Data classification and grading method for vehicle-road-cloud integrated multi-network fusion system

Abstract

The application provides a vehicle-road-cloud integrated multi-network fusion system data classification and grading method, the method realizes automatic data screening and marking by constructing a multi-scene data recognition rule library and analyzing business process characteristics, solves the standardized recognition problem caused by the lack of a unified classification framework for multi-source data; semantic relationship system modeling is realized by means of a multi-layer architecture model and a multi-level model, solving the problems of insufficient semantic abstraction of cross-domain entities and model fragmentation in vehicle-road-cloud; the quantification evaluation of importance and sensitivity is realized by combining the regulatory requirements and the modeled data, solving the problems of missing data classification and grading standards and the imperfect security evaluation system, the method provided by the application forms a closed-loop process of data recognition, semantic modeling, model mapping and grading evaluation, solving the problem that the prior art relies on static rules and cannot dynamically adapt to the changes of data types, interaction types and security requirements in the multi-network fusion scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle-road-cloud data classification and grading, and particularly relates to a vehicle-road-cloud integrated multi-network fusion system data classification and grading method. BACKGROUND

[0002] With the in-depth development of vehicle-road-cloud integration technology, the dynamic security management needs of multi-network fusion data in typical scenarios such as identity authentication and vehicle-road interaction are increasingly prominent.

[0003] In the prior art, a data classification and grading system is constructed by relying on static rules, which is difficult to adapt to the real-time changes of data types, interaction paths and security needs in the multi-network fusion scenario.

[0004] Among them, in the identity authentication scenario, the static rules cannot update the identification rules of vehicle identification data with new authentication protocols, resulting in confusion in the management of sensitive authentication information and ordinary data; in the vehicle-road interaction scenario, the traditional technology uses a fixed semantic analysis template to process unstructured data such as radar point clouds, and when new roadside sensor types are added, entity features cannot be dynamically extracted, resulting in lagging of automatic driving decision data classification; in the vehicle-pile interaction scenario, the static classification model does not include real-time interaction features of charging data, and when the charging protocol version is iterated, the billing data privacy protection strategy cannot be updated synchronously; in the vehicle-cloud interaction scenario, the classification standard of cloud service data relies on artificial preset thresholds, and when the precision of high-precision map slices is upgraded or new traffic event data is accessed, the matching efficiency of security strategies across domains is low.

[0005] In addition, in key business scenarios such as automatic driving, the static risk assessment mechanism lacks real-time analysis capability for the correlation of multi-source data, and it is difficult to dynamically calibrate the importance level of data according to the business process, resulting in disconnection between data security protection and business needs. SUMMARY

[0006] The present application provides a vehicle-road-cloud integrated multi-network fusion system data classification and grading method to solve the problem that the prior art relies on static rules and cannot dynamically adapt to changes in data types, interaction types and security needs in the multi-network fusion scenario.

[0007] In one aspect, the present application provides a vehicle-road-cloud integrated multi-network fusion system data classification and grading method, comprising:

[0008] Obtaining multi-source data of a vehicle-road-cloud interaction scenario, extracting data interaction features of the multi-source data by analyzing business process logic;

[0009] Based on the data interaction features, a multi-scenario data recognition rule library is constructed, the multi-source data is filtered and labeled, and standardized recognition data is output;

[0010] Input the standardized recognition data into a multi-layer architecture model, and output a named entity recognition result through word vector training, bidirectional encoding representation converter semantic extraction, bidirectional long short-term memory network sequence modeling, and conditional random field entity boundary optimization.

[0011] Associate the named entity recognition result with scene semantics, construct a vehicle-road cloud multi-network fusion scene corpus, and form structured semantic data containing entity types and semantic relationships.

[0012] Input the structured semantic data into a multi-level model, and output a modeled data set; the multi-level model includes a concept model formed by abstracting business logic of vehicle-road cloud cross-domain entity objects, a logic model formed by mapping business rules, and a physical model formed by combining data storage and interaction requirements.

[0013] Combine the modeled data set and regulatory requirements to evaluate data importance, sensitivity, and leakage impact, and develop multi-network fusion data classification and grading standards.

[0014] Optionally, the multi-source data of the vehicle-road cloud interaction scene is obtained by analyzing business process logic to extract data interaction features of the multi-source data, including:

[0015] Collecting raw data from vehicle-side interfaces, roadside-side interfaces, and cloud-side interfaces;

[0016] Analyzing the business process of the vehicle-road cloud interaction scene to generate multi-source data;

[0017] Extracting interaction features related to time dimension, space dimension, data structure, and transmission protocol from the multi-source data.

[0018] Optionally, based on the data interaction features, a multi-scene data recognition rule library is constructed to filter and label the multi-source data, and output standardized recognition data, including:

[0019] Based on the interaction features, a multi-scene data recognition rule library containing scene classification rules, data format verification rules, and security level pre-judgment rules is established;

[0020] Using the multi-scene data recognition rule library to perform legality verification on the raw data, and eliminating invalid data, adding scene labels, format labels, and security level preliminary labels to the raw data that passes the verification;

[0021] Converting the labeled raw data into a preset data format;

[0022] Structuring the raw data to generate standardized recognition data containing data identification, scene attributes, security attributes, and raw data content.

[0023] Optionally, the establishing of the multi-scene data recognition rule library comprising scene classification rules and security level pre-determination rules based on the interaction features further comprises:

[0024] establishing a data timeliness rule through a time dimension feature in the interaction features;

[0025] generating a data source verification rule through a space dimension feature in the interaction features;

[0026] establishing a data format verification rule through a data structure in the interaction features;

[0027] generating a data interaction security rule through a transmission protocol feature in the interaction features.

[0028] Optionally, the inputting of the standardized recognition data into the multi-layer architecture model, the outputting of the named entity recognition result through word vector training, bidirectional encoding representation converter semantic extraction, bidirectional long short-term memory network sequence modeling and conditional random field entity boundary optimization comprises:

[0029] performing sub-word segmentation and vector mapping on the standardized recognition data to generate an initial word vector matrix;

[0030] inputting the initial word vector matrix into a pre-trained bidirectional encoding representation converter to extract context-enhanced semantic representation vectors, and outputting feature vectors after multi-layer weighted fusion;

[0031] inputting the feature vectors into a bidirectional long short-term memory network to capture bidirectional dependency relationships of data sequences, and generating intermediate vectors containing time sequence features;

[0032] performing global optimal decoding on the intermediate vectors through a conditional random field layer to output a named entity recognition result containing entity types and position indexes.

