A pollution site portrait construction method and system based on knowledge graph reasoning

Abstract

The application provides a pollution site portrait construction method and system based on knowledge graph reasoning, which comprises the following steps: collecting key data in a pollution site investigation report; obtaining a pollution land block structured data table of each pollution land block based on the pollution site investigation report, and obtaining a knowledge graph ontology structure; integrating the pollution land block structured data table to form a pollution site structured data table; obtaining a node table, a relationship table and an attribute table of each node in the knowledge graph ontology structure to obtain a triple table; establishing a graph database based on the triple table, and performing reasoning based on the graph database to obtain known pollution information of the pollution site; performing potential risk reasoning on the triple table based on a graph neural network to obtain potential pollution information of the pollution site; and obtaining a pollution site portrait based on the known pollution information of the pollution site and the potential pollution information of the pollution site. The method of the application evaluates the pollution site based on known facts and potential risk reasoning, and improves the accuracy of the evaluation.

Description

Technical Field

[0001] This invention relates to the field of environmental governance technology, specifically to a method and system for constructing a profile of a contaminated site based on knowledge graph reasoning. Background Technology

[0002] In recent years, pollution management at production sites has become a major challenge for environmental science and engineering. Various production activities result in a wide variety of pollutants that may be generated at contaminated sites, including heavy metals and organic pollutants. These pollutants can enter the soil through exhaust emissions, wastewater discharge, waste dumping, and groundwater seepage, causing pollution and secondary pollution to the site and its surrounding ecosystem. However, current contaminated site management faces many challenges. First, site data is multi-source and heterogeneous, meaning that a wide variety of data types need to be processed and analyzed, including structured, semi-structured, and unstructured data. Second, the relationships between data are often ambiguous and complex, increasing the difficulty of data analysis. Finally, factors such as resources, data availability, and uncertainty increase the complexity of soil pollution research. Therefore, the accuracy of environmental risk assessments is relatively low, reducing the efficiency and accuracy of contaminated site management. Summary of the Invention

[0003] In view of this, the present invention provides a method and system for constructing a contaminated site profile based on knowledge graph reasoning, which can improve the accuracy of environmental risk assessment.

[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0005] On the one hand, this invention provides a method for constructing a contaminated site profile based on knowledge graph reasoning, the method comprising:

[0006] Collect key data from the contaminated site investigation report for each contaminated site and store the collected key data in the contaminated site database. The key data is used to represent the associated factors of contamination for each contaminated site.

[0007] Based on the contaminated site investigation report, a structured data table for each contaminated site is obtained. The structured data table is used to record key data and the category to which each key data belongs. Key data under the same category is labeled with a tag.

[0008] The knowledge graph ontology structure is obtained based on the structured data tables of multiple contaminated sites. The knowledge graph ontology structure is used to represent the relationships between the tags corresponding to all key data.

[0009] The structured data tables of each contaminated site are integrated to form a structured data table of contaminated sites.

[0010] Obtain the node table, relation table, and attribute table of each node in the knowledge graph ontology structure, and populate the node table, relation table, and attribute table based on the key data in the structured data table of contaminated sites, with each node corresponding to a label;

[0011] A triplet table is obtained based on the node table, relation table, and attribute table. A graph database is built based on the triplet table. Reasoning is performed based on the graph database to obtain known pollution information of the contaminated site.

[0012] Based on graph neural networks, potential risk reasoning is performed on triplet tables to obtain potential pollution information of contaminated sites;

[0013] A profile of a contaminated site is obtained based on known contamination information and potential contamination information.

[0014] The contaminated site profiling method based on knowledge graph reasoning according to embodiments of the present invention extracts data with different structures from contaminated site reports by using a knowledge graph ontology structure and a contaminated site structured data table, and converts it into a unified format within both the knowledge graph ontology structure and the contaminated site structured data table, thereby improving query and retrieval efficiency. Furthermore, by using a graph database and a graph neural network to reason through the triplet table of the knowledge graph, known pollution information and potential pollution information of the contaminated site are obtained respectively, and combined to form a contaminated site profiling, effectively improving the accuracy of environmental risk assessment.

[0015] In one embodiment of the present invention, the method for constructing a contaminated site profile based on knowledge graph reasoning further includes:

[0016] When new key data is updated to the contaminated site database, the knowledge graph ontology structure of the key data and the structured data table of contaminated sites are updated based on the updated contaminated site database.

[0017] In one embodiment of the invention, the contaminated site database is updated in real time. By updating the contaminated site database in real time, the accuracy of environmental risk assessment can be further improved.

[0018] In one embodiment of the present invention, when new key data is updated to the contaminated site database, the knowledge graph ontology structure of the key data and the structured data table of contaminated sites are updated based on the updated contaminated site database, including:

[0019] The labels in the structured data table of contaminated sites are matched with the labels in the updated contaminated site database to obtain updated information, including newly added labels and / or key data added under the labels.

[0020] The information was updated by updating the labels in the structured data tables of nodes and contaminated sites, as well as the key data corresponding to the labels.

[0021] If newly added key data lacks a label, then the key data will be marked as having a missing label.

