Field pollution tracing method and system based on knowledge graph

Abstract

The invention relates to the field of environment supervision, and particularly provides a site pollution tracing method and system based on a knowledge graph. The method comprises the following steps: when an environment abnormal event occurs at a monitoring point, determining path query information through a spatial database based on event information of the environment abnormal event; based on the path query information, performing path search in the knowledge graph, and determining all candidate paths from the potential pollution source node to the monitoring point node; based on a pre-trained scoring model, scoring each candidate path, and sorting the candidate paths according to a scoring result; based on the sorting result, selecting a preset number of candidate paths from high to low as candidate pollution paths; generating a traceability report based on the candidate pollution paths; wherein the spatial database and the knowledge graph are generated based on multi-source heterogeneous data related to environment supervision. According to the site pollution tracing method provided by the invention, pollution sources can be automatically and accurately traced in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of environmental regulatory technology, specifically to a method and system for tracing the source of site pollution based on knowledge graphs. Background Technology

[0002] Industrial enterprises generate various pollutants during their production processes, and the complex composition of emissions results from different industries, raw materials, and process combinations. Currently, the monitoring and prediction of enterprise emissions mainly rely on fixed industry pollutant lists, manual experience, or post-event testing, lacking intelligent correlation and inference of multi-source data, making it difficult to detect abnormal emissions and potential risks in a timely manner.

[0003] Some existing environmental management systems are beginning to incorporate technologies such as knowledge graphs and graph neural networks (GNNs) to attempt to integrate data such as basic enterprise information, process flows, and monitoring history. For example, some methods construct knowledge graphs of contaminated sites and use graph neural networks to infer potential pollution information; other systems construct enterprise-related knowledge graphs and use graph attention networks to extract features and detect abnormal risks. Still other solutions focus on the fusion of multi-source sensor monitoring data, using graph convolutional neural networks to extract spatiotemporal features to predict the spatial distribution of pollutants.

[0004] However, the aforementioned existing technologies still have significant shortcomings. First, there is a lack of a unified framework for source tracing and prediction. Existing solutions often focus on single functions, such as risk warning or spatial distribution prediction, making it difficult to support the complete path inference from abnormal monitoring points to potential pollution sources, nor can they provide a list of possible pollutants and their probability distribution for specific process nodes. Second, the automation level of compliance judgment is low. Few systems structurally embed environmental regulations and emission standards to achieve automatic threshold comparison and compliance judgment, still relying on manual verification. Moreover, there is a lack of multi-model collaboration mechanisms. Existing methods mostly use single models or sequential processing, and have not yet established multi-agent collaborative working mechanisms such as knowledge graph reasoning and dynamic path search, making it difficult to balance prediction accuracy and result interpretability. Therefore, there is an urgent need for a pollution source tracing method that can integrate knowledge graph reasoning and dynamic path search. Summary of the Invention

[0005] In view of this, the present invention provides a site pollution source tracing method and system based on knowledge graph, which can automatically and accurately trace the source of pollution in complex environments.

[0006] To solve at least one of the above-mentioned technical problems, the present invention adopts the following technical solution:

[0007] The first aspect of this invention provides a site pollution source tracing method based on knowledge graphs, comprising:

[0008] When an environmental anomaly occurs at a monitoring point, the path query information is determined through a spatial database based on the event information of the environmental anomaly.

[0009] Based on path query information, path search is performed in the knowledge graph to determine all candidate paths between potential pollution source nodes and monitoring point nodes.

[0010] Based on a pre-trained scoring model, each candidate path is scored, and the candidate paths are ranked according to the scoring results.

[0011] Based on the ranking results, a preset number of candidate paths are selected as candidate contamination paths from high to low.

[0012] Generate source tracing reports based on candidate pollution pathways;

[0013] Among them, the spatial database and knowledge graph are generated based on multi-source heterogeneous data related to environmental regulation.

[0014] In one embodiment of the present invention, the generation of knowledge graphs and spatial databases includes:

[0015] Preprocessing of multi-source heterogeneous data;

[0016] Entity relations are extracted from the preprocessed multi-source heterogeneous data to generate knowledge graph triples;

[0017] Based on a predefined environment ontology model, entities and relations in the knowledge graph triples are inserted into a graph database to construct a knowledge graph.

[0018] Spatial location information is obtained from preprocessed multi-source heterogeneous data, and a spatial database is constructed based on the spatial location information.

[0019] In one embodiment of the present invention, when an environmental anomaly occurs at a monitoring point, path query information is determined through a spatial database based on the event information of the environmental anomaly, including:

[0020] Based on event information, the set of affected areas of an abnormal environmental event is determined through a spatial database. The event information includes anomaly information, location information, and medium information.

[0021] The event information and the set of affected areas are encapsulated into path query information.

[0022] In one embodiment of the present invention, based on path query information, a path search is performed in a knowledge graph to determine all candidate paths between potential pollution source nodes and monitoring point nodes, including:

[0023] Based on path query information, path search is performed in the knowledge graph according to preset search constraints to determine candidate paths between all potential pollution source nodes and monitoring point nodes in the affected area. The search constraints include temporal constraints, spatial / medium topology constraints, and search boundary constraints.

[0024] In one embodiment of the present invention, the timing constraint is used to constrain the occurrence time of the associated events of each path segment in the candidate path to be within a preset occurrence time window of the environmental anomaly event;

[0025] Spatial / medium topology constraints are used to ensure that the topological relationships between nodes in a candidate path conform to preset topology rules;

[0026] Search boundary constraints are used to constrain the maximum depth and maximum breadth when searching for paths.

[0027] In one embodiment of the present invention, based on path query information, a path search is performed in a knowledge graph to determine all candidate paths between potential pollution source nodes and monitoring point nodes, further comprising:

[0028] Based on the distribution density of potential pollution source nodes in the affected area, determine whether the spatial predicate selectivity is lower than a preset threshold;

[0029] When the spatial predicate selectivity is lower than the preset threshold, spatial filtering is performed on the knowledge graph based on path query information to obtain relevant subgraphs, and graph search is performed in the relevant subgraphs to determine all candidate paths between potential pollution source nodes and monitoring point nodes.

[0030] When the spatial predicate selectivity is higher than a preset threshold, a heuristic graph search is performed on the knowledge graph based on path query information to determine all candidate paths between potential pollution source nodes and monitoring point nodes.

[0031] In one embodiment of the present invention, the scoring model includes a mechanism consistency term, an evidence support term, and an uncertainty penalty term;

[0032] The mechanism consistency term is used to evaluate the consistency between each candidate path and the pollutant diffusion mechanism based on the edge attribute information of each candidate path.

[0033] The evidence support items are used to evaluate the degree of consistency between the edge attribute information and the event information of each candidate path;

[0034] The uncertainty penalty term is used to evaluate the degree of uncertainty of each candidate path in the determination process.

