A reliable task distribution method and system for industrial intelligent agents

Abstract

The application provides a trusted industrial intelligent agent task distribution method and system, and relates to the technical field of data processing.The method comprises the following steps: receiving a registration request sent by an industrial intelligent agent developer, storing metadata information of the industrial intelligent agent into a distributed ledger based on the registration request; receiving a task distribution request sent by a task requester based on the metadata information, and matching a candidate industrial intelligent agent and a unique identifier that meet a functional requirement in the distributed ledger.The application stores core information in the distributed ledger, screens trusted intelligent agents through dynamic trust evaluation, and iteratively updates the trust level after execution, so as to realize the trusted traceability of the whole task distribution process and the dynamic optimization of the trust level of the intelligent agent.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a reliable industrial intelligent agent task distribution method and system. Background Technology

[0002] In industrial large-scale model applications, agents typically rely on specific large-scale model service platforms to complete the registration and task distribution process. Currently, many distribution methods have several limitations: First, they primarily rely on static metadata provided during agent registration (such as rated load capacity and theoretical speed) or simply refer to historical task success rates for trust assessment and task matching, without considering real-time performance fluctuations caused by equipment aging and software updates. Second, they lack a trusted evidence storage mechanism based on distributed ledgers; the storage and verification methods for agent metadata are relatively simple, making them susceptible to data tampering or loss, affecting the reliability of task matching. Third, in terms of trust assessment, fixed evaluation indicators are often used, and dynamic trust assessment mechanisms have not been constructed. The spatial and corresponding dynamic trust trajectory analysis mechanism makes it difficult to accurately and in real-time adjust the agent's trust level through trust correction parameters, and it also fails to comprehensively screen based on the trust threshold of the target large model service platform; fourth, after task distribution, there is a lack of multi-dimensional behavior analysis and feedback mechanism for the agent's execution process, making it impossible to update the agent's behavior compliance data and recalibrate the trust level in a timely manner based on the behavior adjustment coefficient; fifth, the evidence storage system for the entire task distribution process (including task request, agent matching, distribution instructions and execution status, etc.) is imperfect, making it difficult to achieve traceable and trustworthy task management.

[0003] For example, in the cargo transfer process of smart warehousing, some scheduling systems, when assigning shelf transfer tasks to AGVs (Automated Guided Vehicles), only refer to fixed parameters such as the rated load and theoretical driving speed entered during AGV registration to match tasks. For instance, assigning a 450kg shelf transfer task to an AGV with a rated load of 500kg fails to consider in real time that the AGV has been running continuously for 3 months, wheel wear has caused a 15% decrease in actual driving speed, or load calibration deviations have occurred after recent software updates. At the same time, the historical operation data of AGVs lacks distributed ledger storage, and their credibility assessment relies only on simple success rate statistics, without being correlated with the credibility threshold of the large model service platform to which they belong, nor is dynamic trust trajectory analysis established during task execution, resulting in the inability to correct the credibility level in a timely manner. After the task is completed, the compliance data is not updated by generating behavior adjustment coefficients through multidimensional analysis, and the entire task distribution process lacks complete storage, which may lead to reduced task execution efficiency and even the risk of shelf tilting, while making it difficult to trace the root cause of the problem and hold the responsible party accountable. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a reliable industrial intelligent agent task distribution method and system, which can track the performance fluctuations and behavior changes of the intelligent agent in real time, dynamically adjust the reliability level, and effectively improve the reliability, compliance and traceability of task distribution and execution.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] Firstly, a reliable task distribution method for industrial intelligent agents, the method comprising:

[0007] Step 1: Receive the registration request sent by the industrial intelligent agent developer, and store the metadata information of the industrial intelligent agent in the distributed ledger based on the registration request; based on the metadata information, receive the task distribution request sent by the task requester, and match the candidate industrial intelligent agents and unique identifiers that meet the functional requirements in the distributed ledger.

[0008] Step 2: Based on the unique identifier, obtain the historical performance data of each candidate industrial intelligent agent as the evaluation benchmark to construct a trust evaluation space; set five evaluation nodes in the trust evaluation space, generate dynamic trust trajectories based on the five evaluation nodes, and calculate the credibility correction parameters based on the dynamic trust trajectories; dynamically adjust the credibility level through the credibility correction parameters, and select distributable intelligent agents from the candidate industrial intelligent agents based on the adjusted credibility level and the platform credibility threshold corresponding to the identifier of the target large model service platform;

[0009] Step 3: Based on the preset scheduling strategy, determine the target industrial intelligent agent from the distributable intelligent agents, generate a distribution instruction, and send it to the target large model service platform;

[0010] Step 4: Perform multidimensional analysis on the status information of the execution and distribution instructions returned by the target large model service platform to generate behavior adjustment coefficients. Update the behavior compliance data of the target industrial intelligent agent based on the behavior adjustment coefficients and recalculate the credibility level of the target industrial intelligent agent.

[0011] Step 5: Record the task distribution request, target industrial intelligent agent, distribution instructions and task execution status information into the distributed ledger to form task distribution and execution evidence.

[0012] Secondly, a reliable industrial intelligent agent task distribution system includes:

[0013] The registration module is used to receive registration requests sent by industrial intelligent agent developers, store the metadata information of industrial intelligent agents in the distributed ledger based on the registration requests, and receive task distribution requests sent by task requesters based on the metadata information, and match candidate industrial intelligent agents and unique identifiers that meet the functional requirements in the distributed ledger.

[0014] The screening module is used to obtain the historical performance data of each candidate industrial intelligent agent based on a unique identifier as an evaluation benchmark to construct a trust evaluation space. Five evaluation nodes are set in the trust evaluation space, and a dynamic trust trajectory is generated based on the five evaluation nodes. The credibility correction parameter is calculated based on the dynamic trust trajectory. The credibility level is dynamically adjusted through the credibility correction parameter, and distributable intelligent agents are selected from the candidate industrial intelligent agents based on the adjusted credibility level and the platform credibility threshold corresponding to the identifier of the target large model service platform.

[0015] The scheduling module is used to determine the target industrial intelligent agent from the distributable intelligent agents based on a preset scheduling strategy, generate distribution instructions, and send them to the target large model service platform.

[0016] The update module is used to perform multi-dimensional analysis on the status information of the execution and distribution instructions returned by the target large model service platform to generate behavior adjustment coefficients, update the behavior compliance data of the target industrial intelligent agent based on the behavior adjustment coefficients, and recalculate the credibility level of the target industrial intelligent agent.

[0017] The evidence storage module is used to record task distribution requests, target industrial intelligent agents, distribution instructions, and task execution status information into a distributed ledger to form task distribution and execution evidence storage.

[0018] Thirdly, a computing device includes:

[0019] One or more processors;

[0020] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0021] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0022] The above-described solution of the present invention has at least the following beneficial effects:

[0023] By storing the metadata of industrial intelligent agents in a distributed ledger, the immutability of the metadata is ensured, and candidate intelligent agents can be matched based on the real-time updated metadata in the ledger. This avoids matching deviations caused by relying solely on initial static metadata. For example, unlike in smart warehousing where tasks are assigned based solely on the initial parameters of AGVs, the ledger can be supplemented with updated equipment status information, making the matching of candidate intelligent agents more closely match actual functional requirements. A trust evaluation space is constructed based on historical performance data. Through five evaluation nodes, a dynamic trust trajectory is generated and correction parameters are calculated. This allows for real-time tracking of performance fluctuations and behavioral changes in intelligent agents, dynamically adjusting the trust level. Compared to simply referring to historical success rates, this approach can promptly reflect the impact of equipment-related factors on the intelligent agent's performance. Performance changes caused by aging and software updates; determining target agents from distributable agents according to preset strategies and sending instructions to avoid potential blind allocation problems in the distribution process, thus improving the quality and efficiency of task completion; generating behavior adjustment coefficients through multi-dimensional analysis of execution status information, updating compliance data and recalculating credibility levels, allowing the credibility assessment of agents to continuously iterate and optimize with the task execution situation, solving the problem of static and difficult-to-reflect real-time status of agents; recording key information of the entire task distribution chain to a distributed ledger to form evidence, so that when task disputes or failures occur, responsibility can be quickly identified and root causes analyzed based on the ledger evidence, enhancing the regulatory oversight of the task distribution process in industrial big data model applications. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a reliable industrial intelligent agent task distribution method provided by an embodiment of the present invention.

[0025] Figure 2 This is a schematic diagram of a reliable industrial intelligent agent task distribution system provided by an embodiment of the present invention. Detailed Implementation

[0026] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0027] like Figure 1 As shown, embodiments of the present invention propose a reliable task distribution method for industrial intelligent agents, the method comprising the following steps:

[0028] Step 1: Receive the registration request sent by the industrial intelligent agent developer, and store the metadata information of the industrial intelligent agent in the distributed ledger based on the registration request; based on the metadata information, receive the task distribution request sent by the task requester, and match the candidate industrial intelligent agents and unique identifiers that meet the functional requirements in the distributed ledger.

