Urban non-point source pollution refined simulation method combining open source information and machine learning

Abstract

The invention discloses an urban non-point source pollution refined simulation method combining open source information and machine learning, and belongs to the field of urban non-point source pollution monitoring and evaluation, and the method comprises the steps: obtaining open source geographic data, including road network, POI / AOI data, and generating spatial feature raster data sets of population density, density of buildings with different functions, traffic intensity index and the like; non-point source pollution load actual measurement data of representative point locations under different early-stage rainy-day-free conditions are obtained through field sampling, so that a data set is constructed; training an input and output mapping relation by adopting a machine learning algorithm, and establishing a prediction model; for any point location to be simulated, corresponding characteristics are given, and a non-point source pollution load quantification result can be quickly obtained by inputting the model. According to the method, open source building information and building inversion population data are systematically applied to non-point source pollution load estimation, the defect that a traditional empirical formula neglects spatial diversity is overcome, and rapid, low-cost and high-precision non-point source pollution load estimation is achieved.

Description

Technical Field

[0001] This invention belongs to the field of urban non-point source pollution monitoring and assessment technology, and in particular relates to a refined simulation method for urban non-point source pollution that combines open source information and machine learning. Background Technology

[0002] With significant progress in controlling point source pollution from urban life and industry, the main problem of urban water pollution has gradually shifted from point source pollution to non-point source pollution. Specifically, urban non-point source pollution refers to the accumulation of pollutants on surfaces such as plazas and roads during dry weather, which are then washed into urban receiving water bodies via surface runoff or drainage networks during the rainy season. The highly complex urban environment leads to significant ambiguity and spatiotemporal heterogeneity in non-point source pollution, attracting widespread attention from scholars. Controlling pollution at its surface sources is an effective way to alleviate urban non-point source pollution. Efficiently and accurately quantifying surface source pollution loads can provide direction and a foundation for source control of urban non-point source pollution, and is of great significance for risk assessment and key management of urban non-point source pollution.

[0003] Currently, the mainstream method for quantitative research on urban non-point source pollution is the mechanistic model, represented by the Swing Model (SWMM). Mechanistic models focus on the formation principles of urban non-point source pollution, incorporating pollutant source accumulation, process erosion, and pipeline migration into the model through mathematical modeling, and the output results have relatively clear physical meaning. However, mechanistic models also have several prominent problems in application: First, model construction requires a large amount of high-precision basic data (such as land use, underlying surface, and other related data), while in reality, most cities generally suffer from sparse, scarce, and difficult-to-obtain data; second, the model has numerous physical parameters, making the calibration process complex and requiring significant effort for model calibration; third, the description of the non-point source pollution accumulation process in the model is too simplistic, using empirical formulas that only consider the influence of the number of rainless days in the preceding time, ignoring spatial differences, resulting in low quantitative accuracy.

[0004] To address the aforementioned issues, many scholars have proposed several technical improvements. For example, data fusion techniques can be employed, such as integrating remote sensing image inversion and assimilating low-precision monitoring data, to supplement the basic data for modeling. However, most of the required data remains difficult to obtain due to practical limitations. Regarding parameter calibration, optimization algorithms have been introduced to improve the efficiency and accuracy of parameter calibration through their adaptive optimization capabilities. However, the sheer number of parameters required for calibration leads to high model uncertainty. To address the insufficient description of the source accumulation process, a differentiated setting of pollutant accumulation coefficients for different functional areas (such as main traffic arteries and commercial areas) based on spatial heterogeneity characteristics has been proposed. However, this still suffers from the problem of overgeneralization of regional data and cannot fully explain the spatial heterogeneity of urban non-point source pollution.

[0005] Therefore, there is an urgent need for a quantitative method for urban non-point source pollution load, especially for estimating pollution load at the source, that clearly considers geographical spatial factors, is applicable to areas with sparse data, and provides fast and detailed calculation results. Summary of the Invention

[0006] To address the problems existing in current technologies, the purpose of this invention is to propose a refined simulation method for urban non-point source pollution that combines open-source information with machine learning. The open-source information utilizes POI and AOI data commonly used in urban planning.

[0007] According to a first aspect of the embodiments of this application, a method for refined simulation of urban non-point source pollution combining open-source information and machine learning is provided, comprising: S1: Obtain multi-source geographic information of the target area through an open-source geographic information platform and perform data preprocessing. The multi-source geographic information includes traffic network data, POI data and AOI data corresponding to buildings. S2: Based on the preprocessed multi-source geographic information, the buildings in the target area are classified, the population is estimated, and the road traffic intensity index is defined, generating building density raster datasets, population density raster data, and traffic intensity index raster data of different functional categories. S3: During the rainless period, obtain sampling samples from several sampling points in the target area, and calculate the non-point source pollution load data corresponding to each sampling sample under different previous rainless days; S4: Based on the raster data obtained in step S2, extract the building density, population density, and traffic intensity index corresponding to each sample, and combine them with the number of rainless days in the previous period corresponding to each sample as feature input variables, and use the non-point source pollution load data corresponding to each sample as output variables, thereby constructing a feature-target dataset; S5: Normalize the data in the feature-target dataset, train the input-output mapping relationship using machine learning algorithms, and establish an urban non-point source pollution prediction model; S6: Spatial gridding is performed on the area to be simulated, and the building density, population density, traffic intensity, and number of rainless days in the previous period are extracted for each grid. The urban non-point source pollution prediction model is used to make predictions to obtain the non-point source pollution load data of the area to be simulated.