[0022] In one embodiment of the present invention, when new key data is updated to the contaminated site database, the knowledge graph ontology structure of the key data and the structured data table of contaminated sites are updated based on the updated contaminated site database, and the method further includes:

[0023] Based on the pollutant property table and industry classification table, the key data corresponding to the key data tags in the structured data table of contaminated sites are supplemented.

[0024] In one embodiment of the present invention, the structured data tables of each contaminated site are integrated to form a structured data table of contaminated sites, including:

[0025] After disambiguating the synonymous labels and key data corresponding to the labels in the structured data of each contaminated site, the structured data tables of each contaminated site are integrated to form a structured data table of contaminated sites.

[0026] In one embodiment of the present invention, potential risk reasoning is performed on a triplet table based on a graph neural network to obtain potential pollution information of a contaminated site, including:

[0027] Vector embedding is performed on the text in the triplet table to obtain the first feature vector of each key data, and the key data is mapped into a multidimensional space based on the first feature vector to form a heterogeneous graph;

[0028] The first feature vector is transformed into a heterogeneous data object using a graph neural network framework, and the heterogeneous data object is then passed into the transformation layer.

[0029] Two sets of graph neural network models are superimposed in the transform layer to form two heterogeneous graph neural network convolutional layers, and the encoder is composed of the two heterogeneous graph neural network convolutional layers.

[0030] The encoder updates the key data in the heterogeneous graph and the edges between the key data to obtain the second feature vector, and merges the second feature vectors of the key data to obtain the combined vector.

[0031] The decoder is used to decode the combined vector and output the link probabilities between the key data points;

[0032] Based on heterogeneous graphs and link probabilities, labels are filled in for key data points marked as missing to obtain potential pollution information about contaminated sites.

[0033] Secondly, the present invention also provides a contaminated site profiling system based on knowledge graph reasoning, comprising:

[0034] The data acquisition module is used to collect key data from the contaminated site investigation report for each contaminated site and store the collected key data in the contaminated site database. The key data is used to represent the related factors of the contamination status of each contaminated site.

[0035] The contaminated site structured data determination module is used to obtain a contaminated site structured data table for each contaminated site based on the contaminated site investigation report. The contaminated site structured data table is used to record key data and the category to which each key data belongs. Key data under the same category are labeled with a tag.

[0036] The ontology building module is used to obtain the knowledge graph ontology structure based on the labels in the structured data tables of multiple contaminated sites. The knowledge graph ontology structure is used to represent the relationship between the labels corresponding to all key data.

[0037] The contaminated site structured data integration module is used to integrate the structured data tables of each contaminated site to form a contaminated site structured data table.

[0038] The ontology information acquisition module is used to acquire the node table, relation table, and attribute table of each node in the knowledge graph ontology structure, and to populate the node table, relation table, and attribute table based on the key data in the contaminated site structured data table, where each node corresponds to a label.

[0039] The knowledge graph reasoning module is used to obtain a triple table based on the populated node table, relation table, and attribute table. It then builds a graph database based on the triple table and performs reasoning based on this database to obtain known pollution information about the contaminated site.

[0040] It is also used to perform potential risk reasoning on triple tables based on graph neural networks to obtain potential pollution information of contaminated sites;

[0041] The contaminated site profiling module is used to obtain a contaminated site profile based on known contamination information and potential contamination information.

[0042] In one embodiment of the present invention, the system further includes:

[0043] The update module is used to update the knowledge graph ontology structure and the structured data table of contaminated sites based on the updated contaminated site database when new key data is added to the contaminated site database.

[0044] The above-described technical solution of the present invention has at least one of the following beneficial effects:

[0045] The present invention provides a method and system for constructing contaminated site profiles based on knowledge graph reasoning. By extracting data of different structures from contaminated site reports through the knowledge graph ontology structure and the structured data table of contaminated sites, and converting them into a unified format, the efficiency of querying and retrieval is improved. In addition, the known pollution information and potential pollution information of the contaminated site are obtained by reasoning through graph database and graph neural network on the triple table of the knowledge graph, and combined to form a contaminated site profile, which effectively improves the accuracy of environmental risk assessment. Attached Figure Description

[0046] Figure 1 This is a flowchart of a method for constructing a contaminated site profile based on knowledge graph reasoning according to an embodiment of the present invention;

[0047] Figure 2 This is a flowchart illustrating a method for constructing a contaminated site profile based on knowledge graph reasoning, according to another embodiment of the present invention.

[0048] Figure 3 This is a knowledge graph ontology structure diagram in the contaminated site profiling method based on knowledge graph reasoning in an embodiment of the present invention.

[0049] Figure 4 This is a schematic diagram of the site profile assessment results in the contaminated site profile construction method based on knowledge graph reasoning according to an embodiment of the present invention.

[0050] Figure 5 This is a schematic diagram of the modules of the contaminated site profiling system based on knowledge graph reasoning, according to an embodiment of the present invention.

[0051] Attached reference numerals: 210, Data acquisition module; 220, Contaminated site structured data determination module; 230, Ontology establishment module; 240, Contaminated site structured data integration module; 250, Ontology information acquisition module; 260, Knowledge graph reasoning module; 270, Contaminated site profile determination module. Detailed Implementation

[0052] 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.

