Air pollution analysis method and system based on motor vehicle exhaust

Abstract

The present invention discloses a method and system for analyzing atmospheric pollution based on motor vehicle exhaust. The method comprises determining the exhaust gas collection amount according to the vehicle operating condition, collecting exhaust gas to obtain automobile exhaust characteristics, constructing an exhaust gas prediction model, collecting traffic monitoring images and atmospheric monitoring indicators, performing image processing and constructing a vehicle knowledge graph, determining comprehensive automobile exhaust characteristics and constructing an atmospheric indicator prediction model, selecting representative monitoring points according to the regional characteristics of the area to be analyzed, obtaining predicted atmospheric indicators for the representative monitoring points, calculating a monitoring pollution score based on the predicted atmospheric indicators and corresponding reference indicators, determining a density weight based on the regional characteristics, and determining an atmospheric pollution score and atmospheric pollution level for the area to be analyzed based on the pollution scores and density weights of each representative monitoring point. This method not only improves the efficiency and accuracy of atmospheric pollution analysis, but also has good interpretability and can be directly applied to atmospheric pollution analysis systems.

Description

Technical Field

[0001] The present invention relates to the field of air pollution, and in particular to an air pollution analysis method and system based on motor vehicle exhaust. Background Art

[0002] In today's society, with the rapid increase in the number of motor vehicles, motor vehicle exhaust emissions have become one of the key causes of air pollution. Accurate and efficient air pollution analysis based on motor vehicle exhaust is a research hotspot and key direction in the current field of environmental monitoring and governance. It is of irreplaceable importance for formulating scientific and reasonable environmental governance strategies, protecting people's physical and mental health, and promoting sustainable development.

[0003] However, traditional vehicle exhaust air pollution analysis technology has many limitations. First, the fixed sampling points and uniform sampling volume make it difficult to accurately collect exhaust gas based on different vehicle operating conditions, resulting in a lack of representative exhaust gas data and a failure to truly reflect actual pollution conditions. Second, in the atmospheric pollution prediction process, it often relies on a single data source or simple empirical models, failing to comprehensively consider multi-dimensional factors such as traffic flow characteristics, vehicle operating conditions, and atmospheric monitoring indicators, resulting in low prediction accuracy. In recent years, the development of new technologies such as artificial intelligence and big data has provided new ideas and directions for atmospheric pollution analysis. This paper proposes a method and system for atmospheric pollution analysis based on vehicle exhaust. By determining the exhaust gas collection volume based on vehicle operating conditions, constructing an exhaust gas prediction model, a vehicle knowledge graph, and an atmospheric indicator prediction model, and utilizing a particle search algorithm to select representative monitoring points, the present invention can comprehensively and accurately analyze and predict atmospheric pollution caused by vehicle exhaust. This overcomes the shortcomings of existing technologies, significantly improves the accuracy of data collection and the precision of analysis and prediction, provides a more reliable decision-making basis for atmospheric pollution prevention and control, and has important practical significance for promoting the improvement of atmospheric environmental quality. Summary of the Invention

[0004] The purpose of the present invention is to provide an atmospheric pollution analysis method and system based on motor vehicle exhaust.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0006] The present invention comprises the following steps:

[0007] Determining the exhaust gas collection amount according to the vehicle operating condition, collecting the exhaust gas, and analyzing the collected exhaust gas to obtain automobile exhaust characteristics; the vehicle operating condition includes a first vehicle operating condition and a second vehicle operating condition;

[0008] Building an exhaust prediction model based on the vehicle operating condition and the automobile exhaust characteristics, collecting traffic flow monitoring images and atmospheric monitoring indicators, performing image processing to obtain traffic flow characteristics and a vehicle statistics set, and building a vehicle knowledge graph based on the second vehicle operating condition and the traffic flow characteristics;

[0009] Determining comprehensive automobile exhaust characteristics based on the automobile exhaust characteristics and the vehicle statistical set, and constructing an atmospheric index prediction model based on the comprehensive automobile exhaust characteristics, the atmospheric monitoring index, and the corresponding environmental conditions;

[0010] Select representative monitoring points based on the regional characteristics of the area to be analyzed, input the traffic flow characteristics of the representative monitoring points into the vehicle knowledge graph to obtain a predicted second vehicle operating condition, obtain a predicted first vehicle operating condition based on the vehicle statistics set of the representative monitoring points, and obtain a predicted comprehensive automobile exhaust characteristic based on the exhaust prediction model;

[0011] The predicted comprehensive automobile exhaust characteristics are input into the atmospheric index prediction model to obtain the predicted atmospheric index, the monitoring pollution score is calculated based on the predicted atmospheric index and the corresponding reference index, the density weight is determined based on the regional characteristics, and the atmospheric pollution score and atmospheric pollution level of the area to be analyzed are determined based on the pollution score and density weight of each representative monitoring point.

[0012] Furthermore, the method for obtaining automobile exhaust characteristics includes:

[0013] The first vehicle operating condition includes energy consumption type, brand and model; the brand and model correspond to engine parameters; the second vehicle operating condition is vehicle-mounted parameters, including vehicle speed, rotational speed and load, obtained through the vehicle-mounted OBD system;

[0014] Determine a baseline exhaust volume according to a first vehicle operating condition, measure actual exhaust volumes under different second vehicle operating conditions, determine the relationship between the actual exhaust volume, the second vehicle operating condition, and the baseline exhaust volume by fitting, determine the exhaust gas collection volume according to the actual exhaust volume, construct a flow control unit according to the exhaust gas collection volume, the first vehicle operating condition, and the second vehicle operating condition, and connect the exhaust gas collection device to the flow control unit and the on-board OBD system to perform variable-flow exhaust gas collection; the baseline exhaust volume is a factory-calibrated value associated with the first vehicle operating condition; the actual exhaust volume is in a fixed ratio to the exhaust gas collection volume; the flow control unit is used to connect to the on-board OBD system to determine the vehicle operating condition, and determine the exhaust gas collection volume according to the vehicle operating condition.

[0015] The collected exhaust gas is analyzed using high-resolution infrared spectroscopy technology and miniaturized proton transfer reaction mass spectrometry technology to obtain automobile exhaust characteristics; the automobile exhaust characteristics include conventional exhaust components, oxygen-containing organic compound components, polycyclic aromatic hydrocarbon components and volatile organic compound components.

[0016] Furthermore, the method for constructing a vehicle knowledge graph includes:

[0017] Different vehicle operating conditions and corresponding automobile exhaust characteristics are combined into a vehicle operating condition exhaust set, and the vehicle operating condition exhaust set is used to build the exhaust prediction model performance;

[0018] Randomly selecting monitoring points to collect traffic monitoring images and atmospheric monitoring indicators; the atmospheric monitoring indicators are related to air pollution, including conventional pollutants, characteristic derivatives, environmental indicators and meteorological elements;

[0019] Preprocessing and vehicle detection and tracking are performed on traffic flow monitoring images to assign a unique ID to each detected vehicle. Image processing is performed on each vehicle image to obtain traffic flow characteristics. The traffic flow characteristics include traffic volume, speed characteristics, spatial characteristics, and traffic flow composition. The speed characteristics include average speed, speed distribution, and speed stability. The spatial characteristics include lane occupancy and vehicle spacing. The traffic flow composition includes vehicle type distribution and vehicle attributes.

