A multi-key homomorphic encryption authentication evaluation method for data grid environment

By combining multi-key homomorphic encryption and trusted execution environment in the data grid environment, the problem of insufficient security in the transmission and evaluation of machine learning models is solved, and the privacy protection of data and models is realized, providing a safe and efficient evaluation platform.

CN117978356BActive Publication Date: 2024-11-08ZHEJIANG NON-LINEAR DIGITAL TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410108324.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-25
Publication Date
2024-11-08
Estimated Expiration
2044-01-25

AI Technical Summary

Technical Problem

In the transmission and evaluation process of machine learning models, the problems of insufficient security of model transmission channels, high complexity of evaluation results aggregation, and incomplete encryption process of model and data features in the prior art, resulting in the inability to effectively protect the privacy of data and models.

Method used

Using a combination of multi-key homomorphic encryption (MKHE) and a trusted execution environment (TEE), data is transmitted through a secure channel to a trusted execution environment for evaluation, and using a secure comparison protocol and a secure counting protocol to ensure privacy of data and models Protect.

Benefits of technology

It realizes high privacy and security protection of machine learning data and models in an untrusted cloud environment, ensures data characteristics security and model privacy, and provides a secure and efficient multi-party data model evaluation platform.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117978356B_ABST
    Figure CN117978356B_ABST
Patent Text Reader

Abstract

The present invention relates to a multi-key homomorphic encryption authentication and evaluation method for a data grid environment. The model is initialized and deployed, the client encrypts the data features, the model provider determines the security of the execution environment, establishes a secure communication channel, completes the transmission of the model and data features, collaboratively evaluates the data features, evaluates the decision tree based on a secure comparison protocol to obtain a preliminary comparison evaluation result, aggregates the results based on a secure counting protocol, and returns the authentication result. The present invention ensures data feature security and machine learning model privacy, and is suitable for multi-party data authentication and evaluation of most machine learning models; TFF is used in combination with multi-key homomorphic encryption technology for authentication and evaluation in a data grid environment to ensure that the confidentiality of data and models is not leaked during the evaluation process, and combined with security protocols and verification mechanisms, a safe and efficient authentication and evaluation platform is provided for all parties involved, meeting the security requirements and needs of multi-party data model evaluation and authentication.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of electronic digital data processing, and in particular to a multi-key homomorphic encryption authentication and evaluation method for a data grid environment in the technical field of computer security and cryptography. Background Art

[0002] Computer security involves almost all areas of our daily lives and is widely used in many fields such as medicine, military, agriculture, and smart cities. Machine learning models are often used in computer-related data processing and prediction. Machine learning models are often regarded as key intellectual property, and model owners are often reluctant to share them. At the same time, user-side data features need to be protected during model evaluation. Security issues are the main obstacle to multi-party data collaborative evaluation and processing. Therefore, it is particularly important to study a multi-key homomorphic encryption authentication evaluation method for data grid environments to perform data feature processing evaluation while ensuring the privacy security of all parties.

[0003] In order to provide privacy for customers and model providers during machine learning inference, encryption technologies such as multi-party computation (MPC), homomorphic encryption (HE), or hybrid HE-MPC technology are usually used, but due to resource limitations and heterogeneity, existing security solutions do not fully solve all problems. Problems with existing technologies include:

[0004] (1) The security of the model transmission channel cannot be guaranteed - previous solutions only ensured the security of the evaluation environment, but did not establish a secure transmission channel and ensure the integrity of model transmission;

[0005] (2) The aggregation of evaluation process results is highly complex;

[0006] (3) The model and data feature encryption process is not perfect.

[0007] Based on this, it is necessary to propose a new security solution for the entire multi-party certification and evaluation system, which needs to meet the following requirements:

[0008] (1) Ensure the security of the model transmission channel of the model party;

[0009] (2) design new ways to evaluate results;

[0010] (3) Improve the model and data feature encryption process. Summary of the invention

[0011] The present invention solves the problems existing in the prior art and provides a multi-key homomorphic encryption authentication and evaluation method for a data grid environment, which is used for privacy-preserving machine learning between two mutually untrusting parties in an untrusted cloud setting, and provides a high degree of privacy and security protection for data and models in the data grid environment.