[0029] Step 2: Based on the unique identifier, obtain the historical performance data of each candidate industrial intelligent agent as the evaluation benchmark to construct a trust evaluation space; set five evaluation nodes in the trust evaluation space, generate dynamic trust trajectories based on the five evaluation nodes, and calculate the credibility correction parameters based on the dynamic trust trajectories; dynamically adjust the credibility level through the credibility correction parameters, and select distributable intelligent agents from the candidate industrial intelligent agents based on the adjusted credibility level and the platform credibility threshold corresponding to the identifier of the target large model service platform;

[0030] Step 3: Based on the preset scheduling strategy, determine the target industrial intelligent agent from the distributable intelligent agents, generate a distribution instruction, and send it to the target large model service platform;

[0031] Step 4: Perform multidimensional analysis on the status information of the execution and distribution instructions returned by the target large model service platform to generate behavior adjustment coefficients. Update the behavior compliance data of the target industrial intelligent agent based on the behavior adjustment coefficients and recalculate the credibility level of the target industrial intelligent agent.

[0032] Step 5: Record the task distribution request, target industrial intelligent agent, distribution instructions and task execution status information into the distributed ledger to form task distribution and execution evidence.

[0033] In this embodiment of the invention, by storing the metadata of industrial intelligent agents in a distributed ledger, matching deviations caused by relying solely on initial static metadata are avoided. For example, tasks are no longer assigned based solely on the initial parameters of AGVs, as is the case in smart warehousing. By incorporating potentially updated equipment status information from the ledger, the matching of candidate intelligent agents becomes more closely aligned with actual functional requirements. A trust evaluation space is constructed based on historical performance data. Through five evaluation nodes, a dynamic trust trajectory is generated and correction parameters are calculated. This allows for real-time tracking of performance fluctuations and behavioral changes in intelligent agents, dynamically adjusting the trust level. Compared to evaluation methods that simply refer to historical success rates, this approach can promptly reflect performance changes caused by equipment aging and software updates. Combined with platform trust threshold filtering, this effectively avoids unreliable intelligent agents caused by evaluation lag. The system addresses the selection of agents; it identifies target agents from distributable agents based on preset strategies and sends instructions, selecting the optimal agent based on actual needs (such as trustworthiness priority), avoiding blind allocation during distribution, and improving task completion quality and efficiency; it generates behavior adjustment coefficients through multi-dimensional analysis of execution status information, updates compliance data and recalculates trustworthiness levels, allowing agent trustworthiness assessment to continuously iterate and optimize with task execution, solving the problem of static assessments that fail to reflect the real-time status of agents; and it records key information throughout the task distribution chain in a distributed ledger for evidence preservation, enabling rapid identification of responsibility and root cause analysis in case of task disputes or failures, enhancing the regulatory oversight of the task distribution process in industrial large-scale model applications.

[0034] In a preferred embodiment of the present invention, step 1 includes:

[0035] In this embodiment of the invention, step 100 involves receiving a registration request from an industrial intelligent agent developer. The registration request includes a unique identifier for the industrial intelligent agent, functional description information, and authentication credentials. Specifically, the large-scale model service platform refers to a professional technical platform possessing core capabilities of industrial large-scale models, capable of providing industrial intelligent agents with end-to-end services such as registration management, task distribution, data storage, and performance monitoring.

[0036] First, industrial intelligent agent developers, such as companies developing AGVs or teams developing production monitoring intelligent agents, need to register on the large-scale model service platform. This platform will pre-build an online registration page or a dedicated registration interface. After logging into their platform accounts via computer, developers will enter the dedicated industrial intelligent agent registration section, fill in the required information, and submit the registration request. The unique identifier for the industrial intelligent agent is generated by the developer according to the platform's preset rules. For example, if the platform requires the identifier to be in the format of company abbreviation-intelligent agent type-6-digit code, then a company like "XX Technology" developing AGVs might generate a unique identifier like "XX Technology-AGV-123456". This identifier will be uniquely verified in the platform's database to ensure it doesn't duplicate intelligent agents registered by other companies. The functional description information requires developers to fill in the actual capabilities of the intelligent agent, using an AGV intelligent agent in a smart warehousing scenario as an example. For example, developers need to specify the weight range of goods the AGV can handle (e.g., 50kg-500kg), the supported path planning methods (e.g., laser navigation, QR code navigation), the types of obstacles it can identify during obstacle avoidance (e.g., personnel, shelves, other equipment), and the types of tasks it can handle (e.g., warehouse shelf transfer, production line material delivery), and the working environment requirements (e.g., temperature 0-40℃, humidity 30%-80%). Identity verification credentials are materials used by developers to prove their legal identity. If the developer is a company, they need to upload a scanned copy of their business license and industry qualification certificates (e.g., intelligent equipment production license). If the developer is an individual or team, they need to upload a photo of the team leader's ID card and platform account real-name authentication information (e.g., facial recognition pass record). After this information is filled out and submitted, the request receiving unit of the large model service platform will obtain the registration request in real time, temporarily store the request data in the platform's temporary database, and generate a temporary request number.

[0037] Step 101: Verify the validity of the identity verification credentials. After successful verification, extract the unique identifier and functional description information as metadata information and add the metadata information to the intelligent agent registration block of the distributed ledger. Specifically, the large model service platform will automatically call the internal identity verification unit to verify the validity of the identity verification credentials in the registration request stored in the temporary database. During the verification, the platform will first retrieve the registered enterprise or personal information from its own legitimate developer information database. For example, if the developer submits the business license of "XX Technology", the platform will find the previously registered business license information of "XX Technology" in the information database and check the enterprise name, unified social credit code, business scope, business license validity period, etc. of the two word by word to confirm that all information is completely consistent. If the developer submits the facial recognition information bound to the platform account, the platform will compare whether the facial data submitted this time matches the facial data left when registering the account, and at the same time check whether the account is in normal use (no violation ban record). If all the above verification items meet the requirements, the platform will determine that the identity verification credentials are valid. If there is a discrepancy in the enterprise name, the qualification certificate has expired, or the identity verification credentials are invalid, the platform will determine that the identity verification credentials are valid. If facial recognition fails or any other reason is met, the platform will determine that the verification has failed and notify the developer via SMS or platform message, clearly stating the reason for the failure. For example, if the business license has expired, the developer should upload the latest certificate. Once the identity verification credential is verified, the platform's data extraction unit will extract the unique identifier and functional description information of the industrial intelligent agent from the registration request, and integrate these two types of information to define metadata information. Next, the platform's ledger interaction unit will establish a connection with the distributed ledger system. This distributed ledger is jointly built by the platform and multiple industrial enterprises, with each participant having a ledger node. In the intelligent agent registration block specifically divided in the distributed ledger, a new record is created for the industrial intelligent agent, and the integrated metadata information is completely written into this record. At the same time, the record creation time, platform identifier, and other information are marked, completing the on-chain storage of the metadata information. After storage, the distributed ledger will automatically synchronize this new record to all ledger nodes, such as the server nodes of various factories and enterprises that participated in building the ledger, to ensure that each node can query the metadata information, and that no single node can modify this record alone, preventing the metadata information from being tampered with.

[0038] Step 102: Based on the metadata in the agent registration block, receive the task distribution request sent by the task requester, parse the task distribution request to obtain the task description, the functional requirements of the required industrial intelligent agent, and the identifier of the target large model service platform. Specifically, the large model service platform will continuously monitor the task sending status of the task requester (which may be a warehouse manager that needs to schedule AGVs, a factory that needs to monitor equipment, etc.) through a dedicated network channel, such as an encrypted API interface. When the task requester has a task requirement, it will log in to its own large model service platform account, enter the task publishing section, fill in and submit the task distribution request. At this point, the platform's request receiving unit first retrieves all stored industrial intelligent agent metadata information from the intelligent agent registration block of the distributed ledger to quickly determine whether the platform has registered an intelligent agent type that can adapt to the task. For example, if the task requester needs an AGV capable of moving 300kg shelves, the platform will check if the metadata contains an AGV intelligent agent whose function description includes moving 300kg goods. After confirming that there is a suitable type, the platform will then receive the task distribution request. Subsequently, the platform's request parsing unit will break down the received task distribution request. The first step is to extract the task description information. For example, if the task requester is a warehouse manager, the task description will specify that before 14:00 on August 28, 2024, 10 300kg shelves in warehouse A should be moved. The first step is to transfer heavy shelving to Zone B, avoiding the construction area in Zone C during the transfer. The specific details, completion time, and special requirements of the task must be clearly defined. The second step is to identify the functional requirements of the required industrial intelligent agent, extracting three essential functionalities from the task description: the ability to handle 300kg of goods, support for obstacle avoidance, and adaptation to the warehouse environment. The third step is to extract the identifier of the target large-scale model service platform. This identifier is specified by the task requester and is the platform that ultimately receives and executes the task. For example, if the task requester has a long-term partnership with the "XX Warehouse Intelligent Service Platform," its identifier might be the platform's unique C001 number or the platform's full name, "XX Warehouse Intelligent Service Platform." After parsing, the platform will store these three types of information in the task information database.