[0008] Further, step S1 includes: Download the urban transportation network of the target area from the open platform; Based on the geographic information platform program interface, access the building information point / information surface database in the platform, and use the Request function to retrieve the POI and AOI data of each building in the target area; , in, P n 、P a 、P t 、P l These are the name, address, category, and latitude / longitude coordinates from the POI and AOI data corresponding to the building, respectively. url Interface coding for accessing open-source geographic information databases; The acquired multi-source geographic information in the geographic information system is preprocessed, including data import, data visualization, unified coordinate system, and data projection.

[0009] Further, in step S2, the buildings in the target area are classified to generate different functional building density grids, specifically: Based on the preprocessed POI data, the buildings in the target area are reclassified; Based on the reclassified functional categories and the AOI data, all buildings in the target area are clustered, and density is calculated for each functional category to obtain a building density raster dataset for different functional categories.

[0010] Further, in step S2, population estimation is performed on the buildings in the target area to generate population density raster data, specifically as follows: Based on the building function category, determine the per capita usable area index A and the building usable area coefficient for different buildings. β ; The building area S is obtained from the AOI data, combined with different per capita usable area indices A and building usable area coefficients. β Estimate the population size P for different buildings: , in, P For the number of people living in the building; S The building area; β This refers to the building's usable area coefficient. A This refers to the per capita usable floor area of ​​buildings. The population count of a building is assigned to its centroid, and kernel density analysis is used to calculate the population density of the target area, resulting in a population density raster for the target area.

[0011] Further, in step S2, a road traffic intensity index is defined for the target area, and traffic intensity index raster data is generated, including: Based on the preprocessed traffic network data, the roads in the target area are divided into several categories; Obtain the single-lane design capacity of different road types, and calculate the traffic intensity index of different road types by combining the number of lanes of different road types; The traffic intensity index is assigned to each road in the target area, and the line density analysis method is applied to perform cluster calculation to obtain the road traffic intensity index raster data of the target area.

[0012] Further, in step S3, the sample is obtained in the following way: Arrangement J Sampling points located around different types of buildings, in areas with different population densities, and on different types of roads; During the rainless period, that is, the time from the end of one rainfall to the start of the next, sampling points are... L Secondary surface dust sample collection; For the obtained U=J*L For each street dust sample, calculate the total surface source pollution load. M and the pollutant load data. M p : , , in, M It is the total load of surface source pollution; m It refers to the quality of street dust samples collected from different locations; a These are the sampling areas at different locations; C p It is the first in the dust of the street P The concentration of the pollutants; M p It is the first in the dust of the street P The load of various pollutants.

[0013] Furthermore, in step S4, the construction of the feature-target dataset can be completed through the following steps: Based on raster data, each sample is obtained. j Different functional building densities B Population density PD Traffic intensity WD Obtain the number of rainless days in the preceding period corresponding to each sample. ADD Construct each sample j eigenvectors SF j : , , in, It is a samplej of I Density information for buildings with different functions; PD j It is a sample j Population density information; WD j It is a sample j Traffic intensity information; ADD j It is a sample j The corresponding number of rainless days in the preceding period; Based on each sample j Based on the feature vectors, construct the feature information matrix of all samples. FM ; , Get each sample j Data on total non-point source pollution load and loads of different types of pollutants were used to construct a sample for each pollutant. j Target vector ST j : , in, It is a sample j Total surface source pollution load; It is a sample j The P Pollutant load, P ≥1; Based on each sample j Target vector, establish target information matrix TM ; .

[0014] According to a second aspect of the embodiments of this application, a refined simulation device for urban non-point source pollution combining open-source information and machine learning is provided, comprising: The open-source information acquisition module is used to acquire multi-source geographic information of the target area through an open-source geographic information platform and perform data preprocessing. The multi-source geographic information includes traffic network data, POI data and AOI data corresponding to buildings. The raster data generation module is used to classify buildings, estimate population, and define road traffic intensity index in the target area based on preprocessed multi-source geographic information, and generate building density raster datasets, population density raster data, and traffic intensity index raster data of different functional categories. The non-point source pollution load calculation module is used to acquire sampling samples from several sampling points in the target area during rainless periods and calculate the non-point source pollution load data corresponding to each sampling sample under different prior rainless days. The dataset construction module is used to extract the building density, population density, and traffic intensity index corresponding to each sample based on the raster data obtained by the raster data generation module. Combined with the number of rainless days in the previous period corresponding to each sample, it is used as the feature input variable, and the non-point source pollution load data corresponding to each sample is used as the output variable, thereby constructing the feature-target dataset. The model training module is used to normalize the data in the feature-target dataset, train the input-output mapping relationship using machine learning algorithms, and establish an urban non-point source pollution prediction model. The prediction module is used to spatially grid the area to be simulated, extract the building density, population density, traffic intensity, and number of rainless days in the previous period for each grid, and use the urban non-point source pollution prediction model to make predictions to obtain the non-point source pollution load data of the area to be simulated.

