Data security management system and method based on cloud computing

Abstract

The invention relates to the technical field of data mining, in particular to a data security management system and method based on cloud computing. The method comprises the following steps: acquiring real-time vehicle driving data, and performing homomorphic modular exponentiation dangerous case sensing fragmentation encryption to obtain dangerous case sensing encrypted fragmentation data; acquiring an Internet of Vehicles data stream, and implementing zero-trust adaptive data stream encryption to obtain a zero-trust authentication credential; implementing federal learning privacy feature mapping on the dangerous case sensing encrypted fragmented data to obtain an anonymized encrypted feature mapping model; performing Merkle tree aggregation on the anonymized encrypted feature mapping model to obtain a block chain global dangerous case fingerprint; extracting a threat behavior quantization vector according to the block chain global dangerous case fingerprint and the anonymization encryption feature mapping model; and carrying out transmission strategy optimization according to the threat behavior quantization vector and the zero-trust authentication credential to obtain a global dynamic transmission strategy matrix. According to the invention, the safety of vehicle data uploading and the overall uploading efficiency can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data mining, and in particular to a data security management system and method based on cloud computing. Background Art

[0002] With the rapid development of automobile intelligence and networking, vehicle systems are gradually integrated into cloud computing platforms, forming a three-dimensional intelligent transportation system of "vehicle-road-cloud". In this system, cars not only rely on local hardware and embedded control systems for driving control, but also exchange data and collaborate with cloud platforms through the vehicle-mounted Internet to achieve real-time road condition monitoring, intelligent navigation, vehicle remote diagnosis and other functions. However, with the popularization of the Internet of Vehicles and the widespread application of cloud computing technology, the large amount of data generated during the driving of cars faces many security risks in the transmission, storage and processing process. First of all, the biggest defect of traditional methods lies in the limitations of data storage and processing. Traditional automobile driving data security management mostly relies on vehicle-mounted equipment and local computing, which makes it difficult to achieve real-time large-scale data sharing and analysis. With the improvement of the level of vehicle intelligence, the amount of data generated by vehicles has shown explosive growth, and the storage and processing requirements of these data far exceed the carrying capacity of traditional methods. In the cloud computing environment, the data generated by cars often needs to be uploaded to the cloud for centralized storage and analysis, but traditional local data security measures are difficult to cope with the security threats of cross-regional and cross-platform data transmission.

[0003] Traditional access control and identity authentication mechanisms cannot meet the needs of intelligent transportation systems in cloud computing environments. In cloud computing platforms, car data is not only generated by the vehicle itself, but also involves multiple collaborative operating entities, such as traffic management systems, remote maintenance service providers, and third-party navigation applications. The identity authentication and authorization management of these multiple visitors is more difficult, and traditional access control methods cannot effectively manage the huge and changeable user identity information, resulting in potential risks of abuse of authority and information leakage. Summary of the invention

[0004] Based on this, it is necessary for the present invention to provide a data security management system and method based on cloud computing to solve at least one of the above technical problems.

[0005] To achieve the above purpose, a data security management method based on cloud computing includes the following steps:

[0006] Step S1: acquiring real-time vehicle driving data, and performing attribute sensitivity boundary classification to obtain sensitivity classified driving data; performing homomorphic modular power hazard-aware shard encryption on the sensitivity classified driving data to obtain hazard-aware encrypted shard data;

[0007] Step S2: Obtain the Internet of Vehicles data stream and build a zero-trust two-way authentication mechanism; use the zero-trust two-way authentication mechanism to implement zero-trust adaptive data stream encryption, obtain zero-trust authentication credentials, and upload them to the Internet of Vehicles for credential storage;

[0008] Step S3: Perform secure multi-party de-identification processing on the risk-aware encrypted shard data to obtain de-identified encrypted shard data; perform federated learning privacy feature mapping on the de-identified encrypted shard data to obtain an anonymized encrypted feature mapping model;

[0009] Step S4: Perform Merkle tree hierarchical hash aggregation on the anonymized encrypted feature mapping model to obtain the global blockchain risk fingerprint; construct a dynamic trust quantification matrix based on the blockchain global risk fingerprint and the anonymized encrypted feature mapping model to extract the threat behavior quantification vector;

[0010] Step S5: Perform gradient optimization of the dangerous situation data transmission strategy based on the threat behavior quantification vector and zero-trust authentication credentials to obtain a global dynamic transmission strategy matrix, and deploy it to the vehicle communication unit in real time.

[0011] The present invention effectively ensures the real-time, confidential and reliable uploading of emergency information through precise multiple encryption and security strategies. When a vehicle encounters a dangerous situation during driving, the vehicle driving data obtained in real time is first classified by sensitivity, and the data is homomorphically encrypted, thereby ensuring that key data (such as emergency braking signals, airbag triggering records, etc.) are not interfered with or tampered with by the outside world during transmission, avoiding the risk of sensitive data leakage and tampering. Based on the zero-trust two-way authentication mechanism, the Internet of Vehicles data stream is authenticated to ensure the security of information transmission. By implementing zero-trust adaptive data encryption, the vehicle-mounted system can dynamically adjust the encryption strategy in real time, give priority to the encryption and transmission of highly sensitive data for different emergency data streams, and effectively prevent the data from being maliciously tampered with or leaked during transmission. Using de-identification and federated learning privacy feature mapping technology, the transmitted emergency data can be intelligently analyzed on the premise of ensuring that the privacy data is not leaked, providing support for subsequent real-time response and decision-making. In addition, blockchain technology generates a global danger fingerprint to ensure that the authenticity and integrity of emergency information can be traced and verified after uploading, effectively avoiding the possibility of data tampering or forgery. Ultimately, based on the dynamic trust quantification matrix and transmission strategy optimization, the transmission strategy of emergency information can be dynamically adjusted to ensure that emergency data can be transmitted to the cloud or related monitoring platforms in the shortest time and in the safest way, thereby providing timely and accurate information support for emergency response. This technical system not only improves the reliability and security of emergency information upload, but also greatly improves the emergency response capability of the traffic emergency system in the face of emergencies, and ensures the safety of vehicles and other entities in the traffic system.

[0012] Optionally, the attribute sensitivity boundary classification in step S1 is specifically:

[0013] Acquire real-time vehicle sensor data through the vehicle-mounted sensor group, and perform multi-modal sensor data fusion on the real-time vehicle sensor data to obtain real-time vehicle driving data;

[0014] Perform vehicle driving pattern recognition on real-time vehicle driving data to obtain a driving behavior semantic label dataset;

[0015] Segment driving events according to the driving behavior semantic label dataset, extract discrete driving events, and obtain discrete driving event data;

[0016] Conduct vehicle driving risk assessment on discrete driving event data, quantify driving collision probability, and obtain a real-time driving risk quantification map;

[0017] Based on the real-time driving risk quantification map, sensitivity is quantified, sensitivity classification thresholds are dynamically calculated, and an adaptive sensitivity classification matrix is ​​obtained;

[0018] A sensitivity label is assigned to the vehicle driving mode data according to the adaptive sensitivity classification matrix to obtain sensitivity classification driving data, wherein the sensitivity classification driving data includes high-sensitivity driving data and low-sensitivity driving data.

[0019] The present invention obtains real-time vehicle sensor data through a vehicle-mounted sensor group, and fuses these multimodal data to ensure that the various states of vehicle driving can be accurately perceived and the information is updated in real time. These fused data provide sufficient data support for the pattern recognition of subsequent driving behaviors, can accurately identify different driving behaviors, and generate semantic labels based on the behaviors, thereby constructing a data set containing multiple driving scenarios. Next, by segmenting driving events and extracting discrete events from these data, the system can identify and clearly distinguish potential dangerous situations, which lays the foundation for the assessment of driving risks. By quantifying the risk of discrete driving event data, the system can evaluate the collision probability in each driving scenario in real time and generate a corresponding driving risk quantification map. This map not only reflects the current driving risk status, but also provides a basis for the vehicle to determine whether emergency response measures need to be taken immediately. Based on these real-time risk information, the system dynamically adjusts the risk data through a sensitivity quantification mechanism and calculates a sensitivity classification threshold to ensure that when a dangerous situation occurs, the system can respond in a timely manner to different risk levels. Finally, according to the adaptive sensitivity classification matrix, the driving data is assigned sensitivity labels, effectively dividing high-sensitivity and low-sensitivity data, ensuring that emergency data can be processed and transmitted first, especially those involving major safety threats, such as emergency braking or pre-collision warning signals. This process ensures the rapid and accurate uploading of emergency information in dangerous situations, while optimizing the data transmission process, ensuring that car owners and traffic management systems can receive key safety information in the shortest time, greatly improving the safety of vehicles and road users.

[0020] Optionally, the homomorphic modular exponentiation hazard-aware shard encryption described in step S1 is specifically:

[0021] Classify the vehicle danger level according to the real-time driving risk quantification map to obtain a vehicle danger level classification table;

[0022] Based on the vehicle danger level classification table and the adaptive sensitivity classification matrix, a danger-sharding dimension mapping rule is constructed, and the modular exponentiation parameters are initialized by homomorphic encryption to obtain a dynamic sharding function.

[0023] The dynamic slicing function and the sensitivity-classified driving data are subjected to classified hazard-aware slicing, and the high-sensitivity driving data is subjected to modular exponentiation slicing to obtain high-sensitivity driving slicing data; the low-sensitivity driving data is subjected to fixed slicing to obtain low-sensitivity driving slicing data;

[0024] Merge the high-sensitivity driving segment data and the low-sensitivity driving segment data to obtain the hazard perception segment data set;

[0025] The hazard-aware sharded data set is subjected to Paillier homomorphic encryption operation, the ciphertext of each shard is calculated, and the hazard label is bound to the ciphertext while retaining the spatiotemporal correlation to generate hazard-aware encrypted sharded data.

[0026] The present invention uses a real-time driving risk quantification map to classify the danger of the vehicle, which can clearly identify and classify potential dangerous situations and ensure that dangerous situations receive priority attention. This classification provides an accurate basis for subsequent data processing and safety decisions. On this basis, combined with the adaptive sensitivity classification matrix, the system constructs a mapping rule of the danger-slicing dimension, and combines the dynamic slicing function to ensure that highly sensitive data can be more finely encrypted and protected during transmission, while reducing the processing complexity of low-sensitivity data and optimizing resource utilization efficiency. The implementation of this strategy ensures the security of sensitive data, especially when it comes to key safety data such as emergency braking and pre-collision warning, it can be encrypted through modular exponentiation slicing to reduce the risk of external attacks and data leakage. Low-sensitivity data is processed in a fixed slicing manner to achieve more efficient transmission. By merging high-sensitivity driving slicing data and low-sensitivity driving slicing data, the system can integrate data of different sensitivities while maintaining its independence and security to ensure that the integrity of the information is not destroyed. Finally, the hazard-aware shard data is encrypted through Paillier homomorphic encryption operations, ciphertext is generated, and the hazard label is bound to the ciphertext, which not only ensures the encryption security of the data, but also retains the temporal and spatial correlation, thereby ensuring that the information can be effectively verified during the transmission process and supporting subsequent real-time security response. This processing flow ensures that the sensitive data generated by the vehicle in an emergency can be protected and transmitted in real time. At the same time, through efficient encryption and sensitivity management, data security is improved, the risk of data leakage and abuse is avoided, and the safety of vehicles and road users is ensured.

