Knowledge graph information extraction method based on LNG receiving station data

Abstract

The invention discloses a knowledge graph information extraction method based on LNG receiving station data, and belongs to the technical field of LNG receiving station data processing, and the method comprises the steps: carrying out the entity recognition of the preprocessed LNG receiving station data through an entity recognition model which is deployed in advance through deep learning, obtaining an entity of the preprocessed LNG receiving station data, and carrying out the recognition of the entity of the preprocessed LNG receiving station data; performing relation identification on the preprocessed LNG receiving station data by adopting a relation identification model pre-deployed by deep learning to obtain a corresponding relation of the preprocessed LNG receiving station data, and finally constructing a knowledge graph according to an entity and the relation of the preprocessed LNG receiving station data, so as to obtain the data of the LNG receiving station. According to the method, information can be efficiently and accurately extracted from data of the LNG receiving station, subsequent intelligent retrieval and data analysis are facilitated, the requirements for data interconnection and knowledge intelligence are met, equipment operation state monitoring, fault early warning and decision support are achieved, and management efficiency and safety are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of LNG receiving station data processing, and specifically relates to a knowledge graph information extraction method based on LNG receiving station data. Background Art

[0002] Liquefied natural gas (LNG) receiving terminals are important facilities for receiving, storing, and regasifying liquefied natural gas (LNG). These terminals are typically located in coastal areas and equipped with dedicated docks to receive LNG from LNG carriers. The received LNG is stored in specialized cryogenic tanks. After being converted to gas by gasification facilities, the LNG is transported via pipelines to the natural gas main network or directly to users. The terminals are also equipped with safety features to ensure operational safety. LNG receiving terminals play a key role in balancing energy supply and ensuring energy security. With the adjustment of energy structure, the amount of data on the construction and operation of LNG receiving terminals, as a clean energy source, is increasing. However, traditional data processing methods struggle to effectively extract useful information from this data. Therefore, an efficient and accurate information extraction method is needed. Summary of the Invention

[0003] The present invention provides a knowledge graph information extraction method based on LNG receiving station data, which is used to solve the problem that traditional data processing methods are difficult to effectively extract useful information from these data.

[0004] A knowledge graph information extraction method based on LNG receiving station data, comprising:

[0005] Acquiring LNG receiving station data, and preprocessing the LNG receiving station data to obtain the preprocessed LNG receiving station data;

[0006] Use the entity recognition model pre-deployed by deep learning to perform entity recognition on the pre-processed LNG receiving station data to obtain the entities in the pre-processed LNG receiving station data;

[0007] Use the relationship recognition model pre-deployed by deep learning to identify relationships in the pre-processed LNG receiving station data and obtain the relationships corresponding to the pre-processed LNG receiving station data;

[0008] Based on the entities and relationships of the pre-processed LNG receiving station data, a knowledge graph is constructed, and a graph database is used to store the knowledge graph to complete the extraction of knowledge graph information.

[0009] Furthermore, obtaining LNG receiving station data and preprocessing the LNG receiving station data to obtain the preprocessed LNG receiving station data includes:

[0010] Acquire LNG receiving station data, and perform deduplication on the LNG receiving station data to obtain deduplicated LNG receiving station data;

[0011] The deduplicated LNG receiving station data is sequentially segmented and stop words are removed to obtain pre-processed LNG receiving station data.

[0012] Furthermore, the entity recognition model is set to a BERT-BiLSTM-CRF composite model.

[0013] Furthermore, the method of pre-deploying an entity recognition model through deep learning includes:

[0014] Initializing hyperparameters of the entity recognition model to obtain a plurality of first hyperparameter codes;

[0015] Repeatedly adopting the boundary information fusion strategy, the multi-encoding co-evolution strategy and the reverse global information search strategy to train multiple first hyperparameter encodings to obtain the first optimal hyperparameter encoding;

[0016] The hyperparameters in the obtained first optimal hyperparameter encoding are used as the final hyperparameters of the entity recognition model, and the entity recognition model is deployed.

[0017] Furthermore, the relationship recognition model is set to a Word2Vec-GRU-Attention-PCNN composite model.

