Decision Support Method for Spatiotemporal Dynamic Data of Internet of Vehicles Enhanced by Knowledge Retrieval

Abstract

The present invention discloses a decision support method for spatio-temporal dynamic data of the Internet of Vehicles based on knowledge retrieval enhancement, including: collecting in-vehicle and out-vehicle perception data, and separating general data and privacy data; preprocessing the data, and compressing and encrypting the privacy data; adopting a data-knowledge dual-drive mode and a large language model for data processing to convert it into structured data; constructing a hyper-relationship knowledge graph and performing vectorization processing based on an embedding vector model; evaluating the importance of nodes based on the hyper-relationship knowledge graph for data update and increment; saving the privacy data locally, uploading the general data to a public database in the cloud, and defining the dynamic access scope of the public database; using a local entity recognition model to extract user question entities, matching the data existing locally and in the cloud, obtaining background knowledge content, and generating enhanced retrieval results. The present invention solves the problems of obtaining decision support information and the security of data use in the Internet of Vehicles environment.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a decision support method for spatio-temporal dynamic data of an Internet of Vehicles enhanced by knowledge retrieval. Background Art

[0002] In the digital age, Internet of Vehicles technology enables information exchange between vehicles and between vehicles and infrastructure by connecting vehicles to the Internet. This not only improves driving safety and efficiency but also brings challenges in data security and privacy protection. In Internet of Vehicles technology, the issue of data security is particularly prominent. During driving, vehicles generate and exchange a large amount of real-time data, including vehicle status information, environmental perception data, and user private data. The privacy requirements of this data necessitate effective protection measures to ensure that while users enjoy intelligent decision support, their privacy and data security are fully safeguarded.

[0003] Under the graph-based retrieval enhancement system, it is also a challenge to achieve real-time update of vehicle information. Most existing retrieval enhancement systems do not support the incremental generation of knowledge based on time series, which means that vehicle data cannot be effectively processed and updated. To address this challenge and be able to continuously and dynamically update hyper-relationship knowledge to ensure the accuracy and effectiveness of vehicle information in the database, a decision support method for spatio-temporal dynamic data of an Internet of Vehicles enhanced by knowledge retrieval needs to be proposed. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to propose a decision support method for spatio-temporal dynamic data of an Internet of Vehicles enhanced by knowledge retrieval.

[0005] To achieve the above technical objectives, the technical solutions adopted by the present invention are as follows:

[0006] The present invention provides a decision support method for spatio-temporal dynamic data of an Internet of Vehicles enhanced by knowledge retrieval, including the following steps:

[0007] Step 1: Collect vehicle internal and external perception data, and separate general data and privacy data from the vehicle internal and external perception data;

[0008] Step 2: Preprocess the general data and privacy data, then compress and encrypt the privacy data, and perform data marking and classification processing on the general data;

[0009] Step 3: Process the processed general data and privacy data with a data and knowledge dual-drive mode and a large language model for data processing based on rule documents, and convert the vehicle internal and external perception data into structured data;

[0010] Step 4: Construct a hyper-relationship knowledge graph based on the structured data, and perform vectorization processing on the indexes of the information in the hyper-relationship knowledge graph based on the embedding vector model;

[0011] Step 5: Conduct importance assessment of nodes based on the hyper-relationship knowledge graph, and perform data update and incremental expansion on the hyper-relationship knowledge graph;

[0012] Step 6: Save the privacy data locally, upload the general data to the public database in the cloud, and define the dynamic access scope of the public database based on the hyper-relationship knowledge graph;

[0013] Step 7: Use the local entity recognition model to extract the user's question entity, match the data existing locally and in the cloud, obtain the background knowledge content, and generate enhanced retrieval results.

[0014] Further, the specific steps of Step 1 include:

[0015] Step 11: Collect the vehicle internal and external perception data, where the vehicle internal and external perception data includes the vehicle's environmental information, vehicle status information, traffic flow information, vehicle networking service platform information, vehicle location information, vehicle ownership location information, vehicle networking user community group information, parking lot layout information, user personal information, vehicle driving trajectory information, and in-vehicle audio interaction information;

[0016] Step 12: Convert the vehicle internal and external perception data into text data for pre-trained language model recognition, and save it in a structured format;

[0017] Step 13: Use the pre-trained language model based on deep learning for pre-training, and for the privacy data types in the vehicle networking, freeze some parameters of the pre-trained language model by the method of freezing the large model and repairing the small model, and only fine-tune the key layers to enable the pre-trained language model to have the function of separating and desensitizing data;

[0018] Step 14: Use the fine-tuned pre-trained language model to recognize the vehicle internal and external perception data, and separate the general data and privacy data. The general data includes the vehicle's environmental information, vehicle status information, traffic flow information, vehicle networking service platform information, vehicle location information, vehicle ownership location information, vehicle networking user community group information, and parking lot layout information; the privacy data includes user personal information, vehicle driving trajectory information, and in-vehicle audio interaction information.

[0019] Further, the specific steps of Step 2 include:

[0020] Step 21: Clean, handle missing values, and handle outliers for the general data and privacy data;

[0021] Step 22: Build a semantic knowledge base based on the privacy data, analyze the semantic meaning of the privacy data in the semantic knowledge base, and construct a Huffman tree according to the semantic features of different users and scenarios and the frequency of character occurrences, assign a code to each character for compression, and save the compressed privacy data and the coding table;

[0022] Step 23: Add noise to the privacy data through differential privacy technology to mask the true value of the original data;

[0023] Step 24: Use the AES encryption algorithm to encrypt the privacy data and retain the encryption identifier of the original text category at the encryption location;

[0024] Step 25: Perform data tagging and classification on the general data to match the data interaction between the local and the cloud in different scenarios.