[0033] Optionally, the training process of the multi-layer architecture model comprises:

[0034] adopting a weighted sum of a cross-entropy loss function and a conditional random field potential energy function as a total loss function;

[0035] iteratively updating model parameters through a back propagation algorithm and a stochastic gradient descent optimizer until the loss of a validation set converges;

[0036] adopting a dynamic learning rate adjustment strategy and Dropout regularization during the training process to prevent model overfitting.

[0037] Optionally, the associating of the named entity recognition result with scene semantics to construct a car-road cloud multi-network fusion scene corpus to form structured semantic data containing entity types and semantic relationships comprises:

[0038] The entity type in the named entity recognition result is matched with semantic features of a corresponding scene to generate a scene-entity association matrix;

[0039] Based on business process logic, an interaction relationship between entities is extracted to construct a semantic relationship graph;

[0040] The entity type, scene semantic information, and the semantic relationship graph are structured and stored in a corpus in the preset data format to form structured semantic data containing entity types, semantic relationships, and scene attributes. The structured semantic data contains data identifiers, entity content, entity types, belonging scenes, associated entities, and interaction action fields.

[0041] Optionally, the structured semantic data is input into a multi-level model to output a modeled data set. The multi-level model includes a concept model formed by abstracting business logic of vehicle-road cloud cross-domain entity objects, a logic model formed by mapping business rules, and a physical model formed by combining data storage and interaction requirements, including:

[0042] The vehicle-road cloud cross-domain entity objects in the structured semantic data are abstracted and classified to define entity sets and attributes, establish a logical relationship graph between entities, and form the concept model;

[0043] The logical relationships in the concept model are mapped to data processing rules and interaction protocols, data lifecycle management rules are defined, and a logic model is generated;

[0044] According to the requirements of the logic model, a data storage structure and an interaction interface are designed, index optimization and data sharding strategies are established, and a physical model containing storage schemes, interface definitions, and performance optimization rules is output;

[0045] The output results of the concept model, the logic model, and the physical model are integrated to form a modeled data set containing entity definitions, rule constraints, and storage interaction specifications.

[0046] Optionally, the modeled data set is combined with regulatory requirements to evaluate data importance, sensitivity, and leakage impact, and multi-network fusion data classification and grading standards are developed, including:

[0047] Based on the entity attributes, relationship structures, and interaction rules of the modeled data set, the functional positioning of data in vehicle-road cloud business processes is identified, and data criticality indicators are extracted;

[0048] According to national data security regulations and industry standards, data attributes are matched with regulatory provisions to determine the sensitive types involved by the data;

[0049] The data leakage influence evaluation model is constructed to perform weighted scoring in three dimensions of personal privacy leakage risk, commercial value loss risk and national security threat risk, and output a data leakage influence score;

[0050] According to the key degree index, the sensitive type and the data leakage influence score, a multi-network fusion data classification grading standard is generated.

[0051] Optionally, it further comprises:

[0052] According to the multi-network fusion data classification grading standard, the original data is divided into public data, sensitive data and important data, and a differentiated security policy is matched;

[0053] According to the data interaction characteristics, the business process logic and the change of the national data security regulations, the multi-network fusion data classification grading standard and the differentiated security policy are automatically updated;

[0054] The differentiated security policy comprises:

[0055] If the original data is public data, an open access mechanism is implemented to perform basic encryption protection;

[0056] If the original data is sensitive data, desensitization processing is performed and fine-grained access control is applied;

[0057] If the original data is important data, encryption transmission, blockchain storage and a cross-network transmission approval mechanism are forcibly implemented.

[0058] In another aspect, the present application also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the vehicle-road-cloud integrated multi-network fusion system data classification grading method of any of the above.

[0059] In another aspect, the present application also provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the vehicle-road-cloud integrated multi-network fusion system data classification grading method of any of the above.

[0060] In another aspect, the present application also provides a computer program product comprising a computer program, wherein the computer program is executable by a processor to implement the vehicle-road-cloud integrated multi-network fusion system data classification grading method of any of the above.

[0061] The application provides a vehicle-road-cloud integrated multi-network fusion system data classification and grading method. BRIEF DESCRIPTION OF DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0063] Figure 1 Fig. 1 is one of process schematic diagrams of a vehicle-road-cloud integrated multi-network fusion system data classification and grading method provided by an embodiment of the application.

[0064] Figure 2 Fig. 2 is another of process schematic diagrams of a vehicle-road-cloud integrated multi-network fusion system data classification and grading method provided by an embodiment of the application.

[0065] Figure 3 Fig. 3 is a structural schematic diagram of an electronic device provided by an embodiment of the application. DETAILED DESCRIPTION

[0066] In order to make the objects, technical solutions and advantages of the application clearer, the technical solutions in the application will be described clearly and completely below with reference to the drawings in the application. Obviously, the described embodiments are some of the embodiments of the application, but not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without any creative effort belong to the protection scope of the application.

[0067] Figure 1 Fig. 1 is one of process schematic diagrams of a vehicle-road-cloud integrated multi-network fusion system data classification and grading method provided by an embodiment of the application.

[0068] As Figure 1As shown, the vehicle-road-cloud integrated multi-network fusion system data classification and grading method provided by the embodiment of the application mainly includes the following steps:

[0069] 101. Obtain multi-source data of a vehicle-road-cloud interaction scene, and extract data interaction features of the multi-source data by analyzing business process logic.

[0070] The multi-source data is from data of a vehicle end, a road end and a cloud end server. After the multi-source data is obtained, the interaction features between the multi-source data are extracted by analyzing the business process logic. Specifically, the steps of obtaining multi-source data of a vehicle-road-cloud interaction scene and extracting data interaction features of the multi-source data by analyzing business process logic include:

[0071] Collect original data from a vehicle end interface, a road side end interface and a cloud end interface.

[0072] Parse a business process of the vehicle-road-cloud interaction scene to generate multi-source data.

[0073] Extract interaction features of time dimensions, space dimensions, data structures and transmission protocols in the multi-source data.