[0027] Optionally, step S2 specifically includes:

[0028] Step S21: obtaining the Internet of Vehicles data stream through the vehicle communication unit, and performing data preprocessing on the Internet of Vehicles data stream to obtain the Internet of Vehicles data stream to be analyzed;

[0029] Step S22: extracting IoV node features from the IoV data stream to be analyzed, and obtaining connected vehicle data, roadside unit data, and cloud platform node identity data;

[0030] Step S23: construct a zero-trust two-way authentication mechanism based on the connected vehicle data, the roadside unit data, and the cloud platform node identity data;

[0031] Step S24: allocating data sensitivity to the Internet of Vehicles data stream based on the sensitivity-classified driving data to obtain a sensitivity-allocated Internet of Vehicles data stream;

[0032] Step S25: Utilize the zero-trust two-way authentication mechanism to implement adaptive encryption on the sensitivity-assigned Internet of Vehicles data stream, obtain zero-trust authentication credentials, and upload them to the Internet of Vehicles for credential storage.

[0033] By acquiring the Internet of Vehicles data stream and performing data preprocessing, the system can provide high-quality data input for subsequent analysis, ensuring the integrity and accuracy of the data. Furthermore, by extracting the characteristics of the Internet of Vehicles nodes, the system can distinguish the identity information of the vehicle, the roadside unit and the cloud platform, and analyze and verify it, which helps to ensure that the source of each data packet is credible and provides a basis for subsequent authentication. In this process, combined with the zero-trust two-way authentication mechanism, not only the communication between the vehicle and the cloud platform is strictly verified, but also the confidentiality of the data can be ensured in real time during the transmission process, preventing external attacks or unauthorized access, and ensuring the safe transmission of sensitive information. By allocating the sensitivity of the Internet of Vehicles data stream, the corresponding encryption strategy can be selected according to the importance and urgency of the data, giving priority to protecting highly sensitive data (such as collision warnings, emergency brake signals, etc.), without wasting too much computing resources on low-sensitivity data. Finally, through the combination of adaptive encryption and zero-trust authentication mechanisms, it is ensured that each piece of emergency data in the Internet of Vehicles can obtain the best encryption protection and authentication verification, while ensuring security, it can also efficiently transmit and store data.

[0034] Optionally, step S23 is specifically:

[0035] Step S231: extract identity features from the connected vehicle data, roadside unit data, and cloud platform node identity data, construct an interactive identity authentication data set, and set an identity trust score threshold to screen highly trusted nodes, thereby generating an initial identity feature matrix;

[0036] Step S232: Based on the initial identity feature matrix, elliptic curve signature bidirectional identity authentication is performed on the connected vehicle data and the roadside unit data, and a timestamp synchronization deviation threshold Δt=50 is set for timeliness verification to generate a dynamic identity authentication parameter set;

[0037] Step S233: Perform variable challenge response authentication on the dynamic identity authentication parameter set, calculate the device fingerprint matching degree, set the matching threshold Tf=0.85 to filter abnormal nodes, and generate a device fingerprint authentication matrix;

[0038] Step S234: training a dynamic trust score model in combination with the interactive identity authentication data set and the device fingerprint authentication matrix, and setting the trust score update step length η=0.05 to calculate the trust weight of the vehicle-cloud platform-roadside unit, and generating a dynamic trust weight matrix;

[0039] Step S235: Couple the zero-knowledge proof verifiable mechanism according to the dynamic trust weight matrix, perform two-way identity confirmation, set the zero-knowledge proof challenge round Nc∈[5,10], and derive the identity credibility based on the identity credibility threshold Ttrust=0.9, thereby generating a zero-trust two-way authentication mechanism.

[0040] The present invention can accurately distinguish each participant, identify their trust, and screen out highly trusted nodes according to the set threshold by extracting the identity features of the connected vehicles, roadside units and cloud platform nodes, which ensures the legitimacy of each participating node and avoids the interference of malicious nodes on the system. Through the introduction of a two-way identity authentication mechanism, the communication between vehicles, roadside units and cloud platforms is strictly verified, and timeliness verification is also performed to ensure that all data uploaded in an emergency can instantly and truly reflect the current dangerous situation. In addition, through device fingerprint matching and abnormal node filtering, the system can accurately identify illegal devices and eliminate potential security threats, further improving the protection capability of the system. On the basis of building a dynamic trust scoring model, the system updates the trust weight in real time, and combines the zero-knowledge proof verifiable mechanism to achieve two-way identity confirmation, avoiding the problem of identity forgery or tampering in data transmission. In dangerous situations, this multiple verification and identity confirmation mechanism can ensure that the transmission and storage of emergency information have extremely high security and accuracy, effectively support the security collaboration between vehicles and cloud platforms and roadside units, and improve the overall security and emergency response capabilities of the Internet of Vehicles.

[0041] Optionally, the federated learning privacy feature mapping described in step S3 is specifically:

[0042] Perform Laplace noise local privacy protection on the de-identified hazard segmented dataset, set the privacy budget Q∈[1,10], and calculate the feature perturbation matrix based on the preset feature dimension d=128;

[0043] The feature perturbation matrix is ​​normalized, and the federated feature transformation matrix is ​​constructed according to the federated learning privacy feature mapping parameters α=0.7, β=0.3, γ=0.9 to obtain the privacy-preserving feature vector;

[0044] Map the privacy-preserving feature vector to a preset prime modulus p=2 2048 Encrypt the computational domain and set the public key parameter g = 2 512 , private key parameter λ = 2 1024 Perform homomorphic encryption transformation to generate an encrypted feature vector set;

[0045] Perform distributed gradient calculation on the encrypted feature vector set, set the training batch B∈[64,128], and use the Adam optimizer with η=0.001, β1=0.9, β2=0.999 to update the gradient, and obtain the privacy-preserving feature gradient matrix;

[0046] Set the noise standard deviation σ∈[e,2e] and adopt the clipping norm Δ∈[1,2] to perform differential privacy stochastic gradient descent on the privacy-preserving feature gradient matrix, thereby constructing a privacy-preserving hazard feature mapping model;

[0047] The privacy-preserving hazard feature mapping model is updated with a zero-knowledge proof model to obtain an anonymized encrypted feature mapping model.

[0048] The present invention introduces Laplace noise to ensure that when processing de-identified dangerous data sets, privacy protection and data practicality can be balanced, and privacy leakage can be prevented while ensuring the validity of the data. This privacy budget-based strategy can dynamically adjust the privacy protection intensity so that the privacy protection requirements in different scenarios are reasonably met. Through the introduction of federated learning privacy feature mapping, multiple participating nodes can be jointly modeled without exposing specific data, which not only protects the data privacy of individual vehicles, but also promotes data collaboration of the entire Internet of Vehicles system. The encryption of privacy-protected feature vectors and the use of homomorphic encryption computing domains make it impossible for malicious third parties to access specific vehicle data even during data transmission, thereby ensuring the confidentiality of the data. Through distributed gradient computing and differential privacy stochastic gradient descent technology, the model training process is further optimized, while privacy leakage is prevented, and privacy protection in data processing and computing is guaranteed. In addition, through the update of the zero-knowledge proof model, data anonymization and verification are achieved, ensuring that the dangerous information uploaded in an emergency situation not only has a high level of privacy protection, but also has high reliability and verifiability, and can support subsequent analysis and emergency response.

[0049] Optionally, step S4 is specifically:

[0050] Step S41: Perform encryption feature hashing according to the anonymized encryption feature mapping model, and construct a Merkle tree hierarchical hash structure to generate a preliminary blockchain risk fingerprint;

[0051] Step S42: Perform breadth-first recursive verification on the preliminary blockchain risk fingerprint to generate an intermediate blockchain risk fingerprint;

[0052] Step S43: Perform hierarchical verification on the intermediate blockchain risk fingerprint according to the Merkle tree hierarchical hash structure to obtain the blockchain global risk fingerprint;

[0053] Step S44: combining the global risk fingerprint of blockchain and the anonymized encrypted feature mapping model, performing data fusion based on the dynamic trust scoring model to obtain a dynamic trust quantification matrix;

[0054] Step S45: quantify the threat behavior on the dynamic trust quantization matrix, set the threat behavior intensity threshold, and thus generate a threat behavior quantization vector.

[0055] The present invention performs encrypted feature hashing on the anonymized encrypted feature mapping model and constructs a Merkle tree hierarchical hash structure, which can efficiently generate a preliminary blockchain danger fingerprint to ensure that each piece of data will not be tampered with or forged when it is safely uploaded. Then, the validity of the intermediate blockchain danger fingerprint is ensured by using breadth-first recursive verification and hierarchical level verification, and the credibility of the data is further enhanced through the gradual blockchain verification process. These measures enable each piece of dangerous information to be strictly verified and tracked even in a complex Internet of Vehicles environment, thereby preventing any tampering or data loss. Next, data fusion based on a dynamic trust scoring model can quantify the trust relationship between each node in the Internet of Vehicles, further improving the credibility and priority judgment of the system for danger data. Finally, by quantifying threat behaviors and setting threat behavior intensity thresholds, not only can potential security threats be identified in real time, but also emergency response strategies can be adjusted in real time according to the quantified vectors, thereby improving the overall system's responsiveness in emergency situations.

[0056] Optionally, step S44 is specifically:

[0057] Step S441: perform feature alignment on the blockchain global risk fingerprint and the anonymized encrypted feature mapping model, set the alignment error tolerance threshold ε=0.05, and thus generate a feature alignment data set;

[0058] Step S442: Based on the feature alignment data set, the global blockchain risk fingerprint is fused with the anonymized encrypted feature mapping model, and the weight parameters α=0.7 and β=0.3 are set to generate a preliminary trust data set;

[0059] Step S443: standardize the preliminary trust data set and set the standardization range to [0,1], thereby generating a standardized trust data set;

[0060] Step S444: Based on the standardized trust data set, a dynamic trust scoring model is used for calculation, and the update step length η is set to 0.05, thereby generating a dynamic trust scoring matrix;

[0061] Step S445: According to the dynamic trust score matrix, a trust score threshold Ttrust=0.85 is set, nodes with trust scores lower than the threshold are removed, and a dynamic trust quantization matrix is ​​generated.