[0018] Furthermore, the method of pre-deploying a relationship recognition model through deep learning includes:

[0019] Initializing the hyperparameters of the relationship recognition model to obtain a plurality of second hyperparameter codes;

[0020] Repeatedly adopting the boundary information fusion strategy, the multi-encoding co-evolution strategy and the reverse global information search strategy to train multiple second hyperparameter encodings to obtain the second optimal hyperparameter encoding;

[0021] The hyperparameters in the obtained second optimal hyperparameter encoding are used as the final hyperparameters of the relationship recognition model, and the relationship recognition model is deployed.

[0022] Furthermore, a knowledge graph is constructed based on the entities and the relationships of the pre-processed LNG receiving station data, including:

[0023] Constructing knowledge based on the entities and the relationships in the preprocessed LNG receiving station data; wherein the knowledge represents entities, relationships between entities, and triples formed by entities;

[0024] Build a knowledge graph based on the acquired knowledge.

[0025] Furthermore, a graph database is used to store the knowledge graph, including: using Neo4j to store the knowledge graph.

[0026] Furthermore, it also includes:

[0027] Using deep learning to build an entity vector extraction model and an importance assessment model, and deploying the entity vector extraction model and the importance assessment model;

[0028] Entity relationship prediction is performed based on the deployed entity vector extraction model and importance assessment model to complete the knowledge graph.

[0029] Furthermore, entity relationship prediction is performed based on the deployed entity vector extraction model and importance assessment model to complete the knowledge graph, including:

[0030] The entity vector extraction model deployed is used to extract the feature vectors of the same knowledge entity and the feature vectors of the neighborhood entity corresponding to the entity in the knowledge graph;

[0031] Among them, the same knowledge entity feature vector represents the feature vector corresponding to other entities in the same knowledge as the entity, the neighborhood entity feature vector represents the feature vector of the entity in the knowledge that has at least one entity in common with the entity, knowledge represents the triple in the knowledge graph, and the same knowledge entity feature vector is different from the neighborhood entity feature vector;

[0032] According to the feature vector of the same knowledge entity corresponding to the entity in the knowledge graph, the first importance corresponding to the same knowledge entity is evaluated by the deployed importance evaluation model;

[0033] According to the feature vector of the neighborhood entity corresponding to the entity in the knowledge graph, the second importance corresponding to the neighborhood entity is evaluated by the deployed importance evaluation model;

[0034] Determine, based on the first importance corresponding to the same knowledge entity, first target knowledge corresponding to the entity in the knowledge graph, and obtain other knowledge corresponding to the same knowledge entity corresponding to the entity in the knowledge graph in the first target knowledge to obtain second target knowledge;

[0035] Determining final target knowledge and a final target entity in the second target knowledge according to the second importance corresponding to the neighborhood entity;

[0036] The knowledge graph is completed by reconstructing knowledge using entities in the knowledge graph, relationships in the final target knowledge, and the final target entities.

[0037] The present invention provides a knowledge graph information extraction method based on LNG receiving station data. The method performs entity recognition on the pre-processed LNG receiving station data through an entity recognition model pre-deployed by deep learning to obtain the entities of the pre-processed LNG receiving station data. Then, the method uses a relationship recognition model pre-deployed by deep learning to perform relationship recognition on the pre-processed LNG receiving station data to obtain the relationships corresponding to the pre-processed LNG receiving station data. Finally, a knowledge graph is constructed based on the entities and the relationships of the pre-processed LNG receiving station data. The method can efficiently and accurately extract information from the LNG receiving station data, convert the original data into structured information, and facilitate subsequent intelligent retrieval and data analysis, thereby meeting the needs of data interconnection and knowledge intelligence, realizing equipment operation status monitoring, fault warning and decision support, and thus improving management efficiency and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0039] Figure 1 A flowchart of a knowledge graph information extraction method based on LNG receiving station data provided in an embodiment of the present invention.

