Intelligent tracing method and system for agricultural non-point source pollution based on knowledge graph

Abstract

The invention relates to the technical field of agricultural traceability, and discloses an agricultural non-point source pollution intelligent traceability method and system based on a knowledge graph, and the method comprises the steps: achieving the system integration of pollution source features through constructing the knowledge graph fusing multi-dimensional data, and building a sky-ground integrated monitoring network to obtain multi-scale dynamic data. An intelligent traceability mechanism is formed based on deep coupling of a knowledge graph and monitoring data, a high-precision pollution identification model is trained in combination with historical data, and real-time response and dynamic traceability of an over-standard pollution area are achieved. And verifying a traceability result through feature matching and semantic reasoning, quantitatively calculating a pollution contribution rate, and finally generating a reliable traceability conclusion containing the position, the type and the contribution rate. According to the scheme, the bottlenecks of data fragmentation, monitoring simplification, extensive analysis and the like of a traditional method are broken through, and the accuracy, timeliness and credibility of pollution traceability are remarkably improved through a series of full-chain technical systems of traceability identification, analysis and positioning.

Description

Technical Field

[0001] The present invention relates to the field of agricultural traceability technology, and more specifically, to an intelligent traceability method and system for agricultural non-point source pollution based on a knowledge graph. Background Art

[0002] With the increasing impact of agricultural non-point source pollution on farmland environmental quality, particularly from fertilizer and pesticide runoff, livestock and poultry waste, and rural domestic sewage, the need to accurately identify pollution sources and develop scientific treatment plans is becoming increasingly urgent. While traditional pollution source tracing methods have made some progress in ground-based monitoring, source investigations, and simple statistical analysis, they still lack systematic solutions to core issues such as the effective integration of multidimensional data, the identification of complex, multi-source pollution stacking, and the in-depth exploration of pollution transmission mechanisms. These challenges make it difficult to meet the generalization needs of different regions, pollution types, and spatial and temporal scales.

[0003] Therefore, how to develop an intelligent tracing method for agricultural non-point source pollution that can integrate multi-source monitoring data, deeply explore the interactive relationship between pollution source-transmission path-receiving water body, accurately capture the spatiotemporal variation characteristics of pollution, and have strong adaptability has become an urgent problem to be solved. Summary of the Invention

[0004] The present invention provides an intelligent tracing method and system for agricultural non-point source pollution based on knowledge graphs, which solves the technical problem in the existing technology of how to develop a method that can integrate multi-source monitoring data, deeply explore the interactive relationship between pollution sources, transmission paths and receiving water bodies, accurately capture the spatiotemporal variation characteristics of pollution, and have strong adaptability.

[0005] The present invention provides a method and system for intelligent tracing of agricultural non-point source pollution based on knowledge graphs, including: First, a method for intelligent tracing of agricultural non-point source pollution based on knowledge graphs, including: Collect multi-dimensional data on agricultural non-point source pollution and construct a knowledge graph of agricultural non-point source pollution; Based on the agricultural non-point source pollution knowledge graph, an integrated sky-ground monitoring network is established to obtain a monitoring data set; Deeply integrate the agricultural non-point source pollution knowledge graph with the monitoring data set to build an intelligent traceability monitoring mechanism; Based on the intelligent source tracing monitoring mechanism and historically acquired monitoring data sets, a pollution source identification model is trained and constructed; Acquire pollution content values ​​based on a pollution source identification model. When the pollution content value exceeds a preset threshold, the pollution source identification model is triggered to identify pollution sources and perform intelligent tracing analysis on the monitoring area where pollution exceeds the standard, thereby obtaining tracing analysis results. The traceability analysis results are matched and verified with the pollution source feature information in the agricultural non-point source pollution knowledge graph, the pollution source type is confirmed through feature similarity comparison and semantic relationship reasoning, and the pollution contribution rate of different agricultural non-point sources is quantitatively calculated, and finally a traceability result including the pollution source location, type, and contribution rate is generated.

[0006] Furthermore, we collected multi-dimensional data on agricultural non-point source pollution and constructed a knowledge graph of agricultural non-point source pollution, including: Collect environmental quality standard data, agricultural production history data, land use status data, and agricultural pollution source distribution data in the monitoring area to build a basic agricultural non-point source data set; Preprocessing and normalizing the agricultural non-point source basic data set to obtain a processed agricultural non-point source data set; Extract farmland entities, livestock and poultry breeding entities, rural life entities, and environmental factor entities from the processed agricultural nonpoint source data set, and identify the association relationships between the entities; Construct knowledge representation units in the form of multiple arrays to form structured agricultural non-point source pollution knowledge units; The structured knowledge units are integrated according to the logical relationship from pollution source to transmission path and then to receptor to establish an agricultural non-point source pollution knowledge network; By designing semantic node index and query relationship index, a knowledge graph of agricultural non-point source pollution that supports semantic reasoning and complex queries is constructed.

[0007] Furthermore, based on the agricultural non-point source pollution knowledge graph, an integrated sky-ground monitoring network is established to form a monitoring network data system, obtain monitoring data sets, and build an intelligent source tracing monitoring mechanism, including: Build a space-based satellite remote sensing monitoring layer and use high-resolution satellites to obtain macro-environmental information such as regional farmland distribution, crop types, and vegetation coverage; Build an airborne drone monitoring layer and deploy drones equipped with multispectral and hyperspectral sensors to obtain plot-level agricultural activity information and pollution source distribution data; Build a ground-based IoT monitoring layer, deploying water quality sensors, soil sensors, and weather stations at key river sections, farmlands, and catchment areas to collect real-time water quality, soil, and meteorological environmental parameters; Based on the knowledge graph of agricultural non-point source pollution, a multi-scale monitoring network data system is constructed, including a space-based remote sensing monitoring layer, an air-based drone monitoring layer, and a ground-based Internet of Things monitoring layer. The space-based remote sensing monitoring layer obtains large-scale vegetation coverage and crop growth information; the air-based drone monitoring layer obtains mesoscale farmland dynamic change monitoring data; and the ground-based Internet of Things monitoring layer obtains small-scale real-time and continuous farmland environmental parameter monitoring. A multi-source data spatiotemporal registration mechanism is established based on a multi-scale monitoring network data system. Using farmland plots as the basic spatial unit and the crop growth period as the temporal scale, the space-based, airborne, and ground-based monitoring data systems are spatially interpolated and temporally synchronized. This integrates monitoring data at different scales to produce a comprehensive monitoring dataset that is consistent in both time and space. Perform spatial expansion analysis on the spatial distribution characteristics of pollutants for each monitoring point in the comprehensive monitoring data set, use time series analysis to identify pollution change trends, and form analysis results that integrate multi-source data; Based on the analysis results, a directional response relationship between pollution load and multiple driving factors such as meteorological conditions, agricultural activities, and soil properties was established; By combining directional response relationships with historical meteorological data, we can obtain pollution washoff thresholds under different rainfall intensities, identify key driving factors affecting changes in agricultural non-point source pollution, and ultimately build an intelligent source tracing monitoring mechanism.

[0008] Furthermore, the intelligent traceability monitoring mechanism includes: The pollution source emission coefficients, soil nutrient balance equations, hydrological transmission parameters and key driving factors of various pollution source parameters in the agricultural non-point source pollution knowledge map are coupled to obtain coupling results, and the nonlinear adsorption and desorption processes between soil and water bodies are integrated. Based on the coupling results, identify the complex pollution pattern of multiple pollution sources to distinguish the superposition effects of different pollution sources such as farmland fertilization, livestock and poultry breeding, and domestic sewage. This will obtain the first identification result; Based on the first identification results, the pollution load response characteristics under different rainfall intensity and duration scenarios are predicted to obtain prediction results; Based on the first identification and prediction results, the migration and diffusion patterns of pollutants under complex terrain conditions are simulated, and the pollutant diffusion trend is obtained through the changes in the flow velocity of surface runoff and base flow; Analyze the pollutant diffusion law based on the pollutant diffusion trend, obtain the distribution of pollution content values ​​in the monitoring area within the preset time, evaluate the pollutant diffusion efficiency and impact range, and obtain the evaluation results; According to the evaluation results, the monitoring network layout and parameter configuration are adjusted through a multi-objective optimization algorithm to form an intelligent traceability monitoring mechanism that adapts to different agricultural regions and pollution types.