[0074] The original data is collected from the vehicle end interface, the road side end interface and the cloud end interface. For example, the vehicle end interface can collect basic data, perception data, decision data, running data and vehicle control data of a vehicle; the road side end interface can collect traffic data, application range data, running states and map data; and the cloud end interface can collect control data, life service data, network monitoring data, vehicle service data and user profile data. The original data collectively constitutes a multi-source data basis of the vehicle-road-cloud interaction scene.

[0075] In the process of parsing the original data, each link and process of the vehicle-road-cloud interaction needs to be deeply understood, such as a communication process of a vehicle and a road side device, an interaction process of the vehicle and the cloud end, etc. By parsing the business process, the original data can be converted into multi-source data with clear business meanings. For example, in vehicle-road interaction, road change data, RSU number data, radar data, target recognition data, etc. can be obtained through the original data. In vehicle-cloud interaction, map messages, traffic flow, ETC (Electronic Toll Collection) billing, weather services, etc. can be obtained through the original data.

[0076] In addition, through vehicle-pole interaction, vehicle-pole positioning, charging billing, current voltage, battery state, etc. can be obtained.

[0077] Then, the interactive features of the multi-source data in the time dimension, the space dimension, the data structure, and the transmission protocol are extracted. In the time dimension, the collection time and the update time of the data can be extracted as features; in the space dimension, the geographical position and the coverage range of the data can be extracted as features; in the data structure, the field type and the field length of the data can be analyzed as features; and in the transmission protocol, the transmission mode and the encryption mode of the data can be parsed as features. These interactive features jointly constitute the basis of the correlation and interaction between the multi-source data.

[0078] Through the implementation of the above steps, comprehensive acquisition and in-depth analysis of the multi-source data in the vehicle-road-cloud interaction scene can be realized, thereby providing strong support for subsequent data classification, grading, and efficient management.

[0079] 102. Based on the data interaction features, a multi-scene data recognition rule library is constructed to filter and label the multi-source data, and standardized recognition data is output.

[0080] The scene data recognition rule library covers different types of vehicle-road-cloud interaction scenes, such as highways, urban roads, and parking lots, ensuring accurate recognition of data in various scenes. Through the filtering and labeling steps, redundant and invalid data can be effectively removed, improving the quality and usability of the data. The output standardized recognition data helps to achieve efficient management and optimization of the vehicle-road-cloud integrated multi-network fusion system.

[0081] Specifically, based on the data interaction features, a multi-scene data recognition rule library is constructed to filter and label the multi-source data, and standardized recognition data is output, which specifically includes:

[0082] 201. Based on the interaction features, a multi-scene data recognition rule library is established, which includes scene classification rules, data format verification rules, and security level prediction rules.

[0083] 202. The multi-scene data recognition rule library is used to perform legality verification on the original data, and invalid data is removed. Scene labels, format labels, and security level preliminary labels are added to the original data that passes the verification.

[0084] 203. The labeled original data is uniformly converted into a preset data format.

[0085] 204. The original data is structured to generate standardized recognition data containing data identification, scene attributes, security attributes, and original data content.

[0086] Among them, in the construction of multi-scene data identification rule library, the scene classification rule is used to distinguish the data in different scenes, to ensure the accurate classification of data; the data format verification rule is used to verify whether the data format meets the expectation, to avoid the problem caused by data format error; the security level preliminary judgment rule is to preliminarily judge the security level of data according to the source, content and other factors of data, to provide reference for subsequent data processing. Through the setting of multi-scene data identification rule library, the fine management of multi-source data is realized.

[0087] When the original data is verified for legality by using the multi-scene data identification rule library, the characteristics of the data will be checked to ensure the authenticity and validity of the data. For invalid data that cannot pass the verification, it will be eliminated to avoid interference of invalid data on subsequent data processing. Adding scene label, format label and security level preliminary label to the original data that passes the verification is helpful to realize the rapid identification and classification of the original data.

[0088] In addition, the original data after marking is uniformly converted into a preset data format, which can ensure the consistency and readability of the data. At the same time, the original data is structured to generate standardized identification data containing data identification, scene attribute, security attribute and original data content, which is helpful to realize the comprehensive management and efficient use of data, and provides support for the efficient operation of the vehicle-road-cloud integrated multi-network fusion system.

[0089] For example, vehicle dynamics data such as vehicle speed and acceleration are collected from the vehicle CAN (Controller Area Network) bus interface, BSM (Basic Safety Message) messages broadcast by vehicles are received through the LTE-V2X module of the roadside unit, charging power and SOC (State of Charge) status are obtained from the OCPP (Open Charge Point Protocol) protocol interface of the charging pile, and high-precision map and traffic flow data are pulled through the cloud RESTful API.

[0090] Taking the vehicle-pile interaction scene as an example, the state conversion logic of charging request→authentication→charging execution→settlement notification is analyzed, and the data interaction types of each node are extracted, such as the association data of vehicle VIN (Vehicle Identification Number) code and charging pile ID that need to be transmitted in the authentication stage, to form the business process state.

[0091] In some embodiments, based on the interaction characteristics, a multi-scene data identification rule library containing scene classification rules and security level preliminary judgment rules is established, which further includes:

[0092] A data timeliness rule is formulated through the time dimension feature in the interaction characteristics.

[0093] Data source verification rules are generated through the spatial dimension feature in the interaction feature.

[0094] Data format verification rules are established through the data structure in the interaction feature.

[0095] Data interaction security rules are generated through the transmission protocol feature in the interaction feature.

[0096] Among them, through in-depth analysis of the interaction features of data, including time dimension, spatial dimension, data structure and transmission protocol, etc., the key information that can accurately reflect the characteristics and source of data is extracted. Key information is the basis for building a multi-scenario data recognition rule library.

[0097] Then, for the time dimension feature, data timeliness rules are developed. Data timeliness rules aim to ensure the timeliness and update frequency of data meet expectations, so that the latest and most accurate information can be obtained in the process of data processing and analysis.

[0098] For the spatial dimension feature, data source verification rules are generated. Data source verification rules verify the reliability of data sources to enhance the credibility of data. For example, data from authoritative institutions or certified data sources can be given higher trust, while data from unknown sources needs additional verification and audit.

[0099] In terms of data structure, data format verification rules are established. Data format verification rules are used to verify whether the format of data meets expectations to avoid problems caused by data format errors. For example, by defining a unified data format standard, consistency and readability of data can be ensured, providing convenience for subsequent data processing.