[0035] In one embodiment of the present invention, based on a pre-trained scoring model, each candidate path is comprehensively scored, and the candidate paths are ranked according to the scoring results, including:

[0036] Each candidate path is evaluated using a mechanism consistency term, an evidence support term, and an uncertainty penalty term, respectively.

[0037] Based on the evaluation results of the mechanism consistency item, evidence support item, and uncertainty penalty item, a comprehensive score for each candidate path is determined by weighted combination;

[0038] The candidate paths are ranked according to the scoring results.

[0039] In one embodiment of the present invention, generating a source tracing report based on candidate pollution pathways includes:

[0040] Based on a predefined set of rules, the emission allowance threshold is judged for each candidate pollution path in turn, and the evidence chain corresponding to each candidate pollution path is determined.

[0041] The judgment results and evidence chain for each candidate pollution path and emission allowance threshold are output as a source tracing report, and action recommendations are determined based on the source tracing report;

[0042] The set of rules includes at least one of the following: emission limit rules based on national regulations, emission limit rules based on local regulations, and enterprise pollution discharge permits.

[0043] In one embodiment of the present invention, based on a predefined set of rules, emission allowance thresholds are sequentially determined for each candidate pollution path, and the evidence chain corresponding to each candidate pollution path is determined, including:

[0044] Based on the edge attribute information of each candidate pollution path, the corresponding emission rules and emission allowance thresholds corresponding to the emission rules are determined from the rule set.

[0045] The event information of each candidate pollution path is compared with the emission rules and emission allowance thresholds in turn to obtain the judgment result of whether emission is allowed;

[0046] The evidence chain corresponding to each candidate contamination path is determined sequentially based on the topological relationships between nodes, event information, and rule set in each candidate contamination path.

[0047] In one embodiment of the present invention, the site pollution source tracing method based on knowledge graphs further includes:

[0048] Obtain feedback information on the source tracing report, and update the knowledge graph and / or scoring model based on the feedback information and according to the preset update constraints;

[0049] Among them, update constraints are used to constrain the update magnitude, update frequency, and update traceability of the update process.

[0050] A second aspect of the present invention provides a site pollution source tracing system based on a knowledge graph, comprising:

[0051] The query generation module is used to determine the path query information based on the event information of the environmental anomaly when an environmental anomaly occurs at the monitoring point, through a spatial database.

[0052] The path search module is used to perform path search in the knowledge graph based on path query information to determine all candidate paths between potential pollution source nodes and monitoring point nodes.

[0053] The path scoring module is used to score each candidate path based on a pre-trained scoring model and rank the candidate paths according to the scoring results.

[0054] The path filtering module is used to select a preset number of candidate paths as candidate contaminated paths based on the sorting results, from high to low.

[0055] The report generation module is used to generate source tracing reports based on candidate pollution pathways;

[0056] Among them, the spatial database and knowledge graph are generated based on multi-source heterogeneous data related to environmental regulation.

[0057] The above-described technical solution of the present invention has at least one of the following beneficial effects:

[0058] This invention presents a knowledge graph-based site pollution tracing method. It utilizes spatial constraints from a spatial database for initial geographical screening to determine path query information, effectively narrowing the subsequent search space of the knowledge graph. Then, based on the path query information, path search is performed through the semantic relationships of the knowledge graph to identify all candidate paths between potential pollution source nodes and monitoring point nodes, thereby significantly improving the accuracy and efficiency of candidate pollution path extraction. Furthermore, this application introduces a pre-trained scoring model to score and rank each candidate path, enabling the identification of the most likely pollution diffusion path and providing quantitative decision support for environmental regulation. Attached Figure Description

[0059] Figure 1 This is a schematic diagram illustrating the implementation environment of a site pollution tracing method based on knowledge graphs in one embodiment of the present invention.

[0060] Figure 2 This is a flowchart of a site pollution source tracing method based on knowledge graphs in one embodiment of the present invention;

[0061] Figure 3 This is a flowchart illustrating the generation of a knowledge graph and a spatial database in a site pollution tracing method based on a knowledge graph, as described in one embodiment of the present invention.

[0062] Figure 4 This is a flowchart illustrating the process of determining path query information in a knowledge graph-based site pollution tracing method according to one embodiment of the present invention.

[0063] Figure 5 This is a flowchart illustrating the process of determining all candidate paths between potential pollution source nodes and monitoring point nodes in a site pollution tracing method based on knowledge graphs in one embodiment of the present invention.

[0064] Figure 6 This is a flowchart illustrating the comprehensive scoring of each candidate path in a knowledge graph-based site pollution tracing method according to one embodiment of the present invention.

[0065] Figure 7 This is a flowchart illustrating the generation of a source tracing report in a knowledge graph-based site pollution source tracing method according to one embodiment of the present invention.

[0066] Figure 8 This is a flowchart illustrating the determination of emission allowance thresholds for each candidate pollution path in a knowledge graph-based site pollution tracing method according to an embodiment of the present invention.

[0067] Figure 9 This is a schematic diagram of a site pollution source tracing system based on knowledge graphs in one embodiment of the present invention. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0069] Reference manual attached Figure 1 , Figure 1 A schematic diagram illustrating the implementation environment of the knowledge graph-based site pollution source tracing method of the present invention is shown. Figure 1As shown, the implementation environment may include a monitoring device 1001 installed at the monitoring point, and a computing device 1002 electrically connected to the monitoring device 1001 via wired / wireless means. The computing device 1002 may be, but is not limited to, various servers, personal computers, laptops, smartphones, tablets, and portable wearable devices. The server may be a standalone server, a server cluster or distributed system composed of multiple servers, or an edge server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0070] The knowledge graph-based site pollution tracing method may include the following steps: When an environmental anomaly occurs at a monitoring point, the computing device 1002 can determine path query information through a spatial database based on the event information of the environmental anomaly, thus narrowing down the search space of the subsequent knowledge graph. Then, based on the path query information, the computing device 1002 can perform a path search in the knowledge graph to determine all candidate paths between potential pollution source nodes and monitoring point nodes, thereby effectively improving the accuracy and efficiency of candidate pollution path extraction. Further, the computing device 1002 can score each candidate path based on a pre-trained scoring model, and rank the candidate paths according to the scoring results. Based on the ranking results, a preset number of candidate paths are selected as candidate pollution paths from high to low. After determining the candidate pollution paths, the computing device 1002 can generate a tracing report based on the candidate pollution paths. This allows for the identification of the most likely pollution diffusion path, providing quantitative decision support for environmental supervision.

[0071] Below, please refer to the attached instruction manual. Figure 2 This illustrates the flowchart of a knowledge graph-based site pollution tracing method provided by an embodiment of the present invention, which can be applied to... Figure 1 In the computing device 1002. Specifically, the method may include the following steps:

[0072] S100. When an environmental anomaly occurs at a monitoring point, the path query information is determined through a spatial database based on the event information of the environmental anomaly.

