A traffic pollutant spatial distribution inference method based on multi-source data

Abstract

The application discloses a traffic pollutant spatial distribution inference method based on multi-source data, which comprises the following steps: collecting hourly pollutant data of a research area, and dividing different time periods to calculate average values as explained variables; extracting potential explained variables, including geographical position, meteorological elements, land use, road density, POI (Point of Interest) data and other pollutants; constructing an LUR model by using a supervised forward stepwise regression algorithm to screen important variables; constructing an xgboost machine learning model to establish a machine learning regression model between pollutant concentration and explained variables; dividing the research area into regular grids by using GIS, extracting the explained variables corresponding to the grid centroid points, and inputting the machine learning model to predict the grid pollutant concentration; and investigating the pollutant concentration of different traffic micro-environments in small areas, so that the high-precision distribution of traffic pollutants can be inferred, scientific basis can be provided for environmental governance, and visual suggestions can be provided for people's travel.

Description

Technical Field

[0001] This invention relates to the field of transportation, and particularly to a method for inferring the spatial distribution of traffic pollutants based on multi-source data. Background Technology

[0002] Methods for constructing the spatiotemporal distribution of pollutants include exposure index variables, interpolation methods, discrete models, and land use regression (LUR) models. Interpolation methods and discrete models are purely mechanical and heavily reliant on dense monitoring networks. LUR models, by establishing multiple linear regression, consider factors such as surrounding population, land use, traffic, and meteorology, selecting significant variables, and exploring the multiple linear relationship between measured air pollutant concentrations and these dependent variables. In recent years, LUR studies have mostly focused on annual or quarterly average concentrations. While offering high spatial resolution, they have low temporal resolution and are only suitable for studying long-term pollution exposure characteristics.

[0003] LUR models fail to capture the nonlinear relationships between pollutants and various dependent variables, often resulting in poor model fit. To address this issue, machine learning algorithms such as Artificial Neural Networks (ANNs), Random Forests, and XGBoost have been used to predict spatial and temporal patterns of air pollutants because they do not employ a functional form of the relationship between pollutant data and predictor variables.

[0004] However, due to factors such as the altitude of the monitoring stations and wind speed, the station detection results differ somewhat from the concentrations of traffic source pollutants in actual traffic environments. For example, there are significant differences in pollutant concentrations in driving, walking, and public transportation environments. Therefore, adjustments need to be made based on the model's prediction results for pollutant concentrations in different traffic environments.

[0005] To address the issues that current models cannot accurately account for the nonlinear relationships between variables and the differences in pollutant concentrations in the traffic microenvironment, there is an urgent need to propose effective solutions. Therefore, this invention provides a method for inferring the spatial distribution of traffic pollutants based on multi-source data. Summary of the Invention

[0006] This invention provides a method for inferring the spatial distribution of traffic pollutants based on multi-source data. It comprehensively utilizes multiple data sources and advanced data processing techniques, combines them with traditional air pollutant distribution simulation models, and considers the nonlinear relationships between factors influencing traffic pollutants. Simultaneously, it can simulate pollutant concentration distribution under different traffic environments.

[0007] To achieve the aforementioned technical objectives, this invention discloses a method for inferring the spatial distribution of traffic pollutants based on multi-source data. The method includes: collecting hourly pollutant data for the study area and calculating the average value for different time periods as the dependent variable; extracting potential explanatory variables, including geographical location, meteorological elements, land use, road density, POI (Point of Interest) data, and other pollutants; constructing a LUR (Learning and Regression) model using a supervised forward stepwise regression algorithm to screen important variables; constructing an XGBoost machine learning model to establish a machine learning regression model between pollutant concentration and explanatory variables; dividing the study area into regular grids using GIS, extracting the dependent variables corresponding to the centroids of the grids, and inputting them into the machine learning model to predict the pollutant concentration in the grids; conducting surveys on pollutant concentrations in different traffic microenvironments in smaller areas, including walking, driving, subway, and bus; and correcting the machine learning regression results based on the machine learning results and measured data to obtain the distribution of pollutant concentrations in different traffic microenvironments.

[0008] To achieve the above-mentioned technical effects, the technical solution of the present invention is as follows:

[0009] Hourly data of pollutants were collected in the study area, and the average values ​​were calculated for different time periods as the explained variable (S1): Hourly data of pollutants were obtained from environmental monitoring stations, including SO2, NO, pollutants, O3, PM10, PM2.5 and CO.

[0010] Furthermore, statistical descriptions of these traffic pollutant concentrations were conducted, analyzing the changes and distribution of pollutant concentrations during different time periods: daytime (9:00-17:00), nighttime (19:00-7:00 the next day), morning peak (7:00-9:00), and evening peak (17:00-19:00).

[0011] Extracting latent explanatory variables, including geographic location, meteorological elements, land use, road density, POI (Point of Interest) data, and other pollutants (S2): Using ArcGIS, Python, Gaode API, and meteorological websites, latent explanatory variables were extracted, including geographic location, meteorological elements, land use, road density, POI data, and other pollutants.

[0012] Furthermore, potential explanatory variables were extracted from different buffer regions of 100m, 200m, 300m, 500m, 750m, 1000m, 2000m, 3000m and 5000m respectively.

[0013] The LUR model is constructed using a supervised forward stepwise regression algorithm to screen important variables (S3): Important variables are selected from numerous explanatory variables, and collinear variables are eliminated. This step improves the efficiency and accuracy of subsequent machine learning models.

[0014] Furthermore, we explored the relationship between pollutant concentration and explanatory variable data, and compared the magnitude of the impact of different factors on pollutant concentration.

