Knowledge graph-based methods, devices, and electronic equipment for tracing the source of maritime violations

Abstract

The present application relates to a kind of sea area illegal event tracing method, device and electronic equipment based on knowledge graph, belong to water traffic technical field, wherein, the method includes: based on sea area illegal event data set, construct target knowledge graph, determine illegal event knowledge triple;According to illegal event knowledge triple, the feature matrix data set of actual sea area illegal data is established;With the feature matrix data set as input, the initial graph neural network model is trained, and the target graph neural network model of training perfection is obtained;According to target graph neural network model and real-time sea area data, sea area illegal event is traced back to source.The present application solves the technical problem that the tracing method of the prior art in sea illegal event is carried out on fixed node, and when new event is dynamically updated, it will lead to large amount of graph calculation, thereby leading to the technical problem of low tracing efficiency.

Description

Technical Field

[0001] This invention relates to the field of water transportation technology, and in particular to a method, apparatus, and electronic device for tracing the source of maritime violations based on knowledge graphs. Background Technology

[0002] While general-purpose knowledge graphs contain a large number of entities, their accuracy is low, making them unsuitable for precise application in maritime illegal incidents. This paper addresses this by using the information fusion results of multi-source heterogeneous data for entity identification, knowledge extraction, and knowledge fusion quality assessment. A knowledge graph model is constructed that integrates illegal incident ontology relationship identification, knowledge extraction and fusion, and knowledge process storage. This model overcomes the shortcomings of general-purpose knowledge graphs in terms of insufficient entity attributes and data patterns, and can accommodate different application scenarios and user needs for various illegal incidents.

[0003] Graph convolutional neural networks (CNNs) are a type of neural network used to efficiently process graph-structured data. Traditional CNN models can only process data with Euclidean structures. Nodes in a graph lack translation invariance, so they cannot be convolved using kernels of the same size. Compared to other graph neural networks, the GraphSAGE (Graph Sample and Aggregate) algorithm is characterized by its powerful inductive learning ability, enabling it to process unseen nodes and quickly generate embeddings for new nodes. Furthermore, through neighborhood sampling and diverse aggregation functions, GraphSAGE achieves efficient processing of large-scale graphs and captures rich features, thus demonstrating greater flexibility and scalability in practical applications.

[0004] Existing methods for clue mining and incident tracing have the following problems: 1. Some researchers have used event graphs to trace the origins of events, but they often only represent events in a structured way in terms of graphs without deep learning; 2. Current incident tracing is basically carried out on graphs with fixed nodes. When a new event occurs, the current graph needs to be relearned, which increases the computational complexity. Summary of the Invention

[0005] In view of this, it is necessary to provide a method, device and electronic device for tracing the source of maritime illegal events based on knowledge graphs, so as to solve the technical problem that the existing methods for tracing the source of maritime illegal events are carried out on fixed nodes, and the large amount of graph computation is caused when dynamically updating new events, resulting in low tracing efficiency.

[0006] To address the aforementioned problems, this invention provides a method for tracing the source of maritime illegal events based on knowledge graphs, comprising:

[0007] A target knowledge graph is constructed based on a dataset of illegal incidents in the sea area, and knowledge triples of illegal incidents are determined.

[0008] Based on the knowledge triplet of illegal events, a feature matrix dataset of actual maritime illegal data is established.

[0009] Using the feature matrix dataset as input, the initial graph neural network is trained to obtain the fully trained target graph neural network;

[0010] Based on target graph neural networks and real-time marine data, the source of illegal events in the marine area is traced.

[0011] In one possible implementation, the step of constructing a target knowledge graph based on a dataset of maritime illegal cases and determining knowledge triples of illegal events includes:

[0012] Obtain historical data on illegal incidents in the sea area. The historical data should include at least the information of the vessels involved, the location of the incident, the incident itself, the personnel involved, and the background of the incident.

[0013] Analyze historical data to determine the ontology relationship model;

[0014] The entity recognition module and relation extraction module of the marine illegality knowledge extraction model are trained based on the ontology relation model;

[0015] A well-trained maritime violation knowledge extraction model was used to extract knowledge from historical data and determine the knowledge triplet of violation events.

[0016] In one possible implementation, the step of establishing a feature matrix dataset of actual maritime illegal data based on the knowledge triplet of illegal events includes:

[0017] Historical data on maritime violations are categorized into general datasets and special datasets.