[0040] The above drawings have shown specific embodiments of the present invention, which will be described in more detail below. These drawings and the text description are not intended to limit the scope of the present invention in any way, but rather to illustrate the concept of the present invention to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0041] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0042] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0043] like Figure 1 As shown, an embodiment of the present invention provides a knowledge graph information extraction method based on LNG receiving station data, comprising:

[0044] S1. Acquire LNG receiving station data, and preprocess the LNG receiving station data to obtain the preprocessed LNG receiving station data;

[0045] The data of LNG receiving stations have significant characteristics, mainly reflected in rapid scale growth, high utilization rate, concentrated regional distribution and obvious seasonal fluctuations. As of the end of 2023, my country has built 28 LNG receiving stations with an annual receiving capacity of more than 116 million tons and a total storage tank capacity of 18 million cubic meters, ranking among the top in the world12. These receiving stations are mainly concentrated in coastal areas, and the operation of the receiving stations has obvious seasonal characteristics. During the peak gas consumption period in winter, gas storage facilities and gasification capacity are under great pressure, while in summer they are relatively stable. This volatility puts higher demands on energy scheduling and emergency supply. Therefore, extracting knowledge graph information from the data of LNG receiving stations can effectively improve data utilization efficiency and find useful data.

[0046] LNG receiving station data may contain a large amount of useless data. Therefore, it is necessary to preprocess the LNG receiving station data to obtain more accurate results. Preprocessing can include steps such as data cleaning, formatting, and feature engineering to ensure data quality and meet model requirements.

[0047] S2. Use the entity recognition model pre-deployed by deep learning to perform entity recognition on the pre-processed LNG receiving station data to obtain the entities in the pre-processed LNG receiving station data;

[0048] Using pre-deployed deep learning entity recognition models to identify entities in pre-processed LNG receiving terminal data can efficiently extract key information from the data. For example, Transformer-based architectures (such as BERT and GPT), trained on large amounts of annotated data, can identify named entities in text, such as locations, times, quantities, and organizations. In LNG receiving terminal data, these entities may include the terminal name, geographic location, storage tank capacity, regasification capacity, import volume, and supplier information. By inputting the pre-processed data into the pre-deployed entity recognition model, the model outputs the identified entities and their locations in the text. This entity information is crucial for subsequent data analysis and decision support. For example, analyzing the import volume and gas storage capacity of different receiving terminals can be used to optimize energy scheduling and emergency response strategies. The results of entity recognition can be further used to construct knowledge graphs, support visual analysis, or serve as input for other machine learning tasks, thereby improving insight into LNG terminal operations and enhancing management efficiency.

[0049] S3. Using a relationship recognition model pre-deployed by deep learning to perform relationship recognition on the pre-processed LNG receiving station data, and obtaining relationships corresponding to the pre-processed LNG receiving station data;

[0050] The pre-deployed deep learning relationship recognition model is used to perform relationship recognition on the pre-processed LNG receiving station data, which can efficiently extract the association information between entities in the data.

[0051] Relationship recognition models (such as those based on Transformer architectures) are trained to learn the relationship patterns between entities, enabling them to identify and classify relationships between entities in text. For example, in LNG receiving terminal data, the model can identify the supply relationship between "receiving terminal A" and "supplier B," or the capacity relationship between "storage tank C" and "gasification capacity D."

[0052] By feeding the preprocessed data into the model, the model outputs entity-relationship triples (e.g., "Receiving Terminal A - Supply - Supplier B"). This relationship information helps construct a knowledge graph, supporting data analysis and decision-making, such as optimizing energy scheduling and emergency supply strategies. The relationship identification results can also be used for further visualization analysis or as input for other machine learning tasks, improving insight into LNG terminal operations and enhancing management efficiency.

[0053] S4. Based on the entities and the relationships of the pre-processed LNG receiving station data, a knowledge graph is constructed, and a graph database is used to store the knowledge graph to complete the extraction of knowledge graph information.

[0054] The present invention provides a knowledge graph information extraction method based on LNG receiving station data. The method performs entity recognition on the pre-processed LNG receiving station data through an entity recognition model pre-deployed by deep learning to obtain the entities of the pre-processed LNG receiving station data. Then, the method uses a relationship recognition model pre-deployed by deep learning to perform relationship recognition on the pre-processed LNG receiving station data to obtain the relationships corresponding to the pre-processed LNG receiving station data. Finally, a knowledge graph is constructed based on the entities and the relationships of the pre-processed LNG receiving station data. The method can efficiently and accurately extract information from the LNG receiving station data, convert the original data into structured information, and facilitate subsequent intelligent retrieval and data analysis, thereby meeting the needs of data interconnection and knowledge intelligence, realizing equipment operation status monitoring, fault warning and decision support, and thus improving management efficiency and safety.