[0015] According to a third aspect of the embodiments of this application, an electronic device is provided, comprising: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors perform the method as described in the first aspect.

[0016] According to a fourth aspect of the embodiments of this application, a computer-readable storage medium is provided that stores computer instructions thereon, which, when executed by a processor, implement the steps of the method as described in the first aspect.

[0017] The technical solutions provided by the embodiments of this application may include the following beneficial effects: (1) This invention abandons the empirical formula for calculating pollution generation from traditional surface sources and proposes a calculation method that uses population density, functional building density, and traffic information to correlate non-point source pollution. Compared with the current formulas that require different settings for cumulative pollution generation in different areas, the method described in this invention can fully explain the spatial heterogeneity of urban non-point source pollution and achieve refined load quantification results; (2) This invention is the first to apply open-source geographic big data such as POI and AOI to the quantification of urban non-point source pollution load. It can quickly obtain the distribution, labels and area of ​​urban buildings through POI and AOI data, and calculate the density of buildings with different functions and the refined population density, which solves the problems of sparse basic data, difficulty in data acquisition and high cost in actual situations. (3) This invention proposes a mapping model based on machine learning algorithms, which can better handle the complex nonlinear relationships of urban non-point source pollution and is suitable for refined spatial quantification of non-point source pollution. At the same time, the model construction process is simple, no physical parameters need to be calibrated, and the prediction results are output quickly, which greatly improves the efficiency of computation.

[0018] (4) This invention proposes a deterministic and universal method to obtain the distribution of surface source pollution loads, taking into account both time and space. In particular, it clarifies many influencing factors of surface source pollution in space, breaking the ambiguity of traditional methods that rely on subjective generalization and empirical zoning in spatial adaptation, and providing new thinking for the quantification of non-point source pollution.

[0019] (5) The technical framework for estimating urban non-point source pollution load based on open source data and machine learning proposed in this invention solves the defects of traditional calculation methods, ensures accuracy, and improves precision. It is an important supplement to the field of urban non-point source pollution research, provides important technical support for the control of urban non-point source pollution, and has great value for promotion and practical engineering application.

[0020] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0021] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0022] Figure 1 This is a flowchart illustrating a refined simulation method for urban non-point source pollution that combines open-source information with machine learning, according to an exemplary embodiment.

[0023] Figure 2 This is a building and road network distribution map of the case area Y town, as illustrated by an exemplary embodiment.

[0024] Figure 3 This is a heat map of the functional building density of case area Y town, as illustrated in an exemplary embodiment.

[0025] Figure 4 This is a population density heat map of case area Y town shown according to an exemplary embodiment.

[0026] Figure 5 This is a traffic intensity index distribution map of case area Y town, as illustrated by an exemplary embodiment.

[0027] Figure 6 This is a street dust sampling point distribution map of case area Y town, as shown according to an exemplary embodiment.

[0028] Figure 7 This is a comparison chart of the predicted and actual load values ​​of sampling points in the case area Y town according to an exemplary embodiment of the invention method.

[0029] Figure 8This is a distribution map of non-point source pollution load in case area Y town when the number of rainless days in the preceding period ADD=2, according to an exemplary embodiment.

[0030] Figure 9 This is a distribution map of non-point source pollution load in case area Y town when the number of rainless days in the preceding period ADD=7, according to an exemplary embodiment.

[0031] Figure 10 This is a block diagram illustrating a refined simulation device for urban non-point source pollution that combines open-source information with machine learning, according to an exemplary embodiment.

[0032] Figure 11 This is a schematic diagram of an electronic device according to an exemplary embodiment. Detailed Implementation

[0033] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0034] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0035] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0036] like Figure 1 As shown, this invention provides a refined simulation method for urban non-point source pollution that combines open-source geographic information and machine learning. The steps of establishing the method are as follows: S1: Obtain multi-source geographic information of the target area through an open-source geographic information platform and perform data preprocessing. The multi-source geographic information includes traffic network data, POI data and AOI data corresponding to buildings. Specifically, a Point of Interest (POI) is a facility location on a map with a specific function or significance. It is one-dimensional data, typically containing latitude and longitude coordinates and semantic attributes such as name, category, and address. An Area of ​​Interest (AOI), on the other hand, is two-dimensional data with clearly defined boundaries. It usually includes polygon coordinates and semantic attributes, and can be used for area analysis, building plan visualization, etc. It can also contain a unique identifier for the POI as one of its internal attributes. This step may include the following sub-steps: S11: Download the urban road network of the target area from open platforms such as OpenStreetMap; S12: Based on the program interfaces of geographic information platforms such as Baidu / Gao De / OpenStreet Map, access the building information point / area database in the platform, and use the Request function to retrieve the POI data and AOI data of each building in the target area; in, P n 、P a 、P t 、P l These are the name, address, category, and latitude / longitude coordinates from the POI and AOI data corresponding to the building, respectively. url Interface coding for accessing open-source geographic information databases; S13: Preprocess the acquired multi-source geographic information in a Geographic Information System (GIS) (such as ArcGIS), including operations such as data import, data visualization, unified coordinate system, and data projection.

