Model training method, device and system and medium

Data processing through interface bus is solved by acquiring and calibration of data, and the problems of high labor costs and poor scalability in traditional industrial AI models are achieved, and the standardization and cost reduction of interface data are achieved.

CN119005357APending Publication Date: 2024-11-22CELLAUTO BIOLOGICAL AUTOMATION CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411085152.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Traditional industrial AI model training requires a lot of manpower to label data and high-cost deployment, and poor scalability in existing software systems, resulting in high interface switching costs.

Method used

The input data is obtained through the interface bus, transmitted to the calculation model for calculation and calibration processing, and feedback the calibration results to improve the standardization of the interface data and reduce the cost of interface switching.

Benefits of technology

It improves the standardization of interface data, reduces the cost of interface switching, supports standardization and standardization of model training, and is easy to expand.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119005357A_ABST
    Figure CN119005357A_ABST
Patent Text Reader

Abstract

The invention provides a model training method, device and system and a medium, and the method comprises the steps: obtaining input data through an interface bus, and transmitting the input data to a calculation model; the calculation model receives input data and calculates the input data to obtain output data; performing calibration processing on the output data to obtain calibration data, and feeding back a calibration result to the interface bus; and the interface bus feeds back calibration data to the AI model for training according to the calibration result. According to the invention, the input data is transmitted to the calculation model for calculation through the interface bus to obtain the output data, the output data is calibrated, and the calibrated data is fed back to the model for training, so that the standardization of the interface data is improved, and the cost of interface switching is reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular to a model training method, device, system and medium. Background Art

[0002] Industrial AI at the traditional industry and enterprise levels is generally based on existing batch data for centralized training. After the training effect is achieved, a unified deployment implementation model is deployed. During batch training, a large amount of training data needs to be labeled and batch training feedback needs to be given, which has a high labor cost. When the AI ​​model is deployed to the production environment for model calling after successful training, the business logic of the original program needs to be modified, which involves the upgrade and deployment of the production environment, and the switching cost is high. When there are many trained models cut into the existing software system, there are problems with standardization, which is not easy to expand. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, an object of the present invention is to provide a model training method that can improve the standardization of interface data and reduce the cost of interface switching.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A first aspect of the present invention provides a model training method, comprising the following steps:

[0006] The interface bus acquires input data and transmits the input data to the calculation model;

[0007] The computing model receives input data and performs calculations to obtain output data;

[0008] Performing calibration processing on the output data to obtain calibration data, and feeding back the calibration result to the interface bus;

[0009] The interface bus feeds back the calibration data to the AI ​​model for training based on the calibration results.

[0010] In one embodiment, the interface bus acquires input data and transmits the input data to the computing model, including:

[0011] Set interface specifications and switch mechanisms to build an interface bus;

[0012] Classify the transmitted data and set index codes for each data type;

[0013] The input data binding index is encoded according to the data type.

[0014] In one embodiment, the interface bus acquires input data and transmits the input data to the computing model, including:

[0015] Encrypting and identifying the input data according to the recognition algorithm, and determining whether the input data conforms to the recognition algorithm;

[0016] When the input data conforms to the recognition algorithm, the interface is allowed to access the bus;

[0017] When the input data does not conform to the recognition algorithm, the interface is not allowed to access the bus.

[0018] In one embodiment, when the input data meets the recognition algorithm, allowing the interface to access the interface bus includes:

[0019] When the input data conforms to the recognition algorithm, the input data is encrypted and transmitted;

[0020] The data weight of the input data is calculated, and the data transmission order is determined according to the data weight and the interface specification.

[0021] In one embodiment, the step of calibrating the output data to obtain calibration data and feeding back the calibration result to the interface bus includes:

[0022] If the output data needs calibration processing, the output data is corrected to obtain calibration data and output the calibration result;

[0023] If the output data does not require calibration processing, a feedback result is output.