[0062] The present invention ensures accurate matching between different data sources by aligning the features of the global danger fingerprint and the anonymized encrypted feature mapping model of the blockchain, thereby effectively reducing the data alignment error and generating an accurate feature alignment data set. This process ensures that the data from different nodes can maintain consistency during transmission and fusion, and avoids potential errors caused by data misalignment. Next, by weighted fusion of the aligned data, combined with the set weight parameters, a preliminary trust data set is generated. This process ensures that the data from different sources can be reasonably fused according to their importance, ensuring that the most trusted nodes have the greatest impact on the data. After that, the standardization process makes the trust data within a unified range, providing a more stable and standardized basis for subsequent trust scoring. The calculation of the dynamic trust scoring model can adjust the trust of each node in real time, and dynamically update the system's trust in each node according to the changes in the trust scoring matrix, further improving the system's responsiveness and accuracy. Finally, the low-trust nodes are eliminated by the set trust scoring threshold, avoiding interference from unreliable nodes in the transmission of emergency data, ensuring that the dynamic trust quantization matrix finally generated accurately reflects the trustworthiness of each node in the system, and providing a solid foundation for subsequent security decisions and emergency responses.

[0063] Optionally, step S5 specifically includes:

[0064] Step S51: extracting data transmission features from the vehicle network data stream to be analyzed to obtain vehicle network transmission feature data;

[0065] Step S52: performing transmission strategy gradient optimization based on vehicle network transmission characteristic data to obtain a preliminary transmission strategy matrix;

[0066] Step S53: dividing the data transmission priority of the preliminary transmission strategy matrix according to the dynamic trust quantization matrix, and performing regularization processing to obtain a regularized transmission strategy matrix;

[0067] Step S54: Adaptively adjust the regularized transmission strategy matrix dynamically based on the zero-trust authentication credentials to obtain a global dynamic transmission strategy matrix, and deploy it to the vehicle communication unit in real time.

[0068] The present invention extracts transmission feature data, which provides an accurate basis for subsequent optimization and ensures the capture and processing of key information during the transmission process. Based on these feature data, a preliminary transmission strategy matrix is ​​obtained through transmission strategy gradient optimization, thereby setting a preliminary plan for subsequent transmission optimization, so that the data transmission process can be adaptively adjusted according to the actual situation to ensure the efficiency and security of data transmission. Next, the transmission strategy matrix is ​​prioritized and regularized according to the dynamic trust quantization matrix. This step reasonably adjusts the transmission priority of emergency information by combining the trust status of each node in the system, ensuring that important data can be transmitted in time, and avoiding low-trust nodes from affecting the reliability of data. Subsequently, the transmission strategy is dynamically adjusted by combining zero-trust authentication credentials, further strengthening the security of the transmission process, ensuring that only authenticated nodes can participate in information transmission, thereby preventing interference from malicious nodes. Finally, after these steps, the generated global dynamic transmission strategy matrix is ​​deployed to the vehicle-mounted communication unit in real time, ensuring that when encountering dangerous situations, the vehicle-mounted system can quickly and effectively upload emergency information, and ensure the security and transmission efficiency of data.

[0069] Optionally, the present specification also provides a cloud computing-based data security management system, which is used to execute the cloud computing-based data security management method as described above, and the cloud computing-based data security management system includes:

[0070] The hazard awareness sharding module is used to obtain real-time vehicle driving data and perform attribute sensitivity boundary classification to obtain sensitivity classified driving data; the sensitivity classified driving data is encrypted by homomorphic modular power hazard awareness sharding to obtain hazard awareness encrypted sharding data;

[0071] The two-way authentication encryption module is used to obtain the Internet of Vehicles data stream and build a zero-trust two-way authentication mechanism; the zero-trust two-way authentication mechanism is used to implement zero-trust adaptive data stream encryption, obtain zero-trust authentication credentials, and upload them to the Internet of Vehicles for credential storage;

[0072] The multi-party de-identification module is used to perform secure multi-party de-identification processing on the risk-aware encrypted shard data to obtain de-identified encrypted shard data; implement federated learning privacy feature mapping on the de-identified encrypted shard data to obtain an anonymized encrypted feature mapping model;

[0073] The threat behavior quantification module is used to perform Merkle tree hierarchical hash aggregation on the anonymized encrypted feature mapping model to obtain the global blockchain risk fingerprint; a dynamic trust quantification matrix is ​​constructed based on the blockchain global risk fingerprint and the anonymized encrypted feature mapping model to extract the threat behavior quantification vector;

[0074] The data transmission strategy analysis module is used to optimize the dangerous situation data transmission strategy gradient according to the threat behavior quantification vector and zero-trust authentication credentials, obtain the global dynamic transmission strategy matrix, and deploy it to the vehicle communication unit in real time.

[0075] The data security management system based on cloud computing of the present invention can realize any data security management method based on cloud computing of the present invention, and is used to combine the operation and signal transmission medium between various modules to complete the data security management method based on cloud computing. The internal modules of the system cooperate with each other, thereby improving the reliability and security of emergency information upload. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0077] Figure 1 A schematic diagram of the steps of the cloud computing-based data security management method of the present invention;

[0078] Figure 2 Detailed step flow diagram of step S2 in the present invention;

[0079] Figure 3 Detailed step flow diagram of step S4 in the present invention;

[0080] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0081] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0082] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0083] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0084] To achieve this, please refer to Figures 1 to 3 The present invention provides a data security management method based on cloud computing, the method comprising the following steps:

[0085] Step S1: acquiring real-time vehicle driving data, and performing attribute sensitivity boundary classification to obtain sensitivity classified driving data; performing homomorphic modular power hazard-aware shard encryption on the sensitivity classified driving data to obtain hazard-aware encrypted shard data;

[0086] In this embodiment, vehicle driving data, including but not limited to speed, acceleration, position, steering angle, and road conditions, are collected in real time through on-board sensors (such as radar, camera, inertial navigation, etc.). According to the collected data, the data is classified by setting the attribute sensitivity boundary (for example, setting collision risk data, lane departure data, etc. as high sensitivity) to obtain high-sensitivity and low-sensitivity classified data. Next, the homomorphic encryption technology is used to perform modular exponentiation sharding on the sensitive data, where the high-sensitivity data is encrypted by the modular exponentiation sharding algorithm (for example, modular exponentiation sharding with key parameter λ=3 and modulus p=5) to obtain hazard-aware encrypted sharding data. These encrypted sharding data will be processed according to the importance, timeliness, and sensitivity of the data to ensure the privacy and security of sensitive data during transmission and storage.

[0087] Step S2: Obtain the Internet of Vehicles data stream and build a zero-trust two-way authentication mechanism; use the zero-trust two-way authentication mechanism to implement zero-trust adaptive data stream encryption, obtain zero-trust authentication credentials, and upload them to the Internet of Vehicles for credential storage;

[0088] In this embodiment, the Internet of Vehicles data stream is obtained through the vehicle communication unit, including vehicle communication data, roadside unit data, and cloud platform node identity data. Through the Internet of Vehicles architecture based on the V2X protocol, a zero-trust two-way authentication mechanism is constructed, which verifies the legal identity of the vehicle, roadside unit (RSU) and cloud platform by using dynamic digital certificates (for example, the certificate expiration time is set to 6 hours) and a two-way mTLS encrypted tunnel. In this process, the vehicle and the cloud platform perform two-way identity authentication to ensure the security of vehicle-mounted communications. When implementing a zero-trust adaptive encryption mechanism, priority encryption is set for highly sensitive data in transmission (such as brake signals) to generate zero-trust authentication credentials. These credentials include encryption information, authentication identifiers, and valid timestamps, and are uploaded to the Internet of Vehicles system for credential storage to ensure the source and integrity of the data.

[0089] Step S3: Perform secure multi-party de-identification processing on the risk-aware encrypted shard data to obtain de-identified encrypted shard data; perform federated learning privacy feature mapping on the de-identified encrypted shard data to obtain an anonymized encrypted feature mapping model;

[0090] In this embodiment, the hazard-aware encrypted shard data is de-identified. During the data transmission process, each hazard-aware encrypted shard data is first subjected to secure multi-party computing processing to strip off the vehicle identity information and vehicle owner information, retaining only important hazard feature data, such as collision acceleration, steering angle changes, etc. On the basis of de-identification, the private data is converted using a federated learning privacy feature mapping algorithm (for example, setting feature mapping parameters α=0.7, β=0.3) to obtain an anonymized encrypted feature mapping model. After training, the model can generate risk predictions for specific driving behaviors (such as sudden braking, sharp turns, etc.) without exposing specific vehicle information, and make the data privacy-protected. Through this process, data sharing does not expose privacy information, thereby ensuring the data security of car owners and users.

[0091] Step S4: Perform Merkle tree hierarchical hash aggregation on the anonymized encrypted feature mapping model to obtain the global blockchain risk fingerprint; construct a dynamic trust quantification matrix based on the blockchain global risk fingerprint and the anonymized encrypted feature mapping model to extract the threat behavior quantification vector;

[0092] In this embodiment, the Merkle tree hierarchical hash aggregation is performed on the anonymized encrypted feature mapping model. A hierarchical hash structure is used (for example, the maximum depth of the tree is set to 4 levels), and the encrypted feature vector generated by the model is hashed multiple times to obtain the global blockchain danger fingerprint. The blockchain global danger fingerprint contains the hash value of each danger event, and data traceability and verification are performed based on these values. Then, the fingerprint data is combined with the anonymized encrypted feature mapping model to calculate the trust matrix based on the dynamic trust scoring model. In this process, a threat behavior quantification vector is set to quantify and calibrate different threat levels (such as high threat, low threat), and the final threat behavior quantification vector is generated by calculating its weighted score (for example, setting the threat intensity threshold to 0.75). The matrix can reflect the trust status of the Internet of Vehicles nodes in real time, further ensuring the security of the Internet of Vehicles system and the reliability of information.

[0093] Step S5: Perform gradient optimization of the dangerous situation data transmission strategy based on the threat behavior quantification vector and zero-trust authentication credentials to obtain a global dynamic transmission strategy matrix, and deploy it to the vehicle communication unit in real time.

[0094] In this embodiment, based on the previously generated threat behavior quantification vector and combined with zero-trust authentication credentials, the dangerous situation data transmission strategy is gradient optimized. By setting the gradient optimization parameters (such as learning rate η = 0.001, batch size B = 64), an adaptive optimization algorithm (such as Adam optimizer) is used to update the transmission strategy matrix. Specifically, for each type of dangerous situation data (such as vehicle emergency braking or rapid acceleration data), gradient calculation is performed according to its risk level and transmission priority, and the data transmission path is optimized to ensure that high-threat data can be transmitted and encrypted first. After the optimization is completed, the generated global dynamic transmission strategy matrix will be adjusted in real time according to the needs of the vehicle communication unit and deployed to the vehicle communication unit to ensure that the vehicle can maintain efficient and secure communication capabilities in the vehicle network environment when facing emergencies.

