Carbon emission checking method and system based on multi-mode inversion model

Abstract

The invention discloses a carbon emission checking method and system based on a multi-mode inversion model, and relates to the technical field of carbon emission checking. Air-based monitoring, foundation monitoring, mobile monitoring and maneuvering monitoring are adopted to form a three-dimensional monitoring network; the intelligent inversion calculation layer fuses global background concentration and regional real-time monitoring data, and adopts a GSI 4DVar algorithm to optimize a three-dimensional concentration field; in the first inversion stage, random disturbance is applied to the prior emission list, multiple emission sets are generated, each emission set corresponds to a grid area of a three-dimensional concentration field, and concentration distribution is simulated through a WRF-CMAQ model; in the second inversion stage, an EnSRF algorithm is adopted to assimilate measured data, and emission parameters are iteratively optimized; according to the invention, various monitoring means, advanced algorithms and data analysis methods are organically integrated, and automation and intelligentization of the whole process from data acquisition, processing and analysis to abnormity determination are realized. The method provides efficient and scientific technical support for carbon emission supervision, and is helpful for promoting the realization of carbon emission fine management and energy conservation and emission reduction targets.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of carbon emission verification, in particular to a carbon emission verification method and system based on a multi-mode inversion model. BACKGROUND

[0002] With the increasingly severe global climate change problem, carbon emission verification, as an important means of assessing greenhouse gas emissions, plays a key role in addressing climate change and promoting the transformation of the low-carbon economy. Carbon emission verification is not only the basis for the effective operation of market mechanisms such as carbon emission rights trading, but also an important embodiment of enterprises fulfilling their social responsibilities and enhancing their international competitiveness. It also helps to promote technological innovation and industrial upgrading, and to promote sustainable economic and social development.

[0003] Currently, carbon emission verification technology has developed from early simple manual sampling and laboratory analysis to modern multi-means comprehensive monitoring, but still faces many challenges and limitations. With the continuous progress of technology, especially breakthroughs in multi-source data fusion, artificial intelligence, satellite remote sensing and other fields, the innovation of carbon emission verification technology provides new possibilities.

[0004] Despite the significant progress made in carbon emission verification technology, there are still many defects and challenges that affect its effectiveness in practical applications. These defects mainly manifest in the following aspects: Traditional calculation methods highly rely on the product of fuel consumption and carbon emission coefficient reported by power plants, which cannot reflect the actual emission differences of different power plants and regions. Due to the lack of unified accounting standards internationally, the estimated results are difficult to compare and verify. The limitation of this method is that it assumes that all power plants have the same emission factor, but in reality, due to differences in technology level, equipment condition, fuel quality and other factors, the actual emissions of different power plants may differ greatly from the results calculated based on the standard emission factor.

[0005] Taking lignite as an example, the difference between its emission factors can be as high as 27%, and the deviation of the emission calculated by some default emission factors based on coal type from the actual emission exceeds 10%. This deviation not only affects the accuracy of carbon emission verification, but also may lead to unfair competition in the carbon market and inconsistency in policy implementation.

[0006] Although satellite remote sensing technology has significant advantages in monitoring key source carbon emissions, it still faces challenges such as background interference, atmospheric stability error and smoke plume lifting deviation, resulting in a high carbon emission inversion error of up to 50%, which is a key vulnerability in the global carbon satellite remote sensing monitoring system. These problems are caused by the complexity of atmospheric radiation transmission and the limitations of remote sensing data itself. SUMMARY

[0007] To solve the above technical problems, the present application provides a carbon emission verification method and system based on a multi-mode inversion model. The following technical solution is adopted: The carbon emission verification method based on a multi-mode inversion model comprises the following steps: Step 1: a stereoscopic monitoring network is formed by adopting air-based monitoring, ground-based monitoring, mobile monitoring and mobile monitoring; Step 2: the intelligent inversion calculation layer fuses global background concentration and real-time monitoring data, and optimizes the three-dimensional concentration field by using the GSI 4DVar algorithm; Step 3: the intelligent inversion calculation layer divides the inversion calculation into two stages; The first inversion stage: random perturbation is applied to the prior emission inventory to generate multiple emission sets, each emission set corresponds to a grid area of the three-dimensional concentration field, and the concentration distribution is simulated by the WRF-CMAQ model; the second inversion stage: the EnSRF algorithm is used to assimilate the measured data to iteratively optimize the emission parameters; Step 4: the background concentration and the stereoscopic monitoring network are sampled multiple times by Monte Carlo sampling, and the confidence interval is output by the probability density function of the emission result, and the carbon emission value exceeding the set confidence interval is the final carbon emission value, and the final carbon emission value of each grid area is output; Step 5: a preset emission anomaly threshold is set, and if the output final carbon emission value is greater than the emission anomaly threshold, the abnormal emission grid is marked, and the mobile monitoring is called to patrol the abnormal emission grid; Step 6: visual analysis is performed on the full-coverage visual patrol data, the specific location area of the enterprise is marked in the visual patrol image data, the emission features including the emission center, the emission path, the chimney height, the chimney mouth emission smoke concentration gradient and the smoke color feature value are identified by using the visual feature algorithm, the corresponding relationship between the emission features and the enterprise GPS coordinates is established, the corresponding enterprise is listed as a suspected abnormal emission enterprise, and a visual evidence package is generated.

[0008] By adopting the above technical solution, the stereoscopic monitoring network combining air-based monitoring, ground-based monitoring, mobile monitoring and mobile monitoring realizes full-range and multi-level coverage of the carbon emission area. The air-based monitoring uses satellite remote sensing equipment and unmanned aerial vehicle groups to quickly obtain gas concentration data in a large range of areas and key areas, which makes up for the limitations of traditional monitoring methods in spatial coverage; the ground-based monitoring provides long-term and stable real-time monitoring data by reasonable arrangement of fixed stations and enterprises; the mobile monitoring vehicle can dynamically collect data in different areas; the mobile monitoring focuses on the problem area. The multiple monitoring methods complement each other to ensure the integrity and accuracy of the data, avoid monitoring blind spots, and provide a reliable data basis for carbon emission verification.

[0009] The intelligent inversion calculation layer adopts the GSI 4DVar algorithm to optimize the three-dimensional concentration field, fully integrates global background concentration and regional real-time monitoring data, minimizes the difference between model prediction and observation data through the variational principle, and dynamically adjusts the concentration field to make the model more consistent with the actual atmospheric environment. In the inversion calculation stage, the first stage applies random perturbations to the prior emission inventory and uses the WRF-CMAQ model to simulate the concentration distribution, fully considering the uncertainty of the emission data; the second stage uses the EnSRF algorithm to assimilate the measured data and iteratively optimize the emission parameters to achieve the best match between the simulation results and the measured data. The application of these algorithms significantly improves the data processing efficiency and model accuracy, making the calculation and prediction of carbon emissions more scientific and reliable.

