Calculation method of agricultural pollution load into water bodies based on data analysis

Abstract

The present application relates to the field of non-point source pollution analysis and management technology, and specifically discloses a method for calculating the load of agricultural pollution sources entering water bodies based on data analysis. After calculating the pollutant influencing factors of multiple dimensions based on a variety of basic data of the area to be studied, it further introduces data analysis technology based on deep learning to semantically embed and encode the attribute descriptions and numerical values ​​of each pollutant influencing factor, so as to achieve a deep understanding of each pollutant influencing factor. Then, by performing semantic correlation interaction analysis on each pollutant influencing factor, the causal relationship and synergistic effect between the pollutant influencing factors are excavated, so as to more accurately estimate the number of water inflows and realize the accurate calculation of the load of agricultural pollution sources entering water bodies. In this way, the correlation between the various pollutant influencing factors can be comprehensively considered, the calculation accuracy of the load of agricultural pollution sources entering water bodies can be improved, and more scientific decision-making support can be provided for pollution control.

Description

Technical Field

[0001] The present application relates to the technical field of non-point source pollution analysis and management, and more specifically, to a method for calculating the load of agricultural pollution sources entering a water body based on data analysis. Background Art

[0002] With the rapid development of agricultural production, the use of pollutants such as fertilizers, pesticides, and livestock and poultry manure has increased significantly. These substances enter nearby water bodies through surface runoff, drainage ditches, and underground seepage, leading to serious non-point source pollution problems. Agricultural non-point source pollution has now become one of the main reasons for the deterioration of water environment quality in many regions. Therefore, quantifying non-point source pollution in the basin to determine the key control areas of pollution, and then accurately controlling and controlling non-point source pollution, has become an important task in current environmental protection work.

[0003] In the prior art, there are already a variety of methods for calculating the load of agricultural pollution sources entering the water body. For example, the invention patent with the publication number CN113011993A proposes a method for calculating the load of agricultural pollution sources entering the water body based on standard data. According to the precipitation and topographic data of the area to be studied, the area to be studied is divided into several watershed divisions, and for each watershed division, multiple influencing factors related to pollutants are calculated, such as rainfall driving factor, topographic driving factor, surface runoff factor, groundwater runoff factor and plant interception factor, etc. After standardizing each influencing factor, the water system number of each watershed division is obtained by cumulative multiplication, and then further combined with the agricultural source pollutant emissions in the area to be studied, the water body load of agricultural pollution sources in each watershed division is calculated.

[0004] However, although existing technologies can provide relatively reliable estimation results on a larger scale, they mainly rely on traditional statistical models and empirical formulas. When dealing with multidimensional complex environmental variables, they fail to give sufficient consideration to the interactions and synergistic effects among the factors affecting the entry of multidimensional pollutants into water. It is difficult to capture the nonlinear relationships and potential causal relationships between different factors, which may lead to inaccurate estimation results.

[0005] Therefore, an optimized method for calculating the load of agricultural pollution sources into water bodies based on data analysis is desired. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a method for calculating the load of agricultural pollution sources entering water bodies based on data analysis. After calculating the pollutant influencing factors of multiple dimensions based on a variety of basic data of the area to be studied, it further introduces data analysis technology based on deep learning to semantically embed and encode the attribute descriptions and numerical values ​​of each pollutant influencing factor, so as to achieve a deep understanding of each pollutant influencing factor. Then, by performing semantic correlation interaction analysis on each pollutant influencing factor, the causal relationship and synergistic effect between each pollutant influencing factor are excavated, so as to more accurately estimate the number of water inflows and realize accurate calculation of the load of agricultural pollution sources entering water bodies. In this way, the correlation between the various pollutant influencing factors can be comprehensively considered, and the calculation accuracy of the load of agricultural pollution sources entering water bodies can be effectively improved, providing more scientific decision-making support for environmental protection and pollution control.

[0007] Accordingly, according to one aspect of the present application, a method for calculating the load of agricultural pollution sources entering water bodies based on data analysis is provided, which includes: obtaining precipitation, water quality, DEM, land use, soil texture, soil type and socioeconomic data of the area to be studied; dividing the area to be studied into several watershed divisions according to the precipitation and topographic data of the area to be studied; calculating the pollutant impact factor of each watershed division based on the data of the area to be studied; and calculating the water inflow data of each watershed division based on the pollutant impact factor of each watershed division. ; Based on the number of water inlet systems in each basin division and the discharge of agricultural pollutants S, calculate the load of agricultural pollution sources into water bodies .

[0008] In the above-mentioned method for calculating the water body load of agricultural pollution sources based on data analysis, the water body load of each watershed partition is calculated based on the pollutant impact factor of each watershed partition. ,include:

[0009] Extracting the attribute description and value of each pollutant influencing factor;

[0010] Performing semantic embedding coding on the attribute description and the value of each pollutant influencing factor to obtain a set of pollutant influencing factor attribute-value embedding coding vectors;

[0011] Performing semantic differentiation enhancement based on the essential condition description on the set of pollutant impact factor attribute-value embedding coding vectors to obtain a set of distinguishable enhanced pollutant impact factor attribute-value embedding coding vectors;

[0012] Performing inter-influencing factor association coding on the set of distinguishable enhanced pollutant influencing factor attribute-value embedded coding vectors to obtain a pollutant influencing factor multi-dimensional interactive coding vector;

[0013] The water inlet system number is estimated based on the multi-dimensional interactive coding vector of the pollutant influencing factor to obtain the water inlet system number .