[0009] Furthermore, based on the intelligent source tracing monitoring mechanism and the historically acquired monitoring data set, a pollution source identification model is trained and constructed, including: Through an intelligent traceability monitoring mechanism, water quality data from farmland drainage outlets, livestock and poultry farms, and rural domestic sewage treatment facility outlets are monitored in real time. Historical rainfall, soil moisture content, and environmental parameters during the crop growth period are simultaneously acquired. Combined with information on land use changes, crop planting structure, and farm distribution obtained through remote sensing, a multidimensional monitoring dataset containing water quality parameters, environmental factors, and spatial attributes is constructed. Based on the relationship network in the agricultural non-point source pollution knowledge graph, the multidimensional monitoring data set is decomposed into time series to obtain decomposition results; the decomposition results include trend items reflecting the agricultural activity cycle, seasonal items reflecting seasonal pollution patterns, and random items representing sudden pollution events; Based on the decomposition results, we identify the water quality variation patterns caused by agricultural activity cycles such as fertilization season, rainfall season, and crop harvest season, and establish a temporal correlation network between pollution content values ​​and agricultural activities. Based on the temporal association network, , obtain the comprehensive index of agricultural non-point source pollution in the monitoring area, quantitatively assess the severity of the current pollution situation, establish a dynamically adjusted classification and pollution content value threshold system based on the surface water environmental quality standards and farmland irrigation water quality standards, identify abnormal patterns and mutation points in water quality data, and obtain the second identification result; among them, is the comprehensive index of agricultural non-point source pollution, is the pollution content value of the i-th type, is the corresponding standard value, is the pollution weight, is the environmental impact weight of agricultural activities, Number the pollutant types; The second recognition result is semantically matched and knowledge inference is performed with the agricultural non-point source pollution knowledge graph, and the reasoning process is further optimized through causal reasoning to establish the correlation between pollution events and pollution sources; Based on the correlation between pollution events and pollution sources, pollution events are classified, and the spatiotemporal characteristics of pollution content values ​​corresponding to pollution events are extracted. A pollution source identification model based on contrastive learning is established to accurately identify the types of pollution sources corresponding to different types of pollution events, achieving end-to-end intelligent identification from monitoring data anomalies to precise positioning of pollution sources. The abnormal pollution data obtained from real-time monitoring is semantically matched with the pollution source characteristics, transmission path relationships, and environmental impact factors in the agricultural non-point source pollution knowledge graph. Combined with environmental parameters such as current rainfall intensity, wind speed and direction, soil moisture, and the intensity of agricultural activities, the Bayesian inference method is used to comprehensively determine the cause of the abnormal event and distinguish between water quality anomalies caused by natural factors (such as rainfall erosion and concentrated drought) and human factors (such as excessive fertilization and illegal discharge). Based on the semantic reasoning capabilities and expert rule base of the agricultural non-point source pollution knowledge graph, the system automatically identifies various abnormal pollution event types, such as excessive fertilization, livestock and poultry manure leakage, and direct discharge of domestic sewage, by using pattern recognition algorithms based on the concentration levels and concentration ratios of pollutants such as total nitrogen, total phosphorus, ammonia nitrogen, and chemical oxygen demand, combined with the temporal and spatial distribution characteristics of pollution duration, impact range, and frequency of occurrence. The identified abnormal pollution events are hierarchically classified according to the type of pollution source, degree of pollution, and scope of impact. An abnormal event knowledge base is established, which includes event triggering conditions, characteristic parameter thresholds, and response and disposal measures. An intelligent pollution source identification model is constructed that integrates multi-source data perception, knowledge graph reasoning, and machine learning prediction to achieve full-process automated identification from data anomaly detection to precise positioning of pollution sources.

[0010] Furthermore, based on the pollution source identification model, a pollution content value is obtained. When the pollution content value exceeds a preset threshold, the pollution source identification model is triggered to perform pollution source identification and intelligent source tracing analysis on the monitoring area where the pollution exceeds the standard, and the source tracing analysis results are obtained, including: The pollution source identification model is used to obtain the comprehensive index of agricultural non-point source pollution in each monitoring area in real time. The dynamic pollution content threshold is set based on the regional background value and seasonal variation characteristics in the agricultural non-point source basic data set. When the pollution content value in the monitoring area exceeds the preset threshold, the pollution source identification program is automatically triggered to identify abnormal events of excessive pollution. Based on the triggered pollution source identification program, the intelligent source tracing analysis process is initiated, calling the relationship network in the agricultural non-point source pollution knowledge map. Combined with the current rainfall intensity, soil moisture content, and environmental parameters of the crop growth period, the graph reasoning algorithm is used to identify the pollution source type and spatial distribution of the pollution exceeding the standard event, and obtain the identification results; Based on the identification results, the water quality fingerprint characteristics of abnormal pollution exceeding the standard in the monitoring area are analyzed, and the concentration values ​​of key pollutants such as total nitrogen, total phosphorus, and chemical oxygen demand are extracted. The contribution rates of different concentration values ​​are obtained, and the contribution of each pollution source such as farmland runoff, livestock and poultry breeding, and domestic sewage to the water quality exceeding the standard is quantitatively evaluated; Based on the degree of contribution, the pollutant transmission path is reversed, and the migration trajectory and transmission time of pollutants from each pollution source to the monitoring point are determined through surface runoff simulation and groundwater seepage calculation; The transmission path obtained by reverse inference is matched and verified with the spatial relationship network in the agricultural non-point source pollution knowledge map, and the traceability confidence of different pollution sources is obtained through spatial overlay analysis and path consistency test; Based on the contribution degree and confidence level of the pollution source to water quality exceeding the standard, a comprehensive traceability analysis result is generated, including the pollution source type, time-varying contribution rate and traceability credibility.

[0011] Furthermore, based on the identification results, the water quality fingerprint characteristics of abnormal pollution exceeding the standard in the monitoring area are analyzed, and the concentration values ​​of key pollutants such as total nitrogen, total phosphorus, and chemical oxygen demand are extracted. The contribution rates of different concentration values ​​are obtained, and the contribution of each pollution source such as farmland runoff, livestock and poultry breeding, and domestic sewage to the water quality exceeding the standard is quantitatively evaluated, including: pass Get the concentration value; where: is the concentration value of the i-th pollutant; is the contribution coefficient of the kth pollution source; is the source component spectrum of the jth pollutant of the kth pollution source; is the residual term; is the number of pollution sources; Number the pollutant types; Number the monitoring points; Number the pollution sources; pass Obtain the pollution source contribution rate; where: is the contribution rate of the kth pollution source; by obtaining the contribution rates of different pollution sources, we can quantitatively evaluate the contribution of each pollution source to the water quality exceeding the standard, as follows: Assess the contribution rate of agricultural runoff pollution sources:

[0012] Assessment of the contribution rate of livestock and poultry breeding pollution sources:

[0013] Assessment of the contribution rate of domestic sewage pollution sources:

[0014] in, Indicates total nitrogen concentration; represents the total phosphorus concentration; represents chemical oxygen demand; 、 、 These are the contribution coefficients of three types of pollution sources: farmland runoff, livestock and poultry breeding, and domestic sewage; 、 、 They are the total nitrogen, total phosphorus and chemical oxygen demand source component spectrum values ​​of agricultural runoff pollution sources; 、 、 They are the total nitrogen, total phosphorus and chemical oxygen demand source component spectrum values ​​of livestock and poultry breeding pollution sources; 、 、 They are the source component spectrum values ​​of total nitrogen, total phosphorus and chemical oxygen demand of domestic sewage pollution sources.

[0015] Furthermore, the traceability analysis results are matched and verified with the pollution source feature information in the agricultural non-point source pollution knowledge graph. The pollution source type is confirmed through feature similarity comparison and semantic relationship reasoning, and the pollution contribution rate of different agricultural non-point sources is quantitatively calculated. Finally, the traceability results containing the pollution source location, type, and contribution rate are generated, including: Based on the results of the source tracing analysis, the key characteristic parameters of the pollution content value ratio, spatiotemporal distribution characteristics, and transmission path information of the pollution event were preliminarily matched with the standard characteristic plates of the three types of pollution sources stored in the agricultural non-point source pollution knowledge graph: farmland runoff, livestock and poultry breeding, and rural life, and a preliminary matching result was obtained; Based on the preliminary matching results, a similarity threshold is obtained, and candidate pollution source types are screened according to the similarity threshold. At the same time, the semantic relationship network in the knowledge graph is used for reasoning and verification to confirm the pollution source type in the current monitoring area; Based on the confirmed pollution source type, the spatial distribution information and attribute characteristics of the corresponding pollution source in the knowledge graph are called up. Through GIS spatial overlay analysis, the pollution transmission path is matched with the location of the pollution source in the actual geographic space. The spatial distance and transmission accessibility of each candidate pollution source to the monitoring point are calculated to determine the precise geographic coordinates of the pollution source. Combining the contribution rate of pollution sources with their precise geographic coordinates, we spatially aggregated the contribution rates of multiple pollution sources of the same type. By correcting the actual contribution of pollution sources at different distances through pollution load attenuation, we obtained the final contribution rate distribution of various agricultural non-point source pollution sources. Based on the final contribution rate allocation results, a comprehensive traceability result report is generated, which includes the precise coordinates of the pollution source, pollution source type identification, time-varying contribution rate value, and traceability confidence evaluation.

[0016] Furthermore, combining the contribution rate of pollution sources with their precise geographic coordinates, we spatially aggregated the contribution rates of multiple pollution sources of the same type. By correcting the actual contribution of pollution sources at different distances using pollution load attenuation, we obtained the final contribution rate distribution of various agricultural non-point source pollution types, including: pass Gets the spatial weights; where: is the spatial weight of the j-th monitoring point to the k-th pollution source; is the Euclidean distance from the jth monitoring point to the kth pollution source; is the distance decay exponent; is the exponential decay coefficient; is the affected area of ​​the kth pollution source; is the number of pollution sources; pass Correct the actual contribution of pollution sources at different distances; among which: is the corrected contribution rate of the kth pollution source; is the original contribution rate; is the time attenuation coefficient; is the time it takes for pollutants to be transported from the kth pollution source to the jth monitoring point; is the maximum effective transmission distance; is the spatial attenuation index; Number the monitoring points; Number the pollution source Final contribution rate distribution through Perform global optimization; Constraints: , ; in: is the number of pollution source types; is the number of the k-th pollution source; is the observed contribution rate of the kth pollution source; is the prior contribution rate based on historical data; is the regularization parameter; is the corrected contribution rate of the kth pollution source. By solving the optimal contribution rate allocation scheme, the final contribution rate distribution of various types of agricultural non-point source pollution is obtained.