[0073] The spatial database is generated based on multi-source heterogeneous data related to environmental regulation. Multi-source heterogeneous data includes, but is not limited to: sensor monitoring data (such as water quality and air indicators), information on polluting enterprises and their discharge permits, enterprise production processes and raw material data, hazardous waste transfer manifest records, meteorological and hydrological observation data, geographic information, and spatial location information (such as discharge outlet coordinates, groundwater well coordinates, river basin grids, etc.).

[0074] In this embodiment, when the environmental monitoring data collected by the monitoring equipment at the monitoring point exceeds a preset threshold or shows an abnormal trend, the computing device identifies the anomaly as an environmental anomaly event. Then, based on the event information of the environmental anomaly event, the computing device can limit the spatially related areas to the monitoring point through a spatial database, thereby generating path query information for subsequent path searching.

[0075] S200: Based on path query information, perform path search in the knowledge graph to determine all candidate paths between potential pollution source nodes and monitoring point nodes.

[0076] The knowledge graph is generated based on multi-source heterogeneous data related to environmental regulation. For details regarding multi-source heterogeneous data, please refer to the above description; further explanation is omitted here.

[0077] In this embodiment, the computing device can perform path search in the knowledge graph based on the path query information. Starting from the potential pollution source node, it traverses the entity nodes and relation edges related to pollution propagation, emission behavior and environmental media layer by layer until it reaches the corresponding monitoring point node, thereby obtaining all candidate paths that meet the conditions between the potential pollution source node and the monitoring point node.

[0078] S300: Based on a pre-trained scoring model, each candidate path is scored, and the candidate paths are ranked according to the scoring results.

[0079] In this embodiment, the computing device can input each candidate path into a pre-trained scoring model to obtain the corresponding path score value, and sort the candidate paths according to the score value to reflect the relative priority between different candidate paths.

[0080] S400. Based on the ranking results, select a preset number of candidate paths as candidate contaminated paths from high to low.

[0081] In this embodiment, the computing device can select a preset number of candidate paths based on the ranking results of the candidate paths, in descending order of scores, as candidate pollution paths for subsequent pollution source tracing analysis.

[0082] S500 generates source tracing reports based on candidate pollution pathways.

[0083] In this embodiment, the computing device can organize and summarize relevant path information based on the determined candidate pollution paths, and generate a source tracing report to reflect the correlation between the candidate pollution paths and environmental anomalies.

[0084] Below, each step will be explained in detail.

[0085] First, the generation of knowledge graphs and spatial databases is introduced. Specifically, such as... Figure 3 As shown, the generation of knowledge graphs and spatial databases includes the following steps S101-S104:

[0086] S101. Preprocess the multi-source heterogeneous data.

[0087] In this embodiment, the computing device can perform preprocessing operations on the acquired multi-source heterogeneous data to improve data consistency and usability. Specifically, preprocessing may include identifying and processing outlier values, invalid values, or data that do not meet preset conditions in the multi-source heterogeneous data. The outlier or invalid values ​​include at least one of the following: negative values, zero values, values ​​below the detection limit, and data that was acquired but not collected or is missing. Specifically, the computing device can employ robust statistical methods, such as interquartile range (IQR), median absolute difference (MAD), or time-series prediction models, such as the ARIMA model or the Prophet model, to replace, correct, or supplement outliers, thereby generating high-quality standardized data entries that can be used for subsequent knowledge graph construction and path analysis. By filtering, correcting, replacing, or labeling the multi-source heterogeneous data and performing spatiotemporal alignment processing, such as standardizing units and measurement systems, standardized data entries that meet the requirements of subsequent analysis are formed.

[0088] S102. Extract entity relationships from the preprocessed multi-source heterogeneous data to generate knowledge graph triples.

[0089] In this embodiment, the computing device can use a semantic extractor to identify and extract entities contained in the preprocessed multi-source heterogeneous data, such as sites, emission outlets, pollutants, processes, permits, monitoring points, and regulatory clauses, as well as the relationships between entities, such as enterprise-emission outlet affiliation, emission outlet-medium location, permit-covered emission outlet, monitoring point-measured pollutant, and pollutant-environmental standard limit. Then, the computing device can perform a structured representation of the extracted entities and their corresponding relationships. This forms a knowledge graph triple consisting of entities, relationships, and entities, which can be used for subsequent knowledge graph construction and reasoning analysis.

[0090] S103. Based on a predefined environment ontology model, entities and relations in the knowledge graph triples are inserted into the graph database to construct the knowledge graph.

[0091] In this embodiment, the computing device can perform normalized mapping of entity types and relation types in knowledge graph triples based on a predefined environment ontology model, and insert the entities and relations in the knowledge graph triples into a graph database for storage according to the definition rules of entities and relations in the environment ontology model, thereby constructing a knowledge graph with a consistent structure. The graph database can be Neo4j, JanusGraph, TigerGraph, NebulaGraph, or RDF based on triples to store the entities and relations in the knowledge graph triples; this application does not impose any restrictions.

[0092] S104. Obtain spatial location information from the preprocessed multi-source heterogeneous data, and construct a spatial database based on the spatial location information.

[0093] In this embodiment, the computing device can obtain corresponding spatial location information from preprocessed multi-source heterogeneous data, organize and process the spatial location information, perform geographic coordinate system unification, and store the spatial location information in a spatial database for unified management, thereby forming a spatial database based on spatial location information. The spatial database can be PostGIS, ESRI ArcGIS Enterprise, GeoMesa / HBase, or ClickHouse; this application does not impose any restrictions.

[0094] (a) Determining path query information.

[0095] Next, we will introduce the steps for determining path query information (i.e., step S100).

[0096] like Figure 4 As shown, step S100 may include the following steps S110-S120:

[0097] S110. Based on event information, determine the set of affected areas of the environmental anomaly event through a spatial database. The event information includes anomaly information, location information, and medium information.

[0098] In this embodiment, when the environmental monitoring data collected by the monitoring equipment at the monitoring point exceeds a preset threshold or shows an abnormal trend, the computing device can delineate the set of influence areas of the environmental anomaly event in the spatial database based on location information and medium information. For example, downstream / downwind areas can be determined based on river flow direction, groundwater flow field, or atmospheric wind direction.