[0053] 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.

[0054] To facilitate understanding of the technical solution of the present invention, the technical problem to be solved by the present invention will be explained first.

[0055] In current environmental risk assessment processes, big data technologies are used to analyze contaminated site investigation reports. However, traditional big data technologies still face numerous challenges. Most valuable empirical data often appears in contaminated site investigation reports in structured, semi-structured, and unstructured formats. For example, structured data can appear in tables or databases, semi-structured data in JSON or XML formats, and unstructured data in text, images, etc. Traditional big data methods present significant challenges with these types of data. While traditional big data technologies can handle large-scale data and uncover hidden patterns, they are less efficient in processing this type of data, particularly in areas such as data integration, cleaning, and standardization. For instance, regarding data structure, site risk assessment requires processing and analyzing large amounts of data, including structured, semi-structured, and unstructured data. This data is stored in various site investigation reports with different structures, making data integration and subsequent query analysis very difficult. Regarding data depth, the length of the data chain obtained from different site reports varies. Some reports are very detailed, clearly outlining the pollution migration logic of the site, while others are vague in certain aspects, creating a data "black box" that makes it impossible to clearly explain the causes of pollution. Regarding the complexity of data logic, traditional big data technologies, especially machine learning methods, are limited in their interpretability, transparency, and ability to handle complex logical relationships when dealing with long logical chains and deductive reasoning. This makes them less suitable for scenarios requiring complex logic and reliance on the knowledge of experienced personnel. To address these issues, this invention provides a method and system for constructing contaminated site profiles.

[0056] The method for constructing a contaminated site profile according to an embodiment of the present invention will be described in detail below with reference to the accompanying drawings.

[0057] refer to Figure 1 , Figure 1 This is a flowchart illustrating a method for constructing a contaminated site profile based on knowledge graph reasoning, according to an embodiment of the present invention. Figure 1 As shown, the method for constructing a contaminated site profile based on knowledge graph reasoning in this embodiment of the invention includes:

[0058] S110. Collect key data from the contaminated site investigation report for each contaminated site and store the collected key data in the contaminated site database.

[0059] The contaminated site investigation report is used to record the investigation data of contaminated plots in the contaminated site. The investigation data includes key data, which is used to indicate the related factors of the contamination status of each contaminated plot, such as sulfides, mercury and other substances in the pollutants, and galvanizing process in the production process.

[0060] S120. Based on the contaminated site investigation report, obtain the structured data table of each contaminated site.

[0061] The structured data table for contaminated sites can be used to record key data and the category to which each key data belongs. Key data under the same category is labeled with a tag. For example, substances such as sulfides and mercury in pollutants are labeled as pollutants and can be used as row labels in the structured data table for contaminated sites. Key data such as sulfides and mercury can be used as data columns corresponding to the row labels.

[0062] S130. Obtain the knowledge graph ontology structure based on the labels in the structured data tables of multiple contaminated sites.

[0063] The knowledge graph ontology structure is used to represent the relationships between the tags corresponding to all key data. The extracted structured key data tags and their corresponding key data are stored in a structured data table of contaminated sites, and the relationships between these tags are represented by the knowledge graph ontology structure, thus establishing the semantic associations between the key data.

[0064] S140. Integrate the structured data tables of each contaminated site to form a structured data table of contaminated sites.

[0065] Because there is duplicate key data in the structured data tables corresponding to different contaminated sites, such as synonymous data within the same label, eliminating identical or synonymous data in each contaminated site's structured data table can avoid redundancy.

[0066] S150. Obtain the node table, relation table, and attribute table of each node in the knowledge graph ontology structure, and populate the node table, relation table, and attribute table based on the key data in the structured data table of the contaminated site.

[0067] By populating the node table, relationship table, and attribute table with key data from the structured data table of contaminated sites, specific key data is assigned to each label in these tables. For example, after populating the label "pollutant," pollutants such as mercury and sulfur will be listed under that label. The node table can include the entity's label, name, address, and attributes. The relationship table can include the relationships between these entities. The attribute table can include the entity's label, name, address, and attributes.

[0068] S160. Obtain a triplet table based on the node table, relation table, and attribute table. Build a graph database based on the triplet table and perform reasoning based on the graph database to obtain known pollution information of the contaminated site.

[0069] The triplet table allows the node table, relation table, and attribute table to be displayed graphically, which facilitates subsequent querying and reasoning.

[0070] S170. Based on graph neural networks, perform potential risk reasoning on triple tables to obtain potential pollution information of contaminated sites.

[0071] By reasoning about the connection probabilities between data in triples, potential pollution information of contaminated sites can be obtained, thereby improving the accuracy of environmental risk assessment.

[0072] S180. Obtain a profile of the contaminated site based on known pollution information and potential pollution information of the contaminated site.

[0073] The contaminated site profiling method based on knowledge graph reasoning according to embodiments of the present invention stores data from contaminated site reports through a knowledge graph ontology structure and a structured data table of contaminated sites, improving query and retrieval efficiency. Simultaneously, by mapping the nodes of the knowledge graph ontology structure to the tags of the contaminated site structured data table, semantic relationships are established while ensuring data consistency. Furthermore, by using a graph database and a graph neural network to reason through the triplet table of the knowledge graph, known pollution information and potential pollution information of the contaminated site are obtained respectively, and combined to form a contaminated site profiling, effectively improving the accuracy of environmental risk assessment.