[0017] In the second aspect, an intelligent tracing system for agricultural non-point source pollution based on a knowledge graph implements an intelligent tracing method for agricultural non-point source pollution based on a knowledge graph, including: Data collection module: used to collect multi-dimensional data on agricultural non-point source pollution and build a knowledge graph of agricultural non-point source pollution; Integrated monitoring module: used to establish an integrated sky-ground monitoring network based on the agricultural non-point source pollution knowledge graph and obtain a monitoring data set; Source tracing monitoring mechanism module: used to deeply integrate the agricultural non-point source pollution knowledge graph with the monitoring data set to build an intelligent source tracing monitoring mechanism; Pollution source identification module: used to train and build a pollution source identification model based on the intelligent source tracing monitoring mechanism and historically acquired monitoring data sets; Source tracing analysis module: used to obtain pollution content values ​​based on the pollution source identification model. When the pollution content value exceeds a preset threshold, the pollution source identification model is triggered to identify the pollution source and conduct intelligent tracing analysis on the monitoring area where the pollution exceeds the standard, and obtain the tracing analysis results; Source tracing result generation module: used to match and verify the source tracing analysis results with the pollution source feature information in the agricultural non-point source pollution knowledge graph, confirm the pollution source type through feature similarity comparison and semantic relationship reasoning, and quantitatively calculate the pollution contribution rate of different agricultural non-point sources, and finally generate a source tracing result including the location, type and contribution rate of the pollution source.

[0018] The beneficial effects of the present invention are as follows: by collecting multidimensional data on agricultural non-point source pollution and constructing a knowledge graph of agricultural non-point source pollution, the present invention solves the technical problems of incomplete data acquisition and incomplete knowledge system in the prior art, and realizes the systematic integration and structured representation of multi-source heterogeneous data; Based on the agricultural non-point source pollution knowledge graph, an integrated sky-ground monitoring network is established to obtain monitoring data sets, solve the technical problem of a single monitoring network, and achieve multi-scale, multi-dimensional three-dimensional monitoring coverage; Deeply integrate the agricultural non-point source pollution knowledge graph with the monitoring data set to build an intelligent traceability monitoring mechanism, solve the technical problem of the low level of intelligence of the traceability mechanism, and achieve an organic combination of knowledge-driven and data-driven; Based on the intelligent source tracing monitoring mechanism and historically acquired monitoring data sets, a pollution source identification model is trained and constructed to solve the technical problem of low pollution source identification accuracy and achieve accurate identification and classification of complex multi-source pollution; The pollution source identification model is used to obtain pollution content values. When the pollution content value exceeds a preset threshold, the pollution source identification model is triggered to identify the pollution source and conduct intelligent tracing analysis on the monitoring area where the pollution exceeds the standard. The tracing analysis results are obtained to achieve real-time response and dynamic tracing of pollution events. The traceability analysis results are matched and verified with the pollution source feature information in the agricultural non-point source pollution knowledge graph, the pollution source type is confirmed through feature similarity comparison and semantic relationship reasoning, and the pollution contribution rate of different agricultural non-point sources is quantitatively calculated. Finally, the traceability results including the location, type and contribution rate of the pollution source are generated, which solves the technical problem of the lack of traceability result verification mechanism and realizes the reliability verification of the traceability results and the quantitative evaluation of pollution contribution.

[0019] The technical solution of the present invention systematically solves the technical defects of traditional methods in data integration, monitoring coverage, intelligent analysis, precision identification, and result verification by constructing a complete knowledge graph system, establishing a three-dimensional monitoring network, integrating multi-source data information, training intelligent recognition models, realizing dynamic traceability analysis, and establishing a verification feedback mechanism, thereby significantly improving the accuracy, timeliness and reliability of agricultural non-point source pollution traceability. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of an intelligent agricultural non-point source pollution tracing method based on a knowledge graph provided in an embodiment of the present invention; Figure 2 It is a schematic diagram of a module of an intelligent tracing system for agricultural non-point source pollution based on a knowledge graph provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described in some examples may be combined in other examples.

[0022] At least one embodiment of the present invention discloses a method and system for intelligent tracing of agricultural non-point source pollution based on a knowledge graph, including: Example 1 like Figure 1 As shown in FIG, a method for intelligent tracing of agricultural non-point source pollution based on a knowledge graph includes the following steps: Step 1: Collect multidimensional data on agricultural non-point source pollution and construct a knowledge graph of agricultural non-point source pollution; Step 2: Based on the agricultural non-point source pollution knowledge graph, establish an integrated sky-ground monitoring network and obtain a monitoring data set; Step 3: Deeply integrate the agricultural non-point source pollution knowledge graph with the monitoring dataset to build an intelligent traceability monitoring mechanism; Step 4: Based on the intelligent source tracing monitoring mechanism and the historically acquired monitoring data set, a pollution source identification model is trained and constructed; Step 5: Obtaining pollution content values ​​based on the pollution source identification model. When the pollution content value exceeds a preset threshold, the pollution source identification model is triggered to identify the pollution source and perform intelligent tracing analysis on the monitoring area where the pollution exceeds the standard, thereby obtaining the tracing analysis results. Step 6: Match and verify the traceability analysis results with the pollution source feature information in the agricultural non-point source pollution knowledge graph, confirm the pollution source type through feature similarity comparison and semantic relationship reasoning, and quantitatively calculate the pollution contribution rate of different agricultural non-point sources, and finally generate a traceability result including the location, type, and contribution rate of the pollution source.

[0023] In this implementation, the multidimensional data of agricultural non-point source pollution include but are not limited to: soil physical and chemical properties data (pH value, organic matter content, nitrogen, phosphorus and potassium content), water quality monitoring data (total nitrogen, total phosphorus, chemical oxygen demand, ammonia nitrogen), meteorological data (rainfall, temperature, humidity, wind speed), agricultural production data (fertilizer amount, fertilization time, crop type, planting area), livestock and poultry breeding data (breeding scale, manure treatment method, emission volume), and geographic spatial data (topography, land use type, water system distribution).

[0024] The knowledge graph is constructed using ontology modeling, establishing an entity-relationship-attribute triple structure. Entities include pollution source entities, environmental factor entities, and receptor entities; relationships include causal, spatial, and temporal relationships; and attributes include both quantitative and qualitative attributes. The knowledge graph is stored in the Neo4j graph database, supporting complex semantic queries and reasoning.

[0025] This method is based on the principle of knowledge-driven intelligent traceability. By constructing a domain knowledge graph as a priori knowledge base and combining it with multi-source monitoring data for knowledge-enhanced machine learning, it achieves a shift from data-driven to knowledge-driven. The knowledge graph provides a causal network of pollution sources, transmission pathways, and receptors, while monitoring data provides real-time status information. The two are integrated to form an intelligent traceability system with reasoning capabilities. This achieves semantic unification and knowledge representation of multi-dimensional heterogeneous data, improving data utilization efficiency. The semantic reasoning capabilities of the knowledge graph enable processing of incomplete and uncertain monitoring data. An explainable traceability reasoning process is established, enhancing the credibility of the traceability results. Compared with traditional methods, traceability accuracy is improved and response time is shortened.

[0026] Example 2 A preferred embodiment of the present invention collects multidimensional data on agricultural non-point source pollution and constructs a knowledge graph of agricultural non-point source pollution, specifically including: Collect environmental quality standard data, agricultural production history data, land use status data, and agricultural pollution source distribution data in the monitoring area to build a basic agricultural non-point source data set; Preprocessing and normalizing the agricultural non-point source basic data set to obtain a processed agricultural non-point source data set; Extract farmland entities, livestock and poultry breeding entities, rural life entities, and environmental factor entities from the processed agricultural nonpoint source data set, and identify the association relationships between the entities; Construct knowledge representation units in the form of multiple arrays to form structured agricultural non-point source pollution knowledge units; The structured knowledge units are integrated according to the logical relationship from pollution source to transmission path and then to receptor to establish an agricultural non-point source pollution knowledge network; By storing the agricultural non-point source pollution knowledge network and designing semantic node indexes and query relationship indexes, an agricultural non-point source pollution knowledge graph that supports semantic reasoning and complex queries is constructed.

[0027] In this implementation, the preprocessing includes data cleaning, missing value filling, outlier detection, and data format standardization. Data cleaning uses the statistical 3σ criterion and boxplot method to identify outliers; missing value filling uses a method that combines time series interpolation and spatial interpolation; and normalization uses a hybrid normalization method that combines Z-score normalization and Min-Max normalization.

[0028] Entity extraction uses named entity recognition (NER) technology combined with domain dictionaries, improving and stabilizing entity recognition accuracy. Relationship identification uses a method based on dependency parsing and semantic role labeling, capable of identifying 12 relationship types, including causal relationships, inclusion relationships, and influence relationships.

[0029] The knowledge representation unit structure in the form of multiple arrays includes five core components: an entity set is used to store various entity objects related to agricultural non-point source pollution, a relationship set is used to describe various association relationships between entities, an attribute set is used to record the characteristic attribute information of entities and relationships, constraints are used to define the logical constraints and integrity rules of knowledge units, and inference rules are used to support knowledge-based automatic reasoning and decision-making processes.

[0030] This embodiment is based on ontology engineering and knowledge engineering theory, and adopts a bottom-up knowledge construction method. First, data quality is ensured through data preprocessing, then structured knowledge is extracted from unstructured data using natural language processing technology, and finally, scattered knowledge units are organized into a knowledge network with logical relationships through ontology modeling. The knowledge graph adopts the RDF triple storage mode and supports SPARQL query language for complex semantic queries. It realizes the automatic conversion from unstructured data to structured knowledge, and the efficiency of knowledge extraction is improved; a standardized agricultural non-point source pollution knowledge representation framework is established to support knowledge sharing and reuse; through semantic indexing technology, the knowledge query response time is controlled within 100ms; the scale of the knowledge graph reaches 100,000+ entities and 500,000+ relationships, covering the entire field of agricultural non-point source pollution knowledge.