[0015] Construct a machine learning model and establish a machine learning regression model between pollutant concentration and explanatory variables (S4): Based on the selected independent variable data, input the independent variable data and the corresponding pollutant concentration data into the machine learning model, select the optimal parameters of the model according to the cross-validation results, and obtain the optimal model.

[0016] Furthermore, a multi-indicator evaluation model for predictive performance is established, including R... 2 The study also used metrics such as MAE and RMSE, and employed built-in machine learning functions to calculate variable importance, exploring the extent to which variables affect pollutant concentrations.

[0017] The study area was divided into regular grids using GIS, and the explained variables corresponding to the centroids of the grids were extracted and input into a machine learning model to predict the concentration of pollutants in the grids (S5): The study area was divided into 500m*500m grids, and based on the above important variables, the explanatory variable data corresponding to the centroids of the grids were extracted and predicted.

[0018] Furthermore, based on the prediction results, the distribution of pollutants in the region is analyzed.

[0019] A survey was conducted in small areas to investigate pollutant concentrations in different traffic microenvironments, including pedestrian, driving, subway, and bus (S6). Due to factors such as station height and wind speed, the station detection results differed somewhat from the actual traffic source pollutant concentrations in real-world traffic environments. Therefore, experiments were needed to obtain the actual pollutant concentrations in driving, walking, bus, and subway environments.

[0020] Furthermore, based on measured data, we analyze pollutant concentrations under different traffic conditions to provide reference and suggestions for residents' travel.

[0021] Based on the machine learning results and measured data, the machine learning regression results are corrected to obtain the pollutant concentration distribution in different traffic microenvironments (S7): Based on the measured values ​​of pollutant concentration in different regions and different traffic microenvironments, the correction coefficients for different regions and different traffic microenvironments are calculated. The correction coefficients for each grid are calculated using the spatial interpolation method to obtain the pollutant concentration distribution in different traffic microenvironments.

[0022] The technical solution of this invention utilizes multi-source data (including meteorological data, location information, road networks, land use, etc.) and advanced machine learning algorithms to provide a precise and practical method for inferring the spatial distribution of traffic pollutants. Through this invention, the high-precision distribution of traffic pollutants can be inferred, providing a scientific basis for environmental governance and offering visualized suggestions for people's travel. Attached Figure Description

[0023] Figure 1 This is an example diagram of the urban road network data distribution according to the present invention;

[0024] Figure 2 This is an example image of the city POIPOI point of interest data of this invention;

[0025] Figure 3 This is an example diagram showing the time-segmented prediction results of urban traffic source pollutants according to the present invention;

[0026] Figure 4 This is an example diagram of the Kriging correction coefficient based on the driving environment factor of this invention;

[0027] Figure 5 A flowchart of a method for inferring the spatial distribution of traffic pollutants based on multi-source data provided by the present invention; Detailed Implementation

[0028] The present invention will be further described in detail below with reference to the accompanying drawings:

[0029] Please see Figure 5 The present invention provides a method for inferring the spatial distribution of traffic pollutants based on multi-source data, comprising the following steps:

[0030] Step 1: Obtain hourly pollutant data. Obtain hourly pollutant data from environmental monitoring stations. Pollutants include SO2, NO, pollutants, O3, PM10, PM2.5 and CO. The average pollutant values ​​for different time periods are used as the dependent variable: daytime (9:00-17:00), nighttime (19:00-7:00 the next day), morning peak (7:00-9:00), and evening peak (17:00-19:00).

[0031] Y it =∑ day C it / N day ;

[0032] Where: Y it This represents the average pollutant concentration at the i-th monitoring station during the t-th time period, in μg / m³. 3 C it Let represent the daily concentration of pollutants at the i-th monitoring station during time period t, in μg / m³. 3 N day This refers to the number of days for pollutant monitoring.

[0033] Step 2: Extract potential explanatory variables. Extract the potential explanatory variables from geographic location, meteorological elements, land use, road density, POI data and other pollutants. The potential explanatory variables are listed in Table 1.

[0034] Table 1 Variable Description

[0035]

[0036]

[0037] Explanation: (1) "-" indicates that the variable is negatively correlated with the potential explanatory variable and its prior direction is negative; (2) "+" indicates that the variable is positively correlated with the potential explanatory variable and its prior direction is positive; (3) "〇" indicates that the prior direction of the variable is not explicitly specified.

[0038] (1) Geographical location

[0039] Geographic location data is used to describe the location information of environmental monitoring stations, including the station's latitude and longitude coordinates, elevation, distance from the coastline, and whether it is located in an urban area. The latitude and longitude coordinates of each station are extracted based on the WGS-84 coordinate system, with the unit being degrees (°).

[0040] Elevation data refers to the numerical values ​​of each station in the Digital Elevation Model (DEM), expressed in meters (m). These values ​​are obtained using ArcGIS's "Spatial Analyst Tool - Extraction - Extract Values ​​to Points". The DEM model data used originates from the United States Geological Survey (USGS) Shuttle RadarTopography Mission 1-arc second Global (SRTM1) file, with a resolution of 30m.

[0041] Coastline distance refers to the straight-line distance between each station and the nearest coastline, measured in meters (m). This value was obtained through analysis using ArcGIS's "Analysis Tools - Proximity - Near". The coastline data used was sourced from the Global Self-consistent, Hierarchical, High-resolution Geography Database (GSHHG).

[0042] Whether the station is located in the city center is a binary variable, representing whether the station is located in the city center. This section divides the local city into two areas: the city center and the suburbs, using the outer ring road as the boundary. For environmental monitoring stations located in the city center, this variable is set to 1; otherwise, it is set to 0.

