An industrial large model security protection method, system, device and medium based on a trusted secret environment

By establishing an encrypted communication channel and hardware isolation environment based on the TEE environment between the industrial gateway and the model server, the security issues of data transmission and inference operation are solved, and hardware-level trusted execution and data privacy protection are achieved.

CN123122860APending Publication Date: 2026-11-17INSPUR YUNZHOU (SHANDONG) IND INTERNET CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610990687.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-03
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

The data transmission between existing industrial gateways and model servers lacks hardware-level security, and the inference runtime environment of the model server lacks a trusted isolation mechanism, making it difficult to resist kernel-level malicious attacks and model theft.

Method used

By establishing an encrypted communication channel based on a TEE environment between the industrial gateway and the model server, using a hardware cryptographic card to encrypt and store the channel key, and running the inference process in the TEE hardware isolation environment of the model server, hardware-level trusted execution is achieved.

Benefits of technology

It effectively prevents key leakage and model theft, ensures the confidentiality and integrity of data transmission, provides a hardware-level trusted execution environment, and improves data transmission efficiency and privacy protection of inference results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123122860A_ABST
    Figure CN123122860A_ABST
Patent Text Reader

Abstract

This application discloses a method, system, device, and medium for security protection of large industrial models based on a trusted confidential environment. It mainly relates to the field of model security protection technology, aiming to solve the problems of insufficient hardware-level security for data transmission between existing industrial gateways and model servers, and the lack of trusted isolation mechanisms in the inference runtime environment of the model server. The method includes: binding remote proof results with transmitted data, transmitting the data to the TNG TEE gateway of the model server via an encrypted communication channel, verifying the encryption integrity and binding relationship of the transmitted data through the TNG TEE gateway; upon successful verification, forwarding the data to a preset inference process running in a CSV virtual machine within the TEE hardware isolation environment on the model server via an internal channel of the TEE environment to obtain the original inference result; processing the original inference result according to a preset strategy to obtain the inference conclusion; and returning the inference conclusion to the industrial gateway via an encrypted communication channel.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of model security protection technology, and in particular to a method, system, device and medium for security protection of large industrial models based on a trusted confidential environment. Background Technology

[0002] With the deepening of industrial intelligent transformation, large-scale industrial security models play a crucial role in scenarios such as equipment fault diagnosis, intrusion detection, and risk warning. These models typically need to process highly sensitive data such as industrial control commands, equipment maintenance privacy information, and core production parameters. Their model architecture and training weights also constitute core intellectual property. In actual deployments in industrial settings, the model server is often located at the edge or in the cloud. The industrial gateway needs to standardize the collected raw data before sending it to the model for inference via the industrial network, and then return the inference conclusions to the field for closed-loop control. In this process, data transmission and model inference constitute the core business chain.

[0003] However, data transmission between existing industrial gateways and model servers lacks hardware-level security. Traditional software encryption schemes rely on the operating system kernel, making them vulnerable to kernel-level malicious attacks and failing to guarantee the confidentiality and integrity of sensitive industrial data during transmission. Furthermore, the generation, storage, and exchange of keys are at risk of leakage. The inference runtime environment of the model server lacks a trusted isolation mechanism. If the inference process of large industrial models runs on a general operating system or virtualization environment, its weights are easily stolen, and the inference process may be subject to model poisoning or member inference attacks. General software isolation schemes cannot provide a hardware-level trusted execution environment, making it difficult to meet the stringent requirements of real-time performance and strong isolation in industrial scenarios. Summary of the Invention

[0004] This application provides a method, system, device, and medium for security protection of large industrial models based on a trusted confidential environment, in order to solve the problems of lack of hardware-level security for data transmission between existing industrial gateways and model servers, and lack of trusted isolation mechanisms for the inference and operation environment of model servers.

[0005] Firstly, this application provides a security protection method for large industrial models based on a trusted confidential environment, the method comprising: Raw industrial data is collected through data acquisition terminals in the industrial field; the raw industrial data is sent to the industrial gateway through the industrial bus; when the industrial gateway receives the raw industrial data, it performs local preprocessing operations on the raw industrial data to obtain standardized data; Before the industrial gateway triggers the inference service, it reads the current security level of the industrial scenario and determines the proof process of the model server. According to the proof process, it receives the remote proof report from the model server. When the remote proof result of the remote proof report is that the verification is passed, the industrial gateway establishes an encrypted communication channel with the TNG TEE gateway of the model server. The channel key of the encrypted communication channel is generated by the TEE environment and stored encrypted by a hardware cryptographic card. Using the generated session key, the industrial gateway processes the standardized data into transmission data, binds the remote verification result to the transmission data, and transmits it to the TNG TEE gateway on the model server through an encrypted communication channel. The TNG TEE gateway verifies the encryption integrity and binding relationship of the transmission data. Once the verification is successful, the data is forwarded to the preset inference process running in the CSV virtual machine in the TEE hardware isolation environment on the model server through the internal channel of the TEE environment to obtain the original inference result; after the original inference result is processed by the preset strategy, the inference conclusion is obtained; and the inference conclusion is returned to the industrial gateway through the encrypted communication channel.