[0074] The steps S110-S180 described above will be explained one by one below with reference to specific embodiments.

[0075] In step S110, each contaminated site investigation report includes investigation data for a contaminated plot within the contaminated site. Key data labels are used to indicate the related factors of contamination status for each plot. For example, key data can be factors directly related to production, such as industry, production area, site, production process, pollutants, and attributes; or geographical location factors such as province, city, and district. Key data can be labeled according to the user's specific needs.

[0076] In one embodiment of the present invention, the method for constructing a contaminated site profile based on knowledge graph reasoning of the present invention further includes: when new key data is updated to the contaminated site database, updating the key data knowledge graph ontology structure and the contaminated site structured data table based on the updated contaminated site database.

[0077] By updating the contaminated site database in real time, and updating the key data knowledge graph ontology structure and the structured data table of contaminated sites based on the updated contaminated site database, the accuracy of environmental risk assessment can be effectively improved.

[0078] In step S120, a structured data table of contaminated sites is obtained for each contaminated site based on the contaminated site investigation report, which may include steps S121-S122:

[0079] S121. Extract the key data and tags of the key data from each contaminated site investigation report;

[0080] S122. Based on key data and corresponding labels, establish a structured data table for each contaminated site. The key data is used to populate the data column corresponding to each label in the structured data table of the contaminated site.

[0081] refer to Figure 2 , Figure 2 This is a flowchart illustrating the method for constructing a contaminated site profile based on knowledge graph reasoning, according to an embodiment of the present invention. Figure 2 As shown, key data from each contaminated site in the investigation report can be manually extracted and tagged. For example, tags can be added to indicate the site, production process, and industry of the contaminated site. Then, the tags and corresponding key data are entered into a structured data table for each contaminated site. Tags serve as row labels, and key data serves as the corresponding data columns. This structured data table visually represents the nodes in the knowledge graph associated with the row labels, improving query and retrieval efficiency.

[0082] In step S130, the knowledge graph ontology structure is obtained based on the tags in the structured data tables of multiple contaminated sites, and may also include S131-S132:

[0083] S131. Extract the labels of key data from the structured data table of contaminated sites;

[0084] S132. Establish a knowledge graph ontology structure based on tags.

[0085] refer to Figure 3 , Figure 3 This is a knowledge graph ontology structure diagram in the contaminated site profiling method based on knowledge graph reasoning, as described in this embodiment of the invention. Figure 3 As shown, tags for key data in the structured data table of contaminated sites can be extracted, or tags for key data in the contaminated site investigation report can be extracted manually. These tags form a preliminary knowledge graph ontology structure, which can include multiple nodes, such as site, production process, and industry. Nodes correspond to tags in the structured data table of contaminated sites, representing the categories of key data. Nodes are interconnected by edges, which represent relationships between nodes; for example, a site node is connected to a production process node via an edge indicating "possesses". Furthermore, nodes can include attributes; for example, the attribute of a site node could be area.

[0086] In one embodiment of the present invention, when new key data is updated to the contaminated site database, the knowledge graph ontology structure of the key data and the structured data table of contaminated sites are updated based on the updated contaminated site database, including:

[0087] The updated contaminated site database is used to match the labels in the structured data table of contaminated sites to obtain updated information. The updated information is then used to update the nodes, the labels in the structured data table of contaminated sites, and the key data corresponding to the labels. The updated information includes newly added labels and / or the key data corresponding to the newly added labels.

[0088] In other words, matching can be performed in the contaminated site database using tags corresponding to nodes, and information associated with those tags can be extracted according to relationships. For example, the "site name" tag can be used to create a relationship index in the contaminated site database, extracting all data from the "site name" and "pollutant" rows, thereby improving query and retrieval efficiency. Then, all data from these two rows can be updated in the knowledge graph ontology structure and the contaminated site structured data table. By updating nodes, tags, and key data corresponding to tags using update information, the accuracy of environmental risk assessment can be further improved.

[0089] In some embodiments of the present invention, if updated key data lacks a label, the data column corresponding to the key data is marked as missing. Specifically, since there are cases where the updated information of contaminated sites stored in the contaminated site database has missing labels, when integrating the key data of the contaminated site into the contaminated site structured data table, the data column corresponding to the key data can be marked as missing.

[0090] In one embodiment of the present invention, the data columns corresponding to key data tags in the structured data table of contaminated sites can be completed based on pollutant information and industry information extracted from the pollutant property table and industry classification table.

[0091] By manually extracting pollutant and industry information from the "National Hazardous Waste List" and the "National Economic Standard Table" to complete the key data corresponding to the label, the nature of the node corresponding to the label can be displayed intuitively.

[0092] In step S140, the structured data tables of each contaminated site are integrated to form a structured data table of contaminated sites. This includes: disambiguating the synonymous labels and key data corresponding to the labels in the structured data of each contaminated site, and then integrating the structured data tables of each contaminated site to form a structured data table of contaminated sites.