[0020] Each detected vehicle is input into the ResNet-18 convolutional neural network for fine-grained vehicle classification, and a vehicle statistics set is obtained based on the vehicle fine-grained classification results; the vehicle statistics set includes vehicle brand and model, vehicle age, and corresponding quantity;

[0021] Constructing a vehicle knowledge graph based on the second vehicle operating condition and traffic flow characteristics. Specific steps include: creating core nodes based on vehicle type, road type, and pollution type; applying a three-step weighting algorithm to calculate edge weights between two nodes and automatically filtering weakly connected edges with weights < 0.2; and setting real-time weights and graph reconstruction frequency.

[0022] The three-step weight algorithm includes spatiotemporal correlation weight calculation, pollution contribution weight calculation, road network topology weight calculation and weight fusion, and the expression is:

[0023]

[0024]

[0025]

[0026] in Contribution weight for pollution, is a set of vehicle types, is the number of vehicle models, is the emission factor of the corresponding vehicle type, is the vehicle dwell time, Monitoring area, is the road network topology weight, is the road grade, is the angle between the road direction and the wind direction, is the maximum impact distance, is the vertical distance from the monitoring point to the road, is the wind speed correction factor, is the wind speed, is the building density index, is the altitude correction factor, is the average building height, is the edge weight determined by the three-step weight algorithm, is the spatiotemporal correlation weight, which is determined based on the time series similarity of the DTW algorithm. 、 、 is the weight adjustment parameter.

[0027] Furthermore, the method for constructing an atmospheric index prediction model includes:

[0028] According to the automobile exhaust characteristics and the vehicle statistics set, the exhaust emissions of all automobiles in the monitoring area of ​​the monitoring point are summed up to obtain the comprehensive automobile exhaust characteristics;

[0029] A monitoring environment set is formed by integrating automobile exhaust characteristics, atmospheric monitoring indicators and corresponding environmental conditions, and an atmospheric indicator prediction model is constructed based on the monitoring environment set; the environmental conditions include temperature, humidity, wind speed, air pressure, and solar radiation; the atmospheric indicator prediction model includes an input layer, a genetic variation hidden layer, an environmental feedback gating layer and an output layer;

[0030] The input layer receives the comprehensive automobile exhaust characteristics and environmental status through the exhaust characteristic channel and the environmental status channel respectively, and preprocesses the input data; the genetic variation hidden layer includes a BP neural network and a genetic variation strategy, the BP neural network is used to capture the dependency relationship between the comprehensive automobile exhaust characteristics and the atmospheric monitoring indicators, and the genetic variation strategy is used to perform multi-path evolution and directed variation; the environmental feedback gating layer determines the gating coefficient according to the environmental status, and corrects the output of the genetic variation hidden layer to output the gating prediction; the gating coefficient includes a first gating coefficient and a second gating coefficient, the first gating coefficient is obtained by modulating the temperature and humidity input sigmod activation function, and the second gating coefficient is a connection weight scaling, which is determined by the deviation of wind speed, air pressure and solar radiation from the corresponding standard values; the output layer receives the gating prediction output by the environmental feedback gating layer and predicts each atmospheric monitoring indicator respectively through a multi-task output head; the output layer mixes the MSE loss function, the MAE loss function and the Huber loss function to improve the prediction accuracy of the atmospheric monitoring indicators.

[0031] Furthermore, the method for selecting representative monitoring points includes:

[0032] Divide the area to be analyzed into multiple square grids, obtain the regional characteristics within each grid, and use the road intersections within each grid as pending monitoring points; the regional characteristics include road network density, population density, and intersection traffic volume;

[0033] When the road network density of the grid is less than the first road network density threshold, the pending monitoring point closest to the grid center is selected as the representative monitoring point. When the road network density of the grid is greater than the first road network density threshold, a grid search is performed on the pending monitoring points in the grid to determine two representative monitoring points. When the road network density of the grid is greater than the second road network density threshold, a grid search is performed on the pending monitoring points in the grid to determine three representative monitoring points.

[0034] Furthermore, the method for performing grid search includes:

[0035] The grid search objective function is determined based on the regional characteristics of all grids in the area to be analyzed and the distance between monitoring points:

[0036]

[0037] in To search for the objective function, is the regional feature weight, is the cost weight, is the set of representative monitoring points determined by grid search, is the set of all representative monitoring points, Representative monitoring points The set of 5 nearest monitoring points in the neighborhood, Representative monitoring points The collection of roads extending outwards, For monitoring points The number of representative monitoring points corresponding to the grid points, For monitoring points The number of monitoring points to be determined in the corresponding grid, For monitoring points Traffic flow at For monitoring points Neighborhood monitoring points of traffic volume, is the standard population density, For monitoring points is the population density within the area with a radius of 100m and the center of the circle. is the area of ​​a circular region with a radius of 100m, Representative monitoring points The road extending out The width, For monitoring points The road within the 100m radius area with the center of the circle length, is the distance cost of monitoring point layout, Representative monitoring points arrive distance;

[0038] All pending monitoring points that need to be searched for grid points are defined as a particle population. The corresponding positions of the nearest pending monitoring point and the farthest one or two pending monitoring points from the center point are selected as the optimal positions of the initial population. The particle population is divided into a core group A and an auxiliary group B, and the particles are encoded. The particles of the core group A are located within a 0.3km radius from the grid center and are used for global exploration. The particles of the auxiliary group B are located outside a 0.3km radius from the grid center and are used for local optimization.

[0039] Update the learning factor according to the particle population, and the expression is:

[0040]

[0041]

[0042] in population The individual learning factor, For population The group learning factor, , is the current iteration number, is the maximum number of iterations, For population The maximum individual learning factor, For population The median value of the individual learning factor, For population The minimum individual learning factor, For population The maximum value of the group learning factor, For population The median value of the group learning factor, For population The minimum value of the group learning factor;

[0043] Update the particle speed and position according to the learning factor. The expression is:

[0044]

[0045]

[0046] in for Population during iteration Neutral particles speed, for Population during iteration Neutral particles location, is the inertia weight, For population The corresponding position of the individual's fitness extreme value, is the position corresponding to the extreme value of the particle population fitness;

[0047] The search objective function is calculated based on the updated optimal position of the population to make a convergence judgment, and quantum annealing perturbation is performed based on the judgment result. The expression is:

[0048]

[0049]

[0050] in is the rotation angle, is the road network density within a 100m radius area with the monitoring point corresponding to the optimal population location as the center, is the maximum road network density, is the search objective function for calculating the historical optimal population position, For the The search objective function for iterative optimal population position calculation, is the annealing acceptance, For the Iteration temperature, is the annealing temperature attenuation coefficient;

[0051] The top 10 optimal solutions of each generation are retained, and the optimal solution is selected based on minimizing the search objective function and maximizing the coverage function, and crossover mutation operations are performed with the population; the coverage function is , is the standard deviation of the distances between all representative monitoring points; the crossover operation is specifically the exchange of two points; the mutation operation is specifically the Gaussian mutation;

[0052] Repeat the above operation and iterate continuously, transferring elite particles every 5 generations until the maximum number of iterations is reached or the search objective function decreases by less than 0.1% for 3 consecutive generations, and then stop the iteration to output the optimal representative monitoring point; the specific operation of transferring elite particles every 5 generations is to use the global optimal particles of core group A to update the particles of auxiliary group B, and use the local optimal particles of auxiliary group B to update the particles of core group A.