[0043] (2) Meteorological elements

[0044] Meteorological elements are sourced from the Meteorological Science Data Center. The Inverse Distance Weighted (IDW) interpolation method is used to simulate the spatial distribution of different meteorological elements in ArcGIS at different time periods. Then, the "Extract Values ​​to Points" geoprocessing tool is used to extract the characteristic values ​​of various meteorological elements at different environmental monitoring stations at different time periods.

[0045]

[0046] In the formula: Z0 represents the estimated value; Z i Let be the attribute value of the i-th sample (i = 1, 2, 3, ..., n); p is the power of the distance, which significantly affects the interpolation result, and its selection criterion is the minimum mean absolute error; D i For distance.

[0047] (3) Land use data

[0048] Land use data includes residential land, commercial land, industrial land, green space, and water bodies. In ArcGIS, buffer zones of 100m, 200m, 300m, 500m, 750m, 1000m, 2000m, 3000m, and 5000m are constructed centered on each environmental monitoring station. The total area of ​​each type of land use within the different buffer zones is extracted and used as a potential explanatory variable.

[0049] (4) Road network data

[0050] like Figure 1 As shown, the road network data comes from publicly available street map (OSM) road data and is divided into expressways and arterial roads. The road lengths of expressways and arterial roads are extracted within buffer zones of 100m, 200m, 300m, 500m, 750m, and 1000m, respectively. Based on this, the road density within the corresponding buffer zones is calculated and used as a potential explanatory variable for the model.

[0051] (5) Points of Interest (POI) data

[0052] like Figure 2As shown, the Point of Interest (POI) data is based on the Baidu Open Map Platform. The names, categories, longitudes, and latitudes of various POIs are crawled according to categories and keywords. Four categories of data are selected: catering services, bus stops, intersections, and gas stations. The number of POIs of each type within buffer zones of 100m, 200m, 300m, 500m, 750m, 1000m, 2000m, 3000m, and 5000m is extracted for each station.

[0053] Step 3: Construct a LUR model using a supervised forward stepwise regression algorithm to screen important variables;

[0054] It should be noted that the LUR model is based on the "regression-inversion" concept, establishing a regression mapping relationship between pollutant concentration and related factors such as meteorological elements, population density, and land use, and inverting the pollutant concentration at any location within the study area. The data related to the potential explanatory variables used in this invention originate from local city official data, totaling 32 fixed environmental monitoring stations, such as... Figure 1 As shown, the local city comprises 16 administrative districts, covering a total area of ​​6,341 square kilometers, encompassing the urban area, and the timeframe covers November 1-30, 2023. Based on the temporal variation pattern of local urban traffic volume throughout the day, the average values ​​of potential explanatory variables for each station during the morning peak (7:00-9:00) and throughout the day were selected as the explained variables for the LUR model.

[0055] Based on previous research findings, 108 potential explanatory variables were extracted from geographical location, meteorological elements, land use, POI data, road network, and other pollutants. These potential explanatory variables were used to construct the LUR model.

[0056] Based on existing research, the prior directions of all potential explanatory variables were determined. For example, since wind speed is conducive to enhancing pollutant dispersion and is negatively correlated with its concentration, the prior direction is wind direction. Next, univariate regression analysis was performed on all potential explanatory variables. (Adjusted R 2 The explanatory variable with the highest value and conforming to the prior direction is used as the initial model for stepwise regression. Finally, other explanatory variables are added to the regression model one by one, if the following conditions are met: (1) If the value increases and the regression coefficient conforms to the prior direction; (2) the direction of the regression coefficients of other explanatory variables in the model remains unchanged; then the explanatory variable is included in the model. Repeat the above process to introduce all potential explanatory variables into the model one by one until the regression model is complete. The increment should not exceed 1%. Finally, a model significance test is performed, and variables with p-values ​​> 0.1 and VIF > 3 (variance inflation factor) are removed.

[0057] Next, model validation is performed. Model validation includes two types: residual analysis and cross-validation. The former is used to verify whether the model results meet the basic assumptions of the regression model, including normality test, equal variance test, residual independence test, and spatial autocorrelation test. The first three types of tests are implemented in SPSS 26.0, while the spatial autocorrelation test is implemented using ArcGIS. Cross-validation is used to verify the accuracy of the model.

[0058] Step 4: Construct a machine learning model and establish a machine learning regression model between pollutant concentration and explanatory variables. Based on the selected independent variable data, input the independent variable data and the corresponding pollutant concentration data into the machine learning model. Use the optuna library in Python to select the optimal parameters of the model based on the cross-validation results to obtain the optimal model. Calculate Ri on the training set and the test set. 2 And RMSE, to evaluate the model.

[0059]

[0060]

[0061] Where: n is the number of samples; y i It is the true value of the i-th sample; It is the predicted value of the i-th sample; It is the mean of the true values.

[0062] Step 5: Divide the study area into a dense grid of 500*500 meters, extract the key variable data corresponding to the centroids of the grid, input them into the model, and obtain high-precision prediction results of pollutant concentrations. The figure below shows an example of pollutant concentration distribution in the local city at different times. Based on the LUR model, estimate the pollutant concentration values ​​at each centroid point during the morning rush hour and throughout the day. Implement spatial visualization in ArcGIS to finally obtain the pollutant concentration distribution map of Shanghai during the morning rush hour and throughout the day from November 1st to 30th, 2023.

[0063] Step 6: Due to factors such as the altitude of the monitoring station and wind speed, the pollutant concentrations detected at the station may differ from those in actual traffic environments. Therefore, experiments need to be conducted under different traffic conditions to obtain pollutant concentrations in driving, walking, public transportation, and subway environments.

