Intelligent network connection automobile data processing system based on data elements and knowledge graph

Abstract

The invention relates to the technical field of knowledge maps, in particular to an intelligent networked automobile data processing system based on data elements and knowledge maps, and the system comprises a data element definition module which analyzes original intelligent networked automobile data elements, distinguishes the original intelligent networked automobile data elements into a vehicle static parameter set and a vehicle dynamic parameter set, and forms a structured map blueprint. According to the method, element analysis is carried out on the intelligent network connection automobile data, and the intelligent network connection automobile data is divided into the static parameter set and the dynamic parameter set, so that the data has hierarchical structural characteristics, and the controllability and definition of subsequent data processing are improved. And in combination with differentiated structure construction of the static storage layer and the dynamic storage layer, a hierarchical data configuration model with clear logic is formed, and the data scheduling efficiency and the system load stability are improved. A geographic grid coding mode is introduced, space division is carried out on entity data with position attributes, spatial dimension information is made to have high-density indexability, and positioning and regionalization operation can be achieved easily.

Description

Technical Field

[0001] The present invention relates to the field of knowledge graph technology, and in particular to an intelligent connected vehicle data processing system based on data elements and knowledge graphs. Background Art

[0002] The field of knowledge graph technology is an interdisciplinary research direction centered on building, managing, and applying semantic information networks. It integrates technical means from multiple fields such as artificial intelligence, databases, semantic networks, natural language processing, and graph computing.

[0003] Traditional knowledge graph technology lacks independent management strategies for dynamically changing attributes when addressing application scenarios with temporal and spatial variations, which can significantly reduce query efficiency. Concurrent retrieval of multidimensional data, limited by linear access paths and shallow indexing, often results in lengthy paths and low hit rates, impacting the real-time and accuracy of relational queries. Sensitive attributes of nodes in graphs are often statically desensitized, which can easily expose private fields during frequent queries or deep traversals, posing a risk of data leakage. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the existing technology and propose an intelligent connected vehicle data processing system based on data elements and knowledge graphs.

[0005] To achieve the above objectives, the present invention adopts the following technical solutions: A data processing system for intelligent connected vehicles based on data elements and knowledge graphs includes:

[0006] The data element definition module analyzes the original intelligent connected vehicle data elements, distinguishes them into vehicle static parameter sets and vehicle dynamic parameter sets, and forms a structured graph blueprint;

[0007] A graph layered storage module, based on the structured graph blueprint, assigns entities of the vehicle static parameter set to the static storage layer and entities of the vehicle dynamic parameter set to the dynamic storage layer to obtain a logical layered data configuration. Based on the logical layered data configuration, the module applies geographic grid coding to spatially segment entities including location attributes to establish an optimized data storage layout.

[0008] A multidimensional index optimization module, based on the optimized data storage layout, defines an access path for entities with spatiotemporal attributes by combining R-tree variant parameters and B+ tree parameters, generates spatiotemporal index construction instructions, and constructs a graph traversal path based on the spatiotemporal index construction instructions by defining an SPO triple index structure and path index parameters to establish a query index system;

[0009] The graph security publishing module applies marginal noise to the graph side information associated with the query based on the received data access instructions and the query index system to obtain preliminary privacy-protected data. Based on the preliminary privacy-protected data, it attaches a query operation traceability code to the anonymized data segment to generate a secure data circulation certificate.

[0010] Preferably, the steps for obtaining the structured graph blueprint are:

[0011] Read the data stream of the original intelligent connected vehicle data elements one by one, extract the timestamp information of each data element, count the number of times each data element changes within a predetermined interval, determine the data elements that exceed the high-frequency threshold as vehicle dynamic parameters, and determine the data elements that do not reach the high-frequency threshold as vehicle static parameters, and generate a vehicle static parameter set and a vehicle dynamic parameter set;

[0012] Based on the vehicle static parameter set and the vehicle dynamic parameter set, the data information of each entity in each parameter set is parsed one by one, the attributes of the entities in the vehicle static parameter set are labeled and classified according to inherent attributes, vehicle model attributes, and production batch attributes, and the attributes of the entities in the vehicle dynamic parameter set are classified according to real-time attribute labels based on real-time displacement, instantaneous speed, real-time position, and real-time temperature, to generate specific attribute labels for each entity;

[0013] Based on the specific attribute labels of each entity, a logical matching relationship is established for the entities in the vehicle static parameter set according to the consistency association rules of vehicle model and production batch, and a logical matching relationship is established for the entities in the vehicle dynamic parameter set according to the spatiotemporal synchronization association rules. In addition, functional dependency rules across attribute labels are established between static entities and dynamic entities to form a structured graph blueprint.

[0014] Preferably, the steps of obtaining the logical hierarchical data configuration are:

[0015] Based on the structured graph blueprint, the entity nodes of the vehicle static parameter set and the entity nodes of the vehicle dynamic parameter set in the structured graph blueprint are parsed item by item, the entity nodes of the vehicle static parameter set with inherent attributes, vehicle model attributes, and production batch attributes are marked one by one, and the marked entity nodes are assigned in batches to the static storage layer to generate a vehicle static parameter storage mapping relationship;

[0016] Based on the structured graph blueprint, the entity nodes of the vehicle dynamic parameter set are retrieved one by one, and the entity nodes of the real-time displacement, instantaneous speed, real-time position and real-time temperature attributes contained in the vehicle dynamic parameter set are extracted. The entity nodes of the real-time attributes are marked one by one and then assigned to the dynamic storage layer in batches to generate a vehicle dynamic parameter storage mapping relationship;

[0017] Based on the vehicle dynamic parameter storage mapping relationship, the real-time data stream corresponding to the entity node of each real-time attribute is obtained, a time window benchmark of a fixed length is set, and the real-time data stream is segmented window by window according to the set time window benchmark, and the entity node data fragments of each real-time attribute within the time window benchmark are segmented to generate a logical hierarchical data configuration.

[0018] Preferably, the step of obtaining the optimized data storage layout is:

[0019] Scanning each entity node included in the logical hierarchical data configuration one by one, extracting entity nodes with location attribute identifiers, parsing longitude and latitude values ​​in the location attributes, separately marking the entity nodes containing the longitude and latitude values, and generating a set of entity nodes with location attributes;

[0020] According to the set of entity nodes with location attributes, bidirectional coordinate encoding is performed on the longitude and latitude values ​​of each entity node, and the longitude and latitude values ​​are respectively mapped to corresponding two-dimensional integer grid numbers. In combination with the preset geographic grid division accuracy, each entity node is classified into a corresponding geographic grid code area to generate an entity space segment corresponding to the geographic grid code;

[0021] According to the physical space segments corresponding to the geographic grid codes, the space segments are numbered and classified in the order of the grid codes, and the classified space segments are submitted to the corresponding storage locations in the preset storage unit in sequence according to the grid numbers to establish an optimized data storage layout.

[0022] Preferably, the steps of obtaining the spatiotemporal index construction instruction are:

[0023] Sequentially searching all entity nodes submitted to a preset storage unit in the optimized data storage layout, screening entity nodes containing both longitude and latitude values ​​and timestamp information, extracting spatial position coordinates and time tags of such entity nodes, and generating a set of entity nodes with spatiotemporal attributes;

[0024] Extracting the longitude, latitude, and timestamp fields of each entity node from the set of entity nodes having spatiotemporal attributes, combining the longitude and latitude values ​​as two-dimensional spatial coordinates, and using the timestamp field as an independent time coordinate to construct a three-dimensional coordinate structure, respectively setting spatial partition dimension parameters of an R-tree variant and time series index parameters of a B+ tree according to the three-dimensional coordinate structure, and generating an entity access path jointly defined by the R-tree variant parameters and the B+ tree parameters;

[0025] According to the entity access path jointly defined by the R-tree variant parameters and the B+ tree parameters, the longitude value, latitude value and timestamp in each set of three-dimensional coordinate structures are extracted, and the three-dimensional coordinate structure is converted into a continuous multidimensional coordinate set through sequential traversal. The multidimensional coordinate set is encoded and mapped according to the preset space-time filling curve identification rules to form a unique one-dimensional key value set, and the one-dimensional key value set is combined with the R-tree variant parameters and B+ tree parameter information to generate a space-time index construction instruction.

[0026] Preferably, the steps of obtaining the query index system are:

[0027] Extract the entity one-dimensional key value, R-tree variant parameter, and B+ tree parameter information carried in the spatiotemporal index construction instruction one by one, restore the entity one-dimensional key value to the original three-dimensional coordinate structure, parse the entity identification information associated with the three-dimensional coordinate structure, and use the entity identification as the primary key to extract the attribute field corresponding to each entity to generate an entity attribute set;

[0028] Based on the entity attribute set, field mapping is performed for each set of entity attribute information according to the subject-predicate-object semantic logic, mapping the entity identifier to the subject field, mapping the attribute label to the predicate field, and mapping the attribute value to the object field. The subject field, predicate field, and object field are combined to construct a corresponding SPO triple. The triple index position parameter is defined in combination with the storage path of the one-dimensional key value to generate an SPO triple index structure.

[0029] According to the SPO triple index structure, the subject field, predicate field and object field index position parameters in the triple index structure are read, the entity node connection information recorded in the index position parameters is extracted, the relationship paths between all SPO triples are combined and arranged, the connection order and traversal route of each entity node in the graph are constructed, the graph traversal path is generated, and a query index system is established according to the graph traversal path.