[0100] In addition, for the transmission protocol feature, data interaction security rules are generated. Interaction security rules focus on the security of data in the transmission process, and through the use of encryption, signature and other technical means, data is prevented from being stolen or tampered with during transmission.

[0101] 103、Input standardized recognition data into a multi-layer architecture model, and output named entity recognition results through word vector training, bidirectional encoding representation converter semantic extraction, bidirectional long short-term memory network sequence modeling and conditional random field entity boundary optimization.

[0102] The named entity recognition result can accurately mark the key information in the text, such as location, name, organization name, etc., and provide a basis for subsequent data processing and analysis. After obtaining the named entity recognition result, the entity is classified and graded in detail according to the preset data classification and grading rules. The classification process aims to classify the entity into its corresponding data category, such as personal information, traffic information, environmental information, etc.; and the grading process assigns different security levels to data of different categories according to the sensitivity and importance of the entity.

[0103] Specifically, the standardized recognition data is input into a multi-layer architecture model, and the named entity recognition result is output through word vector training, bidirectional encoding representation converter semantic extraction, bidirectional long short-term memory network sequence modeling, and conditional random field entity boundary optimization. The specific steps are as follows:

[0104] 301. The standardized recognition data is subjected to sub-word segmentation and vector mapping to generate an initial word vector matrix.

[0105] 302. The initial word vector matrix is input into a pre-trained bidirectional encoding representation converter to extract context-enhanced semantic feature vectors, and the feature vectors are output after multi-layer weighted fusion.

[0106] 303. The feature vectors are input into a bidirectional long short-term memory network to capture the bidirectional dependency of the data sequence and generate intermediate vectors containing time sequence features.

[0107] 304. The intermediate vectors are decoded globally by a conditional random field layer to output named entity recognition results containing entity types and position indexes.

[0108] The standardized recognition data is subjected to sub-word segmentation and vector mapping to generate an initial word vector matrix, which can be subjected to special sub-word segmentation to contain vehicle information, road information, cloud data, etc. Then, the word embedding technology is used to map the sub-words to a high-dimensional vector space to form an initial word vector matrix. For example, taking the sentence "Vehicle A drives on the highway, and the cloud receives its speed data" as an example, first, the sub-word segmentation is performed to obtain "Vehicle / A / In / Highway / On / Drive / , / Cloud / Receive / Its / Speed / Data". Then, the word embedding technology is used to map the sub-words to vectors to form an initial word vector matrix.

[0109] The initial word vector matrix is input into the pre-trained bidirectional encoding converter to extract context-enhanced semantic representation vectors. The bidirectional encoding converter can deeply understand the context information in the text, especially for complex relationships in the vehicle-road-cloud data, such as the relationship between the vehicle and the road, the relationship between the vehicle and the cloud data, etc., to generate more accurate semantic representations. After multi-layer weighted fusion, the feature vector is output. For example, taking the sentence "Vehicle A drives on the highway, and the cloud receives its speed data" as an example, the initial word vector matrix is input into the pre-trained bidirectional encoding converter. The bidirectional encoding converter can capture the association between "Vehicle A" and "highway", "cloud" and "speed data", and generate more accurate semantic representation vectors. After multi-layer weighted fusion, the feature vector is obtained.

[0110] Then, the feature vector is input into the bidirectional long short-term memory network to capture the bidirectional dependency of the data sequence and generate an intermediate vector containing time sequence features. The bidirectional long short-term memory network can consider both forward and backward information of the text, and for the time sequence relationship in the vehicle-road-cloud data, such as the time sequence of vehicle driving and the time sequence of cloud data receiving, it can generate more comprehensive feature representations. For example, taking the sentence "Vehicle A enters the highway entrance at 9 am, leaves the exit at 10 am, and the cloud records its full-speed data" as an example, the feature vector is input into the bidirectional long short-term memory network. The bidirectional long short-term memory network can capture the time sequence of "Vehicle A" from the entrance to the exit and the time sequence relationship of the cloud recording speed data, and generate an intermediate vector containing time sequence features.

[0111] Finally, the intermediate vector is globally optimally decoded by the conditional random field layer to output the named entity recognition result containing entity types and position indexes. The conditional random field layer can consider the constraint relationship between labels to ensure that the output named entity recognition result meets the syntax and semantic rules of the vehicle-road-cloud data. Especially for entity types in the vehicle-road-cloud data, such as vehicle information, road information, and cloud data, the conditional random field layer can accurately label these key information. For example, taking the sentence "Vehicle A enters the highway entrance at 9 am, leaves the exit at 10 am, and the cloud records its full-speed data" as an example, the intermediate vector is input into the conditional random field layer for global optimal decoding. The conditional random field layer considers the constraint relationship between labels such as "Vehicle A (vehicle information, position index: 0-4)", "highway entrance (road information, position index: 9-14)", and "cloud (cloud data, position index: 21-22)", and outputs the named entity recognition result containing entity types and position indexes.

[0112] In some embodiments, the training process of the multi-layer architecture model includes:

[0113] The total loss function is a weighted sum of the cross-entropy loss function and the conditional random field potential function.

[0114] The model parameters are iteratively updated by the backpropagation algorithm and the stochastic gradient descent optimizer until the loss of the validation set converges.

[0115] During training, a dynamic learning rate adjustment strategy and Dropout regularization are used to prevent model overfitting.

[0116] Specifically, the cross-entropy loss function is used to measure the difference between the predicted probability distribution of the model and the true label, while the conditional random field potential function is used to consider the constraint relationship between the labels to ensure that the output of the named entity recognition result is more accurate. During training, the gradient is calculated by the backpropagation algorithm, and the model parameters are iteratively updated by the stochastic gradient descent optimizer to gradually reduce the value of the total loss function. At the same time, a dynamic learning rate adjustment strategy is used to dynamically adjust the learning rate according to the loss change during training to improve the convergence speed and performance of the model. In addition, the Dropout regularization technique is used to randomly discard the output of part of the neurons to reduce the overfitting of the model to the training data and improve the generalization ability of the model.

[0117] 104、The named entity recognition result is associated with the scene semantics to construct a vehicle-road-cloud multi-network fusion scene corpus to form structured semantic data containing entity types and semantic relationships.

[0118] The named entity recognition result is associated with the scene semantics to construct a vehicle-road-cloud multi-network fusion scene corpus to form structured semantic data containing entity types and semantic relationships.