[0012] The technical concept of the present invention is that the multi-party data feature client can only decrypt the evaluation result ciphertext returned by its own data, and finally complete the collaborative evaluation of the data and return the aggregated result, so as to ensure the security of the data features and the privacy of the machine learning model; based on the data grid environment, it relies on the combination of multi-key homomorphic encryption (MKHE) and trusted execution environment (TEE), processes the data based on MKHE, transmits it to the trusted execution environment through a secure channel to complete the security evaluation, and finally returns the aggregated ciphertext, uses MKHE to provide confidentiality guarantee for the customer, and uses TEE to provide confidentiality guarantee for the model provider, protects the integrity of the calculation from the attack of malicious cloud opponents, designs a secure comparison protocol and a secure counting protocol, and uses the stored machine learning model to evaluate the encrypted data sent by the client.

[0013] The technical solution adopted by the present invention is a multi-key homomorphic encryption authentication evaluation method for a data grid environment, the method comprising the following steps:

[0014] S1 model initialization and deployment;

[0015] The S2 client encrypts the data features;

[0016] The S3 model provider determines the security of the execution environment, establishes a secure communication channel with the execution environment, and completes the transmission of the model and data features;

[0017] S4 performs collaborative evaluation on data features, obtains preliminary comparative evaluation results based on the safety comparison protocol evaluation decision tree, and further aggregates the results based on the safety counting protocol;

[0018] S5 authentication result is returned.

[0019] Preferably, in S1, in a data grid environment, a machine learning model is deployed and initialized based on a trusted execution environment; the trusted execution environment is an isolated execution space, and the code and data are invisible to external software;

[0020] The model provider first verifies whether the code can safely implement a certain function. After passing the model verification, the model can be deployed to a trusted execution environment. On the server hosted by the service provider, the model provider uses the code to initialize an execution environment to complete the initialization of the model.

[0021] Preferably, in S2, the client selects a multi-key homomorphic encryption scheme for the data features and generates a corresponding key pair, generates a joint key with the generated public key, and performs multiple encryption processing on the data features in combination with the Advanced Encryption Standard and partially homomorphic encryption.

[0022] Preferably, a corresponding evaluation helper element is generated in cooperation with the key pair for encrypting the second power of the private key.

[0023] Preferably, S3 comprises the following steps:

[0024] S3.1 The model provider M generates a key pair for establishing a secure transport layer protocol, obtains a certificate for its public key from a certificate authority trusted by both the model provider and the client, verifies that the execution environment is a secure execution environment through remote authentication, and establishes a secure communication channel σM with the execution environment;

[0025] S3.2 The model provider M sends the model weight W to the execution environment through σM. M issues a certificate CertES for the public key of the execution channel and sends the array (IF, W, CertM, CertES, π) to the execution environment through the secure channel σM. IF is the model code, CertM is M's certificate, and π is the proof.

[0026] S3.3 Until the data is securely transferred to the execution environment, M exits the protocol and goes offline, and proceeds to S4.

[0027] Preferably, after the data is securely transferred to the execution environment, the data is checked for integrity; the client selects a value X to encrypt, the expected output of X is Y, the client encrypts X and sends it to the execution environment for evaluation, and if the decrypted output Y' matches what the client expected, the data is intact.

[0028] Preferably, S4 comprises the following steps:

[0029] S4.1 The execution environment receives the encrypted feature vector X from the client and evaluates each decision tree in the random forest using a secure comparison protocol;

[0030] After S4.2 completes the evaluation of each decision tree in the random forest, the execution environment aggregates the number of occurrences of each unique class label using a secure counting protocol and sends the class label associated with it and its count to the client.

[0031] Preferably, the security comparison protocol is used to evaluate decision nodes of a decision tree; for each decision node, a check is made to see whether a particular feature is greater than or less than a certain threshold; and the protocol is performed on encrypted data without decrypting the data.

[0032] Preferably, after all decision nodes are evaluated, each leaf node will have a class label associated with it, and the secure counting protocol is used to calculate the count of each class label and perform majority voting to determine the final classification result.

[0033] Preferably, S5 comprises the following steps:

[0034] S5.1 Partial decryption of the result. After the evaluation is completed, the model in the trusted execution environment evaluates a result encrypted under multiple keys. Each model provider participates in the decryption process of the corresponding part and sends its own decryption component to the execution environment to convert the encrypted result into ciphertext that can be decrypted by the client's private key.

[0035] S5.2 The execution environment collects responses from all model providers and aggregates these decryption components to generate a new encryption result, which is decrypted by the client's private key; the execution environment sends the class label associated with it and its count to the client.