[0010] Multiple Monte Carlo sampling is performed on the background concentration and stereoscopic monitoring network, and the probability density function of the emission results is constructed to output the confidence interval. Carbon emissions exceeding the set confidence interval are determined as the final carbon emissions. This method effectively quantifies the uncertainty of the data, considers various possible situations through a large number of random simulations, avoids the bias caused by a single data or model, and enhances the reliability and credibility of the carbon emission value calculation results, providing a solid basis for subsequent abnormal emission determination.

[0011] The pre-set emission anomaly threshold marks the abnormal emission grid, and calls for mobile monitoring for inspection to quickly lock the problem area. Deep visual analysis is performed on the visual inspection data, and advanced image recognition and visual feature algorithms are used to accurately extract key emission features such as emission center and emission path, and associate them with enterprise GPS coordinates to accurately determine the suspected enterprise of abnormal emission. Finally, a visual evidence package is generated to present the enterprise abnormal emission information in an intuitive and clear form, making it easy for regulatory departments to quickly grasp the situation, improve law enforcement efficiency, and achieve precise attack and effective supervision of abnormal emission enterprises.

[0012] Various monitoring methods, advanced algorithms, and data analysis methods are integrated to achieve full-process automation and intelligentization from data collection, processing, analysis to abnormality determination. This reduces human intervention and reduces human error, while improving the efficiency of carbon emission verification, enabling rapid response to environmental changes, timely detection and handling of abnormal emission problems, providing efficient and scientific technical support for carbon emission supervision, and helping to promote fine management of carbon emissions and energy saving and emission reduction targets.

[0013] Optionally, in step 1, the space-based monitoring is synchronously connected to a high-resolution satellite, and the high-resolution satellite carries a greenhouse gas detection spectrometer; The ground-based monitoring is to deploy multiple fixed monitoring stations in the target monitoring area, and the fixed monitoring stations carry gas analyzers and Fourier transform infrared spectrometers; Mobile monitoring is to configure multiple mobile monitoring vehicles equipped with high-precision vehicle-mounted gas monitoring equipment and positioning systems, and to drive on the road according to the preset route or random route to collect real-time gas concentration data during the movement; Mechanical monitoring is to use a portable Fourier transform infrared spectrometer and a high-definition visual camera module carried by a drone to patrol the target area according to the instructions.

[0014] Optionally, in step 2, the pretreated global background concentration data and the real-time monitoring data of the region are input into the GSI 4DVar algorithm for optimization, and the information of time and space dimensions is considered in the optimization process to dynamically adjust the three-dimensional concentration field to generate an optimized three-dimensional concentration field model. The optimized three-dimensional concentration field is output in the form of grid data, and the grid resolution is set to a size according to the monitoring requirements.

[0015] Optionally, the first inversion stage includes the following specific steps: collecting the basic information of enterprises in the target area, establishing a prior emission inventory, and applying random perturbations conforming to a normal distribution to the parameters of each emission source in the prior emission inventory. The perturbation range is set according to the data uncertainty to generate multiple sets of emission collections. Divide the three-dimensional concentration field into multiple grid regions, and each grid region corresponds to an emission collection to establish a mapping relationship between the emission collection and the grid region. Input the multiple sets of emission collections into the WRF-CMAQ model to simulate the concentration distribution of each grid region corresponding to the emission collection at different times, and output the result data of the simulated concentration distribution.

[0016] Optionally, the second inversion stage includes the following specific steps: filtering and preprocessing the measured data obtained by the stereoscopic monitoring network in step 1 to extract effective data corresponding to the simulation time and space of the WRF-CMAQ model; and using the EnSRF algorithm to assimilate the effective data and the result data of the simulated concentration distribution. The EnSRF algorithm updates the estimated value of the emission parameter by calculating the covariance between the collection members. In each iteration process, the emission parameter in the emission collection is adjusted according to the difference between the measured data and the simulated data. The iteration termination condition is set, and the iteration optimization is continuously performed before the termination condition is met. The emission parameter is continuously adjusted, and the optimal emission parameter is obtained after iteration termination.

[0017] Optionally, in step 5, a reasonable emission anomaly threshold is set by referring to the industry emission standard, the regional environmental capacity, the historical carbon emission data, the functional positioning of the region, and the environmental carrying capacity. Compare the final carbon emission value of each grid region output in step 4 with the set emission anomaly threshold. If the final carbon emission value is greater than the emission anomaly threshold, mark the grid region as an abnormal emission grid, and record the grid position and the exceeding value.

[0018] Optionally, in step 6, a deep learning-based target detection algorithm is used to identify the building and chimney landmark features of the enterprise in the visual image, mark the specific location area where the enterprise is located, and at the same time, use a visual feature algorithm to extract the emission center, emission path, chimney height, chimney mouth emission smoke concentration gradient, and smoke color feature value as the emission characteristics; The identified emission characteristics are associated with the GPS coordinates of the enterprise to establish a corresponding relationship, and according to the association result, the corresponding enterprise is listed as a suspected enterprise of abnormal emission, and the relevant emission characteristic data and image evidence are recorded.

[0019] Optionally, in step 6, the basic information, emission characteristic data, visual inspection image, and GPS coordinate information of the suspected enterprise of abnormal emission are integrated to generate a visual evidence package.

[0020] The carbon emission verification system based on the multi-mode inversion model is used to realize the carbon emission verification method based on the multi-mode inversion model, and the carbon emission verification system comprises an air-based monitoring unit, a ground-based monitoring unit, a mobile monitoring unit, a mobile monitoring unit and a carbon emission control center computer, and the carbon emission control center computer is connected with the air-based monitoring unit, the ground-based monitoring unit, the mobile monitoring unit and the mobile monitoring unit through a wireless communication network.

[0021] Optionally, the air-based monitoring unit comprises a plurality of high-resolution satellites carrying greenhouse gas detection spectrometers, the ground-based monitoring unit comprises a plurality of fixed monitoring stations carrying gas analyzers and Fourier transform infrared spectrometers, the mobile monitoring unit comprises a plurality of mobile monitoring vehicles equipped with vehicle-mounted high-precision gas monitoring equipment and positioning systems, and the mobile monitoring unit comprises a plurality of unmanned aerial vehicles carrying portable Fourier transform infrared spectrometers and high-definition visual camera modules.

[0022] In summary, the present application has at least one of the following beneficial technical effects: The present application can provide a carbon emission verification method and system based on a multi-mode inversion model, which realizes all-around and multi-level coverage of the carbon emission area through a three-dimensional monitoring network combining air-based, ground-based, mobile and mobile monitoring. The intelligent inversion calculation layer optimizes the three-dimensional concentration field by using GSI 4DVar algorithm, fully integrates global background concentration and real-time monitoring data, minimizes the difference between model prediction and observation data through variational principle, and dynamically adjusts the concentration field, so that the model is more suitable for the actual atmospheric environment.

[0023] The preset emission anomaly threshold marks an abnormal emission grid, and calls motorized monitoring for inspection, quickly locks the problem area. The visual inspection data is subjected to deep visual analysis, the advanced image recognition and visual feature algorithm is used to accurately extract the key emission features such as emission center and emission path, and is associated with the enterprise GPS coordinates, so that the abnormal emission suspected enterprise is accurately determined. Finally, a visual evidence package is generated to present the abnormal emission information of the enterprise in an intuitive and clear form, so that the supervision department can quickly master the situation, improve the law enforcement efficiency, and realize the accurate attack and effective supervision on the abnormal emission enterprise.