[0095] Optionally, the attribute sensitivity boundary classification in step S1 is specifically:

[0096] Acquire real-time vehicle sensor data through the vehicle-mounted sensor group, and perform multi-modal sensor data fusion on the real-time vehicle sensor data to obtain real-time vehicle driving data;

[0097] In this embodiment, the vehicle-mounted sensor group includes multiple modes, such as millimeter-wave radar, inertial measurement unit (IMU) and vehicle-mounted camera, and the data sampling frequency of each sensor is set to 10Hz. The vehicle's driving data includes speed, acceleration, lane departure, steering angle, etc. In order to achieve data fusion, the Kalman filter algorithm is used to synchronize the radar, IMU and camera data in time and space. The measurement error of the vehicle-mounted sensor is corrected by the Kalman filter, and the mutual information algorithm is used to jointly optimize the data of each modality to ensure data consistency and accuracy. The system will generate fused vehicle driving data in real time, including the vehicle's real-time position, speed, acceleration, obstacle distance, road condition information, etc., as input for subsequent driving behavior recognition and risk assessment.

[0098] Perform vehicle driving pattern recognition on real-time vehicle driving data to obtain a driving behavior semantic label dataset;

[0099] In this embodiment, a deep learning model is used to classify driving behavior. First, the real-time driving data of the vehicle is input into a convolutional neural network (CNN) for feature extraction. The input data includes sensor data such as acceleration, steering angle, vehicle speed, obstacle distance, etc. By training the existing historical driving data, the model learns the characteristics of driving behavior, such as sudden braking, sudden acceleration, constant speed driving, cornering and other modes. In order to improve the accuracy of classification, the model uses an LSTM (long short-term memory) network to further analyze the spatiotemporal association of time series data. The output of the model is a driving behavior semantic label data set, each label represents a driving mode (such as "sudden braking", "sudden acceleration"), and a timestamp is given to each driving behavior. The training of the model uses a label data set with driving scenes, and the accuracy is set to be more than 90%.

[0100] Segment driving events according to the driving behavior semantic label dataset, extract discrete driving events, and obtain discrete driving event data;

[0101] In this embodiment, based on the driving behavior semantic label dataset, the data is segmented into time windows (such as 5 seconds) to extract discrete driving events. 2 ), the driving data is distinguished according to the characteristics of the driving event. Each driving event will be marked separately, such as the acceleration of the emergency braking event is greater than 3m / s 2 When the model extracts this part of the data, it forms discrete driving event data. The data of each event includes the time of the event, duration and related dynamic data, such as vehicle speed, steering wheel angle, etc. Through this process, the system can clearly divide the driving behavior data into independent driving events, providing a clear basis for subsequent risk assessment and sensitivity classification.

[0102] Conduct vehicle driving risk assessment on discrete driving event data, quantify driving collision probability, and obtain a real-time driving risk quantification map;

[0103] In the present embodiment, based on discrete driving event data, a collision probability quantification model based on a multivariate regression model and a Bayesian network is constructed. The model uses a variety of factors that affect the probability of collision (such as vehicle speed, distance from the front vehicle, steering angle, driving behavior) as input variables, and uses linear regression or random forest regression methods to predict the collision risk of the current driving event. Through the training of a large amount of historical collision data, the model has learned the relationship between various driving events and the occurrence of collisions, and outputs a collision probability value (such as between 0-1). The collision probability assessment model sets a preliminary risk threshold (such as 0.5 represents medium risk) during training, and the evaluation result of the model is a real-time driving risk quantification map, which shows the risk level of different driving behaviors. When the model is running, combined with real-time data, the collision risk map is updated every 10ms, and real-time feedback is provided to the driver.

[0104] Based on the real-time driving risk quantification map, sensitivity is quantified, sensitivity classification thresholds are dynamically calculated, and an adaptive sensitivity classification matrix is ​​obtained;

[0105] In this embodiment, a dynamic sensitivity classification method is used to calculate the sensitivity threshold based on the real-time driving risk quantification map. First, based on historical data, the risk level of different driving events (such as sudden braking and sudden acceleration) is calculated, and the sensitivity threshold is dynamically adjusted according to the current driving conditions of the vehicle. For example, when the collision risk exceeds 80%, the data is marked as high-sensitivity data. Through the collision probability in the real-time driving risk quantification map, combined with factors such as the vehicle's current speed, distance, and steering, a sensitivity classification threshold that adapts to the current situation is dynamically calculated. This threshold is updated every 5 minutes to ensure that it adapts to different road conditions and driving scenarios. In this process, the computational complexity of the model is accelerated by the GPU to improve real-time responsiveness.

[0106] A sensitivity label is assigned to the vehicle driving mode data according to the adaptive sensitivity classification matrix to obtain sensitivity classification driving data, wherein the sensitivity classification driving data includes high-sensitivity driving data and low-sensitivity driving data.

[0107] In this embodiment, sensitivity labels are assigned to vehicle driving mode data based on the generated adaptive sensitivity classification matrix. For highly sensitive data, encryption protection measures will be adopted during the transmission process, while for low-sensitivity data, standard secure transmission methods can be used. Through this sensitivity classification, emergency and non-emergency data can be efficiently distinguished, ensuring the security and privacy protection of key data. For highly sensitive data (such as emergency braking or high-speed driving data), encryption measures will be used to ensure the safe transmission of data; while for low-sensitivity data (such as normal driving data), ordinary data transmission methods are used. For example, highly sensitive data (such as emergency braking events) will be encrypted and sent to the cloud, while low-sensitivity data (such as ordinary driving information) will be processed in the car to reduce the use of network bandwidth. Sensitivity label allocation is based on the classification results of the model output, ensuring data privacy protection and bandwidth efficiency.

[0108] Optionally, the homomorphic modular exponentiation hazard-aware shard encryption described in step S1 is specifically:

[0109] Classify the vehicle danger level according to the real-time driving risk quantification map to obtain a vehicle danger level classification table;

[0110] In this embodiment, the vehicle risk level is divided according to the real-time driving risk quantification map. First, by evaluating the real-time driving data of the vehicle (such as vehicle speed, collision probability, steering angle, etc.), the vehicle will be divided into different risk levels based on the set thresholds (such as collision probability exceeding 60% is high risk, 30%-60% is medium risk, and less than 30% is low risk). Specifically, the system evaluates the current risk situation through a multivariate regression model, groups all driving data according to risk levels, and dynamically adjusts the threshold of each level to cope with different driving scenarios. For example, if the vehicle is driving under complex road conditions, the risk threshold may need to be adjusted appropriately. The vehicle risk level classification table will eventually be output based on the evaluation results of each event. The table includes the risk level (high, medium, low), the corresponding risk score and its timestamp.

[0111] Based on the vehicle danger level classification table and the adaptive sensitivity classification matrix, a danger-sharding dimension mapping rule is constructed, and the modular exponentiation parameters are initialized by homomorphic encryption to obtain a dynamic sharding function.

[0112] In this embodiment, based on the vehicle hazard level classification table and the adaptive sensitivity classification matrix, the hazard-sharding dimension mapping rules are constructed. First, based on different hazard levels (high, medium, and low risks), the data in the sensitivity classification matrix is ​​mapped accordingly, and the modular exponentiation encryption level corresponding to high-sensitivity events is set to a higher level, and the encryption level corresponding to low-sensitivity events is set to a lower level. The specific mapping rules are determined according to the real-time risk assessment and sensitivity classification. Then, by applying a homomorphic encryption algorithm (such as Paillier encryption) to the highly sensitive part of the vehicle driving data, the modular exponentiation parameters are initialized to ensure that each data block can maintain its privacy and operability during the encryption process. The set modular exponentiation parameters are adjusted according to the actual driving environment of the vehicle. For example, an initial encryption modular exponentiation value is set to 2 so that the encrypted data can be used for subsequent encryption calculations. Finally, a dynamic sharding function is obtained, which is used to determine how to handle data fragments of different sensitivities.

[0113] The dynamic slicing function and the sensitivity-classified driving data are subjected to classified hazard-aware slicing, and the high-sensitivity driving data is subjected to modular exponentiation slicing to obtain high-sensitivity driving slicing data; the low-sensitivity driving data is subjected to fixed slicing to obtain low-sensitivity driving slicing data;

[0114] In this embodiment, based on the dynamic sharding function and sensitivity classification driving data, high-sensitivity and low-sensitivity data are sharded respectively. For high-sensitivity data (such as emergency braking, high-speed driving, etc.), the modular exponentiation sharding method is applied, and Paillier homomorphic encryption is used to encrypt these data and generate high-sensitivity sharding data. In the specific implementation, the modular exponentiation parameter is automatically calculated and set to the power of 2 for encryption sharding. The size of each shard is 64 bytes to ensure that sensitive data cannot be cracked during transmission. For low-sensitivity data (such as smooth driving or low-risk driving data), a fixed sharding method is used to shard the data according to a predetermined time window. The size of each slice is 128 bytes, which is convenient for subsequent processing. After all high-sensitivity data and low-sensitivity data are processed separately, high-sensitivity driving sharding data and low-sensitivity driving sharding data are generated.

[0115] Merge the high-sensitivity driving segment data and the low-sensitivity driving segment data to obtain the hazard perception segment data set;

[0116] In this embodiment, the high-sensitivity and low-sensitivity data fragments are merged to form a complete hazard awareness fragmented data set. During the merging process, the high-sensitivity and low-sensitivity data are verified to ensure that their timestamps are consistent and there are no omissions. Specifically, the low-sensitivity fragmented data and the high-sensitivity fragmented data are arranged in chronological order to ensure that the spatiotemporal correlation of the data is not destroyed. After the merger, the formed hazard awareness fragmented data set contains high-sensitivity and low-sensitivity data fragments, and each data fragment retains the corresponding timestamp information. These merged data sets can be used for further encryption and transmission to ensure that there is no information leakage between data of different sensitivities.

[0117] The hazard-aware sharded data set is subjected to Paillier homomorphic encryption operation, the ciphertext of each shard is calculated, and the hazard label is bound to the ciphertext while retaining the spatiotemporal correlation to generate hazard-aware encrypted sharded data.

[0118] In this embodiment, the Paillier homomorphic encryption operation is performed on the merged hazard-aware sharded data set. Specifically, each shard data is encrypted to ensure that the encrypted data maintains the original spatiotemporal correlation and can be calculated and operated in the ciphertext state. The Paillier homomorphic encryption algorithm is used to encrypt the data and keep the encrypted data capable of addition and multiplication operations, which is crucial for subsequent data analysis. During encryption, the ciphertext is calculated for each shard, and the ciphertext includes the original data and its hazard label (such as a "high risk" or "low risk" label). For example, assuming that the collision probability of a certain driving event is 70%, this data is encrypted and bound to the corresponding "high risk" label, while ensuring that the spatiotemporal characteristics of the data are not lost. Through this operation, the generated hazard-aware encrypted sharded data not only ensures the privacy of the data, but also provides encryption protection for subsequent analysis and transmission.

