application logic of the forward physics simulation model is as follows:

[0122] Using the factory emission source location, emission rate estimate, and meteorological data as simulation variables, and combined with the terrain constraints in the three-dimensional coordinate system, numerical simulation calculations are performed on the transport and diffusion process of pollutants in the atmosphere. The calculated pollutant concentrations corresponding to the pixel locations are stored in a regular grid form to form the first concentration distribution field matching the target enhancement image.

[0123] Furthermore, the joint fitting of the first concentration distribution field with spatially registered ground monitoring data using a residual correction network includes:

[0124] The actual geographic coordinates of the ground monitoring stations are mapped to the grid coordinates of the first concentration distribution field. The discrete monitoring data are then projected onto a spatial grid consistent with the first concentration distribution field using bilinear interpolation or neighbor interpolation methods to obtain a registered monitoring data field.

[0125] Calculate the grid difference between the first concentration distribution field and the registered monitoring data field, and form a residual field based on the pixel position to characterize the systematic deviation and error characteristics between the physical simulation results and the actual observations;

[0126] The first concentration distribution field and the residual field are used as input factors of the residual correction network. The perturbation factors that the forward physics simulation model failed to capture are fitted and learned by the convolution feature extraction and mapping mechanism, and the deviation correction field is output.

[0127] The first concentration distribution field is superimposed on the deviation correction field pixel by pixel to obtain the second concentration distribution field.

[0128] The cross-scale analysis module is used to decompose and reconstruct the cross-scale residual tensor of the second concentration distribution field at multiple scales, calculate the consistency index to generate parameter adjustment maps, and output a forward physics simulation model that meets the update conditions.

[0129] Specifically, the logic for obtaining the cross-scale residual tensor is as follows:

[0130] The second concentration distribution field is decomposed into multiple scale sub-bands with different spatial resolutions and frequency components using a multi-scale decomposition algorithm; each scale sub-band corresponds to the characteristic manifestation of pollutant concentration changes at a specific spatial scale.

[0131] The residual components at each scale are obtained by performing difference calculations between each scale sub-band and the previous time or historical baseline concentration distribution field at the same scale.

[0132] The residual components at each scale are stacked according to their spatial location to form a cross-scale residual tensor containing scale, location, and amplitude information, which is used to characterize the deviation distribution pattern of the concentration field at different scales.

[0133] Further explanation of the logic for obtaining the parameter adjustment spectrum:

[0134] The global consistency index is calculated based on the statistical characteristics of the cross-scale residual tensor and labeled as the consistency index result. The consistency index is used to quantify the overall matching degree between the current model prediction result and historical trends or actual monitoring data at multiple scales, thereby enhancing all aspects of the target image.

[0135] Based on the cross-scale residual tensor and consistency index results, the parameter adjustment amount corresponding to each spatial location is calculated. The parameter adjustment amount includes the emission source strength correction coefficient, the diffusion coefficient correction factor and / or the wind field correction vector.

[0136] The parameter adjustments are combined into a parameter adjustment map corresponding to a three-dimensional coordinate space grid. This parameter adjustment map is used to update the forward physics simulation model; specifically:

[0137] When the consistency index is higher than the preset update trigger threshold, the parameter adjustment map is invoked to locally or globally update the key parameters of the forward physics simulation model; otherwise, the current model parameters are kept unchanged to prevent frequent fluctuations in the prediction results.

[0138] Output the updated forward physics simulation model parameters, or maintain the existing parameter settings in the frozen state, while archiving the consistency index and parameter adjustment map.

[0139] The prediction assimilation module inputs the updated forward physics simulation model and real-time wind field data into the three-dimensional diffusion simulation model to generate a predicted concentration distribution and performs recursive least squares assimilation with the new remote sensing observations to obtain a prediction correction vector set.

[0140] Specifically, the application logic of the three-dimensional diffusion simulation model is as follows:

[0141] A three-dimensional diffusion simulation model based on the finite volume method or a high-precision convection-diffusion equation solver is used to combine the real-time wind field with the updated physical parameters to predict the diffusion concentration distribution of pollutants in the horizontal and vertical directions, thus obtaining the predicted concentration distribution field.

