A method, system and device for inverting atmospheric pollution emissions based on machine learning

Abstract

The present invention discloses a method, system, and device for inverting atmospheric pollutant emissions based on machine learning. The method comprises: obtaining pollutant concentration observation data within a regional grid; obtaining an influence function and simulated concentration values ​​within the regional grid; establishing an emissions inversion model within the regional grid based on the pollutant concentration observation data, the simulated concentration values ​​within the regional grid, the influence function of the regional grid, and prior emission information; and iteratively calculating the inverted atmospheric pollutant emissions within the regional grid based on the loss function of the emissions inversion model. The technical solution provided by the present invention solves the current problems of low accuracy and resolution of atmospheric pollutant emission inventory data. Furthermore, the atmospheric pollutant emission inversion equipment used is inexpensive, solving the problem of high construction costs for current inversion systems and equipment used.

Description

Technical Field

[0001] The present invention relates to the field of environmental monitoring and measurement technology, and in particular to an atmospheric pollutant emission inversion method, system and equipment based on machine learning. Background Art

[0002] Air pollution occurs when certain substances enter the atmosphere due to human activities or natural processes, reaching sufficient concentrations for a sufficient period of time to endanger human comfort, health, and well-being, or the environment. Air pollution has become one of the most serious environmental problems currently facing the world. Obtaining accurate, high-resolution data on air pollutant emissions is key to achieving precise air pollution control. Currently, a "bottom-up" inventory compilation approach is commonly used. This approach relies on surveys, statistics, and emission factors to obtain information on regional air pollutant emissions. This approach is labor-intensive and financially intensive, with significant data errors. It also suffers from time lags and a lack of spatial and temporal distribution characteristics.

[0003] In recent years, "top-down" atmospheric pollutant emission inversion systems based on methods such as Bayesian statistics and Kalman filtering have achieved optimization of emission source inventories. However, this requires the establishment of high-density, high-precision air quality monitoring stations within the target area, which is extremely costly and difficult to promote. Furthermore, the amount of monitoring data obtained from the establishment of a small number of air quality monitoring stations cannot meet the accuracy requirements of the above-mentioned inversion systems. Although the use of low-cost detectors can significantly reduce costs, due to defects such as poor instrument accuracy and difficulty in calibration, they still cannot replace high-precision air quality monitoring stations for emission inversion calculations and the construction of related inversion systems. Therefore, how to improve the accuracy and resolution of atmospheric pollutant emission data while reducing the construction cost of emission inversion monitoring networks is an urgent problem to be solved in atmospheric environmental monitoring and governance. Summary of the Invention

[0004] In view of this, the present invention provides an atmospheric pollutant emission inversion method, system and equipment based on machine learning, which can provide complete data required for emission inversion, thereby improving the accuracy and resolution of atmospheric pollutant emission data while reducing costs.

[0005] According to a first aspect, an atmospheric pollutant emission inversion method is applied to an atmospheric pollutant emission inversion device, wherein a target area for measuring atmospheric pollutant emissions is divided into a plurality of preset resolution area grids, and the method comprises:

[0006] Obtaining pollutant concentration observation data within the regional grid;

[0007] Obtaining the influence function and simulated concentration value within the regional grid;

[0008] Establishing an emission inversion model within the regional grid based on the pollutant concentration observation data, the simulated concentration value within the regional grid, the regional grid influence function and prior emission information;

[0009] The inverted emissions of atmospheric pollutants in the regional grid are iteratively calculated according to the loss function of the emission inversion model.

[0010] Optionally, an air microstation is deployed in the regional grid, and obtaining the pollutant concentration observation data in the regional grid includes:

[0011] Acquire dynamic data and static data within a regional grid according to the multivariate environmental variables of the air microstation and the target area;

[0012] After cleaning the dynamic data, combine it with the static data to obtain the algorithm sample;

[0013] Dividing the algorithm samples into training samples and prediction samples according to whether a standard air quality monitoring station is set up in the regional grid where the algorithm samples are located;

[0014] Establishing a calibration model based on the training samples in combination with a machine learning algorithm;

[0015] The predicted samples are used as input to the calibration model to obtain output values, which are the pollutant concentration observation data within the regional grid after calibration.