[0014] Compared with the prior art, the method for calculating the load of agricultural pollution sources entering water bodies based on data analysis provided in this application, after calculating the pollutant influencing factors of multiple dimensions based on a variety of basic data of the area to be studied, further introduces data analysis technology based on deep learning to semantically embed and encode the attribute descriptions and numerical values ​​of each pollutant influencing factor, so as to achieve a deep understanding of each pollutant influencing factor, and then, through the semantic association interaction analysis of each pollutant influencing factor, the causal relationship and synergistic effect between each pollutant influencing factor are excavated, so as to more accurately estimate the number of water inflows and realize the accurate calculation of the load of agricultural pollution sources entering water bodies. In this way, the correlation between the various pollutant influencing factors can be comprehensively considered, and the calculation accuracy of the load of agricultural pollution sources entering water bodies can be effectively improved, providing more scientific decision-making support for environmental protection and pollution control. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0016] Figure 1 This is a flow chart of a method for calculating the load of agricultural pollution sources entering a water body based on data analysis according to an embodiment of the present application.

[0017] Figure 2 This is a data flow diagram of a method for calculating the load of agricultural pollution sources entering a water body based on data analysis according to an embodiment of the present application.

[0018] Figure 3 This is a flowchart of step S120 in the method for calculating the load of agricultural pollution sources entering a water body based on data analysis according to an embodiment of the present application.

[0019] Figure 4 This is a flowchart of step S130 in the method for calculating the load of agricultural pollution sources entering a water body based on data analysis according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0021] As mentioned in the background technology, patent CN113011993A proposes a method for calculating the water body load of agricultural pollution sources based on standard data, which includes: obtaining precipitation, water quality, DEM, land use, soil texture, soil type and socio-economic data of the area to be studied; dividing the area to be studied into several watershed divisions according to the precipitation and topographic data of the area to be studied; calculating the pollutant impact factor of each watershed division according to the data of the area to be studied; calculating the water body load of each watershed division based on the pollutant impact factor of each watershed division; ; Based on the number of water inlet systems in each basin division and the discharge of agricultural pollutants S, calculate the load of agricultural pollution sources into water bodies .

[0022] Specifically, when calculating the load of agricultural pollution sources into water bodies, first of all, obtaining precipitation, water quality, DEM (digital elevation model), land use, soil texture, soil type and socioeconomic data of the area to be studied is the basis of the entire accounting method. Precipitation data is of great significance for analyzing the generation and migration process of agricultural non-point source pollution. Precipitation is one of the key factors in the formation of surface runoff. Its intensity, frequency and distribution will directly affect the amount and path of pollutants entering the water body with runoff. For example, in areas with large and concentrated rainfall, a large amount of surface runoff may be triggered, thereby accelerating the transport of pollutants to the water body. In order to obtain reliable precipitation information, it can be obtained through long-term observation records provided by the National Meteorological Administration or other authoritative agencies. In addition, considering the uncertainty that may be caused by local climate change, high-resolution satellite remote sensing images will also be used to supplement the areas that are not covered by ground stations. Through the spatiotemporal interpolation algorithm, a precipitation distribution map covering the entire study area can be constructed.

[0023] Water quality data can reflect the actual degree of water pollution and is an important reference for assessing the impact of agricultural pollution sources on water bodies. Through monitoring and analysis of water quality, we can understand the types and concentrations of pollutants in the current water body, and then compare and verify with the subsequent calculated water body load. Traditional water quality monitoring mainly relies on fixed-point sampling and analysis, that is, setting fixed monitoring points in important water bodies such as rivers and lakes, regularly collecting water samples and sending them to laboratories for testing. With the development of science and technology, more and more automatic online monitoring equipment is now used to transmit data in real time, which improves monitoring efficiency and accuracy. In order to ensure the consistency and comparability of data, all participating units should follow a unified standard operating procedure (SOP) and ensure that instrument calibration and maintenance are in place. In addition, considering the impact of seasonal changes and human activities, it is also necessary to design a reasonable sampling frequency and time series to capture the trend of dynamic changes in water quality.

[0024] DEM data plays a core role in watershed zoning and terrain analysis, and can accurately present the topographic and geomorphic characteristics of the study area, such as slope, aspect and other information. Based on this information, the flow direction and convergence area of ​​the water flow can be further analyzed, providing a key basis for the reasonable division of watershed zoning. At present, the most commonly used DEM sources include the global SRTM (Shuttle Radar Topography Mission) dataset and the more accurate LiDAR (Light Detection and Ranging) scanning results. The latter is particularly suitable for small-scale or specific-purpose research projects because it can provide centimeter-level spatial resolution. After selecting a suitable DEM product, the next step is to import it into the GIS software for preprocessing, such as resampling, clipping boundaries, and removing outliers to ensure that it meets the actual application requirements.

[0025] Land use data reflects the distribution of different land types in the study area, such as farmland, forestland, grassland, construction land, etc. Different land use types have significant differences in the generation, interception and transmission of pollutants. For example, farmland may produce a large amount of fertilizer and pesticide pollutants during fertilization and irrigation, while forestland and grassland may have a certain interception and buffering effect on pollutants. At present, remote sensing technology has become the main means of obtaining land use information over a large area, especially with the help of multispectral and hyperspectral sensors, various vegetation types and non-vegetated surfaces can be distinguished. However, since the classification accuracy is limited by image quality and interpretation algorithms, it is often necessary to combine auxiliary data for verification and correction. For example, use high-resolution optical images or drone aerial photos to refine the classification results; refer to local chronicles, planning documents and other literature to confirm the existence of certain special land uses. The final land use map should be highly current and consistent, so as to facilitate subsequent spatial overlay analysis with other elements.