[0031] Example 3 A preferred embodiment of the present invention establishes an integrated sky-ground monitoring network based on the agricultural non-point source pollution knowledge graph, forms a monitoring network data system, obtains monitoring data sets, and constructs an intelligent source tracing monitoring mechanism, including: Based on the knowledge graph of agricultural non-point source pollution, a multi-scale monitoring network data system is constructed, including a space-based remote sensing monitoring layer, an air-based drone monitoring layer, and a ground-based Internet of Things monitoring layer. The space-based remote sensing monitoring layer obtains large-scale vegetation coverage and crop growth information; the air-based drone monitoring layer obtains mesoscale farmland dynamic change monitoring data; and the ground-based Internet of Things monitoring layer obtains small-scale real-time and continuous farmland environmental parameter monitoring. A multi-source data spatiotemporal registration mechanism is established based on a multi-scale monitoring network data system. Using farmland plots as the basic spatial unit and the crop growth period as the temporal scale, the space-based, airborne, and ground-based monitoring data systems are spatially interpolated and temporally synchronized. This integrates monitoring data at different scales to produce a comprehensive monitoring dataset that is consistent in both time and space. Perform spatial expansion analysis on the spatial distribution characteristics of pollutants for each monitoring point in the comprehensive monitoring data set, use time series analysis to identify pollution change trends, and form analysis results that integrate multi-source data; Based on the analysis results, a directional response relationship between pollution load and multiple driving factors such as meteorological conditions, agricultural activities, and soil properties was established; By combining directional response relationships with historical meteorological data, we can obtain pollution washoff thresholds under different rainfall intensities, identify key driving factors affecting changes in agricultural non-point source pollution, and ultimately build an intelligent source tracing monitoring mechanism.

[0032] In this implementation, the space-based remote sensing monitoring layer uses multispectral satellite data from Landsat-8, Sentinel-2, and other satellites with a spatial resolution of 10-30 meters and a temporal resolution of 5-16 days to acquire parameters such as NDVI, LAI, and soil moisture. The airborne drone monitoring layer utilizes multi-rotor drones equipped with multispectral cameras and water quality sensors, flying at an altitude of 50-120 meters and with a spatial resolution of 0.1-1 meter, enabling on-demand mobile monitoring. The ground-based IoT monitoring layer deploys intelligent sensor nodes, including water quality sensors (pH, DO, turbidity, and conductivity), meteorological sensors (temperature and humidity, wind speed and direction, and rainfall), and soil sensors (temperature and humidity, pH, and EC values). The data collection frequency is adjustable from 1 to 60 minutes.

[0033] The spatiotemporal registration mechanism uses the first reference coordinate system (WGS84) and a unified time base (Beijing Time). Spatial registration utilizes control point matching and geometric correction, while temporal registration utilizes timestamp synchronization and data interpolation. Spatial interpolation methods include inverse distance weighted interpolation (IDW), kriging, and spline interpolation, with the optimal interpolation method adaptively selected based on data characteristics.

[0034] The key driving factor identification adopts a method combining principal component analysis (PCA), correlation analysis and regression analysis to identify key driving factors such as rainfall intensity, soil moisture content, fertilizer application amount, temperature, etc., with a high and stable contribution rate to explained variance.

[0035] This embodiment is based on multi-scale remote sensing monitoring theory and Internet of Things technology to build a "sky-air-ground" three-dimensional monitoring network. Through the multi-source data fusion algorithm, a unified expression of data of different scales and phases is achieved, and the spatiotemporal data mining technology is used to identify the pollution change pattern and driving mechanism. The intelligent traceability monitoring mechanism is based on a data-driven machine learning method to establish a nonlinear response model of pollution load and environmental factors. It has achieved a leap from point monitoring to surface monitoring, and the monitoring coverage rate has been significantly improved; through multi-scale data fusion, the monitoring accuracy and timeliness have been significantly improved; an adaptive monitoring network optimization mechanism has been established, and the monitoring cost has been effectively reduced; and early warning of pollution events has been achieved, with a high and stable warning accuracy, and the warning time is 6-12 hours in advance.

[0036] Example 4 In a preferred embodiment of the present invention, the intelligent traceability monitoring mechanism includes: The pollution source emission coefficients, soil nutrient balance equations, hydrological transmission parameters and key driving factors of various pollution source parameters in the agricultural non-point source pollution knowledge map are coupled to obtain coupling results, and the nonlinear adsorption and desorption processes between soil and water bodies are integrated. Based on the coupling results, identify the complex pollution pattern of multiple pollution sources to distinguish the superposition effects of different pollution sources such as farmland fertilization, livestock and poultry breeding, and domestic sewage. This will obtain the first identification result; Based on the first identification results, the pollution load response characteristics under different rainfall intensity and duration scenarios are predicted to obtain prediction results; Based on the first identification and prediction results, the migration and diffusion patterns of pollutants under complex terrain conditions are simulated, and the pollutant diffusion trend is obtained through the changes in the flow velocity of surface runoff and base flow; Analyze the pollutant diffusion law based on the pollutant diffusion trend, obtain the distribution of pollution content values ​​in the monitoring area within the preset time, evaluate the pollutant diffusion efficiency and impact range, and obtain the evaluation results; According to the evaluation results, the monitoring network layout and parameter configuration are adjusted through a multi-objective optimization algorithm to form an intelligent traceability monitoring mechanism that adapts to different agricultural regions and pollution types.

[0037] In this implementation, the pollution source emission coefficient is expressed as the emission intensity per unit area, and the nitrogen and phosphorus emission coefficient of farmland runoff is 5-15 kg / hm2. 2 a. The livestock and poultry farming emission coefficient is determined based on the scale of production and the method of manure treatment. The domestic sewage emission coefficient is calculated based on population density and treatment level. The soil nutrient balance equation is expressed as soil nutrient change equal to input minus output minus losses. Input includes fertilization and rainfall, output includes crop absorption and runoff loss, and losses include volatilization and leaching.

[0038] The nonlinear adsorption and desorption process adopts the Freundlich isotherm adsorption model, which states that the adsorption amount is equal to the adsorption coefficient multiplied by the nonlinear exponential power of the equilibrium concentration. The desorption process takes into account the hysteresis effect and is described by a double exponential decay model.

[0039] The composite pollution pattern recognition method uses pollution source fingerprinting technology to distinguish the characteristic spectra of different pollution sources through principal component analysis and cluster analysis. Agricultural runoff is characterized by a high nitrogen-to-phosphorus ratio and low organic matter content; livestock and poultry farming is characterized by high ammonia nitrogen content and high fecal coliform counts; and domestic sewage is characterized by a moderate COD / BOD ratio and high surfactant content.

[0040] The pollutant migration and diffusion simulation adopts the coupling of SWAT (Soil and Water Assessment Tool) model and MIKE21 hydrodynamic model, considering the influence of factors such as terrain slope, soil type, and vegetation cover on pollutant transmission.

[0041] This embodiment constructs a multi-process coupled pollutant migration and transformation model based on environmental process simulation and system dynamics theory. By combining the prior knowledge in the knowledge graph with the physical process model, knowledge-driven process simulation is achieved. The multi-objective optimization algorithm adopts a hybrid strategy of genetic algorithm (GA) and particle swarm optimization algorithm (PSO) to simultaneously optimize multiple objectives such as monitoring accuracy, cost-effectiveness, and response time. It achieves accurate simulation of pollutant migration and transformation with multi-process coupling, and the simulation accuracy is significantly improved; through the recognition of composite pollution patterns, it can accurately distinguish the superposition effect of different pollution sources, and the recognition accuracy is high and stable; an adaptive monitoring network optimization mechanism is established, which significantly improves monitoring efficiency and effectively reduces costs; and it achieves pollution load prediction under different scenarios with good prediction accuracy, providing a scientific basis for pollution prevention and control.

[0042] Example 5 A preferred embodiment of the present invention trains and constructs a pollution source identification model based on the intelligent source tracing monitoring mechanism and historically acquired monitoring data sets, including: Through an intelligent traceability monitoring mechanism, water quality data from farmland drainage outlets, livestock and poultry farms, and rural domestic sewage treatment facility outlets are monitored in real time. Historical rainfall, soil moisture content, and environmental parameters during the crop growth period are simultaneously acquired. Combined with information on land use changes, crop planting structure, and farm distribution obtained through remote sensing, a multidimensional monitoring dataset containing water quality parameters, environmental factors, and spatial attributes is constructed. Based on the relationship network in the agricultural non-point source pollution knowledge graph, the multidimensional monitoring data set is decomposed into time series to obtain decomposition results; the decomposition results include trend items reflecting the agricultural activity cycle, seasonal items reflecting seasonal pollution patterns, and random items representing sudden pollution events; Identify the water quality change patterns caused by the agricultural activity cycles of fertilization seasons, rainfall seasons, and crop harvest periods based on the decomposition results, and establish a temporal correlation network between pollution content values and farming activities; Based on the temporal correlation network, through , obtain the comprehensive index of agricultural non-point source pollution in the monitoring area, quantitatively evaluate the severity of the current pollution situation, establish a dynamically adjusted threshold system for graded and classified pollution content values in combination with surface water environmental quality standards and farmland irrigation water quality standards, identify abnormal patterns and mutation points in water quality data, and obtain the second identification result; where is the comprehensive index of agricultural non-point source pollution, is the i-th pollution content value, is the corresponding standard value, is the pollution weight, is the environmental impact weight of farming activities, is the pollutant type number; Semantically match and perform knowledge reasoning on the second identification result with the knowledge graph of agricultural non-point source pollution, and further optimize the reasoning process through causal reasoning to establish the association relationship between pollution events and pollution sources; Based on the association relationship between pollution events and pollution sources, classify pollution events, and at the same time extract the spatio-temporal characteristics of the pollution content values corresponding to pollution events, establish a pollution source identification model based on contrast learning, and accurately identify the pollution source types corresponding to different types of pollution events, so as to obtain end-to-end intelligent identification from monitoring data anomalies to accurate pollution source positioning.

[0043] In this implementation, the multi-dimensional monitoring data set includes water quality parameters (pH, DO, COD, BOD, TN, TP, NH3-N, etc.), environmental factors (temperature, humidity, rainfall, wind speed, etc.), and spatial attributes (latitude and longitude, elevation, slope, land use type, etc.). The data dimension reaches 35 dimensions, the time span covers 3-5 years, and the data volume reaches the TB level.