[0119] Optionally, step S2 specifically includes:

[0120] Step S21: obtaining the Internet of Vehicles data stream through the vehicle communication unit, and performing data preprocessing on the Internet of Vehicles data stream to obtain the Internet of Vehicles data stream to be analyzed;

[0121] In this embodiment, the vehicle communication unit (VCU) is used to obtain the Internet of Vehicles data stream. The vehicle communication unit includes a 5G-V2X (Vehicle-to-Everything) module, a DSRC (Dedicated Short-Range Communications) module, and a CAN (Controller Area Network) bus interface, which can collect dynamic interaction data from vehicles, roadside units (RSUs) and cloud platforms in real time. After the data collection is completed, the Internet of Vehicles data stream is preprocessed, which mainly includes three operations: (1) Data cleaning: remove redundant data packets and correct abnormal data points, such as GPS signal drift, acceleration sensor noise, etc.; (2) Time synchronization: unify the timestamp based on the GNSS (Global Navigation Satellite System) timing module to ensure that all data points are in UTC time format and the maximum error is controlled within 5ms; (3) Format standardization: convert data from different communication protocols (such as ETSI ITS-G5, IEEE 802.11p) into JSON format to ensure data format consistency and facilitate subsequent analysis. After preprocessing, the Internet of Vehicles data stream to be analyzed is output for subsequent feature extraction and authentication mechanism construction.

[0122] Step S22: extracting IoV node features from the IoV data stream to be analyzed, and obtaining connected vehicle data, roadside unit data, and cloud platform node identity data;

[0123] In this embodiment, node feature extraction is performed on the Internet of Vehicles data stream to be analyzed to parse the identity information and behavioral characteristics of different data sources. For connected vehicles, the OBU device ID, MAC address, and operating status data sent by the on-board ECU are parsed through the V2X communication protocol, and the vehicle trajectory is constructed in combination with GPS data. For roadside units (RSUs), the device number, broadcast signal strength, communication delay, and message frequency of the RSU are extracted, and the availability score of the RSU is calculated (based on historical packet loss rate and delay fluctuations). For cloud platform nodes, the TLS handshake protocol is used to parse public key fingerprints, server IPs, and access logs, and anomaly detection algorithms based on machine learning are used to mark abnormal traffic behaviors (such as DDoS attacks or unauthorized access). Finally, structured data storage is generated, and the connected vehicle data, RSU data, and cloud platform identity data are stored in the blockchain account book to ensure the integrity and traceability of subsequent analysis.

[0124] Step S23: construct a zero-trust two-way authentication mechanism based on the connected vehicle data, the roadside unit data, and the cloud platform node identity data;

[0125] In this embodiment, a zero-trust two-way authentication mechanism is constructed based on the node characteristics of the Internet of Vehicles data flow to ensure the legitimacy of the identities of all communication entities. When a vehicle is connected to the Internet of Vehicles for the first time, it is necessary to submit a digital signature based on ECC-256 encryption and verify its registration status through a distributed identity authentication system (DID). When the RSU verifies the identity of the vehicle, it uses zero-knowledge proof (ZKP) technology to avoid exposing the real identity information of the vehicle, and uses smart contracts to verify the historical communication records of the vehicle to ensure that its behavior complies with security policies. On the cloud platform side, the TLS1.3 protocol is used for two-way identity authentication, and combined with the dynamic access control (ABAC) strategy, the device trust is continuously evaluated during the session. Once abnormal behavior is found (such as vehicle ID forgery, RSU broadcast deception), the adaptive trust degradation strategy is immediately executed to limit the communication rights of suspicious devices and send an alarm to the security operation center (SOC). Finally, all legitimate devices must complete the zero-trust authentication process before subsequent communication can be carried out.

[0126] Step S24: allocating data sensitivity to the Internet of Vehicles data stream based on the sensitivity-classified driving data to obtain a sensitivity-allocated Internet of Vehicles data stream;

[0127] In this embodiment, data sensitivity is allocated to the Internet of Vehicles data stream according to the sensitivity classification rules of vehicle driving data. The sensitivity score of driving data is calculated based on the driving behavior analysis model (such as Bayesian network), and the classification threshold is set (such as 0-0.3 for low sensitivity, 0.3-0.7 for medium sensitivity, and 0.7-1.0 for high sensitivity). Then, for highly sensitive data (such as collision warning, emergency braking signal, occupant biometric data), fully homomorphic encryption (FHE) is used for encrypted storage, and access rights are recorded on the distributed ledger. For medium sensitive data (such as driving trajectory, historical speed data), attribute-based encryption (ABE) strategy is adopted, and only authorized users can decrypt and access. For low-sensitivity data (such as weather and road conditions information), AES-GCM encryption is used, and anonymous access is allowed. Finally, all data are stored and distributed according to their sensitivity labels to ensure data privacy and security.

[0128] Step S25: Utilize the zero-trust two-way authentication mechanism to implement adaptive encryption on the sensitivity-assigned Internet of Vehicles data stream, obtain zero-trust authentication credentials, and upload them to the Internet of Vehicles for credential storage.

[0129] In this embodiment, a zero-trust two-way authentication mechanism is used to implement adaptive encryption on the Internet of Vehicles data stream after sensitivity allocation, and generate a zero-trust authentication credential (ZTC). Specifically, after high-sensitivity data (such as emergency event logs) are encrypted by FHE, the MPC (secure multi-party computing) protocol is used to de-identify between multiple cloud nodes and stored in an encrypted database. After medium-sensitive data (such as driving behavior records) are encrypted by ABE, only users who meet the access policy can decrypt them, and the access logs are traced and managed through the blockchain. Low-sensitivity data (such as environmental data) are encrypted using AES-GCM, and short-term storage caching is allowed to improve access efficiency. In the process of generating authentication credentials, a ZTC credential is generated based on the encryption level of each data stream. The credential contains the device ID, data hash summary, access control policy, and timestamp information, and uses zero-knowledge proof (ZKP) technology to enable the credential to be verified without exposing the original data. The storage and access of ZTC credentials are managed by smart contracts, and each data access request must verify whether the ZTC credential is legal through the zero-trust authentication process. Finally, the ZTC credentials are stored in the blockchain ledger to ensure that access audits of Internet of Vehicles data are traceable while guaranteeing data confidentiality and integrity.

[0130] Optionally, step S23 is specifically:

[0131] Step S231: extract identity features from the connected vehicle data, roadside unit data, and cloud platform node identity data, construct an interactive identity authentication data set, and set an identity trust score threshold to screen highly trusted nodes, thereby generating an initial identity feature matrix;

[0132] In this embodiment, identity features are extracted from the connected vehicle data, roadside unit data, and cloud platform node identity data to construct an interactive identity authentication data set. Identity feature extraction adopts a strategy based on multi-factor authentication (MFA), including key attributes such as device MAC address, IP address, communication protocol type, and encryption certificate information. In order to screen highly trusted nodes, the identity trust score threshold is set to 0.8, and only nodes with scores higher than this threshold are retained to enter the subsequent authentication process. The identity trust score is calculated comprehensively based on historical interaction logs, data integrity verification results, abnormal behavior detection and other factors, and the score is dynamically updated using the exponentially weighted moving average (EWMA) method. The calculation method is: TrustScore = α × H s +β×N a +γ×C r ; Among them, H s is the historical authentication success rate, N a is the abnormality rate of the last five interactions, C rThe validity score of the device encryption certificate is scored, and α, β, and γ are weight coefficients (the range is 0.3 to 0.5). Finally, the initial identity feature matrix is ​​generated based on the selected high-trust identity nodes, providing a basis for subsequent two-way identity authentication.

[0133] Step S232: Based on the initial identity feature matrix, elliptic curve signature bidirectional identity authentication is performed on the connected vehicle data and the roadside unit data, and a timestamp synchronization deviation threshold Δt=50 is set for timeliness verification to generate a dynamic identity authentication parameter set;

[0134] In this embodiment, based on the generated initial identity feature matrix, the elliptic curve digital signature algorithm (ECDSA) is used for two-way identity authentication. First, a secure handshake protocol is established between the vehicle, the roadside unit and the cloud platform. Each node signs the identity data with a private key and verifies the identity of the other party through the public key. In order to prevent replay attacks, the timestamp synchronization deviation threshold Δt=50ms is set, and the deviation between the timestamp of all identity authentication data and the system clock is required to be no more than 50ms, otherwise it is considered that the authentication has failed. In addition, a challenge-response mechanism is also adopted. During the authentication process, a challenge message is randomly generated, and the interactive requesting party is required to sign it with a private key and return it to further verify the legitimacy of the identity. After the authentication is completed, a dynamic identity authentication parameter set is generated, which contains the identity credentials, timestamp information and signature verification results of the authenticated node. If the deviation exceeds the threshold, it is determined to be an invalid identity request. In addition, the server initiates a challenge response authentication (Challenge-Response) to the requesting device, the challenge code length is 256-bit, and the device is required to return an encrypted response value within 100ms. All verified devices will be added to the dynamic identity authentication parameter set, and their identity status and authentication success timestamp will be recorded.

[0135] Step S233: Perform variable challenge response authentication on the dynamic identity authentication parameter set, calculate the device fingerprint matching degree, set the matching threshold Tf=0.85 to filter abnormal nodes, and generate a device fingerprint authentication matrix;

[0136] In this embodiment, a variable challenge response authentication is performed on a dynamic authentication parameter set to improve the security of authentication. Variable challenge response authentication is implemented based on a dynamic authentication parameter set. This authentication method introduces a dynamic challenge factor based on the traditional challenge response mechanism to prevent replay attacks and man-in-the-middle attacks. Specifically, the server generates a dynamic challenge code based on the device identity (such as MAC address, device certificate public key hash) and the current timestamp. Set different challenge modes and dynamically adjust them according to the trust score and historical authentication status of the device. Low-risk devices (trust score>0.9) use a single hash challenge, that is, the device directly signs and responds to the Challenge value; medium-risk devices (trust score 0.7~0.9) use a two-factor challenge, such as combining device private key signature and time synchronization verification; high-risk devices (trust score<0.7 or recent abnormalities) use an interactive challenge, that is, the server sends different challenge codes in multiple rounds, requiring the device to provide a correct combined response. For example, the server can randomly insert a dynamic noise parameter so that the device must correctly handle the noise when calculating the response value. After receiving the challenge code, the device calculates the response value in combination with its own private key. After receiving the device response, the server first checks the time synchronization deviation. If the device performs stably in multiple rounds of challenges (such as 5 consecutive authentication successes and the time synchronization deviation is <10ms), the system can reduce the challenge intensity, such as reducing the challenge rounds Nc or ​​simplifying the response calculation method. On the contrary, if the device fails to authenticate recently (such as 3 failure records), the system will increase the challenge intensity, such as increasing the challenge rounds to Nc∈[5,10], or introducing multimodal authentication (such as combining biometric data). The device fingerprint recognition technology is used to extract the hardware features of each node (such as CPU serial number, device startup time, wireless communication characteristics, etc.), and calculate the device fingerprint matching degree. The device fingerprint matching degree is calculated using the cosine similarity algorithm, and the matching threshold Tf=0.85 is set. That is, if the fingerprint matching degree of a device is lower than 0.85, it is regarded as a suspicious node and filtered. In order to prevent the forgery of device fingerprints, the challenge factor is dynamically adjusted during the challenge response process, and a hash chain is used to ensure the uniqueness of each challenge. Finally, all nodes that pass the matching degree screening form a device fingerprint authentication matrix to provide data support for further trust score calculation.