[0099] Furthermore, after collecting environmental monitoring data at the monitoring points, the computing device can determine abnormal environmental events through a multi-trigger fusion mechanism. This includes combinations of various triggering methods such as real-time threshold exceeding the limit trigger, rate of change threshold trigger, rule trigger (e.g., special cases during holidays or at night), and neighbor station consistency trigger, thereby improving the reliability and sensitivity of abnormal event identification. For identified abnormal environmental events, the computing device can generate a set of influence areas based on event information and location information using different delineation strategies. These delineation strategies may include: spatial influence area delineation based on simplified equipotential surface models of hydrodynamics or wind fields, and reachability domain solving based on surface or underground pipe network topology to accurately describe the spatial range of potential pollutant diffusion, thus providing precise spatial constraints for the generation of subsequent path query information.

[0100] S120. Encapsulate the event information and the set of affected areas into path query information.

[0101] In this embodiment, the computing device integrates the event information with the set of affected areas, encapsulates it according to a preset data structure, and forms path query information for subsequent path search, so that the path query information simultaneously reflects the abnormal characteristics of the environmental abnormal event and its corresponding spatial affected area.

[0102] In one embodiment of the present invention, based on path query information, a path search is performed in the knowledge graph to determine all candidate paths between potential pollution source nodes and monitoring point nodes. This includes: based on path query information, a path search is performed in the knowledge graph according to preset search constraints to determine all candidate paths between potential pollution source nodes and monitoring point nodes in the affected area. The search constraints include temporal constraints, spatial / medium topology constraints, and search boundary constraints.

[0103] In this embodiment, the computing device starts from the affected monitoring point nodes and performs a reverse traversal of the knowledge graph towards the potential pollution source nodes, searching for all possible candidate paths. To ensure the controllability and effectiveness of the candidate paths, the computing device applies multiple constraints during the search process.

[0104] Specifically, the timing constraint is used to ensure that the occurrence time of the associated events of each path segment in the candidate path is within the preset occurrence time window of the environmental anomaly event. That is, the event time must match the occurrence time of the environmental anomaly event or be reasonably earlier than that time. Segments that exceed the preset occurrence time window do not participate in path generation, thereby ensuring that the causal chain has temporal consistency.

[0105] Spatial / medium topology constraints are used to ensure that the topological relationships between nodes in a candidate path conform to preset topology rules. This means they must follow actual geographic space or media connectivity; for example, an outlet must pass through a river or groundwater unit to affect downstream sections. If there is no actual media transmission interface between two nodes in the knowledge graph, the corresponding connection does not constitute a valid path.

[0106] Search boundary constraints are used to limit the scale of path search, which computing devices can achieve by limiting the maximum depth and maximum breadth. Specifically, when traversing the knowledge graph, the computing device employs a finite depth and finite branch strategy, limiting the maximum number of edges a path can traverse and the maximum number of branches each node can expand into, where the maximum depth... It can be 3-12, maximum breadth The value is 20-300. Preferably, the maximum depth is... It can be 6-8, maximum breadth The range is 80-150. By setting search boundary constraints, the combinatorial explosion problem in the path search process can be effectively avoided.

[0107] (ii) Candidate path determination.

[0108] The following describes the candidate path determination steps (i.e., step S200). Figure 5 As shown, in one embodiment of the present invention, step S200 (i.e., based on path query information, performing path search in the knowledge graph to determine all candidate paths between potential pollution source nodes and monitoring point nodes) may include the following steps S210-S230:

[0109] S210. Based on the distribution density of potential pollution source nodes in the affected area, determine whether the spatial predicate selectivity is lower than a preset threshold.

[0110] In this embodiment, the computing device can statistically analyze the distribution of potential pollution source nodes in the affected area, calculate spatial predicate selectivity based on this distribution, and compare the calculated spatial predicate selectivity with a preset threshold to determine whether the spatial predicate selectivity is lower than the preset threshold, thus providing a basis for subsequent path search strategy selection. If the spatial predicate selectivity is lower than the preset threshold, proceed to step S220; otherwise, proceed to step S230.

[0111] S220. When the spatial predicate selectivity is lower than the preset threshold, the knowledge graph is spatially filtered based on the path query information to obtain relevant subgraphs, and graph search is performed in the relevant subgraphs to determine all candidate paths between potential pollution source nodes and monitoring point nodes.

[0112] In this embodiment, when the spatial predicate selectivity is lower than a preset threshold At that time, the computing device can perform spatial filtering operations in the knowledge graph based on path query information to select subgraphs related to the affected area. Subsequently, in the obtained relevant subgraphs Within the scope, a spatially prioritized graph search operation is performed, starting from the potential pollution source node and traversing along the nodes and edges in the knowledge graph to determine all candidate paths between the potential pollution source node and the monitoring point node in the affected area.

[0113] S230. When the spatial predicate selectivity is higher than the preset threshold, perform a heuristic graph search on the knowledge graph based on the path query information to determine all candidate paths between potential pollution source nodes and monitoring point nodes.

[0114] In this embodiment, when the spatial predicate selectivity is higher than a preset threshold At this time, the computing device can perform a graph-first heuristic graph search within the global knowledge graph based on path query information. Starting from the potential pollution source node, it traverses towards the monitoring point node, using the geographical distance or cumulative time difference between nodes as a heuristic function to prioritize the search for possible paths. Spatial matching verification is then applied to the generated candidate paths to determine all candidate paths between the potential pollution source node and the monitoring point node in the affected area. Simultaneously, the path depth is guaranteed not to exceed the maximum depth during the search process. The node expansion does not exceed the maximum breadth. This ensures that the complexity remains manageable even in the worst-case scenario.

[0115] In one embodiment of the present invention, a preset threshold is used. The range is 0.2–0.7. Preferably, a preset threshold is used. The range is 0.35–0.55. By determining the relationship between spatial predicate selectivity and a preset threshold based on the distribution density of potential pollution source nodes in the affected area, the computing device can adaptively select a more efficient path search strategy.

[0116] Furthermore, to improve path search efficiency, the computing device can perform candidate source pre-selection operations on potential pollution source nodes in the knowledge graph. By utilizing prior data such as pollution emission permits, bill of materials information, historical violation records, and enterprise production capacity and operating conditions, the candidate source scores are calculated, thereby narrowing down the set of candidate pollution sources and reducing the search space. Simultaneously, the computing device can also set dynamic edge weights for path segments. The edge weights can be updated over time slices to reflect the impact of changes in operating conditions, rainfall events, or other environmental factors. The updated edge weights are written to a temporary layer for prioritizing more probable propagation paths in subsequent path searches, thus improving the accuracy and interpretability of candidate paths while maintaining search controllability.

[0117] (III) Scoring and ranking of candidate paths.

[0118] In other words, after obtaining the candidate paths, each candidate path needs to be scored and sorted (i.e., step S300).

[0119] In one embodiment of the present invention, the scoring model includes a mechanism consistency term, an evidence support term, and an uncertainty penalty term; the mechanism consistency term is used to evaluate the consistency between each candidate path and the pollutant diffusion mechanism based on the edge attribute information of each candidate path; the evidence support term is used to evaluate the degree of conformity between the edge attribute information and event information of each candidate path; and the uncertainty penalty term is used to evaluate the degree of uncertainty of each candidate path in the determination process.