[0024] Various monitoring means, advanced algorithms and data analysis methods are organically integrated to realize the full-process automation and intelligentization from data acquisition, processing, analysis to abnormality determination. The artificial intervention is reduced, the human error is reduced, the efficiency of carbon emission verification is improved, the environmental changes can be quickly responded, the abnormal emission problems can be quickly found and handled, high-efficiency and scientific technical support is provided for carbon emission supervision, and the realization of carbon emission fine management and energy saving and emission reduction goals is promoted. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 It is a flowchart of a carbon emission verification method based on a multi-mode inversion model. DETAILED DESCRIPTION

[0026] The application will be further described in detail below with reference to the drawings.

[0027] The embodiment of the application discloses a carbon emission verification method and system based on a multi-mode inversion model.

[0028] Reference Figure 1 , embodiment 1, the carbon emission verification method based on the multi-mode inversion model comprises the following steps: Step 1, a stereoscopic monitoring network is formed by using air-based monitoring, ground-based monitoring, mobile monitoring and motorized monitoring; Step 2, the intelligent inversion calculation layer fuses global background concentration and regional real-time monitoring data, and optimizes the three-dimensional concentration field by using the GSI 4DVar algorithm; Step 3, the intelligent inversion calculation layer divides the inversion calculation into two stages; The first inversion stage: random disturbance is applied to the prior emission inventory to generate a plurality of emission sets, each emission set corresponds to a grid area of the three-dimensional concentration field, and the concentration distribution is simulated through the WRF-CMAQ model; the second inversion stage: the EnSRF algorithm is used to assimilate the measured data, and the emission parameters are iteratively optimized; Step 4, the background concentration and the stereoscopic monitoring network are subjected to multiple Monte Carlo sampling, the confidence interval is output through the probability density function of the emission result, the carbon emission value exceeding the set confidence interval is the final carbon emission value, and the final carbon emission value of each grid area is output; Step 5, presetting the emission anomaly threshold, if the final carbon emission value output is greater than the emission anomaly threshold, marking the anomaly emission grid, calling for mobile monitoring to patrol the anomaly emission grid; Step 6, visual analysis of full-coverage visual patrol data, marking the specific location area where the enterprise is located in the visual patrol image data, using visual feature algorithm to identify emission features, including emission center, emission path, chimney height, chimney mouth emission smoke concentration gradient, smoke color feature value, establishing the corresponding relationship between emission features and enterprise GPS coordinates, and listing the corresponding enterprises as suspected emission enterprises, and generating a visual evidence package.

[0029] By adopting the above technical scheme, through the combination of space-based, ground-based, mobile and mobile monitoring of the three-dimensional monitoring network, full-coverage, multi-level coverage of carbon emission areas is achieved. Space-based monitoring uses satellite remote sensing equipment and unmanned aerial vehicle groups to quickly obtain gas concentration data in a large range of areas and key areas, making up for the limitations of traditional monitoring methods in spatial coverage; ground-based monitoring provides long-term, stable real-time monitoring data through reasonable layout of fixed sites and enterprise surroundings; mobile monitoring vehicles can dynamically collect data in different areas through flexible deployment; mobile monitoring focuses on the problem areas of abnormal emission grids. The various monitoring methods complement each other to ensure the integrity and accuracy of the data, avoid monitoring blind spots, and provide a reliable data basis for carbon emission verification.

[0030] The intelligent inversion calculation layer uses GSI 4DVar algorithm to optimize the three-dimensional concentration field, fully integrates global background concentration and regional real-time monitoring data, minimizes the difference between model prediction and observation data through variational principle, and dynamically adjusts the concentration field to make the model more consistent with the actual atmospheric environment. In the inversion calculation stage, in the first stage, random perturbation is applied to the prior emission inventory and the concentration distribution is simulated using the WRF-CMAQ model, fully considering the uncertainty of emission data; in the second stage, EnSRF algorithm is used to assimilate measured data, iteratively optimize emission parameters, and realize the best matching of simulation results and measured data. The application of these algorithms significantly improves the data processing efficiency and model accuracy, making the calculation and prediction of carbon emissions more scientific and reliable. The background concentration and three-dimensional monitoring network are sampled multiple times by Monte Carlo, the probability density function of the emission results is output by constructing the confidence interval, and the carbon emission value exceeding the set confidence interval is determined as the final carbon emission value. This method effectively quantifies the uncertainty of the data, considers various possible situations through a large number of random simulations, avoids the bias caused by single data or model, enhances the reliability and credibility of the calculation results of carbon emission value, and provides a solid basis for subsequent abnormal emission judgment.

[0031] The preset emission anomaly threshold marks an abnormal emission grid, and calls for mobile monitoring for inspection, to quickly lock the problem area. Deep visual analysis is performed on the visual inspection data, advanced image recognition and visual feature algorithms are used to accurately extract key emission features such as emission center and emission path, and are associated with enterprise GPS coordinates to accurately determine the suspected enterprise of abnormal emission. Finally, a visual evidence package is generated to present the abnormal emission information of the enterprise in an intuitive and clear form, so that the regulatory department can quickly grasp the situation, improve the efficiency of law enforcement, and achieve precise attack and effective supervision of the abnormal emission enterprise.

[0032] Various monitoring methods, advanced algorithms and data analysis methods are organically integrated to realize the full-process automation and intelligentization from data collection, processing, analysis to abnormality determination. The artificial intervention is reduced, the human error is reduced, the efficiency of carbon emission verification is improved, the environmental changes can be quickly responded, the abnormal emission problems can be found and handled in time, high-efficiency and scientific technical support is provided for carbon emission supervision, and the realization of fine management of carbon emission and the energy saving and emission reduction goal is promoted.

[0033] In step 1 of embodiment 2, the space-based monitoring is synchronous access to a high-resolution satellite, and the high-resolution satellite carries a greenhouse gas detection spectrometer; The ground-based monitoring is to deploy multiple fixed monitoring stations in the target monitoring area, and the fixed monitoring stations carry gas analyzers and Fourier transform infrared spectrometers; The mobile monitoring is to configure multiple mobile monitoring vehicles, equipped with vehicle-mounted high-precision gas monitoring equipment and positioning systems, and to drive on the road according to the preset route or random route to collect gas concentration data in real time during the movement; The mobile monitoring is to configure multiple mobile monitoring vehicles, equipped with vehicle-mounted high-precision gas monitoring equipment and positioning systems, and to drive on the road according to the preset route or random route to collect gas concentration data in real time during the movement;

[0034] By using the above technical scheme, the high-resolution satellite carrying the greenhouse gas detection spectrometer is synchronously accessed, and the greenhouse gas concentration data with large range and high spatio-temporal resolution can be obtained. The satellite monitoring is not limited by geographical environment and human factors, and can effectively monitor the areas such as oceans and plateaus which are difficult to cover by traditional monitoring, to realize dynamic monitoring of carbon emission on a global or large regional scale. At the same time, the high-resolution characteristic makes the monitoring data accurately reflect the subtle differences of carbon emission in different regions, to provide high-precision data support for regional carbon emission total amount accounting and trend analysis, and to help the macro-level carbon emission control decision.

