TEE technology-based single cloud architecture intra-field security data transaction method

Abstract

The invention discloses a single cloud architecture intra-field security data transaction method based on a TEE technology, and relates to the field of security outsourcing computing. The method only relates to three main bodies of a data seller, a data buyer and a cloud proxy server, and a completely trusted third party does not need to be introduced. According to the method, a data protection service is innovatively converted into a data security processing service, in the method, a data seller does not directly sell source data, but provides the right to use the data, and a data buyer only needs to pay for a data analysis service. Specifically, the method allows a data seller to encrypt a data set, and uploads the data to a cloud in a ciphertext form for sale. The cloud proxy server serves as a provider of the data transaction platform, a server equipped with the TEE technology is deployed in the cloud proxy server, and based on a (2-2) threshold Paillier encryption system, safe data analysis service is provided for data buyers, so that safe hosting of data is realized. The designed intra-field security data transaction process comprises five stages of identity authentication between participants, system initialization, transaction data outsourcing, security model training and result returning. In addition, based on the architecture, the invention further provides a detailed ciphertext linear regression training transaction example. According to the invention, the double challenges of difficult original data copy prevention and difficult re-transaction prevention in a traditional data transaction scheme are effectively solved, and finally the data security of both the data seller and the data buyer is protected in an efficient, safe and low-cost manner.

Description

Technical Field

[0001] The present invention relates to the field of secure outsourced computing and proposes a single cloud architecture on-site secure data transaction method based on TEE technology. Background Art

[0002] With the rapid development of information technology, the data economy has become an important economic form after the agricultural economy and the industrial economy. Especially driven by technologies such as cloud computing, big data, and artificial intelligence, data, as the core asset of the digital economy, plays a vital role in promoting data transactions and facilitating the development of a data-driven economy. In the data trading market, end users with limited resources usually entrust computing-intensive tasks to data trading platforms that have the ability to efficiently process large-scale data. Therefore, data trading platforms have become an important bridge between data sellers and data buyers.

[0003] In the existing dataset trading model, buyers usually obtain complete datasets from sellers through intermediary agents. However, in the traditional data trading industry, centralized trading models often lack transparency, which easily leads to the emergence of dishonest buyers. At the same time, intermediary agents as data brokers may also have problems. Specifically, during the data exchange process, the original dataset may be secretly cached or resold without the knowledge or approval of the seller. This poses two major challenges to data trading: one is the problem of preventing private copying of original data, and the other is the difficulty of preventing secondary transactions of data.

[0004] Existing research usually uses privacy computing methods to ensure the security of the original data set during the transaction process. However, these solutions face huge communication overhead in multi-party data exchange under the cloud architecture, insufficient support for secure computing types, and difficulty in preventing the risk of collusion between untrusted users and intermediaries. There is a lack of an efficient, secure and low-cost solution.

[0005] In summary, how to make full use of the powerful computing resources in the cloud to provide a data trading method that can protect the data security of data sellers and buyers while being efficient and secure has become a technical problem that needs to be solved urgently in this field. Summary of the invention

[0006] The present invention aims to solve the problem that it is difficult to prevent the original data from being copied and traded again in traditional data trading solutions, and at the same time protect the data security of both data sellers and data buyers during the transaction process. To this end, an efficient, secure and low-cost on-site secure data trading method based on a single cloud architecture based on TEE technology is provided.

[0007] Specifically, the method allows data sellers to encrypt their data sets and upload the data to the cloud in ciphertext form for sale. As an intermediate broker, the cloud proxy server provides secure data analysis services to data buyers in need with its powerful computing resources, thereby realizing secure data hosting. The system architecture only involves three types of entities: data sellers, data buyers, and cloud proxy servers, without the need to introduce any additional fully trusted third parties. As the provider of the data trading platform, the cloud proxy server deploys a server equipped with a trusted execution environment TEE, which divides code execution into two parts: the rich execution environment REE and the trusted execution environment TEE. Therefore, as two independent computing entities, REE and TEE will collaboratively perform two-party secure computing operations based on the (2-2) threshold Paillier encryption system, and use the encrypted data set of the data seller to provide data buyers with ciphertext data processing services, such as ciphertext model training and ciphertext model reasoning.