[0026] Soil texture and soil type data are crucial for analyzing the migration and transformation of pollutants in soil. Soils of different textures and types have different adsorption and desorption capabilities for pollutants, which affects the rate and amount of pollutants entering water bodies. For example, soils with a high clay content may have a strong adsorption capacity for certain pollutants, reducing their migration into water bodies. In order to accurately obtain soil texture and soil type data, a detailed soil property database must be established, covering multiple physical and chemical indicators from the topsoil to the deep soil. On the one hand, macroscopic information can be quickly obtained by consulting existing soil census reports; on the other hand, it is necessary to carry out highly targeted field surveys, take samples and test at key locations to ensure that microscopic details are not overlooked. It is worth noting that some new detection technologies that have emerged in recent years, such as near-infrared spectrometers (NIRS) and X-ray fluorescence spectrometers (XRF), have greatly shortened on-site measurement time and improved analysis accuracy.

[0027] In addition to natural conditions, the level of social and economic development also has a profound impact on agricultural production methods and environmental protection awareness. Specifically, social and economic data include population size, poultry and livestock breeding numbers, fertilizer and pesticide application data, etc., which are closely related to agricultural pollution sources. Population growth will lead to an increase in demand for agricultural products, which may lead to an increase in the use of fertilizers and pesticides in agricultural production; an increase in the number of poultry and livestock breeding will produce more pollutants such as livestock and poultry manure. Therefore, before formulating prevention and control strategies, it is necessary to collect first-hand materials on the living conditions of local residents. This can be done through questionnaires, interviews, etc., but it is also necessary to make full use of official statistical data, such as various indicators under the national economic accounting system.

[0028] Since precipitation and topography are key factors in determining water flow paths and watershed formation, precipitation forms surface runoff under the guidance of topography and converges in different topographic areas to form watersheds. Therefore, by dividing the study area into multiple watershed sub-areas, the generation, migration and entry of pollutants into water bodies in each sub-area can be analyzed in more detail, improving the accuracy and pertinence of the calculation.

[0029] In the specific implementation process, first, GIS software is used to load and integrate all spatial data layers to create a multidimensional analysis platform. On this platform, the spatial relationship between different variables can be intuitively observed, and a preliminary zoning plan can be formulated accordingly. A common practice is to use the river system as the main framework and divide the entire study area into several independent small watersheds based on the terrain undulation characteristics. Doing so not only helps to simplify the problem, but also better reflects the actual operating mechanism of the hydrological process.

[0030] When determining the boundaries of small watersheds, the principle of natural drainage should be followed, that is, to ensure that all precipitation in each small watershed can converge to the same exit point along the surface or underground path - usually the confluence of the main tributaries or main stream of the river. In order to achieve this goal, the flow direction algorithm is often used, which can automatically identify the direction of water flow based on the height difference in the DEM and track it grid by grid until the lowest point is found. On this basis, the flow accumulation algorithm is used to calculate the maximum amount of water that each grid unit can contribute, and then determine the location of the main river channel. The watershed boundary obtained in this way has a high accuracy and is consistent with the real hydrological phenomenon.

[0031] However, zoning based solely on topographic features may not fully meet the needs of agricultural pollution research. Considering the spatiotemporal heterogeneity of precipitation patterns and their important impact on the migration and transformation of pollutants, it is necessary to further subdivide small watersheds on this basis. Specifically, the existing small watersheds can be divided secondary according to indicators such as multi-year average precipitation, precipitation zoning index, and wetness coefficient to form more refined functional zones. For example, in areas with abundant precipitation, more surface runoff and faster nutrient loss may occur; in arid and semi-arid areas, attention should be paid to irrigation water management and soil moisture conservation measures. By conducting a detailed analysis of the characteristics of different functional zones, guidance can be provided for targeted pollution prevention and control strategies.

[0032] Finally, in order to verify the validity of the zoning results, field surveys and monitoring activities need to be carried out to collect sufficient evidence to prove whether the proposed zoning reasonably reflects the actual situation. For example, fixed monitoring points can be set up in each functional area, and water quality samples can be collected regularly and their chemical composition analyzed to verify the consistency of pollutant sources with expectations. If obvious differences are found within certain zones, the boundaries should be adjusted or the input parameters should be re-evaluated in a timely manner. In addition, a dynamic update mechanism should be established to continuously optimize and improve the zoning system as the time and spatial scales change, so that it always remains up to date and consistent with the actual situation.

[0033] Next, calculating the pollutant impact factors of each watershed division is one of the key steps of this accounting method. Pollutant impact factors mainly include rainfall driving factors, topographic driving factors, surface runoff factors, groundwater runoff factors and plant interception factors, which reflect the driving mechanism and influencing factors of pollutants entering the water body from different aspects. Specifically, the calculation principle of rainfall driving factors is based on the comparison of rainfall conditions in different watershed divisions with the rainfall conditions in the entire study area, as well as the rainfall conditions in different grids within the watershed divisions.

[0034] The calculation basis of the topographic driving factor is the comparison between the average slope in the watershed partition and the average slope of the calculation watershed grid. For example, in areas with larger slopes, the water flow speed is faster, which may cause pollutants to enter the water body faster, and the topographic driving factor will be correspondingly larger. When the pollutant is total nitrogen or ammonia nitrogen, the calculation of the surface runoff driving factor is based on the SCS-CN model. This model takes into account the influence of factors such as rainfall, soil anterior moisture and watershed underlying surface characteristics on surface runoff. First, after determining the soil hydrological group according to the soil stable infiltration rate and soil texture, the SCS manual is consulted to obtain the CN value of the general wetness under different land use conditions, and then the soil anterior moisture condition (AMC) is divided into 3 levels according to the rainfall in the previous 5 days. The CN value suitable for the study area is further calculated, so as to calculate the surface runoff driving factor. For example, when the soil water storage capacity is low (the CN value is large) and the rainfall is large, more surface runoff may be generated, and the corresponding surface runoff driving factor will also be large.