[0039] Step 103: Compare the functional requirements with the functional descriptions of all registered industrial intelligent agents in the distributed ledger, and select industrial intelligent agents whose functional descriptions contain all the essential functional items in the functional requirements as candidate industrial intelligent agents. Obtain the unique identifier for each candidate industrial intelligent agent. Specifically, this includes: First, the ledger data retrieval unit of the large model service platform retrieves the functional descriptions of all registered industrial intelligent agents from the agent registration block through a connection with the distributed ledger. For example, it retrieves the functional description of "XX Technology-AGV-123456" (capable of handling 50kg-500kg goods, laser navigation obstacle avoidance, adaptable to warehouse environments). The functional descriptions of "YY Technology-AGV-654321" (capable of handling 10kg-200kg goods, QR code navigation and obstacle avoidance, adaptability to workshop environment, etc.) are compiled into a complete set of functional description information. Then, the platform's requirement decomposition unit will break down the functional requirements obtained in step 102, such as the ability to handle 300kg goods, support for path obstacle avoidance, and adaptability to warehouse environment, into three independent essential functional items. According to the preset comparison order of core function priority (first compare the goods handling capacity, then compare the obstacle avoidance function, and finally compare the environmental adaptability), each essential functional item is compared with the functional description of each industrial intelligent entity in the functional description information set. The information is compared. For example, the essential function of being able to move 300kg of goods is first compared with the function description of "XX Technology-AGV-123456". It is found that its moving range includes 300kg, which meets the requirement. Next, the function description of supporting obstacle avoidance is compared, and the function description of this AGV includes laser navigation obstacle avoidance, which meets the requirement. Finally, the function description of adapting to the warehouse environment is also clearly adapted, so this AGV is initially included in the candidate range. Then, the same essential function is compared with "YY Technology-AGV-654321". Its maximum moving weight is 200kg, which does not meet the requirement of being able to move 300kg of goods, so this AGV is directly excluded. The above method is used to complete the function comparison. After comparing all industrial intelligent agents in the information set, the platform's candidate screening unit will summarize all industrial intelligent agents that meet the conditions, that is, whose functional descriptions contain all the necessary functional items, and identify them as candidate industrial intelligent agents. Finally, the platform will retrieve and record the unique identifier of each candidate industrial intelligent agent according to the correspondence between industrial intelligent agent and unique identifier stored in the intelligent agent registration block in the distributed ledger. For example, if the ledger records the identifier "XX Technology-AGV-123456" corresponding to the XX Technology AGV that can carry goods from 50kg to 500kg, the unique identifier of each candidate industrial intelligent agent will be formed into a correspondence list of candidate industrial intelligent agent name and unique identifier, and stored in the candidate intelligent agent database.

[0040] This embodiment ensures the integrity of industrial agent registration information and the identifiability of developer identities, avoiding the problem of chaotic task matching caused by missing information during the registration process; it guarantees the reliability of the source of registered industrial agents, and stores metadata information in the agent registration block of the distributed ledger to prevent the metadata information from being arbitrarily modified, solving the problem of static metadata being easily tampered with or lost; it clearly defines the task description, functional requirements, and target platform identifier, avoiding matching deviations caused by ambiguous request information during task distribution; by comparing the functional requirements with the functional description information of the industrial agent item by item, it filters out candidate industrial agents containing all the necessary functional items, avoiding the problem of mismatch between agent functions and task requirements caused by matching only based on initial static parameters, and at the same time obtains the unique identifier of the candidate agent to ensure the accuracy of task distribution.

[0041] In a preferred embodiment of the present invention, step 2 includes:

[0042] In this embodiment of the invention, step 200 involves obtaining historical performance data of each candidate industrial intelligent agent in the latest 100 task executions from the historical task execution record database based on the unique identifier of each candidate industrial intelligent agent. The historical performance data includes task success rate, task execution latency, and resource utilization rate. Specifically, this includes: clarifying that the historical task execution record database is a database within the large model service platform that specifically stores past task execution data for all industrial intelligent agents. Each data entry in the database is bound to the unique identifier of the corresponding industrial intelligent agent and records information such as task execution time, task result, and resource consumption. For each candidate industrial intelligent agent, data is retrieved from the historical task execution record database based on its unique identifier. During the retrieval, the task execution records of the intelligent agent are filtered in ascending order of task execution time, using the current time as a reference, until 100 of the latest valid records are selected. If the intelligent agent has accumulated [number of executions], [further details are needed]. If the number of task executions is less than 100, then all records of executed tasks are retrieved. From each selected record, three types of key data are extracted as historical performance data. The first type is task success rate data, which requires counting the number of successfully completed tasks out of 100 tasks. The success rate is then calculated by dividing the number of successful tasks by the total number of tasks (100; if less than 100, the actual total number is used). The second type is task execution latency data, which records the time consumed from receiving the instruction to completing the task for each task, forming the execution latency value for each task. The third type is resource utilization data, which counts the actual usage of resources such as CPU, memory, and energy by the agent during each task execution. The actual usage is then divided by the agent's rated total resources to obtain the resource utilization value for each task. These three types of data are then organized in order of task execution time to form the historical performance dataset corresponding to the candidate industrial intelligent agent.

[0043] Step 201: Using historical performance data as the evaluation benchmark, construct a two-dimensional trust evaluation space. The horizontal axis represents the time dimension, and the vertical axis represents the task success rate indicator dimension. Specifically, this includes: first, organizing the task success rate and task execution time from the historical performance data; using the execution time of each task as the coordinate value of the time dimension, and the corresponding success rate of each task as the coordinate value of the task success rate indicator dimension. Each set of time and success rate data constitutes an evaluation benchmark. Next, determine the coordinate axis range of the two-dimensional trust evaluation space. The horizontal axis, representing the time dimension, starts with the execution time of the earliest task among the selected 100 tasks and ends with the execution time of the latest task. The endpoint is to divide this time interval evenly into several equal parts, so that the time coordinate of each evaluation benchmark point can correspond to a clear position on the horizontal axis; the vertical axis is the task success rate index dimension, which is set according to the task success rate data range of all candidate industrial intelligent agents, from 0 to 100%, to ensure that the success rate values ​​of all evaluation benchmark points can be fully presented on the vertical axis. All the organized evaluation benchmark points are marked in chronological order on this planar coordinate system with time on the horizontal axis and task success rate on the vertical axis. The position of each benchmark point in the coordinate system is determined by its corresponding time coordinate and success rate coordinate. In this way, a two-dimensional trust evaluation space is constructed for evaluating the candidate industrial intelligent agent.

[0044] Step 202: Set five evaluation nodes within the two-dimensional trust evaluation space. The first evaluation node is located at the center of the high-performance stable zone with a task success rate greater than or equal to 95%. The second evaluation node is located at the center of the performance fluctuation zone with a task success rate between 85% and 94%. The third evaluation node is located at the center of the low-performance zone with a task success rate less than 85%. The fourth evaluation node is located at the critical position inside the trust boundary with a task success rate of 90%. The fifth evaluation node is located at the critical position outside the trust boundary with a task success rate of 80%. Specifically, on the vertical axis task success rate dimension, the area with a success rate greater than or equal to 95% is defined as the high-performance stable zone, the area with a success rate between 85% and 94% is defined as the performance fluctuation zone, and the area with a success rate less than 85% is defined as the low-performance zone. Simultaneously, the position of 90% success rate is set as the internal critical line of the trust boundary, and the position of 80% success rate is set as the external critical line of the trust boundary. For the first evaluation node, first calculate the time coordinate range and success rate coordinate range of all evaluation benchmark points within the high-performance stable zone, and take the time range... The median value is used as the time coordinate of the node, and the median value of the success rate range is taken as the success rate coordinate of the node. This coordinate point is the location of the first evaluation node, and the node must be located in the center of the high-performance stable zone. For the second evaluation node, the time and success rate coordinate ranges of the evaluation benchmark points within the performance fluctuation zone are also calculated, and the median value of the two ranges is taken as the node coordinate, ensuring that the node is located in the center of the performance fluctuation zone. For the third evaluation node, the coordinate ranges of the benchmark points within the low-performance zone are calculated in the same way, and the median value is taken to determine the node location, making it located in the center of the low-performance zone. For the fourth evaluation node, the median value of the time range of all evaluation benchmark points is taken as the time coordinate on the horizontal axis time dimension, and then the vertical axis success rate coordinate is fixed at 90%. The intersection of this time coordinate and the 90% success rate coordinate is the fourth evaluation node, which falls exactly on the inner critical line of the trust boundary. For the fifth evaluation node, the median value of the horizontal axis time range is also taken as the time coordinate, and the vertical axis success rate coordinate is fixed at 80%. The intersection of the two is the fifth evaluation node, which falls on the outer critical line of the trust boundary.