[0006] In one implementation of this application, the data acquisition terminal includes at least: sensors, PLC, DCS, and SCADA; the industrial bus includes: Modbus, OPC UA, PROFINET, and CAN; the raw industrial data includes at least: equipment operating parameters, operating condition data, and control commands. After receiving raw industrial data, the industrial gateway performs local preprocessing operations to obtain standardized data, specifically including: By using a pre-set heterogeneous data conversion mapping program, heterogeneous raw industrial data can be converted into standardized data in the same format.

[0007] In one implementation of this application, before the industrial gateway triggers the inference service, the current security level of the industrial scenario is read, and the proof process of the model server is determined; according to the proof process, a remote proof report from the model server is received; when the remote proof result of the remote proof report is verification passed, the industrial gateway and the TNG TEE gateway of the model server establish an encrypted communication channel, specifically including: Read the current industrial scenario security level, and determine the model server's proof process based on the preset mapping relationship between the industrial scenario security level and the proof process; the proof process includes at least providing CSV virtual machine running status measurement values, TEE environment integrity verification results, industrial security large model running environment configuration and integrity hash value; Receive remote proof reports from the model server; the remote proof reports shall include at least: CSV virtual machine running status measurement values, TEE environment integrity verification results, and industrial security large model running environment configuration and integrity hash values; When the remote verification report meets the preset requirements, the remote verification result is determined to be that the remote verification report has passed verification; When the remote proof result is that the remote proof report has passed verification, the industrial gateway and the TNG TEE gateway of the model server establish an encrypted communication channel based on the remote proof result binding.

[0008] In one implementation of this application, the generated session key is used to process standardized data into transmission data by the industrial gateway. The remote authentication result is then bound to the transmission data and transmitted via an encrypted communication channel to the TNG TEE gateway on the model server. The TNG TEE gateway verifies the encryption integrity and binding relationship of the transmission data. Specifically, this includes: The generated session key is used to encrypt the standardized data to obtain the transmitted data; at the same time, the remote proof result is bound to the transmitted data to form an association verification relationship for a trusted data environment. When the TNG TEE gateway on the model server receives the transmitted data, it verifies the data encryption integrity and the binding relationship between the transmitted data and the remote proof.

[0009] In one implementation of this application, after verification, the data is forwarded to a preset inference process running in a CSV virtual machine within a TEE hardware isolation environment on the model server via an internal channel, to obtain the original inference result; after the original inference result undergoes preset strategy processing, an inference conclusion is obtained, specifically including: Once the verification is successful, the TNG TEE gateway directly injects the encrypted transmission data into the preset inference process running in the CSV virtual machine through the hardware-isolated internal channel. The preset inference process performs secondary processing on the received transmission data in the TEE environment to obtain format data that meets the input requirements of the preset inference model. Input the formatted data into the preset inference model and finally output the original inference result; The original inference results enter the strategic processing module inside the TEE. Based on the dynamic loading verification rules of the industrial safety level, the original inference results are filtered for confidence threshold, outlier removal and multi-source data consistency verification are performed. At the same time, deviation analysis is performed in combination with the historical operating condition baseline of the equipment, and finally the structured inference conclusion is output.

[0010] In one implementation of this application, after returning the reasoning conclusion to the industrial gateway via an encrypted communication channel, the method further includes: After receiving the inference conclusion, the industrial gateway verifies the HSM signature and data integrity. When the HSM signature and data integrity meet the preset qualification requirements, the industrial gateway performs the corresponding operation based on the reasoning conclusion and local configuration policy.

[0011] Secondly, this application provides a security protection system for large industrial models based on a trusted confidential environment, the system comprising: Data acquisition terminal, used to collect raw industrial data from industrial sites; An industrial gateway is used to acquire raw industrial data received via an industrial bus. Upon receiving the raw industrial data, the industrial gateway performs local preprocessing to obtain standardized data. Before triggering inference services, the industrial gateway reads the current industrial scenario security level and determines the model server's proof process. Based on the proof process, it receives the remote proof report from the model server. When the remote proof report indicates successful verification, the industrial gateway establishes an encrypted communication channel with the model server's TNG TEE gateway. The channel key for this encrypted communication channel is generated by the TEE environment and encrypted and stored using a hardware cryptographic card. Using the generated session key, the industrial gateway processes the standardized data into transmission data, binds the remote proof result to the transmission data, and transmits it to the model server's TNG TEE gateway via the encrypted communication channel. The model server is used to verify the encryption integrity and binding relationship of transmitted data through the TNG TEE gateway. After the verification is successful, the data is forwarded to the preset inference process running in the CSV virtual machine in the TEE hardware isolation environment on the model server through the internal channel of the TEE environment to obtain the original inference result. After the original inference result is processed by the preset strategy, the inference conclusion is obtained. The inference conclusion is returned to the industrial gateway through the encrypted communication channel.