[0064] A small, evenly distributed area was selected within the study region, and pollutant concentrations were measured multiple times at different times within the area, taking place while walking, in the subway, driving, or on public transportation. For example... Figure 3As shown, the overall trend is a decrease from the city center to the suburbs. The air quality in the eastern region is generally better than that in the western region, and the air quality in the southern region is better than that in the northern region. The pollutant concentration is significantly higher in the suburbs where logistics parks and industrial parks are more concentrated, while the pollutant concentration is relatively lower in areas with more water bodies and green spaces.

[0065] Step 7: Calculate the correction coefficients for different traffic microenvironments in different regions based on the measured values ​​of pollutant concentrations in different traffic microenvironments.

[0066]

[0067] Where: α d C is the environmental correction factor for the traffic microenvironment d; d The measured pollutant concentration of the traffic microenvironment d is expressed in μg / m³. 3 C ml The simulated concentration of pollutants in the machine learning model, in μg / m³ 3 .

[0068] like Figure 4 As shown, the environmental correction coefficients were interpolated into smooth curves using the Geostatistical Analyst-Kriging tool in ArcGIS. Finally, the correction coefficients were multiplied by the grid pollutant concentrations to obtain the pollutant concentration distributions under different traffic microenvironments.

[0069] Example 2

[0070] For details not described in this embodiment, please refer to Embodiment 1. In reality, with the rapid advancement of urbanization and the surge in motor vehicle ownership, the contribution of road traffic pollution to air pollution will further increase. Long-term commuting exposure to this environment will cause significant damage to residents' health and increase economic and social costs. This embodiment focuses on the typical air pollutant NO2, combining measured data under different traffic environments and a high-precision LUR model to simulate the spatial distribution of NO2 during the morning rush hour. Taking commuters in Jing'an Temple Street, Nanxiang Town, Qibao Town, and Youyi Road Street in Shanghai as examples, it precisely quantifies the exposure risk and characteristics of the entire commuting process. The results show that the exposure risk of driving routes is higher than that of public transportation routes. The overall exposure risk of the group with a commuting time of more than 60 minutes is about 5% lower than that of the groups with a commuting time of less than 15 minutes and less than 45 minutes. The average overall exposure risk and commuting time in Jing'an Temple Street are the lowest in the study area. The average overall exposure risk and commuting time in Nanxiang Town are both relatively high, with the highest average individual exposure at 33.86 μg, followed by Qibao Town. The Youyi Road Subdistrict has the highest overall exposure risk, but also the highest commuting happiness.

[0071] The following study examines the high-precision NO2 simulation distribution based on LUR and measured data correction:

[0072] 1.1 Study Area and Data

[0073] 1.1.1 Study Area

[0074] Shanghai Municipality comprises 16 administrative districts, covering a total area of ​​6,341 square kilometers. Due to the geographical discontinuity of Chongming District compared to other administrative districts in Shanghai, and significant differences in environmental conditions such as road networks and meteorological conditions, this study focuses on the main areas of Shanghai excluding Chongming District. Figure 1 As shown.

[0075] 1.1.2 LUR Model Data

[0076] Land Use Regression (LUR) models, based on the "regression-inversion" concept, establish regression mapping relationships between pollutant concentrations and related factors such as meteorological elements, population density, and land use, and invert pollutant concentrations at any location within the study area. The NO2 concentration data used in this paper originates from official data from Shanghai Municipality, covering a total of 32 fixed environmental monitoring stations (e.g., [missing data]). Figure 1 As shown in the figure, the model covers the urban area of ​​Shanghai (excluding Chongming District) and the time period is from November 1st to 30th, 2023. Based on the time variation pattern of traffic volume in Shanghai throughout the day, the average NO2 concentration at each station during the morning peak (7:00-9:00) and throughout the day is selected as the dependent variable of the LUR model.

[0077] Based on a review of previous research, this paper selects six categories of variables, including geographical location, meteorological elements, land use, distribution of points of interest, road network, and other pollutants. Ultimately, 108 potential explanatory variables are extracted for model construction, and these variables are listed in Table 1.

[0078] 1.1.3 NO2 Measured Data

[0079] Due to factors such as the altitude of environmental monitoring stations and wind speed, the large-scale atmospheric pollution simulation results of the LUR model differ somewhat from the NO2 exposure concentration in the traffic environment. Therefore, the LUR model results need to be corrected and adjusted to improve the accuracy of exposure risk assessment. Cheng Xinghong et al., in simulating NO2 concentrations in North China, found that the spatial distribution of NO2 exhibits significant urban locality. Therefore, using low-resolution measured data can also achieve good source assimilation correction effects. Thus, this paper selected four blocks—Jing'an Temple Street in Jing'an District, Qibao Town in Minhang District, Nanxiang Town in Jiading District, and Youyi Road Street in Baoshan District—as experimental areas for field measurement of NO2 concentrations, as shown in Table 2.

[0080] Table 2. Population and socioeconomic attributes of the experimental area

[0081]

[0082] The experiment was conducted during weekday morning rush hours (7-9 AM) from November 1st to 30th, 2023. The highest temperature ranged from 29℃ to 0℃, resulting in a significant temperature difference. Over 70% of the experimental days had no sustained wind direction. The experimental instrument was a portable nitrogen oxide gas detector, manufactured by Shenzhen Dongri Yingneng Technology Co., Ltd., model SK-800-NOX-K, with a Honeywell original NO2 electrochemical sensor as its built-in sensor. The experiment was designed as a handheld nitrogen oxide detector, using public transportation (bus, subway, and private car) within the study area during peak hours to simultaneously collect NO2 concentration data in the pedestrian environment.