[0018] Based on the knowledge triples of the target knowledge graph, we can derive the general feature set corresponding to a general dataset and the special feature set corresponding to a special dataset.

[0019] A pre-defined synonym mining algorithm was used to perform synonym mining and replacement on both the general feature set and the special dataset to obtain text features.

[0020] By encoding the text features, a feature matrix dataset of actual maritime illegal data is obtained.

[0021] In one possible implementation, the encoding of text features to obtain a feature matrix dataset of actual maritime illegal data includes:

[0022] The text features are encoded using a pre-defined one-hot encoding method to obtain a feature matrix dataset of actual maritime illegal data.

[0023] In one possible implementation, training the initial graph neural network model using the feature matrix dataset as input to obtain a fully trained target graph neural network model includes:

[0024] Local domain information is obtained by sampling adjacent nodes based on the target knowledge graph.

[0025] The aggregated features are obtained by calculating the element-wise mean of the features of all sampled neighboring nodes.

[0026] The adjacency relationships of each node are determined based on the aggregation features, thus obtaining the target graph neural network model.

[0027] In one possible implementation, the sampling of adjacent nodes based on the target knowledge graph to obtain local domain information includes:

[0028] The initial graph neural network model samples the node adjacency list derived from the knowledge graph to obtain the local neighborhood information of the starting node.

[0029] In one possible implementation, the element-wise mean calculation of the features of all sampled neighboring nodes can be expressed by the following formula:

[0030]

[0031] in Represents a node In the Hidden states (or features) of a layer. Represents a node The set of neighboring nodes, This indicates the operation of calculating the mean. This indicates aggregation features.

[0032] Secondly, the present invention also provides a knowledge graph-based device for tracing the source of illegal activities in maritime areas, comprising:

[0033] The knowledge triple determination module is used to construct a target knowledge graph based on a dataset of illegal events in the sea area and determine the knowledge triples of illegal events.

[0034] The feature matrix building module is used to build a feature matrix dataset of actual maritime illegal data based on the knowledge triplet of illegal events.

[0035] The model training module is used to train the initial graph neural network model with the feature matrix dataset as input to obtain the fully trained target graph neural network model.

[0036] The source tracing module is used to trace the source of illegal events in the sea area based on the target graph neural network model and real-time sea area data.

[0037] Thirdly, the present invention also provides an electronic device, comprising: a processor and a memory;

[0038] The memory stores a computer-readable program that can be executed by the processor;

[0039] When the processor executes the computer-readable program, it implements the steps in the knowledge graph-based method for tracing the source of maritime illegal events as described above.

[0040] Fourthly, the present invention also provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the steps in the knowledge graph-based method for tracing the source of maritime illegal events as described above.

[0041] The beneficial effects of this invention are as follows: First, a knowledge graph is constructed, and corresponding knowledge triples are extracted from the dataset of illegal maritime incidents based on the knowledge graph. Then, feature matrix data of actual illegal maritime data is established based on the knowledge triples of illegal incidents. The initial graph neural network is trained using the feature matrix data of illegal maritime data, so that the nodes in the knowledge graph can be aggregated and learned through the graph neural network. It can learn new nodes in the knowledge graph without learning the entire graph, which improves computational efficiency and preserves the rich information of the graph data. Attached Figure Description

[0042] Figure 1 is a flowchart of an embodiment of the method for tracing the source of illegal events in the sea area based on knowledge graphs provided by the present invention;

[0043] Figure 2 for Figure 1 A flowchart of a method according to an embodiment of step S101;

[0044] Figure 3 A schematic diagram of the structure of an embodiment of the relationship extraction model in the knowledge graph-based method for tracing the source of illegal events in the sea provided by the present invention;

[0045] Figure 4 for Figure 1 A flowchart of a method according to an embodiment of step S102;

[0046] Figure 5 This is a schematic diagram of an embodiment of the knowledge graph-based method for tracing the source of illegal events in the sea area provided by the present invention, showing some features of the illegal event.