[0055] In an embodiment of the present invention, obtaining LNG receiving station data and preprocessing the LNG receiving station data to obtain the preprocessed LNG receiving station data includes:

[0056] Acquire LNG receiving station data, and perform deduplication on the LNG receiving station data to obtain deduplicated LNG receiving station data;

[0057] The deduplicated LNG receiving station data is sequentially segmented and stop words are removed to obtain pre-processed LNG receiving station data.

[0058] It is worth noting that the above preprocessing method is only a preferred choice for the implementation of the present invention, and other preprocessing methods can also be set to improve the data processing effect.

[0059] In an embodiment of the present invention, the entity recognition model is set to a BERT-BiLSTM-CRF (pre-trained language model-bidirectional long short-term memory network-conditional random field) composite model.

[0060] First, the input text is preprocessed through the pre-trained language model BERT to extract the dynamic word vector of the input text sequence. Then, the context information of the word vector is captured through the BiLSTM neural network, and finally the label sequence information is obtained by the CRF layer.

[0061] In an embodiment of the present invention, a method for pre-deploying an entity recognition model through deep learning includes:

[0062] Initializing the hyperparameters of the entity recognition model to obtain a plurality of first hyperparameter codes; for example, the hyperparameters may be randomly initialized between their corresponding upper and lower limits, and then encoded into vectors to determine the first hyperparameter codes;

[0063] Repeatedly adopting the boundary information fusion strategy, the multi-encoding co-evolution strategy and the reverse global information search strategy to train multiple first hyperparameter encodings to obtain the first optimal hyperparameter encoding;

[0064] The hyperparameters in the obtained first optimal hyperparameter encoding are used as the final hyperparameters of the entity recognition model, and the entity recognition model is deployed.

[0065] Optionally, repeatedly adopting a boundary information fusion strategy, a multi-encoding co-evolution strategy, and a reverse global information search strategy to train multiple first hyperparameter encodings to obtain a first optimal hyperparameter encoding includes:

[0066] A1. Determine the first optimal hyperparameter encoding in the current training process among all first hyperparameter encodings;

[0067] A2. Based on the first optimal hyperparameter code, the first hyperparameter code is updated using a boundary information fusion strategy to obtain the updated first hyperparameter code:

[0068]

[0069] in, Indicates the d-th dimension hyperparameter of the n-th first hyperparameter encoding in the t-th training process, n=1,2,....,N, N represents the total number of first hyperparameter individuals, d=1,2,....,D, D represents the total dimension of the parameters in the first hyperparameter individual, represents the d-th dimension hyperparameter of the n-th neighborhood hyperparameter encoding within the neighborhood of the n-th first hyperparameter encoding during the t-th training process, Ir represents the total number of neighborhood hyperparameter encodings, represents the d-th dimension hyperparameter of the first hyperparameter encoding after the n-th update, X max,d represents the upper limit of the d-th dimension hyperparameter, X min,d represents the lower limit of the d-th dimension hyperparameter, ξ mn represents the neighborhood search control factor, and λ represents the information fusion coefficient, e represents the natural constant, η represents the information fusion range factor, dist mn represents the Euclidean distance between the nth first hyperparameter encoding and its corresponding mth neighborhood hyperparameter encoding during the tth training process, represents the d-th dimension hyperparameter of the first optimal hyperparameter encoding, α represents the adaptive update coefficient, α max Represents the maximum value of the adaptive update coefficient, α min represents the minimum value of the adaptive update coefficient, T represents the maximum number of training times, γ represents the neighborhood radius, and dist(X max -X min ) indicates the upper limit code X max With lower limit code X min The Euclidean distance between .

[0070] The boundary information fusion strategy provided by the present invention can fuse boundary information and neighborhood hyperparameter encoding in the neighborhood range for search. In the early stage of the algorithm, the population is relatively dispersed, which can greatly improve the global search capability; in the later stage of the algorithm, the population gathers together, which can also expand the search range to a certain extent.