[0016] Optionally, obtaining the influence function and the simulated concentration value in the regional grid includes:

[0017] Driving the backward trajectory diffusion model according to regional meteorological field data, and calculating air particle footprint weights as the influence function matrix;

[0018] The simulated concentration value within the regional grid is calculated based on the influence function matrix and the priori emission information.

[0019] Optionally, the establishing of an emission inversion model within a regional grid by combining the pollutant concentration observation data, the simulated concentration value within the regional grid, the regional grid influence function and the prior emission information includes:

[0020] A loss function of the inversion optimization process is established based on the difference between the simulated concentration value and the pollutant concentration observation data, the difference between the emission posterior information and the prior information, the error covariance of the pollutant concentration observation data and the error covariance with the prior emission information.

[0021] The iterative calculation of the inverted emissions of atmospheric pollutants in the regional grid according to the loss function of the emission inversion model includes:

[0022] Iteratively calculating the minimum value of the loss function;

[0023] The posterior distribution of the corresponding emission when the loss function takes the minimum value is used as the inverted emission of atmospheric pollutants after optimization and calibration.

[0024] The algorithm samples are divided into training samples and prediction samples according to whether a standard air quality monitoring station is set up in the regional grid where the algorithm samples are located, including:

[0025] When the standard air quality monitoring station is set in the regional grid, the corresponding algorithm sample in the regional grid is used as the training sample;

[0026] When there is no standard air quality monitoring station in the regional grid, the corresponding algorithm sample in the regional grid is used as the prediction sample;

[0027] According to a second aspect, an atmospheric pollutant emission inversion system is applied to an atmospheric pollutant emission inversion device, wherein a target area for measuring atmospheric pollutant emissions is divided into a plurality of preset resolution area grids, and the system comprises:

[0028] A machine learning module, which obtains pollutant concentration observation data within the regional grid;

[0029] The trajectory diffusion module applies the backward trajectory diffusion model to calculate the regional grid influence function matrix and combines it with the prior emission information to obtain the simulated concentration value within the regional grid;

[0030] An inversion model module is configured to establish an emission inversion model within a regional grid by combining the pollutant concentration observation data, the simulated concentration value within the regional grid, the regional grid influence function, and the prior emission information;

[0031] The output module iteratively calculates and outputs the inverted emissions of atmospheric pollutants in the regional grid according to the loss function of the emission inversion model.

[0032] According to a third aspect, an atmospheric pollutant emission monitoring device includes: an air micro-station sensor, a main control unit, and a display module, wherein the air micro-station sensor is used to obtain a raw pollutant concentration signal within a preset resolution grid in a region and send the pollutant concentration signal to the main control unit, and the main control unit processes the pollutant concentration signal and sends it to the display module for display, and the main control unit includes:

[0033] A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method described in the first aspect or any optional embodiment of the first aspect by executing the computer instructions.

[0034] The technical solution of the present invention has the following advantages:

[0035] The present invention provides a method and system for inverting atmospheric pollutant emissions, which are applied to atmospheric pollutant emission inversion equipment. The target area for measuring atmospheric pollutant emissions is divided into several high-resolution grid regions, and air microstations are deployed within the regional grids. The method specifically includes: obtaining pollutant concentration observation data within the regional grids; applying a backward trajectory diffusion model to calculate the regional grid influence function matrix, and combining it with prior emission information to obtain simulated concentration values ​​within the regional grid; establishing an emissions inversion model within the regional grid by combining the pollutant concentration observation data, the simulated concentration values ​​within the regional grid, the regional grid influence function, and prior emission information; and iteratively calculating the inverted atmospheric pollutant emissions within the regional grid based on the loss function of the emissions inversion model. Thus, high-resolution grid concentration observation data in the area is obtained through the air micro-station, and the pollutant concentration observation data of the area without standard air quality monitoring stations is trained by machine learning algorithms to calibrate the grid. This solves the problem that the air micro-station has been unable to accurately and effectively serve the urban air pollution prevention and control work due to low measurement accuracy and high uncertainty. The concentration observation data after calibration and optimization is substituted into the emission inversion model, which greatly increases the amount of data of known information in the inversion calculation, realizes the optimization of the influence function and inversion iteration, and then obtains a high-resolution and accurate "top-down" atmospheric pollutant emission inventory, solving the current problem of low accuracy and resolution of atmospheric pollutant emission data. At the same time, the air micro-station used in this patent is cheaper than the high-precision air quality monitoring station, solving the problem of high cost of building the current atmospheric pollutant emission inversion monitoring network. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0037] Figure 1 A schematic diagram of the steps of a method for inverting atmospheric pollutant emissions according to an embodiment of the present invention;

[0038] Figure 2 A schematic flow chart of a method for inverting atmospheric pollutant emissions according to an embodiment of the present invention;

[0039] Figure 3 A schematic diagram of a process for acquiring regional grid pollutant concentration observation data according to an embodiment of the present invention;

[0040] Figure 4 A schematic diagram of the inversion process of an atmospheric pollutant emission inversion method according to an embodiment of the present invention;

[0041] Figure 5 This is a schematic structural diagram of an atmospheric pollutant emission inversion system according to an embodiment of the present invention;

[0042] Figure 6 This is a schematic structural diagram of an atmospheric pollutant emission inversion device according to an embodiment of the present invention;

[0043] Figure 7 Schematic diagram of the main control unit structure of an embodiment of the present invention. DETAILED DESCRIPTION

[0044] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0045] The technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0046] See also Figure 1 and Figure 2 The atmospheric pollutant emission inversion method provided by the embodiment of the present invention is based on the target area of ​​the atmospheric pollutant emission to be measured being pre-divided into a plurality of preset resolution grid areas. The atmospheric pollutant emission inversion method specifically includes the following steps:

[0047] Step S101: Obtain pollutant concentration observation data within the regional grid. Specifically, the division accuracy of the regional grid is within the preset resolution. For example, in the embodiment of the present invention, the grid resolution is set to no less than 3km×3km, and high-resolution regional division is used to obtain high-resolution pollutant observation information and grid prior information, with the purpose of obtaining high-resolution and accurate pollutant inversion emissions. The present invention is not limited to this. Air micro-stations are deployed in the regional grid. Combined with the standard air quality monitoring stations in the area, high-resolution calibrated pollutant concentration observation data in the regional grid are obtained through machine learning algorithms. Therefore, in the subsequent inversion step, the amount of known information is increased by a large amount of high-temporal and spatial resolution concentration observation data, a more accurate inversion result is obtained, and the accuracy of the inversion system is improved.

[0048] Step S102: Obtain the influence function and simulated concentration value within the regional grid. Specifically, first obtain the meteorological field information within each regional grid of the target area according to the NECP / NCAR website or satellite radar and other channels. Apply the backward trajectory diffusion model, and driven by the meteorological field, release a large number of air particles backward based on the position of the air micro-station observation point in each grid to simulate the backward motion trajectory of the gas and obtain the grid influence function matrix. The simulated concentration value within the regional grid can be obtained based on the influence function matrix and prior emission information. In this way, more accurate and complete inversion data are obtained, and the footprint weight (influence function) information volume and simulation accuracy are significantly improved compared to those obtained using only a small amount of standard air quality monitoring station observation data, which plays a vital role in improving the accuracy of subsequent inversion steps.