[0045] Step 203 involves generating a dynamic trust trajectory based on the continuous changes of the five evaluation nodes over time. This includes: sorting the five evaluation nodes by time, arranging them sequentially from the earliest to the latest node based on their time coordinates, forming a time sequence; observing the success rate changes of each node over time (e.g., from the earliest evaluation node to the next, whether the success rate increases, decreases, or remains stable); recording the trend and magnitude of success rate changes between adjacent nodes; and then connecting the five evaluation nodes sequentially by time using a smooth curve. The connection process should fully reflect the trend of success rate changes between adjacent nodes. If the success rate of the previous node is higher than that of the next node, the curve shows a downward trend; if the success rate of the previous node is lower than that of the next node, the curve shows an upward trend; if the success rate remains relatively constant, the curve is close to horizontal. This smooth curve formed by connecting the five evaluation nodes in time sequence is the dynamic trust trajectory of the candidate industrial intelligence agent. The trajectory can intuitively reflect the changing pattern of the agent's success rate over different time periods.

[0046] Step 204: Calculate the credibility correction parameter based on the curvature change rate of the dynamic trust trajectory. The curvature change rate is obtained by the ratio of the change in the tangent angle of the trajectory curve to the change in time. Specifically, this includes: selecting several uniformly distributed sampling points on the dynamic trust trajectory curve. During selection, first determine the total time length corresponding to the trajectory curve. Subtract the earliest time coordinate from the latest time coordinate of the trajectory curve to obtain the total time length. Then, based on the required number of sampling points, divide the total time length by (number of sampling points minus 1) to obtain the time interval between two adjacent sampling points. According to this time interval, starting from the earliest time point of the trajectory curve, determine the time coordinate of each sampling point sequentially. Then, find the corresponding success rate coordinate on the trajectory curve based on the time coordinate, ensuring that each sampling point has a clear and unique time coordinate and success rate coordinate. For example, if the total time length is 100 hours and 11 sampling points are selected, then the time interval between adjacent sampling points is 10 hours, starting from 0 hours, 10 hours, 20 hours... ...The time coordinates of the sampling points are determined sequentially over 100 hours, and the success rate coordinates of each time point are found accordingly. For every two adjacent sampling points, the tangent angle of the trajectory curve at each sampling point is calculated. The tangent angle is the angle between the tangent and the positive direction of the horizontal time dimension. The calculation first determines the slope of the curve at each sampling point. The slope is calculated by selecting two adjacent auxiliary points before and after the sampling point, subtracting the success rate coordinate of the previous auxiliary point from the success rate coordinate of the latter auxiliary point to obtain the change in success rate, then subtracting the time coordinate of the previous auxiliary point from the time coordinate of the latter auxiliary point to obtain the change in time. The change in success rate is divided by the change in time to obtain the slope of the curve at that sampling point. If the slope is positive, it indicates that the tangent is sloping upwards, and the angle between the tangent and the positive direction of the horizontal time dimension is acute. For example, when the slope is 0.5, the angle is approximately 26.56 degrees. If the slope is negative, it indicates that the tangent is sloping downwards, and the angle is obtuse. For example, when the slope is -0.5, the angle is approximately 153 degrees.43 degrees; if the slope is zero, it means the tangent is horizontal and the included angle is zero. This method determines the specific value of the tangent angle at each sampling point. Then, calculate the change in tangent angle between two adjacent sampling points. Specifically, subtract the tangent angle of the previous sampling point from the tangent angle of the later sampling point. The difference is the change in tangent angle between these two sampling points. For example, if the tangent angle of the previous sampling point is 30° and the tangent angle of the later sampling point is 45°, then the angle change is 45° - 30° = 15°. If the tangent angle of the previous sampling point is 45° and the tangent angle of the later sampling point is 30°, then the angle change is 30° - 45° = -15°. Simultaneously, calculate the time change between these two sampling points. Subtract the time coordinate of the previous sampling point from the time coordinate of the later sampling point. The difference is the time change. For example, if the time coordinate of the previous sampling point is 10 hours and the time coordinate of the later sampling point is 20 hours, then the time change is 20 hours - 10 hours. Given a time interval of 10 hours, divide the calculated change in tangent angle by the change in time to obtain the rate of curvature change of the trajectory segment. For example, if the angle change is 15° and the time change is 10 hours, then the rate of curvature change is 15° ÷ 10 hours = 1.5 degrees per hour; if the angle change is -15° and the time change is -15° ÷ 10 hours = -1.5 degrees per hour. Calculate the rate of curvature change for the trajectory between all adjacent sampling points using this method, obtaining several rate of curvature change values. Then, average all the rate of curvature change values. First, add all the rate of curvature change values ​​to get the sum of the rate of curvature changes. Then, divide the sum of the rate of curvature changes by the number of rate of curvature changes (i.e., the number of groups of adjacent sampling points) to obtain the average rate of curvature change. For example, if there are 10 groups of adjacent sampling points, calculate 10 rate of curvature change values, add these 10 values ​​to get the sum, and then divide the sum by 10 to get the average rate of curvature change.

[0047] The confidence correction parameter is determined based on the magnitude and sign of the average rate of change of curvature. If the average rate of change of curvature is positive and large, such as 2.5 degrees per hour, it indicates a significant upward trend in the trajectory curve, signifying continuous improvement in the agent's performance. In this case, the confidence correction parameter is set to a positive value, and its magnitude is positively correlated with the magnitude of the positive value of the average rate of change of curvature; the larger the positive value of the average rate of change of curvature, the larger the positive value of the confidence correction parameter. Conversely, if the average rate of change of curvature is negative and large, such as -2.5 degrees per hour, it indicates... A clear downward trend in the trajectory curve indicates a continuous decline in the agent's performance. In this case, the credibility correction parameter is set to a negative value, and its magnitude is positively correlated with the magnitude of the negative value of the average rate of change of curvature. The larger the negative value of the average rate of change of curvature, the larger the negative value of the credibility correction parameter. If the average rate of change of curvature is close to zero, such as 0.1 degrees per hour or -0.1 degrees per hour, it indicates that the trajectory curve is stabilizing, meaning that the agent's performance remains stable. In this case, the credibility correction parameter is set to a value close to zero to complete the calculation of the credibility correction parameter.

[0048] Step 205: Determine the initial credibility level for each candidate industrial intelligence agent based on historical performance data. Weighted fusion of the credibility correction parameter and the initial credibility level yields a dynamically adjusted credibility level. Specifically, this includes determining the initial credibility level based on three core indicators from historical performance data: task success rate, task execution latency, and resource utilization. Here, historical performance data refers to the relevant data from the latest 100 tasks (or actual execution counts) of the candidate industrial intelligence agent obtained in Step 200. Before calculating the evaluation values ​​for each indicator, clear and feasible scoring standards must be set for each of these three indicators. For the task success rate indicator, it is clearly set that the higher the success rate, the higher the corresponding evaluation value, and the maximum evaluation value for this indicator (e.g., 100) is predetermined. For example, if a candidate intelligence agent achieves a 100% success rate in historical tasks, then this indicator can directly obtain the highest evaluation value of 100; if the success rate is 90%, the evaluation value is reduced proportionally to ensure a positive correlation between the score and the success rate. For task execution latency... For metrics, firstly, based on the needs of the industrial scenario and the capabilities of the intelligent agent, a standard latency for the metric is preset. For example, in an intelligent warehousing scenario, the standard latency for the AGV transfer rack is set to 10 minutes, and a maximum evaluation value (e.g., 50). It is clearly defined that the shorter the actual time consumed, the higher the evaluation value of the corresponding metric. Under the premise that the actual time consumed does not exceed the standard latency, the closer the actual time consumed is to the lower limit of the standard latency, such as 5 minutes, the closer the evaluation value is to the maximum value. If the actual time consumed exceeds the standard latency, the metric is judged to have failed, and the evaluation value is directly set to 0. For resource utilization metrics, firstly, based on the hardware configuration and task requirements of the intelligent agent, an optimal utilization rate is preset. For example, the optimal CPU utilization rate is set to 80%, and the optimal memory utilization rate is set to 75%, and a maximum evaluation value (e.g., 50). It is clearly defined that the closer the actual usage is to the preset optimal usage, the higher the evaluation value of the corresponding metric. For example, when the actual CPU utilization rate is 82%, the evaluation value is higher because the difference from the optimal utilization rate of 80% is small. If the actual utilization rate is 60% or 100%, the evaluation value will be significantly lower because the difference from the optimal value is large.