[0083] 1.2 High-precision NO2 simulation distribution based on LUR model

[0084] This paper constructs a LUR model based on the supervised forward stepwise regression algorithm proposed by Wang Huizi et al. The final results of the LUR model are listed in Table 3.

[0085] Table 3. LUR model results during the morning rush hour.

[0086]

[0087]

[0088] The model validation results of this LUR model conform to the basic assumptions of regression models. The standardized residual mean is 6.65E-15, close to 0, and the standard deviation is 0.916, close to 1; the Durbin-Watson test value is 1.369, between 0 and 4; the spatial Moran's index is 0.90, indicating a very strong positive correlation in the spatial model results. Furthermore, the cross-validation results show that the root mean squared error (RMSE) is 0.728, and the difference from R² is less than 0.1, indicating that the model has good stability.

[0089] A 500m × 500m regular grid was generated in ArcGIS, and the geographic environment data of the centroids of each grid was extracted. Based on the LUR model described above, the NO2 concentration values ​​at each centroid during the morning rush hour and throughout the day were estimated. Spatial visualization was implemented in ArcGIS, ultimately obtaining the NO2 concentration distribution map of Shanghai during the morning rush hour and throughout the day from November 1st to 30th, 2023, as shown below. Figure 3As shown, NO2 concentrations during Shanghai's morning rush hour exhibit significant spatial differentiation. Generally, NO2 concentrations decrease from the city center to the suburbs. Air quality in Pudong is generally better than in Puxi, and air quality in the southern areas, including Jinshan and Fengxian districts, is better than in the northern areas. Suburban areas with a high concentration of logistics and industrial parks, such as Baoshan and Minhang districts, have significantly higher NO2 concentrations, while areas near Dianshan Lake with more water bodies and green spaces have relatively lower NO2 concentrations. Furthermore, within Shanghai's expressway network, NO2 concentrations are concentrated at key nodes, especially in the Hongqiao Business District. As the network extends outwards towards the suburbs, NO2 concentrations gradually decrease.

[0090] 1.3 High-precision simulated NO2 distribution after correction of measured data

[0091] Since the interior environments of subways and buses were all unventilated and air-conditioned, the overall exposure conditions in the four experimental areas were quite similar. The NO2 exposure concentration of public transport passengers fluctuated less and was relatively stable; therefore, the average passenger exposure concentration was used when calculating the NO2 pollution exposure intensity. The average exposure concentration for bus passengers was 57.55 μg / m3, and the average exposure concentration for subway passengers was 55.3 μg / m3.

[0092] For pedestrians, due to direct contact with air, the exposure risk is highly correlated with environmental exposure risk. Meanwhile, the experiment was conducted in November, and during the driving route experiment, the vehicle interior environment was semi-ventilated, with air conditioning off and the air purifier turned off. The exposure risk of the driver / passengers inside the vehicle was highly correlated with the atmospheric environment. Therefore, the exposure risk during commuting was adjusted using walking and driving environment correction factors. The formula follows the correction factor format from the environmental impact assessment, as follows:

[0093]

[0094] Where: α is the traffic environment correction coefficient;

[0095] C represents the NO2 concentration in the traffic environment, expressed in μg / m3.

[0096] C LUR The simulated concentration of NO2 in the LUR model is expressed in μg / m3.

[0097] The average NO2 concentration data and correction factor for walking and driving environments are shown in Table 4.

[0098] Table 4 Average NO2 concentration and correction factor under traffic conditions during morning rush hour

[0099]

[0100] The simulated NO2 concentration distribution results of the LUR model during the morning rush hour were corrected based on regional correction coefficients for pedestrian and driving environments, as well as the average correction coefficient. The results are shown in the figure below. NO2 pollution is severe at the intersection of the Shanghai Ring Expressway, Shenhai Expressway, and Hujia Expressway in Jiading District, especially near the Nanxiang Impression City area. The area outside the outer ring road in Baoshan District has a wide area of ​​high NO2 pollution due to numerous logistics and industrial parks. The area around Hongqiao Business District, as an important logistics and container distribution center, also experiences relatively severe NO2 pollution near the intersection of the Huyu Expressway, Hujin Expressway, and Shenhai Expressway. The areas surrounding Gaodong Town, Waigaoqiao Free Trade Zone, and the Shanghai Free Trade Zone in Pudong District are key pollution areas in northeastern Shanghai.

[0101] 2. Risk assessment of residents' exposure during the morning rush hour commute

[0102] 2.1 Travel Chain Data

[0103] The data used in this study is commuting travel data processed using LBS (Location-Based Services) big data. The LBS big data was provided by Guangzhou Mengxiang Network Technology Co., Ltd., and it comprehensively reports user travel behavior based on multiple signal data types such as Wi-Fi, Bluetooth APs, and GPS. The data content mainly includes anonymized user attribute data and user reporting data in multiple scenarios. The time period covers 7:00-9:00 AM on November 1, 2023, and the data geographical scope is the entire Shanghai.

[0104] The processing steps include: First, the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm is used to cluster high spatial density points, thereby refining the residents' activity areas and identifying travel stops. Then, referring to the user work-residence identification rules of Yang Chao et al., the commuting home and workplace locations are determined. Finally, the commuting process during the morning peak hours is identified based on work-residence identification, and commuting time is extracted. The processed commuting data is presented in 100*100m grid units. Approximately 196,498 travel records from 319,455 users residing in four blocks: Jing'an Temple Street, Qibao Town, Nanxiang Town, and Youyi Road Street, are selected as the research object. The data structure is shown in Table 5.