[0036] The present invention relates to a multi-key homomorphic encryption authentication and evaluation method for a data grid environment. The model is initialized and deployed, the client encrypts the data features, the model provider determines the security of the execution environment, establishes a secure communication channel with the execution environment, completes the transmission of the model and data features, performs collaborative evaluation on the data features, obtains preliminary comparison and evaluation results based on a security comparison protocol evaluation decision tree, further aggregates the results based on a security counting protocol, and returns the authentication results.

[0037] The beneficial effects of the present invention are:

[0038] (1) Ensure data feature security and machine learning model privacy, complete collaborative data evaluation and return aggregated results, and is suitable for multi-party data authentication and evaluation of most machine learning models;

[0039] (2) In a data grid environment, TFF is used in combination with multi-key homomorphic encryption technology to complete the authentication and evaluation. While ensuring the security of multi-party data and models, the evaluation and authentication of each party's data is completed in an encrypted state. Multi-key homomorphic encryption technology ensures that the confidentiality of data and models is not leaked during the evaluation process. It also combines security protocols and verification mechanisms to provide all parties with a secure and efficient authentication and evaluation platform that meets the security requirements and needs of multi-party data model evaluation and authentication. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flow chart of the method of the present invention;

[0041] Figure 2 It is a schematic diagram of the interaction between the client, data grid environment and model end of the present invention;

[0042] Figure 3 This is a schematic diagram of the model initialization deployment process of the present invention;

[0043] Figure 4 This is a schematic diagram of data feature encryption of the present invention;

[0044] Figure 5 This is a schematic diagram of secure data transmission according to the present invention;

[0045] Figure 6 A schematic diagram of collaborative evaluation of a data feature model of the present invention;

[0046] Figure 7 This is a schematic diagram of returning the evaluation and authentication results of the present invention. DETAILED DESCRIPTION

[0047] The present invention is further described in detail below in conjunction with embodiments, but the protection scope of the present invention is not limited thereto.

[0048] like Figure 1~2 The present invention relates to a multi-key homomorphic encryption authentication evaluation method for a data grid environment, the method comprising the following steps:

[0049] S1 model initialization and deployment;

[0050] The S2 client encrypts the data features;

[0051] The S3 model provider determines the security of the execution environment, establishes a secure communication channel with the execution environment, and completes the transmission of the model and data features;

[0052] S4 performs collaborative evaluation on data features, obtains preliminary comparative evaluation results based on the safety comparison protocol evaluation decision tree, and further aggregates the results based on the safety counting protocol;

[0053] S5 authentication result is returned.

[0054] The following describes each step in conjunction with specific content.

[0055] (1) Model initialization deployment

[0056] The "environment preparation" step ensures that the machine learning model is correctly deployed and initialized in a secure environment. This method deploys and initializes the machine learning model based on the trusted execution environment (TEE) in a data grid environment.

[0057] A trusted execution environment, such as Intel SGX, provides an isolated execution space that ensures that code and data are invisible to external software (including the operating system). This provides a secure execution environment for data and models, ensuring their privacy and security during processing. The model side is protected by the trusted execution environment protection and isolation from the interactors.

[0058] The model provider first verifies whether their code implements a certain function securely. Then the model provider uses the code to initialize a secure execution environment on a server hosted by the service provider, where the code can be executed and data can be processed to ensure that the model will not be maliciously attacked or leak information during execution, such as Figure 3 As shown in the figure, the initialization process includes loading the model's weights and parameters, setting any necessary runtime configuration, and preparing any other necessary resources for model evaluation.

[0059] The operation algorithm is described as:

[0060] Set parameters regarding model code, test data for validating model functionality, model weights, runtime configuration, and any other necessary resources;

[0061] (1-1) Model function verification: Model providers first verify whether their code implements a certain function safely. This means that model providers need to ensure that their code does not have any malicious behavior and can correctly perform the required tasks.

[0062] (1-2) After passing the model verification, the model can be deployed to the trusted execution environment, which involves loading the code into the isolated space of the data grid environment and executing the initialization process in this space;

[0063] (1-3) Model initialization: On a server hosted by the service provider, the model provider uses their code to initialize a secure execution environment where the code can be executed and data can be processed; the initialization process includes loading the model's weights and parameters, setting any necessary runtime configuration, and preparing any other necessary resources for model evaluation.