[0008] During the entire data transaction process, neither the cloud proxy server nor the data buyer can access the data seller's source data, ensuring that the data seller's data is not resold. Although the cloud proxy server may copy the encrypted data, it can only operate on the ciphertext and cannot obtain any sensitive information due to the lack of decryption capabilities. It can be said that this data transaction method innovatively transforms data protection services into data security processing services. The data seller does not directly sell the source data, but provides the right to use the data. The data buyer only needs to pay for the data analysis service and break through the limitation of computing resources by introducing a proxy server, thereby achieving more efficient and safer data use and analysis.

[0009] Specifically, the implementation of the on-site secure data trading method includes the following steps:

[0010] Step 1, identity authentication phase: The data seller first authenticates the cloud proxy server and obtains a data transaction authorization in ciphertext. Similarly, the data buyer authenticates the cloud proxy server and obtains a data use authorization in ciphertext. Subsequently, the data seller and the data buyer submit a data transaction request and a data use request to the cloud proxy server respectively. With the cloud proxy server acting as an intermediary, the data seller can learn the identity of the data buyer.

[0011] Step 2, system initialization phase: The data seller initializes (2-2) the threshold Paillier encryption system, generates a pair of public and private keys (pk, sk), and splits the private key sk into two sets of sub-private keys (sk) through a private key splitting algorithm (such as based on the Chinese remainder theorem). 1 ,sk 2 ) and (sk 3 ,sk 4), any set of sub-private keys can complete the collaborative decryption of the ciphertext through the partial decryption algorithm PDec and the joint decryption algorithm TDec.

[0012] Subsequently, the data seller establishes a secure channel with the rich execution environment REE and the trusted execution environment TEE in the cloud proxy server, and a communication channel is also established between the REE and the TEE. 1 ) is sent to the REE through a secure channel, and the key group (pk, sk 2 ) is sent to the TEE through a secure channel. In addition, since the cloud proxy server still returns ciphertext after completing the data processing task, and the data seller needs to confirm the processing result, the department key group (pk, sk 4 ) is sent to the data buyer, ensuring that the final result can only be obtained after joint decryption between the data seller and the data buyer. This mechanism will effectively prevent the cloud proxy server from accessing sensitive data and prevent potential attack risks in untrusted networks.

[0013] Step 3, outsourcing transaction data stage: the data seller encrypts the sensitive data set to be traded by calling the encryption algorithm Enc, and uploads the encrypted ciphertext data set to the cloud proxy server for storage, so that data buyers in need can purchase the right to use this data.

[0014] Step 4, security model training phase: The data buyer first pays the cloud proxy server for the right to use the required data, and encrypts its model training requirements (such as model initial parameters) by calling the encryption algorithm Enc. Subsequently, the data buyer initiates a data usage request and uploads the ciphertext initial model to be trained. After the cloud proxy server confirms the model training requirements of the data buyer, it will use the ciphertext data set provided by the data seller to process the ciphertext initial model uploaded by the data buyer. Specifically, REE and TEE collaborate to perform two-party secure computing based on the threshold Paillier encryption system and mask technology, thereby completing the secure model training task.

[0015] Step 5, return result stage: The cloud proxy server returns the final ciphertext model result after training to the data seller for confirmation. The data seller uses the partial private key sk 3 The final model is partially decrypted using the partial decryption algorithm PDec, and then the partially decrypted ciphertext model is transmitted to the data buyer. The data buyer then uses another private key sk 4 The final decryption is completed through the joint decryption algorithm TDec to obtain the final model of the plaintext, thus completing the secure transaction.

[0016] Furthermore, the (2-2) threshold Paillier encryption system splits the private key sk into two sets of subkeys. To recover the complete private key or decrypt the original ciphertext, at least two relevant parties must work together to decrypt any set of subkeys (i.e., decryption is only possible when the threshold condition is met). Specifically, in the system architecture, REE holds part of the private key sk 1 , TEE holds part of the private key sk 2 , and the data buyer holds part of the private key sk 4 The split private key satisfies the constraint equation: sk 1 +sk 2 =0 mod 2α and sk 1 +sk 2 = 1 mod N, and sk 3 +sk 4 =0 mod 2α and sk 3 +sk 4 =1 mod N, where N is the modulus, sk=α, and α is related to the odd prime number p and the odd prime number q. Therefore, REE and TEE have joint decryption capabilities, and data buyers can only obtain the remaining decryption capabilities after the data sellers have partially decrypted.