[0035] Deploying fixed monitoring stations equipped with multi-component gas analyzers and Fourier transform infrared spectrometers in the target monitoring area can provide long-term, continuous, and stable monitoring of the atmospheric environment in the region. The multi-component gas analyzer can simultaneously measure multiple greenhouse gases in real time, while the Fourier transform infrared spectrometer can accurately identify and analyze gas composition and concentration changes due to its high sensitivity and high resolution. The stable setting of the fixed station provides a long-term and reliable data sequence for the study of regional carbon emission characteristics, which helps to analyze the seasonal and periodic variation of carbon emissions and provides a scientific basis for the formulation of regional carbon emission management policies.

[0036] Configuring multiple mobile monitoring vehicles equipped with high-precision vehicle-mounted gas monitoring equipment and positioning systems can drive along pre-set or random routes on the road according to actual needs, and collect real-time gas concentration data during the movement. This flexible monitoring method can quickly respond to sudden situations and monitoring needs in key areas, fill in the gaps in spatial coverage of fixed monitoring stations, and obtain carbon emission data in dynamic environments such as urban blocks and traffic roads. Through the complementarity of mobile monitoring and fixed monitoring data, a more comprehensive and more realistic regional carbon emission data model can be constructed, improving the accuracy and comprehensiveness of carbon emission verification.

[0037] Using unmanned aerial vehicles equipped with portable Fourier transform infrared spectrometers and high-definition visual camera modules for mobile monitoring can quickly respond to abnormal emission grid instructions and accurately patrol target areas. Portable Fourier transform infrared spectrometers can quickly obtain gas composition and concentration information in target areas, helping to determine the type and intensity of abnormal emissions; high-definition visual camera modules can present visual information such as emission sources and emission paths in target areas through high-definition images and videos, providing visual evidence for abnormal emission analysis. The mobility and flexibility of unmanned aerial vehicles enable them to penetrate complex terrain and dangerous areas for monitoring, greatly improving the efficiency and accuracy of monitoring abnormal emission areas and assisting regulatory authorities in quickly locating abnormal emission enterprises to provide strong support for law enforcement actions.

[0038] The combination of air-based, ground-based, mobile, and mobile monitoring forms a three-dimensional monitoring network from macro to micro, from static to dynamic, and from wide area to local.

[0039] In Example 3, Step 2, the pre-processed global background concentration data and real-time monitoring data of the region are input into the GSI 4DVar algorithm for optimization. During the optimization process, the information of time and space dimensions is considered, and the three-dimensional concentration field is dynamically adjusted to generate an optimized three-dimensional concentration field model. The optimized three-dimensional concentration field is output in the form of grid data, and the grid resolution is set according to the monitoring requirements.

[0040] By adopting the above technical solution, the global background concentration data is screened and interpolated to match the spatial and temporal resolution of the regional real-time monitoring data. The regional real-time monitoring data is subjected to quality control to eliminate abnormal values and invalid data, and is smoothed by Kalman filtering and other methods to improve data accuracy. The preprocessed global background concentration data and regional real-time monitoring data are input into the GSI 4DVar algorithm. Based on the variational principle, the algorithm minimizes the difference between the model prediction and the observation data by constructing a cost function. During the optimization process, the information in the time and space dimensions is considered to dynamically adjust the three-dimensional concentration field, generating a three-dimensional concentration field model that better matches the actual situation. The optimized three-dimensional concentration field is output in the form of grid data, with a grid resolution of 100m x 100m x 10m, providing basic data for subsequent inversion calculations.

[0041] In the first stage of the inversion, the following specific steps are taken: collecting basic information of enterprises in the target area, establishing a prior emission inventory, applying random perturbations conforming to a normal distribution to the parameters of each emission source in the prior emission inventory, and setting the perturbation range according to data uncertainty to generate multiple sets of emissions. The three-dimensional concentration field is divided into multiple grid regions, each corresponding to a set of emissions, and a mapping relationship between the set of emissions and the grid region is established. Each set of emissions is input into the WRF-CMAQ model to simulate the concentration distribution of the grid region corresponding to each set of emissions at different times, and the result data of the simulated concentration distribution is output.

[0042] By adopting the above technical solution, the production process, energy consumption, product output, and other information of enterprises in the region are collected to establish a prior emission inventory. Random perturbations conforming to a normal distribution are applied to the parameters of each emission source (such as emission intensity and emission time) in the prior emission inventory, and the perturbation range is set according to data uncertainty to generate multiple sets of emissions.

[0043] The three-dimensional concentration field is divided into multiple grid regions according to certain rules, each grid region corresponding to a set of emissions. A mapping relationship between the set of emissions and the grid region is established to ensure that the set of emissions for each grid region accurately reflects the emissions in that region.

[0044] Each set of emissions is input into the WRF-CMAQ model. The WRF model is used to simulate the dynamic and thermal processes of the atmosphere, and the CMAQ model is used to simulate the transport, diffusion, and transformation of atmospheric chemical substances. Through model calculation, the concentration distribution of the grid region corresponding to each set of emissions at different times is simulated, and the simulation result data is output.

[0045] The second stage of the inversion embodiment 5 includes the following specific steps: filtering and preprocessing the measured data obtained by the stereoscopic monitoring network in step 1, and extracting effective data corresponding to the simulation time and space of the WRF-CMAQ model; using the EnSRF algorithm to assimilate the effective data and the result data of the simulated concentration distribution, the EnSRF algorithm updates the estimated value of the emission parameter by calculating the covariance between the ensemble members, adjusts the emission parameter in the emission ensemble according to the difference between the measured data and the simulated data in each iteration process, sets the iteration termination condition, and continuously iterates and optimizes before the termination condition is met, and continuously adjusts the emission parameter until the optimal emission parameter is obtained.

[0046] By using the above technical solution, the measured data obtained by the stereoscopic monitoring network in step 1 is filtered and preprocessed, and effective data corresponding to the simulation time and space of the WRF-CMAQ model is extracted. At the same time, the concentration distribution data obtained by the first stage simulation is prepared.

[0047] The EnSRF (ensemble square root filter) algorithm is used to assimilate the measured data and the simulated data. The EnSRF algorithm updates the estimated value of the emission parameter by calculating the covariance between the ensemble members. In each iteration process, the emission parameter in the emission ensemble is adjusted according to the difference between the measured data and the simulated data, so that the simulation result is closer to the actual observation value.

[0048] The iteration termination condition is set, such as the maximum number of iterations or the error between the simulated data and the measured data is less than the set threshold. Before the termination condition is met, the iteration optimization is continuously carried out, and the emission parameter is continuously adjusted until the optimal emission parameter set is obtained, so that the simulated concentration distribution and the measured data are best matched.