[0012] In one implementation of this application, the industrial gateway includes a verification module. This is used to read the current security level of the industrial scenario and determine the proof process of the model server based on the preset mapping relationship between the security level of the industrial scenario and the proof process. The proof process includes at least providing CSV virtual machine running status measurement values, TEE environment integrity verification results, industrial security large model running environment configuration and integrity hash value. Receive remote proof reports from the model server; the remote proof reports shall include at least: CSV virtual machine running status measurement values, TEE environment integrity verification results, and industrial security large model running environment configuration and integrity hash values; When the remote verification report meets the preset requirements, the remote verification result is determined to be that the remote verification report has passed verification; When the remote proof result is that the remote proof report has passed verification, the industrial gateway and the TNG TEE gateway of the model server establish an encrypted communication channel based on the remote proof result binding.

[0013] Thirdly, this application provides a security protection device for a large industrial model based on a trusted confidential environment, the device comprising: processor; And a memory containing executable code, which, when executed, causes the processor to perform a security protection method for a large industrial model based on a trusted confidential environment, as described above.

[0014] Fourthly, this application provides a non-volatile computer storage medium storing computer instructions thereon, which, when executed, implement a security protection method for industrial large-scale models based on a trusted confidential environment as described above.

[0015] As can be seen from the above technical solutions, this application has the following advantages: By constructing an encrypted communication channel between the industrial gateway and the model server, based on a hardware cryptographic card for encrypted storage of the channel key, the technical shortcomings of traditional software encryption schemes—relying on the operating system kernel and being vulnerable to kernel-level malicious attacks—are fundamentally solved. The channel key for the encrypted communication channel is generated by the TEE environment of the model server and directly encrypted and stored via a hardware cryptographic card. This ensures that the entire process of key generation, storage, and exchange is completed within the hardware security boundary, eliminating the risk of key leakage in plaintext form stored in the operating system memory or disk. Before triggering the inference service, the industrial gateway must perform a remote verification process on the model server according to the current security level of the industrial scenario. Only after the remote verification report is verified is the encrypted communication channel established. This mechanism ensures the trustworthiness of the peer's operating environment and effectively prevents man-in-the-middle attacks and spoofed server access. Simultaneously, transmitted data is bound to the remote verification result before transmission, and the TNG TEE gateway verifies the encryption integrity and binding relationship, ensuring that the data is not tampered with or replayed during transmission, thus achieving the confidentiality and integrity of sensitive industrial data during transmission.

[0016] Deploying the inference process of the large industrial model within a CSV virtual machine in a TEE hardware isolation environment on the model server provides a hardware-level trusted execution environment for the inference process. Unlike software isolation solutions in general operating systems or virtualization environments, the TEE hardware isolation environment completely isolates the inference process code, model weights, and intermediate computation results from untrusted operating systems, virtual machine monitors, and other applications through processor-level security isolation mechanisms. This ensures that even if the operating system kernel is compromised, attackers cannot access or steal the model weights within the isolated area, fundamentally defending against model theft attacks.

[0017] By performing local preprocessing operations on raw industrial data through the industrial gateway, the raw data is transformed into standardized data before encrypted transmission. This reduces the bandwidth consumption of invalid and noisy data in encrypted communication, improving data transmission efficiency. In the inference result return stage, the raw inference result undergoes pre-defined strategic processing to form an inference conclusion, which is then returned to the industrial gateway via an encrypted communication channel. This process transforms the raw tensor or probability distribution of the inference output into a decision conclusion that conforms to the semantics of the industrial scenario. This avoids directly exposing potentially sensitive intermediate information hidden in the raw inference result to the transmission channel, further strengthening data privacy protection in the end-to-end process. Attached Figure Description

[0018] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a security protection method for industrial large-scale models based on a trusted confidential environment, provided in an embodiment of this application.

[0020] Figure 2 This is a schematic diagram of the internal structure of an industrial large-scale model security protection system based on a trusted confidential environment, provided in an embodiment of this application.