[0035] When the pollutant is total phosphorus, the calculation of the surface runoff driving factor not only considers surface runoff, but also involves soil loss. This is because the migration of total phosphorus in the soil is closely related to the loss of soil particles. The soil loss is calculated through the comprehensive calculation of annual rainfall erosion factor, soil erodibility factor, slope length factor, slope factor, vegetation and management factor, and soil and water conservation measures factor. For example, in areas with low vegetation coverage, large slopes, and strong rainfall erosion, the soil loss may be large, resulting in a large surface runoff driving factor for total phosphorus.

[0036] The calculation of groundwater runoff factor is based on data such as annual rainfall, non-flood season rainfall and standard runoff curve number of the basin division, reflecting the possibility of pollutants entering the water body through groundwater runoff. The actual infiltration amount of pollutants is approximately described by analyzing the relationship between the infiltration capacity of soil moisture along the soil profile and the intensity of pollutant load. For example, in areas with high annual rainfall and relatively low non-flood season rainfall, groundwater runoff may be more active and the groundwater runoff factor will be correspondingly larger.

[0037] The plant interception factor mainly considers the interception efficiency and average slope of forest and grassland in the watershed division. Different plant types have different interception efficiencies for agricultural source pollutants. Relevant data is obtained by querying the interception efficiency table and calculated in combination with the average slope. Generally, only the interception factor values ​​of forest and grassland are calculated, and other land use types are treated as having no interception efficiency. For example, due to the dense vegetation, forest land has a strong ability to intercept pollutants, and its interception factor may be large, which can effectively reduce the amount of pollutants entering the water body.

[0038] After obtaining the pollutant impact factors of each river basin partition, it can be used to calculate the number of water inflow systems of each river basin partition. In the prior art, it mainly relies on traditional statistical models and empirical formulas to perform mathematical operations on each pollutant impact factor to obtain the number of water inflow systems. Although it can provide relatively reliable estimation results on a larger scale, when dealing with multidimensional complex environmental variables, it does not consider the interactions and synergistic effects between multidimensional pollutant inflow influencing factors. It is difficult to capture the nonlinear relationship and potential causal relationship between different factors, which may lead to inaccurate estimation results. Based on this, the present application proposes a method for calculating the load of agricultural pollution sources entering the water body based on data analysis. After calculating the pollutant impact factors of multiple dimensions based on a variety of basic data of the area to be studied, it further introduces data analysis technology based on deep learning to semantically embed and encode the attribute descriptions and values ​​of each pollutant impact factor, so as to achieve a deep understanding of each pollutant impact factor. Then, by performing semantic correlation interaction analysis on each pollutant impact factor, the causal relationship and synergistic effect between each pollutant impact factor are excavated, so as to more accurately estimate the number of water inflow systems and realize accurate calculation of the load of agricultural pollution sources entering the water body. In this way, the correlation between the factors affecting the entry of various pollutants into water can be comprehensively considered, the calculation accuracy of the load of agricultural pollution sources into water bodies can be effectively improved, and more scientific decision-making support can be provided for environmental protection and pollution control.

[0039] Figure 1 This is a flow chart of a method for calculating the load of agricultural pollution sources entering a water body based on data analysis according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the method for calculating the agricultural pollution source water body load based on data analysis according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the method for calculating the load of agricultural pollution sources entering water bodies based on data analysis according to the embodiment of the present application, the method includes the following steps: S110, extracting the attribute descriptions and values ​​of the various pollutant influencing factors; S120, semantically embedding the attribute descriptions and values ​​of the various pollutant influencing factors to obtain a set of pollutant influencing factor attribute-value embedding coding vectors; S130, semantically distinguishing and strengthening the set of pollutant influencing factor attribute-value embedding coding vectors based on the essential condition description to obtain a set of distinguishable and strengthened pollutant influencing factor attribute-value embedding coding vectors; S140, performing correlation coding between influencing factors on the set of distinguishable and strengthened pollutant influencing factor attribute-value embedding coding vectors to obtain a multidimensional interactive coding vector of pollutant influencing factors; S150, estimating the water influencing system number based on the multidimensional interactive coding vector of the pollutant influencing factors to obtain the water influencing system number .

[0040] In the above-mentioned method for calculating the load of agricultural pollution sources entering water bodies based on data analysis, the step S110 extracts the attribute description and value of each pollutant influencing factor. It should be understood that the attribute description and value of the pollutant influencing factor are the basic data for estimating the load of agricultural pollution sources entering water bodies, including information such as rainfall drive, terrain drive, surface runoff, groundwater runoff and plant interception. Specifically, the attribute description of the pollutant influencing factor can clearly define the essential characteristics and action mechanism of each influencing factor. For example, the attribute description of the rainfall driving factor will involve the influence of the time distribution and intensity change of rainfall on the migration of pollutants; the attribute description of the terrain driving factor focuses on the relationship between terrain features such as terrain slope and height difference and the movement of pollutants; the attribute description of the surface runoff factor focuses on the influence of factors such as surface cover type and soil type on the migration of pollutants; the attribute description of the groundwater runoff factor involves the influence of groundwater level, soil permeability and other factors on the migration path of pollutants; the attribute description of the plant interception factor includes the description of the interception capacity of pollutants such as vegetation type and coverage. By extracting these attribute descriptions and values, we can provide rich input information for the subsequent deep learning model. The values ​​are quantitative representations of the degree of influence of each pollutant influencing factor on the influencing of pollutants into the water. Based on this, by extracting the attribute descriptions and values ​​of each pollutant influencing factor, a comprehensive description of the impact of pollutants entering the water is formed, which can provide an original and reliable data basis for subsequent precise calculations.