[0093] By manually integrating and disambiguating the synonymous labels and corresponding key data in the structured data of each contaminated site, such as combining the synonymous labels "potential pollutants" and "other pollutants," the accuracy of environmental risk assessment can be improved.

[0094] In some embodiments of the present invention, the knowledge graph ontology structure can be evaluated to confirm whether there are any missing relationships between nodes. If there are missing relationships, it is necessary to verify the contaminated site investigation report to modify the relationships between nodes, or to reduce data noise by deleting non-key nodes, thereby further improving the accuracy of environmental risk assessment.

[0095] like Figure 2As shown, in step S150, a Python script can be used to extract data from the contaminated site structured data table according to the node relationships in the knowledge graph ontology structure, and organize them into a node table, a relationship table, and an attribute table. The node table records the labels corresponding to the nodes and the key data corresponding to the labels; the relationship table records the relationships between the labels; and the attribute table records the attributes of the labels and the attributes of the key data. The row labels in the node table can include an entity label (Entity_label), an entity name (Entity_name), an entity number (Entity_id), and a first entity attribute (Properties1). The entity is the key data under the label in the contaminated site structured data table corresponding to the node in the knowledge graph ontology structure. The row labels in the relation table can include the subject entity label (h_label), subject entity name (h_name), subject entity ID (h_id), object entity label (t_label), object entity name (t_name), object entity ID (t_id), and relation. Subject entities and object entities are related through relations. For example, in the knowledge graph ontology structure, there are subject entity labels "site" and object entity labels "production process," and the relationship between them is "use." The row labels in the attribute table can include entity labels (Entity_label), entity names (Entity_name), entity IDs (Entity_id), and second entity attributes (Properties1). The first entity attribute and the second entity attribute can be used to represent the same attribute, but the first entity attribute has its own label. For example, if site A has the attributes "Province A, City B," then the first entity attribute is "Province: Province A, City: City B"; and the second entity attribute is "Province A|City B." By setting the first entity attribute and the second entity attribute separately, data in different formats can be input in subsequent reasoning.

[0096] After obtaining the node table, relation table, and attribute table for each node based on the knowledge graph ontology structure, these tables can be populated with key data from the contaminated site structured data table. Taking the node table as an example, the entity label can be a pollutant, the entity name can be mercury, the entity number can be 1, and the first entity attribute can be mercury-contaminated water. By filling the node table, relation table, and attribute table with key data from the contaminated site structured data table, subsequent reasoning about known and potential pollution information of the contaminated site becomes easier.

[0097] In step S160, the node table, relation table, and attribute table can be merged in the form of "h, t, r", i.e., "head_entity, tail_entity, relation", to obtain a triple table. In the triple table, h represents the "subject entity", t represents the "object entity", and r represents the "relation". The "subject entity" and the "object entity" are related through the "relation".

[0098] Specifically, as shown in Table 1, the row labels in the triplet table can include the subject entity ID (h_id), subject entity label ID (h_nid), subject entity label (h_label), subject entity name (h_name), and subject entity attribute (h_pro) for the subject entity h; the object entity ID (t_id), object entity label ID (t_nid), object entity label (t_label), object entity name (t_name), and object entity attribute (t_pro) for the object entity t; and the relation (relation) and relation name (relation_name) for the relation r. Here, the entity is the key data under the label in the structured data table of contaminated sites corresponding to the nodes in the knowledge graph ontology structure. The subject entity ID and object entity ID are the numbers of the entity in the label. For example, if the subject entity label is "site," and the subject entity names are "site A" and "site B," the subject entity ID of site A is number 0 in the site label, and the subject entity ID of site B is number 1 in the site label. The numbers in different labels all start from 0. The entity label number and object label number are the entity's numbers in all key data.

[0099] For example, the main entity can be a site, and the object entity can be a pollutant. The relationship between the site and the pollutant can be "possessing," indicating that the site contains the pollutant. Further, as shown in row 1 of Table 1, the main entity of the site is named "Site A," the object entity of the pollutant is named "1,2-dichloroethane," and the attribute of the object entity used to indicate its properties is "water pollution." The relationship between "Site A" and "1,2-dichloroethane" is named "dissolved," indicating that "1,2-dichloroethane" is "dissolved" in "Site A." Similarly, row 2 of Table 1 indicates that "lead" is "deposited" in "Site B."

[0100] Table 1. Triplet Table

[0101]

[0102] Furthermore, information from node tables, relation tables, and attribute tables can be used to populate triple tables, thereby establishing a graph database for key data based on the triple tables. For example, a Neo4j graph database can be established, and then the Cypher language in the Neo4j database can be used to query and reason about the subject entity data and object entity data in the triple tables, thereby obtaining known pollution information about the contaminated site. For example, the known pollution information obtained could be: "A certain plot of land belongs to a certain industry, and both the plot of land and the industry have a certain production process. This production process uses a certain facility or equipment, which is prone to oil leakage. The oil leaks onto the ground, and because the ground has not taken anti-seepage measures, the oil enters the soil, causing soil pollution." This leads to the inference that the site is soil polluted and the pollutant is oil.

