Data security gateway method and system based on edge privacy computing

Abstract

The present invention provides a data security gateway method and system based on edge privacy computing, which relates to the field of data security technology, including receiving a data access request from a user device, using a distributed identity authentication protocol to authenticate the user device to determine the data access right; performing adaptive privacy processing on the requested data information according to the data access right, determining the privacy protection level corresponding to the data access right, obtaining privacy protection data, performing secondary privacy protection on the privacy protection data using a privacy protection model, and obtaining secondary privacy protection data; performing homomorphic encryption processing on the secondary privacy protection data using a private key to obtain homomorphically encrypted privacy data; receiving decrypted data fragments after multiple edge nodes decrypt the data fragments using a public key and a verification share, splicing the decrypted data fragments to obtain complete decrypted data, and returning the decrypted data to the user device.

Description

Technical Field

[0001] The present invention relates to data security technology, and in particular to a data security gateway method and system based on edge privacy computing. Background Art

[0002] Traditional data processing methods usually upload data to the cloud for centralized processing, but this method has the following problems: on the one hand, the uploading and storage of large amounts of sensitive data can easily lead to security risks such as data leakage and privacy infringement; on the other hand, the uploading of massive data places high demands on network bandwidth and latency, making it difficult to meet the needs of real-time processing.

[0003] To solve these problems, edge computing came into being. Edge computing deploys computing, storage, and network resources at the edge of the network close to the data source. By processing data locally, it can significantly reduce the amount of data uploaded, reduce network transmission pressure, and improve data processing efficiency. However, directly processing sensitive data on edge nodes still faces the risk of data security and privacy leakage. Especially in scenarios with multiple parties involved, the ownership and usage rights of data among different parties are not clearly defined, and there is a lack of effective privacy protection mechanisms.

[0004] In the existing technology, there are some solutions to the data security and privacy protection issues of edge computing. One type of solution is based on traditional encryption technology, such as symmetric encryption, public key encryption, etc., to encrypt sensitive data before uploading it to the edge node, but this solution cannot support direct processing of encrypted data. The data must be decrypted before calculation, which poses a risk of data leakage. Another type of solution is based on secure multi-party computing technology, such as secret sharing, secure multi-party computing, etc., by dividing the data into multiple fragments and distributing them on different edge nodes to ensure that a single node cannot restore the complete data, but this solution has high communication and computing overhead and is difficult to apply to real-time scenarios. Summary of the invention

[0005] The embodiments of the present invention provide a data security gateway method and system based on edge privacy computing, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] Provides a data security gateway method based on edge privacy computing, including:

[0008] Receive a data access request from a user device, the data access request including a user identifier, requested data information, and a device identifier of the user device; authenticate the user device using a distributed identity authentication protocol according to the user identifier and the device identifier, and obtain data access rights corresponding to the user identifier from a preset rights management database;

[0009] According to the data access permission, adaptive privacy processing is performed on the requested data information, and the privacy protection level corresponding to the data access permission is determined to obtain privacy protection data; a privacy protection model is established between the user device and the data security gateway using a federated learning algorithm, and secondary privacy protection is performed on the privacy protection data using the privacy protection model to obtain secondary privacy protection data;

[0010] Generate a private key and a public key pair corresponding to the user device using a distributed key generation algorithm, send the public key to the user device, and use the private key to perform homomorphic encryption on the secondary privacy protection data to obtain homomorphically encrypted privacy data; divide the homomorphically encrypted privacy data into multiple data slices, and use a secret sharing algorithm to generate a verification share for each data slice, and send the data slices and the verification share to multiple edge nodes corresponding to the user device through a blockchain network;

[0011] Receive the decrypted data slices after the multiple edge nodes decrypt the data slices using the public key and the verification share, splice the decrypted data slices to obtain complete decrypted data, and return the decrypted data to the user device.

[0012] In an optional embodiment,

[0013] According to the data access permission, the requested data information is adaptively processed for privacy, a privacy protection level corresponding to the data access permission is determined, and the obtained privacy protection data includes:

[0014] Determine the privacy protection level based on data access rights, and divide data access rights into three types: full access, partial access, and restricted access, corresponding to low, medium, and high privacy protection levels respectively;

[0015] According to the privacy protection level, select a corresponding privacy protection algorithm from three privacy protection algorithms: differential privacy, homomorphic encryption, and secure multi-party computing;

[0016] Processing the requested data information using the selected privacy protection algorithm to obtain privacy-protected data;

[0017] When the differential privacy algorithm is selected, random noise is added to the requested data information using the Laplace mechanism to obtain privacy-protected data, where the amplitude of the random noise is inversely proportional to the preset privacy protection budget;

[0018] When the homomorphic encryption algorithm is selected, the Paillier encryption scheme is used to encrypt the requested data information, and the ciphertext data is calculated to obtain the privacy-protected data. The Paillier encryption scheme supports addition and multiplication operations of the ciphertext data.

[0019] When a secure multi-party computing algorithm is selected, secret sharing technology is used to divide the requested data information into multiple data slices, and the data slices are sent to multiple participants respectively. The computing task is completed through a secure protocol under the premise of privacy protection to obtain privacy-protected data.

[0020] In an optional embodiment,

[0021] A privacy protection model is established between the user device and the data security gateway using a federated learning algorithm, and the privacy protection model is used to perform secondary privacy protection on the privacy protection data, and the secondary privacy protection data obtained includes:

[0022] Each participant uses a local data set to train a local privacy-preserving teacher model. The local privacy-preserving teacher model is built based on a deep neural network model. A differential privacy mechanism is used during the training process. Laplace noise is added to the gradient of model parameters to control the privacy budget during the training process.

[0023] Each participant uses the local privacy-preserving teacher model to infer the local data set, calculates the conditional probability of each sample in the local data set belonging to the corresponding category through forward propagation, and obtains soft label data containing the learning knowledge of the local privacy-preserving teacher model, and the soft label data is the probability distribution of each sample belonging to the corresponding category;

[0024] Each participant encrypts the soft label data using homomorphic encryption technology, and uploads the homomorphically encrypted soft label data to the data security gateway;

[0025] The data security gateway collects the soft label data uploaded by each participant, constructs a global soft label data set, and uses the global soft label data set to train a global student model. The global student model is constructed using a deep neural network model, and model parameters are compressed based on the local privacy protection teacher model and the same network framework as the local privacy protection teacher model is used.

[0026] The trained global student model is sent to each participant, and each participant uses the global student model to perform secondary protection on the privacy-protected data, calculates the soft label of the privacy-protected data through forward propagation, and uses knowledge distillation technology to perturb the soft label to obtain secondary privacy-protected data.