[0030] Preferably, the steps for obtaining the preliminary privacy protection data are:

[0031] Based on the received data access instructions and the query index system, a graph traversal is performed on each access path to locate all associated graph edges and the start and target nodes in the query path, and a set of original fields with location attribute fields, identity attribute fields, and physiological parameter attribute fields in the start and target nodes are extracted. Fields with a numeric data type and contained in both entities are screened, and the original values ​​in the start and target nodes are recorded to form a set of common numeric sensitive attribute field pairs of the graph edges;

[0032] Calculate the marginal perturbation score of each graph edge according to the set of common numerical sensitive attribute field pairs of the graph edge;

[0033] According to the marginal perturbation score of each graph edge, determine whether the marginal perturbation score is higher than the set perturbation judgment threshold. For all graph edges with marginal perturbation scores higher than the perturbation judgment threshold, generate a pseudo-random offset sequence with a length equal to the normalized value of the marginal perturbation score, and inject the pseudo-random offset sequence into the structure field of the graph edge. Search the corresponding attribute value generalization level for the position attribute field, identity attribute field and physiological parameter attribute field in the starting node and target node connected by the graph edge respectively, replace the field value according to the mapping rule from the original value to the previous level fuzzy interval, merge the updated graph edge structure and node attribute field, and generate preliminary privacy protection data.

[0034] Preferably, the steps for obtaining the secure data circulation certificate are:

[0035] Based on the preliminary privacy-preserving data, locate the source path of each anonymized data segment in the query index system, extract all query operation records associated with the source path, record the execution time, execution terminal code, and path depth level of each query operation, and integrate them to generate a query operation feature parameter set;

[0036] Calculate the query operation traceability code for each segment of data based on the query operation characteristic parameter set;

[0037] Based on the query operation traceability code, the query operation traceability code is attached to the preliminary privacy protection data as a header identification field, the data segment and the query operation traceability code are encapsulated into an independent sealed data unit according to a unified structural format, and the tag number and query path index information are unified to generate a secure data circulation certificate.

[0038] Compared with the prior art, the advantages and positive effects of the present invention are:

[0039] In this invention, by parsing the elements of intelligent connected vehicle data and dividing it into static and dynamic parameter sets, the data is given hierarchical structured features, improving the controllability and clarity of subsequent data processing. By combining the differentiated structural construction of static and dynamic storage layers, a logically clear hierarchical data configuration model is formed, improving data scheduling efficiency and system load stability. A geographic grid encoding method is introduced to spatially partition entity data with location attributes, making spatial dimension information highly indexable, which facilitates positioning and regionalized operations. By combining spatiotemporal attribute entities and binding multidimensional index paths in a combined manner, and jointly defining data access paths using variant and sequence structures, the response speed of cross-dimensional retrieval and large-scale data association is enhanced. By constructing index paths based on entity relationships in structured graphs, graph traversal instructions maintain the advantages of clear direction and shortest path when executing relationship queries. Marginal perturbations are applied during relationship calls to perturb sensitive path data, and a generalization hierarchy is applied to node information to ensure that key privacy fields are not exposed when responding to queries. The operation time, terminal information and path structure in the query behavior are further encoded into a traceability code and encapsulated into the data segment to ensure that the data has a complete traceable path during the circulation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0042] See also Figure 1 The present invention provides a technical solution: an intelligent connected vehicle data processing system based on data elements and knowledge graphs, comprising:

[0043] The data element definition module analyzes the original intelligent connected vehicle data elements, distinguishes them into vehicle static parameter sets and vehicle dynamic parameter sets, and forms a structured graph blueprint;

[0044] The graph layered storage module, based on the structured graph blueprint, assigns entities of the vehicle static parameter set to the static storage layer and entities of the vehicle dynamic parameter set to the dynamic storage layer to obtain a logical layered data configuration. Based on the logical layered data configuration, it uses geographic grid coding to perform spatial segmentation on entities including location attributes to establish an optimized data storage layout;

[0045] The multi-dimensional index optimization module optimizes the data storage layout, defines access paths for entities with spatiotemporal attributes by combining R-tree variant parameters and B+ tree parameters, generates spatiotemporal index construction instructions, and constructs graph traversal paths based on the spatiotemporal index construction instructions by defining SPO triple index structures and path index parameters, thereby establishing a query index system.

[0046] The graph security publishing module applies marginal noise to the graph side information associated with the query based on the received data access instructions and query index system to obtain preliminary privacy-protected data. Based on the preliminary privacy-protected data, it attaches a query operation traceability code to the anonymized data segment to generate a secure data circulation certificate.

[0047] The steps to obtain the structured graph blueprint are:

[0048] Read the data stream of the original intelligent connected vehicle data elements one by one, extract the timestamp information of each data element, count the number of times each data element changes within a predetermined interval, determine the data elements that exceed the high-frequency threshold as vehicle dynamic parameters, and determine the data elements that do not reach the high-frequency threshold as vehicle static parameters, and generate a vehicle static parameter set and a vehicle dynamic parameter set;

[0049] Based on the vehicle static parameter set and the vehicle dynamic parameter set, the data information of each entity in each parameter set is parsed one by one. The attributes of the entities in the vehicle static parameter set are labeled and classified according to inherent attributes, vehicle model attributes, and production batch attributes. The attributes of the entities in the vehicle dynamic parameter set are divided into real-time attribute labels based on real-time displacement, instantaneous speed, real-time position, and real-time temperature to generate specific attribute labels for each entity;

[0050] Based on the specific attribute labels of each entity, a logical matching relationship is established for the entities in the vehicle static parameter set according to the consistency association rules of vehicle models and production batches, and a logical matching relationship is established for the entities in the vehicle dynamic parameter set according to the spatiotemporal synchronization association rules. In addition, functional dependency rules across attribute labels are established between static entities and dynamic entities to form a structured graph blueprint.

[0051] Specifically, the data stream of the original intelligent connected vehicle data elements is read one by one. Specifically, for each data entering the system, the timestamp information carried is first parsed and extracted to record the time when the data is generated. Then, a change counter is initialized for each type of data element, such as vehicle speed, engine speed, vehicle identification code, etc., and a predetermined interval period is set. The predetermined interval period is set according to the expected change rate of the data element and the system processing capacity. For example, it is set to 0.5 seconds for vehicle speed data with faster changes, and it is set to 5 seconds for accumulated mileage with slower changes. Within a continuous observation window, for example, 200 consecutively recorded predetermined interval periods are selected to form an observation window. Statistics are made to see whether the value of the current data element has changed in value or state compared with the value of the previous interval period. If a change occurs, the corresponding change counter is increased by one. After the observation window ends, the total number of changes of each data element is compared with a preset high-frequency threshold. The setting of the high-frequency threshold is based on a large number of historical data samples of various types. Statistical analysis of parameter change frequency. For example, the number of changes in the same type of parameters for all vehicles in the past month is counted, and the average number of changes and standard deviation are calculated. The high-frequency threshold can be set as the average number of changes plus a specific multiple of the standard deviation, such as the average plus 1.5 times the standard deviation. Or, more directly, the number of changes can be sorted and a higher percentile point is taken, such as the number of changes corresponding to the 80th percentile as the threshold. Specifically, if the number of changes of a data element is set to exceed 120 times (i.e., an average of more than 1.2 times per second) within a 100-second observation window consisting of 200 0.5-second intervals, it is determined to be a high-frequency change. Data elements with a number of changes exceeding this high-frequency threshold, such as real-time vehicle speed and GPS coordinates, are classified and determined as vehicle dynamic parameters. Data elements with a number of changes below this high-frequency threshold, such as vehicle identification number (VIN) and engine model, are classified and determined as vehicle static parameters. Finally, based on this classification result, vehicle static parameter sets and vehicle dynamic parameter sets are generated respectively.

[0052] Based on the vehicle static parameter set and vehicle dynamic parameter set generated in the previous step, each data entity in these two sets will be carefully parsed and attribute-labeled. First, for each entity in the vehicle static parameter set, the system will match and classify attributes based on a pre-built parameter metadata knowledge base or a set of configuration rules. This knowledge base or rule set is defined by domain experts in the system initialization phase based on vehicle engineering standards and data specifications, and describes the intrinsic characteristics and classification standards of each static parameter. For example, for unique identifiers such as "Vehicle Identification Number (VIN)" and "Engine Serial Number" that remain unchanged during the life cycle of the vehicle, the system queries the knowledge base to confirm its invariance and then adds it to the vehicle. Classify and assign "inherent attributes" labels. For information describing the basic model characteristics of the vehicle, such as "vehicle brand", "vehicle model name" (for example, "X-type SUV of a certain brand"), and "configuration level" (for example, "prestige version"), the system classifies it and assigns it the label of "model attribute" by matching the model definition information in the knowledge base. For those parameters related to a specific production and manufacturing stage or parts supply, such as "parts supplier code", "in-vehicle infotainment system software version number", and "production date", the system classifies them and assigns them the label of "production batch attribute" based on the batch information definition in the knowledge base; then, for each entity in the vehicle dynamic parameter set, the system will classify it according to its physical The system divides the real-time attribute labels into detailed ones based on the meaning, data source sensor type and real-time change characteristics of the parameters. Specifically, if the parameters directly describe the geographic coordinate sequence of the vehicle in three-dimensional space, such as the longitude, latitude and altitude data periodically reported by the vehicle-mounted GPS module or Beidou module, the system will label it as "real-time position". If the parameters reflect the distance moved by the vehicle in unit time or the vector change of position, such as the short-term driving distance increment calculated by combining the wheel speed sensor data with the tire circumference, or the X, Y, Z axial displacement changes obtained by the inertial measurement unit (IMU), it will be labeled as "real-time displacement". If the parameters clearly indicate the movement of the vehicle at a specific moment, Rate, such as the vehicle speed displayed on the dashboard read from the vehicle's CAN bus or the current ground speed directly sensed by sensors such as radar and cameras, is labeled "instantaneous speed". Parameters that reflect the current working status or environmental status of specific vehicle components, such as the temperature inside the cockpit, the engine coolant temperature, and the maximum / minimum temperature of the power battery pack, are uniformly labeled "real-time temperature" depending on their measurement objects. Through the above precise labeling and real-time attribute label division process, each data entity is given a clear attribute identity that can be used for subsequent processing, and finally a set of specific attribute labels for each entity is generated. This set clearly reveals the nature and classification of each data element.