[0103] In step S170, potential risk reasoning is performed on the triplet table based on graph neural network to obtain potential pollution information of contaminated sites, including S171-S176.

[0104] S171. Perform vector embedding on the text in the triplet table to obtain the first feature vector of each key data, and map the key data in a multidimensional space according to the first feature vector to form a heterogeneous graph.

[0105] In one embodiment of the present invention, the SentenceTransformer or word2vec framework can be used to embed the text in the triplet table into vectors. The SentenceTransformer or word2vec framework can embed the text based on the relations in the text in the triplet table to obtain the first feature vector of each entity, and then map all entities in a multidimensional space to form a heterogeneous graph.

[0106] S172. The first feature vector is transformed into a heterogeneous data object through a graph neural network framework, and the heterogeneous data object is passed into the transformation layer.

[0107] In one embodiment of the present invention, the first feature vector is converted into a heterogeneous data object using the PyG (Pytorch_Geometric, PyG) framework or the DGL (DeepGraph Library, DGL) framework. The heterogeneous data object is a heterogeneous data type. By converting the first feature vector into a heterogeneous data object, it is easier to predict graph links in the heterogeneous graph in the future.

[0108] S173. Two sets of graph neural network models are superimposed in the transform layer to form two heterogeneous graph neural network convolutional layers, and the encoder is composed of the two heterogeneous graph neural network convolutional layers.

[0109] In one embodiment of the present invention, a heterogeneous graph neural network convolutional layer (HeteroGNN) can be formed by stacking GraphAttention Network (GAT) and GraphSample and Aggregated (GraphSAGE) convolutional layers from the PyG framework onto a Transformer layer. The encoder module consists of these two heterogeneous graph neural network convolutional layers. The encoder module can update entities in the heterogeneous graph based on the GAT (or GraphSAGE) layer and the Transformer layer.

[0110] S174. The encoder updates the key data in the heterogeneous graph and the edges between the key data to obtain the second feature vector, and merges the second feature vectors of the key data to obtain the combined vector.

[0111] In one embodiment of the invention, each key data point, i.e., the second feature vector of each entity, can be a weighted combination of the first feature vectors of all its neighboring entities, where the weights are calculated by a GAT (or GraphSAGE) layer. Then, the outputs of all different types of edges are aggregated according to the entity's label to obtain the second feature vector of each entity.

[0112] S175. Use a decoder to decode the combined vector and output the link probability between the entities corresponding to each node.

[0113] In one embodiment of the present invention, a Multi-Layer Perceptron (MLP) algorithm or a Complex Embeddings (ComplEx) algorithm can be used as a decoder to decode the other. This allows for the prediction of whether an edge exists between two entities in the heterogeneous graph. For example, for each pair of entities (i,j), a feature vector formed by concatenating the second feature vectors of the two entities can be received. The link probability is then output by the MLP. A link probability between 0 and 1 indicates the probability that an edge exists between entities i and j, i.e., the link probability between i and j.

[0114] S176. Based on heterogeneous graphs and link probabilities, complete the labels of key data marked as missing to obtain potential pollution information of contaminated sites.

[0115] By using heterogeneous graphs and link probabilities, graph link prediction is performed on key data marked as missing labels. Entities adjacent to the key data with missing labels are found, and their labels are obtained. Then, the labels of the entities with missing labels are completed based on semantic relationships. By completing the labels of the entities with missing labels, potential pollution information of contaminated sites can be obtained, enabling potential pollution risk inference.

[0116] Step S170 will be described below through specific embodiments.

[0117] In a heterogeneous graph, each entity includes an initial feature vector, i.e., the first feature vector. An encoder can be used to aggregate the first feature vectors of each entity to obtain the second feature vector for each entity. Taking entities i and j as an example, after aggregation, the second feature vectors hi and hj for entities i and j, respectively, can be obtained. These two second feature vectors can be combined by performing element-wise difference and element-wise product calculations. The formula for calculating the difference method is as follows:

[0118] d = hi - hj

[0119] The formula for calculating the dot product is:

[0120] p=hi⊙hj

[0121] The merging formula is:

[0122] z = [d; p]

[0123] Here, ";" indicates a connection. Then, z can be input into a multilayer perceptron (MLP). The MLP learns to predict whether a link exists between two entities based on z. z contains the difference and dot product between the features of the two nodes, providing rich information for link prediction. The link probability is output in the last layer of the MLP, and the link probability is between 0 and 1.

[0124] In step S180, a profile of the contaminated site can be obtained based on known pollution information and potential pollution information of the contaminated site, with reference to... Figure 4 , Figure 4 This is a schematic diagram illustrating the site profile assessment results in the knowledge graph-based reasoning-based contaminated site profile construction method of this invention. Figure 4As shown, the known pollution information of the contaminated site obtained in step S160, the potential pollution information of the contaminated site obtained in S170, and the nodes in the knowledge graph can be combined to create a profile to provide relevant pollutant information, thereby offering users feasibility suggestions, such as relocation conditions. Furthermore, referring to Table 2, which is a production process-pollutant-generation probability table, the link probabilities between entities in step S160 can be intuitively displayed through data analysis. For example, the link probability between the entities "production process" and "pollutant" can be presented through data analysis, thus showing the pollutant generation probability of different production processes. This improves the intuitiveness of the profile and the accuracy of user decision-making.