[0027] In an optional embodiment,

[0028] Dividing the homomorphically encrypted private data into multiple data shards, generating a verification share for each data shard using a secret sharing algorithm, and sending the data shards and the verification share to multiple edge nodes corresponding to the user device through a blockchain network includes:

[0029] The homomorphically encrypted private data is vertically divided into multiple data shards. For each data shard, a verification share is generated using a secret sharing algorithm. A polynomial is randomly selected over a finite field so that the value of the polynomial at zero is equal to the data shard. The value of the polynomial is calculated at multiple different points to obtain multiple verification shares.

[0030] The data shards and the corresponding verification shares are sent to multiple edge nodes corresponding to the user devices using the blockchain network. The transmission of the data shards and the verification shares is achieved in the form of encrypted transactions. Each encrypted transaction contains at least one of the sender's address, the receiver's address, the data shard hash, the verification share hash, and the sender's private key's digital signature on the encrypted transaction.

[0031] Encrypted transactions are packaged into blocks and added to the blockchain through a consensus mechanism. After each edge node receives the data shard and verification share, it stores them locally.

[0032] In an optional embodiment,

[0033] The method further comprises:

[0034] When an independent third party needs to obtain homomorphically encrypted private data, the independent third party sends a secret reply request to the edge node storing the homomorphically encrypted private data;

[0035] The edge node that receives the secret reply request sends the locally stored verification shares to an independent third party through the blockchain network. After the independent third party collects a predetermined number of verification shares, it restores the data shards of the encrypted private data through an interpolation formula and splices all the data shards to obtain complete encrypted private data.

[0036] In an optional embodiment,

[0037] Receiving the decrypted data slices after the multiple edge nodes decrypt the data slices using the public key and the verification share, and splicing the decrypted data slices to obtain complete decrypted data includes:

[0038] The data owner generates a decryption request, which contains the request ID, the blockchain address of the data owner, the index list of the data shards to be decrypted, the key for decrypting the verification share, and the digital signature of the data request using the data owner's private key. The decryption request is sent through the blockchain network to multiple edge nodes that store the corresponding data shards.

[0039] After receiving the decryption request, the edge node uses the public key of the data owner to verify the digital signature of the decryption request to ensure that the decryption request has not been tampered with, and checks whether the index list of the data shard in the decryption request has existed in the local storage. If the verification is successful, the corresponding verification share is read from the local storage according to the index list of each data shard, and the key pair for decrypting the verification share is used to decrypt it to obtain the decrypted verification share value. According to the verification share value and the corresponding node number, the original data shard is reconstructed;

[0040] The edge node combines the decrypted verification shares into a decrypted data shard, which contains the index of the original data shard corresponding to the decrypted data shard, the decrypted data shard content, and the digital signature of the decrypted data shard using the edge node private key, and sends the decrypted data shard to the data owner through the blockchain network;

[0041] After the data owner receives the decrypted data slice sent by the edge node, he uses the public key of the edge node to verify the digital signature of the decrypted data slice to ensure that the data has not been tampered with, and checks whether the slice index of the decrypted data slice is consistent with the index in the request. According to the slice index, all the received decrypted data slices are sequentially spliced ​​into a complete plaintext data matrix; if an error occurs in the verification process or the data is incomplete, the data owner resends the decryption request to the corresponding edge node until the complete plaintext data is obtained.

[0042] According to a second aspect of the embodiments of the present invention,

[0043] Provides a data security gateway system based on edge privacy computing, including:

[0044] The first unit is configured to receive a data access request from a user device, the data access request including a user identifier, requested data information, and a device identifier of the user device; authenticate the user device using a distributed identity authentication protocol according to the user identifier and the device identifier, and obtain data access rights corresponding to the user identifier from a preset rights management database;

[0045] The second unit is used to perform adaptive privacy processing on the requested data information according to the data access permission, determine the privacy protection level corresponding to the data access permission, and obtain privacy protection data; use a federated learning algorithm to establish a privacy protection model between the user device and the data security gateway, and use the privacy protection model to perform secondary privacy protection on the privacy protection data to obtain secondary privacy protection data;

[0046] The third unit is used to generate a private key and a public key pair corresponding to the user device by using a distributed key generation algorithm, send the public key to the user device, and use the private key to perform homomorphic encryption on the secondary privacy protection data to obtain homomorphically encrypted privacy data; divide the homomorphically encrypted privacy data into multiple data slices, and use a secret sharing algorithm to generate a verification share for each data slice, and send the data slices and the verification share to multiple edge nodes corresponding to the user device through a blockchain network;

[0047] The fourth unit is used to receive the decrypted data slices after the multiple edge nodes decrypt the data slices using the public key and the verification share, splice the decrypted data slices to obtain complete decrypted data, and return the decrypted data to the user device.

[0048] According to a third aspect of the embodiments of the present invention,

[0049] An electronic device is provided, comprising:

[0050] processor;

[0051] a memory for storing processor-executable instructions;

[0052] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0053] According to a fourth aspect of the embodiments of the present invention,

[0054] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0055] The present invention authenticates the user device by introducing a distributed identity authentication protocol, ensuring that only legitimate users can access the data. At the same time, adaptive privacy processing is performed on the requested data according to the user's data access rights, providing privacy protection of different granularities to meet the privacy protection needs in different scenarios. A privacy protection model is established between the user device and the data security gateway using a federated learning algorithm, and secondary privacy protection is performed on the data that has undergone adaptive privacy processing, further reducing the risk of data leakage and privacy infringement.

[0056] Homomorphic encryption technology is used to encrypt the data after secondary privacy protection to generate homomorphic encrypted privacy data. Through homomorphic encryption, data can be calculated and processed in an encrypted state without decrypting the data, which greatly reduces the risk of data leakage. The secret sharing algorithm is used to divide the homomorphic encrypted privacy data into multiple data shards, and the data shards are distributed to multiple edge nodes for parallel processing, making full use of the distributed resources of edge computing and improving data processing efficiency. At the same time, by increasing the number of edge nodes, the computing power of the system can be easily expanded.

[0057] The present invention utilizes the blockchain network to transmit data fragments and verification shares, and leverages the decentralized and tamper-proof characteristics of the blockchain to ensure the security and integrity of the data transmission process. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A schematic diagram of a data security gateway method based on edge privacy computing according to an embodiment of the present invention;

[0059] Figure 2 This is a structural diagram of a data security gateway system based on edge privacy computing in an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0062] Figure 1 FIG. 1 is a flow chart of a data security gateway method based on edge privacy computing according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0063] S101. Receive a data access request from a user device, the data access request including a user identifier, requested data information, and a device identifier of the user device; authenticate the user device using a distributed identity authentication protocol based on the user identifier and the device identifier, and obtain data access rights corresponding to the user identifier from a preset rights management database;

[0064] Exemplarily, the user device sends a data access request to the data security gateway, and the request includes the following information:

[0065] User identification: used to uniquely identify the user, which can be a user name, user ID, etc. Requested data information: indicates the data resource the user wants to access, which can be a data table name, data file path, etc. Device identification of the user device: used to uniquely identify the user device, which can be a device serial number, MAC address, etc.