[0037] In one embodiment, taking Town Y (approximately 3 square kilometers) in a city in China as an example, the road network information of Town Y is downloaded through the OpenStreet Map platform; based on the open API interface of Baidu / Gao De / OpenStreet Map, the Request function is used to obtain all building POI and AOI data of Town Y, including 1413 POI data and 1100 AOI data. Finally, the various data are imported into ArcGIS to obtain the building and road network distribution map of Town Y. Figure 2 The data is clipped according to the boundaries and then unified into the WGS84 Web Mercator projection coordinate system.

[0038] S2: Based on the preprocessed multi-source geographic information, the buildings in the target area are classified, the population is estimated, and the road traffic intensity index is defined, generating building density raster datasets, population density raster data, and traffic intensity index raster data of different functional categories. Specifically, this step can be divided into three parts: (1) Determine the functional characteristics of urban buildings through the POIs corresponding to the buildings, and then generate a grid of building densities for different functions. In one embodiment, the following steps may be included: (1.1) Based on the preprocessed POI data, reclassify the buildings in the target area according to their functional types; Specifically, the original category labels in the POI data include major, medium, and minor categories, which are numerous and have similar functions. Referring to the "Urban Land Use Classification Standard and Planning Construction Land Use Standard" (GB 50137-2011), and considering that different building functions represent different urban activities (urban activities are closely related to surface source pollution), the POIs corresponding to buildings in the target area are reclassified into different functional types (Category I). The reclassification result of a building POI directly serves as the functional attribute of its corresponding building; that is, when a building POI belongs to a certain functional type, its corresponding building simultaneously belongs to that functional category. In this embodiment, the POI categories are reclassified into 6 functional types, and the specific classification results are shown in Table 1.

[0039] Table 1 Reclassification of Y Town POI Data (1.2) Based on the reclassified functional categories and the AOI data, all buildings in the target area are clustered using a geographic information system, and density is calculated for each functional category to obtain functional building density raster datasets for different building categories. Specifically, kernel density analysis is applied in GIS software (such as ArcGIS) to cluster and calculate the density of information points for different functional buildings, resulting in I different categories of functional building density raster datasets. In this embodiment, six different functional categories of building density distribution raster datasets are calculated and generated, such as... Figure 3 As shown, (a)-(f) respectively display the building density distribution of POI_xzkj, POI_gyqy, POI_gwfw, POI_jmsh, POI_jtdl, and POI_cyfw, with units of buildings / km². 2 .

[0040] (2) Estimating the population of buildings in the target area and generating a population density raster for the target area. In one embodiment, this may include the following steps: (2.1) Determine the per capita usable area index A and the building usable area coefficient for different buildings according to their functional categories. β ; In practical implementation, the indicators and coefficients need to be determined in conjunction with relevant urban planning and construction data. In one embodiment, the per capita indicators and coefficients are all taken from current national standards and the 2020 National Population Census Bulletin, including the "Standards for Classification of Urban Land Use and Standards for Planning and Construction Land Use" (GB 50137-2011), the "Control Indicators for Construction Land Use of Industrial Projects" (Natural Resources Development

[2023] No. 5), the "Design Standard for Catering Buildings" JGJ 64-2017, the "Design Code for Shop Buildings" JGJ 48-2014, the "Design Standard for Office Buildings" JGJ / T67-2019, and the data requirements of the Seventh National Population Census of the National Bureau of Statistics. In practical implementation, they can be obtained based on the actual current standards and bulletins, the latest publicly available data, etc.

[0041] (2.2) Obtain the building area S from the AOI data, and combine it with the per capita usable area index A and the building usable area coefficient for different buildings. β Estimate the population size P in different buildings: in, P For the number of people living in the building; S The building area; β This refers to the building's usable area coefficient. A This refers to the per capita usable floor area of ​​buildings. (2.3) Assign the building population number to the building centroid, use kernel density analysis to calculate the population density of the target area, and obtain the population density raster of the target area; this step can be achieved by GIS software (such as ArcGIS).

[0042] In this embodiment, a suitable per capita usable area index is selected based on urban construction and industry design standards and documents. A and coefficients β。 The population count for each building was further calculated using a formula. Simultaneously, the centroids of all building outlines were extracted using ArcGIS, and the population count for each building was assigned to its corresponding centroid. Finally, based on kernel density analysis, a population density raster was calculated, as shown below. Figure 4 As shown, the unit is cap·km -2 .