[0119] Specifically, named entity recognition is an important task in natural language processing, which can identify entities with specific meanings in text, such as names, place names, and organization names. In the method provided in the embodiments of the present application, named entity recognition is used to identify key entities in a vehicle-road-cloud integrated multi-network fusion system, such as vehicles, roads, and cloud services.

[0120] Then, the identified named entities are associated with the scene semantics. Scene semantics refers to the specific scene described in the text or data and its related context information. By associating named entities and scene semantics, the role and relationship of entities in a specific scene can be further understood.

[0121] Finally, based on the above association information, a vehicle-road-cloud multi-network fusion scene corpus is constructed. The vehicle-road-cloud multi-network fusion scene corpus not only contains information about entity types, but also contains semantic relationships between entities, thereby forming structured semantic data.

[0122] In some embodiments, the named entity recognition results are associated with the scene semantics, a vehicle-road cloud multi-network fusion scene corpus is constructed to form structured semantic data containing entity types and semantic relationships, which includes:

[0123] 401. The entity types in the named entity recognition results are matched with the semantic features of the corresponding scene to generate a scene-entity association matrix.

[0124] 402. Based on the business process logic, the interaction relationship between entities is extracted to construct a semantic relationship graph.

[0125] 403. The entity types, scene semantic information and semantic relationship graph are structured and stored in the corpus according to the preset data format to form structured semantic data containing entity types, semantic relationships and scene attributes.

[0126] Among them, the structured semantic data contains data identification, entity content, entity type, belonging scene, associated entity and interaction action field.

[0127] Specifically, the entity types in the named entity recognition results are matched with the semantic features of the corresponding scene to generate a scene-entity association matrix, which can be realized by natural language processing technology and algorithm, which can automatically identify the named entities in the text and compare and match them with the preset scene semantic features. For example, when identifying the text "Zhang San drives a vehicle on the highway", it can identify "Zhang San" as a personal name entity and "highway" as a place name entity, and associate the personal name entity and place name entity with "drive" and "travel" and other scene semantic features to form a scene-entity association matrix, providing a basis for subsequent analysis.

[0128] Secondly, based on the business process logic, the interaction relationship between entities is extracted to construct a semantic relationship graph, which needs to combine the actual business process of the vehicle-road cloud integrated multi-network fusion system, analyze the interaction behavior and relationship between entities through algorithm, and construct an intuitive semantic relationship graph. For example, when identifying the text "the vehicle interacts with the cloud service for data when passing through the intersection", it can identify the interaction relationship between the vehicle and the cloud service and display it in the semantic relationship graph to help users better understand the operation mechanism of the vehicle-road cloud integrated multi-network fusion system.

[0129] Finally, the entity types, scene semantic information, and semantic relationship graph are structured and stored in the corpus according to the preset data format, forming structured semantic data containing entity types, semantic relationships, and scene attributes. Storing data in the preset data format in the corpus requires appropriate data structures and storage formats to ensure data accuracy and readability. For example, when storing structured semantic data, JSON or XML formats can be used to store entity types, scene semantic information, associated entities, and interaction actions in the form of key-value pairs, facilitating subsequent data analysis and application.

[0130] In addition, structured semantic data contains data identification, entity content, entity type, belonging scene, associated entity, and interaction action fields. Data identification is used to uniquely identify each piece of structured semantic data, ensuring data uniqueness and traceability; entity content records the identified named entity, which is the core part of the data; entity type classifies entities, helping users quickly understand the nature of the entity; the belonging scene describes the specific environment in which the entity is located, providing rich contextual information for data analysis; the associated entity and interaction action fields reveal the interaction and dynamic changes between entities, making it possible to explore the operation mechanism of the vehicle-road cloud integrated multi-network fusion system in depth.

[0131] 105. Input structured semantic data into a multi-level model and output a modeled data set.

[0132] The multi-level model includes a concept model formed by abstracting the business logic of vehicle-road cloud cross-domain entity objects, a logic model formed by mapping business rules, and a physical model formed by combining data storage and interaction needs.

[0133] The multi-level model is a comprehensive data processing framework that covers three levels: concept model, logic model, and physical model. The concept model is a result of summarizing and abstracting entities and relationships. The concept model refines various entities such as vehicles, roads, and cloud services, as well as their interactions and relationships, forming a clear and concise framework.

[0134] The logic model further refines business rules based on the concept model. The logic model maps entities and relationships in the concept model to specific business scenarios, clarifying data meaning, source, and destination. The logic model ensures data accuracy and consistency.

[0135] The physical model optimizes and adjusts the logical model based on data storage and interaction requirements. It considers factors such as the actual data storage environment, access speed, and security, and provides a concrete implementation of the data structures and storage methods in the logical model. The physical model improves data storage efficiency and access speed, enabling data to be read and processed quickly.

[0136] By processing structured semantic data through multi-level models, the structured semantic data is transformed into model-based datasets. These model-based datasets not only have a clear structure and explicit meaning, but also meet the needs of subsequent data analysis and applications.

[0137] Specifically, the input structured semantic data is fed into a multi-level model, and the output modeled dataset includes:

[0138] 501. Abstract and classify cross-domain entity objects of vehicle, road and cloud in structured semantic data, define entity sets and attributes, establish a logical relationship graph between entities, and form a conceptual model.

[0139] 502. Map the logical relationships in the conceptual model to data processing rules and interaction protocols, define data lifecycle management rules, and generate a logical model.

[0140] 503. Based on the logical model requirements, design the data storage structure and interaction interface, establish index optimization and data sharding strategies, and output a physical model that includes storage scheme, interface definition and performance optimization rules.

[0141] 504. Integrate the outputs of the conceptual model, logical model, and physical model to form a modeled dataset that includes entity definitions, rule constraints, and storage interaction specifications.

[0142] Specifically, the cross-domain entities of vehicles, roads, and clouds in structured semantic data are abstracted and categorized. For example, vehicle entities are categorized as "transportation vehicles," and their attributes such as license plate number, vehicle type, and speed are defined; road entities are categorized as "infrastructure," and their attributes such as road name, road segment length, and road surface condition are defined; cloud service entities are categorized as "information technology services," and their attributes such as service type, response time, and data capacity are defined. Next, a logical relationship graph between entities is established, such as the driving relationship between vehicles and roads, and the data interaction relationship between vehicles and cloud services, forming a conceptual model.