[0066] After receiving the data access request, the data security gateway parses the request content and extracts the user ID, the requested data information and the device ID to prepare for the subsequent identity authentication and permission management. The data security gateway uses the distributed identity authentication protocol to authenticate the user device based on the received user ID and device ID. The distributed identity authentication protocol can be implemented using blockchain technology. The specific steps are as follows:

[0067] The data security gateway sends the user ID and device ID to the identity authentication smart contract in the blockchain network. The identity authentication smart contract checks whether there is identity information matching the user ID and device ID in the blockchain ledger. If there is matching identity information, it means that the identity of the user device is legal and the identity authentication is passed; otherwise, the identity authentication fails. The identity authentication smart contract returns the identity authentication result to the data security gateway.

[0068] Through the distributed identity authentication protocol, the single point failure problem of centralized identity authentication is avoided, and the reliability and security of identity authentication are improved.

[0069] The data security gateway obtains the data access rights corresponding to the user ID from the preset rights management database. The rights management database stores the mapping relationship between the user ID and the data access rights. The data access rights specify the range of data resources that the user can access and the types of operations allowed (such as reading, writing, modifying, deleting, etc.). In specific implementation, a relational database (such as MySQL) or a non-relational database (such as MongoDB) can be used to store the rights management information.

[0070] S102. Perform adaptive privacy processing on the requested data information according to the data access permission, determine the privacy protection level corresponding to the data access permission, and obtain privacy protection data; use a federated learning algorithm to establish a privacy protection model between the user device and the data security gateway, and use the privacy protection model to perform secondary privacy protection on the privacy protection data to obtain secondary privacy protection data;

[0071] S103. Generate a private key and a public key pair corresponding to the user device using a distributed key generation algorithm, send the public key to the user device, and use the private key to perform homomorphic encryption on the secondary privacy protection data to obtain homomorphically encrypted privacy data; divide the homomorphically encrypted privacy data into multiple data slices, and use a secret sharing algorithm to generate a verification share for each data slice, and send the data slices and the verification share to multiple edge nodes corresponding to the user device through a blockchain network;

[0072] S014. Receive the decrypted data slices after the multiple edge nodes decrypt the data slices using the public key and the verification share, splice the decrypted data slices to obtain complete decrypted data, and return the decrypted data to the user device.

[0073] The present invention authenticates the user device by introducing a distributed identity authentication protocol, ensuring that only legitimate users can access the data. At the same time, adaptive privacy processing is performed on the requested data according to the user's data access rights, providing privacy protection of different granularities to meet the privacy protection needs in different scenarios. A privacy protection model is established between the user device and the data security gateway using a federated learning algorithm, and secondary privacy protection is performed on the data that has undergone adaptive privacy processing, further reducing the risk of data leakage and privacy infringement.

[0074] Homomorphic encryption technology is used to encrypt the data after secondary privacy protection to generate homomorphic encrypted privacy data. Through homomorphic encryption, data can be calculated and processed in an encrypted state without decrypting the data, which greatly reduces the risk of data leakage. The secret sharing algorithm is used to divide the homomorphic encrypted privacy data into multiple data shards, and the data shards are distributed to multiple edge nodes for parallel processing, making full use of the distributed resources of edge computing and improving data processing efficiency. At the same time, by increasing the number of edge nodes, the computing power of the system can be easily expanded.

[0075] The present invention utilizes the blockchain network to transmit data fragments and verification shares, and leverages the decentralized and tamper-proof characteristics of the blockchain to ensure the security and integrity of the data transmission process.

[0076] In an optional embodiment,

[0077] According to the data access permission, the requested data information is adaptively processed for privacy, a privacy protection level corresponding to the data access permission is determined, and the obtained privacy protection data includes:

[0078] Determine the privacy protection level based on data access rights, and divide data access rights into three types: full access, partial access, and restricted access, corresponding to low, medium, and high privacy protection levels respectively;

[0079] According to the privacy protection level, select a corresponding privacy protection algorithm from three privacy protection algorithms: differential privacy, homomorphic encryption, and secure multi-party computing;

[0080] Processing the requested data information using the selected privacy protection algorithm to obtain privacy-protected data;

[0081] When the differential privacy algorithm is selected, random noise is added to the requested data information using the Laplace mechanism to obtain privacy-protected data, where the amplitude of the random noise is inversely proportional to the preset privacy protection budget;

[0082] When the homomorphic encryption algorithm is selected, the Paillier encryption scheme is used to encrypt the requested data information, and the ciphertext data is calculated to obtain the privacy-protected data. The Paillier encryption scheme supports addition and multiplication operations of the ciphertext data.

[0083] When a secure multi-party computing algorithm is selected, secret sharing technology is used to divide the requested data information into multiple data slices, and the data slices are sent to multiple participants respectively. The computing task is completed through a secure protocol under the premise of privacy protection to obtain privacy-protected data.

[0084] Exemplarily, this application proposes an adaptive privacy processing method, which performs privacy protection on requested data information according to data access rights, determines the privacy protection level corresponding to the data access rights, and thus obtains privacy-protected data. This method can dynamically adjust the intensity of privacy protection according to different data access rights to achieve adaptive privacy protection.

[0085] In order to achieve adaptive privacy protection, we first need to determine the corresponding privacy protection level based on data access permissions. Data access permissions reflect the user's access and use permissions to the data, and different permissions correspond to different privacy protection requirements. This article defines three privacy protection levels: low, medium, and high, which correspond to different data access permissions.

[0086] When users have full access to and use of data, it corresponds to a low level of privacy protection. In this case, users can access complete, unprotected raw data. Low-level privacy protection is mainly applicable to data generated by users themselves or data that users have full access to.

[0087] When users can only access part of the data, or can only perform limited data operations, it corresponds to medium-level privacy protection. In this case, a certain degree of privacy protection is required for the original data, but the protection is relatively weak. Medium-level privacy protection is mainly applicable to data that is partially open to users, and users can access some fields or perform some operations.

[0088] When users can only access data that has been strictly protected, it corresponds to a high level of privacy protection. In this case, strong privacy protection is required for the original data to ensure that users cannot infer the privacy information of the original data from the privacy-protected data. High-level privacy protection is mainly applicable to sensitive data or data involving personal privacy.