[0105] Table 5. Explanation of Commuting Data in the Study Area

[0106]

[0107] Using the data from the Seventh National Population Census (hereinafter referred to as "Seventh Census Data") as a reference, SPSS 26.0 was used to conduct correlation analysis and paired-samples t-tests on the spatial distribution of the permanent commuter population of Jing'an Temple Subdistrict, Qibao Town, Nanxiang Town, and Youyi Road Subdistrict extracted from internet location data. This was to verify whether there were significant differences between the two sets of data, and thus to demonstrate whether internet location data can represent the distribution characteristics of the permanent population in the study area. The results are shown in Table 6. The Pearson correlation coefficient between the two sets of data was 0.990, close to 1, indicating a high degree of linear correlation. Simultaneously, the Levin variance significance of the paired-samples t-test was 0.999, much greater than 0.05, indicating that there was no significant difference between the two sets of data at the 0.05 significance level. Therefore, it can be confirmed that internet location data can characterize the distribution characteristics of the permanent population of Jing'an Temple Subdistrict, Qibao Town, Nanxiang Town, and Youyi Road Subdistrict.

[0108] Table 6 Population distribution in the study area

[0109]

[0110] 2.2 Traffic Modality Sharing

[0111] Regarding traffic modal share, an integer programming optimization equation is constructed based on the average commuting time in the travel chain data and the commuting time of driving and public transport routes crawled from the Gaode API to solve for the public transport commuting ratio. The optimization objective of the optimization equation is to minimize the difference between the average commuting time of the number of people using driving routes and the average commuting time of the number of people using public transport routes and the average commuting time of the identified commuters. This model is based on the following assumptions: First, the commuting times of different routes and modes crawled from the Gaode API route planning are not significantly different from the actual commuting times of residents; second, the commuting strategy set in this study in the Gaode API route planning is to minimize the total travel time, which is consistent with the strategy selection when residents actually commute; and finally, the total number of people with identified commuting times within a grid can represent the current commuting status of residents in that grid.

[0112] The optimization equation is as follows:

[0113] minz=|(N d *T d +N t *T t ) / T c -N c |

[0114] stN d +N t =N c

[0115] 0≤N d N t ≤Nc

[0116]

[0117] Where, N d For the number of commuters on the driving route that needs to be determined, N t Let be the number of commuters on the public transportation route that needs to be determined.

[0118] This study uses the PuLP library in Python to define variables, objective functions, and constraints to solve an integer linear programming problem involving commuters using two different routes. The traffic allocation results are shown in Table 7. The average public transport commuting ratio within the study area is 0.222, and all areas except Nanxiang Town are below this average. Jing'an Temple Subdistrict has the lowest public transport commuting ratio at 0.194, while Qibao Town and Youyi Road Subdistrict have ratios that are almost equal and slightly below the average. Nanxiang Town has the highest public transport commuting ratio at 0.263. According to the 2019 Resident Travel Survey, rail commuting accounts for approximately 17.3% of all commuting modes, which is consistent with the traffic allocation results of this study.

[0119] Table 7. Number of commuters using driving and public transportation routes within the study area.

[0120] Total number of commuters 11007 125379 118504 62040 Commuters on bus routes 2135 26580 31167 13649 Number of commuters on driving routes 8872 98799 87337 48391 Public transportation commuting ratio 0.194 0.212 0.263 0.220

[0121] 2.3 Calculation of Exposure Risks During the Residents' Commute

[0122] The Gaode Maps API was used to crawl the planned driving routes, waypoints, and total commuting time between residences and workplaces within the study area. The strategy was set to 0, prioritizing speed (this route is not necessarily the shortest). The NO2 concentration at each waypoint was obtained from the NO2 concentration distribution results during the morning rush hour based on the LUR model. The formula for calculating the exposure risk throughout the entire commuting process is as follows:

[0123]

[0124] Among them, R d NO2 exposure risk along driving routes; T represents the average NO2 concentration at the route point during the morning rush hour in the driving environment of the i-th driving segment; di Let be the commuting time for the i-th driving segment.

[0125] Public transportation routes include walking, buses, and subways. The commuting time for each waypoint and route segment was obtained. For walking, the NO2 concentration at each waypoint was obtained from the NO2 concentration distribution results during the morning rush hour in a walking environment. For buses and subways, the average passenger exposure concentration during the morning rush hour was used. The formula for calculating the exposure risk throughout the commuting process on public transportation routes is as follows:

[0126]

[0127] Among them, R t NO2 exposure risk on public transportation routes; T represents the average NO2 concentration at the route point during the morning rush hour in the pedestrian environment of the i-th pedestrian segment; wi C represents the commute time for the i-th walking segment; b The average exposure concentration for bus passengers was taken as 57.55 μg / m3 in this study; T bi C represents the commute time for the i-th bus segment; s The average exposure concentration for subway passengers is 55.3 μg / m3 in this study; T si Let be the commuting time for the i-th subway segment.

[0128] Using the traffic mode allocation results from the previous section, the proportion of public transportation, and the commuting time and exposure risk of driving routes and public transportation routes within the study area, the comprehensive commuting time and comprehensive exposure risk of each study area are calculated using the following formula.

[0129]

[0130] Where, α t α represents the public transportation commuting ratio calculated in the previous chapter. d The percentage of commuters by car is summed with the percentage of commuters by public transport, and the sum is 1. R d To mitigate the risks of commuting by car, T d For driving route commute time, N d For the number of people commuting by car, R t To mitigate the risks of commuting via public transportation, T t N represents the commute time for public transportation routes. t N represents the number of commuters using public transportation. c This represents the total number of commuters.

[0131] Furthermore, the formula for calculating overall individual exposure is based on the research findings of Jing Huang et al., as follows:

[0132]

[0133] Among them, E c For individual comprehensive exposure, Bin For inhalation rate, γ congesti0n The congestion index for Shanghai's commuting peak hours in 2023 was 1.928.