[0053] Based on the specific attribute labels generated for each entity in the previous step, the system begins to build logical relationships between entities to outline the framework of the structured graph blueprint. First, when processing entities in the vehicle static parameter set, the logical matching relationship between them will be established strictly in accordance with the preset "model and production batch consistency association rules". The rule clearly states that if the system identifies multiple independent vehicle entities, they share exactly the same "model attribute" label value, for example, they all belong to "a certain brand C-class sedan 2.0T luxury version", and their "production batch attribute" label values ​​are also exactly the same, for example, both If the record is "produced in the fourth quarter of 2023, Factory A", an equivalence or strong similarity association link will be established between these entities representing specific vehicles or their static components. For example, this means that they may share the same set of design blueprints, the same list of standard parts or a unified initial firmware version. Such an association will help in the subsequent commonality analysis of vehicles of the same type; secondly, for the entities in the vehicle dynamic parameter set, the "time-space synchronization association rule" is used to construct the logical matching relationship between them. The core of this rule is to capture the multi-source dynamic data generated by the same physical vehicle at similar time points. The specific operation is to first filter out dynamic data points belonging to the same vehicle using the vehicle's unique identifier (such as the VIN code, which has been labeled as "intrinsic attribute" and associated with the dynamic data stream in the previous step). Then, the "timestamp information" of these data points is checked. If the difference between their timestamps is within a very small and preset time window threshold, for example, the threshold is set to no more than 30 milliseconds based on the data acquisition frequency of the on-board sensors and the transmission delay characteristics of the data bus to ensure that events occurring at almost the same time are captured. At the same time, if these dynamic parameters involve spatial location information, such as the vehicle positions reported by multiple sensors (such as GPS and radar), the system will also compare their "real-time location" tag values ​​to ensure that their spatial distance differences are within a preset proximity threshold. For example, the distance threshold is set to less than 2 meters based on the comprehensive accuracy of the positioning system and the application scenario (such as determining whether they are multiple related readings in the same collision event), then the system will determine that these dynamic parameter entities, such as the speed, acceleration, steering angle, and surrounding obstacle distance (if the spatial location verification passes) of the same vehicle at almost the same time (within 30 milliseconds), are temporally synchronized and establish an association between them.Finally, the system will also actively establish important connections between static entities and dynamic entities based on the "functional dependency rule across attribute labels". This rule defines how dynamic data is attached to or generated in a specific static entity context. For example, a specific "instantaneous speed" reading (as a dynamic parameter entity) will establish a functional dependency edge with this specific static vehicle entity through its associated vehicle VIN code (the VIN code corresponds to a static vehicle entity with an "inherent attribute" label), clearly indicating that the speed data is generated by this vehicle. Similarly, a "real-time temperature" reading, such as "the current temperature of the power battery pack" The "degree" (dynamic parameter entity) is not only associated with the static entity of the vehicle that generated the data, but also, through model or batch information, is further associated with the "specific model power battery pack" (a static component entity with "model attributes" or "production batch attributes") installed in the vehicle. By combining the three association rules mentioned above, the homogeneity between static entities, the concurrency between dynamic entities, and the master-slave or generation relationships between dynamic and static entities are accurately modeled. This creates a structured graph blueprint with rich nodes (data entities) and diverse edges (logical relationships), laying the foundation for subsequent data storage, query, and analysis.

[0054] The steps to obtain the logical hierarchical data configuration are:

[0055] Based on the structured graph blueprint, the entity nodes of the vehicle static parameter set and the entity nodes of the vehicle dynamic parameter set in the structured graph blueprint are parsed item by item, the entity nodes of the vehicle static parameter set with inherent attributes, vehicle model attributes, and production batch attributes are marked one by one, and the marked entity nodes are batch assigned to the static storage layer to generate the vehicle static parameter storage mapping relationship;

[0056] Based on the structured graph blueprint, the entity nodes of the vehicle dynamic parameter set are retrieved one by one, and the entity nodes of the real-time displacement, instantaneous speed, real-time position and real-time temperature attributes contained in the vehicle dynamic parameter set are extracted. The entity nodes of the real-time attributes are marked one by one and then assigned to the dynamic storage layer in batches to generate the vehicle dynamic parameter storage mapping relationship;

[0057] Based on the vehicle dynamic parameter storage mapping relationship, the real-time data stream corresponding to the entity node of each real-time attribute is obtained, a time window benchmark of a fixed length is set, and the real-time data stream is segmented window by window according to the set time window benchmark. The entity node data fragments of each real-time attribute within the time window benchmark are segmented to generate a logical hierarchical data configuration.

[0058] Specifically, based on the structured graph blueprint, first traverse each entity node in the blueprint, and distinguish the entity nodes belonging to the vehicle static parameter set by checking the set identifier pre-associated with each node. Then, for each entity node identified as the vehicle static parameter set, further review all the "specific attribute labels of each entity" it possesses. If the attribute label list of an entity node clearly contains at least one of "inherent attributes", "model attributes" or "production batch attributes", for example, a node has an "inherent attribute" label (such as VIN code), a "model attribute" label (such as "a certain brand X-type SUV") and a "production batch attribute" label (such as "produced in the first quarter of 2023"), then the entity node is confirmed as an object that needs to be assigned to the static storage layer, and the system immediately assigns it to the static storage layer. An internal tag is added to the class entity node, such as setting a metadata flag of "storage layer recommendation: static". After completing the inspection and marking of all static parameter set entity nodes, the system collects all entity nodes with the "storage layer recommendation: static" tag and creates a record for each such entity node in a batch processing manner. The record clearly defines the correspondence between the unique identifier of the entity node (for example, its ID or key inherent attribute value in the graph) and the "static storage layer" to which it should be assigned. This static storage layer logically represents a storage mechanism designed for data with low update frequency and possibly high reading frequency, such as using a specific table structure in a relational database or a dedicated collection in a document database. This series of correspondences together constitute the vehicle static parameter storage mapping relationship.

[0059] Based on the structured graph blueprint, the system performs a one-by-one retrieval operation on the entity nodes in the vehicle dynamic parameter set. Specifically, the system iteratively accesses each entity node classified in the vehicle dynamic parameter set in the structured graph blueprint, and carefully analyzes the "specific attribute labels of each entity" of each node to extract those entity nodes with specific real-time attributes. The system will check whether the attribute labels of these dynamic entity nodes contain any of "real-time displacement", "instantaneous speed", "real-time position" or "real-time temperature". For example, a dynamic entity node representing "current GPS coordinates of the vehicle" will be selected because it has the "real-time position" attribute label, and a dynamic entity node representing "current engine speed" will be selected because it has the "instantaneous speed" (rate class in a broad sense) attribute label. All entity nodes whose entity attribute labels match these predefined real-time attribute categories (i.e., real-time displacement, instantaneous speed, real-time position, and real-time temperature) are selected (or a more precise "real-time rate" class label, if defined). The system assigns an internal label to each entity node, such as "Storage layer recommendation: dynamic". After traversing and labeling all entity nodes that meet the conditions for the dynamic parameter set, the system batches these labeled entity nodes and associates their unique identifiers with a logical "dynamic storage layer". This dynamic storage layer is optimized for high throughput, frequent writes, and time series characteristics. For example, it may point to a time series database cluster or a distributed log system that supports fast appends. In this way, a vehicle dynamic parameter storage mapping relationship is generated.

[0060] Based on the vehicle dynamic parameter storage mapping relationship, the mapping relationship is first used to identify all real-time attribute entity nodes assigned to the dynamic storage layer, and then locate the original real-time data streams corresponding to these entity nodes. For example, for an entity node marked as "real-time location", the system will access the GPS data stream of the corresponding vehicle. Subsequently, the system applies a pre-set "fixed-length time window benchmark" to process these data streams. The setting of the "fixed-length time window benchmark", for example, T w , is based on a comprehensive consideration of data characteristics and application requirements. The determination process is as follows: First, analyze the typical sampling frequency of various types of real-time attribute data. For example, "real-time location" data may be 1Hz, and "instantaneous speed" data may be 10Hz. At the same time, consider the requirements of downstream applications (such as real-time monitoring and short-term trend forecasting) for data freshness and aggregation granularity. For example, if a certain type of analysis requires obtaining trend changes within 5 seconds, and the data processing unit takes about 20 milliseconds to process 100 data points, in order to balance the timeliness of the analysis, the amount of information contained in the data segment, and the processing overhead, a time window benchmark is selected. For example, if T is set w= 10 seconds. For 1Hz "real-time position" data, each window will contain 10 data points. For 10Hz "instantaneous speed" data, it will contain 100 data points. The 10-second setting is based on empirical evaluation, that is, it is believed that the amount of data within 10 seconds is sufficient for most short-term analysis, and the generated segment size (10-100 points) will not be too small to cause excessive management overhead, nor too large to increase processing delay. Once this time window benchmark (for example, 10 seconds) is set, the system will perform continuous, non-overlapping window-by-window analysis on each real-time data stream according to this fixed time length. Time series segmentation is specifically performed by defining a time window every 10 seconds starting from the start time of the data stream and collecting all data points arriving within the window time period, thereby dividing the continuous data stream into a series of discrete "entity node data fragments". Each data fragment contains all the actual observation values ​​of "real-time displacement", "instantaneous speed", "real-time position" or "real-time temperature" of a specific entity node (such as a sensor of a certain vehicle) within a specific 10-second time window and its corresponding timestamp. By performing this segmentation operation on all relevant real-time data streams, a logical hierarchical data configuration is finally formed.