[0120] Specifically, such as Figure 6 As shown, step S300 may include the following steps:

[0121] S310. Each candidate path is evaluated using the mechanism consistency term, evidence support term, and uncertainty penalty term, respectively.

[0122] In this embodiment, the computing device can evaluate each path in the candidate pollution path set separately, including a mechanism consistency item, an evidence support item, and an uncertainty penalty item. The mechanism consistency item is defined as the average of the mechanism scores for each segment, and can be used to measure the degree of agreement between the path and the pollutant diffusion mechanism. The calculation formula is:

[0123]

[0124] in Weights for different mechanistic factors, The value range is 0.1-0.5, and the sum is 1. The default average of 0.25 can be used. For time and distance The attenuation coefficient, with a scale of . Order of magnitude , pollutants In the media The sensitivity coefficient in the model can be determined through environmental experiments or experience. This is a media compatibility indicator function; it is 1 if the contaminant can propagate through the medium, and 0 otherwise. This refers to the half-life of the pollutant in that medium. (Through...) The rationality of the path in terms of spatiotemporal decay and medium matching can be quantitatively evaluated.

[0125] Supporting evidence can be used to measure the degree of agreement between the path and the observed data, and can be obtained through recent trend changes of outliers. The difference in level between outlier values ​​and the background and the correlation between path-related source emission records and receptor observation data. The calculations are performed, taking into account penalties for inconsistent observations, and then processed using a normalization function to reflect the degree to which the observed data supports the path. Specifically, the evidence support term... The calculation formula is:

[0126]

[0127] in, It is a sigmoid normalization function. For each component coefficient, and All values ​​are non-negative, and can be assumed to be 1. This indicates a penalty for inconsistent observations.

[0128] The uncertainty penalty term quantifies the negative impact of uncertainty in path reasoning on the score. It can include the combination of confidence levels of each segment of the path and the normalized variance of uncertainty of each segment, calculated through a weighted combination. This term is used to penalize paths with high overall uncertainty. Specifically, the uncertainty penalty term... The calculation formula is:

[0129]

[0130] in, For the first The confidence level of a segment can be derived from a fusion of factors such as sensor accuracy, relation extraction reliability, or expert verification. This indicates the confidence level that all parts of the entire path are true; This is the normalized variance of the uncertainty of each segment of the path, which is calculated based on the measurement error or model residual of each segment and normalized between 0 and 1. These are the weighting coefficients. The value range is 0.4-0.8. The value range is 0.2-0.6, with a default value of 0.2-0.6. ). A higher score indicates lower overall path certainty and should be penalized. Therefore, by calculating these three scoring criteria for each candidate path, the system can comprehensively evaluate each path's performance in terms of mechanistic rationality, data support, and uncertainty control, providing a quantitative basis for subsequent path ranking.

[0131] In other embodiments of the present invention, path search can employ causal graphs or Bayesian networks to generate potential path priors, which are then refined using heuristic search; alternatively, spatiotemporal graph neural networks can be used to generate candidate paths or dynamic edge weight priors, with compliance and evidence verification performed using graph-space rules; or, physical mechanism approximations can be combined, such as solutions to lightweight hydrodynamic or diffusion equations, for mechanism consistency terms. Weighting is applied to enhance path search accuracy.

[0132] S320. Based on the evaluation results of the mechanism consistency item, evidence support item and uncertainty penalty item, the comprehensive score of each candidate path is determined by weighted combination.

[0133] In this embodiment, the computing device can determine the comprehensive score of each candidate path based on the evaluation results of the mechanism consistency item, the evidence support item, and the uncertainty penalty item through a weighted combination. During the weighted combination process, a weight parameter α can be introduced to control the relative weight between the mechanism consistency item and the evidence support item, and a weight parameter β can be introduced to control the influence of the uncertainty penalty item on the comprehensive score. Through this weighted combination method, when a candidate path has high consistency at the mechanism level, strong evidence support at the observation data level, and low overall uncertainty, the corresponding comprehensive score is high; conversely, when mechanism consistency is insufficient, data evidence is weak, or uncertainty is high, the corresponding comprehensive score is low. Specifically, the comprehensive score... The calculation formula is:

[0134]

[0135] The value of α ranges from 0.4 to 0.8, with a preferred value of 0.6, and the value of β ranges from 0.1 to 0.4, with a preferred value of 0.2.

[0136] Furthermore, when calculating the comprehensive score for each candidate path, the computing device can perform segment-by-segment confidence fusion of the confidence scores for each segment of the path. This confidence fusion can flexibly switch between various methods such as Dempster-Shafer evidence theory, Bayesian product, or logarithmic space accumulation, thereby comprehensively considering multi-source information such as sensor observation accuracy, relation extraction reliability, and expert verification, thus improving the robustness and reliability of the score. Further, during the candidate path ranking process, the computing device can introduce a multi-objective ranking strategy based on the comprehensive score (P), determining the priority of candidate paths through multi-objective weighting or Pareto ranking methods. This improves the scientific rigor and operability of path ranking while better serving environmental regulatory decision-making.

[0137] S330. Sort the candidate paths according to the scoring results.

[0138] In other embodiments of the present invention, the candidate path scoring may employ a learned-to-rank model, such as LambdaMART or XGBoostRank, with mechanistic consistency, evidence support, uncertainty assessment and their interaction features as input, or by directly fitting the path to the true probability P through a probabilistic objective and making a decision by minimizing a threshold or Bayesian cost. The present invention does not impose any limitations.

[0139] (iv) Path determination.

[0140] That is, based on the above scoring and ranking structure, candidate contamination paths are determined (i.e., step S400).

[0141] In step S400, the computing device can select the Top-k candidate paths from the scoring results, from highest to lowest, as the most likely pollution paths and pollution source candidates. The value of Top-k ranges from 3 to 10, with a preferred value of 5.

[0142] (v) Generation of traceability report.

[0143] That is, a source tracing report is generated based on the identified candidate pollution pathways (i.e., step S500).

[0144] Specifically, based on candidate pollution pathways, relevant pathway information is organized and summarized to generate source tracing reports, reflecting the correlation between candidate pollution pathways and abnormal environmental events.

[0145] More specifically, such as Figure 7 As shown, step S500 may include the following steps S510-S520:

[0146] S510. Based on a predefined set of rules, the emission allowance threshold is judged for each candidate pollution path in turn, and the evidence chain corresponding to each candidate pollution path is determined.