[0064] (2) Data feature encryption

[0065] The client selects the multi-key homomorphic encryption (MKHE) scheme for the data features and generates the corresponding key pair, and further uses the generated public key to encrypt its data. At the same time, a hybrid method is adopted, combining the Advanced Encryption Standard (AES) and partially homomorphic encryption (SWHE) to encrypt the data features; this means that even if the data is intercepted during transmission, the attacker cannot decrypt and access the original data because he does not have the corresponding private key. Figure 4 shown.

[0066] The operation algorithm is described as:

[0067] Set parameters for the multi-key homomorphic encryption scheme, the model owner's public key list, the data characteristics to be encrypted, the AES key, and the partially homomorphic encryption scheme;

[0068] (2-1) Key pair generation. The client uses a multi-key homomorphic encryption scheme. Each system user, such as the client and the model owner, will generate a key pair. For the client, it is (pkC, skC), and for the model owner, it is (pkMi, skMi). Both include a public key and a private key. The public key is used to encrypt data, while the private key is used to decrypt data. The corresponding evaluation assistant elements ek0C and ek0Mi are also generated to encrypt the second power of the secret keys skC and skMi. The evaluation assistant element refers to the additional data or parameters generated during the encryption process, which may be used to assist or enhance the encryption process, such as participating in key management, optimization of encryption algorithms, security assessment or other related functions. The second power of the encryption keys skC and skMi here is the mathematical operation transformation of the private key, which is equivalent to pre-encryption preprocessing.

[0069] (2-2) Generate a joint key. The n model owners execute an interactive protocol to establish a joint key pkM by combining their public keys pkM1...pkMn. The joint key is formed based on the client, which builds an array independent of each other, allowing each key element to independently contribute to the final encryption or decryption process.

[0070] (2-3) Data feature encryption: Based on the multi-key homomorphic encryption scheme, a hybrid method is used. That is, the data features are first encrypted using an efficient block cipher (AES) to generate a ciphertext, which is then encrypted using partially homomorphic encryption (SWHE). The hybrid method combines the efficiency of AES and the security of SWHE, so that the data remains secure during transmission and storage.

[0071] (3) Secure data transmission

[0072] The model provider M establishes a secure communication channel σM with the execution environment through remote authentication of the secure execution environment, completes the transmission of the model and data features through σM, further checks the data integrity after data transmission, and implements the integrity check mechanism to provide integrity assurance of the model and data features for outsourced homomorphic encryption calculations; after the model and data features are transmitted to the trusted execution environment through secure data transmission, evaluation calculations can be performed directly on the encrypted data, such as Figure 5 shown.

[0073] The operation algorithm is described as:

[0074] Set parameters for certificate authority, execution environment, secure communication channel, model code, model weights, certificate of model provider M, certificate of execution environment, attestation, client-chosen value X, expected output of X;

[0075] (3-1) Establishing a secure channel. The model provider M first generates a key pair for establishing the Transport Layer Security (TLS) protocol. M obtains the certificate of its public key from the certificate authority CA that is trusted by both the model provider and the client, and verifies through remote authentication that the execution environment ES is a secure execution environment. It then establishes a secure communication channel σM with ES. σM ensures that the communication with the execution environment is secure and that data will not be leaked during transmission.

[0076] (3-2) Data transmission, M sends the model weight W to ES through the secure channel σM, issues a certificate CertES for the public key of ES, and sends (IF, W, CertM, CertES, π) to ES through the secure channel σM, where IF is the model code, W is the model weight, CertM is M's certificate, CertES is ES's certificate, and π is the proof; once the data is securely transferred to the execution environment, the model provider M can exit the protocol and go offline;

[0077] (3-3) Data integrity check, in order to provide additional computational integrity guarantees, the client can choose to use an integrity check mechanism;

[0078] Specifically, the client selects a value X to encrypt, and the client knows the expected output Y of this value. The client encrypts this value X and sends it to ES for evaluation. If the decryption of Y matches what the client expected, then it is clear that the data path has no integrity violations.

[0079] (4) Collaborative evaluation of data features

[0080] Due to the characteristics of MKHE processing, calculations can be performed directly on encrypted data in a data grid environment without decrypting the data. First, the encrypted feature vector is received and the decision tree is evaluated based on the secure comparison protocol SecComp (Secure Comparison) to obtain a preliminary comparison evaluation result, and the result is further aggregated based on the secure counting protocol SecCount (Secure Count).