[0024] In one embodiment, the transmitting the calibration result to the interface bus, and the interface bus feeding back the calibration data to the model for training according to the calibration result, comprises:

[0025] Encoding the calibration data, and transmitting the encoded calibration data to an interface bus;

[0026] The interface bus feeds the calibration data back to the AI ​​model for training.

[0027] In one embodiment, encoding the calibration data and transmitting the encoded calibration data to an interface bus comprises:

[0028] After receiving the encoded calibration data, the state conversion unit transmits the calibration data to the corresponding state processing unit;

[0029] The state processing unit integrates the calibration data according to the state table and the interface specification to generate environmental feedback data;

[0030] The state conversion unit transmits the environmental feedback data to an interface bus.

[0031] A model training device, comprising:

[0032] A data acquisition module, used for acquiring input data through an interface bus, and transmitting the input data to a calculation model;

[0033] A data calculation module, used for the calculation model to receive input data and calculate to obtain output data;

[0034] A calibration processing module, used for performing calibration processing on the output data to obtain calibration data, and feeding back the calibration result to the interface bus;

[0035] The model training module is used to transmit the calibration result to the interface bus, and the interface bus feeds back the calibration data to the model for training according to the calibration result.

[0036] A model training system, the system comprising at least one processor; and

[0037] a memory communicatively connected to the at least one processor; wherein,

[0038] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the model training method.

[0039] A non-volatile computer-readable storage medium stores computer-executable instructions, which, when executed by one or more processors, enable the one or more processors to execute the model training method.

[0040] Beneficial Effects : The embodiments of the present invention disclose a model training method, device, system and medium. Compared with the prior art, the embodiments of the present invention transmit input data to a computing model for calculation through an interface bus to obtain output data, calibrate the output data, and feed the calibrated data back to the model for training, thereby improving the standardization of interface data and reducing the cost of interface switching. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0042] Figure 1 A flowchart of a model training method provided by an embodiment of the present invention;

[0043] Figure 2 A flowchart of the implementation of the model training method provided by the embodiment of the present invention;

[0044] Figure 3 A flowchart of step S100 of the model training method provided in an embodiment of the present invention;

[0045] Figure 4 A data recognition flow chart of step S100 of the model training method provided in an embodiment of the present invention;

[0046] Figure 5 A schematic diagram of the functional modules of a model training device provided by an embodiment of the present invention;

[0047] Figure 6 A schematic diagram of the hardware structure of the model training system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the technical problems, technical solutions and beneficial effects to be solved by the embodiments of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0049] See also Figure 1 , Figure 1 This is a flow chart of an embodiment of the model training method provided by the present invention, and the method specifically comprises the following steps:

[0050] S100, the interface bus acquires input data, and transmits the input data to the calculation model;

[0051] S200, the calculation model receives input data and calculates to obtain output data;

[0052] S300, calibrating the output data to obtain calibration data, and feeding back the calibration result to the interface bus;

[0053] S400, the interface bus feeds back the calibration data to the AI ​​model for training according to the calibration result.

[0054] In this embodiment, Figure 2As shown, when the interface is connected to the bus, the input data of the interface accessing the interface bus is obtained. The interface bus transmits the input data to the computing model (traditional computing and / or AI computing), and the traditional computing and / or AI computing receives the input data and performs calculations to obtain output data. The output data is transmitted to the data presentation and subsequent processing program (also known as the data presentation and subsequent processing module) through the interface bus for calibration processing, and the calibration data is obtained and the calibration result is fed back. The calibration result is transmitted to the interface bus, and the interface bus feeds back the calibration data to the model for training according to the calibration result, and repeats until the model effect meets the requirements. When the AI ​​model training is completed, the input switch of the traditional computing can be turned off, and the output switch of the corresponding AI model can be turned on, that is, the iterative upgrade of the computing is completed. Among them, all traditional computing that needs or potentially needs to be upgraded or modified by the algorithm is connected to the bus according to the interface specification and the data interface is formed. The AI ​​training model or other computing model that is to be upgraded to replace the traditional computing is connected to the bus according to the interface specification and the data interface is formed. All AI models are cut into the applied software system, and only need to be standardized and standardized with the interface bus, which is easy to expand.