[0089] After determining the privacy protection level, you need to select the corresponding privacy protection algorithm according to the privacy protection level. This article adopts three mainstream privacy protection algorithms: differential privacy, homomorphic encryption, and secure multi-party computing, corresponding to low, medium, and high privacy protection levels respectively.

[0090] Differential Privacy is a privacy protection method that adds random noise to the original data so that changes to any record in the data set will not affect the output results. Differential privacy can ensure that attackers cannot infer private information in the original data from the output results. This paper uses the Laplace mechanism to implement differential privacy, adding Laplace noise to the original data. The noise amplitude is inversely proportional to the privacy protection budget. The privacy protection budget reflects the strength of privacy protection. The smaller the budget, the greater the privacy protection strength, but at the same time the data availability will also be reduced.

[0091] Homomorphic encryption is an encryption scheme that allows calculations to be performed directly on ciphertext. Through homomorphic encryption, encrypted data can be directly processed and analyzed without decrypting the data. This paper adopts the Paillier homomorphic encryption scheme to encrypt the original data and then perform calculations on the ciphertext. Paillier homomorphic encryption supports addition and multiplication operations, which can meet most data calculation needs.

[0092] Secure Multi-Party Computation is a protocol that allows multiple parties to jointly compute a function without revealing their private data. Through secure multi-party computation, participants can collaborate to complete computing tasks without exposing their private data. This paper uses secret sharing technology to implement secure multi-party computation, dividing the original data into multiple shards and sending them to different participants. Each participant can only obtain one shard and cannot restore the complete original data. During the calculation process, each participant completes the calculation through a secure protocol and obtains the calculation result.

[0093] After selecting the privacy protection algorithm, the original data is processed using the corresponding algorithm to obtain privacy-protected data. Privacy-protected data can protect the privacy information in the original data to a certain extent, while also supporting subsequent calculations and analysis.

[0094] For data protected by differential privacy, attackers can access data with added noise, but cannot infer the private information in the original data. Differential privacy provides strict privacy protection guarantees, but the introduction of noise may affect the availability of data.

[0095] For data protected by homomorphic encryption, attackers can only access the ciphertext data and cannot decrypt the original data without knowing the key. Homomorphic encryption supports direct calculations on ciphertext, but the calculation efficiency is relatively low, and the encryption and calculation processes are relatively complicated.

[0096] For data protected by secure multi-party computing, each participant can only obtain one shard of the data, and a single shard cannot restore the original data. Secure multi-party computing supports multiple parties to collaborate to complete computing tasks without leaking privacy, but it requires the cooperation of multiple participants, and the communication and computing overhead is relatively large.

[0097] The privacy-preserving data obtained by the adaptive privacy processing method can protect the privacy information in the original data to a certain extent, while also supporting subsequent calculations and analysis. The privacy-preserving data generated by different privacy-preserving algorithms have different uses.

[0098] For data protected by differential privacy, we can directly perform statistical analysis on the privacy-protected data and calculate statistics such as mean, variance, and distribution. Due to the introduction of random noise, there will be a certain error in the analysis results, but we can balance privacy protection and data availability by adjusting the privacy protection budget.

[0099] For data protected by homomorphic encryption, calculations can be performed directly on the ciphertext, such as summation and averaging. The calculation result is still ciphertext, which needs to be decrypted with the private key to get the plaintext result. Homomorphic encryption supports a variety of computing operations, but the computing efficiency is relatively low. In actual application, it is necessary to balance the amount of computing and performance overhead.

[0100] For data protected by secure multi-party computing, all participants need to complete the computing task together through a secure protocol. For example, if you want to calculate the mean of the data of multiple participants, you can use a secure summation protocol. Each participant encrypts his or her own data, and then sends the ciphertext to other participants. Each participant calculates the received ciphertext and finally obtains the encrypted sum result, which is then jointly decrypted to obtain the plaintext result. Secure multi-party computing can complete various computing tasks without leaking privacy, but it requires the cooperation of multiple participants, and the communication and computing overhead is large.

[0101] This application proposes an adaptive privacy processing method, which dynamically adjusts the privacy protection strength according to the data access rights to achieve adaptive privacy protection. The method first determines the privacy protection level according to the data access rights, then selects the corresponding privacy protection algorithm according to the privacy protection level, and uses the selected privacy protection algorithm to process the original data, and finally obtains the privacy-protected data. The three mainstream privacy protection algorithms of differential privacy, homomorphic encryption and secure multi-party computing are adopted, corresponding to the three privacy protection levels of low, medium and high, respectively, which can meet the privacy protection needs in different scenarios.

[0102] In an optional embodiment,

[0103] A privacy protection model is established between the user device and the data security gateway using a federated learning algorithm, and the privacy protection model is used to perform secondary privacy protection on the privacy protection data, and the secondary privacy protection data obtained includes:

[0104] Each participant uses a local data set to train a local privacy-preserving teacher model. The local privacy-preserving teacher model is built based on a deep neural network model. A differential privacy mechanism is used during the training process. Laplace noise is added to the gradient of model parameters to control the privacy budget during the training process.

[0105] Each participant uses the local privacy-preserving teacher model to infer the local data set, calculates the conditional probability of each sample in the local data set belonging to the corresponding category through forward propagation, and obtains soft label data containing the learning knowledge of the local privacy-preserving teacher model, and the soft label data is the probability distribution of each sample belonging to the corresponding category;

[0106] Each participant encrypts the soft label data using homomorphic encryption technology, and uploads the homomorphically encrypted soft label data to the data security gateway;

[0107] The data security gateway collects the soft label data uploaded by each participant, constructs a global soft label data set, and uses the global soft label data set to train a global student model. The global student model is constructed using a deep neural network model, and model parameters are compressed based on the local privacy protection teacher model and the same network framework as the local privacy protection teacher model is used.

[0108] The trained global student model is sent to each participant, and each participant uses the global student model to perform secondary protection on the privacy-protected data, calculates the soft label of the privacy-protected data through forward propagation, and uses knowledge distillation technology to perturb the soft label to obtain secondary privacy-protected data.

[0109] For example, each participant uses its own data set to train a privacy-preserving teacher model locally. The teacher model is constructed using a deep neural network model, such as a convolutional neural network (CNN) and a recurrent neural network (RNN). During the training process, a differential privacy mechanism is introduced to control the privacy budget during model training by adding Laplace noise to the model parameter gradient. The specific steps are as follows:

[0110] Define the privacy budget ε and sensitivity Δf. The privacy budget ε represents the allowed privacy leakage risk, and the sensitivity Δf represents the maximum impact of a single sample on the model output. In each iteration, the gradient value of the model parameters is calculated and the gradient value is clipped to ensure that its L2 norm does not exceed the sensitivity Δf. Laplace noise is added to the clipped gradient value, and the proportion of the noise is determined by the privacy budget ε. The standard deviation of the noise is Δf / ε. The model parameters are updated using the gradient value after adding noise.