[0049] Step S103: An emissions inversion model for the regional grid is established by combining the observed pollutant concentration data, the simulated concentration values ​​within the regional grid, the regional grid influence function, and the prior emissions information. Specifically, a loss function for the inversion optimization process is established based on the difference between the simulated concentration and the observed value at the observation point, the difference between the posterior and prior emissions information, the error covariance of the observed data, and the error covariance with the prior emissions. The inversion model can be established using commonly used inversion algorithms, such as Bayesian theory and Kalman filtering, but the present invention is not limited thereto.

[0050] Step S104: iteratively calculating the inverted emissions of atmospheric pollutants within the regional grid according to the loss function of the emission inversion model.

[0051] Specifically, in one embodiment, the above step S101 specifically includes the following steps:

[0052] Step S201: Dynamic and static data within the regional grid are acquired based on the air microstation and the target area's multivariate environmental variables. Specifically, an air microstation is installed within each regional grid. The present invention utilizes, but is not limited to, an XD-YL-1 sensor. The air microstation is used to acquire atmospheric pollutant concentration monitoring data with high temporal and spatial resolution.

[0053] Static data obtained through field surveys or official websites of relevant departments includes, but is not limited to, land use information, transportation network information, and prior pollutant emission information. Land use information includes the land use type within the grid and the area occupied by each type of land within the grid. Transportation network information includes the road class within the grid and the length of each class of road. Prior pollutant emission information includes the prior emission inventory of each pollutant within the grid. Information such as the altitude of air microstations is also provided.

[0054] The dynamic data includes but is not limited to: meteorological monitoring data, traffic emission data, pollutant concentration monitoring data of air micro-stations and concentration monitoring data of standard air quality monitoring stations. Meteorological monitoring data include: meteorological monitoring data such as temperature, humidity, wind speed, wind direction, etc. with high temporal resolution within the grid. The method of obtaining meteorological data within the grid can be based on the meteorological monitoring data such as temperature, humidity, wind speed, wind direction, etc. with high temporal resolution within the region, and selecting the meteorological station data closest to the grid to complete the grid meteorological data matching; it can also be based on WRF or other meteorological models to obtain various meteorological parameter information under the same spatial grid, but the present invention is not limited to this. Traffic emission data include: for cities that have established traffic road network emission systems, obtaining high temporal resolution, road section-level pollutant traffic emission data within the corresponding grid. The atmospheric pollutant concentration monitoring data of the air micro-station is pollutant concentration monitoring data with high temporal and spatial resolution. In the present invention, high temporal resolution means that the data update frequency is not less than 1 hour, and high spatial resolution means that the grid accuracy is not less than 3km×3km. The concentration monitoring data of the standard air quality monitoring station includes: the high-time-resolution monitoring data of the concentration of each pollutant at the ambient air quality standard station in the region. If the region is equipped with air quality monitoring equipment with the same or higher accuracy as the above-mentioned ambient air quality standard station, the relevant parameters can be obtained at the same time.

[0055] The above-mentioned meteorological field information, land use information, traffic network information, corresponding grid prior emission information, and traffic network emission information are introduced into machine learning training, so that the machine learning training fully considers the impact of factors such as basic environmental characteristics, atmospheric transmission characteristics, and background emission characteristics on the concentration contribution, thereby improving the training effect.

[0056] Step S202: After cleaning the dynamic data, combine it with the static data to obtain an algorithm sample. Specifically, the dynamic variable data in step S201 is subjected to outlier removal and then combined with the static data to generate a machine learning algorithm sample file.

[0057] Step S203: Based on whether a standard air quality monitoring station is installed in the area grid where the algorithm sample is located, the algorithm sample is divided into a training sample and a prediction sample. Specifically, for area grids where a standard air quality monitoring station is installed, the corresponding algorithm sample serves as the training sample for the calibration model. For area grids where a standard air quality monitoring station is not installed, the corresponding algorithm sample serves as the prediction sample for the prediction model.