[0053] Furthermore, the method for determining the air pollution score and air pollution level of the area to be analyzed includes:

[0054] Input the traffic flow characteristics of the representative monitoring point into the vehicle knowledge graph to obtain the predicted second vehicle operating condition of the corresponding representative monitoring point, determine the first vehicle operating conditions of different vehicles corresponding to the representative monitoring point based on the vehicle statistical set of the representative monitoring point, and input the predicted second vehicle operating condition of the representative monitoring point and the first vehicle operating conditions of different vehicles into the exhaust prediction model to obtain the automobile exhaust characteristics of different vehicles;

[0055] Based on the exhaust characteristics of different vehicles at the representative monitoring points and the vehicle statistical set, the comprehensive exhaust characteristics of the representative monitoring points are calculated, and the comprehensive exhaust characteristics of each representative monitoring point are input into the atmospheric index prediction model to obtain the predicted atmospheric index of each representative monitoring point;

[0056] The monitoring pollution score of each representative monitoring point is calculated based on the predicted atmospheric indicators and the corresponding reference indicators. The expression is:

[0057]

[0058] in Representative monitoring points The pollution score, is the number of atmospheric indicators, is a set of atmospheric indicators, Atmospheric indicators The predicted value of Atmospheric indicators Reference value of

[0059] The density weight of the representative monitoring point is determined based on the product of the personnel density of the representative monitoring point and the road network density of the corresponding grid. The corresponding time weight is extracted based on the monitoring time. The air pollution score of the area to be analyzed is obtained by summing the pollution score, density weight and time weight of each representative monitoring point. The corresponding air pollution level is determined based on the air pollution score of the area to be analyzed.

[0060] The second aspect is the atmospheric pollution analysis system based on motor vehicle exhaust, including:

[0061] Exhaust gas module: used to determine the exhaust gas collection volume according to the vehicle operating conditions, and analyze the collected exhaust gas to obtain the vehicle exhaust characteristics;

[0062] Monitoring module: collects traffic monitoring images and atmospheric monitoring indicators, performs image processing to obtain traffic flow characteristics and vehicle statistics, and determines comprehensive vehicle exhaust characteristics based on the vehicle exhaust characteristics and the vehicle statistics;

[0063] Model module: used to construct an exhaust prediction model based on the vehicle operating condition and the automobile exhaust characteristics, and predict the automobile exhaust characteristics of each vehicle at the monitoring point; used to construct a vehicle knowledge graph based on the second vehicle operating condition and the traffic flow characteristics, and predict the vehicle operating condition of each vehicle at the monitoring point; used to construct an atmospheric index prediction model based on the comprehensive automobile exhaust characteristics, the atmospheric monitoring indicators and the corresponding environmental conditions, and predict the atmospheric indicators at the monitoring point;

[0064] Grid search module: used to determine the grid search objective function based on the regional characteristics of all grids in the area to be analyzed and the distance between monitoring points, and to perform grid search on the monitoring points to be determined within the grid to determine the representative monitoring points;

[0065] Scoring module: used to calculate the monitoring pollution score based on the predicted atmospheric indicators and the corresponding reference indicators, determine the density weight according to the regional characteristics, and determine the atmospheric pollution score and atmospheric pollution level of the area to be analyzed based on the monitoring pollution score and density weight of each representative monitoring point;

[0066] Intelligent supervision module: used to store, view and manage the predicted atmospheric indicators, the atmospheric pollution scores and the atmospheric pollution levels, and perform traffic control and atmospheric governance based on the predicted atmospheric indicators, the atmospheric pollution scores and the atmospheric pollution levels.

[0067] The beneficial effects of the present invention are:

[0068] The present invention is a method and system for analyzing atmospheric pollution based on motor vehicle exhaust. Compared with the prior art, the present invention has the following technical effects:

[0069] The present invention can improve the data preprocessing capabilities and enhance the adaptability of the model in atmospheric pollution analysis through image processing, grid search, exhaust gas prediction and atmospheric index prediction, calculation of monitoring pollution scores and model construction steps, thereby improving the efficiency and accuracy of atmospheric pollution analysis. The atmospheric pollution analysis technology is optimized, which can greatly save resources and improve work efficiency. It can realize the precision assessment of atmospheric pollution, provide a scientific basis for atmospheric pollution analysis, effectively improve the accuracy of exhaust gas prediction and atmospheric pollution analysis, and select representative monitoring points for monitoring layout according to the characteristics of the area to be analyzed. It can adapt to different atmospheric pollution analysis systems based on motor vehicle exhaust and the analysis needs of atmospheric pollution based on motor vehicle exhaust of different users, and has a certain universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 The present invention is a flowchart of the steps of the atmospheric pollution analysis method based on motor vehicle exhaust. DETAILED DESCRIPTION

[0071] The present invention will be further described below through specific examples. The illustrative examples and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0072] The present invention provides an atmospheric pollution analysis method and system based on vehicle exhaust, comprising the following steps:

[0073] like Figure 1 As shown, in this embodiment, the following steps are included:

[0074] Determining the exhaust gas collection amount according to the vehicle operating condition, collecting the exhaust gas, and analyzing the collected exhaust gas to obtain automobile exhaust characteristics; the vehicle operating condition includes a first vehicle operating condition and a second vehicle operating condition;

[0075] Building an exhaust prediction model based on the vehicle operating condition and the automobile exhaust characteristics, collecting traffic flow monitoring images and atmospheric monitoring indicators, performing image processing to obtain traffic flow characteristics and a vehicle statistics set, and building a vehicle knowledge graph based on the second vehicle operating condition and the traffic flow characteristics;

[0076] Determining comprehensive automobile exhaust characteristics based on the automobile exhaust characteristics and the vehicle statistical set, and constructing an atmospheric index prediction model based on the comprehensive automobile exhaust characteristics, the atmospheric monitoring index, and the corresponding environmental conditions;

[0077] Select representative monitoring points based on the regional characteristics of the area to be analyzed, input the traffic flow characteristics of the representative monitoring points into the vehicle knowledge graph to obtain a predicted second vehicle operating condition, obtain a predicted first vehicle operating condition based on the vehicle statistics set of the representative monitoring points, and obtain a predicted comprehensive automobile exhaust characteristic based on the exhaust prediction model;

[0078] The predicted comprehensive automobile exhaust characteristics are input into the atmospheric index prediction model to obtain the predicted atmospheric index, the monitoring pollution score is calculated based on the predicted atmospheric index and the corresponding reference index, the density weight is determined based on the regional characteristics, and the atmospheric pollution score and atmospheric pollution level of the area to be analyzed are determined based on the pollution score and density weight of each representative monitoring point.