[0147] In this embodiment, for each candidate pollution path, the computing device can determine the emission allowance threshold based on a predefined set of rules for the pollution emission behaviors involved in the candidate pollution path. The rule set includes at least one of the following: emission limit rules based on national regulations, emission limit rules based on local regulations, and enterprise discharge permits, and is in the form of machine-executable rules. Specifically, the rules in the rule set are described using a condition-conclusion structure to characterize the relationship between pollutant types and corresponding concentration limits, special emission quota constraints applicable to specific time periods or regions, and constraints such as the location of emission outlets and permitted emission amounts specified in the discharge permit. When determining the emission allowance threshold for a candidate pollution path, the computing device matches the corresponding rules in the rule set one by one based on information such as the pollutant type, emission intensity, occurrence time, and spatial location contained in the candidate pollution path, and forms a judgment result accordingly. Simultaneously, the computing device can also associate the rule entries that trigger the judgment result, the matching conditions, and the corresponding data basis to construct an evidence chain corresponding to each candidate pollution path, which supports the subsequent display and interpretation of source tracing analysis results. Specifically, as shown... Figure 8 As shown, step S510 may include the following steps S511-S513:

[0148] S511. Based on the edge attribute information of each candidate pollution path, determine the corresponding emission rule from the rule set, and the emission allowable threshold corresponding to the emission rule.

[0149] In this embodiment, the edge attribute information includes at least the pollutant type, emission medium, emission scenario, region, and time attribute. Based on the edge attribute information of each candidate pollution path, the corresponding emission rule and the emission allowance threshold corresponding to the emission rule are determined from the rule set. When there are multiple potentially applicable emission rules or emission allowance thresholds in the rule set, they are filtered according to the preset priority principles in the rule set to determine the unique or most stringent emission rule or emission allowance threshold, and the emission allowance threshold corresponding to that emission rule is determined accordingly. For example, if there are both national and local standards, the version of the standard is dynamically determined according to the principles of jurisdictional priority (i.e., priority of the more stringent standard under its jurisdiction), time validity priority (i.e., priority of the most recently effective standard), and specific priority (i.e., priority of special provisions for specific pollutants or industries).

[0150] S512. The event information of each candidate pollution path is compared with the emission rules and emission allowance thresholds in turn to obtain the judgment result of whether emission is allowed.

[0151] In this embodiment, for the emission rules and emission allowance thresholds determined in step S511, the computing device sequentially acquires the event information corresponding to each candidate pollution path, and compares and analyzes the anomaly information, location information, and medium information contained in the event information with the corresponding emission rules and emission allowance thresholds to obtain a judgment result on whether emission is permitted. The judgment result is used to characterize whether the event information corresponding to the candidate pollution path meets the requirements of the emission rules and emission allowance thresholds, specifically including judgment conclusions of not exceeding the emission allowance threshold or exceeding the emission allowance threshold. The judgment conclusions are recorded in a structured form for subsequent processing and display, for example, "The concentration of a certain pollutant at emission outlet A involved in the path is XX mg / L, exceeding the applicable standard YY mg / L, and is judged to exceed the standard."

[0152] S513. Based on the topological relationships, event information and rule set between nodes in each candidate contamination path, determine the evidence chain corresponding to each candidate contamination path in sequence.

[0153] In this embodiment, the evidence chain is used to fully describe the basis for determining the emission allowance threshold. Its content includes: abnormal monitoring records obtained based on event information, associated emission sources or enterprises and their discharge permits determined based on the topological relationship between nodes, applicable rule clauses based on rule set matching, and corresponding standard thresholds, measured values, and other information. For example, the evidence chain can be represented as a structured data combination: {monitoring point record ID, permit ID, rule ID, standard threshold, measured value}. By generating such an evidence chain, the system can clearly present "where the abnormality was detected and based on which standard or permit was the determination made," thereby providing complete and traceable evidence support for the compliance judgment conclusion of each candidate pollution path, enhancing the reviewability and persuasiveness of the judgment results.

[0154] In other embodiments of the present invention, a rule-graph hybrid approach may also be used to express complex antecedent relationships using graph matching, or to implement multi-constraint reasoning using constraint satisfaction problems, thereby improving the flexibility and accuracy of emission allowance threshold judgment.

[0155] S520. Output the judgment results and evidence chain of each candidate pollution path and emission allowance threshold as a source tracing report, and determine action recommendations based on the source tracing report.

[0156] In this embodiment, the computing device can provide action suggestions based on the analysis results of the source tracing report. For example, if a company is found to be exceeding emission standards, it is suggested to "issue a rectification notice to company X and increase monitoring intensity." If the pollution source is unknown, it is suggested to "increase monitoring points in region Y." These suggestions are expressed in the form of structured disposal instructions, including action type, relevant rules and basis, evidence chain, and tracking ID, which facilitates integration with subsequent processes.

[0157] In step S500, when managing the rule set, the computing device can adopt a standard synchronization strategy, caching national standards, local standards, and watershed or park-level standards in a hierarchical manner, and prioritizing the synchronization of the latest effective standards according to the principles of legal jurisdiction, time validity, and specific priority. For expired or invalid standards, the computing device can automatically remove them to avoid misuse, thereby ensuring that the rule set and emission allowance thresholds are always up-to-date and traceable. Furthermore, when generating source tracing reports and action recommendations based on candidate pollution pathways, the computing device can generate structured disposal instructions based on rule templates or small models, such as lightweight LLMs, and assign a unique traceable ID to each instruction. The structured instructions include the action type, the emission rule basis involved, evidence chain information, and traceability ID, thereby ensuring that each action recommendation is not only operable but also traceable, facilitating the execution of subsequent regulatory processes.

[0158] In one embodiment of the present invention, the site pollution source tracing method based on knowledge graph further includes: obtaining feedback information on the source tracing report, and updating the knowledge graph and / or scoring model based on the feedback information and according to preset update constraints; wherein, the update constraints are used to constrain the update magnitude, update frequency and update traceability of the update process.

[0159] Specifically, the computing device can acquire feedback information from the source tracing report and, based on this feedback, update the knowledge graph and / or scoring model according to preset update constraints. For new monitoring data or manual verification feedback, such as confirming a candidate pollution path as a real pollution source or rejecting a candidate path, the computing device executes a controlled online update mechanism, incrementally writing the new information back to the knowledge graph and making minor adjustments to the parameters of the scoring model. In a concrete implementation, when a candidate path is confirmed by experts in real time, the system can increase the confidence level of the correlations within that path. If a candidate path is rejected, the confidence level of the correlation is reduced. The update process is subject to update constraints, including limiting the magnitude of individual parameter changes, controlling the update frequency, and ensuring the traceability of the update process. For example, the magnitude of a single parameter adjustment must not exceed a preset threshold. , The value range is 1–10%, with a preferred range of 3%. The cumulative number of updates is throttled within a preset time window. Each update operation, including parameter values ​​before and after the update, time, and reason, is recorded in the audit log library and a version snapshot is saved. Administrators can call the rollback interface as needed to restore the knowledge graph or scoring model to a specified historical version, ensuring that the model evolution process is auditable, traceable, and controllable. Through continuous lightweight updates, the computing device can gradually learn from historical cases and optimize the scoring model parameters, such as dynamically adjusting the weights α, β, or decay coefficient λ of each scoring item, thereby improving the source tracing accuracy in long-term operation and avoiding model drift. In addition, when the computing device performs updates to the knowledge graph or scoring model, it can adopt a semantic version number management mechanism and support blue-green deployment or canary release strategies to ensure the smooth launch of new versions. At the same time, the updated model can be validated based on the scoring replay benchmark set to evaluate the impact of the update on path scoring and candidate path ranking, thereby ensuring that the model evolution process is controllable and traceable.