[0058] Step S204: Establish a calibration model based on the training samples combined with the machine learning algorithm. Specifically, the training samples in step S203 are substituted into the machine learning algorithm model for training the calibration model. For example: XGBoost, random forest, etc., the present invention is not limited to this. This embodiment takes XGBoost as an example, and the specific training process is as follows:

[0059] ① Dataset Splitting: Randomly split the sample dataset and proportionally combine it into a training set and a test set for cross-validation. This example uses ten-fold cross-validation, randomly splitting the training sample data of the calibration model into 10 parts, and rotating 9 of them as training data and 1 as test data for the experiment.

[0060] ② XGBoost model hyperparameter adjustment: Adjust parameters including booster, eta, min_child_weight, max_depth, gamma, subsample, colsample_bytree, and tree_method. Select the hyperparameters that perform best on the 10 test sets as the model hyperparameters. For specific evaluation methods, see the section on training effect evaluation.

[0061] ③ Training effect evaluation: Statistical methods are used to compare the simulated pollutant concentration values ​​and the true values. In this invention, the pollutant concentration values ​​monitored by the standard air quality monitoring station are used as the true values. The statistical indicators used to evaluate the training effect in this implementation method include root mean square error (RMSE), mean absolute error (MAE), Person correlation coefficient, and determination coefficient.

[0062] ④Training the model: Use the model hyperparameters determined based on the 10 test results to train the model and finally determine the calibration model.

[0063] Step S205: The predicted samples are used as input to the calibration model to obtain the output value, which is the pollutant concentration observation data within the calibrated regional grid. Specifically, the predicted samples used for the prediction model are substituted into the calibration model finally determined in step S204 to obtain the output data, thereby calibrating the air microstation monitoring data and obtaining the pollutant concentration observation data within the calibrated regional grid. In the subsequent inversion model construction phase, the regional gridded pollutant concentration observation data with high temporal and spatial resolution after optimization and calibration is used, which greatly increases the amount of known information in the inversion calculation, optimizes the influence function, inversion iteration and other links, and improves the model accuracy.

[0064] Specifically, in one embodiment, the above step S102 includes the following steps:

[0065] Step S206: driving the backward trajectory diffusion model according to the regional meteorological field data, and calculating the air particle footprint weight as the influence function matrix.

[0066] Specifically, meteorological field information for a preset region can be obtained by directly downloading meteorological reanalysis data for the target region from the NCEP / NCAR website. The NCEP / NCAR reanalysis dataset utilizes the most advanced global data assimilation system and comprehensive database, performing quality control and assimilation on observational data from various sources (ground, ship, radiosonde, wind balloon, aircraft, satellite, etc.). It not only contains numerous elements and a wide range, but also extends over a long period of time, with a maximum resolution of 0.25°×0.25°. Preferably, higher-resolution meteorological field output files within the region can also be obtained using mesoscale meteorological field models such as WRF, with a resolution generally reaching a grid accuracy of 500-1000m, but the present invention is not limited to this.

[0067] This embodiment applies the transmission model of Lagrangian random walk theory to link the emission flux upstream of the regional grid observation point with the concentration change of the observation point using footprint weights. Figure 3 As shown, driven by the meteorological field, a large number of air particles are released backward, simulating the backward motion trajectory of the gas driven by turbulence and average wind direction. The footprint weight is calculated by calculating the number of all particles within the boundary layer height upstream of the observation point and the residence time of each particle within the grid, which serves as the influence function matrix for emission inversion. Based on the observed pollutant concentration data within each regional grid obtained by machine learning, the present invention can release particles within each grid and calculate the distribution characteristics of the footprint weight under the scenario of particle release in each grid. Compared with the footprint weight obtained by only using observation data from a small number of standard air quality monitoring stations, the information content and simulation accuracy are significantly improved.