[0044] The time series decomposition adopts the combination of STL decomposition method and X-13-ARIMA-SEATS method, with relatively high and stable decomposition accuracy. The trend term reflects the long-term change trend, the seasonal term reflects the periodic change pattern, and the random term reflects the impact of unexpected events.

[0045] The pollutant toxicity weight pi is determined based on the toxicity equivalent factor (TEF), and the environmental impact weight Wi is determined based on the environmental risk assessment results. API index classification standard: API ≤ 50 is excellent, 50 < API ≤ 100 is slightly polluted, 100 < API ≤ 150 is moderately polluted, and API > 150 is severely polluted.

[0046] This contrastive learning model uses the SimCLR framework and data augmentation techniques to generate positive and negative sample pairs to learn feature representations of contamination events. The model architecture includes a feature extractor (ResNet-50), a projection head (MLP), and a contrastive loss function.

[0047] This embodiment builds an end-to-end pollution source identification model based on deep learning and contrastive learning theory. It separates the change signals of different scales through time series decomposition technology, establishes the correlation between pollution events and pollution sources by using the semantic reasoning ability of knowledge graph, and finally trains the identification model through contrastive learning to achieve direct mapping from original monitoring data to pollution source types. It realizes unified modeling of multi-dimensional heterogeneous data and significantly improves the efficiency of feature extraction. Through time series decomposition, it can accurately identify the pollution change patterns of different time scales with high and stable recognition accuracy. It establishes a dynamically adjusted threshold system to adapt to the changes in pollution characteristics in different regions and seasons. The pollution source identification accuracy is high and stable, which is significantly improved compared with traditional methods, and the recognition time is shortened to seconds.

[0048] Example 6 In a preferred embodiment of the present invention, a pollution concentration value is obtained based on a pollution source identification model. When the pollution concentration value exceeds a preset threshold, the pollution source identification model is triggered to perform pollution source identification and intelligent source tracing analysis on the monitoring area where pollution exceeds the standard, and the source tracing analysis results are obtained, including: The pollution source identification model is used to obtain the comprehensive index of agricultural non-point source pollution in each monitoring area in real time. The dynamic pollution content threshold is set based on the regional background value and seasonal variation characteristics in the agricultural non-point source basic data set. When the pollution content value in the monitoring area exceeds the preset threshold, the pollution source identification program is automatically triggered to identify abnormal events of excessive pollution. Based on the triggered pollution source identification program, the intelligent source tracing analysis process is initiated, calling the relationship network in the agricultural non-point source pollution knowledge map. Combined with the current rainfall intensity, soil moisture content, and environmental parameters of the crop growth period, the graph reasoning algorithm is used to identify the pollution source type and spatial distribution of the pollution exceeding the standard event, and obtain the identification results; Based on the identification results, the water quality fingerprint characteristics of abnormal pollution exceeding the standard in the monitoring area are analyzed, and the concentration values ​​of key pollutants such as total nitrogen, total phosphorus, and chemical oxygen demand are extracted. The contribution rates of different concentration values ​​are obtained, and the contribution of each pollution source such as farmland runoff, livestock and poultry breeding, and domestic sewage to the water quality exceeding the standard is quantitatively evaluated; Based on the degree of contribution, the pollutant transmission path is reversed, and the migration trajectory and transmission time of pollutants from each pollution source to the monitoring point are determined through surface runoff simulation and groundwater seepage calculation; The transmission path obtained by reverse inference is matched and verified with the spatial relationship network in the agricultural non-point source pollution knowledge map, and the traceability confidence of different pollution sources is obtained through spatial overlay analysis and path consistency test; Based on the contribution degree and confidence level of the pollution source to water quality exceeding the standard, a comprehensive traceability analysis result is generated, including the pollution source type, time-varying contribution rate and traceability credibility.

[0049] In this implementation, the dynamic thresholds are set using a quantile regression method. Based on three years of historical monitoring data, the 75th, 90th, and 95th percentiles are calculated as thresholds for light, moderate, and heavy pollution. Seasonal adjustment factors are set based on the month of the year: 1.2 for spring (March-May), 1.5 for summer (June-August), 1.1 for autumn (September-November), and 0.8 for winter (December-February).

[0050] This graph reasoning algorithm uses the Graph Attention Network (GAT) within a Graph Neural Network (GNN) to learn the relationship weights between nodes through a multi-head attention mechanism. The network structure consists of three GAT layers, each with 128 hidden units, using the ReLU activation function and Dropout regularization.

[0051] The water quality fingerprint characteristics include pollutant concentration ratio characteristics (TN / TP, COD / BOD, NH3-N / TN), ion composition characteristics (Ca 2+ Mg 2+ 、SO4 2- 、Cl - ), isotopic characteristics (δ 15 N, δ 18 O) and other characteristic parameters to form a multi-dimensional fingerprint vector.

[0052] The transmission path inversion method uses the Lagrangian particle tracking method, combined with the digital elevation model (DEM) and hydrological network analysis to calculate the transmission trajectory of pollutants in surface runoff. The groundwater seepage calculation uses the MODFLOW model, taking into account aquifer parameters and boundary conditions.

[0053] This embodiment is based on an event-driven intelligent response mechanism, which triggers the traceability analysis process through real-time monitoring data. Graph reasoning algorithms are used to perform reasoning on the knowledge graph, and the transmission path is reversed in combination with the physical process model. Finally, the confidence level of the traceability results is determined through multi-evidence fusion. The entire process realizes an automated process from anomaly detection to traceability analysis. Automatic identification and real-time response to pollution events are achieved, and the response time is significantly shortened. Dynamic threshold setting effectively reduces the false alarm rate and improves the accuracy of anomaly detection. A confidence assessment mechanism for multi-evidence fusion is established, and the credibility of the traceability results is high. Accurate reverse inference of the pollution transmission path is achieved, and the path identification accuracy is good, providing precise guidance for pollution control.

[0054] Example 7 In a preferred embodiment of the present invention, based on the identification results, the water quality fingerprint characteristics of abnormal pollution exceeding the standard in the monitoring area are analyzed, and the concentration values ​​of key pollutants such as total nitrogen, total phosphorus, and chemical oxygen demand are extracted. The contribution rates of different concentration values ​​are obtained, and the contribution degree of each pollution source such as farmland runoff, livestock and poultry breeding, and domestic sewage to the water quality exceeding the standard is quantitatively evaluated, including: pass Get the concentration value; where: is the concentration value of the i-th pollutant; is the contribution coefficient of the kth pollution source; is the source component spectrum of the jth pollutant of the kth pollution source; is the residual term; is the number of pollution sources; Number the pollutant types; Number the monitoring points; Number the pollution sources; pass Obtain the pollution source contribution rate; where: is the contribution rate of the kth pollution source; Assess the contribution rate of agricultural runoff pollution sources:

[0055] Assessment of the contribution rate of livestock and poultry breeding pollution sources:

[0056] Assessment of the contribution rate of domestic sewage pollution sources:

[0057] in, 、 and These are the contribution rate assessment values ​​of farmland runoff, livestock and poultry breeding, and domestic sewage: Indicates total nitrogen concentration; represents the total phosphorus concentration; represents chemical oxygen demand; 、 、 These are the contribution coefficients of three types of pollution sources: farmland runoff, livestock and poultry breeding, and domestic sewage; 、 、 They are the total nitrogen, total phosphorus and chemical oxygen demand source component spectrum values ​​of agricultural runoff pollution sources; 、 、 They are the total nitrogen, total phosphorus and chemical oxygen demand source component spectrum values ​​of livestock and poultry breeding pollution sources; 、 、 They are the source component spectrum values ​​of total nitrogen, total phosphorus and chemical oxygen demand of domestic sewage pollution sources.

[0058] In this implementation, the source composition profile values ​​were obtained through field sampling and laboratory analysis. The characteristic profile for agricultural runoff is TN:TP:COD = 1:0.15:8; for livestock and poultry farming, TN:TP:COD = 1:0.25:12; and for domestic sewage, TN:TP:COD = 1:0.20:10. The source composition profile database contains over 1,000 data sets of typical values ​​for different regions and seasons.

[0059] The contribution coefficients were calculated using the Absolute Principal Component Analysis-Multivariate Linear Regression (APCS-MLR) method, combined with the Non-Negative Matrix Factorization (NMF) algorithm for optimization. The model was solved using the least squares method, with the constraints that the contribution ratios were non-negative and summed to 100%.

[0060] The residual analysis adopted residual diagnostic methods, including normality test (Shapiro-Wilk test), independence test (Durbin-Watson test) and homogeneity of variance test (Breusch-Pagan test) to ensure the quality of model fitting.

[0061] The model was validated using the cross-validation method, with the data set divided into a training set and a test set in a ratio of 7:3. The model goodness of fit R 2 The performance is good and the root mean square error RMSE is controlled within a reasonable range.

[0062] This embodiment is based on the receptor model theory and uses mathematical modeling methods to quantitatively decompose the contributions of different pollution sources. Utilizing the linear superposition principle of pollutant concentrations and combining the characteristics of the source component spectrum, the contribution coefficient of each pollution source is solved through multivariate linear regression. This method can handle complex situations with multiple pollution sources and multiple pollutants, and achieve accurate quantification of the contribution of pollution sources. It achieves accurate decomposition of the contributions of multiple pollution sources, and significantly improves the quantification accuracy; establishes a standardized source component spectrum database to support the identification of pollution sources in different regions; ensures the reliability of the model through residual analysis, and the model fit performance is excellent; provides accurate data support for pollution control, and significantly improves the accuracy of the control effect evaluation.