[0125] Table 2 Production Process - Pollutant - Generation Probability Table

[0126] Production process pollutants Probability of occurrence Nickel plating production process Metal shavings 92％ Nickel plating production process Metal scraps 91％ Nickel plating production process alkaline wastewater 85％ Nickel plating production process sulfuric acid mist 80％ Nickel plating production process lead 55％ Nickel plating production process 2,3-Dimethyl 51％

[0127] In one embodiment of the present invention, after obtaining a profile of a contaminated site, the knowledge graph ontology structure can be iterated based on user feedback, thereby improving the accuracy of environmental risk assessment.

[0128] It should be noted that algorithms or steps not mentioned in this invention can be referred to in the prior art, and will not be described in detail here.

[0129] This invention also provides a system for constructing a contaminated site profile based on knowledge graph reasoning, with reference to... Figure 5 , Figure 5 This is a schematic diagram of the modules of the contaminated site profiling system based on knowledge graph reasoning, according to an embodiment of the present invention. Figure 5As shown, the system includes: a data acquisition module 210, a contaminated site structured data determination module 220, an ontology building module 230, a contaminated site structured data integration module 240, an ontology information acquisition module 250, a knowledge graph reasoning module 260, and a contaminated site profiling module 270. Specifically, the data acquisition module 210 collects key data from the contaminated site investigation report for each contaminated site and stores the collected key data in the contaminated site database. The key data represents the correlation factors of the contaminated status of each contaminated site. The contaminated site structured data determination module 220 obtains a contaminated site structured data table for each contaminated site based on the contaminated site investigation report. The contaminated site structured data table records key data and the category to which each key data belongs. Key data under the same category is labeled with a tag. The ontology building module 230 obtains a knowledge graph ontology structure based on the tags in multiple contaminated site structured data tables. The knowledge graph ontology structure represents the relationship between the tags corresponding to all key data. The contaminated site structured data integration module 240 is used to... The structured data tables of each contaminated site are integrated to form a structured data table of the contaminated site; the ontology information acquisition module 250 is used to acquire the node table, relation table, and attribute table of each node in the knowledge graph ontology structure, and to populate the node table, relation table, and attribute table based on the key data in the structured data table of the contaminated site, wherein the nodes correspond to labels; the knowledge graph reasoning module 260 is used to acquire a triple table based on the populated node table, relation table, and attribute table, to build a graph database based on the triple table, and to perform reasoning based on the graph database to obtain known pollution information of the contaminated site, and is also used to perform potential risk reasoning on the triple table based on the graph neural network to obtain potential pollution information of the contaminated site; the contaminated site profile determination module 270 is used to obtain a profile of the contaminated site based on the known pollution information and potential pollution information of the contaminated site.

[0130] In one embodiment of the present invention, the system further includes an update module. The update module is used to update the knowledge graph ontology structure and the structured data table of contaminated sites based on the updated contaminated site database when new key data is updated to the contaminated site database.

[0131] The present invention relates to a method and system for constructing contaminated site profiles based on knowledge graph reasoning. This method extracts data from different structures in contaminated site reports through the knowledge graph ontology structure and a structured data table of contaminated sites, converting them into a unified format, thereby improving query and retrieval efficiency. Furthermore, by using a graph database and a graph neural network to reason through the triplet table of the knowledge graph, known pollution information and potential pollution information of the contaminated site are obtained respectively, and then combined to form a contaminated site profile, effectively improving the accuracy of environmental risk assessment.

[0132] 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 method for constructing a contaminated site profile based on knowledge graph reasoning, characterized in that, include: Collect key data from the contaminated site investigation report for each contaminated site and store the collected key data in the contaminated site database. The key data is used to represent the related factors of the contamination status of each contaminated site. Based on the contaminated site investigation report, a structured data table for each contaminated site is obtained. The structured data table is used to record the key data and the category to which each key data belongs. The key data under the same category is labeled with a tag. A knowledge graph ontology structure is obtained based on the tags in the structured data tables of multiple contaminated sites. The knowledge graph ontology structure is used to represent the relationship between the tags corresponding to all key data. The structured data tables of each contaminated site are integrated to form a structured data table of contaminated sites. Obtain the node table, relationship table, and attribute table of each node in the knowledge graph ontology structure, and populate the node table, relationship table, and attribute table based on the key data in the contaminated site structured data table, wherein the node corresponds to the tag; Based on the populated node table, relation table, and attribute table, a triplet table is obtained. A graph database is built based on the triplet table, and reasoning is performed based on the graph database to obtain known pollution information of the contaminated site. Based on graph neural networks, potential risk reasoning is performed on the triplet table to obtain potential pollution information of contaminated sites; A profile of the contaminated site is obtained based on the known pollution information and the potential pollution information of the contaminated site. The step of performing potential risk reasoning on the triplet table based on graph neural network to obtain potential pollution information of contaminated sites includes: embedding the text in the triplet table into vectors to obtain a first feature vector for each key data, and mapping the key data in a multidimensional space according to the first feature vector to form a heterogeneous graph. The first feature vector is transformed into a heterogeneous data object using a graph neural network framework, and the heterogeneous data object is then passed into the transformation layer. Two sets of graph neural network models are superimposed in the transform layer to form two heterogeneous graph neural network convolutional layers, and the encoder is composed of the two heterogeneous graph neural network convolutional layers. The encoder updates each key data point in the heterogeneous graph and the edges between each key data point to obtain a second feature vector, and merges the second feature vectors of each key data point to obtain a combined vector; The decoder is used to decode the combined vector and output the link probability between each of the key data points; Based on the heterogeneous graph and the link probability, the key data marked as missing labels are labeled to complete the labeling to obtain potential pollution information of the contaminated site.