[0021] Figure 3 This is a schematic diagram of the internal structure of a large-scale industrial security protection device based on a trusted confidential environment, provided in an embodiment of this application. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Those skilled in the art should understand that the embodiments described below are merely preferred embodiments of this disclosure and do not imply that this disclosure can only be implemented through these preferred embodiments. These preferred embodiments are merely used to explain the technical principles of this disclosure and are not intended to limit the scope of protection of this disclosure. Based on the preferred embodiments provided by this disclosure, all other embodiments obtained by those skilled in the art without creative effort should still fall within the scope of protection of this disclosure.

[0024] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0025] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0026] The embodiment provides a security protection method for large industrial models based on a trusted confidential environment, such as... Figure 1 As shown in the embodiments of this application, the method mainly includes the following steps: Step 110: Collect raw industrial data through the data acquisition terminal in the industrial field; send the raw industrial data to the industrial gateway through the industrial bus; when the industrial gateway receives the raw industrial data, it performs local preprocessing operations on the raw industrial data to obtain standardized data.

[0027] In some embodiments, the data acquisition terminal includes at least: sensors, PLCs (programmable logic controllers), DCSs (distributed control systems), SCADA (supervisory and data acquisition systems), and other terminal devices; the industrial bus includes: Modbus, OPC UA, PROFINET, and CAN; the raw industrial data includes at least: equipment operating parameters, operating condition data, and control commands; After receiving raw industrial data, the industrial gateway performs local preprocessing operations to obtain standardized data, specifically including: By using a pre-set heterogeneous data conversion mapping program, heterogeneous raw industrial data can be converted into standardized data in the same format.

[0028] It should be further explained that the industrial gateway performs preprocessing operations on the raw industrial data sent by the acquisition terminal via the industrial bus locally. Using a preset heterogeneous data conversion mapping program, the heterogeneous raw industrial data, such as device operating parameters, operating condition data, and control commands from different terminal devices such as sensors, PLCs, DCS, and SCADA, and following various industrial bus protocols such as Modbus, OPC UA, PROFINET, and CAN, are uniformly converted into standardized data in the same format. This eliminates the need to consider the protocol differences and data format differences of the underlying devices in the subsequent encrypted transmission and model inference stages, reduces the data parsing complexity of the model server, avoids communication errors or inference deviations caused by inconsistent data formats, and ensures the consistency and availability of data before entering the secure transmission channel.

[0029] Step 120: Before the industrial gateway triggers the inference service, read the current industrial scenario security level and determine the proof process of the model server; according to the proof process, receive the remote proof report from the model server; when the remote proof result of the remote proof report is verified, the industrial gateway establishes an encrypted communication channel with the TNG TEE gateway of the model server.

[0030] The channel key for the encrypted communication channel is generated by the TEE environment and stored encrypted using a hardware cryptographic card.

[0031] This step can be specifically described as follows: Read the current industrial scenario security level, and determine the model server's proof process based on the preset mapping relationship between the industrial scenario security level and the proof process; the proof process includes at least providing CSV virtual machine running status measurement values, TEE environment integrity verification results, industrial security large model running environment configuration and integrity hash value; Receive remote proof reports from the model server; the remote proof reports shall include at least: CSV virtual machine running status measurement values, TEE environment integrity verification results, and industrial security large model running environment configuration and integrity hash values; When the remote verification report meets the preset requirements, the remote verification result is determined to be that the remote verification report has passed verification; When the remote proof result is that the remote proof report has passed verification, the industrial gateway and the TNG TEE gateway of the model server establish an encrypted communication channel based on the remote proof result binding.

[0032] It should be further explained that by reading the current security level of the industrial scenario before triggering the inference service on the industrial gateway, and determining the corresponding proof process based on the preset mapping relationship between the security level and the proof process, scenarios with different security levels can obtain remote proof content of different granularities as needed. This includes CSV virtual machine runtime status measurement values, TEE environment integrity verification results, industrial security large model runtime environment configuration and integrity hash values. This avoids unnecessary deep verification in low-security scenarios, which would waste computing resources, and also prevents security risks from being introduced due to insufficient verification in high-security scenarios. After receiving the remote proof report from the model server, the industrial gateway only determines that the remote proof verification is successful and establishes an encrypted communication channel if all the measurement values ​​and integrity verification results in the report meet the preset requirements. This mechanism ensures that the industrial gateway can confirm that the inference runtime environment of the model server is in a trustworthy state before transmitting data, that is, the CSV virtual machine has not been tampered with, the TEE environment is intact, and the large model runtime environment configuration is consistent with expectations. This effectively prevents the sending of sensitive industrial data to a compromised or forged server. The channel key for the encrypted communication channel is generated by the TEE environment and stored in encrypted form via a hardware cryptographic card. This ensures that the generation and storage of the key are completed within the hardware security boundary, eliminating the possibility of key leakage at the operating system level. At the same time, the encrypted communication channel is bound to the remote proof result, ensuring the traceability of the channel establishment process and preventing man-in-the-middle attacks such as replacing proof reports or replay attacks.