[0017] Furthermore, the encryption algorithm Enc takes the public key pk and the plaintext data m as input, selects a random number r whose bit length is less than or equal to the private key bit length, and outputs the ciphertext c. The specific calculation process is: c←Enc(pk, m, r)=(1+m·N)·(h N mod N 2 ) r mod N 2 . Where h = -y 2β (mod N), β=(P-1)(Q-1) / (4pq), and the four parameters (P, Q, p, q) are all odd prime numbers.

[0018] Furthermore, the partial decryption algorithm PDec uses a partial private key sk i (i∈{1,2}) and a ciphertext c as input, and output a partially decrypted ciphertext C i (i∈{1,2}), the specific calculation process is:

[0019] Furthermore, the joint decryption algorithm TDec uses two partially decrypted ciphertexts C 1 and C 2 As input, output plaintext m, the specific calculation process is:

[0020] Furthermore, the security model training task may be a ciphertext linear regression, and the calculation steps are as follows:

[0021] Step 1: The data seller calls the encryption algorithm Enc to encrypt the feature data set X and the target value Y to obtain the ciphertext and The sample data set has I rows and J columns. The data buyer initializes the model parameters, including the training round epoch, the learning rate η, the bias term b, and the weight W (a total of J dimensions: w 1 , w 2 ,…,w j ), and then call the encryption algorithm Enc to encrypt and in, The ciphertext is represented as any value*. Then, the data seller The data buyer sends the encrypted data to the cloud proxy server as a training sample. Sent to the cloud proxy server as the model to be trained.

[0022] Step 2: After the cloud proxy server receives the ciphertext data from the data seller and the data buyer, it performs the following calculations in sequence according to the training requirements of linear regression gradient descent:

[0023] ⑤ The rich execution environment REE calculates the ciphertext prediction value and error value

[0024]

[0025] SMUL is a secure multiplication protocol that supports inputting two ciphertexts to obtain their product ciphertext.

[0026] ⑥ The Rich Execution Environment REE computes partial derivatives

[0027]

[0028] Furthermore, we need to and Normalization processing, that is, the goal is to get and

[0029] against REE and TEE work together to calculate, and the calculation process is as follows:

[0030] (4) The rich execution environment REE generates a random number r i , then calculate And call the encryption algorithm Enc to encrypt Then calculate and send Give the trusted execution environment TEE.

[0031] (5) The trusted execution environment TEE receives data from REE and calculates and TDec([Δb+r 1 ] 1 , [Δb+r 1 ] 2 )→Δb+r 1 , then calculate Call the encryption algorithm Enc to re-encrypt return Give the rich execution environment REE.

[0032] (6) The rich execution environment is further implemented by computing The normalized result can be obtained.

[0033] against REE and TEE work together to calculate, and the calculation process is as follows:

[0034] (4) The rich execution environment REE generates a random number r j , then calculate And call the encryption algorithm Enc to encrypt Then calculate and send Give the trusted execution environment TEE.

[0035] (5) The trusted execution environment TEE receives data from REE and calculates and TDec([Δw j +r j ] 1 , [Δw j +r j ] 2 )→Δw j +r j , then calculate Call the encryption algorithm Enc to re-encrypt return Give the rich execution environment REE.

[0036] (6) The rich execution environment is further implemented by computing The normalized result can be obtained.

[0037] ⑦ The rich execution environment REE updates the model parameters under ciphertext

[0038]

[0039] ⑧If the training round epoch is reached, exit the loop and the final result is the updated (where j∈[J]) and Otherwise, continue to repeat steps ① to ③.

[0040] Step 3: The cloud proxy server will train the final ciphertext model (where j∈[J]) and The result is first returned to the data seller for confirmation. The data seller uses the partial private key sk 3 The final model is partially decrypted by the partial decryption algorithm PDec, that is, calculation (where j∈[J]) and The partially decrypted ciphertext model is then transmitted to the data buyer.