2. The method for constructing a contaminated site profile based on knowledge graph reasoning according to claim 1, characterized in that, Also includes: When new key data is updated to the contaminated site database, the knowledge graph ontology structure and the structured data table of the contaminated site are updated based on the updated contaminated site database.

3. The method for constructing a contaminated site profile based on knowledge graph reasoning according to claim 2, characterized in that, When new key data is updated to the contaminated site database, the knowledge graph ontology structure and the structured data table of the contaminated sites are updated based on the updated contaminated site database, including: The labels in the structured data table of the contaminated site are matched with the labels in the updated contaminated site database to obtain updated information, including newly added labels and / or newly added key data under the labels. The update information is used to update the nodes in the knowledge graph ontology structure, the tags in the structured data table of the contaminated site, and the key data corresponding to the tags.

4. The method for constructing a contaminated site profile based on knowledge graph reasoning according to claim 3, characterized in that, The step of updating the nodes in the knowledge graph ontology structure, the tags in the structured data table of the contaminated land parcels, and the key data corresponding to the tags using the update information also includes: If the newly added key data lacks the aforementioned label, then the key data will be marked as having a missing label.

5. The method for constructing a contaminated site profile based on knowledge graph reasoning according to claim 2, characterized in that, Also includes: The key data in the structured data table of the polluted site are supplemented based on the pollutant property table and the industry classification table, using the pollutant information and industry information extracted from them.

6. The method for constructing a contaminated site profile based on knowledge graph reasoning according to claim 1, characterized in that, The process of integrating the structured data tables of each contaminated site to form a structured data table for contaminated sites includes: After disambiguating the synonymous labels and the key data corresponding to the labels in the structured data tables of each contaminated site, the structured data tables of each contaminated site are integrated to form a structured data table of contaminated sites.

7. A system for constructing a contaminated site profile based on knowledge graph reasoning, characterized in that, include: The data acquisition module is used to collect key data from the contaminated site investigation report for each contaminated site and store the collected key data in the contaminated site database. The key data is used to represent the related factors of the contamination status of each contaminated site. The contaminated site structured data determination module is used to obtain a contaminated site structured data table for each contaminated site based on the contaminated site investigation report. The contaminated site structured data table is used to record the key data and the category to which each key data belongs. The key data under the same category is labeled with a tag. An ontology building module is used to obtain a knowledge graph ontology structure based on the tags in the structured data tables of multiple contaminated sites. The knowledge graph ontology structure is used to represent the relationship between the tags corresponding to all key data. The contaminated site structured data integration module is used to integrate the structured data tables of each contaminated site to form a contaminated site structured data table. The ontology information acquisition module is used to acquire the node table, relationship table and attribute table of each node in the knowledge graph ontology structure, and to fill the node table, relationship table and attribute table respectively based on the key data in the contaminated site structured data table, wherein the node corresponds to the tag; The knowledge graph reasoning module is used to obtain a triple table based on the filled node table, the relation table and the attribute table, establish a graph database based on the triple table, and perform reasoning based on the graph database to obtain known pollution information of the contaminated site. It is also used to perform potential risk reasoning on the triple table based on the graph neural network to obtain potential pollution information of the contaminated site. The contaminated site profiling module is used to obtain a contaminated site profile based on the known contamination information and the potential contamination information of the contaminated site. The step of performing potential risk reasoning on the triplet table using a graph neural network to obtain potential pollution information of contaminated sites includes: embedding the text in the triplet table into vectors to obtain a first feature vector for each key data point, and mapping the key data into a multidimensional space based on the first feature vector to form a heterogeneous graph. The first feature vector is transformed into a heterogeneous data object using a graph neural network framework, and the heterogeneous data object is then passed into the transformation layer. Two sets of graph neural network models are superimposed in the transform layer to form two heterogeneous graph neural network convolutional layers, and the encoder is composed of the two heterogeneous graph neural network convolutional layers. The encoder updates each key data point in the heterogeneous graph and the edges between each key data point to obtain a second feature vector, and merges the second feature vectors of each key data point to obtain a combined vector; The decoder is used to decode the combined vector and output the link probability between each of the key data points; Based on the heterogeneous graph and the link probability, the key data marked as missing labels are labeled to complete the labeling to obtain potential pollution information of the contaminated site.

8. The contaminated site profiling system based on knowledge graph reasoning according to claim 7, characterized in that, Also includes: The update module is used to update the knowledge graph ontology structure and the structured data table of contaminated sites based on the updated contaminated site database when new key data is updated to the contaminated site database.

Images