[0055] In one embodiment, the step S100 includes:

[0056] Set interface specifications and switch mechanisms to build an interface bus;

[0057] Classify the transmitted data and set index codes for each data type;

[0058] The input data binding index is encoded according to the data type.

[0059] In this embodiment, Figure 3 As shown, the interface bus architecture can be defined first, the basic architecture of the interface bus can be determined, including message passing, routing, conversion and protocol support, interface specifications can be formed, unified interface specifications can be designed, including request and response formats, data types, encoding methods, etc., an interface access mechanism can be implemented, and system administrators or developers can control which interfaces can access the bus.

[0060] The present invention implements a switch mechanism for each access interface to control the inflow and outflow of data, binds the input data to the corresponding index code according to the data type, ensures that the interface can identify and process the data according to the index code, and implements a data subscription mechanism, allowing the system or component to subscribe to the data of interest according to the index code. The input data is encrypted and identified according to a preset encryption identification mechanism, which can be achieved through a configuration file, a database or a management interface. The present invention classifies the input and output data for easy management and indexing. This embodiment designs a unique index code for each data category to build a powerful, flexible and easy-to-manage interface bus system, which can not only control the flow of data, but also ensure the security of the data and the scalability of the system.

[0061] In one embodiment, the step S100 includes:

[0062] Encrypting and identifying the input data according to the recognition algorithm, and determining whether the input data conforms to the recognition algorithm;

[0063] When the input data conforms to the recognition algorithm, the interface is allowed to access the bus;

[0064] When the input data does not conform to the recognition algorithm, the interface is not allowed to access the bus.

[0065] In this embodiment, considering that the bus may need to be accessed remotely or across network segments, a token-based identity authentication mechanism is implemented to ensure that only clients holding valid tokens can access the bus. The token can be a randomly generated string with time validity and distributed through a secure channel. A secure token generation algorithm is designed to ensure the randomness and unpredictability of the token, and a token verification mechanism is implemented to verify the validity of the token when the access request is made. An encrypted token identification mechanism is provided for bus access, and access that meets the token requirements is supported to ensure the security of the interface bus when accessed remotely or across network segments.

[0066] In one embodiment, when the input data meets the recognition algorithm, the interface is allowed to access the interface bus, including:

[0067] When the input data conforms to the recognition algorithm, the input data is encrypted and transmitted;

[0068] The data weight of the input data is calculated, and the data transmission order is determined according to the data weight and the interface specification.

[0069] In this embodiment, Figure 4As shown, when the input data meets the preset encryption identification mechanism, the input data is encrypted and transmitted, and the data connected to the bus can be encrypted and transmitted optionally, and the transmitted data is encrypted to ensure the security of the data during transmission. In this embodiment, a symmetric encryption algorithm (such as AES) or an asymmetric encryption algorithm (such as RSA) can be selected, and the data can be summarized using algorithms such as MD5 to generate summary data, and the summary data can be transmitted together with the data. The receiver can verify the integrity of the data by comparing the summary. When encrypting, the data is divided into a plain text part and a cipher text part. The plain text part includes "data number", "data type", "routing target address", "weight information", etc. The cipher text part includes the encrypted data body, and the plain text data and the encrypted cipher text data are respectively generated. MD5 summaries are generated, and the two summaries are merged to generate an encrypted summary data. By designing a unified data transmission format, including a plain text part, a cipher text part and encrypted summary data, the consistency and resolvability of the data format are ensured. This embodiment supports security protocols such as HTTPS to ensure the encryption and integrity of the data during transmission, and can consider using protocols such as TLS / SSL to encrypt data transmission to avoid tampering during the data transmission process, that is, to maintain integrity.