[0033] Step 130: Using the generated session key, the industrial gateway processes the standardized data into transmission data, binds the remote verification result to the transmission data, and transmits it to the TNG TEE gateway on the model server through an encrypted communication channel. The TNG TEE gateway verifies the encryption integrity and binding relationship of the transmission data.

[0034] This step can be specifically described as follows: The generated session key is used to encrypt the standardized data to obtain the transmitted data; at the same time, the remote proof result is bound to the transmitted data to form an association verification relationship for a trusted data environment. When the TNG TEE gateway on the model server receives the transmitted data, it verifies the data encryption integrity and the binding relationship between the transmitted data and the remote proof.

[0035] It should be further explained that the generated session key is used for encryption during data transmission, with the SM4 national cryptographic algorithm being the preferred encryption algorithm. Simultaneously, the remote verification result is bound to the transmitted data, forming a verification relationship for a trusted data environment. After receiving the data, the TNG TEE gateway on the model server first verifies the data encryption integrity and the binding relationship with the remote verification. Once verification is successful, the data is securely forwarded to the inference process within the CSV virtual machine via an internal TEE environment channel, bypassing the host operating system throughout the process and eliminating the risk of data interception and theft.

[0036] Step 140: After the verification is passed, the data is forwarded to the preset inference process running in the CSV virtual machine in the TEE hardware isolation environment on the model server based on the internal channel of the TEE environment to obtain the original inference result; after the original inference result is processed by the preset strategy, the inference conclusion is obtained; the inference conclusion is returned to the industrial gateway through the encrypted communication channel.

[0037] In some embodiments, after verification, the data is forwarded to a preset inference process running in a CSV virtual machine within the TEE hardware isolation environment on the model server via an internal channel of the TEE environment to obtain the original inference result; after the original inference result is processed by a preset strategy, an inference conclusion is obtained, specifically including: Once the verification is successful, the TNG TEE gateway directly injects the encrypted transmission data into the preset inference process running in the CSV virtual machine through the hardware-isolated internal channel. The preset inference process performs secondary processing on the received transmission data in the TEE environment to obtain format data that meets the input requirements of the preset inference model. Input the formatted data into the preset inference model and finally output the original inference result; The original inference results enter the strategic processing module inside the TEE. Based on the dynamic loading verification rules of the industrial safety level, the original inference results are filtered for confidence threshold, outlier removal and multi-source data consistency verification are performed. At the same time, deviation analysis is performed in combination with the historical operating condition baseline of the equipment, and finally the structured inference conclusion is output.

[0038] After returning the reasoning conclusion to the industrial gateway via an encrypted communication channel, the method also includes: After receiving the inference conclusion, the industrial gateway verifies the HSM signature and data integrity. When the HSM signature and data integrity meet the preset qualification requirements, the industrial gateway performs the corresponding operation based on the reasoning conclusion and local configuration policy.

[0039] It should be further noted that the CSV virtual machine runs entirely within a TEE hardware-isolated environment. The industrial security big data model has been decrypted collaboratively by KBS (Key Management Service) and HSM (Hardware Cryptographic Card) and loaded into the protected memory area of ​​the CSV virtual machine, achieving hardware-level isolation protection throughout the inference process. The inference process performs secondary processing on the received data within the TEE environment, including data parsing, normalization adjustment, and model input format mapping, ensuring that the data adapts to the input requirements of the industrial security big data model. Furthermore, the data during this process always exists in encrypted form within the protected memory. The industrial security big data model performs inference operations based on the input data, focusing on core industrial security scenarios to generate analysis results, specifically including equipment fault prediction, abnormal operating condition detection, process parameter out-of-bounds risk assessment, industrial control attack or abnormal behavior identification, and security situation assessment and level determination. During inference, model parameters, intermediate inference results, and feature data are all stored in the protected memory space of the TEE, and are not exposed to the host operating system, other virtual machines, or external storage devices in any plaintext form. Even if the host machine is compromised, the confidentiality and integrity of the core inference data can still be guaranteed.