[0061] The steps to obtain the optimized data storage layout are:

[0062] Scan each entity node included in the logical hierarchical data configuration one by one, extract the entity nodes with location attribute identifiers, parse the longitude and latitude values ​​in the location attribute, mark the entity nodes containing the longitude and latitude values ​​separately, and generate a set of entity nodes with location attributes;

[0063] According to the set of entity nodes with location attributes, the longitude and latitude values ​​of each entity node are bidirectionally encoded, and the longitude and latitude values ​​are respectively mapped to corresponding two-dimensional integer grid numbers. In addition, each entity node is classified into the corresponding geographic grid coding area in combination with the preset geographic grid division accuracy, and the entity space segment corresponding to the geographic grid code is generated;

[0064] According to the physical space segments corresponding to the geographic grid codes, the space segments are numbered and classified in the order of the grid codes, and the classified space segments are submitted to the corresponding storage locations in the preset storage units in sequence according to the grid numbers to establish an optimized data storage layout.

[0065] Specifically, the system scans each entity node contained in the logical hierarchical data configuration one by one. It will traverse each entity node entry recorded in the logical hierarchical data configuration and check the "specific attribute tags of each entity" carried by each entity node. If the attribute tag of an entity node contains a predefined "real-time location" tag or any other tag that clearly indicates that the entity contains geographic spatial coordinate information (for example, the tag system may be more finely divided into "GPS location", "base station positioning location", etc., but collectively referred to as "location attribute identifier" here), the system will further parse the specific longitude from the data load of the entity node. For example, an entity node with a "real-time location" label may have a data payload of a JSON object containing fields with keys "longitude" and "latitude". The system will extract the values ​​of these two fields and verify whether they are within the legal geographic coordinate range (longitude -180 to 180, latitude -90 to 90). For those entity nodes that successfully extract valid longitude and latitude values, a specific internal tag will be added to them, such as "Include location attribute: yes", and these marked entity nodes will be collected to generate a set of entity nodes with location attributes.

[0066] Based on the set of entity nodes with location attributes generated in the previous step, the system performs a bidirectional coordinate encoding operation on each of the entity nodes. First, the system requires a "preset geographic grid division accuracy". This accuracy determines the size of the grid cells into which the geographic space is divided. The setting basis is the application scenario's requirements for spatial resolution and the expected query efficiency. For example, for fine vehicle management within the city, the grid accuracy is set to an area corresponding to an actual ground distance of 100 meters by 100 meters, while for regional macro traffic flow analysis, it may be set to an area of ​​1 kilometer by 1 kilometer. For example, If the selected geographic grid division accuracy corresponds to a grid division of every 0.001 degrees in the longitude direction and a grid division of every 0.001 degrees in the latitude direction, for a physical node, its longitude value is 116.407395 and its latitude value is 39.904211. The system will perform integer mapping on these two floating-point coordinates respectively. For example, using a linear transformation with rounding down, if the coordinate origin (0, 0) is set to correspond to a fixed reference point in the global range, and the global geographic range is mapped to a sufficiently large integer grid range, then the longitude value 116.407395 can be mapped to an integer grid number. The latitude value 39.904211 can be mapped to an integer grid number Among them, Lon min and Lat minis the minimum longitude and latitude value in the selected geographic coordinate system (or a reference origin). In this way, each entity node with a location attribute is assigned a unique two-dimensional integer grid number (G x ,G y ), this number represents the geographic grid code area where the entity node falls. All entity nodes that fall into the same two-dimensional integer grid number are naturally grouped together to form a physical space segment corresponding to the geographic grid code. This series of entity node sets organized by geographic grid code together constitute the physical space segment corresponding to the geographic grid code.

[0067] Based on the physical space segments corresponding to the geographic grid codes generated in the previous step, the system then numbers and classifies these space segments according to a predetermined grid code sequence. For example, a sequence similar to raster scanning can be used, that is, first sorting the grid numbers from small to large by latitude (or Y axis), and then sorting the grid numbers from small to large by longitude (or X axis) within the same latitude grid. Alternatively, a coding sequence generated by a more complex space filling curve (such as a Z-order curve or a Hilbert curve) can be used to obtain better spatial locality. For example, a raster scanning sequence can be used, and each geographic grid code (for example, (G x ,G y )) itself can be used as part of the number or index of its corresponding physical space segment. The system will traverse all physical space segments and, based on this determined grid coding order, take each classified space segment (i.e., the set of all physical nodes belonging to the same geographic grid code) as a data unit and submit them to the corresponding storage location in the preset storage unit in sequence. The "preset storage unit" here can refer to a specific directory structure in a distributed file system, a specific key prefix in a key-value store, or a specific partition in a database table. Its "corresponding storage location" is the specific storage block or record area mapped according to the grid coding order. For example, the physical space segment corresponding to grid (0, 0) is stored in location A, the physical space segment corresponding to grid (0, 1) is stored in location B, and so on. In this way, data with spatial proximity can be physically organized as much as possible, thereby establishing an optimized data storage layout.

[0068] The steps to obtain the spatiotemporal index construction instructions are:

[0069] All entity nodes submitted to the preset storage unit in the optimized data storage layout are retrieved in sequence, and entity nodes containing longitude, latitude and timestamp information are screened. The spatial position coordinates and time tags of such entity nodes are extracted to generate a set of entity nodes with spatiotemporal attributes.

[0070] Based on a set of entity nodes with spatiotemporal attributes, the longitude, latitude, and timestamp fields of each entity node are extracted. The longitude and latitude values ​​are combined as two-dimensional spatial coordinates, and the timestamp field is used as an independent time coordinate to construct a three-dimensional coordinate structure. Based on the three-dimensional coordinate structure, the spatial partition dimension parameters of the R-tree variant and the time series index parameters of the B+ tree are set respectively, and an entity access path defined jointly by the R-tree variant parameters and the B+ tree parameters is generated.

[0071] According to the entity access path jointly defined by the R-tree variant parameters and the B+ tree parameters, the longitude value, latitude value and timestamp in each set of three-dimensional coordinate structures are extracted, and the three-dimensional coordinate structure is converted into a continuous multidimensional coordinate set through sequential traversal. The multidimensional coordinate set is encoded and mapped according to the preset space-time filling curve identification rules to form a unique one-dimensional key value set, and the one-dimensional key value set is combined with the R-tree variant parameters and B+ tree parameter information to generate a space-time index construction instruction.

[0072] Specifically, all entity nodes in the optimized data storage layout that have been submitted to the preset storage unit are retrieved in turn. Specifically, the system will traverse each entity node recorded in the optimized data storage layout and check its data content to filter out those entity nodes that have longitude, latitude and timestamp information at the same time. For example, the data of an entity node must contain valid fields named "longitude", "latitude" and "timestamp". For such entity nodes that are successfully filtered out, the system will accurately extract the longitude and latitude values ​​they contain as spatial location coordinates, and extract their timestamp information as time tags, for example, longitude value 116.397128, latitude value 39.916527, timestamp 1678886400000 (indicating millisecond-level Unix timestamp). These extracted spatial location coordinates and time tags will be saved together with other metadata of the entity node (such as entity ID) to form a temporary collection, which brings together all qualified entity nodes and their extracted spatiotemporal information to generate a set of entity nodes with spatiotemporal attributes.

[0073] Based on the set of entity nodes with spatiotemporal attributes generated in the previous step, the system processes each entity node in the set. First, the specific longitude, latitude, and timestamp fields are extracted from each entity node. Then, the extracted longitude and latitude values ​​are combined to form a two-dimensional spatial coordinate point (for example, (116.397128, 39.916527)). At the same time, the extracted timestamp field (for example, 1678886400000) is regarded as an independent time coordinate. In this way, a three-dimensional coordinate structure (longitude, latitude, timestamp) is constructed for each entity node. Next, based on this three-dimensional coordinate structure, the system needs to set the spatial partition dimension parameters of the R-tree variant and the time series index parameters of the B+ tree. For the R-tree variant, its spatial partition dimension parameters mainly refer to the maximum and minimum number of entries that its internal nodes and leaf nodes can accommodate. The setting basis of these parameters is the spatial distribution characteristics of the data set and the expected query performance. For example, the maximum number of entries M Rtree It can be estimated based on the average data point density and disk page size. If a disk page can store 50 entries, then M Rtree Can be set to 50, the minimum number of entries is usually set to M Rtree 30% to 40%, for example, 20. These values ​​are selected based on experience of balancing tree height and node utilization. For this scenario, considering the two-dimensional characteristics of vehicle location data, the R-tree variant is mainly responsible for indexing the two-dimensional space of (longitude, latitude). For the B+ tree, its time series index parameter mainly refers to the order (or fan-out rate) of the B+ tree, that is, the maximum number of child nodes that each non-leaf node can have. The setting of this parameter is also based on the distribution range of the timestamp, the time range of the query, and the characteristics of the storage medium. For example, if the timestamp data is distributed relatively evenly and it is expected to quickly locate a specific time point or time period, the order M of the B+ tree is Btree It can be set according to the number of key-value pairs that a single index block can accommodate. For example, if the size of an index block is 4KB, each timestamp key value occupies 8 bytes and the pointer occupies 4 bytes, then a block can accommodate approximately 4096 / (8+4)≈341 entries, so M Btree It can be set to 300. The B+ tree is used here specifically to index the one-dimensional data of timestamp. By configuring the spatial index parameters (such as node capacity) for the R-tree variant and the time index parameters (such as order) for the B+ tree, a combined indexing strategy is actually defined, in which the R-tree variant is responsible for efficient spatial range queries, while the B+ tree is responsible for efficient time point or time range queries. This combination defines the access path of data entities in the spatial and temporal dimensions, and generates the entity access path jointly defined by the R-tree variant parameters and the B+ tree parameters.