[0041] Step 4: The data buyer uses another part of the private key sk 4 calculate (where j∈[J]) and The final decryption is completed by the joint decryption algorithm TDec, that is, TDec([w j ] 1 ,[w j ] 2 )→w j (where j∈[J]) and TDec([b] 1 , [b] 2 )→b, thus obtaining the final model of the plaintext, and completing the secure linear regression training task.

[0042] Furthermore, during the security model training process, since floating-point calculations are involved, and threshold Paillier encryption, as a typical application of homomorphic encryption, only supports calculations in the integer domain, this problem can be solved by converting floating-point numbers to fixed-point representation. In this way, all values ​​are ensured to have uniform precision, thereby providing an effective method for replacing floating-point numbers with integers for floating-point operations. Specifically, by multiplying each value by a constant scaling factor (e.g., 10 eps , where eps represents the precision level), and converts it to a larger integer. This method minimizes the loss of precision while maintaining a certain decimal precision, and only introduces negligible errors.

[0043] Furthermore, the masking technique can obtain the random value r by generating it once and performing homomorphic encryption. Combined with the collaborative decryption between the REE and the TEE, a standard blinding operation is performed without leaking the original information. Finally, by subtracting To restore the original value, it effectively supports the completion of most calculations.

[0044] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0045] 1. The present invention provides a single cloud architecture on-site secure data trading method based on TEE technology, which effectively solves the problem of preventing the original data from being copied and duplicated during the data trading process. Specifically, the present invention innovatively constructs a single cloud architecture based on TEE technology. The system architecture involves three types of entities: data sellers, data buyers, and cloud proxy servers with built-in TEE, without the need to introduce any additional fully trusted third party. Its core idea is to transform the problem of protecting the data set into the challenge of protecting the data processing process. The method is based on the (2-2) threshold Paillier encryption system. Under the designed architecture, the data seller is allowed to encrypt his data set and upload the encrypted data in ciphertext to the cloud for trading. The cloud proxy server, as an intermediate broker, uses its powerful computing resources to provide secure data analysis services to data buyers, thereby realizing secure hosting of data. During the entire data trading process, neither the cloud proxy server nor the data buyer can access the original data of the data seller, ensuring that the data seller's data will not be illegally resold. Although the cloud proxy server may copy the encrypted data, it cannot obtain any sensitive information because it can only operate ciphertext and lacks decryption capabilities, further ensuring the security of the data.

[0046] 2. The present invention proposes a ciphertext linear regression training scheme based on a threshold Paillier encryption system, which can effectively protect the privacy of data buyers and sellers during the data exchange process. Specifically, it protects the privacy of the data seller's source data while also protecting the security of the data buyer's analysis results and initial model. In this process, neither the data agent nor the data buyer can access the seller's original data, and all operations are performed at the ciphertext level to ensure that all parties can only obtain the required analysis results. In addition, thanks to the support of the trusted execution environment TEE, the scheme can ensure that the confidentiality and integrity of the data can be effectively protected even when performing sensitive data computing tasks in an untrusted computing environment. The present invention also proposes floating-point conversion operations and a one-time mask calculation process, which further expands the application field of secure computing. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 Schematic diagram of the system model of the method of the present invention.

[0048] Figure 2 The figure is a schematic diagram of the system flow of the method of the present invention. Specific implementation methods

[0050] In order to make the purpose, technical solution and advantages of the present invention more clear, the technical solution in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment 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.

[0051] like Figure 1 , Figure 2 As shown, the present invention provides a single cloud architecture on-site secure data transaction method based on TEE technology, including the following steps:

[0052] Step 1, identity authentication phase: The data seller first authenticates the cloud proxy server and obtains a data transaction authorization in ciphertext. Similarly, the data buyer authenticates the cloud proxy server and obtains a data use authorization in ciphertext. Subsequently, the data seller and the data buyer submit a data transaction request and a data use request to the cloud proxy server respectively. With the cloud proxy server acting as an intermediary, the data seller can learn the identity of the data buyer.