[0040] After the industrial safety big data model outputs its raw inference results, the inference service performs refined processing based on preset industrial safety policies, while generating tamper-proof audit logs. Standardized audit logs are generated for key inference conclusions, including at least the model identifier and version number, inference timestamp, input data hash digest, inference conclusion digest, and CSV / TEE remote proof identifier. Finally, the HSM (Hardware Model Service) performs SM2 digital signatures on the audit logs to ensure their authenticity, integrity, and immutability, meeting industrial safety compliance audit requirements. The policy-processed inference conclusions are then returned to the industrial gateway through an established trusted security channel, completing the inference process loop. Upon receiving the inference conclusions, the industrial gateway first verifies the HSM signature and data integrity, confirming that the results originate from the trusted model service and have not been tampered with, preventing malicious results from misleading industrial field operations. Based on the inference conclusions and local configuration policies, the industrial gateway executes corresponding operations, including real-time reporting of alarm information to the superior system, triggering local control policies, and recording local security logs and alarm events, ensuring rapid implementation of inference results and improving industrial field safety response efficiency.

[0041] To further elaborate, after the confidential computing appliance starts up, the TEE chip first completes its own integrity verification, then loads the CSV virtual machine hardware module, verifies the virtual machine image signature through the hardware cryptographic card, and builds a trusted virtualization environment. Simultaneously, an initial key pair is generated; the root key is stored in the hardware cryptographic card, and the session key is held and encrypted by the TEE environment. Industrial terminal equipment uploads data through the interface module, the industrial adapter submodule parses industrial protocols and extracts sensitive information, and the encryption calculation module performs SM4 symmetric encryption on sensitive data through the hardware cryptographic card, while non-sensitive data undergoes format standardization. The industrial security large-scale model starts within the CSV virtual machine. During the training phase, the encrypted training data is decrypted only within the TEE environment, model weights are encrypted and stored using the SM2 algorithm, and the training process log is recorded and signed in real time by the security verification module. During the inference phase, input data is decrypted by the TEE and sent to the model; the inference result is encrypted and output to the industrial terminal or backend system, while simultaneously verifying data integrity. The hardware cryptographic card automatically generates a new session key every hour and synchronizes it to each CSV virtual machine through the TEE environment. The backend management platform monitors the model's running status and device hardware status in real time through a secure channel. If a tampering attack or hardware failure is detected, the security verification module immediately triggers an alarm, and the hardware password card performs key destruction and virtual machine locking operations to prevent data leakage.

[0042] In addition, this application Figure 2 This application provides an industrial large-scale model security protection system based on a trusted confidential environment. For example... Figure 2 As shown in the embodiments of this application, the system mainly includes: Data acquisition terminal 210 is used to collect raw industrial data from industrial sites.

[0043] Industrial gateway 220 is used to acquire raw industrial data received via the industrial bus. After receiving the raw industrial data, industrial gateway 220 performs local preprocessing operations on the raw industrial data to obtain standardized data. Before triggering the inference service, industrial gateway 220 reads the current industrial scenario security level and determines the proof process of model server 230. According to the proof process, it receives the remote proof report from model server 230. When the remote proof result of the remote proof report is verified, industrial gateway 220 establishes an encrypted communication channel with the TNG TEE gateway of model server 230. The channel key of the encrypted communication channel is generated by the TEE environment and encrypted and stored by a hardware cryptographic card. Using the generated session key, industrial gateway 220 processes the standardized data into transmission data, binds the remote proof result with the transmission data, and transmits it to the TNG TEE gateway of model server 230 through the encrypted communication channel.

[0044] The industrial gateway 220 includes a verification module. Used to read the current industrial scenario security level, and determine the proof process of model server 230 based on the preset mapping relationship between the industrial scenario security level and the proof process; wherein, the proof process includes at least providing CSV virtual machine running status measurement values, TEE environment integrity verification results, industrial security large model running environment configuration and integrity hash value; Receive remote verification report from model server 230; wherein, the remote verification report includes at least: CSV virtual machine running status measurement value, TEE environment integrity verification result, industrial security large model running environment configuration and integrity hash value; When the remote verification report meets the preset requirements, the remote verification result is determined to be that the remote verification report has passed verification; When the remote verification result is that the remote verification report has passed, the industrial gateway 220 and the TNG TEE gateway of the model server 230 establish an encrypted communication channel based on the remote verification result binding.

[0045] The model server 230 is used to verify the encryption integrity and binding relationship of the transmitted data through the TNG TEE gateway. After the verification is successful, the data is forwarded to the preset inference process running in the CSV virtual machine in the TEE hardware isolation environment in the model server 230 based on the internal channel of the TEE environment to obtain the original inference result. After the original inference result is processed by the preset strategy, the inference conclusion is obtained. The inference conclusion is returned to the industrial gateway 220 through the encrypted communication channel.

[0046] The above are method embodiments of this application. Based on the same inventive concept, this application also provides a security protection device for large industrial models based on a trusted confidential environment. Figure 3 As shown, the device includes: a processor; and a memory storing executable code thereon, which, when executed, causes the processor to perform a security protection method for an industrial large-scale model based on a trusted confidential environment as described in the above embodiments.