[0049] In step 5 of embodiment 6, a reasonable emission anomaly threshold is set by referring to the industry emission standard, the regional environmental capacity, the historical carbon emission data, the functional positioning of the region and the environmental carrying capacity. The final carbon emission value of each grid region output by step 4 is compared with the set emission anomaly threshold. If the final carbon emission value is greater than the emission anomaly threshold, the grid region is marked as an abnormal emission grid, and the grid position and the exceeding value are recorded.

[0050] By adopting the technical scheme, the emission abnormal threshold is set by referring to the industry emission standard, the regional environmental capacity, historical carbon emission data, the functional positioning of the region and the environmental carrying capacity, so that the threshold has a solid scientific basis and realistic pertinence. The industry emission standard provides a basic specification for threshold setting, ensuring that the enterprise emission requirements conform to the unified standard; the regional environmental capacity and carrying capacity reflect the accommodation limit of the local natural environment to carbon emissions, avoiding damage to the ecological environment due to a too high threshold; the historical carbon emission data can analyze the regional carbon emission trend, and dynamically adjust the threshold in combination with the current situation; the functional positioning of the region (such as industrial area, residential area, etc.) determines the difference in carbon emission demand of different regions, making the threshold setting more in line with the actual scene. This multi-dimensional reference method effectively avoids the one-sidedness and randomness of threshold setting, improves the rationality and applicability of the threshold, and provides a reliable standard for accurately determining abnormal emissions.

[0051] In step 6 of embodiment 7, a deep learning-based target detection algorithm is used to identify the building and chimney landmark features of the enterprise in the visual image, marking the specific location area where the enterprise is located. At the same time, a visual feature algorithm is used to extract the emission center, emission path, chimney height, chimney mouth emission smoke concentration gradient, and smoke color feature value as emission features; The identified emission features are associated with the GPS coordinates of the enterprise to establish a corresponding relationship. According to the association result, the corresponding enterprise is listed as a suspected enterprise of abnormal emissions, and relevant emission feature data and image evidence are recorded.

[0052] By adopting the above technical scheme, the high-definition visual image data obtained by mobile monitoring is subjected to noise reduction processing. The median filter algorithm is used to remove salt and pepper noise in the image, and the bilateral filter algorithm is used to reduce Gaussian noise while preserving edge information. The image is uniformly adjusted to a size suitable for input of a deep learning model, such as 512x512 pixels, and is subjected to normalization processing to map the pixel value to the [0,1] interval, improving the model training efficiency and detection accuracy.

[0053] Advanced target detection algorithms such as YOLOv8 or EfficientDet are selected, a large number of visual images containing enterprise buildings and chimneys are collected as training samples, and the position information of the buildings and chimneys is labeled. In the training process, transfer learning technology is used to initialize the model weights pre-trained on a large-scale image dataset (such as COCO), reducing training time and data requirements. By adjusting learning rate, batch size and other hyperparameters, multiple rounds of iterative training are performed on the self-built dataset, using mean average precision (mAP) as the evaluation indicator to continuously optimize the model performance, ensuring that the model achieves high accuracy and recall rate in enterprise building and chimney detection.

[0054] The pre-processed visual images are input into the trained target detection model, which automatically identifies the enterprise buildings and chimneys in the images and generates detection results containing target positions (bounding box coordinates) and class labels. Based on the detected building and chimney positions, the specific location area of the enterprise is calculated through an algorithm, and different colored polygon frames are marked on the image to record the geometric coordinate information of the area.

[0055] Emission center determination: For the identified chimney area image, an image segmentation algorithm (such as the U-Net model based on deep learning) is used to separate the chimney from the background. The centroid coordinates of the chimney area are calculated, which are taken as the emission center. To improve accuracy, multiple consecutive frames of images are processed, and the average value of the centroid coordinates is taken as the final emission center position.

[0056] Emission path tracking: Based on time series visual images, the optical flow method is used to track the motion trajectory of the smoke. The optical flow field is calculated for adjacent frames of images to obtain the motion vector of the smoke pixel points, and the clustering algorithm (such as DBSCAN) is used to cluster the pixel points with similar motion vectors to form the emission path. At the same time, combined with the position information of the chimney and the wind direction data (which can be obtained from meteorological monitoring equipment), the emission path is corrected to make it more consistent with the actual situation.

[0057] Chimney height measurement: The perspective relationship in the visual image and the known camera parameters (such as focal length, shooting height) are used to calculate the chimney height through the principle of triangulation. Feature points at the bottom and top of the chimney are selected in the image, and the actual three-dimensional space coordinates are calculated based on the pixel coordinates of the feature points in the image and the camera parameters, and then the chimney height is obtained. To reduce errors, multiple images can be taken at different angles for measurement, and the average value is taken as the final result.

[0058] Smoke concentration gradient and color feature extraction: The chimney mouth area image is converted to HSV color space, and the V (brightness) channel value of each pixel point is analyzed to calculate the brightness change within a unit distance to obtain the smoke concentration gradient. At the same time, the mean and variance of the H (hue), S (saturation), and V (brightness) channels in the HSV color space of the chimney mouth area image are calculated as the smoke color feature values.

[0059] Coordinate matching and association: The coordinate information of the extracted emission features (emission center, emission path, etc.) is matched with the enterprise GPS coordinates. Using geographic information system (GIS) technology, the coordinates in the visual image are converted to geographic coordinates, the distance between the emission feature coordinates and the enterprise GPS coordinates is calculated, and a distance threshold (such as 50 meters) is set. When the distance is less than the threshold, the corresponding relationship between the emission feature and the enterprise is established.

[0060] Suspect enterprise determination and record: according to the correlation result, the enterprise with the corresponding emission characteristics is listed as a suspected enterprise of abnormal emission. The name, GPS coordinates, emission characteristic data (such as emission center coordinates, chimney height, smoke concentration gradient, etc.) and original visual image evidence containing emission conditions of the suspected enterprise of abnormal emission are recorded to form a detailed enterprise abnormal emission record file, which provides a comprehensive basis for subsequent investigation and processing.

[0061] In step 6 of embodiment 8, the basic information, emission characteristic data, visual inspection image and GPS coordinate information of the suspected enterprise of abnormal emission are integrated to generate a visual evidence package.

[0062] Embodiment 9, a carbon emission verification system based on a multi-mode inversion model, is used to implement a carbon emission verification method based on a multi-mode inversion model. The carbon emission verification system includes an air-based monitoring unit, a ground-based monitoring unit, a mobile monitoring unit, a motorized monitoring unit and a carbon emission control center computer. The carbon emission control center computer is connected to the air-based monitoring unit, the ground-based monitoring unit, the mobile monitoring unit and the motorized monitoring unit through a wireless communication network.

[0063] Embodiment 10, the air-based monitoring unit includes a plurality of high-resolution satellites carrying greenhouse gas detection spectrometers, the ground-based monitoring unit includes a plurality of fixed monitoring stations equipped with gas analyzers and Fourier transform infrared spectrometers; the mobile monitoring unit includes a plurality of mobile monitoring vehicles equipped with vehicle-mounted high-precision gas monitoring equipment and positioning systems; and the motorized monitoring unit includes a plurality of unmanned aerial vehicles equipped with portable Fourier transform infrared spectrometers and high-definition visual camera modules.