[0111] Through the differential privacy mechanism, privacy security can be guaranteed during model training and sensitive information leakage can be prevented.

[0112] Each participant uses the trained local privacy-preserving teacher model to infer the local data set and calculates the conditional probability of each sample belonging to different categories through forward propagation. The inferred conditional probability is used as soft label data. Soft label data represents the probability distribution of each sample belonging to different categories and contains the knowledge learned by the teacher model. Each participant encrypts the soft label data using homomorphic encryption technology to protect data privacy. The Paillier homomorphic encryption algorithm can be used to encrypt soft label data.

[0113] The data security gateway aggregates soft-label data and trains the global student model;

[0114] Each participant uploads the homomorphically encrypted soft label data to the data security gateway. The data security gateway collects the soft label data uploaded by each participant and constructs a global soft label dataset. The global soft label dataset is used to train the global student model. The student model uses the same deep neural network framework as the teacher model, but the model parameters are compressed to reduce the model size. During the training process, the student model learns the knowledge of the teacher model by minimizing the cross entropy loss function between the soft label and the student model.

[0115] Each participant uses the global student model for secondary privacy protection;

[0116] The data security gateway sends the trained global student model to each participant. Each participant uses the global student model to perform secondary protection on the privacy-protected data. Through forward propagation, the soft labels of the privacy-protected data are calculated. The soft labels are perturbed using knowledge distillation technology. Specifically, the temperature parameter T is introduced to smooth the soft labels and obtain the perturbed soft labels. The temperature parameter T controls the intensity of the perturbation.

[0117] Using the perturbed soft labels as secondary privacy-preserving data can further reduce the risk of privacy leakage. Secondary privacy-preserving data can be used for various data analysis and machine learning tasks, such as classification, clustering, anomaly detection, etc., to fully utilize the value of data while protecting privacy.

[0118] Through the above detailed technical solution, federated learning and knowledge distillation technology are used to establish a privacy protection model between user devices and data security gateways to perform secondary protection on privacy data, effectively reducing the risk of privacy leakage and achieving efficient, secure, and privacy-protected distributed data processing.

[0119] In an optional embodiment,

[0120] Dividing the homomorphically encrypted private data into multiple data shards, generating a verification share for each data shard using a secret sharing algorithm, and sending the data shards and the verification share to multiple edge nodes corresponding to the user device through a blockchain network includes:

[0121] The homomorphically encrypted private data is vertically divided into multiple data shards. For each data shard, a verification share is generated using a secret sharing algorithm. A polynomial is randomly selected over a finite field so that the value of the polynomial at zero is equal to the data shard. The value of the polynomial is calculated at multiple different points to obtain multiple verification shares.

[0122] The data shards and the corresponding verification shares are sent to multiple edge nodes corresponding to the user devices using the blockchain network. The transmission of the data shards and the verification shares is achieved in the form of encrypted transactions. Each encrypted transaction contains at least one of the sender's address, the receiver's address, the data shard hash, the verification share hash, and the sender's private key's digital signature on the encrypted transaction.

[0123] Encrypted transactions are packaged into blocks and added to the blockchain through a consensus mechanism. After each edge node receives the data shard and verification share, it stores them locally.

[0124] For example, in order to further improve the security of private data during transmission, this application proposes a private data sharding transmission scheme based on homomorphic encryption and secret sharing. The scheme first uses homomorphic encryption technology to encrypt private data, then divides the encrypted private data into multiple data shards, and uses a secret sharing algorithm to generate a verification share for each data shard, and finally sends the data shards and verification shares to multiple edge nodes corresponding to the user device through the blockchain network. The specific steps are as follows:

[0125] Use homomorphic encryption technology to encrypt private data and obtain homomorphically encrypted private data. Homomorphic encryption allows calculations to be performed directly on encrypted data, and the calculation results are consistent with the results of the same calculations on plaintext data after decryption. Let the original private data be m, and the homomorphically encrypted private data be E(m), then:

[0126] E(m1+m2)=E(m1)+E(m2);

[0127] E(m1×m2)=E(m1)×E(m2);

[0128] Among them, m1 and m2 represent different private data. Commonly used homomorphic encryption algorithms include Paillier encryption and BGV encryption. Homomorphic encryption can protect the confidentiality of private data during transmission and storage.

[0129] Divide the homomorphically encrypted private data E(m) into n data shards {E(m)1, E(m)2, ..., E(m) n}, where n is the number of edge nodes. Data segmentation can be done vertically or horizontally. Vertical segmentation divides data according to feature dimensions, while horizontal segmentation divides data according to samples. Through data segmentation, private data can be stored in multiple edge nodes in a distributed manner, improving data availability and fault tolerance.

[0130] A secret sharing algorithm is used to generate a verification share for each data shard. The secret sharing algorithm allows a secret to be split into multiple shares and distributed to multiple participants. Only when enough participants combine their shares can the original secret be restored. Commonly used secret sharing algorithms include Shamir's secret sharing algorithm and Blakley's secret sharing algorithm.

[0131] Taking the Shamir secret sharing algorithm as an example, let the data slice E(m)i be the secret to be shared, and randomly select a k-1 degree polynomial f(x) such that f(0) = E(m)i, where k is the minimum number of shares required for secret recovery. Calculate the value of the polynomial at x = 1, 2, ..., n to obtain n verification shares {f(1), f(2), ..., f(n)}. Through the Lagrange interpolation formula, only any k verification shares are needed to recover the original secret E(m) i .

[0132] The data shards and corresponding verification shares are sent to multiple edge nodes corresponding to the user's device through the blockchain network. The blockchain network uses consensus mechanisms and encryption algorithms to ensure the integrity and immutability of data during transmission.

[0133] Each edge node receives a data shard and the corresponding verification share and stores them locally. When the original private data needs to be restored, the edge node submits the collected verification share to the data security gateway, which uses the secret sharing algorithm to restore the encrypted private data shards, and then combines the shards into complete encrypted private data, and finally uses the homomorphic decryption algorithm to obtain the original private data.

[0134] The private data sharding transmission scheme based on homomorphic encryption and secret sharing can provide multi-level privacy protection. First, the original private data is encrypted using homomorphic encryption technology to protect the confidentiality of the data. Even if the attacker obtains the encrypted data, the original private data cannot be restored because there is no decryption key. Secondly, the encrypted data is divided into multiple data shards and stored in multiple edge nodes. The attacker needs to break into enough edge nodes to obtain the complete encrypted data. Data sharding improves the fault tolerance and availability of the data. Finally, the secret sharing algorithm is used to generate verification shares for the data shards. The verification shares themselves do not contain any information about the private data. The encrypted data shards can only be restored when enough edge nodes submit their verification shares. Secret sharing further improves the security of the scheme and prevents a single node from leaking privacy.