[0137] Step S234: training a dynamic trust score model in combination with the interactive identity authentication data set and the device fingerprint authentication matrix, and setting the trust score update step length η=0.05 to calculate the trust weight of the vehicle-cloud platform-roadside unit, and generating a dynamic trust weight matrix;

[0138] In this embodiment, the dynamic trust scoring model is trained by combining the interactive identity authentication data set and the device fingerprint authentication matrix. The trust score adopts the Bayesian dynamic update method to calculate the comprehensive trust value based on factors such as the device's historical interaction records, identity authentication success rate, and abnormal behavior detection. The trust score update step size η is set to 0.05, that is, when the behavior of a device complies with the security rules, its trust score is increased by 0.05, otherwise it is decreased by 0.05. Based on this mechanism, the trust weights between the vehicle-cloud platform-roadside unit are calculated, and a dynamic trust weight matrix is ​​generated. This matrix is ​​used in the subsequent zero-knowledge proof authentication link to ensure that only highly trusted nodes can pass identity authentication.

[0139] Step S235: Couple the zero-knowledge proof verifiable mechanism according to the dynamic trust weight matrix, perform two-way identity confirmation, set the zero-knowledge proof challenge round Nc∈[5,10], and derive the identity credibility based on the identity credibility threshold Ttrust=0.9, thereby generating a zero-trust two-way authentication mechanism.

[0140] In this embodiment, a zero-knowledge proof (ZKP) verifiable mechanism coupling is performed according to the dynamic trust weight matrix to complete two-way identity confirmation. First, the zero-knowledge proof challenge round Nc∈[5,10] is set, that is, each identity authentication request requires 5 to 10 rounds of challenges to prevent replay attacks and forged identity attacks. During the challenge process, the elliptic curve Diffie-Hellman (ECDH) protocol is used to generate session keys, and the legitimacy of the identity is verified based on the ZKP protocol. The calculation of identity credibility is based on the dynamic trust weight matrix, and the identity credibility threshold Ttrust=0.9 is set. Only nodes with a credibility greater than 0.9 can pass the identity authentication. Finally, a zero-trust two-way authentication mechanism is generated based on the zero-knowledge proof results to achieve strict identity authentication and dynamic trust management of all participating nodes in the Internet of Vehicles.

[0141] Optionally, the federated learning privacy feature mapping described in step S3 is specifically:

[0142] Perform Laplace noise local privacy protection on the de-identified hazard segmented dataset, set the privacy budget Q∈[1,10], and calculate the feature perturbation matrix based on the preset feature dimension d=128;

[0143] In this embodiment, Laplace noise local privacy protection is performed on the de-identified hazard segmented data set. First, a privacy budget Q∈[1,10] is selected, and Laplace noise is generated based on the privacy budget. In practical applications, it is assumed that the privacy budget Q=5, and the scale parameter of the noise is calculated based on this budget. For each feature dimension d=128, Laplace noise is added to each data point. For example, for a certain eigenvalue x_i, the perturbed data is x'_i=x_i+\text{Lap}(\mu=0,b=\frac{Q}{\epsilon}), where \epsilon is the adjustment parameter of the privacy budget. This step ensures that when publishing or training the model, the data contains perturbations without exposing the true value, thereby effectively protecting the privacy of the data.

[0144] The feature perturbation matrix is ​​normalized, and the federated feature transformation matrix is ​​constructed according to the federated learning privacy feature mapping parameters α=0.7, β=0.3, γ=0.9 to obtain the privacy-preserving feature vector;

[0145] In this embodiment, the perturbed feature perturbation matrix is ​​normalized. Each feature vector X = [x1, x2, ..., xd] is linearly normalized in the range of [0, 1], that is, the formula x'_i = \frac{x_i-\min(X)}{\max(X)-\min(X)} is used to ensure that all features have the same scale. After normalization, the federated feature transformation matrix is ​​constructed based on the federated learning privacy feature mapping parameters α = 0.7, β = 0.3, γ = 0.9. This matrix generates a privacy-preserving feature vector by combining the weighted sum of different features through the linear transformation Y = \alphaX_1+\beta X_2+\gamma X_3, thereby enhancing the feature expression capability while ensuring privacy.

[0146] Map the privacy-preserving feature vector to a preset prime modulus p=2 2048 Encrypt the computational domain and set the public key parameter g = 2 512 , private key parameter λ = 2 1024 Perform homomorphic encryption transformation to generate an encrypted feature vector set;

[0147] In this embodiment, the privacy-protected feature vector is mapped to a preset prime modulus p=2 2048 Encrypt the computational domain. In practice, choose a large prime number p = 2 2048 and public key parameter g = 2 512 , private key parameter λ = 2 1124, performing Paillier homomorphic encryption. During the encryption process, each data element x'_i in the feature vector is encrypted as E(x'_i) = g^{x'_i}\mod p, thus ensuring that even when the data is transmitted or processed, the content remains undecryptable unless decrypted using the private key.

[0148] Perform distributed gradient calculation on the encrypted feature vector set, set the training batch B∈[64,128], and use the Adam optimizer with η=0.001, β1=0.9, β2=0.999 to update the gradient, and obtain the privacy-preserving feature gradient matrix;

[0149] In this embodiment, distributed gradient calculation is performed on the encrypted feature vector set. During training, the batch size B∈[64,128] is selected and the training data is distributed to multiple computing nodes. On each node, the encrypted data set is trained using the Adam optimizer, with a learning rate η=0.001, momentum coefficients β1=0.9 and β2=0.999. For example, in the first training batch B=64, each node independently calculates the gradient of the encrypted data and aggregates the results to the central server. In this way, the data always remains encrypted, and all calculations during training can be performed without leaking the original data.

[0150] Set the noise standard deviation σ∈[e,2e] and adopt the clipping norm Δ∈[1,2] to perform differential privacy stochastic gradient descent on the privacy-preserving feature gradient matrix, thereby constructing a privacy-preserving hazard feature mapping model;

[0151] In this embodiment, the noise standard deviation σ∈[e,2e] is set and the clipping norm Δ∈[1,2] is used to perform differential privacy stochastic gradient descent on the privacy-preserving feature gradient matrix. In actual operation, the noise standard deviation σ=1.5e and the clipping norm Δ=1.5 are selected, and then random noise is injected into each gradient to ensure the privacy protection of each model parameter. During the training process, the gradient g_i of the model will be calculated through the formula

[0152] g_i'=\text{clip}(g_i,-\Delta,\Delta)+\mathcal{N}(0,\sigma^2) performs differential privacy processing, thereby effectively preventing the leakage of sensitive information of any single data point.

[0153] The privacy-preserving hazard feature mapping model is updated with a zero-knowledge proof model to obtain an anonymized encrypted feature mapping model.

[0154] In this embodiment, a zero-knowledge proof model is updated for the privacy-protected hazard feature mapping model. The zero-knowledge proof method is used for verification. After each update, the model will not expose any internal data or parameters, and only its correctness will be verified. In this process, ZKP(f) is used to verify the model update, where f is the model update process. Set the zero-knowledge proof challenge round Nc=5. In each round, the challenger generates a random challenge and proves the correctness of the model through the responder without revealing any specific data. Finally, after multiple rounds of challenge verification, the updated model is converted into an anonymized encrypted feature mapping model to ensure that all data and calculations remain private.

[0155] Optionally, step S4 is specifically:

[0156] Step S41: Perform encryption feature hashing according to the anonymized encryption feature mapping model, and construct a Merkle tree hierarchical hash structure to generate a preliminary blockchain risk fingerprint;

[0157] In this embodiment, an encrypted feature hash process is generated according to the anonymized encrypted feature mapping model, and a Merkle tree hierarchical hash structure is constructed. Specifically, a hash algorithm, such as SHA-256, is first applied to the privacy-preserving feature vector generated by the anonymized encrypted feature mapping model to perform an encrypted hash process on it. Each hash value represents an encrypted fingerprint of a feature set, and then these fingerprints are organized into a Merkle tree structure. The leaf nodes of the Merkle tree are feature hash values, and the non-leaf nodes are parent node hash values. The nodes are aggregated through multiple recursions, and the root hash of the final generated tree is used as a preliminary version of the blockchain risk fingerprint. In this step, the maximum depth of the Merkle tree is set to 10 layers to ensure the efficiency and encryption strength of the data structure.

[0158] Step S42: Perform breadth-first recursive verification on the preliminary blockchain risk fingerprint to generate an intermediate blockchain risk fingerprint;

[0159] In this embodiment, the preliminary blockchain risk fingerprint is verified by breadth-first recursive verification. Specifically, the breadth-first search (BFS) algorithm starts from the root node of the Merkle tree and verifies the nodes of each layer one by one to ensure that the hash value of each node is correct. By performing a consistency check during the verification process, it is ensured that the hash path from the root to the leaf is consistent and has not been tampered with. During the verification process, the maximum number of recursive layers is set to 5 to ensure the validity and time limit of the calculation. This operation ensures that the verification speed can be maintained efficiently even when the nodes are widely distributed and the amount of data is large.

[0160] Step S43: Perform hierarchical verification on the intermediate blockchain risk fingerprint according to the Merkle tree hierarchical hash structure to obtain the blockchain global risk fingerprint;

[0161] In this embodiment, the intermediate blockchain risk fingerprint is hierarchically verified according to the Merkle tree hierarchical hash structure to obtain the global blockchain risk fingerprint. By using the Merkle tree hierarchical hash structure, the nodes of each layer of the intermediate blockchain risk fingerprint are compared layer by layer to verify the integrity of the hash value. In this process, the hash value of each layer is recursively verified starting from the root node. The verification process is strictly carried out in accordance with the hierarchical structure, and the hierarchical comparison threshold is set to 80% during the comparison process. Ensure that each layer of nodes correctly points to the lower layer and that the hash results of each layer are completely consistent. Finally, the global blockchain risk fingerprint is generated to ensure the credibility of each layer of data fingerprint.

[0162] Step S44: combining the global risk fingerprint of blockchain and the anonymized encrypted feature mapping model, performing data fusion based on the dynamic trust scoring model to obtain a dynamic trust quantification matrix;

[0163] In this embodiment, data fusion is performed based on a dynamic trust scoring model in combination with the blockchain global danger fingerprint and anonymized encrypted feature mapping model. The specific operation is to input data into the trust scoring model according to the generated global danger fingerprint and encrypted feature mapping model, including trust scores, hash values, and feature data from different blockchain nodes. The trust scoring model adopts a weighted average method, sets weight parameters α2=0.6 and β2=0.4, and evaluates the trust of the node based on the historical behavior of the blockchain node and the credibility of the current data. Through the dynamic learning mechanism of the model, the trust weight is continuously adjusted to achieve data fusion, generate a dynamic trust quantization matrix, and represent the relative trust value between each blockchain node, which helps to identify false positives and false negatives data.