[0079] In this embodiment, the method for obtaining automobile exhaust characteristics includes:

[0080] The first vehicle operating condition includes energy consumption type, brand and model; the brand and model correspond to engine parameters; the second vehicle operating condition is vehicle-mounted parameters, including vehicle speed, rotational speed and load, obtained through the vehicle-mounted OBD system;

[0081] Determine a baseline exhaust volume according to a first vehicle operating condition, measure actual exhaust volumes under different second vehicle operating conditions, determine the relationship between the actual exhaust volume, the second vehicle operating condition, and the baseline exhaust volume by fitting, determine the exhaust gas collection volume according to the actual exhaust volume, construct a flow control unit according to the exhaust gas collection volume, the first vehicle operating condition, and the second vehicle operating condition, and connect the exhaust gas collection device to the flow control unit and the on-board OBD system to perform variable-flow exhaust gas collection; the baseline exhaust volume is a factory-calibrated value associated with the first vehicle operating condition; the actual exhaust volume is in a fixed ratio to the exhaust gas collection volume; the flow control unit is used to connect to the on-board OBD system to determine the vehicle operating condition, and determine the exhaust gas collection volume according to the vehicle operating condition.

[0082] High-resolution infrared spectroscopy and miniaturized proton transfer reaction mass spectrometry are used to analyze the collected exhaust gas to obtain automobile exhaust characteristics; the automobile exhaust characteristics include conventional exhaust components, oxygen-containing organic compound components, polycyclic aromatic hydrocarbon components, and volatile organic compound components;

[0083] In the actual evaluation, the first vehicle operating condition of a certain car is taken as an example. The brand model is JLDH-GL 1.5T, the energy consumption types include: displacement 1.5L, fuel type 92#, rated speed 2000rpm, and unloaded mass 1350kg. The corresponding benchmark exhaust volume is 8.72 m³ / min measured by bench test at the factory according to GB 18352.6-2016 standard.

[0084] The actual exhaust volume was acquired in real time through the OBD system under 27 operating condition combinations, including {30, 60, 90} (km / h), {1500, 3000, 4500} (rpm), and {no load, half load, full load}. The relationship between the actual exhaust volume, the second vehicle operating condition, and the baseline exhaust volume was determined using least squares fitting. Based on the sampling flow requirements of ISO 16183:2002 and ensuring that the sample gas residence time at the sensor was ≥0.5 seconds, the exhaust gas collection volume / actual exhaust volume was determined to be 1 / 200. A flow control unit was constructed based on the exhaust gas collection volume, the first vehicle operating condition, and the second vehicle operating condition. The flow control unit collected the second vehicle operating condition from the onboard OBD system and calculated the exhaust gas collection volume based on the first vehicle operating condition to control the collection speed of the exhaust gas collection device. For example, for a JLDH-GL1.5T vehicle under the conditions of 90 km / h, 4000 rpm, and full load, the actual exhaust volume was determined to be 13.18 m³ / min, corresponding to an exhaust gas collection volume of 65.6 L / min.

[0085] High-resolution infrared spectroscopy is used to scan the characteristic absorption spectra of various gaseous pollutants in motor vehicle exhaust to determine the components of conventional exhaust (CO, CO2, HC, NO, NO2, NO3), oxygen-containing organic compounds and polycyclic aromatic hydrocarbons;

[0086] Miniaturized proton transfer reaction mass spectrometry was used to determine the volatile organic compound (VOCs) components.

[0087] In this embodiment, the method for constructing a vehicle knowledge graph includes:

[0088] Different vehicle operating conditions and corresponding automobile exhaust characteristics are combined into a vehicle operating condition exhaust set, and the vehicle operating condition exhaust set is used to build the exhaust prediction model performance;

[0089] Randomly selecting monitoring points to collect traffic monitoring images and atmospheric monitoring indicators; the atmospheric monitoring indicators are related to air pollution, including conventional pollutants, characteristic derivatives, environmental indicators and meteorological elements;

[0090] Preprocessing and vehicle detection and tracking are performed on traffic flow monitoring images to assign a unique ID to each detected vehicle. Image processing is performed on each vehicle image to obtain traffic flow characteristics. The traffic flow characteristics include traffic volume, speed characteristics, spatial characteristics, and traffic flow composition. The speed characteristics include average speed, speed distribution, and speed stability. The spatial characteristics include lane occupancy and vehicle spacing. The traffic flow composition includes vehicle type distribution and vehicle attributes.

[0091] Each detected vehicle is input into the ResNet-18 convolutional neural network for fine-grained vehicle classification, and a vehicle statistics set is obtained based on the vehicle fine-grained classification results; the vehicle statistics set includes vehicle brand and model, vehicle age, and corresponding quantity;

[0092] Constructing a vehicle knowledge graph based on the second vehicle operating condition and traffic flow characteristics. Specific steps include: creating core nodes based on vehicle type, road type, and pollution type; applying a three-step weighting algorithm to calculate edge weights between two nodes and automatically filtering weakly connected edges with weights < 0.2; and setting real-time weights and graph reconstruction frequency.

[0093] The three-step weight algorithm includes spatiotemporal correlation weight calculation, pollution contribution weight calculation, road network topology weight calculation and weight fusion, and the expression is:

[0094]

[0095]

[0096]

[0097] in Contribution weight for pollution, is a set of vehicle types, is the number of vehicle models, is the emission factor of the corresponding vehicle type, is the vehicle dwell time, Monitoring area, is the road network topology weight, is the road grade, is the angle between the road direction and the wind direction, is the maximum impact distance, is the vertical distance from the monitoring point to the road, is the wind speed correction factor, is the wind speed, is the building density index, is the altitude correction factor, is the average building height, is the edge weight determined by the three-step weight algorithm, is the spatiotemporal correlation weight, which is determined based on the time series similarity of the DTW algorithm. 、 、 is the weight adjustment parameter;

[0098] In the actual evaluation, the exhaust prediction model includes an input layer, a gradient boosting decision tree layer, and an output layer. The gradient boosting decision tree layer includes a decision tree group, multiple output heads, and Huber loss. The decision tree group determines the decision rules based on the vehicle operating conditions and automobile exhaust characteristics, the multiple output heads perform simultaneous predictions of multiple automobile exhaust characteristics, and the Huber loss is used to resist outlier interference.

[0099] Atmospheric monitoring indicators specifically include conventional pollutants (determination of PM2.5 / PM10 mass concentration by β-ray absorption method, and determination of NO2 / O3 ground concentration by ultraviolet differential absorption), characteristic derivatives (determination of peroxyacetyl nitrate by gas chromatography, and determination of secondary organic aerosols by aerosol mass spectrometry), environmental indicators (measurement of atmospheric oxidizing properties by free radicals, and determination of extinction coefficient by turbidimeter), and meteorological elements (determination of boundary layer height by lidar, and determination of inversion intensity by temperature profile radar).