[0047] Specifically, the server collects raw industrial data through data acquisition terminals in the industrial field; it then sends the raw industrial data to the industrial gateway via an industrial bus; upon receiving the raw industrial data, the industrial gateway performs local preprocessing operations to obtain standardized data; before triggering the inference service, the industrial gateway reads the current industrial scenario security level and determines the model server's proof process; based on the proof process, it receives the remote proof report from the model server; when the remote proof result in the remote proof report is verified as passed, the industrial gateway establishes an encrypted communication channel with the model server's TNG TEE gateway; the channel key for the encrypted communication channel is generated by the TEE environment and encrypted and stored using a hardware cryptographic card; using the generated session key, the industrial gateway processes the standardized data into transmission data, binds the remote proof result to the transmission data, and transmits it to the model server's TNG TEE gateway through the encrypted communication channel, via TNG... The TEE gateway verifies the encryption integrity and binding relationship of the transmitted data. Once the verification is successful, the data is forwarded to the preset inference process running in the CSV virtual machine in the TEE hardware isolation environment on the model server through the internal channel of the TEE environment to obtain the original inference result. After the original inference result is processed by the preset strategy, the inference conclusion is obtained. The inference conclusion is returned to the industrial gateway through the encrypted communication channel.

[0048] In addition, this application embodiment also provides a non-volatile computer storage medium storing executable instructions, which, when executed, implement the above-described industrial large-scale model security protection method based on a trusted confidential environment.

[0049] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A security protection method for large industrial models based on a trusted confidential environment, characterized in that, The method includes: Raw industrial data is collected through data acquisition terminals in the industrial field; the raw industrial data is sent to the industrial gateway through the industrial bus; when the industrial gateway receives the raw industrial data, it performs local preprocessing operations on the raw industrial data to obtain standardized data; Before the industrial gateway triggers the inference service, it reads the current security level of the industrial scenario and determines the proof process of the model server. According to the proof process, it receives the remote proof report from the model server. When the remote proof result of the remote proof report is that the verification is passed, the industrial gateway establishes an encrypted communication channel with the TNG TEE gateway of the model server. The channel key of the encrypted communication channel is generated by the TEE environment and stored encrypted by a hardware cryptographic card. Using the generated session key, the industrial gateway processes the standardized data into transmission data, binds the remote verification result to the transmission data, and transmits it to the TNG TEE gateway on the model server through an encrypted communication channel. The TNG TEE gateway verifies the encryption integrity and binding relationship of the transmission data. Once the verification is successful, the data is forwarded to the preset inference process running in the CSV virtual machine in the TEE hardware isolation environment on the model server through the internal channel of the TEE environment to obtain the original inference result; after the original inference result is processed by the preset strategy, the inference conclusion is obtained; and the inference conclusion is returned to the industrial gateway through the encrypted communication channel.

2. The industrial large-scale model security protection method based on a trusted confidential environment according to claim 1, characterized in that, The data acquisition terminals include at least: sensors, PLCs, DCSs, and SCADA; industrial buses include: Modbus, OPC UA, PROFINET, and CAN; raw industrial data includes at least: equipment operating parameters, operating condition data, and control commands. After receiving the raw industrial data, the industrial gateway performs local preprocessing operations to obtain standardized data, specifically including: By using a pre-set heterogeneous data conversion mapping program, heterogeneous raw industrial data can be converted into standardized data in the same format.

3. The industrial large-scale model security protection method based on a trusted confidential environment according to claim 1, characterized in that, Before the industrial gateway triggers the inference service, it reads the current security level of the industrial scenario and determines the proof process of the model server; according to the proof process, it receives the remote proof report from the model server. When the remote verification report indicates that the remote verification has passed, the industrial gateway establishes an encrypted communication channel with the TNG TEE gateway on the model server, specifically including: Read the current industrial scenario security level, and determine the model server's proof process based on the preset mapping relationship between the industrial scenario security level and the proof process; the proof process includes at least providing CSV virtual machine running status measurement values, TEE environment integrity verification results, industrial security large model running environment configuration and integrity hash value; Receive remote proof reports from the model server; the remote proof reports shall include at least: CSV virtual machine running status measurement values, TEE environment integrity verification results, and industrial security large model running environment configuration and integrity hash values; When the remote verification report meets the preset requirements, the remote verification result is determined to be that the remote verification report has passed verification; When the remote proof result is that the remote proof report has passed verification, the industrial gateway and the TNG TEE gateway of the model server establish an encrypted communication channel based on the remote proof result binding.