[0053] Step 2, system initialization phase: The data seller initializes (2-2) the threshold Paillier encryption system, generates a pair of public and private keys (pk, sk), and uses the private key splitting algorithm (i.e., satisfies the constraint equation: sk 1 +sk 2 =0 mod 2α and sk 1 +sk 2 = 1 mod N, and sk 3 +sk 4 =0 mod 2α and sk 2 +sk 4 =1 mod N, where N is the modulus, sk = α, α is related to the odd prime number p and the odd prime number q) split the private key sk into two sets of sub-private keys (sk 1 ,sk 2 ) and (sk 3 ,sk 4 ), any set of sub-private keys can complete the collaborative decryption of the ciphertext through the partial decryption algorithm PDec and the joint decryption algorithm TDec.

[0054] Subsequently, the data seller establishes a secure channel with the rich execution environment REE and the trusted execution environment TEE in the cloud proxy server, and a communication channel is also established between the REE and the TEE. 1 ) is sent to the REE through a secure channel, and the key group (pk, sk2 ) is sent to the TEE through a secure channel. In addition, since the cloud proxy server still returns ciphertext after completing the data processing task, and the data seller needs to confirm the processing result, the department key group (pk, sk 4 ) is sent to the data buyer, thereby ensuring that the final result can only be obtained after joint decryption between the data seller and the data buyer.

[0055] Step 3, outsourcing transaction data stage: Assume that in the description of this specific embodiment, the data seller encrypts the sensitive feature data set X (where the number of rows and columns of the sample data set is I and J) and the target value Y as the transaction data set by calling the encryption algorithm Enc, and sends the encrypted ciphertext data set to and Upload to the cloud proxy server for storage so that data buyers with demand can purchase the right to use this data.

[0056] Step 4, security model training phase: Assume that in the description of this specific embodiment, the data buyer pays the cloud proxy server for the right to use the required data and requires the proxy to perform the ciphertext linear regression training task. Therefore, the data buyer initializes the model parameters, including the training round epoch, the learning rate η, the bias term b, and the weight W (total J dimensions: w 1 , w 2 ,…,w j ), and then call the encryption algorithm Enc to encrypt and Subsequently, the data buyer initiates a data usage request and uploads the ciphertext initial model to be trained. After the cloud proxy server confirms the model training requirements of the data buyer, it will use the encrypted data set provided by the data seller. and The ciphertext initial model uploaded by the data buyer is trained.

[0057] Specifically, the REE and the TEE perform the following calculations in sequence based on the threshold Paillier encryption system and mask technology and according to the training requirements of linear regression gradient descent:

[0058] ① The rich execution environment REE calculates the ciphertext prediction value and error value

[0059]

[0060] SMUL is a secure multiplication protocol that supports inputting two ciphertexts to obtain the ciphertext of their product. Assume that the two ciphertexts are and The specific calculation steps are as follows:

[0061] Step 1: The rich execution environment REE randomly selects a random number r and encrypts it as Then calculate:

[0062]

[0063] in, and are two input ciphertexts involved in the calculation. Then, the rich execution environment REE will Send to the trusted execution environment TEE.

[0064] Step 2: The trusted execution environment TEE receives the calculation result of step 1 from the REE Then, perform the following calculation:

[0065]

[0066] at this time That is to return the result, that is

[0067] ② The Rich Execution Environment REE computes partial derivatives

[0068]

[0069] Furthermore, we need to and Normalization processing, that is, the goal is to get and

[0070] against REE and TEE work together to calculate, and the calculation process is as follows:

[0071] (1) The rich execution environment REE generates a random number r 1 , then calculate And call the encryption algorithm Enc to encrypt Then calculate and send Give the trusted execution environment TEE.

[0072] (2) The trusted execution environment TEE receives data from REE and calculates and TDec([Δb+r 1 ] 1 , [Δb+r 1 ] 2 )→Δb+r 1 , then calculate Call the encryption algorithm Enc to re-encrypt return Give the rich execution environment REE.

[0073] (3) The rich execution environment is implemented by computing The normalized result can be obtained.