[0070] According to the interface specification, according to the importance and urgency of the data, the data transmission weight or priority of the transmission can be defined, and a multi-level weight (for example, 1-5 levels) can be defined to clarify the meaning and usage scenario of each weight level. The transmitted data is classified and the corresponding weight is assigned to each type of data. The weight can be determined according to the attributes such as data type, source, and purpose. The interface specification is updated to clarify the definition and use of the data transmission weight, and ensure that all accessed interfaces follow this specification. In the data transmission process, when the weight level of each data packet is clearly identified in the present invention, it can be achieved by adding a weight field in the header or metadata of the data packet, thereby realizing a transmission queue management system, sorting the data according to the weight of the data, giving priority to the data with higher weight, and ensuring its timely transmission. By designing an efficient transmission scheduling algorithm, the transmission order is determined according to the weight of the data. In this embodiment, the use of priority queues or weight polling algorithms can be considered to ensure that the data is transmitted in the bus with different priorities. After receiving all the data to be transmitted, the bus transmits the data in an orderly manner according to the transmission weight, ensuring that the data transmission in the interface bus is both efficient and orderly, while meeting the priority requirements of different data.

[0071] In one embodiment, the step S300 includes:

[0072] If the output data needs calibration processing, the output data is corrected to obtain calibration data and output the calibration result;

[0073] If the output data does not require calibration processing, a feedback result is output.

[0074] In this embodiment, whether it is traditional calculation or AI calculation, when the output data is presented to the end user or a secondary calculation is performed, the data is calibrated and fed back manually or programmatically, and the purpose and process of the data calibration feedback are clarified, and the triggering conditions and processing methods of the calibration feedback are determined to distinguish between manual calibration and programmed calibration. The calibration feedback can be manually editing and submitting new data on the page, while the programmed calibration involves a predefined program to automatically correct the data. Develop a predefined program to automatically correct the data according to established rules, and then submit it to the system, and detect whether the output data needs calibration processing. If the input data needs calibration processing, the output data is corrected to obtain the calibration data and output the calibration result. If the output data does not need calibration processing, the feedback does not require calibration results. If it is manually processed, that is, the output data is passed to the system without correcting the output data, ensuring that the program can handle various data anomalies and errors.

[0075] In one embodiment, the step S400 includes:

[0076] Encoding the calibration data, and transmitting the encoded calibration data to an interface bus;

[0077] The interface bus feeds the calibration data back to the AI ​​model for training.

[0078] In this embodiment, a coding system is designed to transmit calibration feedback data and encode the calibration data. The encoded calibration data is transmitted to the interface bus. The code should contain information such as data identification, calibration type, and calibration results, so as to realize the transmission mechanism of data calibration feedback and ensure that the data can be safely and reliably transmitted to the system. The processing logic of data calibration feedback is implemented in the system. According to the feedback type (manual or programmed) and calibration results, the data in the system is updated, and the detailed information of each data calibration feedback is recorded, including time, user, calibration type, etc., to protect the security of user data and system data.

[0079] Specifically, when traditional calculations and AI calculations of the same calculation target (which may be the calculation outputs of multiple models) are simultaneously output to the data presentation and subsequent processing program through the bus, the program can process multiple candidate calculation data, receive calibration data feedback generated manually or in a programmed manner, and thereby correct the multiple candidate calculation data. The correction results are encoded and transmitted to the bus through the state conversion unit, and then fed back to the training model through the bus to complete one training of the AI ​​model. When the model training is completed, the input switch of the traditional calculation can be turned off, and the output switch of the corresponding AI model can be turned on, thus completing the iterative upgrade of the calculation. Based on this method, it not only supports the iterative upgrade of traditional calculations to AI models, but also supports the iterative upgrade of any calculation to non-AI models.

[0080] Furthermore, reinforcement models, such as deep reinforcement models, can be trained and executed at the same time, so that the desired effect can be achieved without redeploying the training result model. Moreover, they are in a long-term training state and can better adapt to changes in the environment. For non-reinforcement models, it is best to deploy the model after training for commercial applications.