[0143] On the basis of the conceptual model, the logical relationships in the conceptual model are mapped to data processing rules and interaction protocols. For example, define the collection frequency, transmission format and verification rules of vehicle driving data; formulate the request and response format, error handling mechanism and data update strategy of cloud service data, etc. At the same time, define the data life cycle management rules such as data storage period, backup strategy and deletion conditions, and generate a logical model.

[0144] According to the requirements of the logical model, design the data storage structure and interaction interface. For example, design a time series database storage structure for vehicle driving data to facilitate quick retrieval and analysis by time sequence; design a spatial database storage structure for road status data to facilitate query and display by geographic location. At the same time, establish index optimization strategies such as establishing indexes for frequently used query conditions to improve retrieval efficiency; develop data sharding strategies such as sharding by data generation time or geographic location to improve data processing parallelism and scalability. Finally, output the physical model containing storage scheme, interface definition and performance optimization rules.

[0145] Integrate the output results of the conceptual model, logical model and physical model to form a modeled data set containing entity definitions, rule constraints, storage and interaction specifications. For example, integrate the definitions and attributes of entities such as vehicles, roads and cloud services into a unified data dictionary; integrate data processing rules, interaction protocols and life cycle management rules into a data governance framework; integrate data storage structures, interaction interfaces and performance optimization rules into the data storage and access layer.

[0146] 106、Combine the modeled data set with regulatory requirements to evaluate data importance, sensitivity and leakage impact, and develop multi-network fusion data classification and grading standards.

[0147] The modeled data set provides a detailed description of data entities, attributes, rules and storage interaction specifications. Combining the modeled data set with regulatory requirements to evaluate data importance, sensitivity and leakage impact can provide a more comprehensive classification of data. Regulations include data protection laws, industry standards and enterprise internal data management systems, etc. Regulations specify specific requirements for data processing, storage, transmission and access, providing clear guidance and constraints for data classification and grading.

[0148] Combining the modeled data set with regulatory requirements for evaluation, multi-network fusion data classification and grading standards can be developed. Multi-network fusion data classification and grading standards divide data of different importance into different levels and develop corresponding protection measures and management requirements for each level. This helps to ensure the safe storage and compliant use of data, while improving the efficiency and accuracy of data management.

[0149] Specifically, the data importance, sensitivity and leakage impact are evaluated in combination with the modeling data set and regulatory requirements, and the multi-network fusion data classification and grading standards are formulated, including:

[0150] 601. Based on the entity attributes, relationship structure and interaction rules of the modeling data set, the functional positioning of the data in the vehicle-road cloud business process is identified, and the data criticality index is extracted.

[0151] Among them, starting from the modeling data set, the entity attributes, relationship structure and interaction rules in the modeling data set are analyzed. By studying the entity attributes, relationship structure and interaction rules, the function and position of the data in the vehicle-road cloud business process are determined. For example, if a certain data directly affects the automatic driving decision or plays a key role in vehicle-road coordination scheduling, the data occupies an important functional position in the business process, and accordingly the corresponding data criticality index is extracted to measure its indispensability and importance to business operation.

[0152] According to the national data security regulations and industry standards, the data attributes are matched with the regulatory provisions to determine the sensitive types involved in the data.

[0153] Among them, the data attributes are matched one by one with the regulatory provisions. For example, if the data contains personal biological characteristics, whereabouts trajectory, etc., it is known from the regulations that it belongs to personal sensitive information; if it involves enterprise core technology, trade secrets, etc., it corresponds to the type of commercial sensitive data, and if it involves national security map information, it corresponds to the type of national security data. Through accurate matching, the sensitive types involved in the data are clearly defined.

[0154] 602. A data leakage impact evaluation model is constructed to weight score in three dimensions of personal privacy leakage risk, commercial value loss risk and national security threat risk, and output the data leakage impact score.

[0155] Among them, around the three dimensions of personal privacy leakage risk, commercial value loss risk and national security threat risk, an evaluation model is constructed. Each dimension is given a reasonable weight, and the score is calculated according to the actual data. For example, if the data leakage may lead to exposure of a large amount of personal identity information, a higher score is given in the dimension of personal privacy leakage risk; if it involves enterprise core commercial secrets, the commercial value loss risk dimension scores higher. The scores of each dimension are multiplied by the corresponding weight and summed to output the data leakage impact score, quantifying the harm degree that data leakage may cause.

[0156] 603. According to the criticality index, sensitive type and data leakage impact score, the multi-network fusion data classification and grading standards are generated.

[0157] Based on the comprehensive indicators of data criticality, sensitivity type, and data leakage impact score, multi-network converged data is classified and graded. Data with high criticality, high sensitivity, and high leakage impact score is classified as the highest level and given the most stringent protection measures; conversely, data with low criticality, low sensitivity, and low leakage impact are subject to relatively lenient management strategies, thus forming a complete set of multi-network converged data classification and grading standards.

[0158] For example, vehicle sensor data contains real-time information such as vehicle speed, braking, and steering. In the autonomous driving decision-making process, vehicle sensor data directly determines the vehicle's driving status and safe operation, and is the core basis for the autonomous driving system to make decisions such as acceleration, deceleration, and avoidance. Its functional positioning in the vehicle-road-cloud business process is crucial, thus assigning a high score to the data criticality index. In accordance with national data security regulations, vehicle sensor data does not involve sensitive content such as personal biometrics or trade secrets, and is therefore classified as a non-sensitive data type.

[0159] Vehicle sensor data receives a low score because it does not involve personal privacy information; the impact of publicly releasing vehicle sensor data on a company's commercial value is minimal, resulting in a medium score; vehicle sensor data generally does not pose a threat to national security, hence the low score. After comprehensive weighting, the impact of data leakage is considered low. Considering the high scores on the data criticality index, the non-sensitive nature of the data, and the low impact of data leakage, vehicle sensor data is classified as medium-level, requiring appropriate data protection measures such as routine access control and data backup.

[0160] For example, vehicle-to-everything (V2X) user registration information is used to identify users and associate them with services. In the vehicle-road-cloud business process, it serves as the foundational data for providing personalized services and management, but it doesn't directly impact core business operations; therefore, its data criticality index is medium. User registration information includes personal information such as name, ID number, and contact information, which, according to regulations, clearly falls under the category of sensitive personal information. Leakage of user registration information would seriously threaten user privacy and security, resulting in a high score; user registration information has a relatively small impact on the company's commercial value, resulting in a medium score; user registration information generally does not threaten national security, resulting in a low score. After weighted scoring, the impact score for data leakage is high. Combining the medium data criticality index, the sensitive personal information category, and the high impact score for data leakage, V2X user registration information is classified as high-level, and strict data encryption and access control measures are implemented to protect it.