[0134] 2.4 Analysis of the characteristics of residents' exposure risk throughout the entire process

[0135] 2.4.1 Commuting Location

[0136] In terms of transportation location, Jing'an Temple Subdistrict is located within the Inner Ring Road, with subway stations Jing'an Temple Station and Wuding Road Station, and subway lines including Lines 2, 7, and 14. It also benefits from the Yan'an Road Elevated Highway and the North-South Elevated Highway, making its overall transportation infrastructure highly developed. Qibao Town is mostly outside the Outer Ring Road, with a small portion located between the Middle and Outer Ring Roads. It has subway lines including Line 9, which directly connects to Caohejing and Xujiahui, Line 12, which directly connects to Xuhui Riverside, and the under-construction Airport Link Line, which will directly connect to Hongqiao Airport and Pudong Airport. It is also close to the Middle Ring Road, Outer Ring Road, and Yan'an Road Elevated Highway, making transportation convenient. Nanxiang Town currently has Line 11, which directly connects to Jiangsu Road and Xujiahui, and the Jia-Min Line, which is under construction and leads to the Hongqiao Business District. Access to the Middle and Outer Ring Roads is also very convenient via the Shanghai-Jiading Expressway and Shanghai-Nanxiang Expressway. Youyi Road Subdistrict is located in the Songbao area, with only Line 3 currently serving the area, and no future subway lines are planned. The outer ring road, the suburban ring road, and the Yixian Road elevated highway that leads directly to the city center make driving convenient, but traffic congestion is severe during the evening rush hour.

[0137] Using the formulas for calculating the distribution of comprehensive exposure concentration and duration mentioned above, the average comprehensive exposure risk and duration distribution were calculated for Jing'an Temple Subdistrict in Jing'an District, Qibao Town in Minhang District, Nanxiang Town in Jiading District, and Youyi Road Subdistrict in Baoshan District. The results are shown in Table 8. The average comprehensive exposure risk of Jing'an Temple Subdistrict was 61.46 μg / m³, and the average comprehensive commuting time was 26.78 minutes, both the lowest in the study area. Nanxiang Town had the second highest average comprehensive exposure risk and commuting time in the study area. Youyi Road Subdistrict had the highest average comprehensive exposure risk at 65.13 μg / m³, but its average comprehensive commuting time was shorter at 30.19 minutes, only slightly higher than that of Jing'an Temple Subdistrict.

[0138] Table 8. Average composite exposure concentration and time distribution within the study area.

[0139]

[0140] The average individual exposure level comprehensively considered commuting congestion, exposure risk, and commuting time. In terms of this indicator, Jing'an Temple Subdistrict had the lowest average individual exposure level in the study area, at only 29.45 μg during the morning rush hour. Nanxiang Town Subdistrict had the highest average individual exposure level in the study area, at 33.86 μg. Qibao Town was second only to Jing'an Temple Subdistrict, with an average individual exposure level of 33.43 μg. Although Youyi Road Subdistrict had the highest overall exposure risk due to its shorter commuting time, its average individual exposure level was similar to that of Qibao Town, at 33.13 μg.

[0141] The distribution of comprehensive exposure risk for commuters in Jing'an Temple Subdistrict of Jing'an District, Qibao Town of Minhang District, Nanxiang Town of Jiading District, and Youyi Road Subdistrict of Baoshan District is shown in the figure below. In Jing'an Temple Subdistrict, the largest number of commuters have a comprehensive exposure risk of 60–62 μg / m³, with over 60% having a risk below 62 μg / m³. In Qibao Town, the largest number of commuters have a risk of 60–63 μg / m³, accounting for over 40% of the total, while over 90% have a risk below 70 μg / m³. Nanxiang Town and Youyi Road Subdistrict, both former industrial areas, have similar comprehensive exposure risk distribution structures. The highest peak exposure risk in Youyi Road Subdistrict is 68 μg / m³, slightly higher than Nanxiang Town's 66 μg / m³.

[0142] 2.4.2 Commuting time

[0143] In October 2014, Shanghai pioneered the concept of a "15-minute community living circle," which was explicitly implemented in the "Shanghai 2035" master plan, aiming to create a "livable, workable, tourist-friendly, education-friendly, and retirement-friendly" community living circle. Therefore, the proportion of the commuting population with a commute time of less than 15 minutes was chosen to represent the proportion of residents with short commutes. Simultaneously, referencing New York City's urban planning which sets a 90% commute rate within 45 minutes as a goal for urban prosperity, equity, and sustainable development, and Singapore's transportation planning which sets an 80%–90% commute rate within 45 minutes as an urban development goal, this study selects the proportion of the commuting population within 45 minutes as the proportion of medium commuters. Furthermore, referring to commuting monitoring reports of major Chinese cities, the proportion of the population with a commute time of more than 60 minutes is considered the proportion of extreme commuters. The commuting time evaluation indicators for Jing'an Temple Subdistrict in Jing'an District, Qibao Town in Minhang District, Nanxiang Town in Jiading District, and Youyi Road Subdistrict in Baoshan District are shown in Table 9.

[0144] Table 9 Evaluation Indicators of Commuting Time within the Study Area

[0145]

[0146] Jing'an Temple Subdistrict has nearly 30% of its residents with commutes of less than 15 minutes, second only to Youyi Road Subdistrict. Its extreme commuting rate is the lowest at less than 6%, and its medium commuting rate of 86.53% is the highest in the study area. Qibao Town has the lowest short commuting rate at 23.73%, but its medium commuting rate is close to 80%, second only to Jing'an Temple Subdistrict. Nanxiang Town has the lowest medium commuting rate at 74.63%, and the highest extreme commuting rate at 12.73%, indicating significant development needs in future transportation planning. Youyi Road Subdistrict has the highest short commuting rate at 32.71%, and a relatively high medium commuting rate, but also a relatively high extreme commuting rate of nearly 12%. Nanxiang Town and Youyi Road Subdistrict should focus on addressing the issue of extreme commuting in the future.