[0047] Figure 6 A schematic diagram of the structure of an embodiment of the GraphSAGE model sampling process in the knowledge graph-based method for tracing the source of illegal events in the sea provided by the present invention;

[0048] Figure 7 This is a schematic diagram of an embodiment of the knowledge graph-based device for tracing the source of illegal events in the sea provided by the present invention;

[0049] Figure 8 This is a schematic diagram of the operating environment of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0050] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0051] A knowledge graph is a graphical structure used to represent large amounts of knowledge and information, typically composed of entities (such as people, places, and things), concepts (such as categories and attributes), and the relationships between them. Its purpose is to help machines understand and process information, supporting intelligent reasoning and decision-making. The creation of knowledge graphs involves technologies such as data extraction, semantic modeling, and knowledge representation, and is widely used in artificial intelligence fields such as semantic search, intelligent question-answering systems, and recommender systems. Through knowledge graphs, data can be better organized, connected, and utilized, promoting the discovery and application of information.

[0052] GraphSAGE (Graph Sample and Aggregate) is a node representation learning algorithm for graph neural networks, designed to learn low-dimensional vector representations of nodes in a graph. The GraphSAGE algorithm consists of two core steps: sampling neighboring nodes and aggregating neighboring node features.

[0053] The sampling of neighbor nodes: GraphSAGE obtains the neighbor nodes of each node by sampling the nodes in the graph. The sampled neighbor nodes can be directly connected first-order neighbors or more distant higher-order neighbors. The purpose of this step is to obtain the local neighbor information of the nodes.

[0054] Aggregating Neighbor Node Features: Once the neighbor nodes of a node are obtained, the GraphSAGE algorithm updates the representation of the target node by aggregating the features of the neighbor nodes. The aggregation process can employ different methods, such as average pooling, max pooling, and LSTM, to integrate the features of the neighbor nodes and generate a new representation of the target node.

[0055] The GraphSAGE algorithm excels at efficiently processing large graph datasets while learning the representation of each node, capturing both structural and semantic information within the graph. This enables GraphSAGE to achieve superior performance in tasks such as node classification and link prediction, and it has been widely applied in fields such as social network analysis and recommender systems.

[0056] Tracing the source of maritime violations refers to the process of identifying and tracking illegal activities occurring in the marine environment. By tracing the source, the root cause, reason, and responsible parties of the illegal activities can be found, which helps to uncover the truth behind the incident and thus take appropriate measures to stop, punish, or correct it.

[0057] When tracing the source of maritime violations, the first step is to collect and analyze various information and data related to the violation, including key information such as the nature, location, and time of the possible illegal activity. The collected information is then further analyzed to identify valid evidence proving the illegal activity. This evidence may include document records, surveillance footage, and on-site investigations. Finally, on-site investigations are conducted to determine the specific details and scope of the illegal activity, ensuring the completeness and reliability of the evidence.

[0058] A specific embodiment of the present invention discloses a method for tracing the source of maritime illegal events based on knowledge graphs. Please refer to [link to relevant documentation]. Figure 1 ,include:

[0059] S101. Construct a target knowledge graph based on a dataset of illegal incidents in the sea area, and determine the knowledge triples of illegal incidents.

[0060] S102. Based on the knowledge triplet of illegal events, establish a feature matrix dataset of actual maritime illegal data;

[0061] S103. Using the feature matrix dataset as input, train the initial graph neural network to obtain the fully trained target graph neural network.

[0062] S104. Based on the target graph neural network and real-time sea area data, trace the source of illegal events in the sea area.

[0063] In this embodiment, a knowledge graph is first constructed, and corresponding knowledge triples are extracted from the dataset of illegal maritime incidents based on the knowledge graph. Then, feature matrix data of actual illegal maritime data is established based on the knowledge triples of illegal incidents. The initial graph neural network is trained using the feature matrix data of illegal maritime data. The graph neural network aggregates and learns nodes in the knowledge graph, and can learn new nodes in the knowledge graph without learning the entire graph. This improves computational efficiency and preserves the rich information of the graph data.

[0064] It should be noted that marine illegal incident datasets refer to datasets that record and analyze illegal activities occurring in the marine environment. These datasets may include various types of information, such as the geographical location and time of the illegal incident, the type of violation (e.g., illegal fishing, pollution), and the vessels or individuals involved. This data is of great significance for marine conservation, law enforcement, and policy making.

[0065] An illegal incident refers to an act that violates laws, regulations, or international agreements in the marine environment. These incidents can involve a variety of illegal activities, such as illegal fishing, unlicensed mining activities, marine pollution, illegal dumping of waste, and unlicensed navigation or anchorage. These illegal activities can have negative impacts on marine ecosystems, resource conservation, and human health.