[0074] against REE and TEE work together to calculate, and the calculation process is as follows:

[0075] (1) The rich execution environment REE generates a random number r j , then calculate And call the encryption algorithm Enc to encrypt Then calculate and send Give the trusted execution environment TEE.

[0076] (2) The trusted execution environment TEE receives data from REE and calculates and TDec([Δw j +r j ] 1 , [Δw j +r j ] 2 )→Δw j +r j , then calculate Call the encryption algorithm Enc to re-encrypt return Give the rich execution environment REE.

[0077] (3) The rich execution environment is implemented by computing The normalized result can be obtained.

[0078] ③ The rich execution environment REE updates the model parameters under ciphertext

[0079]

[0080] ④If the training round epoch is reached, exit the loop and the final result is the updated (where j∈[J]) and Otherwise, continue to repeat steps ① to ③.

[0081] Step 5: Return result stage: The cloud proxy server will train the final ciphertext model (where j∈[J]) and The result is first returned to the data seller for confirmation. The data seller uses the partial private key sk 3 The final model is partially decrypted by the partial decryption algorithm PDec, that is, calculation (where j∈[J]) and The partially decrypted ciphertext model is then transmitted to the data buyer. The data buyer then uses another part of the private key sk 4 calculate (where j∈[J]) and The final decryption is completed by the joint decryption algorithm TDec, that is, TDec([w j ] 1 ,[w j ] 2 )→w j (where j∈[J]) and TDec([b] 1 , [b] 2 )→b, thus obtaining the final model of the plaintext, and completing the secure linear regression training task.

[0082] The above implementation examples are only used to illustrate the technical solution of the present invention rather than to limit it. A person skilled in the art may modify or convert the technical solution of the present invention without departing from the spirit and scope of the present invention. The scope of protection of the present invention shall be based on the claims.

Claims

1. A single cloud architecture on-site secure data transaction method based on TEE technology, characterized in that: This method allows data sellers to encrypt their data sets and upload the data to the cloud in ciphertext form for sale. As an intermediate broker, the cloud proxy server provides secure data analysis services to data buyers in need with its powerful computing resources, thereby realizing secure data hosting. The system architecture only involves three types of entities: data sellers, data buyers, and cloud proxy servers, without the need to introduce any additional fully trusted third parties. As the provider of the data trading platform, the cloud proxy server deploys a server equipped with a trusted execution environment TEE, which divides code execution into two parts: the rich execution environment REE and the trusted execution environment TEE. Therefore, as two independent computing entities, REE and TEE will collaboratively perform two-party secure computing operations based on the (2-2) threshold Paillier encryption system, and use the encrypted data set of the data seller to provide data buyers with ciphertext data processing services, such as ciphertext model training and ciphertext model reasoning. During the entire data transaction process, neither the cloud proxy server nor the data buyer can access the data seller's source data, ensuring that the data seller's data is not resold. Although the cloud proxy server may copy the encrypted data, it can only operate on the ciphertext and cannot obtain any sensitive information due to the lack of decryption capabilities. It can be said that this data transaction method innovatively transforms data protection services into data security processing services. The data seller does not directly sell the source data, but provides the right to use the data. The data buyer only needs to pay for the data analysis service and break through the limitation of computing resources by introducing a proxy server, thereby achieving more efficient and safer data use and analysis.