[0064] The following describes the implementation principle of the present application by means of specific embodiments: Specific embodiment 1, a large industrial concentrated area with an area of about 500 square kilometers in a certain province is selected as a target monitoring area. The area contains more than 200 enterprises of high carbon emission industries such as chemical industry, steel industry and power industry. In the past, there were problems such as insufficient data coverage and inaccurate abnormal emission determination in carbon emission monitoring. Now, the carbon emission verification method based on the multi-mode inversion model is used to verify the carbon emission of the area.

[0065] Air-based monitoring: 3 high-resolution satellites are simultaneously accessed. The greenhouse gas detection spectrometer carried by the satellite completes a comprehensive scan of the target area every 3 days to obtain the overall concentration distribution data of carbon dioxide, methane and other gases in the area. At the same time, 5 unmanned aerial vehicle groups are arranged every week to fly at a height of 100-500 meters above the area according to a grid path to monitor the low-altitude area and obtain near-surface high-precision gas concentration data.

[0066] Ground monitoring: 20 fixed monitoring stations are deployed in the target monitoring area, of which 5 are distributed in the core area of the industrial zone, 10 are distributed around the industrial zone, and 5 are distributed at the junction of the residential area and the industrial zone. Each fixed monitoring station is equipped with multi-component gas analyzer and Fourier transform infrared spectrometer to collect real-time concentration data of various greenhouse gases in the atmosphere at intervals of 1 minute.

[0067] Mobile monitoring: 8 mobile monitoring vehicles are configured to drive along the main roads and enterprise surrounding roads in the region according to the preset grid route, complete 2 cycles of monitoring per day, and real-time collect gas concentration data and position information during movement by vehicle-mounted high-precision gas monitoring equipment and positioning system.

[0068] Mobile monitoring: 8 mobile monitoring vehicles are configured to drive along the main roads and enterprise surrounding roads in the region according to the preset grid route, complete 2 cycles of monitoring per day, and real-time collect gas concentration data and position information during movement by vehicle-mounted high-precision gas monitoring equipment and positioning system.

[0069] The carbon emission control center computer receives data obtained by air-based, ground-based, and mobile monitoring, and also obtains global background concentration data. The global background concentration data is processed for spatial and temporal resolution matching, and a bilinear interpolation method is used to adjust its resolution to be consistent with the real-time monitoring data of the region. The real-time monitoring data of the region is processed for noise reduction and smoothing by using Kalman filter algorithm, and abnormal data points are removed.

[0070] The preprocessed data is input into the GSI 4DVar algorithm, which is based on the variational principle and constructs a cost function containing time and spatial dimensions. By continuously adjusting the three-dimensional concentration field parameters, the difference between the model prediction and the observation data is minimized. After 30 iterations of calculation, an optimized three-dimensional concentration field model is generated, and the grid data is output with a grid resolution of 100 meters x 100 meters x 10 meters.

[0071] Inversion first stage: Collect the basic information of more than 200 enterprises in the target area, such as production process, energy consumption, product yield, etc., and establish the prior emission inventory. For each emission source parameter in the prior emission inventory, such as emission intensity and emission time, a normal distribution random disturbance with mean value of 0 and standard deviation of 10% is applied to generate 100 emission sets.

[0072] The three-dimensional concentration field is divided into 5000 grid areas, and each grid area is assigned an emission set to establish a clear mapping relationship. 100 groups of emission sets are input into the WRF-CMAQ model, the WRF model simulates the dynamic and thermal processes of the atmosphere, and the CMAQ model simulates the transport, diffusion and transformation of atmospheric chemical substances. After 48 hours of simulation calculation, the concentration distribution data of each grid area corresponding to the emission set at different times is output.

[0073] The second stage of inversion: The measured data obtained by the stereoscopic monitoring network is screened and preprocessed, and the effective data corresponding to the simulation time and space of the WRF-CMAQ model are extracted. The EnSRF algorithm is used to assimilate the effective data with the simulation concentration distribution results.

[0074] In the assimilation process, the EnSRF algorithm updates the estimated value of the emission parameter by calculating the covariance between the 100 groups of emission set members. The iteration termination condition is set to 50 times of maximum iteration or the root mean square error of the simulation data and the measured data is less than 5%. After 35 iterations, the optimal emission parameter set is obtained, and the simulation concentration distribution and the measured data are best matched.

[0075] 1000 times of Monte Carlo sampling are performed on the background concentration and stereoscopic monitoring network data to simulate carbon emission processes under different conditions. According to the emission results obtained by each sampling, the probability density function of the emission results is constructed. Set the confidence level to 95%, calculate the corresponding confidence interval, and determine the carbon emission value exceeding the upper limit of the confidence interval as the final carbon emission value. The final carbon emission value of each grid area is output.

[0076] Referring to the national and local industry emission standards, the environmental capacity evaluation report of the region, the historical carbon emission data of the past 3 years, combined with the functional positioning and environmental carrying capacity of the region as an industrial concentration area, set the emission anomaly threshold. For example, for the chemical industry grid area, the carbon dioxide emission anomaly threshold is set to 500 tons / square kilometer / month.

[0077] Compare the final carbon emission value of each grid area with the set emission anomaly threshold, find that the final carbon emission value of 30 grid areas is greater than the emission anomaly threshold, mark the 30 grid areas as abnormal emission grid, and record the position coordinates and over-standard value of each grid area.

[0078] Call the mobile monitoring unit to conduct full coverage visual inspection on the 30 abnormal emission grids, and transmit the high-definition visual image data obtained by the unmanned aerial vehicle to the carbon emission control center computer.

[0079] The visual image is processed by using a deep learning-based YOLOv8 target detection algorithm. After training on a self-built dataset containing 100,000 enterprise building and chimney images, the detection accuracy reaches 92%. The algorithm automatically identifies the landmark features of enterprise buildings and chimneys in the image and marks the specific location area of the enterprise.

[0080] Emission center determination: The U-Net image segmentation algorithm is used to segment the chimney area image, calculate the centroid coordinates of the chimney area, and take the average of the centroid coordinates of the last 5 frames of images to determine the emission center.

[0081] Emission path tracking: The optical flow method is used to calculate the motion vector of the smoke pixel points in the adjacent frame images, and the DBSCAN clustering algorithm is used to cluster the pixel points with similar motion vectors. Combined with the chimney position information and the wind direction data provided by the meteorological department, the emission path is corrected.

[0082] Chimney height measurement: According to the known parameters such as the UAV shooting height of 100 meters and the camera focal length of 20mm, the feature points at the bottom and top of the chimney are selected, and the chimney height is calculated by the triangulation principle. Take the average value after measuring from 3 different angles.

[0083] Smoke concentration gradient and color feature extraction: Convert the chimney mouth area image to HSV color space, analyze the V channel value to calculate the smoke concentration gradient, and calculate the mean and variance of H, S, and V channels as the smoke color feature value.