[0071] A3. Use a multi-coding co-evolution strategy to update the first hyperparameter code after the first update, and obtain the first hyperparameter code after the second update as:

[0072]

[0073] in, represents the kth first hyperparameter encoding during the tth training process, represents the first hyperparameter encoding after the kth secondary update, π represents the pi, r1 represents the first random number between (0,1), represents the first random hyperparameter encoding, Represents the second random hyperparameter encoding, r2 represents the second random number between (0,1), and r3 represents the third random number between (0,1);

[0074] The multi-coding co-evolution strategy provided by the present invention first performs a collision avoidance operation on the first hyperparameter coding after an update, and then learns the information in two different random hyperparameter codings in two different ways. This can provide more global search possibilities in the early stage of the algorithm, and gradually converts to a fine search as the algorithm proceeds, thereby ensuring the convergence ability of the algorithm.

[0075] A4. Use the reverse global information search strategy to update the first hyperparameter code after the second update, and obtain the first hyperparameter code after the third update as:

[0076]

[0077] in, represents the first hyperparameter encoding after the mth secondary update during the tth training process, represents the first hyperparameter encoding after the mth triple update, Represents the first hyperparameter encoding The corresponding reverse encoding, represents the third random hyperparameter encoding, r4 represents the fourth random number between (0,1), Cauchy (0,σ 2 ) represents the random number generated by Cauchy mutation, σ represents the adaptive Cauchy mutation factor, and r5 represents the fifth random number between (0,1).

[0078] The reverse global information search strategy provided by this invention provides a powerful global search capability, giving the algorithm ample opportunities to escape local optima. While the global search capability gradually decreases as the algorithm progresses, it remains consistently effective, significantly improving the algorithm's training effectiveness. An annealing simulation algorithm can be introduced to control the reverse global information search strategy to ensure algorithm convergence.

[0079] A5. Determine whether the current number of training times is greater than or equal to the preset maximum number of training times. If so, re-obtain the first optimal hyperparameter encoding based on the first hyperparameter encoding after three updates and output it. Otherwise, return to step A1.

[0080] Optionally, after each strategy is executed, an out-of-limit processing may be performed, for example, the out-of-limit parameter is randomly generated between its corresponding upper limit and lower limit.

[0081] The mutual coordination of the above strategies can effectively improve the deep learning effect, ultimately improve the accuracy of entity recognition, and effectively extract entities from LNG receiving station data.

[0082] In an embodiment of the present invention, the relationship recognition model is set to a Word2Vec-GRU-Attention-PCNN (a variant of the word vector representation model-gated recurrent unit-attention mechanism-convolutional neural network) composite model.

[0083] First, the input text sequence is converted into word vectors and position vectors through the Word2Vec model; then the context information is learned through the GRU module, and the weights of the word vectors are dynamically adjusted using the Attention mechanism; finally, the convolution operation is used in the PCNN layer to capture the local semantic information in the sequence, and the convolution features are divided into three regions: left, middle, and right through Piecewise Pooling, and the maximum pooling operation is performed on each region. Finally, the relationship classification result is output through the Softmax function.

[0084] In an embodiment of the present invention, a method for pre-deploying a relationship recognition model through deep learning includes:

[0085] Initializing the hyperparameters of the relationship recognition model to obtain a plurality of second hyperparameter codes;

[0086] Repeatedly adopting the boundary information fusion strategy, the multi-encoding co-evolution strategy and the reverse global information search strategy to train multiple second hyperparameter encodings to obtain the second optimal hyperparameter encoding;

[0087] The hyperparameters in the obtained second optimal hyperparameter encoding are used as the final hyperparameters of the relationship recognition model, and the relationship recognition model is deployed.

[0088] The process of obtaining the second optimal hyperparameter encoding is similar to the process of obtaining the first optimal hyperparameter encoding, and will not be repeated here.

[0089] In an embodiment of the present invention, a knowledge graph is constructed based on the entities and the relationships in the preprocessed LNG receiving station data, including:

[0090] Constructing knowledge based on the entities and the relationships in the preprocessed LNG receiving station data; wherein the knowledge represents entities, relationships between entities, and triples formed by entities;

[0091] Build a knowledge graph based on the acquired knowledge.

[0092] In an embodiment of the present invention, a graph database is used to store the knowledge graph, including: using Neo4j to store the knowledge graph.