[0066] A knowledge triple for an illegal event refers to the three basic elements used to describe an illegal event based on a knowledge graph: subject, predicate, and object. These triples can be used to construct a semantic representation of an illegal event, facilitating understanding and processing by computer programs and databases.

[0067] For example, if we want to describe an illegal event related to illegal fishing, a possible knowledge triple might look like this:

[0068] Subject: Fishing Boat A

[0069] Predicate: Illegal fishing was carried out.

[0070] Object: Marine Protected Area B

[0071] This ternary combination describes fishing vessel A engaging in illegal fishing in marine protected area B.

[0072] When constructing a semantic knowledge base for illegal events, a series of triples can be used to describe a complex event, including time, location, type of violation, subject and object involved, etc.

[0073] In some embodiments, the construction of a target knowledge graph based on a dataset of maritime illegal cases to determine knowledge triples of illegal events is described in the following reference: Figure 2 ,include:

[0074] S201. Obtain historical data on illegal incidents in the sea area, including at least the information of the vessels involved, the location of the incident, the incident itself, the persons involved, and the background of the incident.

[0075] S202. Analyze historical data to determine the ontology relationship model;

[0076] S203. Train the entity recognition module and relation extraction module of the maritime illegal knowledge extraction model based on the ontology relation model;

[0077] S204. Use a well-trained maritime violation knowledge extraction model to extract knowledge from historical data and determine the knowledge triplet of violation events.

[0078] In step S201, a web crawler script is written using Python to collect and store relevant maritime violation cases from the Internet. The crawled text data of maritime violation cases is deconstructed, and the characteristics of the maritime violation incidents, such as the vessels involved, the location of the incident, the time of the incident, the personnel involved, the background of the incident, and the consequences of the incident, are analyzed to determine the ontology relationship model. The crawled text data of maritime violation cases is cleaned and preprocessed. First, the illegal data in the txt format file is cleaned by deleting the table of contents, deleting the header, footer, and page number of the document, and removing meaningless spaces and line breaks, etc., to form the original dataset of maritime violation cases.

[0079] In step S202, text annotation is performed using Label Studio, and entities and relationships in a portion of the original dataset of maritime illegal cases are plotted using the "BIO" format.

[0080] In the BIO method, a piece of text is typically labeled BX, IX, O. X represents an entity; when this entity is time, X represents the time when the illegal event occurred.

[0081] B stands for "begin," a Chinese character-based named entity; I stands for "Inside," representing the middle and end parts of the named entity; O stands for "Other," representing a portion of an unspecified entity. See Tables 1 and 2 for examples.

[0082] Table 1: "BIO" labeling strategy (partial)

[0083]

[0084] Table 2: Examples of "BIO" annotation strategies

[0085]

[0086] In step S103, the "BIO" format maritime illegal knowledge extraction model is trained using the training set generated after text annotation work with Label Studio. The model's entity recognition module and interest extraction module are trained. After the extraction accuracy of both the entity recognition module and the interest extraction module reaches 85%, the maritime illegal knowledge extraction model performs dynamic knowledge extraction on the remaining maritime illegal knowledge dataset to form illegal knowledge triples.

[0087] In one specific embodiment, BERT and BI-LSTM models are used to extract knowledge about maritime violations. Figure 3As can be seen, the model consists of four parts: the input layer, the BERT-based word embedding layer, the feature extraction layer, and the classification layer.

[0088] (1) Input layer, mainly to perform data preprocessing, transforming the original data into word vector representations and inputting them into the model.

[0089] (2) The word embedding layer based on BERT: BERT inputs three types of vectors, namely, word vectors obtained from context information by Token Embedding (MLM task), sentence vectors used to process sentence pairs to judge semantic similarity (NSP task), and position vectors that record the sequence order by Position Embedding. BERT adds them together to obtain the vector representation of each word in the sentence.

[0090] (3) Feature extraction layer: The BiLSTM model is also used to extract features and perform semantic encoding. The forward LSTM unit can output a forward feature vector; the backward LSTM unit outputs a backward feature vector, and finally they are merged to obtain the feature vector output by BiLSTM.

[0091] (4) Classifier layer: The Softmax classifier is used to classify the feature vectors to obtain the prediction results of the model. After further labeling the data results extracted from the relation, the training set of the model is obtained.