[0100] Preprocess the traffic monitoring images (illumination compensation, dehazing, and perspective transformation correction) to obtain optimized traffic monitoring images. Use the YOLOv5 target detection algorithm to locate vehicles in each frame, and use the DeepSORT multi-target tracking algorithm to establish vehicle tracks. Each detected vehicle is assigned a unique ID.

[0101] Image processing for each vehicle image includes: extracting features related to vehicle quantity using inter-frame difference method and connected domain analysis, extracting features related to vehicle speed using optical flow method and distance calibration, extracting features related to distance using lane segmentation and nearest neighbor analysis, and extracting features related to area using semantic segmentation and pixel statistics;

[0102] The specific steps of constructing the vehicle knowledge graph include: creating core nodes based on vehicle type (such as diesel trucks, electric cars, etc.), road type (national highways, urban roads, etc.) and pollution type (such as high NOx concentration areas), applying a three-step weighting algorithm (weight adjustment parameters 、 、 Take 0.3, 0.3, 0.4 respectively, and the road grade According to the highway / urban expressway / main road / secondary road / branch road / street, the maximum impact distance is 0.9 / 0.8 / 0.7 / 0.5 / 0.3 / 0.1 respectively, and the intersection is 1.2. , wind speed correction factor , altitude correction factor ) Calculate the edge weight between two nodes and automatically filter out weakly associated edges with weight < 0.2, set the real-time weight to once per hour and the full graph reconstruction frequency to once every 3 days.

[0103] In this embodiment, the method for constructing an atmospheric index prediction model includes:

[0104] According to the automobile exhaust characteristics and the vehicle statistics set, the exhaust emissions of all automobiles in the monitoring area of ​​the monitoring point are summed up to obtain the comprehensive automobile exhaust characteristics;

[0105] A monitoring environment set is formed by integrating automobile exhaust characteristics, atmospheric monitoring indicators and corresponding environmental conditions, and an atmospheric indicator prediction model is constructed based on the monitoring environment set; the environmental conditions include temperature, humidity, wind speed, air pressure, and solar radiation; the atmospheric indicator prediction model includes an input layer, a genetic variation hidden layer, an environmental feedback gating layer and an output layer;

[0106] The input layer receives the comprehensive automobile exhaust characteristics and environmental status through the exhaust characteristic channel and the environmental status channel respectively, and preprocesses the input data; the genetic variation hidden layer includes a BP neural network and a genetic variation strategy, the BP neural network is used to capture the dependency relationship between the comprehensive automobile exhaust characteristics and the atmospheric monitoring indicators, and the genetic variation strategy is used to perform multi-path evolution and directed variation; the environmental feedback gating layer determines the gating coefficient according to the environmental status, and corrects the output of the genetic variation hidden layer to output the gating prediction; the gating coefficient includes a first gating coefficient and a second gating coefficient, the first gating coefficient is obtained by modulating the temperature and humidity input sigmoid activation function, and the second gating coefficient is a connection weight scaling, which is determined by the deviation of wind speed, air pressure and solar radiation from the corresponding standard values; the output layer receives the gating prediction output by the environmental feedback gating layer and predicts each atmospheric monitoring indicator respectively through a multi-task output head; the output layer hybrid MSE loss function, MAE loss function and Huber loss function improves the prediction accuracy of atmospheric monitoring indicators;

[0107] In the actual evaluation, the exhaust characteristic channel in the input layer normalizes the exhaust characteristics of the vehicle, and the environmental state channel performs Gaussian standardization on the environmental state; a 128-node fully connected BP neural network is used in the genetic variation hidden layer, and the genetic variation strategy specifically includes: population initialization (parallel training of 10 groups of weight matrices), fitness evaluation (calculation of validation set MAE after each round of training), elite selection (retaining the best 3 groups of weights and eliminating 70% after elimination), chromosome crossover (segmented recombination of the optimal weight matrix), directed mutation (applying Gaussian perturbation to low fitness areas), and every 50 epo ch performs a complete genetic iteration; in the environmental feedback gating layer, the weight of the VOCs conversion path is enhanced when the temperature is high (greater than 30°C), the particulate matter generation path is suppressed when the humidity is high (greater than 80%), and the pollutant diffusion channel is opened when the wind is strong (greater than 5m / s); in the output layer, the MSE loss function is used to evaluate the difference between the predicted value and the true value of conventional pollutants, the MAE loss function is used to evaluate the difference between the predicted value and the true value of feature derivatives, and the Huber loss function is used to evaluate the difference between the predicted value and the true value of environmental indicators. The corresponding weights are 1, 0.7, and 0.5, respectively.

[0108] In this embodiment, the method for selecting representative monitoring points includes:

[0109] Divide the area to be analyzed into multiple square grids, obtain the regional characteristics within each grid, and use the road intersections within each grid as pending monitoring points; the regional characteristics include road network density, population density, and intersection traffic volume;

[0110] When the road network density of the grid is less than the first road network density threshold, the pending monitoring point closest to the grid center is selected as the representative monitoring point. When the road network density of the grid is greater than the first road network density threshold, a grid search is performed on the pending monitoring points in the grid to determine two representative monitoring points. When the road network density of the grid is greater than the second road network density threshold, a grid search is performed on the pending monitoring points in the grid to determine three representative monitoring points.

[0111] In this embodiment, the method for performing grid search includes:

[0112] The grid search objective function is determined based on the regional characteristics of all grids in the area to be analyzed and the distance between monitoring points:

[0113]

[0114] in To search for the objective function, is the regional feature weight, is the cost weight, is the set of representative monitoring points determined by grid search, is the set of all representative monitoring points, Representative monitoring points The set of 5 nearest monitoring points in the neighborhood, Representative monitoring points The collection of roads extending outwards, For monitoring points The number of representative monitoring points corresponding to the grid points, For monitoring points The number of monitoring points to be determined in the corresponding grid, For monitoring points Traffic flow at For monitoring points Neighborhood monitoring points of traffic volume, is the standard population density, For monitoring points is the population density within the area with a radius of 100m and the center of the circle. is the area of ​​a circular region with a radius of 100m, Representative monitoring points The road extending out The width, For monitoring points The road within the 100m radius area with the center of the circle length, is the distance cost of monitoring point layout, Representative monitoring points arrive distance;

[0115] All pending monitoring points that need to be searched for grid points are defined as a particle population. The corresponding positions of the nearest pending monitoring point and the farthest one or two pending monitoring points from the center point are selected as the optimal positions of the initial population. The particle population is divided into a core group A and an auxiliary group B, and the particles are encoded. The particles of the core group A are located within a 0.3km radius from the grid center and are used for global exploration. The particles of the auxiliary group B are located outside a 0.3km radius from the grid center and are used for local optimization.