[0043] (3) Define a road traffic intensity index for the target area and generate traffic intensity index raster data. In one embodiment, this may include the following steps: (3.1) Based on the preprocessed traffic network data, the roads in the target area are divided into several categories; Referring to the classification standards in the "Urban Comprehensive Transportation System Planning Standard" (GBT51328-2018) and combining the road information contained in the road network data, the road network of the target area is divided into five major categories: urban expressways, urban arterial roads, urban secondary arterial roads, urban branch roads, and internal roads of buildings. Note that some urban areas have not planned to build urban expressways, and the specific number of categories N depends on the actual situation. (3.2) Obtain the single-lane design capacity of different types of roads, and calculate the traffic intensity index of different types of roads in combination with the number of lanes of different types of roads; By consulting the "Code for Design of Urban Road Engineering" CJJ 37-2012 (2016 edition) and the "Standard for Urban Comprehensive Transportation System Planning" (GBT51328-2018), the single-lane design capacity q of different road categories is obtained. Simultaneously, combined with the road lane information n from the road network data, the traffic intensity index W for different road categories is defined. in, W i For the first i Traffic intensity index of road type; q i For the first i Single-lane design capacity for similar roads; n i For the first i The number of lanes in each road category. In this embodiment, the road network within the area is classified into four different categories: urban arterial roads, urban secondary arterial roads, urban branch roads, and internal building roads. Based on the formula described above, the traffic intensity index for each road category is calculated, as shown in Table 2.

[0044] The classification is based on road capacity, taking into account different traffic flows (road design capacity per hour) to reflect the impact of urban traffic activities on non-point source pollution.

[0045] Table 2 Classification of Urban Roads in Y Town (3.3) Assign the traffic intensity index to each road in the target area, apply the line density analysis method to perform cluster calculation, and obtain the road traffic intensity index raster data of the target area; In this embodiment, based on urban road design and planning, and considering actual lane conditions, the regional road network is divided into five categories. A traffic intensity index is calculated for each category to differentiate the traffic congestion levels of different roads. 。 Furthermore, using ArcGIS's assignment function, corresponding indices are assigned to road elements. Using line density analysis tools, traffic intensity index rasters are calculated, such as... Figure 5 As shown, the unit is km·km -2 .

[0046] S3: During the rainless period, obtain sampling samples from several sampling points in the target area, and calculate the non-point source pollution load data corresponding to each sampling sample under different previous rainless days; It should be noted that the antecedent dry days (ADD) in this scheme focus on the cumulative duration of the dry period, and are primarily used to characterize the accumulation cycle of non-point source pollutants on the land surface, i.e., the duration of continuous no effective rainfall before sampling. Generally, for the same location, non-point source pollution accumulates as the number of dry days increases, exhibiting a rapid initial accumulation followed by a slower accumulation, and then tending to reach dynamic stability after reaching a certain value. Most literature generally shows that this cycle is between 7 and 10 days. Specifically, the acquisition of the non-point source pollution load data may include the following steps. S31: Arrangement J There are 15 sampling points. These sampling points are representative, covering areas surrounding different types of buildings, areas with different population densities, and different types of roads. The number of sampling points can be set according to actual conditions; in this embodiment, 15 points are arranged. Figure 6 As shown.

[0047] S32: During the rainless period, i.e., the time from the end of the previous rainfall to the start of the next rainfall, sampling points are... L Secondary surface dust samples were collected by sampling along the roadside curb at sampling points, 1-2 meters away. 2 Street dust was collected in the grid area using a vacuum cleaner and a fine brush. Three parallel samples were taken and combined to form the final sample for that sampling point. The number of rainless days prior to the sampling was also recorded. ADD The number of sampling times can be set according to the actual situation; in this embodiment, it is set to 6 times.

[0048] S33: Regarding the obtained U=J*L For each street dust sample, calculate the total surface source pollution load. M and the pollutant load data. M p : in, M It is the total load of surface source pollution; m It refers to the quality of street dust samples collected from different locations; a These are the sampling areas at different locations; C p It is the first in the dust of the street P The concentration of the pollutants; M pIt is the first in the dust of the street P The load of various pollutants; In this embodiment, 15 sampling points were set up, covering different types of functional buildings, areas with different population densities, and roads with different traffic flows. A total of 6 street dust samplings were conducted, obtaining 90 street dust samples. The mass and concentration of five pollutant components in each sample were measured, including total phosphorus (TP), total nitrogen (TN), heavy metals chromium (Cr), zinc (Zn), and lead (Pb). The pollution load data for each sample was calculated using formulas: M, M... TP M TN M Cr M Zn M Pb .