[0025] Further, the specific steps of step 3 are as follows:

[0026] Step 31: In the data-driven part, use the vehicle internal and external perception data to train and optimize the large language model to enable the large language model to automatically identify and extract entities and relationships in the vehicle internal and external perception data; in the knowledge-driven part, introduce the knowledge of experts, set up rule documents, including creating rule documents, restricting the output structure of the large language model, providing extraction cases of entities and relationships, evaluation cases of relationship importance ratings, and evaluation cases of category attribute discrimination, to guide the data processing and relationship evaluation of the large language model, and then combine the data-driven and knowledge-driven to form a data-knowledge dual-driven mode;

[0027] Step 32: Select a large language model, deploy the large language model on the cloud, and the local vehicle uses the API to call the large language model from the cloud; or deploy the large language model inside the vehicle networking block and use the private intranet to call the large language model through port forwarding;

[0028] Step 33: Adopt the data-knowledge dual-driven mode and the large language model to evaluate the vehicle internal and external perception data, learn patterns and rules from the actual vehicle internal and external perception data, provide rule documents to restrict the output content and format of model interaction, formulate a strategy for document chunking, and use prior knowledge to guide the output of the large language model;

[0029] Step 34: Make a decision on the chunk size according to the amount of information in the vehicle internal and external perception data. For the vehicle internal and external perception data with an amount of information greater than the preset value, use small chunks, and for the vehicle internal and external perception data with an amount of information not greater than the preset value, use large chunks;

[0030] Step 35: After interacting with the large language model based on the rule document through chunked data, obtain entity information, extract the entity information required to construct the hyper-relationship knowledge graph, and convert it into structured data.

[0031] Further, it is characterized in that the specific steps of step 35 include:

[0032] Step 351: When interacting with the large language model for the first time, record the data chunk position of the information source.

[0033] Step 352: Extract knowledge according to the knowledge density and chunk size of the data to complete the first interaction with the large language model.

[0034] Step 353: After the first interaction with the large language model, transmit the extracted data to the large language model for secondary judgment, evaluate the relationship between nodes and edges in the hyper-relationship knowledge graph, and output the rating result according to the relationship strength rating.

[0035] Step 354: For the data with changed edge description relationship rating in the relationship strength rating, conduct a third judgment, extract the data chunk position of the information source recorded during the first interaction, interact with the large language model again to obtain entity information, and convert it into structured data. The structured data includes entities, relationships, and their attribute information.

[0036] Further, the specific steps of step 4 include:

[0037] Step 41: Construct a hyper-relationship knowledge graph based on the structured data, represent the entities and relationships in the structured data as nodes and edges in the hyper-relationship knowledge graph respectively. Each node contains entity information, and the edge represents the relationship between entities. The hyper-relationship consists of a main triple and several auxiliary attribute value descriptions. The main triple is represented as (subject, relationship, object), and the auxiliary attribute value description is represented in the form of key-value pairs. The hyper-relationship knowledge graph is represented as:

[0038]

[0039] Among them, is the node set, and each node contains the attribute , represents the node key-value category, and saves multiple pairs of key-values in the form of a list. Among them, includes the key and the value ;

[0040] is the edge set, defined as , that is, the edge is an ordered pair of two nodes in the node set ; Edge Connection node And node Edge Contains attributes And attributes , Describes the relationship of the edge, Describes the relationship strength rating;

[0041] Step 42: Add descriptive information to each node and edge, and use key-value pairs to save the supplementary information of the node;

[0042] Step 43: For nodes with multiple relationships, support multi-edge descriptions, mark and distinguish the information of the same nodes and edges, evaluate the importance of different information, and mark the importance of the information;

[0043] Step 44: Capture context information through the hyper-relationship knowledge graph, enhance the context understanding ability of the hyper-relationship knowledge graph, identify and integrate the multi-dimensional relationships between entities, as well as the attribute information of entities and relationships, and finally obtain a hyper-relationship knowledge graph with rich semantics;

[0044] Step 45: Use the embedding vector model to perform embedded vector encoding on the keys, nodes and relationships in the hyper-relationship knowledge graph, and write the vector information into the lists of each key, node and relationship to complete the vectorization of the index.

[0045] Furthermore, the specific steps of step 5 include:

[0046] Step 51: Analyze the hyper-relationship knowledge graph using EasyGraph technology;

[0047] Step 52: Adopt a method for evaluating the importance of time-series network nodes based on inter-layer neighborhood information entropy to determine the importance of each node and edge, and write it into For importance evaluation during the final information matching selection in combination with the user's weight preference;

[0048] Step 53: Perform incremental expansion on the constructed hyper-relationship knowledge graph, and perform another round of data processing based on the rule document to form the original hyper-relationship knowledge graph and the new hyper-relationship knowledge graph. The original hyper-relationship knowledge graph is used as the ontology graph, and the new hyper-relationship knowledge graph is used as the incremental graph;

[0049] Step 54: Define the scope of entity alignment according to the method for evaluating the importance of time-series network nodes, set the block size according to the scale of the ontology graph, and divide the ontology graph into blocks according to the scope of entity alignment and the block size to form multiple node blocks;

[0050] Step 55: With Taking the important nodes identified in the ontology graph as the center, match the node blocks of the ontology graph with the nodes of the incremental graph to obtain the node block with the highest similarity;

[0051] Step 56. In the structured data, use a large language model to , and for intelligent fusion of node information, and for knowledge fusion of the same entity. Expand the unmatched part of the incremental graph and the ontology graph as new structured data onto the ontology graph to complete the incremental generation of the ontology graph for the extended content.