[0041] In the above method for calculating the agricultural pollution source water body load based on data analysis, the step S120 performs semantic embedding coding on the attribute description and value of each pollutant impact factor to obtain a set of pollutant impact factor attribute-value embedding coding vectors. Figure 3 FIG. 1 is a flow chart of step S120 in the method for calculating the agricultural pollution source water body load based on data analysis according to an embodiment of the present application. Figure 3 As shown, the step S120 includes: S121, performing semantic embedding coding on the attribute descriptions of the various pollutant influencing factors to obtain a set of pollutant influencing factor attribute semantic embedding coding vectors; S122, adding the numerical values ​​of the various pollutant influencing factors to the tail of the various pollutant influencing factor attribute semantic embedding coding vectors to obtain a set of pollutant influencing factor attribute-numerical embedding coding vectors.

[0042] Here, considering that the attribute description of the pollutant influencing factor usually exists in the form of text, in order to facilitate efficient processing and analysis by the computer, the present application further performs semantic embedding coding on it and converts it into a numerical vector representation. In a specific example of the present application, the step S121 includes: using a semantic embedding encoder based on the Bert model to perform semantic embedding coding on the attribute description of each pollutant influencing factor to obtain a set of semantic embedding coding vectors of the pollutant influencing factor attributes. It should be known that the Bert model is pre-trained based on a large-scale text corpus and has learned rich language knowledge and semantic representations. When semantically encoding the attribute description of the pollutant influencing factor, the Bert model can accurately capture the vocabulary, grammatical structure and semantic relationship in the attribute description based on the semantic representation learned in the pre-training process, understand its key semantic information, and convert it into a fixed-length vector representation in a high-dimensional semantic space, generating a set of semantic embedding coding vectors of pollutant influencing factor attributes. Through semantic embedding coding processing, the potential associations and similarities between the attributes of different influencing factors can be better explored in the high-dimensional semantic space, thereby providing strong support for further calculation of water inflow system coefficients.

[0043] Next, in order to comprehensively consider the qualitative and quantitative information of each pollutant influencing factor, the numerical value of each pollutant influencing factor is further added to the tail of the semantic embedding coding vector of each pollutant influencing factor attribute to generate a set of pollutant influencing factor attribute-numerical embedding coding vectors. In this way, the organic combination of pollutant influencing factor attribute information and numerical information can be achieved, thereby more completely describing the comprehensive characteristics of each pollutant influencing factor and providing more comprehensive and accurate data support for the calculation of water inflow system coefficients.

[0044] In the above-mentioned method for calculating the load of agricultural pollution sources entering water bodies based on data analysis, the step S130 performs semantic differentiation enhancement based on the essential condition description on the set of pollutant influencing factor attribute-numerical embedding coding vectors to obtain a set of distinguishable enhanced pollutant influencing factor attribute-numerical embedding coding vectors. That is, in order to improve the distinguishability and sensitivity of the model to the characteristics of different pollutant influencing factors, the present application further performs feature differentiation enhancement processing on the attribute-numerical embedding coding vectors of each pollutant influencing factor, and performs personalized feature enhancement on the attribute-numerical embedding coding vectors of each pollutant influencing factor based on the unique characteristics of the attribute-numerical embedding coding vectors of each pollutant influencing factor, so that the model can more clearly identify and distinguish the unique attributes of different pollutant influencing factors, and more clearly reveal the unique mechanism of action of each pollutant influencing factor on the impact of pollutants entering water.

[0045] Figure 4FIG. 1 is a flow chart of step S130 in the method for calculating the agricultural pollution source water body load based on data analysis according to an embodiment of the present application. Figure 4 As shown, the step S130 includes: S131, extracting the essential characteristics of the set of pollutant influencing factor attribute-numerical value embedded coding vectors to obtain the pollutant influencing factor attribute-numerical value semantic field essential characteristic vector; S132, based on the pollutant influencing factor attribute-numerical value semantic field essential characteristic vector, performing semantic differentiation and enhancement processing on each pollutant influencing factor attribute-numerical value embedded coding vector in the set of pollutant influencing factor attribute-numerical value embedded coding vectors to obtain the set of distinguishable enhanced pollutant influencing factor attribute-numerical value embedded coding vectors.

[0046] Specifically, the step S131 includes: first, performing field mapping on each pollutant impact factor attribute-value embedded coding vector in the set of pollutant impact factor attribute-value embedded coding vectors to obtain a set of pollutant impact factor attribute-value semantic field feature vectors, which is expressed as follows:

[0047]

[0048] in, The first one in the set of attribute-value embedding coding vectors of the pollutant impact factor is represented The pollutant impact factor attribute-value embedding coding vector, and Represent the first linear mapping matrix and the second linear mapping matrix respectively, Indicates the The corresponding pollutant influencing factor attribute-value semantic field feature vector. That is, firstly, the field mapping operation is performed on each pollutant influencing factor attribute-value embedding coding vector, and it is mapped to a unified field space, so that the inherent semantic association between the attribute-value features of each pollutant influencing factor can be better reflected in the unified field space.

[0049] Next, the field depth factor of each pollutant impact factor attribute-numerical semantic field feature vector in the set of pollutant impact factor attribute-numerical semantic field feature vectors is calculated to obtain a set of pollutant impact factor attribute-numerical semantic field depth factors, which is expressed by the formula:

[0050]

[0051] in, represents the L2 norm of the vector, represents the logarithmic function with base 2, is the field depth measurement function, Indicates the The corresponding pollutant impact factor attribute-numerical semantic field depth factor. That is, by further calculating the field depth factor of each pollutant impact factor attribute-numerical semantic field feature vector obtained after mapping, the information density and feature importance in the field space are measured.