[0049] The calculation method for the task success rate assessment is as follows: The actual success rate of the candidate industrial intelligent agent, such as 98%, is calculated by dividing the number of successful tasks in history by the total number of tasks, multiplied by the highest assessment value set for the task success rate indicator, 100. For example, if the highest assessment value set for the task success rate is 100, and the actual success rate of an intelligent agent is 98%, then the assessment value for this indicator is 98% × 100, directly reflecting the proportional relationship between the success rate and the assessment value. The calculation method for the task execution latency assessment is as follows: First, calculate the difference between the standard latency and the actual latency. If the actual latency exceeds the standard latency, this difference is negative, and the assessment value for this indicator is directly determined to be 0. Then, divide this difference by the standard latency to obtain the latency compliance rate. Finally, multiply this rate by the highest assessment value set for the latency indicator to obtain the final assessment value. For example, if the highest assessment value set for the latency indicator is 50, the standard latency is 10 minutes, and the actual latency of an intelligent agent is 8 minutes, first calculate 10 - 8 = 2, then multiply by the highest assessment value set for the latency indicator. 2 ÷ 10 = 0.2, yielding the latency compliance rate. Finally, 0.2 × 50 is used to obtain the evaluation value of this indicator, clearly demonstrating the impact of the difference between actual latency and standard latency on the score. The evaluation value of resource utilization is calculated as follows: First, calculate the absolute value of the difference between actual utilization and optimal utilization. For example, the difference between the actual CPU utilization of 85% and the optimal utilization of 80% is 5%, with an absolute value of 5%. Then, subtract this absolute value from 1 to obtain the utilization matching degree. Finally, multiply this matching degree by the highest evaluation value set for the resource utilization indicator to obtain the final evaluation value. For example, if the highest evaluation value set for resource utilization is 50, the optimal utilization is 80%, and the actual utilization of a certain agent is 85%, first calculate the difference between 85% and 80% as 5%, with an absolute value of 5%. Then, subtract 5% from 1 to obtain a utilization matching degree of 0.95. Finally, multiply 0.95 by 50 to obtain the evaluation value of this indicator, intuitively reflecting the degree of fit between the actual utilization and the optimal value.

[0050] Next, different weights are assigned to the three indicators. The weighting should be based on the importance of each indicator in the industrial scenario. Since task success rate directly determines whether a task can be completed, it has the highest weight. Task execution latency affects task efficiency, and resource utilization affects cost and equipment lifespan; their weights decrease sequentially. The sum of the weights of the three indicators is fixed at 1 (to ensure rigorous weighted calculation logic). For example, the weight of task success rate can be set to 0.5, task execution latency to 0.3, and resource utilization to 0.2, clearly defining the importance of each indicator in the overall process. In the comprehensive evaluation, the weighted results of the three indicators are calculated by multiplying the indicator evaluation value by its corresponding weight. For example, if the task success rate evaluation value of a certain intelligent agent is 98 and the weight is 0.5, the corresponding weighted result is 98 multiplied by 0.5, which equals 49; the latency evaluation value is 10 and the weight is 0.3, which equals 3; and the resource utilization evaluation value is 47.5 and the weight is 0.2, which equals 9.5. The three weighted results are then added together to obtain the candidate indicator. The comprehensive evaluation results of industrial intelligent agents allow the comprehensive score to fully reflect the multi-dimensional performance of the agents. Initial trust levels are determined based on the range of the comprehensive evaluation results. The classification criteria must be combined with the industrial scenario's requirements for the trust level of the agents. For example, a comprehensive evaluation result of 90 or above corresponds to initial trust level 1 (high trust), 80 to 89 corresponds to level 2 (relatively high trust), 70 to 79 corresponds to level 3 (medium trust), 60 to 69 corresponds to level 4 (basic trust), and below 60 corresponds to level 5 (low trust). If the comprehensive evaluation result of an agent in the above example is 61.5, falling within the range of 60 to 69, the initial trust level of the agent can be determined as level 4 (basic trust), ensuring that the level classification matches the comprehensive performance. The initial trust level is then converted into a corresponding numerical value. The conversion rule must be consistent with the level classification logic. For example, a value of 100 is set for level 1, 90 for level 2, 80 for level 3, 70 for level 4, and 60 for level 5. Taking the initial trust level of level 4 as an example, its corresponding numerical value is 70, clearly defining the correspondence between level and value.

[0051] After determining the initial credibility level value, the weighted fusion stage begins. First, weights are assigned to both the initial credibility level value and the credibility correction parameter. The initial credibility level value has a significantly higher weight than the credibility correction parameter to ensure its fundamental role. For example, the initial credibility level value might have a weight of 0.9, and the credibility correction parameter a weight of 0.1. Multiplying the initial credibility level value by its corresponding weight yields the weighted result of the initial credibility level; multiplying the credibility correction parameter by its corresponding weight yields the weighted result of the credibility correction parameter. Finally, the two weighted results are summed to obtain the final value. This is the dynamically adjusted credibility level value. For example, if the initial credibility level value is 70 and the weight is 0.9, the corresponding weighted result is 70 multiplied by 0.9, which equals 63. If the credibility correction parameter is 5 and the weight is 0.1, the corresponding weighted result is 5 multiplied by 0.1, which equals 0.5. Adding the two together gives 63.5, which is the dynamically adjusted credibility level value. Based on the range of the dynamically adjusted credibility level value, and referring to the classification standard of the initial credibility level, the final dynamically adjusted credibility level is determined. For example, if 63.5 is still in the range of 60 to 69, the corresponding dynamically adjusted credibility level is level four.

[0052] Step 206 involves comparing the dynamically adjusted credibility levels of each candidate industrial intelligent agent with the platform credibility threshold of the target large model service platform. Industrial intelligent agents with credibility levels greater than or equal to the platform credibility threshold are selected to form a distributable agent set. Specifically, this includes determining the platform credibility threshold of the target large model service platform. This threshold is determined by the target large model service platform based on its own operational needs, such as the platform's minimum requirements for task delivery quality, customer agreements on result reliability, and the reliability requirements of the industrial scenarios it serves. For example, in high-precision component manufacturing scenarios, agent errors can lead to production line shutdowns and product scrapping, requiring extremely high reliability. In ordinary data acquisition scenarios, minor data deviations can be corrected through subsequent verification, requiring relatively relaxed reliability. It also considers past task execution risk data, such as the number of task failures due to insufficient agent credibility and fault handling costs. The platform credibility threshold is stored in the target large model service platform's system database in a clearly defined numerical form as a unified benchmark for selection.

[0053] For example, the platform trust threshold is set to 80 for high-precision production scenarios and 60 for ordinary data acquisition scenarios. Then, the dynamically adjusted trust level value of each candidate industrial intelligent agent is retrieved from the system one by one. This value is calculated based on the agent's historical performance data and dynamic trust trajectory. Simultaneously, the platform trust threshold value of the target large model service platform is extracted from the database. The two values ​​are compared one-to-one. If the dynamically adjusted trust level value of a candidate industrial intelligent agent is greater than or equal to the platform trust threshold value, it indicates that the agent's trust level meets the target platform's trust requirements for the task execution subject, reducing task execution risk, and the agent is included in the candidate range. If the dynamically adjusted trust level value of a candidate industrial intelligent agent is less than the platform trust threshold value, it indicates that its trust level does not meet the platform requirements. Such agents have a higher probability of errors or compliance issues when executing tasks, and are therefore excluded from consideration. If there is a risk in task execution, the agent is directly excluded from the subsequent selection. For example, if the target platform's trust threshold is 60, and a candidate agent's dynamically adjusted trust level is 63.5, which is greater than 60, it means that its current trust status meets the platform's requirements and it is included in the candidate range. Another candidate agent's dynamically adjusted trust level is 58, which is less than 60, indicating that its trust status does not meet the standard and there is an execution risk, so it is excluded. After all candidate industrial agents have been compared with the platform's trust threshold, the information of all agents included in the candidate range is summarized. The system retrieves the unique identifiers, functional descriptions, and the dynamically calculated trust levels of these agents stored in the distributed ledger. These are then organized into a structured list according to the correspondence between the unique identifier, functional description, and current trust level. This list, which contains information on all agents that meet the platform's trust requirements, is the distributable agent set.