[0160] The knowledge graph-based site pollution source tracing method of the present invention will be described in detail below through specific embodiments.

[0161] When downstream monitoring station R detects an abnormal increase in the concentration of a pollutant, the computing device initiates source tracing analysis to identify potential upstream pollution sources. The computing device searches the knowledge graph and identifies a candidate path S→A→B→R from the suspected source S through intermediate links (such as tributary inflow points or treatment facilities) to the receptor R. This path consists of three edges (…). It consists of ) segments, each containing a time difference. Spatial distance Media type and segment confidence Attributes such as connectivity and temporal sequence are considered. The computing device ensures that each pair of adjacent nodes in the path is geographically connected and chronologically ordered. For example, wastewater discharged from point S flows through point A, reaches point B in about 1 hour, and then reaches point R in about 2 hours.

[0162] The computing device then calculates a comprehensive score for the path, including mechanistic consistency C(P), uncertainty U(P), and evidence support E(P). When calculating mechanistic consistency, the computing device calculates the exponential time decay for each path segment. Spatial exponential decay Medium sensitivity weight and chemical half-life decay Then, the scores for each segment are summed according to the preset weights, and finally, the average of the scores for the three segments is taken to obtain C(P).

[0163] If the path is reasonable in terms of temporal, spatial, and media continuity, then C(P) is close to 1; otherwise, if there is an anomaly in a certain period of time or distance, C(P) will decrease. When calculating uncertainty, the computing device first subtracts the product of the confidence levels of each segment from 1 to obtain U1, then calculates the normalized variance U2 of the path parameters, and finally weights them to obtain U(P). If the confidence levels of each segment of the path are high, then U(P) is low; if there are links with significant uncertainty, U(P) will increase, thus deducting from the total score. Evidence supporting E(P) is obtained by the computing device combining the trend, level, and correlation indicators of the monitoring data, using a Sigmoid function mapping. If the pollutant concentration at point R continues to rise and is significantly higher than the background value, and the emission records of company S are highly correlated with the changes at point R, then E(P) is close to 1; if the evidence is insufficient or contradictory, E(P) approaches 0. Furthermore, the computing device synthesizes C(P), E(P), and U(P) into a final score Score(P) according to a weighted formula, where C(P) and E(P) contribute positive scores, and U(P) serves as a penalty deduction. For example, in this embodiment, assuming C(P) = 0.8, E(P) = 0.5, U(P) = 0.2, and weights α = 0.6 and β = 0.2, the final score Score(P) is:

[0164] .

[0165] The computing device simultaneously outputs the contribution results for each sub-item: mechanism consistency contribution 0.48, data evidence contribution 0.20, and uncertainty deduction 0.04, thereby explaining the scoring basis to regulatory personnel and improving decision-making transparency. Subsequently, the computing device uses a knowledge graph to determine compliance for emission events involved in the path and generate an evidence chain. The computing device associates the monitoring record M at the end of the path with the permit information P corresponding to the starting pollution source S, obtains the exceeding rule R1 and boundary violation rule R2 from the rule database, and determines the applicable standards and thresholds based on a dynamic standard selection strategy. The computing device executes the rule judgments sequentially: comparing the monitoring value with the standard limit; if it exceeds the threshold, it is determined to be exceeding the standard; checking whether the emission behavior is within the permitted range; if it does not comply, it is determined to be boundary violation. The computing device incorporates the triggered rule ID and related thresholds, monitoring values, and other information into the evidence chain to form a structured record to support verification and archiving. Regardless of whether the judgment result is non-compliant, the computing device generates a complete evidence chain EC, which includes information such as the monitoring record ID, permit ID, triggered rule ID, threshold and unit, and actual monitoring value, thereby ensuring that the judgment process is traceable and transparent.

[0166] Upon receiving feedback, the computing device temporarily stores the feedback data and processes it according to a preset throttling strategy. For positive feedback, the computing device appropriately increases the confidence level of the relevant path segments. Alternatively, the priority of corresponding rules can be increased; for negative feedback, the computing device can reduce the confidence of relevant edge segments or add corresponding constraints to avoid the recurrence of similar misjudgments, and record the sample as a negative sample for subsequent model adjustments. The computing device can also appropriately adjust the scoring weights or time / space constraint parameters based on the continuous deviation trend, ensuring that the parameter adjustment magnitude |Δ| does not exceed a preset upper limit, such as 5%, to avoid drastic model fluctuations. After the update is completed, the computing device generates a new version number and records the audit log. If an anomaly or poor effect occurs after the update, the computing device can restore to any historical stable version through the rollback interface, thereby ensuring the reliability and controllability of the update process. Through the above process, the computing device can automatically complete pollution source path identification, path scoring, compliance determination, evidence chain generation, and online model updates based on receiving real-time monitoring data, achieving highly automated, transparent, and traceable pollution source tracing analysis.

[0167] In summary, the knowledge graph-based site pollution tracing method of this invention utilizes the spatial constraints of a spatial database for initial geographical screening to determine path query information, effectively narrowing the subsequent search space of the knowledge graph. Then, based on the path query information, path search is performed through the semantic relationships of the knowledge graph to determine all candidate paths between potential pollution source nodes and monitoring point nodes, thereby effectively improving the accuracy and efficiency of candidate pollution path extraction. Furthermore, this application introduces a pre-trained scoring model to score and rank each candidate path, providing objective evidence for regulatory decisions and identifying the most likely pollution diffusion paths, thus providing quantitative decision support for environmental regulation.

[0168] A second aspect of this invention also provides a site pollution source tracing system based on knowledge graphs, such as... Figure 9 As shown, the site pollution source tracing system 1000 may include:

[0169] The query generation module 1100 is used to determine the path query information through a spatial database based on the event information of the environmental anomaly event when an environmental anomaly event occurs at the monitoring point.

[0170] The path search module 1200 is used to perform path search in the knowledge graph based on path query information to determine all candidate paths between potential pollution source nodes and monitoring point nodes.

[0171] The path scoring module 1300 is used to score each candidate path based on a pre-trained scoring model and rank the candidate paths according to the scoring results.