4. The industrial large-scale model security protection method based on a trusted confidential environment according to claim 1, characterized in that, Using the generated session key, the industrial gateway processes standardized data into transmission data, binds the remote authentication result to the transmission data, and transmits it to the TNG TEE gateway on the model server via an encrypted communication channel. The TNG TEE gateway verifies the encryption integrity and binding relationship of the transmission data, specifically including: The generated session key is used to encrypt the standardized data to obtain the transmitted data; at the same time, the remote proof result is bound to the transmitted data to form an association verification relationship for a trusted data environment. When the TNG TEE gateway on the model server receives the transmitted data, it verifies the data encryption integrity and the binding relationship between the transmitted data and the remote proof.

5. The industrial large-scale model security protection method based on a trusted confidential environment according to claim 1, characterized in that, Once the verification is successful, the data is forwarded to the preset inference process running in the CSV virtual machine in the TEE hardware isolation environment on the model server through the internal channel of the TEE environment to obtain the original inference result. After the original reasoning results are processed using a pre-defined strategy, a reasoning conclusion is obtained, specifically including: Once the verification is successful, the TNG TEE gateway directly injects the encrypted transmission data into the preset inference process running in the CSV virtual machine through the hardware-isolated internal channel. The preset inference process performs secondary processing on the received transmission data in the TEE environment to obtain format data that meets the input requirements of the preset inference model. Input the formatted data into the preset inference model and finally output the original inference result; The original inference results enter the strategic processing module inside the TEE. Based on the dynamic loading verification rules of the industrial safety level, the original inference results are filtered for confidence threshold, outlier removal and multi-source data consistency verification are performed. At the same time, deviation analysis is performed in combination with the historical operating condition baseline of the equipment, and finally the structured inference conclusion is output.

6. The industrial large-scale model security protection method based on a trusted confidential environment according to claim 1, characterized in that, After returning the reasoning conclusion to the industrial gateway via an encrypted communication channel, the method further includes: After receiving the inference conclusion, the industrial gateway verifies the HSM signature and data integrity. When the HSM signature and data integrity meet the preset qualification requirements, the industrial gateway performs the corresponding operation based on the reasoning conclusion and local configuration policy.

7. A security protection system for a large industrial model based on a trusted confidential environment, characterized in that, The system includes: Data acquisition terminal, used to collect raw industrial data from industrial sites; An industrial gateway is used to acquire raw industrial data received via an industrial bus. Upon receiving the raw industrial data, the industrial gateway performs local preprocessing to obtain standardized data. Before triggering inference services, the industrial gateway reads the current industrial scenario security level and determines the model server's proof process. Based on the proof process, it receives the remote proof report from the model server. When the remote proof report indicates successful verification, the industrial gateway establishes an encrypted communication channel with the model server's TNG TEE gateway. The channel key for this encrypted communication channel is generated by the TEE environment and encrypted and stored using a hardware cryptographic card. Using the generated session key, the industrial gateway processes the standardized data into transmission data, binds the remote proof result to the transmission data, and transmits it to the model server's TNG TEE gateway via the encrypted communication channel. The model server is used to verify the encryption integrity and binding relationship of transmitted data through the TNG TEE gateway. After the verification is successful, the data is forwarded to the preset inference process running in the CSV virtual machine in the TEE hardware isolation environment on the model server through the internal channel of the TEE environment to obtain the original inference result. After the original inference result is processed by the preset strategy, the inference conclusion is obtained. The inference conclusion is returned to the industrial gateway through the encrypted communication channel.

8. The industrial large-scale model security protection system based on a trusted confidential environment according to claim 7, characterized in that, The industrial gateway includes a verification module. This is used to read the current security level of the industrial scenario and determine the proof process of the model server based on the preset mapping relationship between the security level of the industrial scenario and the proof process. The proof process includes at least providing CSV virtual machine running status measurement values, TEE environment integrity verification results, industrial security large model running environment configuration and integrity hash value. Receive remote proof reports from the model server; the remote proof reports shall include at least: CSV virtual machine running status measurement values, TEE environment integrity verification results, and industrial security large model running environment configuration and integrity hash values; When the remote verification report meets the preset requirements, the remote verification result is determined to be that the remote verification report has passed verification; When the remote proof result is that the remote proof report has passed verification, the industrial gateway and the TNG TEE gateway of the model server establish an encrypted communication channel based on the remote proof result binding.

9. A security protection device for a large industrial model based on a trusted confidential environment, characterized in that, The device includes: processor; And a memory having executable code stored thereon, which, when executed, causes the processor to perform a security protection method for industrial large-scale models based on a trusted confidential environment as described in any one of claims 1-6.

10. A non-volatile computer storage medium, characterized in that, It stores computer instructions, which, when executed, implement the industrial large-scale model security protection method based on a trusted confidential environment as described in any one of claims 1-6.