[0052] Then, based on the set of pollutant impact factor attributes-numerical semantic field depth factors, the field essential characteristics of the set of pollutant impact factor attributes-numerical semantic field feature vectors are calculated to obtain the pollutant impact factor attributes-numerical semantic field essential feature vectors. More specifically, after the set of pollutant impact factor attributes-numerical semantic field depth factors is normalized, it is used as a set of weight coefficients to perform weighted aggregation on the set of pollutant impact factor attributes-numerical semantic field feature vectors to obtain the pollutant impact factor attributes-numerical semantic field essential feature vectors, which is expressed as:

[0053]

[0054]

[0055] in, represents the natural exponential function with base e, Indicates the The corresponding normalized pollutant impact factor attribute - numerical semantic field depth factor, Represents the essential feature vector of the pollutant influencing factor attribute-numerical semantic field. That is, based on the field depth factor of each pollutant influencing factor attribute-numerical semantic field feature vector, the field effect normalization mechanism is used to convert it into a weight coefficient, and each pollutant influencing factor attribute-numerical semantic field feature vector is weightedly aggregated by position, so as to extract a semantic feature representation that comprehensively reflects the overall basic attributes of each pollutant influencing factor, that is, the pollutant influencing factor attribute-numerical semantic field essential feature vector.

[0056] Specifically, the step S132 includes: inputting the pollutant influencing factor attribute-numerical semantic field essential feature vector and the pollutant influencing factor attribute-numerical embedding coding vector into a semantic difference feature extraction network based on a sigmoid activation function to obtain the pollutant influencing factor attribute-numerical semantic difference feature vector; calculating the position point multiplication between the pollutant influencing factor attribute-numerical semantic difference feature vector and the pollutant influencing factor attribute-numerical embedding coding vector to obtain the distinguishable enhanced pollutant influencing factor attribute-numerical embedding coding vector, which is expressed as follows:

[0057]

[0058] in, and are two learnable weight parameter matrices of the semantic difference feature extraction network, represents the bias vector, represents the sigmoid activation function, It means point multiplication by position. Indicates the The corresponding distinguishable and enhanced pollutant influencing factor attribute-value embedded coding vector. That is, taking the pollutant influencing factor attribute-value semantic field essential feature vector as the benchmark, a neural network model is further introduced to learn the difference information of each pollutant influencing factor attribute-value embedded coding vector relative to the pollutant influencing factor attribute-value semantic field essential feature vector, and through element-by-element distinguishable and enhanced processing, the unique information in each pollutant influencing factor attribute-value embedded coding vector is specifically amplified, thereby ensuring that in the subsequent calculation process of pollutant inflow water system numbers, the differences in different pollutant influencing factors can be effectively distinguished, and the unique contribution of each influencing factor can be captured more accurately.

[0059] In the above-mentioned method for calculating the load of agricultural pollution sources into water bodies based on data analysis, the step S140 performs inter-influence factor correlation coding on the set of distinguishable and enhanced pollutant influencing factor attribute-value embedded coding vectors to obtain a multi-dimensional interactive coding vector of pollutant influencing factors. In a specific example of the present application, the step S140 includes: inputting the set of distinguishable and enhanced pollutant influencing factor attribute-value embedded coding vectors into an inter-influence factor correlation encoder based on a multi-layer perceptron model to obtain a multi-dimensional interactive coding vector of the pollutant influencing factor. It should be understood that in the actual environment, the various pollutant influencing factors influence and interact with each other. For example, rainfall driving factors may affect the size and change of surface runoff factors, and terrain driving factors will change the flow path of rainfall and surface runoff, etc. Therefore, in order to further capture the multi-dimensional interactions between the various pollutant influencing factors, the present application further uses a multi-layer perceptron model to construct an inter-influence factor correlation encoder to perform correlation coding on the set of distinguishable and enhanced pollutant influencing factor attribute-value embedded coding vectors. It should be known that the multi-layer perceptron model has powerful nonlinear mapping capabilities. By comprehensively considering the attribute-value embedding coding vectors of multiple distinguishable and enhanced pollutant influencing factors, it can effectively capture the synergy and mutual constraints between various pollutant influencing factors, thereby encoding and generating multi-dimensional interactive coding vectors of pollutant influencing factors, revealing the complex mechanism of the impact of pollutants entering water at a higher level.

[0060] In the above-mentioned method for calculating the agricultural pollution source water body load based on data analysis, the step S150 estimates the water body load based on the multi-dimensional interactive coding vector of the pollutant impact factor to obtain the water body load In a specific example of the present application, the step S150 includes: inputting the multi-dimensional interactive coding vector of the pollutant impact factor into the water inlet system number estimation module based on the decoder to obtain the water inlet system number . It should be understood that after the above processing, the multi-dimensional interactive coding vector of the pollutant influencing factor integrates the interaction effects between multiple pollutant influencing factors. The decoder model uses deep learning technology to analyze it layer by layer to learn the comprehensive mechanism of each pollutant's influence on the process of pollutants entering the water, and finally estimates the water inlet and outlet system number λ. The water inlet system number λ not only reflects the independent effect of a single pollutant influencing factor, but also fully considers the interaction between multiple pollutant influencing factors, thereby providing a more comprehensive and accurate quantitative indicator for the prediction and evaluation of the impact of pollutants entering the water.

[0061] The multi-dimensional interactive coding vector of the pollutant impact factor is input into the water inlet system number estimation module based on the decoder to obtain the water inlet system number include:

[0062] Determine the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the multidimensional interactive coding vector of the pollutant impact factor, and divide the median eigenvalue by the difference between the maximum eigenvalue and the minimum eigenvalue to obtain the multidimensional interactive coding distribution probability value of the pollutant impact factor, that is, ,in, is the probability value of the multi-dimensional interactive coding distribution of pollutant impact factors, , and are respectively the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the multidimensional interactive coding vector of the pollutant influencing factor;

[0063] The multidimensional interactive coding vector of the pollutant influencing factor is normalized to the maximum value to obtain the multidimensional interactive coding probability feature vector of the pollutant influencing factor ;