[0063] Example 8 In a preferred embodiment of the present invention, the source tracing analysis results are matched and verified with the pollution source feature information in the agricultural non-point source pollution knowledge graph, the pollution source type is confirmed through feature similarity comparison and semantic relationship reasoning, and the pollution contribution rate of different agricultural non-point source pollution is quantitatively calculated, and finally a source tracing result containing the pollution source location, type, and contribution rate is generated, including: Based on the results of the source tracing analysis, the key characteristic parameters of the pollution content value ratio, spatiotemporal distribution characteristics, and transmission path information of the pollution event were preliminarily matched with the standard characteristic plates of the three types of pollution sources stored in the agricultural non-point source pollution knowledge graph: farmland runoff, livestock and poultry breeding, and rural life, and a preliminary matching result was obtained; Based on the preliminary matching results, a similarity threshold is obtained, and candidate pollution source types are screened according to the similarity threshold. At the same time, the semantic relationship network in the knowledge graph is used for reasoning and verification to confirm the pollution source type in the current monitoring area; Based on the confirmed pollution source type, the spatial distribution information and attribute characteristics of the corresponding pollution source in the knowledge graph are called up. Through GIS spatial overlay analysis, the pollution transmission path is matched with the location of the pollution source in the actual geographic space. The spatial distance and transmission accessibility of each candidate pollution source to the monitoring point are calculated to determine the precise geographic coordinates of the pollution source. Combining the contribution rate of pollution sources with their precise geographic coordinates, we spatially aggregated the contribution rates of multiple pollution sources of the same type. By correcting the actual contribution of pollution sources at different distances through pollution load attenuation, we obtained the final contribution rate distribution of various agricultural non-point source pollution sources. Based on the final contribution rate allocation results, a comprehensive traceability result report is generated, which includes the precise coordinates of the pollution source, pollution source type identification, time-varying contribution rate value, and traceability confidence evaluation.

[0064] In this embodiment, the feature similarity calculation adopts a method combining cosine similarity and Euclidean distance. The calculation formula is:

[0065] Where α is the weight coefficient (valued at 0.6), θ is the eigenvector angle, d is the Euclidean distance, and σ is the scale parameter. The similarity threshold is set at 0.75, and candidate pollution sources above the threshold proceed to the next step of verification.

[0066] The semantic relationship reasoning adopts a rule-based reasoning engine, and the reasoning rules include: When TN / TP>10 and COD / BOD<2, it is inferred to be a source of agricultural runoff pollution; When NH3-N / TN>0.6 and fecal coliform group>10 4 , it is inferred that the pollution source is livestock and poultry breeding; When the COD / BOD is between 2-4 and contains surfactants, it is inferred to be a domestic sewage pollution source.

[0067] The spatial overlay analysis adopts a method combining buffer zone analysis and network analysis to establish an impact buffer zone with a radius of 2 km centered on the monitoring point, and calculates the spatial distribution density and transmission accessibility index of various pollution sources in the buffer zone.

[0068] The pollution load attenuation correction takes into account factors such as distance attenuation, terrain barriers, and vegetation interception. The attenuation correction coefficient The calculation formula is:

[0069] in, is the basic attenuation coefficient, slope is the average slope, NDVI is the vegetation index, and distance is the transmission distance.

[0070] This embodiment is based on pattern recognition and spatial analysis theory, and ensures the accuracy of traceability results through multi-level matching verification. First, feature similarity is used for preliminary screening, then logical verification is performed through semantic reasoning, and finally the precise location is determined in combination with spatial analysis. The entire process realizes a complete traceability chain from qualitative identification to quantitative analysis. Multi-level matching verification is achieved, and the traceability accuracy rate is significantly improved; the logic of traceability is enhanced through semantic reasoning, effectively reducing the misjudgment rate; an accurate spatial positioning mechanism is established, and the position accuracy reaches the meter level; a standardized traceability report is generated, providing a scientific basis for environmental management.

[0071] Example 9 In a preferred embodiment of the present invention, the contribution rates of multiple pollution sources of the same type are spatially aggregated by combining the pollution source contribution rates and the precise geographic coordinates of the pollution sources. The actual contribution levels of pollution sources at different distances are corrected by pollution load attenuation to obtain the final contribution rate distribution of various agricultural non-point source pollution types, including: pass Gets the spatial weights; where: is the spatial weight of the j-th monitoring point to the k-th pollution source; is the Euclidean distance from the jth monitoring point to the kth pollution source; is the distance decay exponent; is the exponential decay coefficient; is the affected area of ​​the kth pollution source; is the number of pollution sources; pass Correct the actual contribution of pollution sources at different distances; among which: is the corrected contribution rate of the kth pollution source; is the original contribution rate; is the time attenuation coefficient; is the time it takes for pollutants to be transported from the kth pollution source to the jth monitoring point; is the maximum effective transmission distance; is the spatial attenuation index; Number the monitoring points; Number the pollution source Final contribution rate distribution through Perform global optimization; Constraints: , ; in: is the number of pollution source types; is the number of the k-th pollution source; is the observed contribution rate of the kth pollution source; is the prior contribution rate based on historical data; is the regularization parameter; is the corrected contribution rate of the kth pollution source. By solving the optimal contribution rate allocation scheme, the final contribution rate distribution of various types of agricultural non-point source pollution is obtained.

[0072] In this implementation, the distance decay exponent α is determined based on the pollutant type: α = 1.5 for nutrients such as nitrogen and phosphorus, α = 1.2 for organic matter, and α = 2.0 for heavy metals. The exponential decay coefficient β is adjusted based on the terrain: β = 0.001 in plains, β = 0.002 in hilly areas, and β = 0.003 in mountainous areas.

[0073] The calculation of the affected area Aj takes into account the scale of the pollution source and the scope of influence. The affected area of ​​farmland runoff is calculated based on the area of ​​the field, the affected area of ​​livestock and poultry breeding is calculated based on 1.5 times the breeding scale, and the affected area of ​​domestic sewage is calculated based on the service population density.

[0074] The time decay coefficient λ is determined based on the degradation rate of pollutants in the environment. For nitrogen and phosphorus pollutants, λ is 0.05 / day, for organic matter, λ is 0.1 / day, and for pathogenic microorganisms, λ is 0.2 / day. The maximum effective transmission distance Dmax is determined by topographic and hydrological conditions and is generally 2-5 km.

[0075] The multi-objective optimization algorithm uses an improved non-dominated sorting genetic algorithm (NSGA-II) with a population size of 100, 500 generations, a crossover probability of 0.9, and a mutation probability of 0.1. The optimization objectives include minimizing fitting error, maximizing spatial rationality, and minimizing deviation from prior knowledge.

[0076] The regularization parameter μ was determined through cross-validation, ranging from 0.1 to 1.0, with the optimal value obtained through grid search. The prior contribution rates were determined based on historical statistical data and expert knowledge, with a priori contribution rates of 40-60% for farmland runoff, 20-35% for livestock and poultry farming, and 15-25% for domestic sewage.

[0077] This embodiment is based on spatial statistics and multi-objective optimization theory. It quantifies the degree of influence of different pollution sources on monitoring points through spatial weight functions, uses the spatiotemporal attenuation model to correct the influence of distance and time on pollution transmission, and finally solves the optimal contribution rate allocation scheme under the premise of satisfying physical constraints through a multi-objective optimization algorithm. This method fully considers the spatiotemporal characteristics of pollutant transmission and the superposition effect of multiple sources. It realizes the precise allocation of the contribution rates of multiple pollution sources, and the allocation accuracy is significantly improved; through spatiotemporal attenuation correction, it more accurately reflects the actual impact of the pollution sources; establishes a multi-objective optimization allocation mechanism, balances fitting accuracy and spatial rationality; provides a scientific basis for precise governance, improves the efficiency of governance resource allocation, and reduces governance costs.

[0078] like Figure 2 As shown in the figure, an intelligent tracing system for agricultural non-point source pollution based on knowledge graph includes: Data collection module: used to collect multi-dimensional data on agricultural non-point source pollution and build a knowledge graph of agricultural non-point source pollution; Integrated monitoring module: used to establish an integrated sky-ground monitoring network based on the agricultural non-point source pollution knowledge graph and obtain a monitoring data set; Source tracing monitoring mechanism module: used to deeply integrate the agricultural non-point source pollution knowledge graph with the monitoring data set to build an intelligent source tracing monitoring mechanism; Pollution source identification module: used to train and build a pollution source identification model based on the intelligent source tracing monitoring mechanism and historically acquired monitoring data sets; Source tracing analysis module: used to obtain pollution content values ​​based on the pollution source identification model. When the pollution content value exceeds a preset threshold, the pollution source identification model is triggered to identify the pollution source and conduct intelligent tracing analysis on the monitoring area where the pollution exceeds the standard, and obtain the tracing analysis results; Source tracing result generation module: used to match and verify the source tracing analysis results with the pollution source feature information in the agricultural non-point source pollution knowledge graph, confirm the pollution source type through feature similarity comparison and semantic relationship reasoning, and quantitatively calculate the pollution contribution rate of different agricultural non-point sources, and finally generate a source tracing result including the location, type and contribution rate of the pollution source.

[0079] Application example 1: Comprehensive management of agricultural non-point source pollution in a certain county.

[0080] In a major agricultural county, the proposed method was used to conduct intelligent source tracing analysis of agricultural non-point source pollution across 1,200 square kilometers. By constructing a knowledge graph encompassing 32,000 farmland plots, 156 large-scale livestock farms, and 89 rural domestic sewage treatment facilities, a county-wide integrated air-ground monitoring network was established, encompassing 68 automatic water quality monitoring stations, 12 drone patrol monitoring points, and full satellite remote sensing coverage.

[0081] During spring fertilization, the system detected an abnormally high total nitrogen concentration in a watershed, with the API index reaching 132, triggering intelligent source tracing analysis. Using a pollution source identification model, the pollution sources were identified as farmland runoff (58.3% contribution), livestock and poultry farming (31.2% contribution), and domestic sewage (10.5% contribution). Further spatial aggregation analysis pinpointed three key upstream farmland areas and two livestock farms as the primary pollution sources, achieving a confidence level of 89.6%.