[0084] Use GIS technology to match the extracted emission feature coordinate information with the enterprise GPS coordinates, set the distance threshold to 50 meters, establish the correspondence between the emission features and the enterprises, determine 65 enterprises as suspected enterprises of abnormal emissions, and record the basic information, emission feature data and original visual image evidence of these enterprises. Integrate to generate a visual evidence package.

[0085] Through this carbon emission verification, the suspected enterprises of abnormal emissions in the industrial concentration area are accurately located, providing detailed and reliable basis for subsequent environmental supervision and law enforcement. Compared with the previous monitoring method, the coverage of data collection has increased by 80%, and the accuracy of abnormal emission enterprise determination has increased from 65% to 88%, effectively promoting the fine management of carbon emissions and energy saving and emission reduction in the region.

[0086] The above examples fully present the application of the technical scheme. If you want to adjust the case scenario, data details, etc., or supplement other aspects of content, you can tell me at any time.

[0087] Specific embodiment 2, step 1: build a three-dimensional carbon emission monitoring network in a certain province; Integrate air-based, ground-based, mobile, and mobile monitoring methods to achieve full coverage: Air-based monitoring: Access high-resolution satellites equipped with greenhouse gas detection spectrometers (such as Sentinel-5P) to obtain CO2 and CH4 column concentration data for a certain province at a 10km x 10km grid. Simultaneously, use a fleet of drones to conduct low-altitude patrols in key areas such as A port and B industrial zone twice a week, collecting 1km x 1km high-resolution concentration data.

[0088] Ground-based monitoring: Set up 42 fixed monitoring stations (such as A station and B paper industry station) in 11 prefecture-level cities in a certain province, equipped with Fourier transform infrared spectrometers to monitor CO2 (minute-level data) and CH4 (hourly data) concentrations in real time, and simultaneously record weather parameters such as temperature and wind speed.

[0089] Mobile monitoring: Deploy 10 mobile monitoring vehicles to patrol along major highways such as G60 and Hangyong, collecting CO2 concentration gradient data within a 500m range along the road, supplementing the spatial blind spots of fixed stations.

[0090] Mobile monitoring: For the marked abnormal emission grid after inversion (such as A district local area), dispatch drones equipped with portable infrared spectrometers and high-definition cameras for special inspections, covering a 2km x 2km area per inspection, and obtaining detailed images of emission sources.

[0091] Step 2: Optimize the three-dimensional concentration field based on GSI4DVar; Data fusion: Match global background concentration data (monthly average CO2 concentration of 420ppm and CH4 concentration of 1.9ppm at Mauna Loa station in a certain country) with real-time data from ground-based monitoring in a certain province (such as hourly average of 447ppm at a certain fragrance station), eliminate outliers, and smooth the data through Kalman filtering.

[0092] 4DVar optimization: Use GSI4DVar algorithm with a 6-hour time window, combine WRF simulated meteorological field (wind, temperature, and humidity three-dimensional distribution), and dynamically adjust the CO2 and CH4 concentration field of a certain province at a 3km x 3km x 30m grid. The objective function is set to minimize the deviation between the simulated value and the observed value, and the final output is the optimized three-dimensional concentration field model.

[0093] Step 3: Two-stage inversion calculation; First inversion stage: Emission set simulation; Prior list construction: Integrate EDGAR (2021) global list (0.1° resolution) and YRD (2017) Yangtze River Delta list (4km resolution) to establish a prior emission list for enterprises in a certain province, including CO2 emissions of 12,000 industrial enterprises (such as B city steel plant annual emission of 1.2 million tons of CO2).

[0094] Emission perturbation: Apply a normal distribution random perturbation of ±15% to the emission intensity of each enterprise in the prior inventory, generating 50 sets of emission ensembles.

[0095] Concentration simulation: Divide the province into 3km x 3km grids (totaling 1200 grids), each grid corresponding to one set of emission ensembles, input WRF-CMAQ model (turn off gas phase chemical module, keep transport and diffusion process), simulate monthly CO2 concentration distribution in 2023 (e.g. the monthly average concentration of the northeast grid in City A is 450 ppm).

[0096] Inversion phase 2: EnSRF assimilation optimization; Data assimilation: Extract hourly average concentration data from ground-based monitoring stations (e.g. 60m station in City A 449.5 ppm), and perform spatio-temporal matching with WRF-CMAQ simulation results, calculate ensemble covariance using EnSRF algorithm, and iteratively optimize emission parameters (iterate 30 times, convergence threshold set to simulation error <5%).

[0097] Parameter update: For example, the prior emission of a certain grid in City B is 8000gCm -2 yr -1 , after assimilation, the posterior emission is adjusted to 9200gCm -2 yr -1 , and the optimization result is used as the prior emission of the next time window.

[0098] Step 4: Uncertainty quantification and final emission value determination; Monte Carlo sampling: Randomly sample background concentration and monitoring data 1000 times to generate probability density function of emission results, calculate 95% confidence interval (e.g. City B CO2 emission confidence interval is 43.60±1.2TgC).

[0099] Final value determination: Select grid emission values above the upper limit of the confidence interval as the final result, for example, the emission value of a certain grid in Wenzhou is 10500gCm -2 yr -1 (exceeding the upper limit of 9800gCm -2 yr -1 ), included in the final accounting.

[0100] Step 5: Abnormal emission grid labeling and inspection; Threshold setting: Referring to the carbon emission intensity target of the province (unit GDP CO2 emission reduction of 18%), set the abnormal threshold to 10000gCm -2 yr -1 .

[0101] Abnormality labeling: 127 abnormal grids were labeled in 2023, mainly concentrated in a certain district of B city, AB district, and C port district.

[0102] Mobile inspection: dispatch a UAV to inspect the abnormal grids in a certain district of B city, and find 3 high-concentration emission points (smoke concentration gradient > 0.8 ppm / m).

[0103] Step 6: visual analysis and suspect enterprise locking; Feature recognition: YOLOv8 target detection is performed on the UAV image, and 3 chimneys of 3 enterprises are identified (heights are 80m, 60m, and 40m respectively), and emission features are extracted: emission center GPS (121.8°E, 29.9°N), smoke color HSV mean (H=0.12, S=0.3, V=0.6), and diffusion path extending 500m to the southeast.

[0104] Correlation positioning: match the emission features with the enterprise GPS to lock B city a certain paper company (30m away from the emission center) as the suspect enterprise, and generate a visual evidence package containing concentration curves, UAV images, and emission parameters.

[0105] The correlation coefficient R between the inversion results and the online monitoring data 2 is 0.76 (certain aromatic 60m station), which is 3% higher than the prior simulation. Refer to Table 1; Table 1 Accuracy evaluation of CO2 prior and posterior simulation concentration in a certain province in 2023

[0106] Accurate identification of multiple abnormal emission enterprises, 8 of which were confirmed as over-standard emission by on-site verification, and the law enforcement response time was shortened to 48 hours.

[0107] Through this example, the whole-process intelligent verification of carbon emission from global monitoring to precise tracing in a certain province is realized, which provides technical support for the "double carbon" target supervision.