[0064] The power function of each eigenvalue of the multidimensional interactive coding probability eigenvector of the pollutant impact factor is calculated by subtracting the multidimensional interactive coding distribution probability value of the pollutant impact factor from one to obtain the first pollutant impact factor multidimensional interactive coding convergence eigenvector ,in, is the convergent eigenvector of the multidimensional interactive coding of the first pollutant influencing factor, It represents the power function of calculating each eigenvalue in the eigenvector by subtracting the multi-dimensional interactive coding distribution probability value of the pollutant impact factor from one as the exponent;

[0065] The power function of each eigenvalue of the point difference eigenvector between the unit eigenvector and the multidimensional interactive coding probability eigenvector of the pollutant impact factor is calculated with the multidimensional interactive coding distribution probability value of the pollutant impact factor as an exponent to obtain the second pollutant impact factor multidimensional interactive coding convergence eigenvector , Represents the convergence eigenvector of multidimensional interactive coding of the second pollutant influencing factor, is the unit eigenvector, Indicates the spread, It means that the power function of each eigenvalue in the eigenvector is calculated with the probability value of the multi-dimensional interactive coding distribution of the pollutant impact factor as the exponent;

[0066] The multidimensional interactive coding probability feature vector of the pollutant influencing factor is multiplied by the difference between one minus the multidimensional interactive coding distribution probability value of the pollutant influencing factor to obtain the first pollutant influencing factor multidimensional interactive coding limited feature vector ,in, Define the feature vector for multi-dimensional interaction coding of the first pollutant influencing factor;

[0067] The point difference feature vector is multiplied by the multidimensional interactive coding distribution probability value of the pollutant influencing factor to obtain a second pollutant influencing factor multidimensional interactive coding limited feature vector. ,in, Define the feature vector for multi-dimensional interaction coding of the second pollutant influencing factor;

[0068] After multiplying the first pollutant influencing factor multidimensional interactive coding convergence feature vector by the second pollutant influencing factor multidimensional interactive coding convergence feature vector, multiplying the first pollutant influencing factor multidimensional interactive coding limit feature vector by the second pollutant influencing factor multidimensional interactive coding limit feature vector to obtain an optimized pollutant influencing factor multidimensional interactive coding vector ,in, It is the multi-dimensional interaction coding vector of the optimized pollutant impact factors;

[0069] The optimized multi-dimensional interactive coding vector of pollutant impact factors is input into the water inlet system number estimation module based on the decoder to obtain the water inlet system number. .

[0070] Here, considering that each pollutant influencing factor attribute-value embedding coding vector in the set of the pollutant influencing factor attribute-value embedding coding vector respectively represents the attribute description of each pollutant influencing factor and the low-dimensional embedding splicing features of the value, after feature distinguishability enhancement, the feature significance of each pollutant influencing factor attribute-value embedding coding vector is highlighted, but this will also make it so that when the input is based on the multi-layer perceptron model-based influencing factor association coding, the feature association capture between each pollutant influencing factor attribute-value embedding coding vector is suppressed, resulting in the multi-dimensional interaction coding vector of the pollutant influencing factor having the probability convergence and divergence of the multi-layer fully connected aggregation coding, thereby affecting the water influencing system number obtained by the water influencing system number estimation module based on the decoder. accuracy.

[0071] Based on this, by using the cross entropy form power series of the distribution analytical probability of the multidimensional interactive coding vector of the pollutant influencing factors as the probability distribution convergence limit, the quasi-probability convergence approximation of the multidimensional interactive coding vector of the pollutant influencing factors is performed on the basis of the combination of the feature set distribution and the probability density distribution of the multidimensional interactive coding vector of the pollutant influencing factors, so as to avoid the conservative convergence control of the fatal potential divergence of the multidimensional interactive coding vector of the pollutant influencing factors, further use the probability distribution cross entropy of the multidimensional interactive coding vector of the pollutant influencing factors as the objective function to guide the limited recovery strategy, so as to achieve the common agility of the convergence of the probability convergence and divergence of the feature set of the multidimensional interactive coding vector of the pollutant influencing factors to the probability density distribution space, and improve the water inflow system number obtained by the water inflow system number estimation module based on the decoder of the multidimensional interactive coding vector of the pollutant influencing factors. accuracy.

[0072] Finally, based on the number of water inlet systems and the amount of pollutants discharged from agricultural sources in each watershed division, the amount of agricultural pollution load entering the water body can be calculated. The amount of pollutants discharged from agricultural sources is calculated based on the population density and per capita pollutant emission intensity in the study area. By multiplying the calculated number of water inlet systems with the amount of pollutant emissions, the amount of agricultural pollution load entering the water body can be obtained. This calculation method comprehensively considers the amount of pollutants generated and the difficulty of entering the water body, and can more accurately evaluate the pollution load of agricultural pollution sources on the water body.