[0074] According to the entity access path defined by the R-tree variant parameters and B+-tree parameters generated in the previous step, the system further processes each set of three-dimensional coordinate structures (longitude, latitude, and timestamp) converted from the entity nodes. First, the three numerical components of longitude, latitude, and timestamp are extracted again from each set of three-dimensional coordinate structures. Then, through a certain sequential traversal method, for example, in a fixed order of (longitude, latitude, and timestamp), these three components are regarded as a continuous multi-dimensional coordinate point. Next, this multi-dimensional coordinate set (i.e., three-dimensional coordinate point) is traversed according to a "preset space-time filling curve identification rule". Coordinate points) are encoded and mapped. The rule specifies what type of space-time filling curve to use (for example, Z-order curve, Hilbert curve or its variants optimized for space-time data, such as the encoding method combining GeoHash with timestamp) and the parameters of the curve (such as the encoding precision or number of bits). The role of the space-time filling curve is to map points in multidimensional space to one-dimensional space, while keeping the proximity relationship in the multidimensional space as much as possible in the one-dimensional space. For example, the encoding method based on the Z-order curve is selected to convert the normalized longitude, latitude and timestamp (for example, all converted to integers within a specific range) into one-dimensional space. ) are interleaved and merged to form a single binary number, which is the one-dimensional key value corresponding to the three-dimensional coordinate point. Under this encoding method, points that are relatively close in space and time tend to have similar one-dimensional key values. By applying this space-time filling curve encoding rule to each three-dimensional coordinate structure, the system generates a unique one-dimensional key value for each entity node, collects the one-dimensional key values ​​corresponding to all entity nodes, and forms a unique one-dimensional key value set. Finally, this unique one-dimensional key value set is combined with the previously set R-tree variant parameters (such as node capacity 50) and B+ tree Specifically, each one-dimensional key value will serve as the primary key or index key of the data record, and the construction and query operations of the R-tree and B+ tree will revolve around these one-dimensional key values. The R-tree variant parameters and B+ tree parameters guide the specific construction process and performance characteristics of these index structures. For example, the node splitting strategy of the R-tree will use the node capacity parameter, and the node management of the B+ tree will use the order parameter. These one-dimensional key values, R-tree variant parameters and B+ tree parameters are packaged together to form instructions for the subsequent actual construction of the spatiotemporal index structure and the generation of spatiotemporal index construction instructions.

[0075] The steps to obtain the query index system are:

[0076] Extract the entity one-dimensional key value, R-tree variant parameters, and B+ tree parameter information carried in the spatiotemporal index construction instructions one by one, restore the entity one-dimensional key value to the original three-dimensional coordinate structure, parse the entity identification information associated with the three-dimensional coordinate structure, and use the entity identification as the primary key to extract the attribute fields corresponding to each entity to generate an entity attribute set;

[0077] Based on the entity attribute set, field mapping is performed for each set of entity attribute information according to the subject-predicate-object semantic logic. The entity identifier is mapped to the subject field, the attribute label is mapped to the predicate field, and the attribute value is mapped to the object field. The subject field, predicate field, and object field are combined to construct the corresponding SPO triple. The triple index position parameter is defined in combination with the storage path of the one-dimensional key value to generate the SPO triple index structure.

[0078] According to the SPO triple index structure, the subject field, predicate field and object field index position parameters in the triple index structure are read, the entity node connection information recorded in the index position parameters is extracted, the relationship paths between all SPO triples are combined and arranged, the connection order and traversal route of each entity node in the graph are constructed, the graph traversal path is generated, and the query index system is established according to the graph traversal path.

[0079] Specifically, the entity one-dimensional key value, R-tree variant parameter and B+ tree parameter information carried in the spatiotemporal index construction instruction are extracted one by one. The specific method is to parse each spatiotemporal index construction instruction and separate the previously generated entity one-dimensional key value, such as a long integer or string encoding. At the same time, the R-tree variant parameter (such as node capacity 50) and B+ tree parameter (such as order 300) associated with the key value are also recorded. Then, the system uses the preset spatiotemporal filling curve identification rule (the rule is exactly the same as the rule used when generating the one-dimensional key value) to perform a reverse decoding operation on the entity one-dimensional key value and restore it to the original three-dimensional coordinate structure (longitude, latitude, timestamp). For example, if the one-dimensional key value is obtained by interlacing the binary bits of longitude, latitude and timestamp through the Z-order curve, the reverse decoding is to separate these interlaced bits and reconstruct them back to the three original values. After successfully restoring the original After the three-dimensional coordinate structure is formed, the entity identification information that is implicitly or explicitly associated at the beginning of the formation of the three-dimensional coordinate structure is further parsed. This entity identification information is usually the vehicle's unique identification number (VIN) or the unique ID assigned to the data point by the internal system. The system uses this entity identifier as the primary key and backtracks to find all other attribute fields and their values ​​corresponding to the entity node in the original data element. For example, in addition to latitude and longitude and timestamp, it also contains information such as vehicle speed, engine temperature, and driving behavior. All attribute fields directly associated with the entity identifier and their corresponding values ​​are collected to form a key-value pair set, where the key is the attribute name (that is, the "specific attribute label of each entity") and the value is the specific observation value of the attribute. For each entity, this attribute extraction process is performed, and finally a set containing all its non-temporal and spatiotemporal attributes is generated for each entity, that is, the entity attribute set.

[0080] Based on the entity attribute set generated in the previous step, the system processes each set of entity attribute information, aiming to convert it into the SPO (Subject-Predicate-Object) triple format that conforms to the knowledge graph storage and query specifications. During the specific conversion, the system follows the predefined semantic logic to perform field mapping: the entity identifier extracted in the previous step (for example, the vehicle VIN code "LNVSA2023XYZ001") is used as the subject (Subject) field in the triple, representing the core object described. Then, each attribute field in the entity attribute set of the entity is traversed, and the attribute name or label (that is, "the specific attribute label of each entity", such as "real-time speed" and "engine temperature") is used as the predicate (Predicate) field in the triple, representing a certain characteristic of the subject or its relationship with other objects. Finally, the specific attribute value corresponding to the attribute field (for example, the speed value is 60km / h, the temperature value is 85℃) is used as the object (Object) field in the triple, representing the specific measurement of the characteristic or the associated object. Through this mapping Each original "entity-attribute-value" correspondence is converted into one or more SPO triples, for example, ("LNVSA2023XYZ001", "real-time speed", "60km / h"), ("LNVSA2023XYZ001", "engine temperature", "85℃"). After constructing these SPO triples, they need to be associated with the storage path information of the previously obtained entity one-dimensional key value. This one-dimensional key value (for example, the key value obtained by encoding the space-time filling curve) itself implies the entity's location in space and time. The logical storage location or quick retrieval path in the index (such as R-tree and B+ tree combined index). Therefore, the system uses this one-dimensional key value or some pointer or offset derived from it as a parameter to define the index position of the SPO triple. For example, the corresponding one-dimensional key value can be recorded in the metadata of the SPO triple, or a mapping from the one-dimensional key value to the relevant SPO triple set can be established in an auxiliary index table. By combining the SPO triple with the storage path of the one-dimensional key value, the effective link between the graph data and the spatiotemporal index is realized, and the SPO triple index structure is generated.

[0081] According to the SPO triple index structure generated in the previous step, the system first reads the index position parameters of the subject field, predicate field and object field of each SPO triple from the structure. This index position parameter actually refers to the storage information or retrieval path associated with the one-dimensional key value defined in the previous step. Next, the system needs to use these index position parameters and combine them with the content of the SPO triple itself (especially the predicate field, which may represent a direct relationship between entities, such as "isLocatedAt") to extract the entity node connection information recorded therein. The connection information of the entity node is reflected in the SPO triple. When the object (Object) of a triple is itself also the subject (Subject) of another or more triples, or when multiple triples share the same subject but point to different objects through different predicates, a connection between entities is formed. The system will combine and arrange the relationship paths that may be formed between all SPO triples. For example, starting from a starting entity, through its An SPO triple serving as a subject finds a predicate and an object. If this object is also the subject of another SPO triple, traversal can continue along this path. This process essentially builds an adjacency list of a graph represented by SPO triples or other data structures required for graph traversal. By systematically analyzing all SPO triples, we identify which entities are connected to each other through which attributes (predicates), thereby clarifying the connection order of each entity node in the graph and the potential traversal routes. The collection of these connection orders and traversal routes, after organization and optimization (for example, precalculation or caching for frequently queried paths), constitutes a graph traversal path. Finally, based on these carefully constructed graph traversal paths, the system establishes an efficient query index system. This system may include fast search indexes for subjects, predicates, and objects (such as hash indexes or B-tree indexes), as well as path indexes or subgraph indexes for complex graph patterns (such as multi-hop queries), allowing users or applications to quickly retrieve information in the graph based on various conditions.

[0082] The steps for obtaining preliminary privacy protection data are:

[0083] Based on the received data access instructions and query index system, a graph traversal is performed on each access path to locate all associated graph edges and the start and target nodes in the query path. The original field sets with location attribute fields, identity attribute fields, and physiological parameter attribute fields in the start and target nodes are extracted. Fields with numeric data types and contained in both entities are screened, and the original values ​​in the start and target nodes are recorded to form a set of common numeric sensitive attribute field pairs of the graph edges.

[0084] Based on the set of common numerical sensitive attribute field pairs of the graph edges, the marginal perturbation score of each graph edge is calculated using the following formula:

[0085]

[0086] Among them, Q k represents the marginal perturbation score of the k-th graph edge, F k Indicates the access frequency of the kth graph edge in the past 30 days, n k Indicates the number of numeric sensitive fields shared between the starting node and the target node of the k-th graph edge. and The original value of the field in the starting node and the target node, Range i is the value range of the field on all entities in the entire graph, L k is the nested level position of the edge in the path structure, R k The number of repeated visits to the side in the past 30 days;

[0087] According to the marginal perturbation score of each graph edge, determine whether the marginal perturbation score is higher than the set perturbation judgment threshold. For all graph edges with marginal perturbation scores higher than the perturbation judgment threshold, generate a pseudo-random offset sequence with a length equal to the normalized value of the marginal perturbation score, and inject the pseudo-random offset sequence into the structure field of the graph edge. For the position attribute field, identity attribute field and physiological parameter attribute field in the starting node and target node connected by the graph edge, respectively, search for the corresponding attribute value generalization level, replace the field value according to the mapping rule from the original value to the previous level fuzzy interval, merge the updated graph edge structure and node attribute field, and generate preliminary privacy protection data.