[0068] Step S207: Calculate the simulated concentration value within the regional grid based on the influence function matrix combined with the prior emission information. Specifically, in this embodiment, the prior emission information is first vectorized to form the emission state matrix. Then, the simulated concentration contribution value is calculated, where the simulated concentration contribution value = influence function matrix × emission state matrix + error term. Then, the background concentration of pollutants in the region is obtained. The above regional pollutant background concentration can be obtained through websites such as EDGAR, CarbonTracker, Vulcan, and local survey data. The final simulated concentration value within the regional grid is:

[0069] Simulated concentration value = simulated concentration contribution value + regional pollutant background concentration

[0070] The obtained simulated concentration values ​​with high temporal and spatial resolution provide a large number of data samples for subsequent inversion calculations, increase the amount of data for known information in inversion calculations, and improve the accuracy of the inversion results.

[0071] Specifically, in one embodiment, the above step S103 specifically includes the following steps:

[0072] Step S208: An inversion model for emissions within the regional grid is established by combining the observed pollutant concentration data, the simulated concentration values ​​within the regional grid, the regional grid influence function, and the prior emissions information. Specifically, this embodiment combines the Bayesian inversion method and the Kalman filter concept to establish and illustrate the inversion model, but the present invention is not limited to this.

[0073] Specifically, in one embodiment, the above step S104 includes the following steps:

[0074] Step S209: Calculate the minimum value of the loss function. Figure 4 As shown in the figure, specifically, high-time-resolution pollutant concentration observation data are used to constrain emission information that may evolve over time, that is, the Kalman filter algorithm is introduced, and the concentration observation data at the new moment are combined to obtain a new gain matrix, posterior emission flux, and posterior emission error covariance. The posterior calculated value at the previous moment is then used as the prior information at the next moment and substituted into the loss function equation. The process is iterated continuously to calculate the optimal solution of the loss function.

[0075] The observation data (new information) at each moment in the present invention are all calibrated values ​​of the observed concentration in the regional grid obtained through XGBoost machine learning. At the same time, the influence function of each grid observation point is substituted into the Kalman gain matrix. In this way, each iterative solution process introduces the observation values ​​calibrated by machine learning optimization and the influence function of each grid (observation point), and finally obtains the optimal posterior emission information, thereby improving the accuracy of the calculation results.

[0076] Step S210: The posterior distribution of the corresponding emission when the loss function reaches the minimum value is the optimized inverted emission.

[0077] By executing the above steps, the embodiment of the present invention optimizes the observation data of low-cost air micro-stations in the region through machine learning training, obtains gridded concentration observation data with high spatiotemporal resolution after calibration in the region, and further applies the calibration data of pollutant concentration observation values ​​after machine learning training to the influence function, loss function and inversion iterative calculation, etc., which greatly increases the amount of data of known information in the inversion calculation, ensures the accuracy and reliability of the inverted emission data after calibration, and finally obtains a high-resolution and accurate pollutant emission inventory, solving the current problem of poor accuracy and low resolution of emission inventory data. At the same time, air micro-stations are cheaper than standard air quality monitoring stations, solving the problem of high cost.

[0078] like Figure 5 As shown, this embodiment also provides an atmospheric pollutant emission inversion system, which is applied to an atmospheric pollutant emission inversion device. The target area based on the atmospheric pollutant emission to be measured is divided into a plurality of preset resolution area grids. The system includes:

[0079] The machine learning module 101 obtains the pollutant concentration observation data within the regional grid. For details, please refer to the relevant description of step S101 in the above method embodiment, which will not be repeated here.

[0080] The trajectory diffusion module 102 obtains the influence function and the simulated concentration value in the regional grid. For details, please refer to the relevant description of step S102 in the above method embodiment, which will not be repeated here.

[0081] Inversion model module 103 combines the observed pollutant concentration data, the simulated concentration values ​​within the regional grid, the regional grid influence function, and the prior emission information to establish an emission inversion model within the regional grid. For details, see the description of step S103 in the above method embodiment and will not be repeated here.