[0054] This embodiment, by acquiring historical performance data from the candidate industrial agent's latest multiple task executions, can more comprehensively reflect the agent's recent actual performance, avoiding inaccurate performance judgments due to limited data. Constructing a two-dimensional trust evaluation space transforms historical performance data into intuitive coordinate points, solving the problem of scattered data and difficulty in intuitive analysis in static evaluation. Setting five evaluation nodes with different functions covers the agent's possible performance range and trust boundaries, enabling multi-dimensional capture of the agent's performance characteristics and avoiding the one-sidedness of evaluation caused by focusing only on a single performance point. Generating a dynamic trust trajectory intuitively displays the agent's performance change patterns over time, providing a more comprehensive view compared to static evaluation. By reflecting performance fluctuations, it can promptly detect rising or falling trends in agent performance, making the assessment more timely. Based on dynamic trust trajectories, it calculates credibility correction parameters, quantifying the dynamic changes in agent performance into parameters that can be used to adjust credibility. This solves the problem that static assessments cannot adjust trust judgments based on real-time performance changes, making credibility assessments more closely reflect the actual operating state of the agent. Through weighted fusion, it dynamically adjusts the credibility level, enabling the credibility level to dynamically adapt to changes in agent performance, improving assessment accuracy. It filters distributable agents based on the platform's credibility threshold, ensuring that the selected agents meet the target platform's trust requirements, effectively excluding agents with insufficient credibility, and reducing task execution risks.

[0055] In a preferred embodiment of the present invention, step 3 includes:

[0056] In this embodiment of the invention, step 300, obtaining the current credibility level and real-time performance indicators of all industrial intelligent agents in the distributable intelligent agent set, specifically includes: clarifying that the distributable intelligent agent set is a list of industrial intelligent agents selected in step 206 whose credibility level values ​​meet the threshold requirements of the target large model service platform, and the list has recorded the unique identifier of each intelligent agent; for each industrial intelligent agent in the set, retrieving the current credibility level from the system database of the large model service platform based on its unique identifier. This level is the credibility level after dynamic adjustment in step 205, which can reflect the latest credibility status of the intelligent agent. When retrieving, it is necessary to confirm whether the value corresponding to the level and the level description are complete. For example, if the unique identifier of an intelligent agent is "XX Technology-AGV-123456", its current credibility level is retrieved as level four, and the corresponding value is 70; at the same time, retrieving the real-time performance indicators of each intelligent agent. These metrics are real-time data on the current operating status of the agent, including real-time task response speed (time from receiving the test command to feedback preparation completion), real-time resource utilization (the percentage obtained by dividing the current actual CPU and memory usage by the rated total), and real-time online status (whether it is in normal standby or executable task state, without fault alarms). They are obtained through the real-time data transmission channel between the platform and the agent during retrieval, ensuring that the data is the latest data within 1 minute before the retrieval time, avoiding the use of outdated performance data to affect subsequent scheduling judgments. For example, if the real-time response speed of "XX Technology-AGV-123456" is 2 seconds, the real-time CPU utilization is 30%, and the real-time online status is normal, the unique identifier, current trust level, including numerical values, and real-time performance metrics of each agent are organized one by one to form a data table of identifier, trust level, and performance association for distributable agents.

[0057] Step 301a: Extract the dynamically adjusted trust level value of each industrial agent in the distributable agent set; sort all trust level values ​​in descending order to generate a preliminary sorting sequence, specifically including: extracting the dynamically adjusted trust level value of each industrial agent from the identifier, trust level, and performance association data table organized in step 300, for example, the extracted values ​​are 85, 70, 92, 70, and 88, while ensuring that each value is bound to the corresponding agent's unique identifier; arranging all extracted trust level values ​​in descending order, i.e., sorting them in descending order, first finding the maximum value among all values ​​and placing the corresponding agent first; then... Find the second largest value from the remaining values ​​and place the corresponding agent in the second position. Continue in this manner until all agents are sorted according to their corresponding values. For example, the largest value is 92, and the corresponding agent is in the first position; the second largest value is 88, and the corresponding agent is in the second position; then it is 85, and the corresponding agent is in the third position; finally, there are two 70s, and the corresponding agents are in the fourth and fifth positions. This sorting result forms a preliminary sorting sequence. Each position in the sequence records the unique identifier of the agent and its corresponding credibility level value. For example, the preliminary sorting sequence is identifier A (92), identifier B (88), identifier C (85), identifier D (70), and identifier E (70).

[0058] Step 301b: For industrial agents with the same confidence level in the initial sorting sequence, obtain their recorded historical task success rates. Sort the industrial agents with the same confidence level in a second descending order based on their historical task success rates, placing those with higher historical task success rates first, thus generating the final ordered set of distributable agents. Specifically, this includes: first, checking the initial sorting sequence to identify groups of industrial agents with the same confidence level. For example, in the initial sequence, identifiers D and E both have a confidence level of 70, forming a group of agents with the same value. For each group of agents with the same value, retrieve their historical task success rate from the historical task execution record database based on their unique identifier. The historical task success rate is the percentage obtained by dividing the number of successful tasks (or actual executions) in the latest 100 tasks counted in step 200 by the total number of tasks. For example, identifier D has a historical task success rate of 95%, and identifier E has a historical task success rate of 95%. The historical task success rate is 90%. For agents within the same group, they are sorted in descending order of historical task success rate. Agents with higher historical task success rates are placed at the front of the group. For example, if 95% of identifier D is higher than 90% of identifier E, then after the second sorting, identifier D is placed before identifier E, and the group order becomes identifier D (70, 95%) and identifier E (70, 90%). All groups that have completed the second sorting are integrated according to the order in the original initial sorting sequence. Agents with the same value do not remain in their original positions. Finally, a complete ordered set of distributable agents is formed. For example, in the example above, the integrated ordered set is identifier A (92), identifier B (88), identifier C (85), identifier D (70, 95%), and identifier E (70, 90%). The set clearly records the unique identifier, credibility level value, and historical task success rate of each agent (only groups with the same value need to be recorded).

[0059] Step 302: Select the industrial intelligent agent with the highest credibility level from the sorted set of distributable intelligent agents as the target industrial intelligent agent. This specifically includes: viewing the final ordered set of distributable intelligent agents, determining the industrial intelligent agent ranked first in the set, which is the intelligent agent with the highest credibility level (or the highest historical task success rate when the values ​​are the same), and meeting the core requirements of the preset credibility priority scheduling strategy. For example, the first one in the ordered set is identifier A, whose credibility level is 92, which is the highest among all intelligent agents. Therefore, the industrial intelligent agent corresponding to identifier A is determined as the target industrial intelligent agent. After determining the target industrial intelligent agent, its unique identifier, current credibility level, real-time performance indicators, and historical task success rate are retrieved again for information verification to ensure that the information of the selected intelligent agent is consistent with the records in the ordered set, and that the real-time performance indicators meet the basic requirements of the current task, such as being online in real time and having resource utilization not exceeding the warning value, to avoid selection errors due to information errors or changes in real-time status. After verification, the intelligent agent is officially confirmed as the target industrial intelligent agent.

[0060] Step 303: Based on the unique identifier of the target industrial intelligent agent and the task description in the task distribution request, generate a structured distribution instruction. This includes: first, extracting the unique identifier from the confirmed target industrial intelligent agent information, ensuring the identifier is accurate. For example, the extracted unique identifier might be "XX Technology-AGV-123456," which will be used to accurately locate the corresponding intelligent agent in the target large-scale model service platform; then, extracting the task description information from the task distribution request parsed in Step 102. The task description includes the specific task content, such as transferring 10 300kg shelves from warehouse A to warehouse B; the task completion deadline, such as before 14:00 on August 28, 2024; the task execution standards, such as the shelf tilt angle not exceeding 5 degrees during the transfer; and the task docking node, such as completing data interaction with the shelf receiving device in warehouse B. This ensures that the extracted task description is complete and covers all key information. The system covers all requirements necessary for task execution; it generates distribution instructions according to a preset structured format, which must include fixed fields: target agent identifier, task number, task content, completion deadline, execution standard, integration requirements, and instruction generation time. The target agent identifier is a unique identifier extracted from the system; the task number is generated by the platform according to the rule "date - 6 random digits"; the task content, completion deadline, execution standard, and integration requirements are filled with the corresponding information extracted from the task description; the instruction generation time is the specific time the instruction was generated, such as 9:00:00 on August 28, 2024. After generating the instruction, the content is validated to check for missing fields or errors, such as incorrect identifiers or unclear deadlines. If the validation is successful, a platform-specific digital signature is added to the instruction (used by the target platform to verify the instruction's authenticity), forming the final structured distribution instruction.