[0093] A graph database is a database system designed for storing and querying graph data models that organizes and represents data in the form of graphs. Neo4j is the most widely used graph database, providing a rich set of graph database functions and tools. In Neo4j, nodes represent entities, and edges represent relationships between entities. In a graph database, nodes are the basic units for organizing and managing data, and are used to represent entities or objects in the graph; edges are used to connect nodes and have types and attributes to enrich the description of relationships; attributes are used to store detailed information about entities and relationships, including different types of data such as text, data, and dates. Neo4j uses the Cypher query language. Cypher queries are based on graph pattern matching, which is relatively intuitive and readable, and can better meet users' query needs for graph data.

[0094] In an embodiment of the present invention, the following further comprises:

[0095] Using deep learning to build an entity vector extraction model and an importance assessment model, and deploying the entity vector extraction model and the importance assessment model;

[0096] Entity relationship prediction is performed based on the deployed entity vector extraction model and importance assessment model to complete the knowledge graph.

[0097] Optionally, the entity vector extraction model can be set to a graph attention network, and the importance evaluation model can be set to a model consisting of a fully connected neural network layer, multiple aggregation layers, an averaging layer, and a nonlinear activation function layer in sequence. After the initial score is calculated by the fully connected neural network layer, the score aggregation is independently performed through multiple aggregation layers, and then the score is averaged through the averaging layer and output through the nonlinear activation function layer.

[0098] In an embodiment of the present invention, entity relationship prediction is performed based on the deployed entity vector extraction model and importance assessment model to complete the knowledge graph, including:

[0099] The entity vector extraction model deployed is used to extract the feature vectors of the same knowledge entity and the feature vectors of the neighborhood entity corresponding to the entity in the knowledge graph;

[0100] Among them, the same knowledge entity feature vector represents the feature vector corresponding to other entities in the same knowledge as the entity, the neighborhood entity feature vector represents the feature vector of the entity in the knowledge that has at least one entity in common with the entity, knowledge represents the triple in the knowledge graph, and the same knowledge entity feature vector is different from the neighborhood entity feature vector;

[0101] According to the feature vector of the same knowledge entity corresponding to the entity in the knowledge graph, the first importance corresponding to the same knowledge entity is evaluated by the deployed importance evaluation model;

[0102] According to the feature vector of the neighborhood entity corresponding to the entity in the knowledge graph, the second importance corresponding to the neighborhood entity is evaluated by the deployed importance evaluation model;

[0103] According to the first importance corresponding to the same knowledge entity, determine the first target knowledge corresponding to the entity in the knowledge graph (such as determining multiple first target knowledge with the greatest first importance), and obtain other knowledge corresponding to the same knowledge entity corresponding to the entity in the knowledge graph in the first target knowledge to obtain the second target knowledge;

[0104] Determining, in the second target knowledge, a final target knowledge (i.e., the second target knowledge with the greatest second importance) and a final target entity (i.e., the final target entity in the final target knowledge) according to the second importance corresponding to the neighborhood entity;

[0105] The knowledge graph is completed by reconstructing knowledge using entities in the knowledge graph, relationships in the final target knowledge, and the final target entities.

[0106] The method for completing the knowledge graph provided in an embodiment of the present invention can effectively increase information utilization efficiency and improve the knowledge graph information extraction effect of LNG receiving station data.

[0107] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0108] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0109] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0110] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0111] Those skilled in the art will understand that all or part of the steps in implementing the above facts and methods can be completed by instructing relevant hardware through a program, and the program involved or the program can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: the corresponding method steps are then brought out, and the storage medium can be ROM / RAM, a disk, an optical disk, etc.

[0112] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A knowledge graph information extraction method based on LNG receiving station data, characterized in that: include: Acquiring LNG receiving station data, and preprocessing the LNG receiving station data to obtain the preprocessed LNG receiving station data; Use the entity recognition model pre-deployed by deep learning to perform entity recognition on the pre-processed LNG receiving station data to obtain the entities in the pre-processed LNG receiving station data; Use the relationship recognition model pre-deployed by deep learning to identify relationships in the pre-processed LNG receiving station data and obtain the relationships corresponding to the pre-processed LNG receiving station data; Based on the entities and relationships of the pre-processed LNG receiving station data, a knowledge graph is constructed, and a graph database is used to store the knowledge graph to complete the extraction of knowledge graph information.

2. The knowledge graph information extraction method based on LNG receiving station data according to claim 1 is characterized in that: Acquiring LNG receiving station data and preprocessing the LNG receiving station data to obtain the preprocessed LNG receiving station data, including: Acquire LNG receiving station data, and perform deduplication on the LNG receiving station data to obtain deduplicated LNG receiving station data; The deduplicated LNG receiving station data is sequentially segmented and stop words are removed to obtain pre-processed LNG receiving station data.