[0164] Step S45: quantify the threat behavior on the dynamic trust quantization matrix, set the threat behavior intensity threshold, and thus generate a threat behavior quantization vector.

[0165] In this embodiment, the threat behavior is quantified for the dynamic trust quantification matrix, and a threat behavior intensity threshold is set. Specifically, by analyzing the trust scores of each node in the dynamic trust quantification matrix, the threat behavior related to the node behavior (such as malicious node attacks, data tampering, etc.) is evaluated. According to the characteristics and performance of the threat behavior, the threat behavior intensity threshold is set to 0.7, and the node behavior below this threshold will be regarded as normal behavior. During the quantification process, the K-means clustering algorithm is used to analyze the behavior pattern, and the high-intensity threat behavior is quantified into a threat behavior vector, which represents the severity, type and possible impact of the threat behavior on the system. This quantification vector will provide a basis for subsequent security protection and decision-making.

[0166] Optionally, step S44 is specifically:

[0167] Step S441: perform feature alignment on the blockchain global risk fingerprint and the anonymized encrypted feature mapping model, set the alignment error tolerance threshold ε=0.05, and thus generate a feature alignment data set;

[0168] In this embodiment, feature alignment is performed on the global risk fingerprint of the blockchain and the anonymous encrypted feature mapping model. According to the various feature data of the global risk fingerprint of the blockchain and the anonymous encrypted feature mapping model, these feature data differ in dimension, data type or format. In order to ensure that the data can be effectively fused, the alignment operation is necessary. Specifically, by setting the alignment error tolerance threshold to ε = 0.05, an interpolation algorithm (such as linear interpolation or spline interpolation) is used to align the data features of the two. This operation ensures that data from different sources can be compared and fused under the same standard, thereby generating a feature alignment data set to ensure that subsequent calculations are more accurate.

[0169] Step S442: Based on the feature alignment data set, the global blockchain risk fingerprint is fused with the anonymized encrypted feature mapping model, and the weight parameters α=0.7 and β=0.3 are set to generate a preliminary trust data set;

[0170] In this embodiment, based on the feature alignment data set, the global blockchain risk fingerprint and the anonymized encrypted feature mapping model are fused. In the specific operation, a weighted fusion strategy is adopted to weight the data after feature alignment according to the preset weight parameters α=0.7 (for blockchain global risk fingerprint data) and β=0.3 (for anonymized encrypted feature mapping model data). The weighted combination process includes weighted summation of the values ​​of each feature to generate a preliminary trust data set. In this way, the influence of blockchain data and encrypted feature data in the model can be better balanced to ensure that the fused data is more comprehensive and has practical significance.

[0171] Step S443: standardize the preliminary trust data set and set the standardization range to [0,1], thereby generating a standardized trust data set;

[0172] In this embodiment, the preliminary trust data set is standardized. For all trust score data from the preliminary trust data set. These data may have different dimensions and value ranges, so they need to be normalized to the same standard range through the standardization process. The specific operation is to use the minimum-maximum normalization method to proportionally adjust all data values ​​to the interval [0,1] to ensure that the influence of each feature is equal and is not affected by the original data range. The standardized data is more convenient for further analysis to ensure the accuracy of the trust score calculation.

[0173] Step S444: Based on the standardized trust data set, a dynamic trust scoring model is used for calculation, and the update step length η is set to 0.05, thereby generating a dynamic trust scoring matrix;

[0174] In this embodiment, a dynamic trust scoring model is used for calculation based on a standardized trust data set. The role of the dynamic trust scoring model in this step is to calculate the trust score of each layer of node information of each blockchain fingerprint in the standardized trust data set, and dynamically adjust the trust value of the node according to the set update step η=0.05. The trust score of each blockchain node is continuously updated based on its block transaction data, hash, and smart contract. This process is iteratively updated so that the trust score of the node more accurately reflects its current credibility and security status. Finally, a dynamic trust scoring matrix is ​​generated, which contains the trust scores of all nodes in the blockchain and can be used as a basis for subsequent decision-making.

[0175] Step S445: According to the dynamic trust score matrix, a trust score threshold Ttrust=0.85 is set, nodes with trust scores lower than the threshold are removed, and a dynamic trust quantization matrix is ​​generated.

[0176] In this embodiment, according to the dynamic trust score matrix, the trust score threshold Ttrust=0.85 is set, and the nodes with trust scores below the threshold are eliminated. Specifically, first, according to the dynamic trust score matrix generated in step S444, the trust scores of all nodes are compared. For nodes with trust scores lower than the threshold Ttrust=0.85, it is considered that there are problems, such as malicious behavior or data tampering, and they need to be eliminated from the trust calculation. This operation optimizes the trust score matrix by removing untrustworthy nodes, ensuring that the final generated dynamic trust quantization matrix is ​​more reliable.

[0177] Optionally, step S5 specifically includes:

[0178] Step S51: extracting data transmission features from the vehicle network data stream to be analyzed to obtain vehicle network transmission feature data;

[0179] In this embodiment, by extracting features from the vehicle network data stream to be analyzed, key transmission features such as data packet size, transmission delay fluctuation, packet loss rate, etc. are extracted using data acquisition equipment and algorithms (such as Fourier transform or time domain analysis). For the extraction of data transmission features, the feature extraction window size is set to w = 30ms and the sampling rate is set to f = 1kHz to ensure that the collected data is sufficiently timely and representative. The extracted vehicle network transmission feature data provides basic data support for the subsequent optimization of transmission strategies.

[0180] Step S52: performing transmission strategy gradient optimization based on vehicle network transmission characteristic data to obtain a preliminary transmission strategy matrix;

[0181] In this embodiment, the transmission strategy of the vehicle network is optimized by analyzing the extracted vehicle network transmission feature data and using a machine learning algorithm, such as deep reinforcement learning (DRL). The specific operation is to train a policy network based on the transmission feature data, and use the gradient descent algorithm to optimize the network weights so that the transmission efficiency of the transmission strategy under different environmental conditions is maximized. The goal of the transmission strategy is to reduce transmission delay, improve the success rate of data transmission, and optimize the utilization of spectrum resources. During the optimization process, the optimization parameters of the transmission strategy are set, such as the learning rate η = 0.01 and the batch size B = 64. The optimization process will be continuously updated in each training round to ensure that the network can adapt to changes in the vehicle network environment.

[0182] Step S53: dividing the data transmission priority of the preliminary transmission strategy matrix according to the dynamic trust quantization matrix, and performing regularization processing to obtain a regularized transmission strategy matrix;

[0183] In this embodiment, the data transmission priority of the preliminary transmission strategy matrix is ​​divided according to the dynamic trust quantization matrix, and regularization is performed. According to the obtained preliminary transmission strategy matrix, and in combination with the dynamic trust quantization matrix (such as calibrating each node by trust score), the data transmission priority of different nodes is divided. By adjusting the priority of nodes with low trust scores, it is ensured that important data and nodes with high trust are given priority for data transmission. After priority division, in order to avoid overfitting data and improve the generalization ability of the strategy, the preliminary transmission strategy matrix needs to be regularized. The regularization operation will make the transmission strategy matrix smoother by adding the L2 regularization term, avoiding unreasonable fluctuations during the policy update process. The regularization parameter is set to λ=0.01 to ensure the fairness and rationality of data transmission.

[0184] Step S54: Adaptively adjust the regularized transmission strategy matrix dynamically based on the zero-trust authentication credentials to obtain a global dynamic transmission strategy matrix, and deploy it to the vehicle communication unit in real time.

[0185] In this embodiment, the regularized transmission strategy matrix is ​​combined with zero-trust authentication credentials to perform adaptive dynamic transmission strategy adjustment. Using the zero-trust authentication mechanism, the security of data transmission is determined by verifying the identity of the vehicle communication unit. By verifying the authentication credentials of each communication node (such as dynamic digital certificates or two-way TLS authentication), the transmission strategy is dynamically adjusted in combination with the zero-trust strategy. The specific operation is to perform adaptive adjustments based on the authentication information and real-time network status of each communication node. For example, if the authentication credentials of a node pass the verification, but the trust score is low, the transmission strategy will be appropriately adjusted to reduce the priority of its data transmission. At this time, real-time policy adjustments are performed in combination with the zero-trust mechanism to ensure that data transmission is not only efficient but also secure. Finally, a global dynamic transmission strategy matrix is ​​obtained, which is optimized and deployed in the vehicle communication unit to ensure effective data transmission and efficiency in different communication scenarios.

[0186] Optionally, the present specification also provides a cloud computing-based data security management system, which is used to execute the cloud computing-based data security management method as described above, and the cloud computing-based data security management system includes:

[0187] The hazard awareness sharding module is used to obtain real-time vehicle driving data and perform attribute sensitivity boundary classification to obtain sensitivity classified driving data; the sensitivity classified driving data is encrypted by homomorphic modular power hazard awareness sharding to obtain hazard awareness encrypted sharding data;

[0188] The two-way authentication encryption module is used to obtain the Internet of Vehicles data stream and build a zero-trust two-way authentication mechanism; the zero-trust two-way authentication mechanism is used to implement zero-trust adaptive data stream encryption, obtain zero-trust authentication credentials, and upload them to the Internet of Vehicles for credential storage;

[0189] The multi-party de-identification module is used to perform secure multi-party de-identification processing on the risk-aware encrypted shard data to obtain de-identified encrypted shard data; implement federated learning privacy feature mapping on the de-identified encrypted shard data to obtain an anonymized encrypted feature mapping model;

[0190] The threat behavior quantification module is used to perform Merkle tree hierarchical hash aggregation on the anonymized encrypted feature mapping model to obtain the global blockchain risk fingerprint; a dynamic trust quantification matrix is ​​constructed based on the blockchain global risk fingerprint and the anonymized encrypted feature mapping model to extract the threat behavior quantification vector;

[0191] The data transmission strategy analysis module is used to optimize the dangerous situation data transmission strategy gradient according to the threat behavior quantification vector and zero-trust authentication credentials, obtain the global dynamic transmission strategy matrix, and deploy it to the vehicle communication unit in real time.