[0061] Step 304 involves sending the structured distribution instruction to the target large-scale model service platform specified in the task distribution request. Specifically, this includes: extracting the identifier of the target large-scale model service platform from the task distribution request parsed in step 102, such as the platform's unique ID C001 or the platform name "XX Warehouse Intelligent Service Platform," thus identifying the instruction receiving platform; establishing a connection with the target large-scale model service platform through a pre-set secure data transmission channel between platforms, using encrypted transmission to prevent instruction tampering or leakage; after the connection is established, first sending an instruction transmission request to the target platform; and then, after the target platform returns a response signal indicating readiness to receive, transmitting the complete structured distribution instruction to the target platform; and finally, receiving the instruction reception confirmation information returned by the target platform. The confirmation message must include the time the target platform received the instruction, the instruction number, and the reception status. If the instruction was successfully received, the instruction number is 20240828-123456. If no confirmation message is received within the preset time, such as 30 seconds, or if the confirmation message shows that the reception failed, the instruction must be resent, and the number of resentments must be recorded (maximum of 3 resentments). If the instruction is still unsuccessful after 3 resentments, a notification of instruction transmission error should be sent to the platform administrator, and the error should be recorded, such as 3 failed attempts to send the instruction to platform C001 between 9:01 and 9:02 on August 28, 2024. After receiving the successful reception confirmation message from the target platform, the instruction sending time, target platform identifier, instruction number, reception confirmation time, and other information should be recorded in the system log.

[0062] This embodiment, by acquiring the agent's current trust level and real-time performance metrics, ensures that the scheduling criteria align with the agent's latest state, avoiding scheduling errors caused by using outdated data. Through two descending sorts—first by trust level value and then by historical task success rate—it prioritizes agents with higher trust levels and better past performance. Compared to scheduling methods without clear sorting criteria, this makes the scheduling logic more transparent, the selection criteria more comprehensive, and reduces scheduling arbitrariness. It generates structured distribution instructions, clearly defining task requirements and agent identifiers, avoiding execution deviations caused by inconsistent instruction formats or missing information, ensuring that the target platform and the agent clearly understand the task requirements. Sending instructions through a secure channel and confirming receipt ensures the security and traceability of instruction transmission.

[0063] In a preferred embodiment of the present invention, step 4 includes:

[0064] In this embodiment of the invention, step 400 involves receiving the status information of the target industrial intelligent agent performing the task returned by the target large model service platform. The status information includes resource utilization data, task execution latency data, result accuracy data, and anomaly handling record data. Specifically, this includes: clarifying that the target large model service platform will collect the target industrial intelligent agent's operating status data in real time throughout the entire process of the target industrial intelligent agent's task execution, and after the task execution is completed (or according to a preset period, such as every 20% of the task progress is completed), feeding back the summarized status information to the platform that initiated the task distribution through a secure data transmission channel. When receiving the status information, the integrity and authenticity of the data transmission are first verified, and it is checked whether the feedback information contains the four core data types: resource utilization data, task execution latency data, result accuracy data, and anomaly handling record data. At the same time, it is verified whether the unique identifier of the target industrial intelligent agent and the task number in the data are consistent with the previously distributed instruction information to avoid receiving status data from the wrong platform or the wrong task.

[0065] Among them, resource utilization rate data refers to the actual percentage of CPU, memory, and energy consumption used by the target industrial intelligent agent during task execution. For example, the average CPU utilization rate is 60%, the peak memory utilization rate is 75%, and the energy consumption per unit time is 1.2 kWh. Task execution latency data refers to the total time from when the intelligent agent receives the instruction to when it completes the task, as well as the time spent on each sub-stage of the task, such as data acquisition, processing, and result feedback. For example, the total latency is 15 minutes, of which data processing takes 8 minutes. Result accuracy data refers to the degree of matching between the task results output by the intelligent agent and the preset standard results. For example, in a parts inspection task, 98 qualified parts and 2 unqualified parts are correctly identified, and a total of 100 parts are inspected. The accuracy data is the ratio of 98 / 100. Anomaly handling record data refers to abnormal situations that occur during task execution, such as equipment jamming, data loss, and the intelligent agent's response measures. For example, if a data transmission jam occurs in the 10th minute, the intelligent agent automatically switches to a backup transmission channel, with a recovery time of 2 minutes. After confirming that the data is complete and matched, the four types of status data are organized and stored according to the task number, intelligent agent identifier, data type, and specific value / record format.

[0066] Step 401 involves analyzing resource utilization data to obtain the deviation between resource utilization and preset resource thresholds; analyzing task execution latency data to obtain the degree of conformity between task execution latency and preset latency standards; analyzing result accuracy data to obtain the degree of achievement of result accuracy with preset accuracy benchmarks; and analyzing anomaly handling record data to obtain the degree of matching between anomaly handling and preset anomaly handling specifications. Specifically, this includes: first, determining preset resource thresholds, which are upper limits for resource usage set based on the hardware configuration and task requirements of the target industrial intelligent agent, such as a CPU utilization threshold of 70% and a memory utilization threshold of 80%; and then averaging the collected resource utilization data. If there are multiple types of resources, they need to be calculated separately, for example, an average CPU utilization of 60%. The average memory utilization rate is 75%. The deviation value is calculated by subtracting the corresponding preset resource threshold from the average utilization rate of each resource type. If the result is positive, it means the resource utilization exceeds the threshold, and the deviation value is positive; if the result is negative, it means the resource utilization is below the threshold, and the deviation value is negative; if it equals the threshold, the deviation value is 0. For example, subtracting the threshold from the average CPU utilization rate of 60% gives a deviation value of -10%; the average memory utilization rate of 75% - the threshold of 80% = deviation value -%. Finally, a weighted average is taken for the deviation values ​​of each resource type (with weights set according to resource importance, such as CPU weight 0.6 and memory weight 0.4), to obtain the comprehensive resource utilization deviation value. For example, (-10% × 0.6) + (-5% × 0.4) results in -8%, which is the comprehensive deviation value of -8%.

[0067] First, determine the preset latency standard, which is the maximum allowable time to complete the task. If the task contains multiple sub-stages, such as data acquisition, data processing, and result feedback, a latency standard must be set for each sub-stage. For example, the standard for data acquisition could be 5 minutes, data processing 10 minutes, and result feedback 5 minutes. Simultaneously, assign weights to each sub-stage based on its impact on task completion. The weight values ​​should be determined according to the importance of the sub-stage; for example, the data processing sub-stage directly determines the quality of the task result and has the greatest impact, so its weight is set to 0. 0.5; The data acquisition sub-stage is the basic stage, with a weight of 0.3; the result feedback sub-stage has a smaller impact, with a weight of 0.2, and the sum of the weights of all sub-stages is 1; extract the actual time consumed by each sub-stage from the task execution latency data, for example, the actual time consumed by the data acquisition sub-stage is 4 minutes, the actual time consumed by the data processing sub-stage is 8 minutes, and the actual time consumed by the result feedback sub-stage is 3 minutes. First, calculate the degree of conformity between the latency of each sub-stage and the standard of the corresponding sub-stage. If the actual time consumed by a sub-stage is less than or equal to the standard of that sub-stage, divide the actual time consumed by the sub-stage by the standard of the corresponding sub-stage to obtain the degree of conformity of that sub-stage (the result is less than or equal to 1); if the actual time consumed by a sub-stage exceeds the standard of that sub-stage, the degree of conformity of that sub-stage is 0.

[0068] For example, the compliance rate of the data acquisition sub-stage is 0.8 when 4 minutes are divided by 5 minutes; the compliance rate of the data processing sub-stage is 0.8 when 8 minutes are divided by 10 minutes; and the compliance rate of the result feedback sub-stage is 0.6 when 3 minutes are divided by 5 minutes. The total compliance rate is then calculated by taking a weighted average based on the sub-stage weights. The compliance rate of each sub-stage is multiplied by its corresponding weight to obtain the weighted compliance rate of each sub-stage. All the weighted compliance rates of the sub-stages are then summed to obtain the total compliance rate of the task execution delay. If the task has no clearly defined sub-stages, the total time consumed in the task execution delay data is directly extracted, for example, 15 minutes. If the total time consumed is less than or equal to the preset total delay standard, for example, 20 minutes, the total time consumed is divided by the preset total delay standard to obtain the compliance rate (the result is less than or equal to 1). If the total time consumed exceeds the preset total delay standard, the compliance rate is 0; for example, 15 minutes ÷ 20 minutes yields a compliance rate of 0.75.