[0073] In summary, according to the embodiment of the present application, the method for calculating the load of agricultural pollution sources entering water bodies based on data analysis is explained. After calculating the pollutant influencing factors of multiple dimensions based on a variety of basic data of the area to be studied, it further introduces data analysis technology based on deep learning to semantically embed and encode the attribute descriptions and numerical values ​​of each pollutant influencing factor, so as to achieve a deep understanding of each pollutant influencing factor. Then, by performing semantic correlation interaction analysis on each pollutant influencing factor, the causal relationship and synergistic effect between each pollutant influencing factor are excavated, so as to more accurately estimate the number of water inflows and realize the accurate calculation of the load of agricultural pollution sources entering water bodies. In this way, the correlation between various pollutant influencing factors can be comprehensively considered, and the calculation accuracy of the load of agricultural pollution sources entering water bodies can be effectively improved, providing more scientific decision-making support for environmental protection and pollution control.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for calculating the load of agricultural pollution sources into water bodies based on data analysis, including: Obtain precipitation, water quality, DEM, land use, soil texture, soil type and socioeconomic data for the area to be studied; According to the precipitation and topographic data of the area to be studied, the area to be studied is divided into several watershed divisions; according to the data of the area to be studied, the pollutant impact factor of each watershed division is calculated; The number of water inflow systems in each watershed partition is calculated based on the pollutant impact factor of each watershed partition. ; Based on the number of water inlet systems in each basin division and the discharge of agricultural pollutants S, calculate the load of agricultural pollution sources into water bodies , characterized in that the number of water inflow systems of each watershed partition is calculated based on the pollutant impact factor of each watershed partition ,include: Extract the attribute description and value of each pollutant influencing factor; Performing semantic embedding coding on the attribute description and the value of each pollutant influencing factor to obtain a set of pollutant influencing factor attribute-value embedding coding vectors; The set of pollutant influencing factor attribute-value embedded coding vectors is subjected to semantic differentiation enhancement based on the essential condition description to obtain a set of distinguishable enhanced pollutant influencing factor attribute-value embedded coding vectors, wherein a field mapping operation is performed on each pollutant influencing factor attribute-value embedded coding vector to map it to a unified field space, and the field depth factor of each pollutant influencing factor attribute-value semantic field feature vector obtained after mapping is calculated, and after the set of pollutant influencing factor attribute-value semantic field depth factors is normalized, it is used as a set of weight coefficients to perform weighted aggregation on the set of pollutant influencing factor attribute-value semantic field feature vectors to obtain the pollutant influencing factor attribute-value semantic field essential feature vector, and the pollutant influencing factor attribute-value semantic field essential feature vector and the pollutant influencing factor attribute-value embedded coding vector are input into a semantic difference feature extraction network based on a sigmoid activation function for feature extraction to obtain a distinguishable enhanced pollutant influencing factor attribute-value embedded coding vector; Performing inter-influence factor association coding on the set of distinguishable and enhanced pollutant influencing factor attribute-value embedded coding vectors to obtain a pollutant influencing factor multidimensional interactive coding vector, wherein the set of distinguishable and enhanced pollutant influencing factor attribute-value embedded coding vectors is input into an inter-influence factor association encoder based on a multi-layer perceptron model to obtain the pollutant influencing factor multidimensional interactive coding vector; The water inlet system number is estimated based on the multi-dimensional interactive coding vector of the pollutant influencing factor to obtain the water inlet system number , wherein the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the multidimensional interaction coding vector of the pollutant influencing factor are determined, and the median eigenvalue is divided by the difference between the maximum eigenvalue and the minimum eigenvalue to obtain the multidimensional interaction coding distribution probability value of the pollutant influencing factor, the multidimensional interaction coding vector of the pollutant influencing factor is normalized to the maximum value to obtain the multidimensional interaction coding probability eigenvector of the pollutant influencing factor, and the first pollutant influencing factor multidimensional interaction coding convergence eigenvector, the second pollutant influencing factor multidimensional interaction coding convergence eigenvector, and the second pollutant influencing factor multidimensional interaction coding convergence eigenvector are obtained based on the multidimensional interaction coding distribution probability value of the pollutant influencing factor and the multidimensional interaction coding probability eigenvector of the pollutant influencing factor. The coding convergence feature vector, the first pollutant influencing factor multidimensional interactive coding limit feature vector and the second pollutant influencing factor multidimensional interactive coding limit feature vector are obtained by multiplying the first pollutant influencing factor multidimensional interactive coding convergence feature vector with the second pollutant influencing factor multidimensional interactive coding convergence feature vector, and then performing point multiplication with the first pollutant influencing factor multidimensional interactive coding limit feature vector and the second pollutant influencing factor multidimensional interactive coding limit feature vector to obtain an optimized pollutant influencing factor multidimensional interactive coding vector, and the optimized pollutant influencing factor multidimensional interactive coding vector is input into a decoder-based water influencing system number estimation module to obtain the water influencing system number. .

2. The method for calculating the agricultural pollution source water load based on data analysis according to claim 1 is characterized in that: The attribute description and the value of each pollutant influencing factor are semantically embedded and encoded to obtain a set of pollutant influencing factor attribute-value embedded encoding vectors, including: Performing semantic embedding coding on the attribute descriptions of the respective pollutant influencing factors to obtain a set of semantic embedding coding vectors of the pollutant influencing factor attributes; The numerical values ​​of the pollutant impact factors are added to the tail of the semantic embedding coding vector of each pollutant impact factor attribute to obtain a set of the pollutant impact factor attribute-numerical embedding coding vectors.

3. The method for calculating the agricultural pollution source water load based on data analysis according to claim 2 is characterized in that: The attribute descriptions of the various pollutant impact factors are semantically embedded and encoded to obtain a set of pollutant impact factor attribute semantic embedded encoding vectors, including: The attribute descriptions of the various pollutant influencing factors are semantically embedded and encoded using a semantic embedding encoder based on the Bert model to obtain a set of semantic embedding encoding vectors of the pollutant influencing factor attributes.

4. The method for calculating the agricultural pollution source water load based on data analysis according to claim 3 is characterized in that: The set of pollutant impact factor attribute-value embedding coding vectors is subjected to semantic differentiation enhancement based on the essential condition description to obtain a set of distinguishable enhanced pollutant impact factor attribute-value embedding coding vectors, including: Extracting the essential features of the set of pollutant impact factor attribute-value embedded coding vectors to obtain the pollutant impact factor attribute-value semantic field essential feature vector; Based on the pollutant influencing factor attribute-numerical value semantic field essential feature vector, each pollutant influencing factor attribute-numerical value embedded coding vector in the set of pollutant influencing factor attribute-numerical value embedded coding vectors is semantically differentiated and enhanced to obtain the set of distinguishable enhanced pollutant influencing factor attribute-numerical value embedded coding vectors.

Images