[0088] Specifically, based on the received data access instructions and query index system, the system first parses the data access instruction, such as a request submitted through an application program interface (API) that specifies a query range (such as the behavior data of a specific vehicle group within a specific time period) or a specific graph pattern (such as finding vehicles that have had a common stop point with multiple other vehicles). Using the query index system constructed in the previous step, the system performs an efficient traversal operation in the knowledge graph for each potential access path related to the logical query plan parsed from the instruction. This traversal may be a fast jump based on the index, or it may be the execution process of a graph algorithm (such as the shortest path or pattern matching algorithm). During the traversal process, the nodes that constitute the current query plan will be identified and located. All graph edges of the previous query path results, as well as the start node and target node entities connected by these graph edges, for each located start node (for example, an entity representing a vehicle) and target node (for example, an entity representing another vehicle or a location), the system will extract the original field set based on a pre-configured "sensitive attribute definition list". The list is defined by the data security administrator based on privacy regulations and business scenarios, and clearly defines attributes such as "vehicle precise GPS coordinate sequence", "vehicle VIN code", "driver associated mobile phone number", "driver driving behavior score", "vehicle real-time energy consumption data" as sensitive attributes. The system will retrieve whether the node contains the "location attribute field" listed in the list. "segment" (for example, the field name is geolocation_history, and its value is a JSON array containing a series of latitude, longitude, and timestamps), "identity attribute field" (for example, the field name is vehicle_unique_identifier, and its value is the vehicle's VIN code; or the field name is driver_session_id, and its value is a temporary session identifier related to the driver) and "physiological parameter attribute field" (here refers to a highly sensitive numerical parameter related to the vehicle status or driver status, for example, if the driver's biometric information is collected, it is driver_fatigue_level_score, and the numerical range is 100). The system then compares the raw field values ​​extracted from the start and target nodes of each identified graph edge and selects fields that meet both of the following conditions: First, the field data type must be numeric (such as integer or floating point, for example, "instantaneous speed" is 67.5, but not "weather condition" is "clear"); Second, the attribute field name must exist in the attribute lists of both the start and target nodes (i.e., it must be a common attribute for both). For example, if the start node has an "average fuel consumption" value of 7.2 liters / 100 kilometers, and the target node (which could be a snapshot of the same vehicle at a different time, or another closely related vehicle) also has an "average fuel consumption" of 7.5 liters / 100 kilometers, then "average fuel consumption" is selected. For each public numeric sensitive attribute field thus screened, the system faithfully records its original value in the starting node and the original value in the target node, and organizes these paired records (specifically, attribute name, value on the starting node, and value on the target node) to form a corresponding set of public numeric sensitive attribute field pairs for each graph edge being analyzed.

[0089] formula: The formula is beneficial in that it dynamically assesses the marginal information leakage risk that each graph edge may bring during data query by comprehensively considering the access frequency characteristics of the edge, the degree of difference in sensitive attribute values ​​between the nodes connecting the two ends of the edge, the structural position of the edge in the query path, and the total number of historical repeated accesses to the edge. It is designed to identify edges that are not frequently accessed but have significant differences in sensitive information or are located in critical path positions. Once the original information of these edges is leaked, it may be easier to associate and analyze valuable private content. The multiple parameters in the formula work together: The item makes the edges with fewer daily access days obtain higher basic score weights, and the information leakage risk of the edges with occasional access is relatively greater; The term quantifies the sum of the normalized differences between the nodes at both ends of the edge in terms of the shared sensitive attributes. The larger the difference, the greater the potential amount of information and the higher the risk. The item introduces the influence of path structure and total number of historical visits. The deeper the path (L k Large) means more specific information, while the greater the total number of historical visits (R k Large) means that the information is relatively common, and the marginal risk of a single visit is reduced. This multi-dimensional combination makes the risk assessment more comprehensive and adaptive;

[0090] F k Indicates the access frequency of the kth graph edge in the past 30 days, specifically the number of days in the past 30 consecutive days that the graph edge was accessed at least once. The acquisition steps are: maintain an access log to record all graph edges accessed daily and their timestamps. For a specific kth graph edge, query the access log of the past 30 days and count the number of independent days it appears. For example, if a graph edge X has been accessed on the 1st, 3rd, 10th, 11th, and 25th day in the past 30 days, then its F k =5 days;

[0091] n kIndicates the number of shared numeric sensitive fields between the starting node and the target node of the k-th graph edge. This parameter comes directly from the "set of common numeric sensitive attribute field pairs of graph edges" generated in the previous step. The acquisition steps are as follows: for the k-th graph edge, directly count the number of entries contained in its corresponding "set of common numeric sensitive attribute field pairs of graph edges". Each entry represents a shared numeric sensitive field. For example, if the set contains two records ("instantaneous speed", [value 1s, value 1t]) and ("battery temperature", [value 2s, value 2t]), then n k =2;

[0092] and They represent the original values ​​of the i-th common numerical sensitive field on the k-th graph edge in the starting node and the target node respectively. These two parameters also come from the “common numerical sensitive attribute field pair set of graph edges”. The acquisition steps are as follows: for the k-th graph edge and its i-th common numerical sensitive field (for example, “instantaneous speed”), extract the value of the starting node from its corresponding entry in the “common numerical sensitive attribute field pair set of graph edges” as Extract the value of the target node as For example, for "Instantaneous Speed", if the starting node value is 60.5 (km / h) and the target node value is 68.0 (km / h), then

[0093] Range i It represents the value range of the i-th common numeric sensitive field on all entities in the entire graph, that is, the difference between the global maximum and minimum values ​​of the attribute. The acquisition steps are: regularly (for example, once a day) or after the knowledge graph data is updated, for all defined numeric sensitive fields in the "Sensitive Attribute Definition List", traverse all entity nodes in the graph that contain the field and find the global maximum observation value Max of the field. i and the global minimum observation value Min i , then Range i =Max i -Min i For example, for the "instantaneous speed" attribute, by analyzing historical data, it is found that in all vehicle records, the minimum speed is 0km / h and the maximum speed is 220km / h, so its value range is Range 速度 =220-0=220; For the "battery temperature" attribute, the minimum observed value is -10℃ and the maximum observed value is 60℃, so its value range is Range 温度 =60-(-10)=70;

[0094] L kIndicates the nested hierarchical position of the kth graph edge in the query path structure formed by the currently evaluated data access instruction. The hierarchical level is counted from 1. The acquisition steps are: when the system executes the graph traversal according to the data access instruction and locates the specific query path, the depth of each edge in the path relative to the query starting point is recorded. The L of the first edge k =1, L of the second side k =2, and so on. For example, if the current query path is node A->edge 1->node B->edge 2 (that is, the kth edge)->node C, then for edge 2, its L k =2;

[0095] R k R represents the total number of repeated visits to the kth graph edge in the past 30 days, that is, the total number of times the graph edge has been visited in the past 30 consecutive days. The acquisition steps are: query the access logs of the past 30 days and accumulate the total number of times the kth graph edge appears, regardless of whether it is in the same day or the same query. For example, if a graph edge Y is included in the query path a total of 150 times in the past 30 days, then its R k =150 times;

[0096] Calculation process:

[0097] The parameters of the k-th graph edge are obtained as follows:

[0098] F k = 5 days (visited 5 days in the past 30 days);

[0099] n k =2 (there are two common numerical sensitive fields: instantaneous speed and battery temperature);

[0100] For field 1 (instantaneous velocity): Range1 = 220 km / h;

[0101] For Field 2 (Battery Temperature): Range2=70℃;

[0102] L k =2 (the position of the second layer edge in the current query path);

[0103] R k = 150 times (a total of 150 visits in the past 30 days);

[0104] Calculation process:

[0105] Calculate the frequency impact factor:

[0106] Calculate the sum of normalized attribute differences

[0107] For field 1 (instantaneous velocity): For Field 2 (Battery Temperature):

[0108] The sum of normalized attribute differences is: 0.0340909+0.05=0.0840909;

[0109] Calculate the path and repeat visit impact factor:

[0110] Calculate the total marginal perturbation score Q k :Q k =0.408248·(0.0840909+0.0132450)Q k =0.408248

[0111] 0.0973359Q k ≈0.039736;

[0112] The result shows that the marginal perturbation score Q of the k-th graph edge is k It is approximately 0.039736. This value quantifies the potential information leakage risk carried by the graph edge under the current query conditions. The higher the value, the greater the risk and the greater the possibility of privacy protection processing.