[0142] Using the predicted concentration distribution field as the prior field and the newly acquired remote sensing observation data as the observation field, a mapping matrix between the state vector and the observation vector is constructed.

[0143] The recursive least squares algorithm is used to update the state estimate time by time, and the observation confidence weight is introduced to reduce the impact of outliers. The output prediction correction vector set represents the amount of correction to be applied to each grid point. The prediction correction vector set is weighted and superimposed on the prediction concentration distribution field to obtain the fused prediction result, which is marked as the third concentration distribution field, and serves as the final concentration field after assimilation.

[0144] Further explanation includes that, in step 104, if the final concentration field after multiple assimilations does not improve or the improvement is less than a set threshold, the prediction correction vector set is frozen to avoid invalid correction; when freezing, the prediction trend features are extracted and sent back to S103 to adjust the consistency index calculation strategy and update the triggering conditions, and when outputting the third concentration distribution field, the frozen state mark and the prediction trend features are marked.

[0145] The weighted closed-loop update module generates an observation weighted field by combining the parameter adjustment map and the prediction correction vector set in a bounded bilinear combination. Under preset conditions, it updates the fused weighted field, forming a double closed-loop structure of the model update path and the data weighting path.

[0146] Specifically, the parameter adjustment map and the predicted correction vector set are calculated using a bounded bilinear combination formula:

[0147] W(x,y,z)=α·P(x,y,z)+β·C(x,y,z);

[0148] Where: P(x,y,z) is the parameter adjustment spectrum, C(x,y,z) is the prediction correction vector set, W(x,y,z) is the observation weighted field; α and β are non-negative weight coefficients, and satisfy α+β≤1.

[0149] This embodiment provides an intelligent inversion system for factory exhaust emissions based on satellite remote sensing imagery, used to execute the intelligent inversion method for factory exhaust emissions based on satellite remote sensing imagery provided in the above embodiments of the present invention. The specific methods and processes for implementing the corresponding functions of each structure in the intelligent inversion system for factory exhaust emissions based on satellite remote sensing imagery are detailed in the embodiments of the intelligent inversion method for factory exhaust emissions based on satellite remote sensing imagery, and will not be repeated here.

[0150] Example 3

[0151] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0152] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless network. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0153] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0154] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0155] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only one method, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0156] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0157] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0158] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A smart inversion method for factory exhaust gas emissions based on satellite remote sensing imagery, characterized in that, Includes the following steps: S101: Based on terrain data, a three-dimensional coordinate system is constructed. Multi-source satellite remote sensing images and meteorological data are combined using a spatiotemporal reflectance fusion model based on radiometric calibration normalization, bidirectional reflectance distribution function correction, and local weight adaptation to generate target enhanced images and fusion weight fields corresponding to the spatial location of the images. S102: The target enhancement image is used to generate a first concentration distribution field through a forward physical simulation model, and the first concentration distribution field is jointly fitted with the spatially registered ground monitoring data through a residual correction network to generate a bias-corrected second concentration distribution field. S103: Multiscale decomposition and reconstruction of the second concentration distribution field yields the cross-scale residual tensor; the consistency index is calculated to generate the parameter adjustment map; and a forward physics simulation model that satisfies the update conditions is output. S104: Input the updated forward physics simulation model and real-time wind field data into the three-dimensional diffusion simulation model to generate the predicted concentration distribution and perform recursive least squares assimilation with the new remote sensing observations to obtain the prediction correction vector set. S105: The parameter adjustment map and the prediction correction vector set are combined in a bounded bilinear manner to generate the observation weighted field. The fusion weighted field is updated under preset conditions to form a double closed-loop structure of model update path and data weighting path.