[0092] Furthermore, after extracting the knowledge triples, in order to enable efficient knowledge retrieval in the graph structure, Neo4j is used for storage in this embodiment.

[0093] In some embodiments, the step of establishing a feature matrix dataset of actual maritime illegal data based on the knowledge triplet of illegal events is described in the following reference: Figure 4 ,include:

[0094] S401. Classify historical data on illegal incidents in the sea area to obtain general datasets and special datasets;

[0095] S402. Based on the knowledge triples of the target knowledge graph, derive the general feature set corresponding to the general dataset and the special feature set corresponding to the special dataset.

[0096] S403. Use a preset synonym mining algorithm to perform synonym mining and replacement on the general feature set and the special dataset respectively to obtain text features;

[0097] S404. Encode the text features to obtain a feature matrix dataset of actual maritime illegal data.

[0098] In step S401, considering the potential issues of matrix sparsity and excessive matrix dimensionality in the later-stage feature matrix, the accident features are divided into general accident features and specific features for different types of accidents based on the data types that may be received in practice. Please refer to [link / reference]. Figure 5 This diagram illustrates the construction of some general and special datasets. General accident features include common indicators such as environmental visibility, traffic flow density, whether vessel lights and shapes are normal, and whether vessel structure is normal. Different feature indicators are selected based on different types of accidents. For example, for smuggling accidents, indicators such as whether AIS was suddenly turned off and whether there was illegal mooring are added; for collision incidents, indicators such as whether the vessel made a sharp turn and whether its speed changed abruptly are added; and for illegal fishing incidents, whether the mooring area and fishing gear are compliant are included as special features. Finally, a feature indicator system is formed, combining general features with special features for different accidents.

[0099] In step S403, considering the differences in descriptions of different features in actual accident investigation reports—for example, regarding visibility, accident investigation reports often use different descriptions such as "good visibility," "visibility less than 5 meters," and "poor visibility"—the synonym knowledge base method in the synonym mining algorithm is used to replace the feature data exported from the knowledge graph with synonyms, completing the initial processing of the text information. Then, regular expressions are constructed to extract features from the text information. This method yields an event feature chain consisting of an accident node ID and text features, where "0" represents the event ID, the header contains features such as date and time, accident consequences, and event level, and the corresponding specific feature descriptions are the text information in the second column.

[0100] Furthermore, since most of the features in the accident investigation report are textual descriptions, which are difficult to use directly for training graph neural networks, after obtaining the features after synonym replacement, the one-hot encoding method is used to digitize various features and store them in an array in the form of 0s and 1s to form a feature matrix.

[0101] Furthermore, in step S103, the step of training the initial graph neural network model using the feature matrix dataset as input to obtain a fully trained target graph neural network model includes:

[0102] Local domain information is obtained by sampling adjacent nodes based on the target knowledge graph.

[0103] The aggregated features are obtained by calculating the element-wise mean of the features of all sampled neighboring nodes.

[0104] The adjacency relationships of each node are determined based on the aggregation features, thus obtaining the target graph neural network model.

[0105] It should be noted that in the node sampling part, the model will sample based on the node adjacency list exported from the knowledge graph to obtain the local neighborhood information of the starting node.

[0106] In hierarchical sampling, the sampling process builds upon the nodes sampled in the previous layer, continuing to sample their neighboring nodes. As the number of layers increases, the receptive field (i.e., the portion of the graph it can influence) of each starting node gradually expands. However, the number of nodes sampled in each layer typically decreases gradually to control computational complexity. The entire sampling process is as follows: Figure 6 As shown.

[0107] Furthermore, in this invention, the aggregation method adopted is mean aggregation. The element-wise mean of the features of all sampled neighbor nodes is calculated. That is, for each feature dimension, the average value of all neighbor nodes along that dimension is calculated. This new feature vector is the aggregated feature of the neighbor nodes. This process can be expressed mathematically as follows:

[0108]

[0109] in Represents a node In the Hidden states (or features) of a layer. Represents a node The set of neighboring nodes, This indicates the operation of calculating the mean. It's important to note that different aggregation methods can be selected based on the characteristics of the data.

[0110] Finally, the F1 score is used to evaluate the features of the newly embedded nodes. Formula 1.2 represents precision, 1.3 represents recall, and 1.4 represents the F1 score after considering both precision and recall.