[0052] Further, the method for evaluating the importance of the temporal network nodes in step 52 specifically includes:

[0053] Step 521. Calculate the inter-layer similarity of the edges where the node and its neighbors continuously appear according to the neighbor topological overlap coefficient :

[0054]

[0055] Among them, represents the th node, represents the th node, represents the th time slice, represents the adjacency matrix at the th time slice and the edge between node and node in it;

[0056] Step 522. Construct a temporal super-adjacency matrix according to the influence between different time slices:

[0057]

[0058] Among them, represents the adjacency matrix of the th time slice, and is a positive integer, s represents the s-th time slice, s is a positive integer and s < t, represents the influence matrix between the s-th time slice and the th time slice, is the temporal super-adjacency matrix, which is used to describe the relationship or influence of each node between the s-th time slice and the t-th time slice;

[0059] Step 523: Based on the constructed temporal super adjacency matrix Calculate the eigenvector corresponding to the maximum eigenvalue, that is, the eigenvector centrality of each node in each time slice, which is used to judge the importance of the node in the graph network.

[0060] Further, the specific steps of step 6 are as follows:

[0061] Step 61: Save the privacy data as a local super-relation graph file locally, and upload the general data to the public database in the cloud to support multi-vehicle access.

[0062] Step 62: Assign a permission scope based on the rule document to each semantic record, and open different ranges of semantic library sets for different permissions: for neighbor nodes in a similar geographical area, open the traffic geographical information on the local map in the publicly shared semantics; for neighbor nodes with the same unit group relationship, open privacy data such as geographical tags, exclusive parking spaces, and historical trajectories of neighbor vehicles in the shared unit park to support multi-vehicle access.

[0063] Step 63: Through the constructed super-relation knowledge graph, identify the clustering relationship between vehicles based on various key-value pairs of node key-value categories, and divide vehicles with similar characteristics or behaviors into the same group, enabling vehicles to access each other without authorization within the privacy and security range, and the data for mutual access are all valid data strongly correlated with their respective time and space.

[0064] Step 64: The administrator loads and updates the cloud data in combination with the infrastructure within the user group to achieve the sharing and retrieval of general data.

[0065] Further, the specific steps of step 7 are as follows:

[0066] Step 71: Extract the entities and expressions of the problem based on the local entity recognition model.

[0067] Step 72: Vectorize the entities and expressions of the problem through the embedding vector model to convert the text information into vector information.

[0068] Step 73: Identify the clustering relationship between vehicles through the clustering information in the super-relation knowledge graph and dynamically determine the access scope.

[0069] Step 74: Perform similarity matching between the vector information and the index vectors in the local super-relation graph file and the public database within the permissions in the cloud to obtain the background knowledge content.

[0070] Step 75: Combine the obtained background knowledge content with the user's question content to form a prompt content and interact with the large language model.

[0071] Step 76: Return the response after interacting with the large language model to the local, and perform reverse decryption based on the encryption process in Step 2 to obtain the final enhanced retrieval result.

[0072] Adopting the above technical solution, compared with the prior art, the beneficial effects of the present invention are as follows:

[0073] The present invention uses a pre-trained language model based on deep learning for pre-training to capture the complexity, diversity, and spatio-temporal correlation of data in the vehicle networking environment. For privacy data types, a labeled data set is used to fine-tune the model to improve the recognition and extraction performance of privacy data. Protect the initially processed privacy data and general data, including missing value processing and noise data processing, to ensure the security and accuracy of the data, build a semantic knowledge base, and achieve data compression and privacy protection. Combine a large language model in a dual-drive mode of data processing rules and data knowledge to convert vehicle data into structured data, and build a hyper-relationship knowledge graph, which contains attribute information of nodes and edges. Build an embedding vector model based on BERT to encode vehicle knowledge and general knowledge, use the hyper-relationship knowledge graph to mine the spatio-temporal correlation between neighbor nodes in the vehicle networking, introduce a temporal network node importance evaluation method to determine the relationship strength between nodes, and complete knowledge increment generation. Upload the general data to the public database in the cloud to achieve common access and data sharing by multiple vehicles. Through this series of technical steps, an intelligent system that can protect the privacy of vehicles and users while efficiently processing and utilizing vehicle networking data is constructed.

[0074] In summary, the decision support method for spatio-temporal dynamic data of vehicle networking based on knowledge retrieval enhancement proposed by the present invention combines data knowledge dual-drive, incremental retrieval enhancement generation with knowledge graph, semantic knowledge base, hyper-relationship, node importance, and dynamic attention mechanism of data with the requirements of vehicle networking. It not only effectively solves the problems of real-time and efficient data processing and security in vehicle networking, but also provides new ideas and technical means for users' intelligent decision-making, and has important significance for promoting the intelligence and security improvement of vehicle networking. Description of the Drawings

[0075] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0076] Figure 1 It is an execution flowchart of a decision support method for spatio-temporal dynamic data of vehicle networking based on knowledge retrieval enhancement provided by an embodiment of the present invention.

[0077] Figure 2 It is a schematic diagram of the knowledge graph provided by an embodiment of the present invention.

[0078] Figure 3 It is the structural framework provided by an embodiment of the present invention.