[0082] Based on the source-tracing results, a targeted treatment plan was developed: soil testing and formula fertilization were implemented in key farmland areas, reducing nitrogen fertilizer use by 25%; problematic livestock farms were required to install manure treatment facilities to ensure compliance with discharge standards. Three months after the treatment was implemented, water quality in the basin improved significantly, with total nitrogen concentrations dropping by 42% and the API index dropping to 76, demonstrating significant improvement.

[0083] Application example 2: Accurate identification of lake eutrophication pollution sources.

[0084] A cyanobacterial bloom in a lake, a key drinking water source, posed a threat to water supply security. Using this method, we conducted an emergency source analysis across a 500-square-kilometer catchment area surrounding the lake. The resulting knowledge graph contained 21,000 pollution source entities and 150,000 relationship links, encompassing various pollution sources, including farmland, livestock farms, and rural residential areas.

[0085] Real-time analysis using an intelligent source-tracing monitoring mechanism revealed that the total phosphorus concentration in the lake increased dramatically from 0.08 mg / L to 0.35 mg / L over a short period of time, far exceeding the Class III standard for surface water. Pollution source identification models indicate that the pollution incident was primarily caused by non-point source pollution washout triggered by heavy rainfall, with livestock and poultry farming contributing 67.8%, agricultural runoff 23.4%, and domestic sewage 8.8%.

[0086] Through reverse engineering of pollutant transmission pathways and spatial overlay analysis, the primary pollution sources were pinpointed to a large pig farm on the northwest shore of the lake and two dairy farms on the southeast shore, with a confidence level of 92.3%. On-site verification revealed that the manure treatment facilities at these farms overflowed during heavy rainfall, discharging large amounts of high-phosphorus wastewater directly into the lake.

[0087] Based on the precise traceability results, emergency measures were immediately initiated: the problematic farms were shut down, pollutants were cleaned up, and phosphorus removal agents were introduced. A long-term remediation plan was also developed, requiring all farms to install rainwater and sewage diversion systems and emergency storage tanks. Following the remediation efforts, the lake's water quality returned to normal within a month, successfully ensuring water supply security.

[0088] Application example 3: Collaborative governance of cross-regional river pollution.

[0089] A trans-provincial river basin, encompassing three prefecture-level cities in the upper, middle, and lower reaches, has long suffered from agricultural non-point source pollution, with limited success across local authorities. Using the method presented in this paper, a basin-wide intelligent source tracing system was established, constructing a knowledge graph of agricultural non-point source pollution covering the entire 8,000 square kilometers of the basin, encompassing 58,000 pollution source entities and 320,000 relationship networks.

[0090] System operation revealed that water quality in downstream cities has long exceeded standards, with the primary source of pollution coming from upstream areas. Intelligent source tracing analysis quantified the contribution of each region to downstream water quality: Agricultural runoff in upstream City A contributed 43.2%, livestock and poultry farming in midstream City B 31.8%, and domestic sewage in downstream City C 25.0%. Further analysis revealed that corn cultivation in upstream City A and concentrated pig farming in midstream City B were the primary sources of pollution.

[0091] Based on the precise contribution rate allocation results, the three cities established a cross-regional collaborative governance mechanism: City A focused on reducing nitrogen and phosphorus emissions from farmland and implemented a project to replace chemical fertilizers with organic fertilizers; City B strengthened the resource utilization of livestock and poultry manure and established a regional manure treatment center; and City C improved rural sewage collection and treatment facilities. An ecological compensation mechanism was also established, with downstream cities providing financial support for governance upstream.

[0092] After two years of collaborative governance, river water quality has significantly improved, with water quality at key sections rising from Class V to Class III, and agricultural non-point source pollution load reduced by over 35%. This case study provides a successful example of cross-regional collaborative governance of agricultural non-point source pollution, achieving the goals of "precise source tracing, scientific governance, and collaborative win-win results."

[0093] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. An intelligent tracing method for agricultural non-point source pollution based on knowledge graph, characterized in that: include: Collect multi-dimensional data on agricultural non-point source pollution and construct a knowledge graph of agricultural non-point source pollution; Based on the agricultural non-point source pollution knowledge graph, an integrated sky-ground monitoring network is established to obtain a monitoring data set; Deeply integrate the agricultural non-point source pollution knowledge graph with the monitoring data set to build an intelligent traceability monitoring mechanism; Based on the intelligent source tracing monitoring mechanism and historically acquired monitoring data sets, a pollution source identification model is trained and constructed; Acquire pollution content values ​​based on a pollution source identification model. When the pollution content value exceeds a preset threshold, the pollution source identification model is triggered to identify pollution sources and perform intelligent tracing analysis on the monitoring area where pollution exceeds the standard, thereby obtaining tracing analysis results. The traceability analysis results are matched and verified with the pollution source feature information in the agricultural non-point source pollution knowledge graph, the pollution source type is confirmed through feature similarity comparison and semantic relationship reasoning, and the pollution contribution rate of different agricultural non-point sources is quantitatively calculated, and finally a traceability result including the pollution source location, type, and contribution rate is generated.

2. The method for intelligent tracing of agricultural non-point source pollution based on knowledge graph according to claim 1 is characterized in that: Collect multi-dimensional data on agricultural non-point source pollution and construct a knowledge graph of agricultural non-point source pollution, including: Collect environmental quality standard data, agricultural production history data, land use status data, and agricultural pollution source distribution data in the monitoring area to build a basic agricultural non-point source data set; Preprocessing and normalizing the agricultural non-point source basic data set to obtain a processed agricultural non-point source data set; Extract farmland entities, livestock and poultry breeding entities, rural life entities, and environmental factor entities from the processed agricultural nonpoint source data set, and identify the association relationships between the entities; Construct knowledge representation units in the form of multiple arrays to form structured agricultural non-point source pollution knowledge units; The structured knowledge units are integrated according to the logical relationship from pollution source to transmission path and then to receptor to establish an agricultural non-point source pollution knowledge network; By designing semantic node index and query relationship index, a knowledge graph of agricultural non-point source pollution that supports semantic reasoning and complex queries is constructed.

3. The method for intelligent tracing of agricultural non-point source pollution based on knowledge graph according to claim 1 is characterized in that: Based on the agricultural non-point source pollution knowledge graph, an integrated sky-ground monitoring network is established to form a monitoring network data system and obtain monitoring data sets to build an intelligent source tracing monitoring mechanism, including: Based on the knowledge graph of agricultural non-point source pollution, a multi-scale monitoring network data system is constructed, including a space-based remote sensing monitoring layer, an air-based drone monitoring layer, and a ground-based Internet of Things monitoring layer. The space-based remote sensing monitoring layer obtains large-scale vegetation coverage and crop growth information; the air-based drone monitoring layer obtains mesoscale farmland dynamic change monitoring data; and the ground-based Internet of Things monitoring layer obtains small-scale real-time and continuous farmland environmental parameter monitoring. A multi-source data spatiotemporal registration mechanism is established based on a multi-scale monitoring network data system. Using farmland plots as the basic spatial unit and the crop growth period as the temporal scale, the space-based, airborne, and ground-based monitoring data systems are spatially interpolated and temporally synchronized. This integrates monitoring data at different scales to produce a comprehensive monitoring dataset that is consistent in both time and space. Perform spatial expansion analysis on the spatial distribution characteristics of pollutants for each monitoring point in the comprehensive monitoring data set, use time series analysis to identify pollution change trends, and form analysis results that integrate multi-source data; Based on the analysis results, a directional response relationship between pollution load and multiple driving factors such as meteorological conditions, agricultural activities, and soil properties was established; By combining directional response relationships with historical meteorological data, we can obtain pollution washoff thresholds under different rainfall intensities, identify key driving factors affecting changes in agricultural non-point source pollution, and ultimately build an intelligent source tracing monitoring mechanism.

4. The method for intelligent tracing of agricultural non-point source pollution based on knowledge graph according to claim 3 is characterized in that: The intelligent traceability monitoring mechanism includes: The pollution source emission coefficients, soil nutrient balance equations, hydrological transmission parameters and key driving factors of various pollution source parameters in the agricultural non-point source pollution knowledge map are coupled to obtain coupling results, and the nonlinear adsorption and desorption processes between soil and water bodies are integrated. Based on the coupling results, identify the complex pollution pattern of multiple pollution sources to distinguish the superposition effects of different pollution sources such as farmland fertilization, livestock and poultry breeding, and domestic sewage. This will obtain the first identification result; Based on the first identification results, the pollution load response characteristics under different rainfall intensity and duration scenarios are predicted to obtain prediction results; Based on the first identification and prediction results, the migration and diffusion patterns of pollutants under complex terrain conditions are simulated, and the pollutant diffusion trend is obtained through the changes in the flow velocity of surface runoff and base flow; Analyze the pollutant diffusion law based on the pollutant diffusion trend, obtain the distribution of pollution content values ​​in the monitoring area within the preset time, evaluate the pollutant diffusion efficiency and impact range, and obtain the evaluation results; According to the evaluation results, the monitoring network layout and parameter configuration are adjusted through a multi-objective optimization algorithm to form an intelligent traceability monitoring mechanism that adapts to different agricultural regions and pollution types.