[0111] (1.2)

[0112] (1.3)

[0113] (1.4)

[0114] After the features of the newly embedded node are aggregated and optimized, the features and adjacency relationships of the new node are input into the classifier, which can quickly obtain the type of the event.

[0115] Based on the above-described knowledge graph-based method for tracing the source of maritime illegal activities, this application also provides a knowledge graph-based device for tracing the source of maritime illegal activities. Please refer to [link to relevant documentation]. Figure 7 ,include:

[0116] The knowledge triple determination module 710 is used to construct a target knowledge graph based on a dataset of illegal events in the sea area and determine the knowledge triples of illegal events.

[0117] The feature matrix building module 720 is used to build a feature matrix dataset of actual maritime illegal data based on the knowledge triplet of illegal events.

[0118] The model training module 730 is used to train the initial graph neural network model with the feature matrix dataset as input to obtain a fully trained target graph neural network model.

[0119] The source tracing module 740 is used to trace the source of illegal events in the sea area based on the target graph neural network model and real-time sea area data.

[0120] like Figure 8 As shown, based on the above-described knowledge graph-based method for tracing the source of maritime illegal events, this invention also provides an electronic device, which can be a mobile terminal, desktop computer, laptop, handheld computer, server, or other computing electronic device. The electronic device includes a processor 810, a memory 820, and a display 830. Figure 8 Only some components of the electronic device are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0121] In some embodiments, memory 820 may be an internal storage unit of the electronic device, such as a hard disk or memory. In other embodiments, memory 820 may be an external storage device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, memory 820 may include both internal and external storage devices. Memory 820 is used to store application software and various types of data installed on the electronic device, such as program code installed on the electronic device. Memory 820 can also be used to temporarily store data that has been output or will be output. In one embodiment, memory 820 stores a knowledge graph-based maritime violation incident tracing program 840, which can be executed by processor 810 to implement the knowledge graph-based maritime violation incident tracing method of the various embodiments of this application.

[0122] In some embodiments, processor 810 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 820 or process data, such as executing a knowledge graph-based method for tracing the source of maritime illegal events.

[0123] In some embodiments, display 830 may be an LED display, a liquid crystal display, a touch-screen liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 830 is used to display information from the knowledge graph-based maritime violation tracing electronic device and to display a visual user interface. Components 810-830 of the electronic device communicate with each other via a system bus.

[0124] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0125] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for tracing a sea area illegal event based on a knowledge graph, characterized in that, The method comprises the following steps: Adopting BERT and BI An LSTM model is used to construct a target knowledge graph based on a sea area illegal event data set to determine illegal event knowledge triples; the sea area illegal event data set refers to a data set recording and analyzing illegal behaviors occurring in a marine environment, including geographical location, time, illegal type, involved ships, and personal information of illegal events; the illegal event knowledge triples refer to three basic elements of a illegal event based on a knowledge graph, including a subject, a relationship, and an object; According to the illegal event knowledge triple, a feature matrix data set of actual sea area illegal data is established; The feature matrix data set of actual sea area illegal data is established according to the illegal event knowledge triple, which comprises the following steps: The historical data of sea area illegal events are classified to obtain a general data set and a special data set; a general feature set corresponding to the general data set and a special feature set corresponding to the special data set are derived based on the knowledge triple of the target knowledge graph; the general features include environmental visibility, traffic flow density, whether the ship light type is normal, and whether the ship structure is normal; the special features include smuggling accidents, collision events and illegal fishing events; the smuggling accidents include whether the AIS is suddenly closed and whether the illegal parking occurs; the collision events include whether the ship makes a large turning and whether the ship speed suddenly changes; the illegal fishing events include whether the parking area is compliant and whether the fishing tool is compliant; The general feature set and the special data set are replaced by the preset synonym mining algorithm to obtain text features; The text features are encoded to obtain the feature matrix data set of actual sea area illegal data; The initial graph neural network model is trained by taking the feature matrix data set as input to obtain a trained target graph neural network model; The initial graph neural network model is trained by taking the feature matrix data set as input to obtain a trained target graph neural network model, which comprises the following steps: The local field information is obtained by sampling the adjacent nodes based on the nodes of the target knowledge graph; The aggregated features are obtained by performing element-wise mean calculation on the features of all sampled adjacent nodes; The adjacent relationship of each node is determined according to the aggregated features to obtain the target graph neural network model; The element-wise mean calculation on the features of all sampled adjacent nodes can be represented by the following formula: wherein representing a node In a first hidden state of a layer, representing a node a set of neighbor nodes of the node, representing an operation of averaging, representing an aggregated feature; The sea area illegal event is traced according to the target graph neural network model and the real-time sea area data.