[0082] Output module 104 iteratively calculates and outputs the inverted emissions of atmospheric pollutants in the regional grid based on the loss function of the emission inversion model. For details, please refer to the relevant description of step S104 in the above method embodiment, which will not be repeated here.

[0083] The atmospheric pollutant emission inversion system provided in an embodiment of the present invention is used to execute the atmospheric pollutant emission inversion method provided in the above embodiment. Its implementation method and principle are the same. For details, please refer to the relevant description of the above method embodiment and will not be repeated here.

[0084] Figure 6 As shown, this embodiment also provides an atmospheric pollutant emission inversion device, which includes: an air micro-station sensor 1, a main control unit 2 and a display module 3, wherein the air micro-station sensor 1 is used to obtain the original pollutant concentration signal in the regional high-resolution grid and send the pollutant concentration signal to the main control unit 2, and the main control unit 2 calculates the inversion system process for the pollutant concentration signal and other relevant static and dynamic data stored and received in this patent, and sends the inversion system output result file to the display module 3 for display. Preferably, the display module 3 can combine the relevant functions of the geographic information system (GIS) to display the optimized high-resolution pollutant inversion emission results on a visual map. At the same time, according to the application requirements of different scenarios, sub-modules at multiple levels such as large screen, PC, and mobile phone can be established.

[0085] The main control unit includes: a processor 901 and a memory 902, which can be connected via a bus or other means. Figure 7 The bus connection is taken as an example.

[0086] The processor 901 may be a central processing unit (CPU). The processor 901 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0087] The memory 902 is a non-transient computer-readable storage medium that can be used to store non-transient software programs, non-transient computer executable programs, and modules, such as the program instructions / modules corresponding to the methods in the above-mentioned method embodiments. The processor 901 executes various functional applications and data processing of the processor by running the non-transient software programs, instructions, and modules stored in the memory 902, that is, implementing the methods in the above-mentioned method embodiments. In this embodiment, the REST API interface is called for the national environmental monitoring data, and the received JSON data is parsed. After the data is cleaned, it is stored in the database by calling the JDBC driver; the air microstation data uses NETTY to create a local TCP server, continuously listens to a local specific port, and after the data is cleaned of abnormal values, it is stored in the database by calling the JDBC driver; the meteorological data uses VSFTPD to create a local FTP server, uses FLUME to listen to a specific resource directory, and stores the data in the database through FLUME's custom sink module; dynamic data such as traffic flow and emissions are received via TCP or OPEN API, with links level as the storage range, and the data is stored in the MySQL database and synchronously backed up to the Hadoop cluster.

[0088] The memory 902 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created by the processor 901, etc. In addition, the memory 902 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 902 may optionally include a memory remotely located relative to the processor 901, and these remote memories may be connected to the processor 901 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0089] One or more modules are stored in the memory 902 and, when executed by the processor 901 , perform the method in the above method embodiment.

[0090] The specific details of the above-mentioned atmospheric pollutant emission inversion equipment can be understood by referring to the corresponding relevant descriptions and effects in the above-mentioned method embodiments, and will not be repeated here.

[0091] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by hardware associated with computer program instructions. The implemented program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD), or a solid-state drive (SSD). The storage medium can also include a combination of the above-mentioned types of memory.

[0092] By applying the above-mentioned atmospheric pollutant emission inversion system and equipment, the automatic calculation and storage of the patented atmospheric pollutant emission inversion method can be realized, and the obtained calibrated high-resolution atmospheric pollutant emission inventory can be visualized, helping the management and control departments to fully grasp the characteristics of urban air pollution emissions, lock in the pollution sources, and identify high-emission areas, so as to formulate accurate and effective emission reduction policies and governance plans.