[0049] S4: Based on the building density raster dataset, population density raster data, and traffic intensity index raster data, extract the building density, population density, and traffic intensity index corresponding to each sample. Combine the number of rainless days in the previous period corresponding to each sample with the input feature vector, and use the non-point source pollution load data corresponding to each sample as the output target vector to construct a feature-target dataset. Specifically, the construction of the feature-target dataset can be accomplished through the following steps: (1) Based on raster data, obtain each sample. j Different functional building densities B Population density PD Traffic intensity WD Simultaneously, obtain the number of rainless days in the preceding period corresponding to each sample. ADD Each sample is constructed using the following formula. j eigenvectors SF j : in, It is a sample j of I The sampling point contains density information for buildings with different functions. There may be buildings of different functional types around the sampling point. If a certain type does not exist, it is assigned a value of 0. PD j It is a sample j Population density information; WD j It is a sample j Traffic intensity information; ADD j It is a sample j The corresponding number of rainless days in the preceding period; SF j Sampling points jeigenvectors; (2) Obtain each sample j After identifying the features, construct a feature information matrix for all samples. FM , as input variables; (3) Obtain each sample j Data on total non-point source pollution load and loads of different types of pollutants were used to construct the data for each sample. j Target pollution data vector ST j : in, ST j It is a sample j The target vector; It is a sample j Total surface source pollution load; It is a sample j The P Different pollutant loads, P ≥1; (4) Establish the target information matrix TM , as an output variable; In this embodiment, there are a total of 90 samples. Feature data: each sample has 9 features, including the number of rainless days in the preceding period, population density, different functional building densities, and road traffic intensity index. A 90×9 feature matrix is ​​constructed using the NumPy library in Python. Area source pollution target data: each sample has 6 predicted targets, including cumulative total load and five types of pollution loads. A 90×6 target matrix is ​​constructed using the NumPy library in Python.

[0050] S5: Normalize the data in the feature-target dataset, train the input-output mapping relationship using machine learning algorithms, and establish an urban non-point source pollution prediction model; Specifically, the sample data was divided into datasets and normalized. A feature input-target output mapping model was constructed based on the XGboost machine learning algorithm (Extreme Gradient Boosting, an improvement on gradient boosting decision trees, which improves prediction performance by progressively building a series of weak learners and combining them into a strong learner). Simultaneously, an optimization algorithm combined with an iterative method of cross-validation was used to tune the model parameters, and the model performance was tested using test data. Finally, an urban non-point source pollution prediction model based on extreme gradient boosting was established.

[0051] In this embodiment, the statsmodels library is used to perform multicollinearity testing on the original sample dataset to prevent collinearity of variables from affecting model stability. Then, the `train_test_split` function from the python-sklearn library is used to split the dataset, with 80% used as the training set and 20% as the test set. The `StandardScaler` function is used to standardize the data using Z-scores. The xgboost library is then used to construct the XGboost regression algorithm. XGBoost uses the default squared loss (the square of the difference between the true value and the algorithm's predicted value) as the training loss. Gradient descent is used to optimize the training loss. Initial values ​​for hyperparameters such as tree depth, minimum threshold for node splitting, and learning rate are defined, and their ranges are given. Finally, the GridSearchCV grid optimization function, combined with 5-fold cross-validation (KFold), is used for automatic hyperparameter optimization. The mean mean squared error (MSE) is used as the evaluation metric, and the algorithm is iterated 5000 times to obtain the optimal hyperparameters. On the other hand, to verify the performance of the above optimization algorithm, it is compared with two commonly used machine learning algorithms, multiple linear regression and support vector machine, under the same dataset and evaluation process. The results show that the model based on the XGboost algorithm outperforms other comparative algorithms in prediction performance.

[0052] Furthermore, by applying optimal hyperparameters, an urban non-point source pollution prediction model based on extreme gradient boosting is established, and the model's performance on the training and test sets is obtained, such as... Figure 7 Figures (a)-(f) show the model for M and M, respectively. TP M TN M Cr M Pb M Zn The model's prediction performance is as follows. Overall, the average NSE of the model on the training set of various pollution load data can reach about 0.9, while the average NSE on the test set can reach about 0.6, indicating that the model has good generalization ability and can accurately quantify the spatial distribution of ground source pollution loads under different rainless periods in complex urban environments.

[0053] S6: Spatial gridding is performed on the area to be simulated, and the building density, population density, traffic intensity, and number of rainless days in the previous period are extracted for each grid. The urban non-point source pollution prediction model is used to make predictions to obtain the non-point source pollution load data of the area to be simulated. Given the characteristics of any location in an area, extrapolation prediction using a model yields the spatiotemporal quantification results of non-point source pollution load. Specifically, the process of predicting non-point source pollution load in an unknown area includes: Spatial gridding of the target area; Extract information such as building density, population density, and traffic intensity for each grid, given the number of rainless days in the previous period; An urban non-point source pollution prediction model based on extreme gradient enhancement is applied to extrapolate and predict the total amount of non-point source pollution and the load of different pollutants in the target area.