2. The knowledge graph-based sea area illegal event tracing method according to claim 1, characterized in that, The target knowledge graph is constructed based on the sea area illegal event data set by using the BERT and BI-LSTM models to determine the illegal event knowledge triple, which comprises the following steps: The historical data of sea area illegal events are obtained, wherein the historical data at least includes the information of the ship involved in the event, the location of the event, the event, the personnel involved in the event and the background of the event; The historical data are analyzed to determine the ontology relationship model; The entity recognition module and the relationship extraction module of the sea area illegal knowledge extraction model are trained according to the ontology relationship model; The illegal event knowledge triple is determined by using the trained sea area illegal knowledge extraction model to extract knowledge from the historical data.

3. The knowledge graph-based sea area illegal event tracing method according to claim 1, characterized in that, The text features are encoded to obtain the feature matrix data set of actual sea area illegal data, which comprises the following steps: The text features are encoded by using the preset one-hot encoding method to obtain the feature matrix data set of actual sea area illegal data.

4. The knowledge graph-based sea area illegal event tracing method according to claim 1, characterized in that, The local field information is obtained by sampling the adjacent nodes based on the nodes of the target knowledge graph, which comprises the following steps: The initial graph neural network model samples the node adjacency list derived based on the knowledge graph to obtain the local field information of the starting node.

5. A sea area illegal event tracing device based on a knowledge graph, characterized in that, The method comprises the following steps: A knowledge triple determination module is configured to determine knowledge triples by using BERT and BI An LSTM model is used to construct a target knowledge graph based on a sea area illegal event data set, and determine illegal event knowledge triples; the sea area illegal event data set refers to a data set recording and analyzing illegal behaviors occurring in the marine environment, including the geographic location, time, illegal type, involved ships, and personal information of illegal events; the illegal event knowledge triples refer to three basic elements describing an illegal event based on a knowledge graph, including the subject, relationship, and object; The feature matrix establishment module is configured to establish a feature matrix dataset of actual sea area illegal data according to the illegal event knowledge triples; The feature matrix dataset of actual sea area illegal data is established according to the illegal event knowledge triples, and includes: The historical data of the sea area illegal event is classified to obtain a general dataset and a special dataset; a general feature set corresponding to the general dataset and a special feature set corresponding to the special dataset are derived based on the knowledge triples of the target knowledge graph; the general features include environmental visibility, traffic flow density, whether the ship light type is normal, and whether the ship structure is normal; the special features include smuggling accidents, collision events, and illegal fishing events; the smuggling accidents include whether the AIS is suddenly closed and whether illegal berthing occurs; the collision events include whether the ship makes a large turning and whether the ship speed suddenly changes; the illegal fishing events include whether the berthing area is compliant and whether the fishing tool is compliant; The preset synonym mining algorithm is used to replace the general feature set and the special dataset with synonyms to obtain text features; The text features are encoded to obtain the feature matrix dataset of the actual sea area illegal data; The model training module is configured to train an initial graph neural network model by taking the feature matrix dataset as input to obtain a trained target graph neural network model; The initial graph neural network model is trained by taking the feature matrix dataset as input to obtain a trained target graph neural network model, and includes: The nodes based on the target knowledge graph are used to sample adjacent nodes to obtain local field information; The features of all sampled adjacent nodes are calculated by element-wise mean to obtain aggregated features; The adjacent relationship of each node is determined according to the aggregated features to obtain a target graph neural network model; The element-wise mean calculation of the features of all sampled adjacent nodes can be represented by the following formula: wherein representing a node In a first hidden state of a layer, representing a node a set of neighbor nodes of the node, representing an averaging operation, representing an aggregated feature; The tracing module is configured to trace the sea area illegal event according to the target graph neural network model and real-time sea area data.

6. An electronic device, comprising: It includes: A processor and a memory; The memory stores a computer readable program that can be executed by the processor; The processor executes the computer readable program to implement the steps of the knowledge graph based sea area illegal event tracing method according to any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The computer readable storage medium stores one or more programs that can be executed by one or more processors to implement the steps of the knowledge graph based sea area illegal event tracing method according to any one of claims 1-4.

Images