[0161] For example, core algorithm data for road traffic management is crucial for achieving efficient traffic scheduling and optimizing road resource allocation. It directly impacts the overall vehicle-road-cloud integrated multi-network fusion system's ability to manage road traffic, making its functional positioning in the business process extremely important, resulting in a high score for the data criticality indicator. Core algorithm data for road traffic management involves core enterprise technologies and key traffic management strategies, classifying it as a commercially sensitive data type. Core algorithm data for road traffic management does not involve personal privacy, resulting in a lower score; however, leakage of this data would damage the enterprise's core competitiveness, resulting in a high score; malicious use of this data could affect traffic order and even national security, resulting in a high score. After weighted scoring, the impact of data leakage is high. Given the high score for data criticality, commercial sensitivity, and the high impact of data leakage, core algorithm data for road traffic management is classified as the highest level, implementing the strictest data protection measures, including multiple encryptions, isolated storage, and dedicated personnel management.

[0162] In some embodiments, such as Figure 2 As shown, the data classification and grading method for the vehicle-road-cloud integrated multi-network fusion system provided by the present invention further includes:

[0163] 107. Based on the multi-network converged data classification and grading standards, the raw data is divided into public data, sensitive data, and important data, and differentiated security strategies are matched accordingly.

[0164] 108. Automatically update the multi-network converged data classification and grading standards and differentiated security strategies based on changes in data interaction characteristics, business process logic, and national data security regulations.

[0165] The differentiated security strategies include:

[0166] If the original data is public data, an open access mechanism is implemented with basic encryption protection.

[0167] If the original data is sensitive, it will be de-identified and fine-grained access control will be applied.

[0168] If the original data is important, then encrypted transmission and blockchain notarization will be enforced, and a cross-network transmission approval mechanism will be established.

[0169] In this process, the raw data is divided into public data, sensitive data, and important data according to the multi-network converged data classification and grading standard, and a differentiated security strategy is matched accordingly.

[0170] Specifically, public data does not involve sensitive information and key business elements, and the security requirement is relatively low. For public data, an open access mechanism is adopted to allow qualified subjects to freely obtain data, while basic encryption protection is implemented to prevent data from being illegally stolen or tampered with during transmission and storage, and to ensure the basic security of data in open use.

[0171] Sensitive data includes data containing sensitive information such as personal privacy and trade secrets. In order to reduce the risk of data leakage, sensitive data is desensitized to remove or conceal sensitive information, such as partially shielding personal identification numbers and bank card numbers. At the same time, fine-grained access control is applied to accurately control the access range and operation type of sensitive data according to the identity, role, and operation permission of the user, and strictly limit the use scenarios of sensitive data.

[0172] Important data has a key influence on business operation, national security, or core interests of an enterprise. For important data, encrypted transmission is enforced to ensure that data cannot be easily cracked even if intercepted during transmission in the network; blockchain storage technology is used to ensure data integrity and authenticity by using the tamper-proof nature of the blockchain, facilitating data tracing and responsibility identification; a cross-network transmission approval mechanism is established, and any cross-network data transmission behavior needs to go through a strict approval process to ensure the security and compliance of data transmission.

[0173] The data classification and grading method of the vehicle-road-cloud integrated multi-network fusion system provided by the application also considers the dynamic changes of data interaction features, business process logic, and national data security regulations, has an automatic updating capability, and continuously monitors data interaction features such as data transmission frequency and changes in interactive objects; tracks adjustments to business process logic such as the development of new businesses and optimization of existing business processes; and real-time attention to updates of national data security regulations to timely obtain the latest regulatory requirements and standards. Once changes in data interaction features, business process logic, or national data security regulations are detected, the update mechanism is automatically triggered. The algorithm and model analyze the change factors to assess their impact on data classification and grading standards and security policies.

[0174] According to the analysis result, the multi-network fusion data classification and grading standard is revised, and the data categories or grading levels are re-divided. At the same time, the differential security policy is updated synchronously to ensure that the security policy matches the new classification and grading standard, so that the data security management always meets the actual needs and regulatory requirements.

[0175] Figure 3 is a structural schematic diagram of an electronic device provided by an embodiment of the application.

[0176] As Figure 3As shown, the electronic device can include a processor 310, a communications interface 320, a memory 330, and a communications bus 340, wherein the processor 310, the communications interface 320, and the memory 330 complete communications with each other through the communications bus 340. The processor 310 can invoke a logical instruction in the memory 330 to execute the vehicle-road-cloud integrated multi-network fusion system data classification and grading method.

[0177] In addition, the logical instruction in the memory 330 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer-readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0178] On the other hand, the present application also provides a computer program product, which includes a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program is executed by a processor, and the computer can execute the vehicle-road-cloud integrated multi-network fusion system data classification and grading method provided by the above-mentioned methods.

[0179] In yet another aspect, the present application also provides a non-transitory computer readable storage medium having a computer program stored thereon, which is executed by a processor to implement the vehicle-road-cloud integrated multi-network fusion system data classification and grading method provided by the above-mentioned methods.

[0180] The device embodiments described above are only schematic, wherein the units shown as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, i.e., they can be located in one place, or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the present embodiment scheme according to actual needs. Those skilled in the art can understand and implement it without creative labor.