2. The intelligent inversion method for factory exhaust gas emissions based on satellite remote sensing imagery according to claim 1, characterized in that, The application logic of the spatiotemporal reflectivity fusion model is as follows: Radiometric calibration and normalization involves converting the radiometric values ​​of pixels in multi-source satellite remote sensing images to unify the original digital values ​​into surface reflectance, thereby obtaining a radiometrically calibrated image set to eliminate response differences between different sensors. BRDF correction utilizes topographic data and meteorological conditions to calculate the observation geometry information of pixels, including solar azimuth, sensor azimuth, and zenith angle. Based on land cover type parameters, it corrects for angle effects on the radiometrically calibrated image set, generating a spectrally consistent image set to ensure the consistency of multi-source images in spectral space. Local weight adaptive calculation: In the three-dimensional coordinate system, the residual variance of the spectral consistency image set in the spectral space is calculated in units of the pixel neighborhood; according to the inverse variance weighting principle, the fusion weight of each image in the neighborhood is dynamically allocated to obtain the local optimal fusion coefficient set; Spatiotemporal reflectance fusion and weight field generation: The spectral consistency image set and the local optimal fusion coefficient set are input into the spatiotemporal reflectance fusion algorithm. The complementarity of high spatial resolution data and high temporal resolution data is used to generate target enhancement images. During the fusion process, the multi-source data weights of each pixel are recorded to form a fusion weight field that corresponds one-to-one with the pixel position.

3. The intelligent inversion method for factory exhaust gas emissions based on satellite remote sensing imagery according to claim 2, characterized in that, The application logic of the forward physics simulation model: Using the factory emission source location, emission rate estimate, and meteorological data as simulation variables, and combined with the terrain constraints in the three-dimensional coordinate system, numerical simulation calculations are performed on the transport and diffusion process of pollutants in the atmosphere. The calculated pollutant concentrations corresponding to the pixel locations are stored in a regular grid form to form the first concentration distribution field matching the target enhancement image.

4. The intelligent inversion method for factory exhaust gas emissions based on satellite remote sensing imagery according to claim 3, characterized in that, The joint fitting of the first concentration distribution field with spatially registered ground monitoring data using a residual correction network includes: The actual geographic coordinates of the ground monitoring stations are mapped to the grid coordinates of the first concentration distribution field. The discrete monitoring data are then projected onto a spatial grid consistent with the first concentration distribution field using bilinear interpolation or neighbor interpolation methods to obtain a registered monitoring data field. Calculate the grid difference between the first concentration distribution field and the registered monitoring data field, and form a residual field based on the pixel position to characterize the systematic deviation and error characteristics between the physical simulation results and the actual observations; The first concentration distribution field and the residual field are used as input factors of the residual correction network. The perturbation factors that the forward physics simulation model failed to capture are fitted and learned by the convolution feature extraction and mapping mechanism, and the deviation correction field is output. The first concentration distribution field is superimposed on the deviation correction field pixel by pixel to obtain the second concentration distribution field.

5. The intelligent inversion method for factory exhaust gas emissions based on satellite remote sensing imagery according to claim 4, characterized in that, The logic for obtaining the cross-scale residual tensor is as follows: The second concentration distribution field is decomposed into multiple scale sub-bands with different spatial resolutions and frequency components using a multi-scale decomposition algorithm; each scale sub-band corresponds to the characteristic manifestation of pollutant concentration changes at a specific spatial scale. The residual components at each scale are obtained by performing difference calculations between each scale sub-band and the previous time or historical baseline concentration distribution field at the same scale. The residual components at each scale are stacked according to their spatial location to form a cross-scale residual tensor containing scale, location, and amplitude information, which is used to characterize the deviation distribution pattern of the concentration field at different scales.

6. The intelligent inversion method for factory exhaust gas emissions based on satellite remote sensing imagery according to claim 5, characterized in that, The logic for obtaining the parameter adjustment map is as follows: The global consistency index is calculated based on the statistical characteristics of the cross-scale residual tensor and labeled as the consistency index result. The consistency index is used to quantify the overall matching degree between the current model prediction result and historical trends or actual monitoring data at multiple scales, thereby enhancing all aspects of the target image. Based on the cross-scale residual tensor and consistency index results, the parameter adjustment amount corresponding to each spatial location is calculated. The parameter adjustment amount includes the emission source strength correction coefficient, the diffusion coefficient correction factor and / or the wind field correction vector. The parameter adjustments are combined into a parameter adjustment map corresponding to a three-dimensional coordinate space grid. This parameter adjustment map is used to update the forward physics simulation model; specifically: When the consistency index is higher than the preset update trigger threshold, the parameter adjustment map is invoked to locally or globally update the key parameters of the forward physics simulation model; otherwise, the current model parameters are kept unchanged to prevent frequent fluctuations in the prediction results. Output the updated forward physics simulation model parameters, or maintain the existing parameter settings in the frozen state, while archiving the consistency index and parameter adjustment map.