[0135] In addition, the blockchain network is used to transmit data shards and verify shares. The blockchain's consensus mechanism and encryption algorithm ensure the integrity and immutability of the data transmission process, preventing data from being tampered with or forged.

[0136] In summary, the private data sharding transmission scheme based on homomorphic encryption and secret sharing provides privacy protection at multiple levels, greatly improves the security of private data transmission, and lays the foundation for privacy-preserving data analysis and machine learning tasks.

[0137] This application proposes a privacy data sharding transmission scheme based on homomorphic encryption and secret sharing to further improve the security of privacy data during transmission. The scheme first uses homomorphic encryption technology to encrypt privacy data to protect the confidentiality of the data; then the encrypted data is divided into multiple data shards to improve the availability and fault tolerance of the data; then the secret sharing algorithm is used to generate verification shares for each data shard to prevent a single node from leaking privacy; finally, the data shards and verification shares are sent to multiple edge nodes through the blockchain network to ensure the integrity and immutability of the data transmission process.

[0138] In an optional embodiment,

[0139] The method further comprises:

[0140] When an independent third party needs to obtain homomorphically encrypted private data, the independent third party sends a secret reply request to the edge node storing the homomorphically encrypted private data;

[0141] The edge node that receives the secret reply request sends the locally stored verification shares to an independent third party through the blockchain network. After the independent third party collects a predetermined number of verification shares, it restores the data shards of the encrypted private data through an interpolation formula and splices all the data shards to obtain complete encrypted private data.

[0142] For example, when an independent third party needs to obtain homomorphically encrypted private data, it sends a secret reply request to the edge node storing the data shard. The secret reply request contains the following information:

[0143] The identity of an independent third party is used to verify the legitimacy of the requester. The identity of the data shard requested specifies which data shards need to be obtained. The timestamp of the request prevents replay attacks. The digital signature of the request ensures the integrity and non-repudiation of the request.

[0144] After receiving the secret reply request, the edge node first verifies the legitimacy of the request, including the identity of the requester, the timestamp of the request, and the digital signature. If the verification is successful, the edge node sends the verification share corresponding to the requested data shard stored locally to an independent third party through the blockchain network. The edge node writes the sending record of the verification share to the blockchain to ensure the security and traceability of the sending process.

[0145] The independent third party collects the verification shares sent by the edge nodes through the blockchain network. After collecting a predetermined number of verification shares (secret sharing threshold), the independent third party uses the Lagrange interpolation formula to restore the encrypted private data shards based on the verification shares. The independent third party splices all the restored data shards according to the predetermined order and rules. The splicing of data shards can be simple sequential splicing, or fault-tolerant coding technologies such as Erasure Code can be used to improve the reliability of data.

[0146] After splicing, the independent third party obtains the complete privacy data that has been homomorphically encrypted. The independent third party can process and analyze the privacy data in an encrypted state to give full play to the value of the data. Due to the use of homomorphic encryption technology, the independent third party can directly calculate the ciphertext data without decrypting the data, such as statistical analysis, machine learning, etc. Homomorphic encryption ensures the privacy security of the data during processing and analysis, and prevents the leakage of sensitive information.

[0147] Through the above detailed technical solutions, independent third parties can safely and efficiently obtain homomorphically encrypted private data and process and analyze the data under the premise of protecting privacy. The combination of secret sharing and blockchain technology further enhances the security, reliability and traceability of the data sharing process.

[0148] In an optional embodiment,

[0149] Receiving the decrypted data slices after the multiple edge nodes decrypt the data slices using the public key and the verification share, and splicing the decrypted data slices to obtain complete decrypted data includes:

[0150] The data owner generates a decryption request, which contains the request ID, the blockchain address of the data owner, the index list of the data shards to be decrypted, the key for decrypting the verification share, and the digital signature of the data request using the data owner's private key. The decryption request is sent through the blockchain network to multiple edge nodes that store the corresponding data shards.

[0151] After receiving the decryption request, the edge node uses the public key of the data owner to verify the digital signature of the decryption request to ensure that the decryption request has not been tampered with, and checks whether the index list of the data shard in the decryption request has existed in the local storage. If the verification is successful, the corresponding verification share is read from the local storage according to the index list of each data shard, and the key pair for decrypting the verification share is used to decrypt it to obtain the decrypted verification share value. According to the verification share value and the corresponding node number, the original data shard is reconstructed;

[0152] The edge node combines the decrypted verification shares into a decrypted data shard, which contains the index of the original data shard corresponding to the decrypted data shard, the decrypted data shard content, and the digital signature of the decrypted data shard using the edge node private key, and sends the decrypted data shard to the data owner through the blockchain network;

[0153] After the data owner receives the decrypted data slice sent by the edge node, he uses the public key of the edge node to verify the digital signature of the decrypted data slice to ensure that the data has not been tampered with, and checks whether the slice index of the decrypted data slice is consistent with the index in the request. According to the slice index, all the received decrypted data slices are sequentially spliced ​​into a complete plaintext data matrix; if an error occurs in the verification process or the data is incomplete, the data owner resends the decryption request to the corresponding edge node until the complete plaintext data is obtained.

[0154] For example, in a private data shard transmission scheme based on homomorphic encryption and secret sharing, when the data owner needs to analyze and calculate the private data, he needs to receive decrypted data shards from multiple edge nodes and splice them into complete plaintext data.

[0155] The data owner first generates a decryption request, which contains the following fields:

[0156] Request ID: unique identifier of the decryption request; Data owner address: blockchain address of the data owner; Data shard index: index list of data shards that need to be decrypted; Decryption key: key used to decrypt the verification share; Digital signature: sign the request using the data owner's private key.

[0157] The data owner sends the decryption request to the edge node that stores the corresponding data shard through the blockchain network. After receiving the decryption request, the edge node first verifies the authenticity and integrity of the request. The specific steps are as follows:

[0158] Verify digital signature: Use the public key of the data owner to verify the digital signature of the request to ensure that the request has not been tampered with. Verify data ownership: Check whether the data owner address in the request is consistent with the owner address of the stored data shard. Verify data shard index: Check whether the data shard index in the request exists in the local storage. If the verification is successful, the edge node will proceed to the next step of decryption; otherwise, the edge node rejects the request.

[0159] For each data shard index in the request, the edge node reads the corresponding verification share from local storage and decrypts it using the decryption key. The decryption process of the Shamir secret sharing algorithm is as follows:

[0160] Apply the decryption function to the verification share to obtain the decrypted verification share value. Reconstruct the original data shard based on the verification share value and the corresponding node number.