[0054] In this embodiment, the entire spatial area of ​​Y town is gridded using ArcGIS's Fishnet tool at a resolution of 50m. Further, using area statistics tools, based on the spatial feature raster data obtained in S2, geospatial features such as population and functional buildings for each grid are extracted and input into the machine learning model constructed in S5. This quickly predicts the non-point source pollution load for all grids under different antecedent dry days (ADD=2&7), including cumulative total load and pollutant load. Finally, using ArcGIS's raster tools, accurate and detailed simulation results of urban non-point source pollution under different antecedent dry days are obtained for the entire area. Figure 8 , Figure 9 As shown, (a)-(f) respectively represent M (g / m³) under different preceding rainless days. 2 M TP (mg / m 2 M TN (mg / m 2 M Cr (mg / m 2 M Pb (mg / m 2 M Zn (mg / m 2 The distribution of non-point source pollution is shown in the figure. In summary, the urban non-point source pollution quantification method based on open-source geographic information and machine learning proposed in this invention estimates the population corresponding to buildings and calculates population density using open-source data such as POI / AOI. It also characterizes the density of buildings with different functions and the traffic intensity of the road network, making up for the shortcomings of traditional cumulative formulas that ignore spatial influences. It solves the long-standing complexity and uncertainty of the process of quantifying surface source load, and achieves accurate and refined results. It provides important technical support for solving problems such as risk identification and source control of urban non-point source pollution, and has practical engineering application value.

[0055] Corresponding to the aforementioned embodiments of the refined simulation method for urban non-point source pollution combining open source information and machine learning, this application also provides embodiments of a refined simulation device for urban non-point source pollution combining open source information and machine learning.

[0056] Figure 10 This is a block diagram illustrating a refined simulation device for urban non-point source pollution that combines open-source information with machine learning, according to an exemplary embodiment. (Refer to...) Figure 10 The device may include: The open-source information acquisition module 21 is used to acquire multi-source geographic information of the target area through the open-source geographic information platform and perform data preprocessing. The multi-source geographic information includes traffic network data, POI data and AOI data corresponding to buildings. The raster data generation module 22 is used to classify buildings, estimate population, and define road traffic intensity index in the target area based on preprocessed multi-source geographic information, and generate building density raster datasets, population density raster data, and traffic intensity index raster data of different functional categories. The non-point source pollution load calculation module 23 is used to acquire sampling samples from several sampling points in the target area during a rainless period and calculate the non-point source pollution load data corresponding to each sampling sample under different previous rainless days. The dataset construction module 24 is used to extract the building density, population density, and traffic intensity index corresponding to each sample based on the raster data obtained by the raster data generation module. It combines the number of rainless days in the previous period corresponding to each sample as the feature input variable and the non-point source pollution load data corresponding to each sample as the output variable, thereby constructing the feature-target dataset. Model training module 25 is used to normalize the data in the feature-target dataset, train the input-output mapping relationship using machine learning algorithms, and establish an urban non-point source pollution prediction model. Prediction module 26 is used to spatially grid the area to be simulated, extract the building density, population density, traffic intensity, and number of rainless days in the previous period for each grid, and use the urban non-point source pollution prediction model to make predictions to obtain the non-point source pollution load data of the area to be simulated.

[0057] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0058] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0059] Accordingly, this application also provides a computer program product, including a computer program / instruction that, when executed by a processor, implements the above-described method for refined simulation of urban non-point source pollution that combines open-source information with machine learning.

[0060] Accordingly, this application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; and, when the one or more programs are executed by the one or more processors, causing the one or more processors to implement the above-described method for refined simulation of urban non-point source pollution combining open-source information and machine learning. Figure 11 The diagram shown is a hardware structure diagram of any data processing-capable device, including an urban non-point source pollution fine simulation device combining open-source information and machine learning, provided by an embodiment of the present invention. (Except for...) Figure 11 In addition to the processor, memory, and network interface shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0061] Accordingly, this application also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the refined simulation method for urban non-point source pollution combining open-source information and machine learning as described above. The computer-readable storage medium can be an internal storage unit of any data-processing device as described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data-processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data-processing device, and can also be used to temporarily store data that has been output or will be output.

[0062] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

Claims

1. A refined simulation method for urban non-point source pollution combining open-source information and machine learning, characterized in that, include: S1: Obtain multi-source geographic information of the target area through an open-source geographic information platform and perform data preprocessing. The multi-source geographic information includes traffic network data, POI data and AOI data corresponding to buildings. S2: Based on the preprocessed multi-source geographic information, the buildings in the target area are classified, the population is estimated, and the road traffic intensity index is defined, generating building density raster datasets, population density raster data, and traffic intensity index raster data of different functional categories. S3: During the rainless period, obtain sampling samples from several sampling points in the target area, and calculate the non-point source pollution load data corresponding to each sampling sample under different previous rainless days; S4: Based on the raster data obtained in step S2, extract the building density, population density, and traffic intensity index corresponding to each sample, and combine them with the number of rainless days in the previous period corresponding to each sample as feature input variables, and use the non-point source pollution load data corresponding to each sample as output variables, thereby constructing a feature-target dataset; S5: Normalize the data in the feature-target dataset, train the input-output mapping relationship using machine learning algorithms, and establish an urban non-point source pollution prediction model; S6: Spatial gridding is performed on the area to be simulated, and the building density, population density, traffic intensity, and number of rainless days in the previous period are extracted for each grid. The urban non-point source pollution prediction model is used to make predictions to obtain the non-point source pollution load data of the area to be simulated.