7. The intelligent inversion method for factory exhaust gas emissions based on satellite remote sensing imagery according to claim 6, characterized in that, The application logic of the three-dimensional diffusion simulation model is as follows: A three-dimensional diffusion simulation model based on the finite volume method or a high-precision convection-diffusion equation solver is used to combine the real-time wind field with the updated physical parameters to predict the diffusion concentration distribution of pollutants in the horizontal and vertical directions, thus obtaining the predicted concentration distribution field. Using the predicted concentration distribution field as the prior field and the newly acquired remote sensing observation data as the observation field, a mapping matrix between the state vector and the observation vector is constructed. The recursive least squares algorithm is used to update the state estimate time by time, and the observation confidence weight is introduced to reduce the impact of outliers. The output prediction correction vector set represents the amount of correction to be applied to each grid point. The prediction correction vector set is weighted and superimposed on the prediction concentration distribution field to obtain the fused prediction result, which is marked as the third concentration distribution field, and serves as the final concentration field after assimilation.

8. The intelligent inversion method for factory exhaust gas emissions based on satellite remote sensing imagery according to claim 7, characterized in that, It also includes the following steps: in step 104, if the final concentration field after multiple assimilations does not improve or the improvement is less than a set threshold, the prediction correction vector set is frozen to avoid invalid correction; when freezing, the prediction trend features are extracted and sent back to S103 to adjust the consistency index calculation strategy and update the triggering conditions, and the frozen state and prediction trend features are marked when the third concentration distribution field is output.

9. The intelligent inversion method for factory exhaust gas emissions based on satellite remote sensing imagery according to claim 8, characterized in that, The parameter adjustment map and the predicted correction vector set are calculated using the bounded bilinear combination formula: W(x,y,z)=α·P(x,y,z)+β·C(x,y,z); Where: P(x,y,z) is the parameter adjustment spectrum, C(x,y,z) is the prediction correction vector set, W(x,y,z) is the observation weighted field; α and β are non-negative weight coefficients, and satisfy α+β≤1.

10. A smart inversion system for factory exhaust emissions based on satellite remote sensing imagery, implemented according to any one of claims 1-8, characterized in that, It includes a 3D construction module, a physical simulation correction module, a cross-scale analysis module, a prediction assimilation module, and a weighted closed-loop update module. Data is transmitted between the modules via wired and / or wireless means. The 3D construction module is used to construct a 3D coordinate system based on terrain data. It generates target enhanced images and fusion weight fields corresponding to the spatial location of the images by using a spatiotemporal reflectance fusion model based on radiometric calibration normalization, bidirectional reflectance distribution function correction and local weight adaptation from multi-source satellite remote sensing images and meteorological data. The physical simulation correction module is used to generate a first concentration distribution field from the target enhancement image through a forward physical simulation model, and then use a residual correction network to jointly fit the first concentration distribution field with spatially registered ground monitoring data to generate a bias-corrected second concentration distribution field. The cross-scale analysis module is used to decompose and reconstruct the cross-scale residual tensor of the second concentration distribution field at multiple scales, calculate the consistency index to generate parameter adjustment maps, and output a forward physics simulation model that meets the update conditions. The prediction assimilation module inputs the updated forward physics simulation model and real-time wind field data into the three-dimensional diffusion simulation model to generate a predicted concentration distribution and performs recursive least squares assimilation with the new remote sensing observations to obtain a prediction correction vector set. The weighted closed-loop update module generates an observation weighted field by combining the parameter adjustment map and the prediction correction vector set in a bounded bilinear combination. Under preset conditions, it updates the fused weighted field, forming a double closed-loop structure of the model update path and the data weighting path.

Images