[0113] The marginal perturbation score of each graph edge calculated in the previous step, such as the marginal perturbation score Q calculated in the previous example k ≈0.039736, and compare it with a pre-set disturbance determination threshold. The setting process of the disturbance determination threshold is: first, collect the marginal disturbance scores of each graph edge calculated under a large number of historical data access scenarios in the system to form a score distribution sample, and then select a specific percentile point in the distribution as the threshold based on the acceptable privacy leakage risk level and the expected data availability. For example, select the value corresponding to the 85th percentile of all marginal disturbance scores. If this value is 0.05, then set the disturbance determination threshold to 0.05. For Q k ≈0.039736. Since it is less than 0.05, this edge will not be considered a high-risk edge in this judgment. If the Q of the other edge kIf the value is calculated as 0.07, then because it is greater than 0.05, the graph edge is determined to require privacy processing. For all graph edges determined to require privacy processing (i.e., edges whose marginal perturbation scores are higher than the perturbation determination threshold), the system first normalizes their marginal perturbation scores. This normalization aims to map the scores to a unified scale that can be used to control the subsequent perturbation intensity. For example, if the actual effective range of the marginal perturbation score is determined to be between 0 and 0.2 (scores higher than 0.2 are extremely rare or have been specially handled), and the target perturbation intensity level is set to an integer from 1 to 10, then the normalization formula can be: Normalized perturbation intensity S Q =ceil(((Q k -T Q ) / (Q upper_bound -T Q ))·9)+1, where T Q is the disturbance judgment threshold (e.g. 0.05), Q upper_bound is the upper limit of the score (for example, 0.2), ensuring that S Q On a scale of 1 to 10, k =0.07, then S Q =ceil(((0.07-0.05) / (0.2-0.05))·9)+1=

[0114] ceil((0.02 / 0.15)·9)+1=ceil(1.2)+1=2+1=3, and the system then generates a pseudo-random offset sequence based on this standardized perturbation intensity (for example, 3). The characteristics of the sequence (such as length, numerical range, and distribution type) are determined by the intensity value. For example, an intensity of 3 may mean adding a random number drawn from a Gaussian distribution with a mean of 0 and a standard deviation of a specific small value (such as 0.01 multiplied by the intensity value) to a certain numerical attribute of the edge (such as the weight, if any), or if the edge represents a spatial path, offset several intermediate points on its path, and the offset of each point is also related to the intensity, and inject this carefully constructed pseudo-random offset into the structural description field of the atlas edge, or modify its associated attribute value; at the same time, the system queries the pre-established "location attribute field", "identity attribute field" and "physiological parameter attribute field" identified in the starting node and the target node connected by this high-risk atlas edge. The "Attribute Value Generalization Hierarchy Comparison Table" defines multiple levels of abstraction for each sensitive attribute, from the most precise original value to progressively more fuzzy values. For example, the hierarchy for a "location attribute field" might be: original latitude and longitude -> 1-kilometer grid ID -> administrative division name -> city name. The system strictly follows the "mapping rules from original value to upper-level fuzzy interval" to replace the original values ​​of these fields with the fuzzy interval or categorical value of the immediately upper level (or the number of levels of upward generalization determined by the perturbation intensity) in the corresponding generalization hierarchy comparison table. For example, the original latitude and longitude are replaced with the 1-kilometer grid ID where they are located, and the user's specific age is replaced with an age range (e.g., 30-39 years old). The modified graph edge structure information, which has undergone pseudo-random offset sequences, is then merged with the attribute field values ​​in the nodes after undergoing generalization hierarchical processing to form updated graph data fragments with reduced privacy leakage risks. All these processed data fragments together constitute the initial privacy-preserving data.

[0115] The steps to obtain the secure data circulation certificate are:

[0116] Based on preliminary privacy-preserving data, we locate the source path of each anonymized data segment in the query index system, extract all query operation records associated with the source path, record the execution time, execution terminal code, and path depth level of each query operation, and integrate them to generate a set of query operation feature parameters.

[0117] Based on the query operation characteristic parameter set, the query operation traceability code of each data segment is calculated using the following formula:

[0118]

[0119] Among them, Z d Indicates the query operation traceability code corresponding to the d-th segment of data, r dIndicates the number of query operation records associated with the d-th segment of data, Γ dj Indicates the depth level of the j-th query operation record in the path (in integer), U dj represents the execution timestamp of the j-th query operation (in milliseconds), Λ d Indicates the upper limit of the path hash space of the terminal to which the d-th segment of data belongs (a fixed positive integer);

[0120] Based on the query operation traceability code, the query operation traceability code is attached to the preliminary privacy protection data as a header identification field. The data segment and the query operation traceability code are encapsulated into independent sealed data units according to a unified structural format, and the tag number and query path index information are unified to generate a secure data circulation certificate.

[0121] Specifically, based on the preliminary privacy protection data, first for each anonymized data segment, such as a vehicle trajectory segment that has been processed with marginal noise injection and attribute value generalization, or a group of interaction records of related vehicles, use the original data source information or metadata tags retained during the generation process to trace back in the previously constructed query index system to accurately locate which specific data access instructions (i.e., query operations) originally generated the data segment. This positioning process may involve searching the SPO triple index structure for records related to the entities in the data segment, and further tracking the generation timestamps and associated query session IDs of these records. Once the source path for generating the data segment is determined (i.e., a series of related query operation sequences), the system extracts detailed records of all related query operations on this source path from the system log or query history records. For each extracted query operation record, the system will accurately record its specific The "execution time" of the query body (for example, a Unix timestamp accurate to the millisecond, such as 1678886400123) is used to identify the precise moment when the operation occurred; the "execution terminal code" is recorded (for example, a unique string or numeric ID, such as "Terminal_App_007" or the IP address "192.168.1.101", which is used to identify the device or application instance that initiated the query); and the "path depth level" of the query operation in the complex query logic chain that constitutes the final data segment is recorded (an integer counting starting from 1, indicating the order of the operation in the multi-step query. For example, if the data segment is obtained by query A and further filtered by query B, the depth of A is 1 and the depth of B is 2). All this information (execution time, execution terminal code, and path depth level) related to the source path of the current data segment is collected and integrated one by one to generate a corresponding query operation feature parameter set for each segment of preliminary privacy-preserving data.

[0122] formula: The benefit of the formula is that it gives each piece of anonymized data a unique traceability code that can reflect its query history. This traceability code integrates the depth and time information of the query operation and is calculated in combination with the terminal-specific hash space. In this way, during the data circulation process, even if the data content itself has been processed for privacy protection, the source and processing path of the data can still be effectively tracked and audited based on this traceability code. dj The term gives greater weight to deep query operations. Deeper queries (usually meaning more refined or more targeted data extraction) contribute more to the formation of the final data segment. round(log2U dj ) item takes the logarithm of the timestamp and rounds it off, which not only preserves the relative influence of the time sequence, but also compresses and discretizes the absolute size of the timestamp, reducing the excessive sensitivity of the traceability code to small changes in the timestamp, while introducing nonlinear characteristics; the final modular operation modΛ d Constraining the calculation results to a limited range defined by the upper limit of the terminal hash space ensures a fixed length of the traceability code and a certain degree of collision resistance. This combination enables the traceability code to reflect the complexity and timing of the query operation while also being practical and secure.

[0123] r d Indicates the number of query operation records associated with the d-th segment of data. This parameter comes directly from the "query operation characteristic parameter set" generated in the previous step. The acquisition steps are: for the d-th segment of preliminary privacy-protected data, directly count the total number of query operation records contained in the corresponding "query operation characteristic parameter set". For example, if the "query operation characteristic parameter set" contains three independent query operation records, corresponding to the three query or data processing steps in the process of generating this data segment, then r d =3;

[0124] Γ dj Indicates the depth level of the j-th query operation record in the path (in integers). This parameter also comes from the "path depth level" information of each record in the "query operation characteristic parameter set" generated in the previous step. The acquisition steps are: for the j-th query operation record associated with the d-th segment data, directly read the "path depth level" value of the record from the "query operation characteristic parameter set". For example, if the j-th query operation is the first operation in the entire query chain, then Γ dj =1, if it is the second one, then Γ dj =2, and so on. For the dth segment of data, the depth levels of the three associated query operation records are Γ d1 =1,Γ d2 =2,Γ d3 =1;

[0125] U dj Indicates the execution timestamp of the j-th query operation (in milliseconds). This parameter also comes from the "execution time" information of each record in the "query operation characteristic parameter set". The acquisition steps are: for the j-th query operation record associated with the d-th segment data, directly read the "execution time" value of the record from the "query operation characteristic parameter set". The value is a long integer number representing the number of milliseconds since January 1, 1970 00:00:00 GMT. For example, the execution timestamp of the first query operation is U d1 =1678886400123 milliseconds, the second one is U d2 =1678886405789 milliseconds, the third one is U d3 =1678886410345 milliseconds;

[0126] Λ d Indicates the upper limit of the path hash space of the terminal to which the d-th segment of data belongs (a fixed positive integer). This parameter is a system-preset configuration value related to the terminal or user group that performs the query operation. It is used to control the value range of the traceability code. The acquisition steps are as follows: the system queries a pre-configured "terminal hash space upper limit mapping table" based on the "execution terminal code" involved in the process of generating the d-th segment of data. The table specifies a fixed positive integer as its path hash space upper limit for each terminal code or a group of terminal codes. The setting of this value needs to balance the uniqueness of the traceability code and the storage space. For example, for ordinary analysis terminals, a smaller upper limit such as 2 can be set. 16 =65536. For terminals with high security levels or requiring more detailed tracing, a larger upper limit can be set, such as 2 32 , here we set a general value, such as Λ d =1000000;

[0127] The query operation traceability code for a specific data segment calculated based on the previous formula, such as Z d=164, the system will use this traceability code as a key metadata and attach it to the preliminary privacy-preserving data to which the data segment belongs. Specifically, in the structured representation of the preliminary privacy-preserving data (for example, if the data is in JSON format, a new key-value pair is added; if it is XML, a new tag element is added; if it is in binary format, it is filled in the reserved header area), the calculated query operation traceability code is explicitly embedded as a "header identification field". For example, it can be named query_trace_code and its value is 164. Subsequently, the system will encapsulate the data segment containing the original (but anonymized) data content and the newly attached query operation traceability code in a unified, pre-defined data structure format to form a logically independent, self-contained "sealed data unit". The encapsulation format also includes other metadata, such as data generation time, data type identifier, encryption information (if applicable), etc. Finally, in order to facilitate management and quick retrieval, the system uniformly marks each sealed data unit generated in this way with a unique internal number (for example, a globally incremented serial number or a UUID), and records the key index information of the original query path corresponding to the data unit (for example, the top-level query ID directly related to the generation of the data segment or several keywords that best represent its content characteristics). This information will be used to quickly locate the relevant secure data circulation certificate through the query path when needed, thereby generating the final secure data circulation certificate, which is a sealed data unit containing anonymized data, traceability code and necessary management information, which can be shared or exchanged under the premise of meeting security and privacy requirements.