[0081] Specifically, the feasible execution environment receives the encrypted feature vector and evaluates the decision tree based on the secure comparison protocol SecComp. SecComp directly generates an encrypted bit, which is used for the branch in the decision tree evaluation. This process reduces the complexity of the round and is highly parallelized in the evaluation process to improve efficiency. Based on the preliminary comparison evaluation results, the results are further aggregated through the secure counting protocol SecCount. SecCount not only calculates the count of each class label, but also performs majority voting to determine which class label is the final classification result, such as Figure 6 shown.

[0082] The operation algorithm is described as:

[0083] Set the client's encryption feature vector, random forest model, security comparison protocol, evaluation results of each decision tree, and parameters of the security counting protocol;

[0084] (4-1) Receive the encrypted feature vector to complete the evaluation. The execution environment first receives the encrypted feature vector X from the client and evaluates each decision tree in the random forest. Each decision tree has a Boolean function on its decision node. In order to evaluate the Boolean function in the decision tree, the execution environment uses the secure comparison protocol SecComp to evaluate the decision nodes of the decision tree, allowing the execution environment to determine whether a certain feature of the feature vector is less than or equal to the threshold of the decision node without knowing the actual value, ensuring the privacy of the data during the evaluation process; the SecComp protocol allows these comparisons to be performed securely on encrypted data without decrypting the data, which means that these comparisons can be performed even if the specific threshold and decision logic of the model are unknown to the execution environment. In addition, the evaluation of SecComp can be highly parallelized to improve efficiency;

[0085] SecComp is a protocol for performing secure comparison in an encrypted environment. In machine learning or other data analysis applications, it allows participants to compare values ​​without exposing actual data values. It is particularly suitable for scenarios where data privacy needs to be protected, such as evaluating machine learning models while keeping data encrypted. In the application scenario of the present invention, the SecComp protocol is used to compare elements in encrypted feature vectors and thresholds in decision tree models. SecComp allows such comparisons to be performed on encrypted data, which means that even if the environment performing the comparison (such as a server or a system performing machine learning evaluation) cannot access the actual value of the data, it can determine whether an encrypted feature value is less than, equal to, or greater than a certain threshold.

[0086] It should be noted that although the SecComp protocol is technically used to compare numerical values, in the context of feature vectors, it can be applied to each element of the vector. For example, in each decision tree of a random forest, SecComp can be used to compare each eigenvalue of the feature vector with the threshold of the decision tree node one by one.

[0087] (4-2) Result aggregation. Once the evaluation of each decision tree in the random forest is completed, the execution environment uses the secure counting protocol SecCount to ignorantly aggregate the number of occurrences of each unique class label to ensure the privacy of the data during the counting process. Specifically, once all decision nodes are evaluated, each leaf node will have a class label associated with it. The SecCount protocol is used to calculate the count of each class label and perform a majority vote to determine which class label is the final classification result. The execution environment then sends the class label associated with it and its count to the client.

[0088] (5) Authentication result returned

[0089] In the data grid environment, the evaluation results are encrypted and returned to the client, and the client uses its private key to decrypt the results to obtain the plaintext output;

[0090] The operation algorithm is described as:

[0091] Set the client's encryption feature vector, random forest model, security comparison protocol, evaluation results of each decision tree, and parameters of the security counting protocol;

[0092] (5-1) Partial decryption of results. After the evaluation is completed, the model in the trusted execution environment will evaluate an encrypted result. This result is encrypted under multiple keys based on the characteristics of multi-key homomorphic encryption (MKHE). In order for the client to be able to decrypt this result, each model owner needs to participate in the partial decryption process; that is, each model owner will send a decryption component to the execution environment to convert the encrypted result into a ciphertext that can be decrypted by the client's private key;

[0093] (5-2) Result aggregation and return. The execution environment collects responses from all model owners and aggregates these decryption components. Then, the execution environment can generate a new encrypted result that can be decrypted by the client's private key. The execution environment sends the class label and its count associated with it to the client. These results are encrypted to ensure security during transmission. The client uses its private key to securely decrypt these results. Figure 7 shown.

[0094] The present invention also relates to a computer-readable storage medium in its application, on which is stored a multi-key homomorphic encryption authentication and evaluation program for a data grid environment, which, when executed by a processor, implements the multi-key homomorphic encryption authentication and evaluation method for a data grid environment.

[0095] The present invention also relates to a computer device in its application, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the multi-key homomorphic encryption authentication and evaluation method for a data grid environment is implemented.