[0079] Figure 4 It is a schematic diagram of enhanced retrieval provided by an embodiment of the present invention. Specific embodiments

[0080] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be specifically noted that the following embodiments are only used to illustrate the present invention, but do not limit the scope of the present invention. Similarly, the following embodiments are only partial embodiments of the present invention rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0081] Please refer to Figures 1-4 , a decision support method for spatio-temporal dynamic data of an Internet of Vehicles based on enhanced knowledge retrieval of the present invention, includes the following steps:

[0082] Step 1: Collect vehicle internal and external perception data, and separate general data and privacy data from the vehicle internal and external perception data;

[0083] In this embodiment, step 1 specifically includes:

[0084] Step 11: Collect vehicle internal and external perception data, where the vehicle internal and external perception data includes vehicle environmental information, vehicle status information, traffic flow information, Internet of Vehicles service platform information, vehicle location information, vehicle attribution information, Internet of Vehicles user community group information, parking lot layout information, user personal information, vehicle driving trajectory information, and in-vehicle audio interaction information;

[0085] Step 12: Convert the vehicle internal and external perception data into text data for pre-trained language model recognition and save it in a structured format;

[0086] Step 13: Use a pre-trained language model based on deep learning for pre-training to learn the complexity, diversity, and spatio-temporal correlation of data in the spatio-temporal dynamic complex environment of the Internet of Vehicles. For the privacy data types in the Internet of Vehicles, freeze some parameters of the pre-trained language model by the method of freezing the large model and repairing the small model, and only fine-tune the key layers, while reducing the computational amount and maintaining the model performance, so as to enable the pre-trained language model to have the function of separating and desensitizing data; use the labeled data set to fine-tune the model to improve the model's performance in identifying and extracting privacy data;

[0087] Step 14: Use the fine-tuned pre-trained language model to identify the vehicle's internal and external perception data, and separate the general data and privacy data. The general data includes the vehicle's environmental information, vehicle status information, traffic flow information, vehicle networking service platform information, vehicle location information, vehicle ownership information, vehicle networking user community group information, and parking lot layout information; the privacy data includes user personal information, vehicle driving trajectory information, and in-vehicle audio interaction information.

[0088] Step 2: Preprocess the general data and privacy data, then perform compression and encryption processing on the privacy data, and perform data tagging and classification processing on the general data;

[0089] In this embodiment, the specific steps of Step 2 include:

[0090] Step 21: Clean, handle missing values, and handle outliers for the general data and privacy data to improve data quality and ensure data security and accuracy;

[0091] Data cleaning: Remove stop words, perform spelling correction, correct spelling mistakes, ensure text accuracy, remove special characters and numbers, clean non-text elements and unnecessary numbers in the text, and perform stemming and lemmatization.

[0092] The method for handling missing values is mode interpolation. Mode interpolation: For categorical labels or phrases, if there are missing values, mode interpolation can be used. For example, if the labels in a dataset are "A", "B", and "C", and a certain label is missing, the most frequently occurring label is used to fill in.

[0093] The outlier detection method uses method. , where is the first quartile, is the third quartile. If a data point is below or above , it is considered an outlier.

[0094] Step 22: Build a semantic knowledge base based on the privacy data, analyze the semantic meaning of the privacy data in the semantic knowledge base, and construct a Huffman tree according to the semantic features of different users and scenarios and the frequency of character occurrences, assign a code to each character for compression, and save the compressed privacy data and the code table for use during decompression;

[0095] Step 23: Add noise to the private data through differential privacy technology to mask the true value of the original data; in this way, even if the data is leaked, the original data cannot be directly identified, protecting the original nature of the data from being directly revealed, ensuring the security of data transmission and storage, and implementing a personalized privacy protection strategy.

[0096] Step 24: Use the AES encryption algorithm to encrypt the private data and retain the encrypted identifier of the original text category during encryption; ensure the security and privacy of the data during sharing and use;

[0097] Encryption algorithm (AES): Among them, is the ciphertext, is the encryption function using the key and is the plaintext. Encrypt and clean the data for subsequent local and external encrypted indexes to prevent the leakage of private data.

[0098] Step 25: Perform data tagging and classification on the general data to match the data interaction between the local and the cloud in different scenarios.

[0099] Step 3: Process the processed general data and private data with the large language model in a data-knowledge dual-drive mode based on the rule document, and convert the vehicle internal and external perception data into structured data;

[0100] In this embodiment, the specific steps of step 3 include:

[0101] In the data-driven part, use the vehicle internal and external perception data to train and optimize the large language model to enable the large language model to automatically identify and extract entities and relationships in the vehicle internal and external perception data; in the knowledge-driven part, introduce the knowledge of experts and set up rule documents, including creating rule documents, restricting the output structure of the large language model, providing extraction cases of entities and relationships, evaluation cases of relationship importance ratings, and evaluation cases of category attribute discrimination, to guide the data processing and relationship evaluation of the large language model, and then combine the data-driven and knowledge-driven to form a data-knowledge dual-drive mode;

[0102] Step 32: Select a large language model, deploy the large language model on the cloud, and the local vehicle uses the API to call the large language model from the cloud; or deploy the large language model inside the vehicle networking block and use the private intranet to call the large language model through port forwarding;

[0103] Step 33: Use the data and knowledge dual-drive mode and large language model to evaluate the vehicle's internal and external perception data. Learn patterns and rules from the actual vehicle's internal and external perception data, provide a rule document to limit the output content and format of the model interaction, formulate a strategy for document chunking, and use prior knowledge to guide the output of the large language model;

[0104] Step 34: Make a decision on the chunk size according to the information volume of the vehicle's internal and external perception data. For the vehicle's internal and external perception data with an information volume greater than the preset value, use small chunks; for the vehicle's internal and external perception data with an information volume not greater than the preset value, use large chunks to improve the information extraction efficiency and credibility of the large language model;

[0105] Step 35: After the large language model interaction based on the rule document through the chunked data, obtain entity information, extract the entity information required for constructing the hyper-relationship knowledge graph, and convert it into structured data.