[0161] The decrypted verification share value is an element on a finite field and needs to be converted into a real number or integer form for subsequent data concatenation operations.

[0162] The edge node combines the decrypted verification shares into a decrypted data shard and sends it to the data owner. The decrypted data shard contains the following fields:

[0163] Shard index: the index of the original data shard corresponding to the decrypted data shard; Shard data: the decrypted data shard content; Digital signature: use the edge node private key to sign the decrypted data shard. The decrypted data shard is sent to the data owner through the blockchain network to ensure the security and integrity of the transmission process.

[0164] After the data owner receives the decrypted data fragments sent by the edge node, he performs the following verification and splicing operations:

[0165] Verify digital signature: Use the public key of the edge node to verify the digital signature of the decrypted data shard to ensure that the data has not been tampered with. Verify shard index: Check whether the shard index of the decrypted data shard is consistent with the index in the request. Splice decrypted data shards: According to the shard index, splice all received decrypted data shards in sequence into a complete plaintext data matrix.

[0166] If an error occurs during the verification process or the data is incomplete, the data owner can resend the decryption request to the corresponding edge node until the complete plaintext data is obtained. Once the data owner obtains the complete plaintext data, it can be analyzed, mined, and machine learned. In order to protect the security of plaintext data, the data owner should take the following measures:

[0167] Access control: Strict permission control is implemented on access to plaintext data, and only authorized users are allowed to access. Data encryption: Encrypt and store plaintext data, using strong encryption algorithms and secure key management solutions. Data desensitization: Desensitize sensitive data, such as data obfuscation and data anonymization, to reduce the risk of data leakage. Data auditing: Record all access and operation logs to plaintext data, conduct regular security audits, and promptly detect and handle abnormal behaviors.

[0168] Through these data protection measures, data owners can fully utilize plaintext data while minimizing the risk of data leakage and abuse. The process of decrypting and splicing data shards involves cryptographic operations and data transmission, which may affect system performance. In order to improve the efficiency of decryption and splicing, the following optimization measures can be adopted: Edge nodes can use multi-threading or distributed computing frameworks to decrypt multiple verification shares in parallel, fully utilize computing resources, and reduce decryption latency.

[0169] When transmitting decrypted data slices, the data can be compressed to reduce the data size and save network bandwidth. At the same time, after receiving the compressed decrypted data slices, the data owner needs to perform a decompression operation to restore the original data.

[0170] Before sending a decryption request, the data owner can pre-verify the legitimacy of the request, such as checking whether the data shard index is valid, whether the decryption key is correct, etc. Pre-verification can avoid unnecessary network communication and computing overhead.

[0171] When a large number of data shards need to be decrypted and spliced, the data owner can package multiple decryption requests into a batch request and send it to the edge node at one time. After receiving the batch request, the edge node decrypts and sends multiple data shards in batches, reducing the number of communications and waiting delays.

[0172] This application details the decryption and splicing process of data shards in a privacy data sharding transmission scheme based on homomorphic encryption and secret sharing. The data owner generates a decryption request, obtains the decrypted data shards from the edge node, and splices them into complete plaintext data. During the decryption and splicing process, digital signatures and access control measures are used to ensure the security and integrity of the data.

[0173] Figure 2 FIG. 1 is a schematic diagram of the structure of a data security gateway system based on edge privacy computing according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0174] The first unit is configured to receive a data access request from a user device, the data access request including a user identifier, requested data information, and a device identifier of the user device; authenticate the user device using a distributed identity authentication protocol according to the user identifier and the device identifier, and obtain data access rights corresponding to the user identifier from a preset rights management database;

[0175] The second unit is used to perform adaptive privacy processing on the requested data information according to the data access permission, determine the privacy protection level corresponding to the data access permission, and obtain privacy protection data; use a federated learning algorithm to establish a privacy protection model between the user device and the data security gateway, and use the privacy protection model to perform secondary privacy protection on the privacy protection data to obtain secondary privacy protection data;

[0176] The third unit is used to generate a private key and a public key pair corresponding to the user device by using a distributed key generation algorithm, send the public key to the user device, and use the private key to perform homomorphic encryption on the secondary privacy protection data to obtain homomorphically encrypted privacy data; divide the homomorphically encrypted privacy data into multiple data slices, and use a secret sharing algorithm to generate a verification share for each data slice, and send the data slices and the verification share to multiple edge nodes corresponding to the user device through a blockchain network;

[0177] The fourth unit is used to receive the decrypted data slices after the multiple edge nodes decrypt the data slices using the public key and the verification share, splice the decrypted data slices to obtain complete decrypted data, and return the decrypted data to the user device.

[0178] According to a third aspect of the embodiments of the present invention,

[0179] An electronic device is provided, comprising:

[0180] processor;

[0181] a memory for storing processor-executable instructions;

[0182] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0183] According to a fourth aspect of the embodiments of the present invention,

[0184] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0185] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0186] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data security gateway method based on edge privacy computing, characterized in that: include: Receiving a data access request from a user device, the data access request including a user identifier, requested data information, and a device identifier of the user device; Authentication of the user device is performed using a distributed identity authentication protocol according to the user identification and the device identification, and obtaining data access rights corresponding to the user identification from a preset rights management database; According to the data access permission, adaptive privacy processing is performed on the requested data information, and the privacy protection level corresponding to the data access permission is determined to obtain privacy protection data; a privacy protection model is established between the user device and the data security gateway using a federated learning algorithm, and secondary privacy protection is performed on the privacy protection data using the privacy protection model to obtain secondary privacy protection data; Generate a private key and a public key pair corresponding to the user device using a distributed key generation algorithm, send the public key to the user device, and use the private key to perform homomorphic encryption on the secondary privacy protection data to obtain homomorphically encrypted privacy data; divide the homomorphically encrypted privacy data into multiple data slices, and use a secret sharing algorithm to generate a verification share for each data slice, and send the data slices and the verification share to multiple edge nodes corresponding to the user device through a blockchain network; Receive the decrypted data slices after the multiple edge nodes decrypt the data slices using the public key and the verification share, splice the decrypted data slices to obtain complete decrypted data, and return the decrypted data to the user device.