[0192] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0193] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A data security management method based on cloud computing, characterized in that: The following steps are involved: Step S1: acquiring real-time vehicle driving data, and performing attribute sensitivity boundary classification to obtain sensitivity classified driving data; Perform homomorphic modular exponentiation hazard-aware shard encryption on the sensitivity-classified driving data to obtain hazard-aware encrypted shard data; Step S2: Obtain the Internet of Vehicles data stream and build a zero-trust two-way authentication mechanism; use the zero-trust two-way authentication mechanism to implement zero-trust adaptive data stream encryption, obtain zero-trust authentication credentials, and upload them to the Internet of Vehicles for credential storage; Step S3: Perform secure multi-party de-identification processing on the risk-aware encrypted shard data to obtain de-identified encrypted shard data; Implement federated learning privacy feature mapping on the de-identified encrypted sharded data to obtain an anonymized encrypted feature mapping model; Step S4: Perform Merkle tree hierarchical hash aggregation on the anonymized encrypted feature mapping model to obtain the global blockchain risk fingerprint; construct a dynamic trust quantification matrix based on the blockchain global risk fingerprint and the anonymized encrypted feature mapping model to extract the threat behavior quantification vector; Step S5: Perform gradient optimization of the dangerous situation data transmission strategy based on the threat behavior quantification vector and zero-trust authentication credentials to obtain a global dynamic transmission strategy matrix, and deploy it to the vehicle communication unit in real time.

2. The data security management method based on cloud computing according to claim 1 is characterized in that: The attribute sensitivity boundary classification described in step S1 is specifically: Acquire real-time vehicle sensor data through the vehicle-mounted sensor group, and perform multi-modal sensor data fusion on the real-time vehicle sensor data to obtain real-time vehicle driving data; Perform vehicle driving pattern recognition on real-time vehicle driving data to obtain a driving behavior semantic label dataset; Segment driving events according to the driving behavior semantic label dataset, extract discrete driving events, and obtain discrete driving event data; Conduct vehicle driving risk assessment on discrete driving event data, quantify driving collision probability, and obtain a real-time driving risk quantification map; Based on the real-time driving risk quantification map, sensitivity is quantified, sensitivity classification thresholds are dynamically calculated, and an adaptive sensitivity classification matrix is ​​obtained; A sensitivity label is assigned to the vehicle driving mode data according to the adaptive sensitivity classification matrix to obtain sensitivity classification driving data, wherein the sensitivity classification driving data includes high-sensitivity driving data and low-sensitivity driving data.

3. The data security management method based on cloud computing according to claim 2 is characterized in that: The homomorphic modular exponentiation hazard-aware shard encryption described in step S1 is specifically as follows: Classify the vehicle danger level according to the real-time driving risk quantification map to obtain a vehicle danger level classification table; Based on the vehicle danger level classification table and the adaptive sensitivity classification matrix, a danger-sharding dimension mapping rule is constructed, and the modular exponentiation parameters are initialized by homomorphic encryption to obtain a dynamic sharding function. Perform classified hazard-aware sharding on the dynamic sharding function and the sensitivity-classified driving data, perform modular exponentiation sharding on the high-sensitivity driving data, and obtain high-sensitivity driving sharding data; The low-sensitivity driving data is fixedly sliced ​​to obtain low-sensitivity driving sliced ​​data; Merge the high-sensitivity driving segment data and the low-sensitivity driving segment data to obtain the hazard perception segment data set; The hazard-aware sharded data set is subjected to Paillier homomorphic encryption operation, the ciphertext of each shard is calculated, and the hazard label is bound to the ciphertext while retaining the spatiotemporal correlation to generate hazard-aware encrypted sharded data.

4. The data security management method based on cloud computing according to claim 1 is characterized in that: Step S2 is specifically as follows: Step S21: obtaining the Internet of Vehicles data stream through the vehicle communication unit, and performing data preprocessing on the Internet of Vehicles data stream to obtain the Internet of Vehicles data stream to be analyzed; Step S22: extracting IoV node features from the IoV data stream to be analyzed, and obtaining connected vehicle data, roadside unit data, and cloud platform node identity data; Step S23: construct a zero-trust two-way authentication mechanism based on the connected vehicle data, the roadside unit data, and the cloud platform node identity data; Step S24: allocating data sensitivity to the Internet of Vehicles data stream based on the sensitivity-classified driving data to obtain a sensitivity-allocated Internet of Vehicles data stream; Step S25: Utilize the zero-trust two-way authentication mechanism to implement adaptive encryption on the sensitivity-assigned Internet of Vehicles data stream, obtain zero-trust authentication credentials, and upload them to the Internet of Vehicles for credential storage.

5. The data security management method based on cloud computing according to claim 4 is characterized in that: Step S23 is specifically as follows: Step S231: extract identity features from the connected vehicle data, roadside unit data, and cloud platform node identity data, construct an interactive identity authentication data set, and set an identity trust score threshold to screen highly trusted nodes, thereby generating an initial identity feature matrix; Step S232: Based on the initial identity feature matrix, elliptic curve signature bidirectional identity authentication is performed on the connected vehicle data and the roadside unit data, and a timestamp synchronization deviation threshold Δt=50 is set for timeliness verification to generate a dynamic identity authentication parameter set; Step S233: Perform variable challenge response authentication on the dynamic identity authentication parameter set, calculate the device fingerprint matching degree, set the matching threshold Tf=0.85 to filter abnormal nodes, and generate a device fingerprint authentication matrix; Step S234: training a dynamic trust score model in combination with the interactive identity authentication data set and the device fingerprint authentication matrix, and setting the trust score update step length η=0.05 to calculate the trust weight of the vehicle-cloud platform-roadside unit, and generating a dynamic trust weight matrix; Step S235: Couple the zero-knowledge proof verifiable mechanism according to the dynamic trust weight matrix, perform two-way identity confirmation, set the zero-knowledge proof challenge round Nc∈[5,10], and derive the identity credibility based on the identity credibility threshold Ttrust=0.9, thereby generating a zero-trust two-way authentication mechanism.

6. The data security management method based on cloud computing according to claim 1, characterized in that: The federated learning privacy feature mapping described in step S3 is specifically: Perform Laplace noise local privacy protection on the de-identified hazard segmented dataset, set the privacy budget Q∈[1,10], and calculate the feature perturbation matrix based on the preset feature dimension d=128; The feature perturbation matrix is ​​normalized, and the federated feature transformation matrix is ​​constructed according to the federated learning privacy feature mapping parameters α=0.7, β=0.3, γ=0.9 to obtain the privacy-preserving feature vector; Map the privacy-preserving feature vector to a preset prime modulus p=2 2048 Encrypt the computational domain and set the public key parameter g = 2 512 , private key parameter λ = 2 1024 Perform homomorphic encryption transformation to generate an encrypted feature vector set; Perform distributed gradient calculation on the encrypted feature vector set, set the training batch B∈[64,128], and use the Adam optimizer with η=0.001, β1=0.9, β2=0.999 to update the gradient, and obtain the privacy-preserving feature gradient matrix; Set the noise standard deviation σ∈[e,2e] and adopt the clipping norm Δ∈[1,2] to perform differential privacy stochastic gradient descent on the privacy-preserving feature gradient matrix, thereby constructing a privacy-preserving hazard feature mapping model; The privacy-preserving hazard feature mapping model is updated with a zero-knowledge proof model to obtain an anonymized encrypted feature mapping model.

7. The data security management method based on cloud computing according to claim 1, characterized in that: Step S4 is specifically as follows: Step S41: Perform encryption feature hashing according to the anonymized encryption feature mapping model, and construct a Merkle tree hierarchical hash structure to generate a preliminary blockchain risk fingerprint; Step S42: Perform breadth-first recursive verification on the preliminary blockchain risk fingerprint to generate an intermediate blockchain risk fingerprint; Step S43: Perform hierarchical verification on the intermediate blockchain risk fingerprint according to the Merkle tree hierarchical hash structure to obtain the blockchain global risk fingerprint; Step S44: combining the blockchain global risk fingerprint and the anonymized encrypted feature mapping model, performing data fusion based on the dynamic trust scoring model to obtain a dynamic trust quantification matrix; Step S45: quantify the threat behavior on the dynamic trust quantization matrix, set the threat behavior intensity threshold, and thus generate a threat behavior quantization vector.

8. The data security management method based on cloud computing according to claim 7 is characterized in that: Step S44 is specifically as follows: Step S441: perform feature alignment on the blockchain global risk fingerprint and the anonymized encrypted feature mapping model, set the alignment error tolerance threshold ε=0.05, and thus generate a feature alignment data set; Step S442: Based on the feature alignment data set, the global blockchain risk fingerprint is fused with the anonymized encrypted feature mapping model, and the weight parameters α=0.7 and β=0.3 are set to generate a preliminary trust data set; Step S443: standardize the preliminary trust data set and set the standardization range to [0,1], thereby generating a standardized trust data set; Step S444: Based on the standardized trust data set, a dynamic trust scoring model is used for calculation, and the update step length η is set to 0.05, thereby generating a dynamic trust scoring matrix; Step S445: According to the dynamic trust score matrix, a trust score threshold Ttrust=0.85 is set, nodes with trust scores lower than the threshold are eliminated, and a dynamic trust quantization matrix is ​​generated.

9. The data security management method based on cloud computing according to claim 1, characterized in that: Step S5 is specifically as follows: Step S51: extracting data transmission features from the vehicle network data stream to be analyzed to obtain vehicle network transmission feature data; Step S52: performing transmission strategy gradient optimization based on vehicle network transmission characteristic data to obtain a preliminary transmission strategy matrix; Step S53: dividing the data transmission priority of the preliminary transmission strategy matrix according to the dynamic trust quantization matrix, and performing regularization processing to obtain a regularized transmission strategy matrix; Step S54: Adaptively adjust the regularized transmission strategy matrix dynamically based on the zero-trust authentication credentials to obtain a global dynamic transmission strategy matrix, and deploy it to the vehicle communication unit in real time.

10. A data security management system based on cloud computing, characterized in that: Used to execute the cloud computing-based data security management method as claimed in claim 1, the cloud computing-based data security management system comprises: The hazard awareness sharding module is used to obtain real-time vehicle driving data and perform attribute sensitivity boundary classification to obtain sensitivity classified driving data; the sensitivity classified driving data is encrypted by homomorphic modular power hazard awareness sharding to obtain hazard awareness encrypted sharding data; The two-way authentication encryption module is used to obtain the Internet of Vehicles data stream and build a zero-trust two-way authentication mechanism; the zero-trust two-way authentication mechanism is used to implement zero-trust adaptive data stream encryption, obtain zero-trust authentication credentials, and upload them to the Internet of Vehicles for credential storage; The multi-party de-identification module is used to perform secure multi-party de-identification processing on the risk-aware encrypted shard data to obtain de-identified encrypted shard data; implement federated learning privacy feature mapping on the de-identified encrypted shard data to obtain an anonymized encrypted feature mapping model; The threat behavior quantification module is used to perform Merkle tree hierarchical hash aggregation on the anonymized encrypted feature mapping model to obtain the global blockchain risk fingerprint; a dynamic trust quantification matrix is ​​constructed based on the blockchain global risk fingerprint and the anonymized encrypted feature mapping model to extract the threat behavior quantification vector; The data transmission strategy analysis module is used to optimize the dangerous situation data transmission strategy gradient according to the threat behavior quantification vector and zero-trust authentication credentials, obtain the global dynamic transmission strategy matrix, and deploy it to the vehicle communication unit in real time.

Images