[0106] In this embodiment, the specific steps of Step 35 include:

[0107] Step 351: When using the large language model for the first interaction, record the data chunk position of the information source;

[0108] Step 352: Extract knowledge according to the knowledge density and chunk size of the data to complete the first large language model interaction;

[0109] Step 353: After the first large language model interaction, transmit the extracted data to the large language model for a second judgment, evaluate the relationship between the nodes and edges in the hyper-relationship knowledge graph, and output the rating result according to the relationship strength rating;

[0110] Step 354: For the data with a changed edge description relationship rating in the relationship strength rating, conduct a third judgment, extract the data chunk position of the information source recorded during the first interaction, and interact with the large language model again to obtain entity information and convert it into structured data. The structured data includes entities, relationships, and their attribute information.

[0111] Step 4: Construct a hyper-relationship knowledge graph based on the structured data, use the hyper-relationship knowledge graph to realize the hyper-relationship mining of the spatio-temporal correlation of data between neighbor nodes in the vehicle network, and perform vectorization processing on the indexing of information in the hyper-relationship knowledge graph based on the embedding vector model;

[0112] In this embodiment, the specific steps of Step 4 include:

[0113] Step 41. Construct a hyper-relation knowledge graph based on structured data, representing entities and relationships in the structured data as nodes and edges in the hyper-relation knowledge graph respectively. Each node contains entity information, and the edge represents the relationship between entities. The hyper-relation knowledge graph is as shown in Figure 2 shown, Figure 2 where the circles in

[0114]

[0115] represent nodes and the lines represent edges. The hyper-relation knowledge graph is represented as: where is a set of nodes, and each node contains attributes which represents the node key-value category and stores multiple key-value pairs in a list form. Among them, includes a key and a value ;

[0116] is a set of edges, defined as , that is, an edge is an ordered pair of two nodes in the node set . The edge connects the node and the node . The edge contains attributes and attribute . describes the relationship of the edge, and describes the relationship strength rating;

[0117] Among them, a hyper-relation consists of a main triple and several auxiliary attribute value descriptions. The main triple is represented as (subject, relation, object), which are respectively represented by in Step 4, while the auxiliary attribute value descriptions are represented in the form of key-value pairs. By introducing the concept of hyper-relations, the relationships between entities are extended to complex relationships that can involve multiple entities, which involves identifying and integrating the multi-dimensional relationships between entities and the attribute information of entities and relationships.

[0118] Among them, the hyper-relation format is as follows: If there is a piece of knowledge: "Zhang San participated in the meeting held at the Beijing International Convention Center on November 11, 2024, and the reason for participating in this meeting is the annual summary." The hyper-relation content is as shown in the following table:

[0119]

[0120] Step 42. Add descriptive information to each node and edge, and use key-value pairs to save the supplementary information of the nodes.

[0121] Step 43: For nodes with multiple relationships, support multi - side description, mark and distinguish information of the same nodes and edges, evaluate the importance of different information, and mark the importance level of the information.

[0122] Step 44: Capture context information such as source documents, time qualifiers, and confidence scores through the hyper - relational knowledge graph to provide a more detailed knowledge representation for the system. This information will be used to enhance the context understanding ability of the hyper - relational knowledge graph, identify and integrate multi - dimensional relationships between entities, as well as attribute information of entities and relationships, and finally obtain a hyper - relational knowledge graph with rich semantics; realize hyper - relational mining of the spatio - temporal correlation of data between neighbor nodes in the vehicle - to - everything network.

[0123] Step 45: Use an embedding vector model (BERT - based embedding vector model) to perform embedded vector encoding on keys, nodes, and relationships within the hyper - relational knowledge graph, and write the vector information into the lists of each key, node, and relationship to complete the vectorization of the index. Combining the structured data in the hyper - relational knowledge graph , use a BERT - based embedding vector model to construct a more comprehensive vector representation, which not only contains the semantic information of the text, but also incorporates relationship information between entities and attribute information such as key - value pairs, providing richer structured data for constructing the index and the node description list of the knowledge graph.

[0124] Step 5: Evaluate the importance of nodes based on the hyper - relational knowledge graph, and perform data update and incremental expansion on the hyper - relational knowledge graph.

[0125] In this embodiment, the specific steps of Step 5 include:

[0126] Step 51: Use EasyGraph technology to analyze the hyper - relational knowledge graph, and utilize its hybrid programming characteristics to accelerate graph processing and graph analysis, effectively manage and query information such as the real - time status, driving path, and environmental interaction of vehicles in the graph. Combining user decision - making requirements, improve the response speed and decision - making efficiency of decision support.

[0127] Step 52: Adopt a method for evaluating the importance of temporal network nodes based on inter - layer neighborhood information entropy to determine the importance level of each node and edge, and write it into , and combine the user's weight preference for importance evaluation during the final information matching selection; accurately measure the influence of nodes in the network.

[0128] Step 53: Perform incremental expansion on the already constructed hyper - relational knowledge graph, and adopt another round of data processing based on rule documents to form the original hyper - relational knowledge graph and the new hyper - relational knowledge graph. The original hyper - relational knowledge graph is used as the ontology graph, and the new hyper - relational knowledge graph is used as the incremental graph.