5. The method for intelligent tracing of agricultural non-point source pollution based on knowledge graph according to claim 4 is characterized in that: Based on the intelligent source tracing monitoring mechanism and historically acquired monitoring data sets, a pollution source identification model is trained and constructed, including: Through an intelligent traceability monitoring mechanism, water quality data from farmland drainage outlets, livestock and poultry farms, and rural domestic sewage treatment facility outlets are monitored in real time. Historical rainfall, soil moisture content, and environmental parameters during the crop growth period are simultaneously acquired. Combined with information on land use changes, crop planting structure, and farm distribution obtained through remote sensing, a multidimensional monitoring dataset containing water quality parameters, environmental factors, and spatial attributes is constructed. Based on the relationship network in the agricultural non-point source pollution knowledge graph, the multidimensional monitoring data set is decomposed into time series to obtain decomposition results; the decomposition results include trend items reflecting the agricultural activity cycle, seasonal items reflecting seasonal pollution patterns, and random items representing sudden pollution events; Based on the decomposition results, we identify the water quality variation patterns caused by agricultural activity cycles such as fertilization season, rainfall season, and crop harvest season, and establish a temporal correlation network between pollution content values ​​and agricultural activities. Based on the temporal association network, , obtain the comprehensive index of agricultural non-point source pollution in the monitoring area, quantitatively assess the severity of the current pollution situation, establish a dynamically adjusted classification and pollution content value threshold system based on the surface water environmental quality standards and farmland irrigation water quality standards, identify abnormal patterns and mutation points in water quality data, and obtain the second identification result; among them, is the comprehensive index of agricultural non-point source pollution, is the pollution content value of the i-th type, is the corresponding standard value, is the pollution weight, is the environmental impact weight of agricultural activities, Number the pollutant types; The second recognition result is semantically matched and knowledge inference is performed with the agricultural non-point source pollution knowledge graph, and the reasoning process is further optimized through causal reasoning to establish the correlation between pollution events and pollution sources; Based on the correlation between pollution events and pollution sources, pollution events are classified, and the spatiotemporal characteristics of pollution content values ​​corresponding to pollution events are extracted. A pollution source identification model based on contrastive learning is established to accurately identify the types of pollution sources corresponding to different types of pollution events, and obtain end-to-end intelligent identification from monitoring data anomalies to precise positioning of pollution sources.

6. The method for intelligent tracing of agricultural non-point source pollution based on knowledge graph according to claim 5 is characterized in that: The pollution source identification model is used to obtain pollution content values. When the pollution content value exceeds a preset threshold, the pollution source identification model is triggered to perform pollution source identification and intelligent source tracing analysis on the monitoring area where pollution exceeds the standard, and the source tracing analysis results are obtained, including: The pollution source identification model is used to obtain the comprehensive index of agricultural non-point source pollution in each monitoring area in real time. The dynamic pollution content threshold is set based on the regional background value and seasonal variation characteristics in the agricultural non-point source basic data set. When the pollution content value in the monitoring area exceeds the preset threshold, the pollution source identification program is automatically triggered to identify abnormal events of excessive pollution. Based on the triggered pollution source identification program, the intelligent source tracing analysis process is initiated, calling the relationship network in the agricultural non-point source pollution knowledge map. Combined with the current rainfall intensity, soil moisture content, and environmental parameters of the crop growth period, the graph reasoning algorithm is used to identify the pollution source type and spatial distribution of the pollution exceeding the standard event, and obtain the identification results; Based on the identification results, the water quality fingerprint characteristics of abnormal pollution exceeding the standard in the monitoring area are analyzed, and the concentration values ​​of key pollutants such as total nitrogen, total phosphorus, and chemical oxygen demand are extracted. The contribution rates of different concentration values ​​are obtained, and the contribution of each pollution source such as farmland runoff, livestock and poultry breeding, and domestic sewage to the water quality exceeding the standard is quantitatively evaluated; Based on the degree of contribution, the pollutant transmission path is reversed, and the migration trajectory and transmission time of pollutants from each pollution source to the monitoring point are determined through surface runoff simulation and groundwater seepage calculation; The transmission path obtained by reverse inference is matched and verified with the spatial relationship network in the agricultural non-point source pollution knowledge map, and the traceability confidence of different pollution sources is obtained through spatial overlay analysis and path consistency test; Based on the contribution degree and confidence level of the pollution source to water quality exceeding the standard, a comprehensive traceability analysis result is generated, including the pollution source type, time-varying contribution rate and traceability credibility.

7. The method for intelligent tracing of agricultural non-point source pollution based on knowledge graph according to claim 6 is characterized in that: Based on the identification results, the water quality fingerprint characteristics of abnormal pollution exceeding the standard in the monitoring area are analyzed, and the concentration values ​​of key pollutants such as total nitrogen, total phosphorus, and chemical oxygen demand are extracted. The contribution rate of different concentration values ​​is obtained, and the contribution degree of each pollution source such as farmland runoff, livestock and poultry breeding, and domestic sewage to the water quality exceeding the standard is quantitatively evaluated, including: pass Get the concentration value; where: is the concentration value of the i-th pollutant; is the contribution coefficient of the kth pollution source; is the source component spectrum of the jth pollutant of the kth pollution source; is the residual term; is the number of pollution sources; Number the pollutant types; Number the monitoring points; Number the pollution sources; pass Obtain the pollution source contribution rate; where: is the contribution rate of the kth pollution source. By obtaining the contribution rates of different pollution sources, the contribution degree of each pollution source to the water quality exceeding the standard is quantitatively evaluated.

8. The method for intelligent tracing of agricultural non-point source pollution based on knowledge graph according to claim 7 is characterized in that: The traceability analysis results are matched and verified with the pollution source feature information in the agricultural non-point source pollution knowledge graph. The pollution source type is confirmed through feature similarity comparison and semantic relationship reasoning, and the pollution contribution rate of different agricultural non-point sources is quantitatively calculated. Finally, the traceability results containing the pollution source location, type, and contribution rate are generated, including: Based on the results of the source tracing analysis, the key characteristic parameters of the pollution content value ratio, spatiotemporal distribution characteristics, and transmission path information of the pollution event were preliminarily matched with the standard characteristic plates of the three types of pollution sources stored in the agricultural non-point source pollution knowledge graph: farmland runoff, livestock and poultry breeding, and rural life, and a preliminary matching result was obtained; Based on the preliminary matching results, a similarity threshold is obtained, and candidate pollution source types are screened according to the similarity threshold. At the same time, the semantic relationship network in the knowledge graph is used for reasoning and verification to confirm the pollution source type in the current monitoring area; Based on the confirmed pollution source type, the spatial distribution information and attribute characteristics of the corresponding pollution source in the knowledge graph are called up. Through GIS spatial overlay analysis, the pollution transmission path is matched with the location of the pollution source in the actual geographic space. The spatial distance and transmission accessibility of each candidate pollution source to the monitoring point are calculated to determine the precise geographic coordinates of the pollution source. Combining the contribution rate of pollution sources with their precise geographic coordinates, we spatially aggregated the contribution rates of multiple pollution sources of the same type. By correcting the actual contribution of pollution sources at different distances through pollution load attenuation, we obtained the final contribution rate distribution of various agricultural non-point source pollution sources. Based on the final contribution rate allocation results, a comprehensive traceability result report is generated, which includes the precise coordinates of the pollution source, pollution source type identification, time-varying contribution rate value, and traceability confidence evaluation.

9. The method for intelligent tracing of agricultural non-point source pollution based on knowledge graph according to claim 8 is characterized in that: Combining the contribution rate of pollution sources with their precise geographic coordinates, we spatially aggregated the contribution rates of multiple pollution sources of the same type. By correcting the actual contribution of pollution sources at different distances using pollution load attenuation, we obtained the final contribution rate distribution of various agricultural non-point source pollution types, including: pass Gets the spatial weights; where: is the spatial weight of the j-th monitoring point to the k-th pollution source; is the Euclidean distance from the jth monitoring point to the kth pollution source; is the distance decay exponent; is the exponential decay coefficient; is the affected area of ​​the kth pollution source; is the number of pollution sources; pass Correct the actual contribution of pollution sources at different distances; among which: is the corrected contribution rate of the kth pollution source; is the original contribution rate; is the time attenuation coefficient; is the time it takes for pollutants to be transported from the kth pollution source to the jth monitoring point; is the maximum effective transmission distance; is the spatial attenuation index; Number the monitoring points; Number the pollution source Final contribution rate distribution through Perform global optimization; Constraints: , ; in: is the number of pollution source types; is the number of the k-th pollution source; is the observed contribution rate of the kth pollution source; is the prior contribution rate based on historical data; is the regularization parameter; is the corrected contribution rate of the kth pollution source. By solving the optimal contribution rate allocation scheme, the final contribution rate distribution of various types of agricultural non-point source pollution is obtained.

10. An intelligent tracing system for agricultural non-point source pollution based on knowledge graph, used to execute an intelligent tracing method for agricultural non-point source pollution based on knowledge graph according to any one of claims 1 to 9, characterized in that: include: Data collection module: used to collect multi-dimensional data on agricultural non-point source pollution and build a knowledge graph of agricultural non-point source pollution; Integrated monitoring module: used to establish an integrated sky-ground monitoring network based on the agricultural non-point source pollution knowledge graph and obtain a monitoring data set; Source tracing monitoring mechanism module: used to deeply integrate the agricultural non-point source pollution knowledge graph with the monitoring data set to build an intelligent source tracing monitoring mechanism; Pollution source identification module: used to train and build a pollution source identification model based on the intelligent source tracing monitoring mechanism and historically acquired monitoring data sets; Source tracing analysis module: used to obtain pollution content values ​​based on the pollution source identification model. When the pollution content value exceeds a preset threshold, the pollution source identification model is triggered to identify the pollution source and conduct intelligent tracing analysis on the monitoring area where the pollution exceeds the standard, and obtain the tracing analysis results; Source tracing result generation module: used to match and verify the source tracing analysis results with the pollution source feature information in the agricultural non-point source pollution knowledge graph, confirm the pollution source type through feature similarity comparison and semantic relationship reasoning, and quantitatively calculate the pollution contribution rate of different agricultural non-point sources, and finally generate a source tracing result including the location, type and contribution rate of the pollution source.

Images