[0093] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for inverting atmospheric pollution emissions based on machine learning, applied to an atmospheric pollutant emission inversion device, characterized in that: The target area for measuring atmospheric pollutant emissions is divided into a plurality of preset resolution area grids, wherein air micro stations are deployed in the area grids. The method includes: Obtaining pollutant concentration observation data within the regional grid; Obtaining the influence function and simulated concentration value within the regional grid; obtaining the influence function and simulated concentration value within the regional grid includes: driving a backward trajectory diffusion model based on regional meteorological field data to calculate air particle footprint weights as an influence function matrix; and calculating the simulated concentration value within the regional grid based on the influence function matrix combined with prior emission information; Establishing an emission inversion model within the regional grid based on the pollutant concentration observation data, the simulated concentration value within the regional grid, the regional grid influence function and prior emission information; The inverted emissions of atmospheric pollutants in the regional grid are iteratively calculated according to the loss function of the emission inversion model.

2. The method according to claim 1, characterized in that The obtaining of pollutant concentration observation data within the regional grid includes: Acquire dynamic data and static data within a regional grid according to the multivariate environmental variables of the air microstation and the target area; After cleaning the dynamic data, combine it with the static data to obtain the algorithm sample; Dividing the algorithm samples into training samples and prediction samples according to whether a standard air quality monitoring station is set up in the regional grid where the algorithm samples are located; Establishing a calibration model based on the training samples in combination with a machine learning algorithm; The predicted samples are used as input to the calibration model to obtain output values, which are the pollutant concentration observation data within the regional grid after calibration.

3. The method according to claim 1, characterized in that The method of establishing an emission inversion model within a regional grid by combining the pollutant concentration observation data, the simulated concentration value within the regional grid, the regional grid influence function, and the prior emission information includes: A loss function of the inversion optimization process is established based on the difference between the simulated concentration value and the pollutant concentration observation data, the difference between the emission posterior information and the prior information, the error covariance of the pollutant concentration observation data and the error covariance with the prior emission information.

4. The method according to claim 1, wherein The iterative calculation of the inverted emissions of atmospheric pollutants in the regional grid according to the loss function of the emission inversion model includes: Iteratively calculating the minimum value of the loss function; The posterior distribution of the corresponding emission when the loss function takes the minimum value is used as the inverted emission of atmospheric pollutants after optimization and calibration.

5. The method according to claim 2, characterized in that The algorithm samples are divided into training samples and prediction samples according to whether a standard air quality monitoring station is set up in the regional grid where the algorithm samples are located, including: When the standard air quality monitoring station is set in the regional grid, the corresponding algorithm sample in the regional grid is used as the training sample; When there is no standard air quality monitoring station in the regional grid, the corresponding algorithm sample in the regional grid is used as the prediction sample.

6. A machine learning-based atmospheric pollution emission inversion system, applied to atmospheric pollutant emission inversion equipment, characterized in that: The target area for measuring atmospheric pollutant emissions is divided into a preset resolution area grid. The system includes: A machine learning module, which obtains pollutant concentration observation data within the regional grid; A trajectory diffusion module applies a backward trajectory diffusion model to calculate a regional grid influence function matrix and obtains a simulated concentration value within the regional grid in combination with prior emission information; obtaining the influence function and simulated concentration value within the regional grid includes: driving the backward trajectory diffusion model according to regional meteorological field data, calculating air particle footprint weights as the influence function matrix; and calculating the simulated concentration value within the regional grid based on the influence function matrix and prior emission information; An inversion model module is configured to establish an emission inversion model within a regional grid by combining the pollutant concentration observation data, the simulated concentration value within the regional grid, the regional grid influence function, and the prior emission information; The output module iteratively calculates and outputs the inverted emissions of atmospheric pollutants in the regional grid according to the loss function of the emission inversion model.

7. An atmospheric pollutant emission inversion device based on machine learning, characterized in that: include: An air micro-station sensor, a main control unit, and a display module. The air micro-station sensor is used to obtain the original pollutant concentration signal within the preset resolution grid of the area and send the pollutant concentration signal to the main control unit. The main control unit processes the pollutant concentration signal and sends it to the display module for display. The main control unit includes: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method according to any one of claims 1 to 5 by executing the computer instructions.

Images