[0096] The present invention relates to a system in application, comprising:

[0097] One or more model providers;

[0098] One or more clients;

[0099] A trusted execution environment, built on a data grid environment;

[0100] After verifying the security of the data grid environment, the model provider initializes and deploys the model in the trusted execution environment. The client encrypts the data features. The model provider determines the security of the execution environment, establishes a secure communication channel with the execution environment, completes the transmission of the model and data features, performs collaborative evaluation of the data features, and returns the authentication results.

[0101] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0102] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0103] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0104] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0105] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0106] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A multi-key homomorphic encryption authentication evaluation method for a data grid environment, characterized by: The method comprises the following steps: S1 model initialization and deployment; In a data grid environment, deploy and initialize a machine learning model based on a trusted execution environment (TEE), which is an isolated execution space where code and data are invisible to external software. The S2 client encrypts the data features; The client selects a multi-key homomorphic encryption scheme for the data features and generates a corresponding key pair, and generates a corresponding evaluation assistant element with the key pair to encrypt the second power of the private key; The S3 model provider determines the security of the execution environment, establishes a secure communication channel with the execution environment, and completes the transmission of the model and data features; S3 includes the following steps: S3.1 The model provider M generates a key pair for establishing a secure transport layer protocol, obtains a certificate for its public key from a certificate authority trusted by both the model provider and the client, verifies that the execution environment is a secure execution environment through remote authentication, and establishes a secure communication channel σM with the execution environment; S3.2 The model provider M sends the model weight W to the execution environment through σM. M issues a certificate CertES for the public key of the execution channel and sends the array (IF, W, CertM, CertES, π) to the execution environment through the secure channel σM. IF is the model code, CertM is M's certificate, and π is the proof. S3.3 until the data is securely transferred to the execution environment, the data is checked for integrity; M exits the protocol and goes offline, and proceeds to S4; S4 performs collaborative evaluation on the data features, obtains preliminary comparative evaluation results based on the security comparison protocol evaluation decision tree, and further aggregates the results based on the security counting protocol; S4 includes the following steps: S4.1 The execution environment receives the encrypted feature vector X from the client and evaluates each decision tree in the random forest using a secure comparison protocol; S4.2 After completing the evaluation of each decision tree in the random forest, the execution environment aggregates the number of occurrences of each unique class label using a secure counting protocol, performs a majority vote, and sends the class label associated with it and its count to the client; S5 authentication result is returned.

2. According to claim 1, a multi-key homomorphic encryption authentication and evaluation method for a data grid environment is characterized in that: In S1, the model provider first verifies whether the code can safely implement a certain function. After passing the model verification, the model can be deployed to a trusted execution environment. On the server hosted by the service provider, the model provider uses the code to initialize an execution environment to complete the initialization of the model.

3. According to claim 1, a multi-key homomorphic encryption authentication and evaluation method for a data grid environment is characterized in that: In S2, the generated public key is used to generate a joint key, and the data features are multiply encrypted by combining the advanced encryption standard and partially homomorphic encryption.

4. According to claim 1, a multi-key homomorphic encryption authentication and evaluation method for a data grid environment is characterized in that: The client selects a value X to encrypt. The expected output of X is Y. The client encrypts X and sends it to the execution environment for evaluation. If the decrypted output Y' matches what the client expected, the data is complete.

5. According to claim 1, a multi-key homomorphic encryption authentication and evaluation method for a data grid environment is characterized in that: The secure comparison protocol is used to evaluate the decision nodes of a decision tree; for each decision node, a specific feature is checked to see if it is greater than or less than a certain threshold; the protocol is performed on encrypted data without decrypting the data.

6. The multi-key homomorphic encryption authentication evaluation method for a data grid environment according to claim 1, characterized in that: After all decision nodes are evaluated, each leaf node will have a class label associated with it, and the secure counting protocol is used to calculate the count of each class label and perform majority voting to determine the final classification result.

7. The multi-key homomorphic encryption authentication evaluation method for a data grid environment according to claim 1, characterized in that: S5 includes the following steps: S5.1 Partial decryption of the result. After the evaluation is completed, the model in the trusted execution environment evaluates a result encrypted under multiple keys. Each model provider participates in the decryption process of the corresponding part and sends its own decryption component to the execution environment to convert the encrypted result into ciphertext that can be decrypted by the client's private key. S5.2 The execution environment collects responses from all model providers and aggregates these decryption components to generate a new encryption result, which is decrypted by the client's private key; the execution environment sends the class label associated with it and its count to the client.