[0081] In one embodiment, encoding the calibration data and transmitting the encoded calibration data to an interface bus comprises:

[0082] After receiving the encoded calibration data, the state conversion unit transmits the calibration data to the corresponding state processing unit;

[0083] The state processing unit integrates the calibration data according to the state table and the interface specification to generate environmental feedback data;

[0084] The state conversion unit transmits the environmental feedback data to an interface bus.

[0085] In this embodiment, the state conversion unit can also be connected to the interface bus to further normalize and standardize the data interface management. Among them, the state conversion unit includes different state subsets and state processing units, and defines different state subsets. Each subset contains a set of possible states. Each state subset should be associated with a specific computing model or data processing requirement, and a corresponding state processing unit is developed for each state subset. The state processing unit should be able to identify and process the encoded correction results to adapt to different computing models (including AI models). When the state conversion unit receives the encoded calibration data, it passes the calibration data to the corresponding internal state processing unit. The state processing unit finds the state in the corresponding state subset according to the state table in the state subset (a state table is defined in each state subset to clarify the attributes and behaviors of each state. The state table should include information such as state name, state attributes, and trigger conditions.), and then integrates the state into environmental feedback data according to the bus interface specification and transmits it back to the bus, ensuring that the state conversion unit can adapt to different computing models, process the encoded correction results, and generate environmental feedback data that conforms to the bus interface specification.

[0086] A model training device, comprising:

[0087] A data acquisition module 11 is used to acquire input data through an interface bus and transmit the input data to a calculation model;

[0088] A data calculation module 12, used for the calculation model to receive input data and calculate to obtain output data;

[0089] A calibration processing module 13, used for performing calibration processing on the output data to obtain calibration data, and feeding back the calibration result to the interface bus;

[0090] The model training module 14 is used for the interface bus to feed back the calibration data to the AI ​​model for training according to the calibration results.

[0091] like Figure 5 As shown, the data acquisition module 11, the data calculation module 12, the calibration processing module 13 and the model training module 14 are connected in sequence. The module referred to in the present invention refers to a series of computer program instruction segments that can complete specific functions, which are more suitable for the execution process of model training than programs. For the specific implementation methods of each module, please refer to the corresponding method embodiments above, which will not be repeated here.

[0092] Another embodiment of the present invention provides a model training system, the system comprising at least one processor; and

[0093] a memory communicatively connected to the at least one processor; wherein,

[0094] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the model training method.

[0095] One or more processors 110 and memory 120, Figure 6 A processor 110 is used as an example for description. The processor 110 and the memory 120 may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0096] The processor 110 is used to complete various control logics of the control system 10. It can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a single-chip microcomputer, an ARM (Acorn RISC Machine) or other programmable logic device, a discrete gate or transistor logic, a discrete hardware component, or any combination of these components. In addition, the processor 110 can also be any traditional processor, microprocessor or state machine. The processor 110 can also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors combined with a DSP and / or any other such configuration.

[0097] The memory 120, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions corresponding to the model training method in the embodiment of the present invention. The processor 110 executes various functional applications and data processing of the control system 10 by running the non-volatile software programs, instructions and units stored in the memory 120, that is, implements the model training method in the above method embodiment.

[0098] The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created according to the use of the control system 10, etc. In addition, the memory 120 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 120 may optionally include a memory remotely arranged relative to the processor 110, and these remote memories may be connected to the control system 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0099] One or more units are stored in the memory 120, and when executed by one or more processors 110, the model training method in any of the above method embodiments is executed, for example, the model training method described above is executed. Figure 1 The method comprises steps S100 to S400.

[0100] An embodiment of the present invention provides a non-volatile computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed by one or more processors, for example, to execute the above-described Figure 1 The method comprises steps S100 to S400.