[0108] The above are preferred embodiments of the present application, and are not intended to limit the protection scope of the present application, therefore: any equivalent changes made in structure, shape, principle, etc. according to the present application should be covered within the protection scope of the present application.

Claims

1. A carbon emission verification method based on a multi-mode inversion model, characterized in that, Includes the following steps: Step 1: A three-dimensional monitoring network is formed by using airborne monitoring, ground-based monitoring, mobile monitoring, and mobile monitoring. Step 2: The intelligent inversion calculation layer integrates global background concentration and regional real-time monitoring data, and uses the GSI 4DVar algorithm to optimize the three-dimensional concentration field; Step 3: The intelligent inversion calculation layer divides the inversion calculation into two stages; The first stage of the inversion involves applying random perturbations to the prior emission inventory to generate multiple emission sets, each corresponding to a grid region in the three-dimensional concentration field. The concentration distribution is then simulated using the WRF-CMAQ model. The second stage of the inversion involves assimilating the measured data using the EnSRF algorithm and iteratively optimizing the emission parameters. Step 4: Perform multiple Monte Carlo samplings on the background concentration and the three-dimensional monitoring network. Output the confidence interval through the probability density function of the emission results. The carbon emission value that exceeds the set confidence interval is the final carbon emission value. Output the final carbon emission value for each grid area. Step 5: Preset emission anomaly threshold. If the final carbon emission value output is greater than the emission anomaly threshold, it is marked as an abnormal emission grid, and the mobile monitoring is called to inspect the abnormal emission grid. Step 6: Perform visual analysis on the full-coverage visual inspection data, mark the specific location area of ​​the enterprise in the visual inspection image data, use visual feature algorithms to identify emission characteristics, including emission center, emission path, chimney height, smoke concentration gradient at the chimney outlet, and smoke color feature value, establish the correspondence between emission characteristics and enterprise GPS coordinates, list the corresponding enterprises as suspected enterprises of abnormal emissions, and generate a visual evidence package.

2. The carbon emission verification method based on a multi-mode inversion model according to claim 1, characterized in that, In step 1, space-based monitoring involves synchronous access to a high-resolution satellite, which carries a greenhouse gas detection spectrometer. Ground-based monitoring involves deploying multiple fixed monitoring stations in the target monitoring area. Each fixed monitoring station is equipped with a gas analyzer and a Fourier transform infrared spectrometer. Mobile monitoring involves deploying multiple mobile monitoring vehicles equipped with vehicle-mounted high-precision gas monitoring equipment and positioning systems. These vehicles travel on roads along preset or random routes, collecting gas concentration data in real time during their movement. Mobile monitoring uses drones equipped with portable Fourier transform infrared spectrometers and high-definition visual camera modules to inspect target areas according to instructions.

3. The carbon emission verification method based on a multi-mode inversion model according to claim 1, characterized in that, In step 2, the preprocessed global background concentration data and regional real-time monitoring data are input into the GSI 4DVar algorithm for optimization. During the optimization process, information in the time and spatial dimensions is considered, and the three-dimensional concentration field is dynamically adjusted to generate an optimized three-dimensional concentration field model. The optimized three-dimensional concentration field is output in the form of gridded data, and the grid resolution is set according to the monitoring requirements.

4. The carbon emission verification method based on a multi-mode inversion model according to any one of claims 1-3, characterized in that, The first stage of the inversion includes the following specific steps: collecting basic information on enterprises within the target area, establishing a priori emission inventory, applying random perturbations that conform to a normal distribution to the parameters of each emission source in the priori emission inventory, with the perturbation range set according to the data uncertainty, and generating multiple emission sets; The three-dimensional concentration field is divided into multiple grid regions, each grid region corresponds to an emission set, and a mapping relationship between the emission set and the grid region is established. Multiple emission sets are input into the WRF-CMAQ model to simulate the concentration distribution of the grid region corresponding to each emission set at different times, and output the simulated concentration distribution results.

5. The carbon emission verification method based on a multi-mode inversion model according to any one of claims 1-3, characterized in that, The second stage of the inversion includes the following specific steps: The measured data obtained from the three-dimensional monitoring network in step 1 are screened and preprocessed to extract effective data corresponding to the time and space simulation of the WRF-CMAQ model; the EnSRF algorithm is used to assimilate the effective data with the simulated concentration distribution results. The EnSRF algorithm updates the estimated values ​​of emission parameters by calculating the covariance between the members of the set. In each iteration, the emission parameters in the emission set are adjusted according to the difference between the measured data and the simulated data. An iteration termination condition is set, and iterative optimization continues until the termination condition is met, continuously adjusting the emission parameters until the iteration terminates and the optimal emission parameters are obtained.

6. The carbon emission verification method based on a multi-mode inversion model according to any one of claims 1-3, characterized in that, In step 5, reasonable emission anomaly thresholds are set with reference to industry emission standards, regional environmental capacity, historical carbon emission data, regional functional positioning, and environmental carrying capacity. The final carbon emission value of each grid area output in step 4 is compared with the set emission anomaly threshold. If the final carbon emission value is greater than the emission anomaly threshold, the grid area is marked as an abnormal emission grid, and the grid location and the value exceeding the standard are recorded.

7. The carbon emission verification method based on a multi-mode inversion model according to any one of claims 1-3, characterized in that, In step 6, a deep learning-based target detection algorithm is used to identify the distinctive features of the company's buildings and chimneys in the visual image, marking the specific location of the company. At the same time, a visual feature algorithm is used to extract emission center, emission path, chimney height, smoke concentration gradient at the chimney outlet, and smoke color feature values ​​as emission features. The identified emission characteristics are correlated with the company's GPS coordinates to establish a correspondence. Based on the correlation results, the corresponding companies are listed as suspected companies of abnormal emissions, and the relevant emission characteristic data and image evidence are recorded.

8. The carbon emission verification method based on a multi-mode inversion model according to claim 7, characterized in that, In step 6, the basic information, emission characteristic data, visual inspection images, and GPS coordinate information of the suspected abnormal emission enterprises are integrated to generate a visual evidence package.

9. A carbon emission verification system based on a multi-mode inversion model, characterized in that: To implement the carbon emission verification method based on the multi-mode inversion model as described in any one of claims 1-8, the carbon emission verification system includes an airborne monitoring unit, a ground-based monitoring unit, a mobile monitoring unit, a mobile monitoring unit, and a carbon emission control center computer. The carbon emission control center computer is connected to the airborne monitoring unit, the ground-based monitoring unit, the mobile monitoring unit, and the mobile monitoring unit respectively through a wireless communication network.

10. The carbon emission verification system based on a multi-mode inversion model according to claim 9, characterized in that: The airborne monitoring unit includes multiple high-resolution satellites carrying greenhouse gas detection spectrometers; the ground-based monitoring unit includes multiple fixed monitoring stations equipped with gas analyzers and Fourier transform infrared spectrometers; the mobile monitoring unit includes multiple mobile monitoring vehicles equipped with vehicle-mounted high-precision gas monitoring equipment and positioning systems; and the mobile monitoring unit includes multiple drones equipped with portable Fourier transform infrared spectrometers and high-definition visual camera modules.

Images