3. The knowledge graph information extraction method based on LNG receiving station data according to claim 1 is characterized in that: The entity recognition model is set to a BERT-BiLSTM-CRF composite model.

4. The knowledge graph information extraction method based on LNG receiving station data according to claim 3 is characterized in that: Methods for pre-deploying entity recognition models through deep learning include: Initializing hyperparameters of the entity recognition model to obtain a plurality of first hyperparameter codes; Repeatedly adopting the boundary information fusion strategy, the multi-encoding co-evolution strategy and the reverse global information search strategy to train multiple first hyperparameter encodings to obtain the first optimal hyperparameter encoding; The hyperparameters in the obtained first optimal hyperparameter encoding are used as the final hyperparameters of the entity recognition model, and the entity recognition model is deployed.

5. The knowledge graph information extraction method based on LNG receiving station data according to claim 4 is characterized in that: The relationship recognition model is set to a Word2Vec-GRU-Attention-PCNN composite model.

6. The knowledge graph information extraction method based on LNG receiving station data according to claim 5 is characterized in that: Methods for pre-deploying relationship recognition models through deep learning include: Initializing the hyperparameters of the relationship recognition model to obtain a plurality of second hyperparameter codes; Repeatedly adopting the boundary information fusion strategy, the multi-encoding co-evolution strategy and the reverse global information search strategy to train multiple second hyperparameter encodings to obtain the second optimal hyperparameter encoding; The hyperparameters in the obtained second optimal hyperparameter encoding are used as the final hyperparameters of the relationship recognition model, and the relationship recognition model is deployed.

7. The knowledge graph information extraction method based on LNG receiving station data according to claim 1 is characterized in that: Constructing a knowledge graph based on the entities and relationships of the pre-processed LNG receiving station data includes: Constructing knowledge based on the entities and the relationships in the preprocessed LNG receiving station data; wherein the knowledge represents entities, relationships between entities, and triples formed by entities; Build a knowledge graph based on the acquired knowledge.

8. The knowledge graph information extraction method based on LNG receiving station data according to claim 1 is characterized in that: Using a graph database to store the knowledge graph includes: using Neo4j to store the knowledge graph.

9. The knowledge graph information extraction method based on LNG receiving station data according to claim 1 is characterized in that: Also includes: Using deep learning to build an entity vector extraction model and an importance assessment model, and deploying the entity vector extraction model and the importance assessment model; Entity relationship prediction is performed based on the deployed entity vector extraction model and importance assessment model to complete the knowledge graph.

10. The knowledge graph information extraction method based on LNG receiving station data according to claim 9 is characterized in that: Entity relationship prediction is performed based on the deployed entity vector extraction model and importance assessment model to complete the knowledge graph, including: The entity vector extraction model deployed is used to extract the feature vectors of the same knowledge entity and the feature vectors of the neighborhood entity corresponding to the entity in the knowledge graph; Among them, the same knowledge entity feature vector represents the feature vector corresponding to other entities in the same knowledge as the entity, the neighborhood entity feature vector represents the feature vector of the entity in the knowledge that has at least one entity in common with the entity, knowledge represents the triple in the knowledge graph, and the same knowledge entity feature vector is different from the neighborhood entity feature vector; According to the feature vector of the same knowledge entity corresponding to the entity in the knowledge graph, the first importance corresponding to the same knowledge entity is evaluated by the deployed importance evaluation model; According to the feature vector of the neighborhood entity corresponding to the entity in the knowledge graph, the second importance corresponding to the neighborhood entity is evaluated by the deployed importance evaluation model; Determine, based on the first importance corresponding to the same knowledge entity, first target knowledge corresponding to the entity in the knowledge graph, and obtain other knowledge corresponding to the same knowledge entity corresponding to the entity in the knowledge graph in the first target knowledge to obtain second target knowledge; Determining final target knowledge and a final target entity in the second target knowledge according to the second importance corresponding to the neighborhood entity; The knowledge graph is completed by reconstructing knowledge using entities in the knowledge graph, relationships in the final target knowledge, and the final target entities.

Images