[0116] Update the learning factor according to the particle population, and the expression is:

[0117]

[0118]

[0119] in population The individual learning factor, For population The group learning factor, , is the current iteration number, is the maximum number of iterations, For population The maximum individual learning factor, For population The median value of the individual learning factor, For population The minimum individual learning factor, For population The maximum value of the group learning factor, For population The median value of the group learning factor, For population The minimum value of the group learning factor;

[0120] Update the particle speed and position according to the learning factor. The expression is:

[0121]

[0122]

[0123] in for Population during iteration Neutral particles speed, for Population during iteration Neutral particles location, is the inertia weight, For population The corresponding position of the individual's fitness extreme value, is the position corresponding to the extreme value of the particle population fitness;

[0124] The search objective function is calculated based on the updated optimal position of the population to make a convergence judgment, and quantum annealing perturbation is performed based on the judgment result. The expression is:

[0125]

[0126]

[0127] in is the rotation angle, is the road network density within a 100m radius area with the monitoring point corresponding to the optimal population location as the center, is the maximum road network density, is the search objective function for calculating the historical optimal population position, For the The search objective function for iterative optimal population position calculation, is the annealing acceptance, For the Iteration temperature, is the annealing temperature attenuation coefficient;

[0128] The top 10 optimal solutions of each generation are retained, and the optimal solution is selected based on minimizing the search objective function and maximizing the coverage function, and crossover mutation operations are performed with the population; the coverage function is , is the standard deviation of the distances between all representative monitoring points; the crossover operation is specifically the exchange of two points; the mutation operation is specifically the Gaussian mutation;

[0129] Repeat the above operation and iterate and transfer elite particles every 5 generations until the maximum number of iterations is reached or the search objective function decreases by less than 0.1% for 3 consecutive generations, then stop the iteration and output the optimal representative monitoring point; the specific operation of transferring elite particles every 5 generations is to use the global optimal particles of core group A to update the particles of auxiliary group B, and use the local optimal particles of auxiliary group B to update the particles of core group A;

[0130] In the actual evaluation, the area to be analyzed is divided into multiple 1km×1km square grids, and the particle population (a total of 110 pending monitoring points) is divided into core group A (40) and auxiliary group B (70). , cost weight , inertia weight , maximum number of iterations , annealing temperature attenuation coefficient Perform a grid search to obtain representative monitoring points.

[0131] In this embodiment, the method for determining the air pollution score and air pollution level of the area to be analyzed includes:

[0132] Input the traffic flow characteristics of the representative monitoring point into the vehicle knowledge graph to obtain the predicted second vehicle operating condition of the corresponding representative monitoring point, determine the first vehicle operating conditions of different vehicles corresponding to the representative monitoring point based on the vehicle statistical set of the representative monitoring point, and input the predicted second vehicle operating condition of the representative monitoring point and the first vehicle operating conditions of different vehicles into the exhaust prediction model to obtain the automobile exhaust characteristics of different vehicles;

[0133] Based on the exhaust characteristics of different vehicles at the representative monitoring points and the vehicle statistical set, the comprehensive exhaust characteristics of the representative monitoring points are calculated, and the comprehensive exhaust characteristics of each representative monitoring point are input into the atmospheric index prediction model to obtain the predicted atmospheric index of each representative monitoring point;

[0134] The monitoring pollution score of each representative monitoring point is calculated based on the predicted atmospheric indicators and the corresponding reference indicators. The expression is:

[0135]

[0136] in Representative monitoring points The pollution score, is the number of atmospheric indicators, is a set of atmospheric indicators, Atmospheric indicators The predicted value of Atmospheric indicators Reference value of

[0137] The density weight of the representative monitoring point is determined by multiplying the personnel density of the representative monitoring point by the road network density of the corresponding grid. The corresponding time weight is extracted according to the monitoring time. The air pollution score of the area to be analyzed is obtained by summing the pollution score, density weight and time weight of each representative monitoring point. The corresponding air pollution level of the area to be analyzed is determined based on the air pollution score of the area to be analyzed.

[0138] In the actual evaluation, the traffic flow characteristics of the representative monitoring point are input into the vehicle knowledge graph to obtain the predicted second vehicle operating condition of the corresponding representative monitoring point. The first vehicle operating condition of different vehicles corresponding to the representative monitoring point is determined based on the vehicle statistics set of the representative monitoring point. The predicted second vehicle operating condition of the representative monitoring point and the first vehicle operating conditions of different vehicles are input into the exhaust prediction model to obtain the exhaust characteristics of different vehicles.

[0139] The correlation between monitoring time and time weight is as follows: morning peak (7:00-9:00) / 1.3, daytime (9:00-17:00) / 1, evening peak (17:00-20:00) / 1.2, night active period (20:00-23:00) / 0.9, late night quiet period (23:00-7:00) / 0.7;

[0140] The corresponding relationship between the air pollution score and the air pollution level is: excellent (level 1) / 0-30, good (level 2) / 31-60, light pollution (level 3) / 61-75, moderate pollution (level 4) / (75-90), and heavy pollution (level 5) / (greater than 90).

[0141] The second aspect is the atmospheric pollution analysis system based on motor vehicle exhaust, including:

[0142] Exhaust gas module: used to determine the exhaust gas collection volume according to the vehicle operating conditions, and analyze the collected exhaust gas to obtain the vehicle exhaust characteristics;

[0143] Monitoring module: collects traffic monitoring images and atmospheric monitoring indicators, performs image processing to obtain traffic flow characteristics and vehicle statistics, and determines comprehensive vehicle exhaust characteristics based on the vehicle exhaust characteristics and the vehicle statistics;

[0144] Model module: used to construct an exhaust prediction model based on the vehicle operating condition and the automobile exhaust characteristics, and predict the automobile exhaust characteristics of each vehicle at the monitoring point; used to construct a vehicle knowledge graph based on the second vehicle operating condition and the traffic flow characteristics, and predict the vehicle operating condition of each vehicle at the monitoring point; used to construct an atmospheric index prediction model based on the comprehensive automobile exhaust characteristics, the atmospheric monitoring indicators and the corresponding environmental conditions, and predict the atmospheric indicators at the monitoring point;

[0145] Grid search module: used to determine the grid search objective function based on the regional characteristics of all grids in the area to be analyzed and the distance between monitoring points, and to perform grid search on the monitoring points to be determined within the grid to determine the representative monitoring points;

[0146] Scoring module: used to calculate the monitoring pollution score based on the predicted atmospheric indicators and the corresponding reference indicators, determine the density weight according to the regional characteristics, and determine the atmospheric pollution score and atmospheric pollution level of the area to be analyzed based on the monitoring pollution score and density weight of each representative monitoring point;

[0147] Intelligent supervision module: used to store, view and manage the predicted atmospheric indicators, the atmospheric pollution scores and the atmospheric pollution levels, and perform traffic control and atmospheric governance based on the predicted atmospheric indicators, the atmospheric pollution scores and the atmospheric pollution levels.