2. The method according to claim 1, characterized in that According to the data access permission, the requested data information is adaptively processed for privacy, a privacy protection level corresponding to the data access permission is determined, and the obtained privacy protection data includes: Determine the privacy protection level based on data access rights, and divide data access rights into three types: full access, partial access, and restricted access, corresponding to low, medium, and high privacy protection levels respectively; According to the privacy protection level, select a corresponding privacy protection algorithm from three privacy protection algorithms: differential privacy, homomorphic encryption, and secure multi-party computing; Processing the requested data information using the selected privacy protection algorithm to obtain privacy-protected data; When the differential privacy algorithm is selected, random noise is added to the requested data information using the Laplace mechanism to obtain privacy-protected data, where the amplitude of the random noise is inversely proportional to the preset privacy protection budget; When the homomorphic encryption algorithm is selected, the Paillier encryption scheme is used to encrypt the requested data information, and the ciphertext data is calculated to obtain the privacy-protected data. The Paillier encryption scheme supports addition and multiplication operations of the ciphertext data. When a secure multi-party computing algorithm is selected, secret sharing technology is used to divide the requested data information into multiple data slices, and the data slices are sent to multiple participants respectively. The computing task is completed through a secure protocol under the premise of privacy protection to obtain privacy-protected data.

3. The method according to claim 1, characterized in that A privacy protection model is established between the user device and the data security gateway using a federated learning algorithm, and the privacy protection model is used to perform secondary privacy protection on the privacy protection data, and the secondary privacy protection data obtained includes: Each participant uses a local data set to train a local privacy-preserving teacher model. The local privacy-preserving teacher model is built based on a deep neural network model. A differential privacy mechanism is used during the training process. Laplace noise is added to the gradient of model parameters to control the privacy budget during the training process. Each participant uses the local privacy-preserving teacher model to infer the local data set, calculates the conditional probability of each sample in the local data set belonging to the corresponding category through forward propagation, and obtains soft label data containing the learning knowledge of the local privacy-preserving teacher model, and the soft label data is the probability distribution of each sample belonging to the corresponding category; Each participant encrypts the soft label data using homomorphic encryption technology, and uploads the homomorphically encrypted soft label data to the data security gateway; The data security gateway collects the soft label data uploaded by each participant, constructs a global soft label data set, and uses the global soft label data set to train a global student model. The global student model is constructed using a deep neural network model, and model parameters are compressed based on the local privacy protection teacher model and the same network framework as the local privacy protection teacher model is used. The trained global student model is sent to each participant, and each participant uses the global student model to perform secondary protection on the privacy-protected data, calculates the soft label of the privacy-protected data through forward propagation, and uses knowledge distillation technology to perturb the soft label to obtain secondary privacy-protected data.

4. The method according to claim 1, characterized in that: Dividing the homomorphically encrypted private data into multiple data shards, generating a verification share for each data shard using a secret sharing algorithm, and sending the data shards and the verification share to multiple edge nodes corresponding to the user device through a blockchain network includes: The homomorphically encrypted private data is vertically divided into multiple data shards. For each data shard, a verification share is generated using a secret sharing algorithm. A polynomial is randomly selected over a finite field so that the value of the polynomial at zero is equal to the data shard. The value of the polynomial is calculated at multiple different points to obtain multiple verification shares. The data shards and the corresponding verification shares are sent to multiple edge nodes corresponding to the user devices using the blockchain network. The transmission of the data shards and the verification shares is achieved in the form of encrypted transactions. Each encrypted transaction contains at least one of the sender's address, the receiver's address, the data shard hash, the verification share hash, and the sender's private key's digital signature on the encrypted transaction. Encrypted transactions are packaged into blocks and added to the blockchain through a consensus mechanism. After each edge node receives the data shard and verification share, it stores them locally.

5. The method according to claim 4, characterized in that The method further comprises: When an independent third party needs to obtain homomorphically encrypted private data, the independent third party sends a secret reply request to the edge node storing the homomorphically encrypted private data; The edge node that receives the secret reply request sends the locally stored verification shares to an independent third party through the blockchain network. After the independent third party collects a predetermined number of verification shares, it restores the data shards of the encrypted private data through an interpolation formula and splices all the data shards to obtain complete encrypted private data.

6. The method according to claim 1, characterized in that Receiving the decrypted data slices after the multiple edge nodes decrypt the data slices using the public key and the verification share, and splicing the decrypted data slices to obtain complete decrypted data includes: The data owner generates a decryption request, which contains the request ID, the blockchain address of the data owner, the index list of the data shards to be decrypted, the key for decrypting the verification share, and the digital signature of the data request using the data owner's private key. The decryption request is sent through the blockchain network to multiple edge nodes that store the corresponding data shards. After receiving the decryption request, the edge node uses the public key of the data owner to verify the digital signature of the decryption request to ensure that the decryption request has not been tampered with, and checks whether the index list of the data shard in the decryption request has existed in the local storage. If the verification is successful, the corresponding verification share is read from the local storage according to the index list of each data shard, and the key pair for decrypting the verification share is used to decrypt it to obtain the decrypted verification share value. According to the verification share value and the corresponding node number, the original data shard is reconstructed; The edge node combines the decrypted verification shares into a decrypted data shard, which contains the index of the original data shard corresponding to the decrypted data shard, the decrypted data shard content, and the digital signature of the decrypted data shard using the edge node private key, and sends the decrypted data shard to the data owner through the blockchain network; After the data owner receives the decrypted data slice sent by the edge node, he uses the public key of the edge node to verify the digital signature of the decrypted data slice to ensure that the data has not been tampered with, and checks whether the slice index of the decrypted data slice is consistent with the index in the request. According to the slice index, all the received decrypted data slices are sequentially spliced ​​into a complete plaintext data matrix; if an error occurs in the verification process or the data is incomplete, the data owner resends the decryption request to the corresponding edge node until the complete plaintext data is obtained.

7. A data security gateway system based on edge privacy computing, used to implement the method described in any one of claims 1 to 6, characterized in that: include: A first unit is used to receive a data access request from a user device, where the data access request includes a user identifier, requested data information, and a device identifier of the user device; Authentication of the user device is performed using a distributed identity authentication protocol according to the user identification and the device identification, and obtaining data access rights corresponding to the user identification from a preset rights management database; The second unit is used to perform adaptive privacy processing on the requested data information according to the data access permission, determine the privacy protection level corresponding to the data access permission, and obtain privacy protection data; use a federated learning algorithm to establish a privacy protection model between the user device and the data security gateway, and use the privacy protection model to perform secondary privacy protection on the privacy protection data to obtain secondary privacy protection data; The third unit is used to generate a private key and a public key pair corresponding to the user device by using a distributed key generation algorithm, send the public key to the user device, and use the private key to perform homomorphic encryption on the secondary privacy protection data to obtain homomorphically encrypted privacy data; divide the homomorphically encrypted privacy data into multiple data slices, and use a secret sharing algorithm to generate a verification share for each data slice, and send the data slices and the verification share to multiple edge nodes corresponding to the user device through a blockchain network; The fourth unit is used to receive the decrypted data slices after the multiple edge nodes decrypt the data slices using the public key and the verification share, splice the decrypted data slices to obtain complete decrypted data, and return the decrypted data to the user device.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images