[0129] Step 54: Define the scope of entity alignment according to the importance evaluation method of time-series network nodes, set the block size according to the scale of the ontology graph, and divide the ontology graph into blocks according to the scope of entity alignment and the block size to form multiple node blocks;

[0130] Step 55: Centered on the important nodes identified in match the node blocks of the ontology graph with the nodes of the incremental graph to obtain the node block with the highest similarity;

[0131] Step 56: In structured data, use a large language model to , and for intelligent fusion of node information, and for knowledge fusion of the same entities, and use the parts of the incremental graph that do not match the ontology graph as new structured data to extend to the ontology graph, completing the incremental generation of the extended content on the ontology graph.

[0132] In this embodiment, the importance evaluation method of time-series network nodes in step 52 specifically includes:

[0133] Step 521: Calculate the inter-layer similarity of the continuous edges between a node and its neighbors according to the neighbor topological overlap coefficient :

[0134]

[0135] where represents the th node, represents the th node, represents the th time slice, represents the edge between node and node in the adjacency matrix and node at the

[0136] Step 522: Construct a time-series super-adjacency matrix according to the influence between different time slices:

[0137]

[0138] where represents the adjacency matrix of the th time slice, and is a positive integer, s represents the s-th time slice, s is a positive integer and s < t, represent the influence matrix between the s-th time slice and the t-th time slice, which is the temporal super-adjacency matrix of, used to describe the relationship or influence between nodes at the s-th time slice and the t-th time slice;

[0139] Step 523. Based on the constructed temporal super-adjacency matrix calculate the eigenvector corresponding to the maximum eigenvalue, that is, the eigenvector centrality of each node within each time slice, which is used to judge the importance of the node in the graph network.

[0140] Step 6. Save the private data locally, upload the general data to the public database in the cloud, and define the dynamic access scope of the public database based on the super-relationship knowledge graph;

[0141] In this embodiment, the specific steps of Step 6 include:

[0142] Step 61. Save the private data as a local super-relationship graph file locally, and upload the general data to the public database in the cloud to support joint access by multiple vehicles;

[0143] Step 62. Mark each semantic record with a permission scope based on a rule document, and open different ranges of semantic library sets for different permissions: for neighbor nodes in a similar geographical range, open the traffic geographical information on the local map in the publicly shared semantics; for neighbor nodes with the same unit group relationship, open the privacy data such as geographical labels, exclusive parking spaces, and historical trajectories of neighbor vehicles within the unit park to support joint access by multiple vehicles;

[0144] Step 63. Through the constructed super-relationship knowledge graph, identify the clustering relationship between vehicles according to various key-value pairs of node key-value categories, and divide vehicles with similar characteristics or behaviors into the same group, so that vehicles can access each other without authorization within the privacy and security scope, and the data for mutual access are all valid data strongly correlated with their respective time and space, so as to more accurately control the data sharing scope and more efficiently realize the rapid retrieval of useful data;

[0145] Step 64. The administrator loads and updates the cloud data in combination with the infrastructure within the user group to realize the sharing and retrieval of general data. In this way, it not only protects the privacy of vehicles, but also realizes the sharing and retrieval of general knowledge, and greatly improves the secondary knowledge construction rate of the system.

[0146] Step 7. Use the local entity recognition model to extract the user's question entity, match the data existing locally and in the cloud, obtain the background knowledge content, and generate an enhanced retrieval result.

[0147] In this embodiment, the specific steps of Step 7 include:

[0148] Step 71: Extract the entities and expressions of the problem based on the local entity recognition model;

[0149] Step 72: Vectorize the entities and expressions of the problem through the embedding vector model, and convert the text information into vector information;

[0150] Step 73: Identify the clustering relationship between vehicles through the clustering information in the hyper-relationship knowledge graph, and dynamically determine the access scope;

[0151] Step 74: Perform similarity matching between the vector information and the index vectors in the local hyper-relationship graph file and the public database within the permission in the cloud to obtain the background knowledge content; this process considers the importance of nodes and context information to retrieve the most relevant background knowledge, improving the accuracy of decision-making background support;

[0152] Step 75: Combine the obtained background knowledge content with the user's question content to form a prompt content and interact with the large language model;

[0153] Step 76: Return the reply after the interaction with the large language model to the local, and decrypt it reversely based on the encryption process in Step 2 to obtain the final enhanced retrieval result. Complete the closed loop of data usage, meeting the user's data usage security in the vehicle networking environment and the intelligent acquisition requirement of decision-making support information.

[0154] The above are only some embodiments of the present invention, and thus do not limit the protection scope of the present invention. Any equivalent device or equivalent process transformation made by using the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present invention.