[0148] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for analyzing atmospheric pollution based on motor vehicle exhaust, characterized in that: The following steps are involved: S1. Determine an exhaust gas collection amount based on a vehicle operating condition, collect exhaust gas, and analyze the collected exhaust gas to obtain vehicle exhaust gas characteristics; the vehicle operating condition includes a first vehicle operating condition and a second vehicle operating condition; the first vehicle operating condition includes energy consumption type and brand model; the brand model corresponds to engine parameters; the second vehicle operating condition is vehicle-mounted parameters, including vehicle speed, rotational speed, and load, obtained through an onboard OBD system; S2. Build an exhaust gas prediction model based on the vehicle operating condition and the automobile exhaust characteristics, collect traffic flow monitoring images and atmospheric monitoring indicators, perform image processing to obtain traffic flow characteristics and a vehicle statistics set, and build a vehicle knowledge graph based on the second vehicle operating condition and the traffic flow characteristics; S3. Determine comprehensive automobile exhaust characteristics based on the automobile exhaust characteristics and the vehicle statistics set, and construct an atmospheric index prediction model based on the comprehensive automobile exhaust characteristics, the atmospheric monitoring indicators, and the corresponding environmental conditions; S4. Select representative monitoring points based on regional characteristics of the area to be analyzed, input the traffic flow characteristics of the representative monitoring points into the vehicle knowledge graph to obtain a predicted second vehicle operating condition, obtain a predicted first vehicle operating condition based on the vehicle statistics set of the representative monitoring points, and obtain a predicted comprehensive automobile exhaust characteristic based on the exhaust prediction model; S5. Inputting the predicted comprehensive automobile exhaust characteristics into the atmospheric index prediction model to obtain predicted atmospheric indexes, calculating a monitoring pollution score based on the predicted atmospheric indexes and corresponding reference indexes, determining a density weight based on the regional characteristics, and determining an atmospheric pollution score and an atmospheric pollution level for the area to be analyzed based on the pollution scores and density weights of each representative monitoring point; The method for selecting representative monitoring points includes: Divide the area to be analyzed into multiple square grids, obtain the regional characteristics within each grid, and use the road intersections within each grid as pending monitoring points; the regional characteristics include road network density, population density, and intersection traffic volume; When the road network density of the grid is less than the first road network density threshold, the pending monitoring point closest to the grid center is selected as the representative monitoring point; when the road network density of the grid is greater than the first road network density threshold, a grid search is performed on the pending monitoring point within the grid to determine two representative monitoring points; when the road network density of the grid is greater than the second road network density threshold, a grid search is performed on the pending monitoring point within the grid to determine three representative monitoring points; The method for performing grid point search comprises: The grid search objective function is determined based on the regional characteristics of all grids in the area to be analyzed and the distance between monitoring points: in To search for the objective function, is the regional feature weight, is the cost weight, is the set of representative monitoring points determined by grid search, is the set of all representative monitoring points, Representative monitoring points The set of 5 nearest monitoring points in the neighborhood, Representative monitoring points The collection of roads extending outwards, For monitoring points The number of representative monitoring points corresponding to the grid points, For monitoring points The number of monitoring points to be determined in the corresponding grid, For monitoring points Traffic flow at For monitoring points Neighborhood monitoring points of traffic volume, is the standard population density, For monitoring points is the population density within the area with a radius of 100m and the center of the circle. is the area of ​​a circular region with a radius of 100m, Representative monitoring points The road extending out The width, For monitoring points The road within the 100m radius area with the center of the circle length, is the distance cost of monitoring point layout, Representative monitoring points arrive distance; All pending monitoring points that need to be searched for grid points are defined as a particle population. The corresponding positions of the nearest pending monitoring point and the farthest one or two pending monitoring points from the center point are selected as the optimal positions of the initial population. The particle population is divided into a core group A and an auxiliary group B, and the particles are encoded. The particles of the core group A are located within a 0.3km radius from the grid center and are used for global exploration. The particles of the auxiliary group B are located outside a 0.3km radius from the grid center and are used for local optimization. Update the learning factor according to the particle population, and the expression is: in population The individual learning factor, For population The group learning factor, , is the current iteration number, is the maximum number of iterations, For population The maximum individual learning factor, For population The median value of the individual learning factor, For population The minimum individual learning factor, For population The maximum value of the group learning factor, For population The median value of the group learning factor, For population The minimum value of the group learning factor; Update the particle speed and position according to the learning factor. The expression is: in for Population during iteration Neutral particles speed, for Population during iteration Neutral particles location, is the inertia weight, For population The corresponding position of the individual's fitness extreme value, is the position corresponding to the extreme value of the particle population fitness; The search objective function is calculated based on the updated optimal position of the population to make a convergence judgment, and quantum annealing perturbation is performed based on the judgment result. The expression is: in is the rotation angle, is the road network density within a 100m radius area with the monitoring point corresponding to the optimal population location as the center, is the maximum road network density, is the search objective function for calculating the historical optimal population position, For the The search objective function for iterative optimal population position calculation, is the annealing acceptance, For the Iteration temperature, is the annealing temperature attenuation coefficient; The top 10 optimal solutions of each generation are retained, and the optimal solution is selected based on minimizing the search objective function and maximizing the coverage function, and crossover mutation operations are performed with the population; the coverage function is , is the standard deviation of the distances between all representative monitoring points; the crossover operation is specifically the exchange of two points; the mutation operation is specifically the Gaussian mutation; Repeat the above operation and iterate continuously, transferring elite particles every 5 generations until the maximum number of iterations is reached or the search objective function decreases by less than 0.1% for 3 consecutive generations, and then stop the iteration to output the optimal representative monitoring point; the specific operation of transferring elite particles every 5 generations is to use the global optimal particles of core group A to update the particles of auxiliary group B, and use the local optimal particles of auxiliary group B to update the particles of core group A.

2. The atmospheric pollution analysis method based on motor vehicle exhaust according to claim 1, characterized in that: The method for obtaining automobile exhaust characteristics comprises: Determine a baseline exhaust volume according to a first vehicle operating condition, measure actual exhaust volumes under different second vehicle operating conditions, determine the relationship between the actual exhaust volume, the second vehicle operating condition, and the baseline exhaust volume by fitting, determine the exhaust gas collection volume according to the actual exhaust volume, construct a flow control unit according to the exhaust gas collection volume, the first vehicle operating condition, and the second vehicle operating condition, and connect the exhaust gas collection device to the flow control unit and the on-board OBD system to perform variable-flow exhaust gas collection; the baseline exhaust volume is a factory-calibrated value associated with the first vehicle operating condition; the actual exhaust volume is in a fixed ratio to the exhaust gas collection volume; the flow control unit is used to connect to the on-board OBD system to determine the vehicle operating condition, and determine the exhaust gas collection volume according to the vehicle operating condition. The collected exhaust gas is analyzed using high-resolution infrared spectroscopy technology and miniaturized proton transfer reaction mass spectrometry technology to obtain automobile exhaust characteristics; the automobile exhaust characteristics include conventional exhaust components, oxygen-containing organic compound components, polycyclic aromatic hydrocarbon components and volatile organic compound components.