2. According to claim 1, a single cloud architecture intra-venue secure data transaction method based on TEE technology is characterized in that: The specific implementation of the on-site secure data trading method includes the following steps: Step 1, identity authentication phase: The data seller first authenticates the cloud proxy server and obtains a data transaction authorization in ciphertext. Similarly, the data buyer authenticates the cloud proxy server and obtains a data use authorization in ciphertext. Subsequently, the data seller and the data buyer submit a data transaction request and a data use request to the cloud proxy server respectively. With the cloud proxy server acting as an intermediary, the data seller can learn the identity of the data buyer. Step 2, system initialization phase: The data seller initializes (2-2) the threshold Paillier encryption system, generates a pair of public and private keys (pk, sk), and splits the private key sk into two sets of sub-private keys (sk1, sk2) and (sk3, sk4) through a private key splitting algorithm (such as based on the Chinese remainder theorem), any set of which can complete the collaborative decryption of the ciphertext through the partial decryption algorithm PDec and the joint decryption algorithm TDee. Subsequently, the data seller establishes a secure channel with the rich execution environment REE and the trusted execution environment TEE in the cloud proxy server, and a communication channel is also established between the REE and the TEE. Furthermore, the data seller sends the key group (pk, sk1) to the REE through a secure channel, and sends the key group (pk, sk2) to the TEE through a secure channel. In addition, since the cloud proxy server still returns ciphertext after completing the data processing task, and the data seller needs to confirm the processing result, the department key group (pk, sk4) is sent to the data buyer to ensure that the final result must be obtained through joint decryption between the data seller and the data buyer. This mechanism will effectively prevent the cloud proxy server from accessing sensitive data and guard against potential attack risks in untrusted networks. Step 3, outsourcing transaction data stage: the data seller encrypts the sensitive data set to be traded by calling the encryption algorithm Enc, and uploads the encrypted ciphertext data set to the cloud proxy server for storage, so that data buyers in need can purchase the right to use this data. Step 4, security model training phase: The data buyer first pays the cloud proxy server for the right to use the required data, and encrypts its model training requirements (such as model initial parameters) by calling the encryption algorithm Enc. Subsequently, the data buyer initiates a data usage request and uploads the ciphertext initial model to be trained. After the cloud proxy server confirms the model training requirements of the data buyer, it will use the ciphertext data set provided by the data seller to process the ciphertext initial model uploaded by the data buyer. Specifically, REE and TEE collaborate to perform two-party secure computing based on the threshold Paillier encryption system and mask technology, thereby completing the secure model training task. Step 5, return result stage: The cloud proxy server returns the final ciphertext model result after training to the data seller for confirmation. The data seller uses the partial private key sk3 to partially decrypt the final model through the partial decryption algorithm PDec, and then transmits the partially decrypted ciphertext model to the data buyer. The data buyer then uses another partial private key sk4 to complete the final decryption through the joint decryption algorithm TDec, thereby obtaining the final model of the plaintext, and thus completing the secure transaction.

3. According to claim 2, a single cloud architecture on-site secure data transaction method based on TEE technology is characterized in that: The (2-2) threshold Paillier encryption system splits the private key sk into two sets of subkeys. Any set of subkeys must be decrypted by at least two relevant parties in order to restore the complete private key or decrypt the original ciphertext (that is, it can only be decrypted when the threshold condition is met). Specifically, in the system architecture, REE holds a partial private key sk1, TEE holds a partial private key sk2, and the data buyer holds a partial private key sk4. The split partial private keys satisfy the constraint equations: sk1+sk2=0 mod2α and sk1+sk2=1 mod N, as well as sk3+sk4=0 mod 2α and sk3+sk4=1 mod N, where N is the modulus, sk=α, and α is related to the odd prime number p and the odd prime number q. As a result, REE and TEE have joint decryption capabilities, and the data buyer can only obtain the decryption capability of the remaining part after the data seller performs partial decryption.

4. According to claim 2, a single cloud architecture on-site secure data transaction method based on TEE technology is characterized in that: The encryption algorithm Enc takes a public key pk and a plaintext data m as input, selects a random number r whose bit length is less than or equal to the private key bit length, and outputs a ciphertext c. The specific calculation process is: c←Enc(pk, m, r)=(1+m·N)·(h N mod N 2 ) r mod N 2 . Where h = -y 2β (mod N), β=(P-1)(Q-1) / (4pq), and the four parameters (P, Q, p, q) are all odd prime numbers.