Claims

1. A decision support method for spatio-temporal dynamic data in the Internet of Vehicles enhanced by knowledge retrieval, characterized in that It includes the following steps: Step 1: Collect the perception data inside and outside the vehicle, and separate the general data and privacy data from the perception data inside and outside the vehicle; Step 2: Preprocess the general data and privacy data, then compress and encrypt the privacy data, and perform data marking and classification processing on the general data; Step 3: Use the data knowledge dual-drive mode and the large language model to perform data processing based on the rule document on the processed general data and privacy data, and convert the perception data inside and outside the vehicle into structured data; Step 4: Construct a hyper-relationship knowledge graph based on the structured data, and perform vectorization processing on the index of the information in the hyper-relationship knowledge graph based on the embedding vector model; Specifically include: Step 41: Construct a hyper-relationship knowledge graph based on the structured data, represent the entities and relationships in the structured data as nodes and edges in the hyper-relationship knowledge graph respectively, each node contains entity information, and the edge represents the relationship between entities; The hyper-relationship consists of a main triple and several auxiliary attribute value descriptions. The main triple is represented as (subject, relationship, object), and the auxiliary attribute value description is represented in the form of key-value pairs; The hyper-relationship knowledge graph is represented as: Among them, is a set of nodes, and each node contains attributes , represents the node key-value category and stores multiple pairs of key-values in a list form. Among them, includes a key and a value ; is an edge set, defined as , that is, an edge is an ordered pair of two nodes in the node set ; the edge connects node and node , the edge contains attributes and attribute , describes the relationship of the edge, describes the relationship strength rating; Step 42: Add descriptive information to each node and edge, and use key-value pairs to save the supplementary information of the node; Step 43: For nodes with multiple relationships, support multi-edge descriptions, mark and distinguish the information of the same nodes and edges, evaluate the importance of different information, and mark the importance of the information; Step 44: Capture context information through the hyper-relationship knowledge graph, enhance the context understanding ability of the hyper-relationship knowledge graph, identify and integrate the multi-dimensional relationships between entities, as well as the attribute information of entities and relationships, and finally obtain a hyper-relationship knowledge graph with rich semantics; Step 45: Use the embedding vector model to perform embedded vector encoding on the keys, nodes and relationships in the hyper-relationship knowledge graph, and write the vector information into the lists of each key, node and relationship to complete the vectorization of the index; Step 5: Perform importance evaluation of nodes based on the hyper-relationship knowledge graph, and perform data update and incremental expansion on the hyper-relationship knowledge graph; Specifically include: Step 51: Use the EasyGraph technology to analyze the hyper-relationship knowledge graph; Step 52: Use the time-series network node importance evaluation method based on the inter-layer neighborhood information entropy to determine the importance levels of each node and edge, and write them into For importance evaluation during the final information matching selection in combination with the user's weight preference; Step 53: Perform incremental expansion on the constructed hyper-relationship knowledge graph, and perform another round of data processing based on the rule document to form the original hyper-relationship knowledge graph and the new hyper-relationship knowledge graph. The original hyper-relationship knowledge graph is used as the ontology graph, and the new hyper-relationship knowledge graph is used as the incremental graph; Step 54: Define the range of entity alignment according to the importance evaluation method of time-series network nodes, set the block size according to the scale of the ontology graph, and divide the ontology graph into blocks according to the range of entity alignment and the block size to form multiple node blocks; Step 55: Taking the important nodes identified therein as the center, matching the node blocks of the ontology graph with the nodes of the incremental graph, and obtaining the node block with the highest similarity; Step 56. In the structured data, use a large language model to , and for intelligent fusion of node information, and for knowledge fusion of the same entity, and use the part of the incremental graph that does not match the ontology graph as new structured data to extend to the ontology graph, completing the incremental generation of the ontology graph for the extended content; Step 6: Save the privacy data locally, upload the general data to the public database in the cloud, and define the dynamic access range of the public database based on the hyper-relationship knowledge graph; Step 7: Use the local entity recognition model to extract the user's question entity, match the data existing locally and in the cloud, obtain the background knowledge content, and generate an enhanced retrieval result.

2. The decision support method for spatio-temporal dynamic data of the vehicle networking enhanced by knowledge retrieval as claimed in claim 1, wherein Step 1 specifically includes the following: Step 11: Collect the perception data inside and outside the vehicle, where the perception data inside and outside the vehicle includes the vehicle's environmental information, vehicle's status information, traffic flow information, vehicle networking service platform information, vehicle's location information, vehicle ownership location information, vehicle networking user community group information, layout information of the parking lot, user personal information, vehicle driving trajectory information, and in-vehicle audio interaction information; Step 12: Convert the perception data inside and outside the vehicle into text data for pre-trained language model recognition and save it in a structured format; Step 13: Use a pre-trained language model based on deep learning for pre-training, and for the privacy data types in vehicle networking, freeze some parameters of the pre-trained language model by the method of freezing the large model to repair the small model, and only fine-tune the key layers to enable the pre-trained language model to have the function of separating and desensitizing data; Step 14: Use the fine-tuned pre-trained language model to recognize the perception data inside and outside the vehicle and separate the general data and privacy data. The general data includes the vehicle's environmental information, vehicle's status information, traffic flow information, vehicle networking service platform information, vehicle's location information, vehicle ownership location information, vehicle networking user community group information, and layout information of the parking lot; the privacy data includes user personal information, vehicle driving trajectory information, and in-vehicle audio interaction information.

3. The decision support method for spatio-temporal dynamic data of an Internet of Vehicles enhanced by knowledge retrieval as claimed in claim 1, wherein Step 2 specifically includes the following: Step 21: Clean, handle missing values, and handle outliers for the general data and privacy data; Step 22: Build a semantic knowledge base based on the privacy data, analyze the semantic meaning of the privacy data in the semantic knowledge base, and construct a Huffman tree according to the semantic features of different users and scenarios and the frequency of character occurrences, assign a code to each character for compression, and save the compressed privacy data and the code table; Step 23: Add noise to the privacy data through differential privacy technology to mask the true value of the original data; Step 24: Use the AES encryption algorithm to encrypt the privacy data and retain the encryption identifier of the original text category during encryption; Step 25: Perform data tagging and classification on the general data to match the data interaction between the local and the cloud in different scenarios.