2. The method according to claim 1, characterized in that, Step S1 includes: Download the urban transportation network of the target area from the open platform; Based on the geographic information platform program interface, access the building information point / information surface database in the platform, and use the Request function to retrieve the POI and AOI data of each building in the target area; ， in, P n 、P a 、P t 、P l These are the name, address, category, and latitude / longitude coordinates from the POI and AOI data corresponding to the building, respectively. url Interface coding for accessing open-source geographic information databases; The acquired multi-source geographic information in the geographic information system is preprocessed, including data import, data visualization, unified coordinate system, and data projection.

3. The method according to claim 1, characterized in that, In step S2, the buildings in the target area are classified, and different functional building density grids are generated, specifically as follows: Based on the preprocessed POI data, the buildings in the target area are reclassified; Based on the reclassified functional categories and the AOI data, all buildings in the target area are clustered, and density is calculated for each functional category to obtain a building density raster dataset for different functional categories.

4. The method according to claim 1, characterized in that, In step S2, population estimation is performed on the buildings in the target area to generate population density raster data, specifically as follows: Based on the building function category, determine the per capita usable area index A and the building usable area coefficient for different buildings. β ; The building area S is obtained from the AOI data, combined with different per capita usable area indices A and building usable area coefficients. β Estimate the population size P in different buildings: ， in, P For the number of people living in the building; S The building area; β This refers to the building's usable area coefficient. A This refers to the per capita usable floor area of ​​buildings. The population count of a building is assigned to its centroid, and kernel density analysis is used to calculate the population density of the target area, resulting in a population density raster for the target area.

5. The method according to claim 1, characterized in that, In step S2, a road traffic intensity index is defined for the target area, and traffic intensity index raster data is generated, including: Based on the preprocessed traffic network data, the roads in the target area are divided into several categories; Obtain the single-lane design capacity of different road types, and calculate the traffic intensity index of different road types by combining the number of lanes of different road types; The traffic intensity index is assigned to each road in the target area, and the line density analysis method is applied to perform cluster calculation to obtain the road traffic intensity index raster data of the target area.

6. The method according to claim 1, characterized in that, In step S3, the sample is obtained in the following way: Arrangement J Sampling points located around different types of buildings, in areas with different population densities, and on different types of roads; During the rainless period, that is, the time from the end of one rainfall to the start of the next, sampling points are... L Secondary surface dust sample collection; For the obtained U=J*L For each street dust sample, calculate the total surface source pollution load. M and the pollutant load data. M p : ， ， in, M It is the total load of surface source pollution; m It refers to the quality of street dust samples collected from different locations; a These are the sampling areas at different locations; C p It is the first in the dust of the street P The concentration of the pollutants; M p It is the first in the dust of the street P The load of various pollutants.

7. The method according to claim 1, characterized in that, In step S4, the construction of the feature-target dataset can be completed through the following steps: Based on raster data, each sample is obtained. j Different functional building densities B Population density PD Traffic intensity WD Obtain the number of rainless days in the preceding period corresponding to each sample. ADD Construct each sample j eigenvectors SF j : ， ， in, It is a sample j of I Density information for buildings with different functions; PD j It is a sample j Population density information; WD j It is a sample j Traffic intensity information; ADD j It is a sample j The corresponding number of rainless days in the preceding period; Based on each sample j Based on the feature vectors, construct the feature information matrix of all samples. FM ; ， Get each sample j Data on total non-point source pollution load and loads of different types of pollutants were used to construct a sample for each pollutant. j Target vector ST j : ， in, It is a sample j Total surface source pollution load; It is a sample j The P Pollutant load, P ≥1; Based on each sample j Target vector, establish target information matrix TM ; 。 8. A refined simulation device for urban non-point source pollution that combines open-source information and machine learning, characterized in that, include: The open-source information acquisition module is used to acquire multi-source geographic information of the target area through an open-source geographic information platform and perform data preprocessing. The multi-source geographic information includes traffic network data, POI data and AOI data corresponding to buildings. The raster data generation module is used to classify buildings, estimate population, and define road traffic intensity index in the target area based on preprocessed multi-source geographic information, and generate building density raster datasets, population density raster data, and traffic intensity index raster data of different functional categories. The non-point source pollution load calculation module is used to acquire sampling samples from several sampling points in the target area during rainless periods and calculate the non-point source pollution load data corresponding to each sampling sample under different prior rainless days. The dataset construction module is used to extract the building density, population density, and traffic intensity index corresponding to each sample based on the raster data obtained by the raster data generation module. Combined with the number of rainless days in the previous period corresponding to each sample, it is used as the feature input variable, and the non-point source pollution load data corresponding to each sample is used as the output variable, thereby constructing the feature-target dataset. The model training module is used to normalize the data in the feature-target dataset, train the input-output mapping relationship using machine learning algorithms, and establish an urban non-point source pollution prediction model. The prediction module is used to spatially grid the area to be simulated, extract the building density, population density, traffic intensity, and number of rainless days in the preceding period for each grid, and use the urban non-point source pollution prediction model to make predictions to obtain the non-point source pollution load data of the area to be simulated.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium storing computer instructions thereon, characterized in that, When executed by the processor, this instruction implements the steps of the method as described in any one of claims 1-7.

Images