[0181] Those skilled in the art can clearly understand the technical solutions of the various embodiments from the above description of the embodiments, and the various embodiments can be implemented by means of software with the necessary general hardware platforms, and of course, can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part of the prior art that makes a contribution, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, and the like, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0182] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features therein; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A vehicle-road-cloud integrated multi-network fusion system data classification and grading method, characterized in that, The method comprises the following steps: acquiring multi-source data of a vehicle-road-cloud interaction scene, extracting data interaction features of the multi-source data by analyzing business process logic; based on the data interaction features, constructing a multi-scene data recognition rule library, filtering and labeling the multi-source data, and outputting standardized recognition data; inputting the standardized recognition data into a multi-layer architecture model, and outputting named entity recognition results through word vector training, bidirectional encoding representation converter semantic extraction, bidirectional long short-term memory network sequence modeling, and conditional random field entity boundary optimization; associating the named entity recognition results with scene semantics, constructing a vehicle-road-cloud multi-network fusion scene corpus to form structured semantic data containing entity types and semantic relationships; abstracting and classifying vehicle-road-cloud cross-domain entity objects in the structured semantic data, defining entity sets and attributes, establishing a logical relationship graph between entities, and forming a concept model; mapping the logical relationships in the concept model to data processing rules and interaction protocols, defining data lifecycle management rules, and generating a logical model; designing data storage structures and interaction interfaces according to the requirements of the logical model, establishing index optimization and data sharding strategies, and outputting a physical model containing storage schemes, interface definitions, and performance optimization rules; integrating the output results of the concept model, the logical model, and the physical model to form a modeled data set containing entity definitions, rule constraints, and storage interaction specifications; combining the modeled data set with regulatory requirements to evaluate data importance, sensitivity, and leakage impact, and developing multi-network fusion data classification and grading standards. 2.The method according to claim 1, characterized in that, The method of acquiring multi-source data of a vehicle-road-cloud interaction scene, extracting data interaction features of the multi-source data by analyzing business process logic, comprises the following steps: collecting raw data from vehicle-side interfaces, roadside-side interfaces, and cloud-side interfaces; analyzing the business process of the vehicle-road-cloud interaction scene to generate multi-source data; extracting interaction features related to time dimension, space dimension, data structure, and transmission protocol from the multi-source data. 3.The method of claim 2, wherein, The method of constructing a multi-scene data recognition rule library based on the data interaction features to filter and label the multi-source data, and outputting standardized recognition data, comprises the following steps: based on the interaction features, establishing a multi-scene data recognition rule library containing scene classification rules, data format verification rules, and security level preliminary judgment rules; using the multi-scene data recognition rule library to perform legality verification on the raw data, eliminating invalid data, adding scene labels, format labels, and security level preliminary labels to the raw data that passes the verification; unifying the labeled raw data into a preset data format; performing structured processing on the raw data to generate standardized recognition data containing data identifiers, scene attributes, security attributes, and raw data content. 4.The method of claim 3, wherein, The method of establishing a multi-scene data recognition rule library containing scene classification rules and security level preliminary judgment rules based on the interaction features, further comprises the following steps: formulating data time limit rules through the time dimension features in the interaction features; generating data source verification rules through the space dimension features in the interaction features; establishing data format verification rules through the data structure in the interaction features; The transmission protocol feature in the interaction feature generates a data interaction security rule. 5.The method according to claim 1, characterized in that, The standardized recognition data is input into a multi-layer architecture model, and a named entity recognition result is output through word vector training, bidirectional encoding representation converter semantic extraction, bidirectional long short-term memory network sequence modeling, and conditional random field entity boundary optimization, including: The standardized recognition data is subjected to sub-word segmentation and vector mapping to generate an initial word vector matrix; The initial word vector matrix is input into a pre-trained bidirectional encoding representation converter to extract context-enhanced semantic representation vectors, and a feature vector is output after multi-layer weighted fusion; The feature vector is input into a bidirectional long short-term memory network to capture bidirectional dependency relationships of data sequences and generate an intermediate vector containing timing features; The intermediate vector is subjected to global optimal decoding by a conditional random field layer to output a named entity recognition result containing entity types and position indexes. 6.The method of claim 1, wherein, The training process of the multi-layer architecture model includes: A weighted sum of cross-entropy loss function and conditional random field potential function is used as a total loss function; Model parameters are iteratively updated by a back propagation algorithm and a stochastic gradient descent optimizer until the loss of a validation set converges; A dynamic learning rate adjustment strategy and Dropout regularization are used in the training process to prevent model overfitting.

7. The data classification and grading method of the vehicle-road-cloud integrated multi-network fusion system according to any one of claim 3, characterized in that, The named entity recognition result is associated with scene semantics to construct a car cloud multi-network fusion scene corpus to form structured semantic data containing entity types and semantic relationships, including: The entity types in the named entity recognition result are matched with semantic features of corresponding scenes to generate a scene-entity association matrix; Based on business process logic, interaction relationships between entities are extracted to construct a semantic relationship graph; The entity types, scene semantic information, and the semantic relationship graph are subjected to structured processing and stored in the corpus in the preset data format to form structured semantic data containing entity types, semantic relationships, and scene attributes; the structured semantic data contains data identification, entity content, entity types, belonging scenes, associated entities, and interaction action fields. The transmission protocol feature in the interaction feature generates a data interaction security rule. The standardized recognition data is input into a multi-layer architecture model, and a named entity recognition result is output through word vector training, bidirectional encoding representation converter semantic extraction, bidirectional long short-term memory network sequence modeling, and conditional random field entity boundary optimization, including: The standardized recognition data is subjected to sub-word segmentation and vector mapping to generate an initial word vector matrix; The initial word vector matrix is input into a pre-trained bidirectional encoding representation converter to extract context-enhanced semantic representation vectors, and a feature vector is output after multi-layer weighted fusion; The feature vector is input into a bidirectional long short-term memory network to capture bidirectional dependency relationships of data sequences and generate an intermediate vector containing timing features; The intermediate vector is subjected to global optimal decoding by a conditional random field layer to output a named entity recognition result containing entity types and position indexes. The training process of the multi-layer architecture model includes: A weighted sum of cross-entropy loss function and conditional random field potential function is used as a total loss function; Model parameters are iteratively updated by a back propagation algorithm and a stochastic gradient descent optimizer until the loss of a validation set converges; A dynamic learning rate adjustment strategy and Dropout regularization are used in the training process to prevent model overfitting. The named entity recognition result is associated with scene semantics to construct a car cloud multi-network fusion scene corpus to form structured semantic data containing entity types and semantic relationships, including: The entity types in the named entity recognition result are matched with semantic features of corresponding scenes to generate a scene-entity association matrix; Based on business process logic, interaction relationships between entities are extracted to construct a semantic relationship graph; The entity types, scene semantic information, and the semantic relationship graph are subjected to structured processing and stored in the corpus in the preset data format to form structured semantic data containing entity types, semantic relationships, and scene attributes; the structured semantic data contains data identification, entity content, entity types, belonging scenes, associated entities, and interaction action fields.

Images