3. The atmospheric pollution analysis method based on motor vehicle exhaust according to claim 1, characterized in that: The method for constructing a vehicle knowledge graph includes: Different vehicle operating conditions and corresponding automobile exhaust characteristics are combined into a vehicle operating condition exhaust set, and the vehicle operating condition exhaust set is used to build the exhaust prediction model performance; Randomly selecting monitoring points to collect traffic monitoring images and atmospheric monitoring indicators; the atmospheric monitoring indicators are related to air pollution, including conventional pollutants, characteristic derivatives, environmental indicators and meteorological elements; Preprocessing and vehicle detection and tracking are performed on traffic flow monitoring images to assign a unique ID to each detected vehicle. Image processing is performed on each vehicle image to obtain traffic flow characteristics. The traffic flow characteristics include traffic volume, speed characteristics, spatial characteristics, and traffic flow composition. The speed characteristics include average speed, speed distribution, and speed stability. The spatial characteristics include lane occupancy and vehicle spacing. The traffic flow composition includes vehicle type distribution and vehicle attributes. Each detected vehicle is input into the ResNet-18 convolutional neural network for fine-grained vehicle classification, and a vehicle statistics set is obtained based on the vehicle fine-grained classification results; the vehicle statistics set includes vehicle brand and model, vehicle age, and corresponding quantity; Constructing a vehicle knowledge graph based on the second vehicle operating condition and traffic flow characteristics. Specific steps include: creating core nodes based on vehicle type, road type, and pollution type; applying a three-step weighting algorithm to calculate edge weights between two nodes and automatically filtering weakly connected edges with weights < 0.2; and setting real-time weights and graph reconstruction frequency. The three-step weight algorithm includes spatiotemporal correlation weight calculation, pollution contribution weight calculation, road network topology weight calculation and weight fusion, and the expression is: in Contribution weight for pollution, is a set of vehicle types, is the number of vehicle models, is the emission factor of the corresponding vehicle type, is the vehicle dwell time, Monitoring area, is the road network topology weight, is the road grade, is the angle between the road direction and the wind direction, is the maximum impact distance, is the vertical distance from the monitoring point to the road, is the wind speed correction factor, is the wind speed, is the building density index, is the altitude correction factor, is the average building height, is the edge weight determined by the three-step weight algorithm, is the spatiotemporal correlation weight, which is determined based on the time series similarity of the DTW algorithm. 、 、 is the weight adjustment parameter.

4. The atmospheric pollution analysis method based on motor vehicle exhaust according to claim 1, characterized in that: The method for constructing an atmospheric index prediction model comprises: According to the automobile exhaust characteristics and the vehicle statistics set, the exhaust emissions of all automobiles in the monitoring area of ​​the monitoring point are summed up to obtain the comprehensive automobile exhaust characteristics; A monitoring environment set is formed by integrating automobile exhaust characteristics, atmospheric monitoring indicators and corresponding environmental conditions, and an atmospheric indicator prediction model is constructed based on the monitoring environment set; the environmental conditions include temperature, humidity, wind speed, air pressure, and solar radiation; the atmospheric indicator prediction model includes an input layer, a genetic variation hidden layer, an environmental feedback gating layer and an output layer; The input layer receives the comprehensive automobile exhaust characteristics and environmental status through the exhaust characteristic channel and the environmental status channel respectively, and preprocesses the input data; the genetic variation hidden layer includes a BP neural network and a genetic variation strategy, the BP neural network is used to capture the dependency relationship between the comprehensive automobile exhaust characteristics and the atmospheric monitoring indicators, and the genetic variation strategy is used to perform multi-path evolution and directed variation; the environmental feedback gating layer determines the gating coefficient according to the environmental status, and corrects the output of the genetic variation hidden layer to output the gating prediction; the gating coefficient includes a first gating coefficient and a second gating coefficient, the first gating coefficient is obtained by modulating the temperature and humidity input sigmod activation function, and the second gating coefficient is a connection weight scaling, which is determined by the deviation of wind speed, air pressure and solar radiation from the corresponding standard values; the output layer receives the gating prediction output by the environmental feedback gating layer and predicts each atmospheric monitoring indicator respectively through a multi-task output head; the output layer mixes the MSE loss function, the MAE loss function and the Huber loss function to improve the prediction accuracy of the atmospheric monitoring indicators.

5. The atmospheric pollution analysis method based on motor vehicle exhaust according to claim 1, characterized in that: The method for determining the air pollution score and air pollution level of the area to be analyzed includes: Based on the exhaust characteristics of different vehicles at the representative monitoring points and the vehicle statistical set, the comprehensive exhaust characteristics of the representative monitoring points are calculated, and the comprehensive exhaust characteristics of each representative monitoring point are input into the atmospheric index prediction model to obtain the predicted atmospheric index of each representative monitoring point; The monitoring pollution score of each representative monitoring point is calculated based on the predicted atmospheric indicators and the corresponding reference indicators. The expression is: in Representative monitoring points The pollution score, is the number of atmospheric indicators, is a set of atmospheric indicators, Atmospheric indicators The predicted value of Atmospheric indicators Reference value of The density weight of the representative monitoring point is determined based on the product of the personnel density of the representative monitoring point and the road network density of the corresponding grid. The corresponding time weight is extracted based on the monitoring time. The air pollution score of the area to be analyzed is obtained by summing the pollution score, density weight and time weight of each representative monitoring point. The corresponding air pollution level is determined based on the air pollution score of the area to be analyzed.

6. The atmospheric pollution analysis system based on motor vehicle exhaust is characterized by: Used to perform the method according to any one of claims 1 to 5, comprising: Exhaust gas module: used to determine the exhaust gas collection volume according to the vehicle operating conditions, and analyze the collected exhaust gas to obtain the vehicle exhaust characteristics; Monitoring module: collects traffic monitoring images and atmospheric monitoring indicators, performs image processing to obtain traffic flow characteristics and vehicle statistics, and determines comprehensive vehicle exhaust characteristics based on the vehicle exhaust characteristics and the vehicle statistics; Model module: used to construct an exhaust prediction model based on the vehicle operating condition and the automobile exhaust characteristics, and predict the automobile exhaust characteristics of each vehicle at the monitoring point; used to construct a vehicle knowledge graph based on the second vehicle operating condition and the traffic flow characteristics, and predict the vehicle operating condition of each vehicle at the monitoring point; used to construct an atmospheric index prediction model based on the comprehensive automobile exhaust characteristics, the atmospheric monitoring indicators and the corresponding environmental conditions, and predict the atmospheric indicators at the monitoring point; Grid search module: used to determine the grid search objective function based on the regional characteristics of all grids in the area to be analyzed and the distance between monitoring points, and to perform grid search on the monitoring points to be determined within the grid to determine the representative monitoring points; Scoring module: used to calculate the monitoring pollution score based on the predicted atmospheric indicators and the corresponding reference indicators, determine the density weight according to the regional characteristics, and determine the atmospheric pollution score and atmospheric pollution level of the area to be analyzed based on the monitoring pollution score and density weight of each representative monitoring point; Intelligent supervision module: used to store, view and manage the predicted atmospheric indicators, the atmospheric pollution scores and the atmospheric pollution levels, and perform traffic control and atmospheric governance based on the predicted atmospheric indicators, the atmospheric pollution scores and the atmospheric pollution levels.

Images