[0172] The path filtering module 1400 is used to select a preset number of candidate paths as candidate contaminated paths based on the sorting results, from high to low.

[0173] The report generation module 1500 is used to generate source tracing reports based on candidate pollution pathways.

[0174] Among them, the spatial database and knowledge graph are generated based on multi-source heterogeneous data related to environmental regulation.

[0175] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus provided in the above embodiments and the corresponding method embodiments belong to the same concept, and the specific implementation process can be found in the corresponding method embodiments, which will not be repeated here.

[0176] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "connected" or "linked" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship also changes accordingly.

[0177] The above are preferred embodiments of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A site pollution source tracing method based on knowledge graphs, characterized in that, include: When an environmental anomaly occurs at a monitoring point, the path query information is determined through a spatial database based on the event information of the environmental anomaly. Based on the path query information, a path search is performed in the knowledge graph to determine all candidate paths between potential pollution source nodes and monitoring point nodes. Based on a pre-trained scoring model, each candidate path is scored, and the candidate paths are ranked according to the scoring results. Based on the ranking results, a preset number of candidate paths are selected as candidate contaminated paths from high to low. A source tracing report is generated based on the candidate pollution pathways; The spatial database and the knowledge graph are generated based on multi-source heterogeneous data related to environmental regulation.

2. The site pollution source tracing method based on knowledge graphs according to claim 1, characterized in that, The generation of the knowledge graph and spatial database includes: The multi-source heterogeneous data is preprocessed; Entity relations are extracted from the preprocessed multi-source heterogeneous data to generate knowledge graph triples; Based on a predefined environment ontology model, entities and relations in the knowledge graph triples are inserted into a graph database to construct the knowledge graph. The spatial location information of the preprocessed multi-source heterogeneous data is obtained, and the spatial database is constructed based on the spatial location information.

3. The site pollution source tracing method based on knowledge graphs according to claim 1, characterized in that, When an environmental anomaly occurs at a monitoring point, based on the event information of the environmental anomaly, path query information is determined through a spatial database, including: Based on the event information, the set of affected areas of the environmental anomaly event is determined through the spatial database. The event information includes anomaly information, location information, and medium information. The event information and the set of affected areas are encapsulated into the path query information.

4. The site pollution source tracing method based on knowledge graphs according to claim 3, characterized in that, Based on the path query information, a path search is performed in the knowledge graph to determine all candidate paths between potential pollution source nodes and monitoring point nodes, including: Based on the path query information, a path search is performed in the knowledge graph according to preset search constraints to determine candidate paths between all potential pollution source nodes and monitoring point nodes in the affected area. The search constraints include temporal constraints, spatial / medium topology constraints, and search boundary constraints.

5. The site pollution source tracing method based on knowledge graphs according to claim 4, characterized in that, Based on the path query information, a path search is performed in the knowledge graph to determine all candidate paths between potential pollution source nodes and monitoring point nodes, including: Based on the distribution density of the potential pollution source nodes in the affected area, determine whether the spatial predicate selectivity is lower than a preset threshold; When the spatial predicate selectivity is lower than a preset threshold, the knowledge graph is spatially filtered based on the path query information to obtain relevant subgraphs, and graph search is performed in the relevant subgraphs to determine all the candidate paths between the potential pollution source node and the monitoring point node. When the spatial predicate selectivity is higher than a preset threshold, a heuristic graph search is performed on the knowledge graph based on the path query information to determine all the candidate paths between the potential pollution source node and the monitoring point node.

6. The site pollution source tracing method based on knowledge graphs according to claim 1, characterized in that, The scoring model includes a mechanism consistency term, an evidence support term, and an uncertainty penalty term. The mechanism consistency term is used to evaluate the consistency between each candidate path and the pollutant diffusion mechanism based on the edge attribute information of each candidate path. The evidence support item is used to evaluate the degree of consistency between the edge attribute information and the event information of each candidate path; The uncertainty penalty term is used to evaluate the degree of uncertainty of each candidate path during the determination process.

7. The site pollution source tracing method based on knowledge graphs according to claim 6, characterized in that, Based on a pre-trained scoring model, each candidate path is comprehensively scored, and the candidate paths are ranked according to the scoring results, including: Each candidate path is evaluated using the mechanism consistency term, the evidence support term, and the uncertainty penalty term, respectively. Based on the evaluation results of the mechanism consistency term, the evidence support term, and the uncertainty penalty term, a comprehensive score for each candidate path is determined by weighted combination. The candidate paths are sorted according to the scoring results.

8. The site pollution source tracing method based on knowledge graphs according to claim 7, characterized in that, A source tracing report is generated based on the candidate pollution pathways, including: Based on a predefined set of rules, emission allowance thresholds are sequentially determined for each candidate pollution path, and the evidence chain corresponding to each candidate pollution path is determined. The results of each candidate pollution path, the determination of the emission allowance threshold, and the chain of evidence are output as a source tracing report, and action recommendations are determined based on the source tracing report. The set of rules includes at least one of the following: emission limit rules based on national regulations, emission limit rules based on local regulations, and enterprise pollution discharge permits.

9. The site pollution source tracing method based on knowledge graphs according to claim 8, characterized in that, Based on a predefined set of rules, emission allowance thresholds are sequentially determined for each candidate pollution path, and the evidence chain corresponding to each candidate pollution path is determined, including: Based on the edge attribute information of each candidate pollution path, the corresponding emission rule and the emission allowance threshold corresponding to the emission rule are determined from the rule set. The event information of each candidate pollution path is compared with the emission rules and the emission allowance threshold in turn to obtain a judgment result on whether emission is allowed. The evidence chain corresponding to each candidate contaminated path is determined sequentially based on the topological relationship between nodes in each candidate contaminated path, the event information, and the rule set.

10. The site pollution source tracing method based on knowledge graphs according to claim 1, characterized in that, Also includes: Obtain feedback information on the source tracing report, and update the knowledge graph and / or the scoring model based on the feedback information and according to preset update constraints; The update constraints are used to constrain the update magnitude, update frequency, and update traceability of the update process.

11. A site pollution source tracing system based on knowledge graphs, characterized in that, include: The query generation module is used to determine path query information through a spatial database based on the event information of the environmental anomaly when an environmental anomaly occurs at the monitoring point. The path search module is used to perform path search in the knowledge graph based on the path query information to determine all candidate paths between potential pollution source nodes and monitoring point nodes. The path scoring module is used to score each candidate path based on a pre-trained scoring model and to rank the candidate paths according to the scoring results. The path filtering module is used to select a preset number of candidate paths as candidate contaminated paths based on the sorting results, from high to low. The report generation module is used to generate a source tracing report based on the candidate pollution paths; The spatial database and the knowledge graph are generated based on multi-source heterogeneous data related to environmental regulation.

Images