4. The decision support method for spatio-temporal dynamic data of the vehicle networking enhanced by knowledge retrieval as claimed in claim 1, wherein Step 3 specifically includes the following: Step 31: In the data-driven part, use the perception data inside and outside the vehicle to train and optimize the large language model to enable the large language model to automatically recognize and extract entities and relationships in the perception data inside and outside the vehicle; in the knowledge-driven part, introduce the knowledge of experts and set rule documents, including creating rule documents, restricting the output structure of the large language model, providing extraction cases of entities and relationships, evaluation cases of relationship importance ratings, and evaluation cases of category attribute discrimination, to guide the data processing and relationship evaluation of the large language model, and then combine the data-driven and knowledge-driven to form a data-knowledge dual-driven mode; Step 32: Select a large language model, deploy the large language model in the cloud, and the local vehicle uses the API to call the large language model from the cloud; or deploy the large language model inside the vehicle networking block and use the private intranet to call the large language model through port forwarding; Step 33: Adopt a data and knowledge dual-driven mode and the large language model to evaluate the vehicle's internal and external perception data, learn patterns and rules from the actual vehicle's internal and external perception data, provide a rule document to limit the output content and format of the model interaction, formulate a strategy for document chunking, and use prior knowledge to guide the output of the large language model; Step 34: Make a decision on the chunk size according to the amount of information in the vehicle's internal and external perception data. For the vehicle's internal and external perception data with an information amount greater than the preset value, small chunks are used, and for the vehicle's internal and external perception data with an information amount not greater than the preset value, large chunks are used; Step 35: After the large language model interaction based on the rule document through the chunked data, obtain entity information, extract the entity information required to construct the hyper-relationship knowledge graph, and convert it into structured data.

5. The decision support method for spatio-temporal dynamic data of the vehicle networking enhanced by knowledge retrieval as claimed in claim 4, wherein, The specific content of Step 35 includes: Step 351: When using the large language model for the first interaction, record the data chunk position of the information source; Step 352: Extract knowledge according to the knowledge density and chunk size of the data to complete the first large language model interaction; Step 353: After the first large language model interaction, send the extracted data to the large language model for a second judgment, evaluate the relationship between the nodes and edges in the hyper-relationship knowledge graph, and output the rating result according to the relationship strength rating; Step 354: For the data with a changed edge description relationship rating in the relationship strength rating, conduct a third judgment, extract the data chunk position of the information source recorded during the first interaction, interact with the large language model again to obtain entity information, and convert it into structured data. The structured data includes entities, relationships, and their attribute information.

6. The decision support method for spatio-temporal dynamic data of the vehicle networking enhanced by knowledge retrieval as claimed in claim 1, wherein The specific method for evaluating the importance of the temporal network nodes in Step 52 includes: Step 521. Calculate the inter-layer similarity of the continuously occurring edges between a node and its neighbors according to the neighbor topological overlap coefficient : Among them, represents the th node, represents the th node, represents the th time slice, represents the edge between node and node in the adjacency matrix of the th time slice; and node Step 522: Construct a temporal super adjacency matrix based on the influence between different time slices : Among them, represents the adjacency matrix of the th time slice, and is a positive integer, s represents the s-th time slice, s is a positive integer and s < t, represents the influence matrix between the s-th time slice and the th time slice, is 's temporal hyper-adjacency matrix, used to describe the relationship or influence of each node between the s-th time slice and the t-th time slice; Step 523: Based on the constructed temporal super-adjacency matrix Calculate the eigenvector corresponding to the maximum eigenvalue, that is, the eigenvector centrality of each node in each time slice, which is used to judge the importance of the node in the graph network.

7. The decision support method for spatio-temporal dynamic data of the vehicle networking based on knowledge retrieval enhancement according to claim 1, wherein The specific content of Step 6 includes: Step 61: Save the privacy data as a local hyper-relationship graph file locally, and upload the general data to the public database in the cloud to support multi-vehicle common access; Step 62: Assign a permission scope based on the rule document to each semantic record, and open different ranges of semantic library sets for different permissions: for neighbor nodes in a similar geographical area, open the traffic and geographical information on the local map in the public sharing semantics; for neighbor nodes with the same unit group relationship, open the privacy data such as geographical tags, exclusive parking spaces, and historical trajectories of neighbor vehicles within the shared unit park to support multi-vehicle common access; Step 63: Through the constructed hyper-relationship knowledge graph, identify the clustering relationship between vehicles based on various key-value pairs of the node key-value categories, divide vehicles with similar characteristics or behaviors into the same group, so that vehicles can access each other without authorization within the privacy and security range, and the accessed data are all valid data strongly correlated with their respective time and space; Step 64. The administrator loads and updates the cloud data in combination with the infrastructure within the user group to achieve the sharing and retrieval of general data.

8. The decision support method for spatio-temporal dynamic data of the vehicle networking enhanced by knowledge retrieval as claimed in claim 1, wherein The specific steps of Step 7 are as follows: Step 71. Extract the entities and expressions of the question based on the local entity recognition model; Step 72. Vectorize the entities and expressions of the question through the embedding vector model to convert the text information into vector information; Step 73. Identify the clustering relationships between vehicles through the clustering information in the hyper-relation knowledge graph and dynamically determine the access scope; Step 74. Perform similarity matching between the vector information and the index vectors in the local hyper-relation graph file and the public database in the cloud within the permissions to obtain the background knowledge content; Step 75. Combine the obtained background knowledge content with the user's question content to form a prompt content and interact with the large language model; Step 76. Return the reply after the interaction with the large language model to the local and decrypt it reversely based on the encryption process in Step 2 to obtain the final enhanced retrieval result.

Images