5. According to claim 2, a single cloud architecture on-site secure data transaction method based on TEE technology is characterized in that: The partial decryption algorithm PDec uses a partial private key sk i (i∈{1,2}) and a ciphertext c as input, and output a partially decrypted ciphertext C i (i∈{1,2}), the specific calculation process is:

6. According to claim 2, a single cloud architecture on-site secure data transaction method based on TEE technology is characterized in that: The joint decryption algorithm TDec takes two partially decrypted ciphertexts C1 and C2 as input and outputs a plaintext m. The specific calculation process is:

7. According to claim 2, a single cloud architecture intra-venue secure data transaction method based on TEE technology is characterized in that: The security model training task may be a ciphertext linear regression, and the calculation steps are as follows: Step 1: The data seller calls the encryption algorithm Enc to encrypt the feature data set X and the target value Y to obtain the ciphertext and The sample data set has I rows and J columns. The data buyer initializes the model parameters, including the training round epoch, the learning rate η, the bias term b, and the weight W (a total of J dimensions: w1, w2, ..., w j ), and then call the encryption algorithm Enc to encrypt and in, The ciphertext is represented as any value*. Then, the data seller The data buyer sends the encrypted data to the cloud proxy server as a training sample. Sent to the cloud proxy server as the model to be trained. Step 2: After the cloud proxy server receives the ciphertext data from the data seller and the data buyer, it performs the following calculations in sequence according to the training requirements of linear regression gradient descent: ① The rich execution environment REE calculates the ciphertext prediction value and error value SMUL is a secure multiplication protocol that supports inputting two ciphertexts to obtain their product ciphertext. ② The Rich Execution Environment REE computes partial derivatives Furthermore, we need to and Normalization processing, that is, the goal is to get and against REE and TEE work together to calculate, and the calculation process is as follows: (1) The rich execution environment REE generates a random number r1 and then calculates And call the encryption algorithm Enc to encrypt Then calculate and send Give the trusted execution environment TEE. (2) The trusted execution environment TEE receives data from REE and calculates and TDec([Δb+r1]1,[Δb+r1]2)→Δb+r1, then calculate Call the encryption algorithm Enc to re-encrypt return Give the rich execution environment REE. (3) The rich execution environment is implemented by computing The normalized result can be obtained. against REE and TEE work together to calculate, and the calculation process is as follows: (1) The rich execution environment REE generates a random number r j , then calculate And call the encryption algorithm Enc to encrypt Then calculate and send Give the trusted execution environment TEE. (2) The trusted execution environment TEE receives data from REE and calculates and TDec([Δw j +r j ]1,[Δw j +r j ]2)→Δw j +r j , then calculate Call the encryption algorithm Enc to re-encrypt return Give the rich execution environment REE. (3) The rich execution environment is implemented by computing The normalized result can be obtained. ③ The rich execution environment REE updates the model parameters under ciphertext ④If the training round epoch is reached, exit the loop and the final result is the updated (where j∈[J]) and Otherwise, continue to repeat steps ① to ③. Step 3: The cloud proxy server will train the final ciphertext model (where j∈[J]) and The result is first returned to the data seller for confirmation. The data seller uses the partial private key sk3 to partially decrypt the final model through the partial decryption algorithm PDec, that is, calculate (where j∈[J]) and The partially decrypted ciphertext model is then transmitted to the data buyer. Step 4: The data buyer uses another part of the private key sk4 to calculate (where j∈[J]) and The final decryption is completed by the joint decryption algorithm TDec, that is, TDec([w j ]1,[w j ]2)→w j (where j∈[J]) and TDec([b]1, [b]2)→b, thus obtaining the final model of the plaintext, thus completing the secure linear regression training task.

8. According to claim 2, a single cloud architecture on-site secure data transaction method based on TEE technology is characterized in that: During the security model training process, since floating-point calculations are involved, and threshold Paillier encryption, as a typical application of homomorphic encryption, only supports calculations in the integer domain, this problem can be solved by converting floating-point numbers to fixed-point representation. In this way, all values ​​are ensured to have uniform precision, thereby providing an effective method for replacing floating-point numbers with integers for floating-point operations. Specifically, by multiplying each value by a constant scaling factor (e.g., 10 eps , where eps represents the precision level), and converts it to a larger integer. This method minimizes the loss of precision while maintaining a certain decimal precision, and only introduces negligible errors.

9. According to claim 2, a single cloud architecture intra-venue secure data transaction method based on TEE technology is characterized in that: The masking technique can obtain the random value r by generating it once and performing homomorphic encryption Combined with the collaborative decryption between the REE and the TEE, a standard blinding operation is performed without leaking the original information. Finally, by subtracting To restore the original value, it effectively supports the completion of most calculations.

Images