[0101] As an example, non-volatile storage media can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) as external cache memory. By way of illustration and not limitation, RAM can be obtained in many forms such as synchronous RAM (SRAM), dynamic RAM, (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The disclosed memory components or memories of the operating environment described herein are intended to include one or more of these and / or any other suitable types of memory.

[0102] In summary, the present invention discloses a model training method, device, system and medium, including: an interface bus obtains input data, and transmits the input data to a computing model; the computing model receives the input data and calculates to obtain output data; the output data is calibrated to obtain calibration data, and the calibration result is fed back to the interface bus; the interface bus feeds back the calibration data to the AI ​​model for training according to the calibration result. The present invention transmits the input data to the computing model through the interface bus for calculation to obtain output data, calibrates the output data, and feeds back the calibrated data to the model for training, thereby improving the standardization of the interface data and reducing the cost of interface switching.

[0103] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium, and the computer program can include the processes of the above-mentioned method embodiments when executed. The storage medium can be a memory, a disk, a floppy disk, a flash memory, an optical storage device, etc.

[0104] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A model training method, characterized in that: The steps include: The interface bus acquires input data and transmits the input data to the calculation model; The computing model receives input data and performs calculations to obtain output data; Performing calibration processing on the output data to obtain calibration data, and feeding back the calibration result to the interface bus; The interface bus feeds back the calibration data to the AI ​​model for training based on the calibration results.

2. The model training method according to claim 1, characterized in that: The interface bus acquires input data and transmits the input data to the calculation model, including: Set interface specifications and switch mechanisms, and build an interface bus; Classify the transmitted data and set index codes for each data type; The input data binding index is encoded according to the data type.

3. The model training method according to claim 1, characterized in that: The interface bus acquires input data and transmits the input data to the calculation model, including: Encrypting and identifying the input data according to the recognition algorithm, and determining whether the input data conforms to the recognition algorithm; When the input data conforms to the recognition algorithm, the interface is allowed to access the bus; When the input data does not conform to the identification algorithm, the interface is not allowed to access the bus.

4. The model training method according to claim 3 is characterized in that: When the input data meets the recognition algorithm, the interface is allowed to access the interface bus, including: When the input data conforms to the recognition algorithm, the input data is encrypted and transmitted; The data weight of the input data is calculated, and the data transmission order is determined according to the data weight and the interface specification.

5. The model training method according to claim 1, characterized in that: The step of calibrating the output data to obtain calibration data and feeding back the calibration result to the interface bus includes: If the output data needs calibration processing, the output data is corrected to obtain calibration data and output the calibration result; If the output data does not require calibration processing, a feedback result is output.

6. The model training method according to claim 5, characterized in that: The interface bus feeds back the calibration data to the AI ​​model for training according to the calibration result, including: Encoding the calibration data, and transmitting the encoded calibration data to an interface bus; The interface bus feeds the calibration data back to the AI ​​model for training.

7. The model training method according to claim 1, characterized in that: The step of encoding the calibration data and transmitting the encoded calibration data to an interface bus comprises: After receiving the encoded calibration data, the state conversion unit transmits the calibration data to the corresponding state processing unit; The state processing unit integrates the calibration data according to the state table and the interface specification to generate environmental feedback data; The state conversion unit transmits the environmental feedback data to an interface bus.

8. A model training device, characterized in that: The device comprises: A data acquisition module, used for acquiring input data through an interface bus, and transmitting the input data to a calculation model; A data calculation module, used for the calculation model to receive input data and calculate to obtain output data; A calibration processing module, used for performing calibration processing on the output data to obtain calibration data, and feeding back the calibration result to the interface bus; The model training module is used for the interface bus to feed back the calibration data to the AI ​​model for training according to the calibration results.

9. A model training system, characterized in that: The system includes at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the model training method described in any one of claims 1-7.

10. A non-volatile computer-readable storage medium, characterized in that: The non-volatile computer-readable storage medium stores computer-executable instructions, which, when executed by one or more processors, enable the one or more processors to execute the model training method described in any one of claims 1-7.