Claims

1. An intelligent connected vehicle data processing system based on data elements and knowledge graph, characterized in that: The system comprises: The data element definition module analyzes the original intelligent connected vehicle data elements, distinguishes them into vehicle static parameter sets and vehicle dynamic parameter sets, and forms a structured graph blueprint; A graph layered storage module, based on the structured graph blueprint, assigns entities of the vehicle static parameter set to the static storage layer and entities of the vehicle dynamic parameter set to the dynamic storage layer to obtain a logical layered data configuration. Based on the logical layered data configuration, the module applies geographic grid coding to spatially segment entities including location attributes to establish an optimized data storage layout. A multidimensional index optimization module, based on the optimized data storage layout, defines an access path for entities with spatiotemporal attributes by combining R-tree variant parameters and B+ tree parameters, generates spatiotemporal index construction instructions, and constructs a graph traversal path based on the spatiotemporal index construction instructions by defining an SPO triple index structure and path index parameters to establish a query index system; The graph security publishing module applies marginal noise to the graph side information associated with the query based on the received data access instructions and the query index system to obtain preliminary privacy-protected data. Based on the preliminary privacy-protected data, it attaches a query operation traceability code to the anonymized data segment to generate a secure data circulation certificate.

2. The intelligent connected vehicle data processing system based on data elements and knowledge graph according to claim 1 is characterized in that: The steps for obtaining the structured graph blueprint are: Read the data stream of the original intelligent connected vehicle data elements one by one, extract the timestamp information of each data element, count the number of times each data element changes within a predetermined interval, determine the data elements that exceed the high-frequency threshold as vehicle dynamic parameters, and determine the data elements that do not reach the high-frequency threshold as vehicle static parameters, and generate a vehicle static parameter set and a vehicle dynamic parameter set; Based on the vehicle static parameter set and the vehicle dynamic parameter set, the data information of each entity in each parameter set is parsed one by one, the attributes of the entities in the vehicle static parameter set are labeled and classified according to inherent attributes, vehicle model attributes, and production batch attributes, and the attributes of the entities in the vehicle dynamic parameter set are classified according to real-time attribute labels based on real-time displacement, instantaneous speed, real-time position, and real-time temperature, to generate specific attribute labels for each entity; Based on the specific attribute labels of each entity, a logical matching relationship is established for the entities in the vehicle static parameter set according to the consistency association rules of vehicle model and production batch, and a logical matching relationship is established for the entities in the vehicle dynamic parameter set according to the spatiotemporal synchronization association rules. In addition, functional dependency rules across attribute labels are established between static entities and dynamic entities to form a structured graph blueprint.

3. The intelligent connected vehicle data processing system based on data elements and knowledge graph according to claim 1 is characterized in that: The steps for obtaining the logical hierarchical data configuration are: Based on the structured graph blueprint, the entity nodes of the vehicle static parameter set and the entity nodes of the vehicle dynamic parameter set in the structured graph blueprint are parsed item by item, the entity nodes of the vehicle static parameter set with inherent attributes, vehicle model attributes, and production batch attributes are marked one by one, and the marked entity nodes are assigned in batches to the static storage layer to generate a vehicle static parameter storage mapping relationship; Based on the structured graph blueprint, the entity nodes of the vehicle dynamic parameter set are retrieved one by one, and the entity nodes of the real-time displacement, instantaneous speed, real-time position and real-time temperature attributes contained in the vehicle dynamic parameter set are extracted. The entity nodes of the real-time attributes are marked one by one and then assigned to the dynamic storage layer in batches to generate a vehicle dynamic parameter storage mapping relationship; Based on the vehicle dynamic parameter storage mapping relationship, the real-time data stream corresponding to the entity node of each real-time attribute is obtained, a time window benchmark of a fixed length is set, and the real-time data stream is segmented window by window according to the set time window benchmark, and the entity node data fragments of each real-time attribute within the time window benchmark are segmented to generate a logical hierarchical data configuration.

4. The intelligent connected vehicle data processing system based on data elements and knowledge graph according to claim 1, characterized in that: The steps for obtaining the optimized data storage layout are: Scanning each entity node included in the logical hierarchical data configuration one by one, extracting entity nodes with location attribute identifiers, parsing longitude and latitude values ​​in the location attributes, separately marking the entity nodes containing the longitude and latitude values, and generating a set of entity nodes with location attributes; According to the set of entity nodes with location attributes, bidirectional coordinate encoding is performed on the longitude and latitude values ​​of each entity node, and the longitude and latitude values ​​are respectively mapped to corresponding two-dimensional integer grid numbers. In combination with the preset geographic grid division accuracy, each entity node is classified into a corresponding geographic grid code area to generate an entity space segment corresponding to the geographic grid code; According to the physical space segments corresponding to the geographic grid codes, the space segments are numbered and classified in the order of the grid codes, and the classified space segments are submitted to the corresponding storage locations in the preset storage unit in sequence according to the grid numbers to establish an optimized data storage layout.

5. The intelligent connected vehicle data processing system based on data elements and knowledge graph according to claim 1 is characterized in that: The steps for obtaining the spatiotemporal index construction instruction are: Sequentially searching all entity nodes submitted to a preset storage unit in the optimized data storage layout, screening entity nodes containing both longitude and latitude values ​​and timestamp information, extracting spatial position coordinates and time tags of such entity nodes, and generating a set of entity nodes with spatiotemporal attributes; Extracting the longitude, latitude, and timestamp fields of each entity node from the set of entity nodes having spatiotemporal attributes, combining the longitude and latitude values ​​as two-dimensional spatial coordinates, and using the timestamp field as an independent time coordinate to construct a three-dimensional coordinate structure, respectively setting spatial partition dimension parameters of an R-tree variant and time series index parameters of a B+ tree according to the three-dimensional coordinate structure, and generating an entity access path jointly defined by the R-tree variant parameters and the B+ tree parameters; According to the entity access path jointly defined by the R-tree variant parameters and the B+ tree parameters, the longitude value, latitude value and timestamp in each set of three-dimensional coordinate structures are extracted, and the three-dimensional coordinate structure is converted into a continuous multidimensional coordinate set through sequential traversal. The multidimensional coordinate set is encoded and mapped according to the preset space-time filling curve identification rules to form a unique one-dimensional key value set, and the one-dimensional key value set is combined with the R-tree variant parameters and B+ tree parameter information to generate a space-time index construction instruction.

6. The intelligent connected vehicle data processing system based on data elements and knowledge graph according to claim 1 is characterized in that: The steps for obtaining the query index system are: Extract the entity one-dimensional key value, R-tree variant parameter, and B+ tree parameter information carried in the spatiotemporal index construction instruction one by one, restore the entity one-dimensional key value to the original three-dimensional coordinate structure, parse the entity identification information associated with the three-dimensional coordinate structure, and use the entity identification as the primary key to extract the attribute field corresponding to each entity to generate an entity attribute set; Based on the entity attribute set, field mapping is performed for each set of entity attribute information according to the subject-predicate-object semantic logic, mapping the entity identifier to the subject field, mapping the attribute label to the predicate field, and mapping the attribute value to the object field. The subject field, predicate field, and object field are combined to construct a corresponding SPO triple. The triple index position parameter is defined in combination with the storage path of the one-dimensional key value to generate an SPO triple index structure. According to the SPO triple index structure, the subject field, predicate field and object field index position parameters in the triple index structure are read, the entity node connection information recorded in the index position parameters is extracted, the relationship paths between all SPO triples are combined and arranged, the connection order and traversal route of each entity node in the graph are constructed, the graph traversal path is generated, and a query index system is established according to the graph traversal path.

7. The intelligent connected vehicle data processing system based on data elements and knowledge graph according to claim 1 is characterized in that: The steps for obtaining the preliminary privacy protection data are: Based on the received data access instructions and the query index system, a graph traversal is performed on each access path to locate all associated graph edges and the start and target nodes in the query path, and a set of original fields with location attribute fields, identity attribute fields, and physiological parameter attribute fields in the start and target nodes are extracted. Fields with a numeric data type and contained in both entities are screened, and the original values ​​in the start and target nodes are recorded to form a set of common numeric sensitive attribute field pairs of the graph edges; Calculate the marginal perturbation score of each graph edge according to the set of common numerical sensitive attribute field pairs of the graph edge; According to the marginal perturbation score of each graph edge, determine whether the marginal perturbation score is higher than the set perturbation judgment threshold. For all graph edges with marginal perturbation scores higher than the perturbation judgment threshold, generate a pseudo-random offset sequence with a length equal to the normalized value of the marginal perturbation score, and inject the pseudo-random offset sequence into the structure field of the graph edge. Search the corresponding attribute value generalization level for the position attribute field, identity attribute field and physiological parameter attribute field in the starting node and target node connected by the graph edge respectively, replace the field value according to the mapping rule from the original value to the previous level fuzzy interval, merge the updated graph edge structure and node attribute field, and generate preliminary privacy protection data.

8. The intelligent connected vehicle data processing system based on data elements and knowledge graph according to claim 1 is characterized in that: The steps for obtaining the secure data circulation certificate are as follows: Based on the preliminary privacy-preserving data, locate the source path of each anonymized data segment in the query index system, extract all query operation records associated with the source path, record the execution time, execution terminal code, and path depth level of each query operation, and integrate them to generate a query operation feature parameter set; Calculate the query operation traceability code for each segment of data based on the query operation characteristic parameter set; Based on the query operation traceability code, the query operation traceability code is attached to the preliminary privacy protection data as a header identification field, the data segment and the query operation traceability code are encapsulated into an independent sealed data unit according to a unified structural format, and the tag number and query path index information are unified to generate a secure data circulation certificate.

Images