[0147] Using evaluation indicators of different commuting durations as a starting point, the average comprehensive exposure risk for different commuting periods within the study area was calculated, and the results are shown in Table 10. Overall, the comprehensive exposure risk for extremely long commutes was about 4% lower than that for short and medium commutes, possibly related to the higher proportion of public transportation use. The medium-commuter group in Youyi Road Subdistrict of Baoshan District had the highest comprehensive exposure risk, while the extremely long-commuter group in Jing'an Temple Subdistrict had the lowest comprehensive exposure risk.

[0148] Table 10 Average composite exposure risk for different commuting times within the study area.

[0149]

[0150] Under driving conditions, the NO2 concentration distribution on Shanghai's expressway network shows an overall trend of high in the north and low in the south, with higher concentrations at intersections and lower concentrations at the edges. NO2 pollution is severe on the expressway network near the Hongqiao Business Center. The areas surrounding the intersection of the Shanghai Ring Expressway and Yun Chuan Highway in Baoshan District, as well as the North-South Elevated Road and Yixian Elevated Road, all experience relatively severe NO2 pollution. Drivers (or passengers) in semi-ventilated private cars have more frequent contact with vehicle exhaust fumes, making driving routes a higher exposure risk compared to commuting by public transport. The Jing'an Temple Subdistrict within the Inner Ring Road has the lowest exposure risk. Nanxiang Town has the highest individual exposure among commuters due to its high overall exposure concentration and long commute time. The mid-commuter group in Youyi Road Subdistrict has the highest overall exposure risk.

[0151] This embodiment addresses the shortcomings of the LUR model, such as discrepancies between large-scale simulation results and traffic environment exposure, and insufficient description of the temporal evolution of pollution. It innovatively uses traffic environment correction coefficients and travel trajectory data to overcome these descriptive biases and dynamic deficiencies, constructing a refined exposure risk assessment model for the entire commuting process of residents, thus improving the accuracy of exposure risk assessment. Meanwhile, research on the fairness of traffic pollution exposure in my country is still in its early stages. Whether different locations and groups bear unequal exposure risks, and how to quantify such unfairness, remains unclear. This paper focuses on NO2, a typical traffic source pollutant, and delves into the differences in the impact of traffic pollution on the exposure risks of different modes of transportation and different regional groups. It provides theoretical basis and data support for cities to formulate refined air pollution control policies, and has broad reference value and certain applicability.

[0152] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0153] In conclusion, 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 within the protection scope of the present invention.

Claims

1. A method for inferring the spatial distribution of traffic pollutants based on multi-source data, characterized in that: Includes the following steps: S1: Collect hourly data on pollutants in the study area and calculate the average value for different time periods as the explained variable; S2: Extract potential explanatory variables, including geographic location, meteorological elements, land use, road density, and POI (Point of Interest) data; S3: Construct a LUR model using a supervised forward stepwise regression algorithm to screen important variables; Screening important variables using the LUR model includes: constructing a LUR model using a supervised forward stepwise regression algorithm to screen out important variables that meet the corresponding conditions; S4: Construct a machine learning model and establish a machine learning regression model between pollutant concentration and explanatory variables; input important dependent variable data into the machine learning model, train the model, evaluate the model based on R2, RMSE, and MAE, and select the machine learning model with the best prediction performance based on the cross-validation results. S5: Use GIS to divide the study area into regular grids, extract the explained variables corresponding to the centroids of the grids, and input them into a machine learning model to predict the pollutant concentration in the grids; the selection logic for the machine learning model with the best prediction effect is as follows: divide the study area into a dense grid of 500*500 meters, extract the important variable data corresponding to the centroids of the grids, input them into the model, and obtain the prediction result with the highest pollutant concentration. S6: Conduct a survey on the concentration of pollutants in different traffic microenvironments in small areas, including walking, driving, subway, and bus; using pollutant sensors, select multiple small areas to survey the concentration of pollutants in different traffic microenvironments, including walking, driving, subway, and bus, to obtain pollutant concentration data in different traffic microenvironments in multiple areas; S7: Based on the machine learning results and measured data, correct the machine learning regression results to obtain the pollutant concentration distribution in different traffic microenvironments; calculate the correction coefficients for different traffic microenvironments in different regions based on the measured pollutant concentrations in different traffic microenvironments. ； in: The environmental correction coefficient for the traffic microenvironment d; The measured pollutant concentration in the traffic microenvironment d; The simulated concentration of pollutants in the machine learning model; The environmental correction factor was interpolated into a smooth curve using the Geostatistical Analyst-Kriging tool in ArcGIS. Finally, the correction factor was multiplied by the grid pollutant concentration to obtain the pollutant concentration distribution under different traffic microenvironments.

2. The method for inferring the spatial distribution of traffic pollutants based on multi-source data according to claim 1, characterized in that: The data in step S1 includes hourly traffic pollutant data, which is divided into daytime, nighttime, morning peak, and evening peak according to the time variation pattern of traffic volume, and the average value is calculated.

3. The method for inferring the spatial distribution of traffic pollutants based on multi-source data according to claim 1, characterized in that: In step S2, potential explanatory variables are extracted using ArcGIS, Python Gaode API, and meteorological websites, including geographic location, meteorological elements, land use, road density, and POI (Point of Interest) data.

Images