[0069] First, determine the preset accuracy benchmark, which is the minimum accuracy required for the task, such as 95%. Extract the actual accuracy from the result accuracy data, such as 98%. If the actual accuracy is greater than or equal to the preset accuracy benchmark, divide the actual accuracy by the preset accuracy benchmark to obtain the achievement level (if the result is greater than or equal to 1, calculate it as 1 to avoid excessively exceeding the benchmark and having too much impact on the result). If the actual accuracy is lower than the preset accuracy benchmark, divide the actual accuracy by the preset accuracy benchmark to obtain the achievement level (the result is less than 1), for example, 98% ÷ 95% = 1.03, calculate the achievement level as 1. If the actual accuracy is 90% divided by 95%, the achievement level is approximately 0.95. First, determine the preset exception handling specifications, which include common exception types and their corresponding handling methods. The standard handling process is as follows: for example, if data transmission is stuck, a backup channel must be switched within 5 minutes; if equipment malfunctions, an alarm signal must be sent immediately and the task must be suspended. Each abnormal situation and corresponding response in the abnormal handling record data is checked. If the handling response of an abnormal record fully complies with the specifications, it is scored as 1 point; if it partially complies, such as the handling response being correct but the time taken exceeds the specifications (e.g., the specifications require 5 minutes to recover, but it actually takes 6 minutes), it is scored as 0.5 points; if it does not comply at all, such as no handling or incorrect handling response, it is scored as 0 points. The total score of all abnormal records is calculated and divided by the total number of abnormal records to obtain the abnormal handling matching degree. For example, if there are 2 abnormal records, one gets 1 point and the other gets 0.5 points, the total is 1.5 points, and dividing by 2 gives a matching degree of 0.75.

[0070] Step 402a involves weighting and fusing the four parameters—deviation value, compliance, achievement, and matching degree—with preset weighting coefficients to generate a scalar behavior adjustment coefficient. Specifically, this includes setting preset weighting coefficients for the four parameters. The weights are determined based on the degree of influence of each parameter on the compliance of the agent's behavior. For example, the achievement degree of result accuracy directly determines task quality and has the highest weight, such as 0.4; the compliance degree of task execution latency affects efficiency and has the second highest weight, such as 0.3; the matching degree of anomaly handling affects risk control and has the next lowest weight, such as 0.2; and the deviation value of resource utilization affects resource efficiency and has the lowest weight, such as 0.1. The sum of the four weights is 1. For each... Each parameter undergoes preprocessing. The resource utilization deviation value needs to be converted into a positive indicator first (because a negative deviation value indicates that the resource utilization is below the threshold, which is better). The processing method is to use 1 + deviation value (if the deviation value is negative, 1 + deviation value is less than 1; if it is positive, 1 + deviation value is greater than 1, and it needs to be truncated to 1 to avoid exceeding the reasonable range). For example, if the deviation value is -6%, after processing it is 1 + (-6%) = 0.94. The compliance, achievement, and matching degrees are positive indicators of 0-1 and do not need to be processed. Multiply each preprocessed parameter by the corresponding weight coefficient to obtain four weighted results. Add the four weighted results together, and the sum is the scalar form of the behavior adjustment coefficient.

[0071] Step 402b: Obtain the original behavioral compliance data of the target industrial intelligent agent. Merge the behavioral adjustment coefficient with each indicator item in the original behavioral compliance data sequentially to obtain a set of updated preliminary indicator values. Specifically, this includes: retrieving the original behavioral compliance data of the target industrial intelligent agent from the historical database of the large model service platform, based on the target industrial intelligent agent's unique identifier. This data contains multiple indicator items, such as historical resource compliance rate, historical latency compliance rate, historical accuracy compliance rate, and historical anomaly handling compliance rate. Each indicator item is a value between 0 and 1. For example, if the original data is [0.85, 0.9, 0.95, 0.8], the value is calculated as: original indicator item value × (original data weight 0.5 + behavioral adjustment coefficient × behavioral adjustment coefficient weight 0). 5) Calculate in order of the indicators. For example, the first indicator, historical resource compliance rate (value 0.85), is calculated as 0.85×(0.5+0.869×0.5); the second indicator, historical latency compliance rate, is calculated as 0.9×(0.5+0.869×0.5); the third indicator, historical accuracy compliance rate, is calculated as 0.95×(0.5+0.869×0.5); the fourth indicator, historical anomaly handling compliance rate, is calculated as 0.8×(0.5+0.869×0.5). Arrange the values ​​of all the indicators in the original order to obtain a set of updated preliminary indicator values, such as [0.794, 0.841, 0.888, 0.747].

[0072] Step 402c involves normalizing the updated preliminary indicator values ​​to ensure all values ​​are within a closed interval of 0 to 1, ultimately generating an updated behavior compliance dataset. This process includes: identifying the maximum and minimum values ​​among the updated preliminary indicator values. For example, in the preliminary indicator values ​​[0.794, 0.841, 0.888, 0.747], the maximum value is 0.888 and the minimum value is 0.747. If the maximum value equals the minimum value (all indicator values ​​are the same), then all indicator values ​​remain unchanged (already within the 0-1 interval). If the maximum value does not equal the minimum value, normalize each indicator value using the formula (indicator value - minimum value) ÷ (maximum value - minimum value) to ensure the result falls within the 0-1 closed interval. All normalized indicator values ​​are then rearranged according to the original indicator item order to form the updated behavior compliance dataset, for example, [0.333, 0.943, 1, 0], where each value is within the 0-1 closed interval.

[0073] Step 403: Based on the updated behavioral compliance dataset, and combined with preset weights, comprehensively calculate the various compliance indicators to redetermine the credibility level of the target industrial intelligent agent. Specifically, this includes: assigning preset weights to each indicator in the updated behavioral compliance dataset, with the weights matching the importance of the indicator. For example, the historical accuracy compliance rate, which affects task result quality, has the highest weight (e.g., 0.4); the historical latency compliance rate, which affects efficiency, has the next highest weight (e.g., 0.3); the historical anomaly handling compliance rate, which affects risk control, has the next lowest weight (e.g., 0.2); and the historical resource compliance rate, which affects resource efficiency, has the lowest weight (e.g., 0.1). The total weights are 1. The normalized value of each compliance indicator is multiplied by its corresponding preset weight to obtain the values ​​for each indicator. The weighted scores are calculated as follows: for example, index 1 (0.333) × weight 0.1; index 2 (0.943) × weight 0.3; index 3 (1) × weight 0.4; index 4 (0) × weight 0.2. The weighted scores of all indicators are added together to obtain the comprehensive score. The credibility level is divided according to the range of the comprehensive score. The division standard is the same as in step 205 (e.g., a comprehensive score of 0.9 and above is level 1, 0.8-0.89 is level 2, 0.7-0.79 is level 3, 0.6-0.69 is level 4, and below 0.6 is level 5). For example, the comprehensive score of 0.7162 is in the range of 0.7-0.79. Therefore, the credibility level of the target industrial intelligent agent is redefined as level 3. The credibility level is updated, and the updated level and calculation process are recorded in the system database.

[0074] This embodiment comprehensively receives multi-dimensional state information of the target agent. Compared to focusing only on task results and ignoring process data, it can fully grasp the agent's performance throughout the entire task execution process. It calculates parameters such as deviation value and compliance for four types of state data, transforming the data into quantifiable evaluation indicators. This makes the agent's behavior evaluation more accurate, avoiding evaluation bias caused by a single analysis dimension and compensating for the lack of quantitative standards in evaluation. By generating behavior adjustment coefficients through weighted fusion, it comprehensively considers multi-dimensional behavioral performance, making behavior adjustment more reasonable compared to single-dimensional adjustment methods. Based on updated compliance data, it recalculates the credibility level, achieving dynamic iteration of credibility evaluation. Compared to static evaluation that cannot be updated, it can reflect the impact of changes in agent behavior on credibility in real time.

[0075] In a preferred embodiment of the present invention, step 5, which records the task distribution request, the target industrial intelligent agent, the distribution instruction, and the task execution status information into a distributed ledger to form a task distribution and execution record, may include:

[0076] In this embodiment of the invention, the scope of core information that needs to be recorded in the distributed ledger is clearly defined to ensure coverage of key node data throughout the task distribution process. Specifically, this includes the task distribution request initiated by the task requester, containing task description, execution requirements, deadline, and initiation time; the target industrial intelligent agent information determined in step 302, including a unique identifier, current trust level, and functional matching description; the distribution instruction generated in step 303, including the instruction's structured content, instruction generation time, and digital signature information; and the task execution status information returned by the target large-scale model service platform in step S4, including four types of core data such as resource utilization, task execution latency, result accuracy, and exception handling records, as well as the status information feedback time. The above four types of information are standardized in format, uniformly categorized by evidence number and information type. The structure includes type, specific content, timestamp, and associated identifier. The evidence storage number is generated according to the rule of "task number + distributed ledger node identifier + random 6-digit number". For example, if the task number is "T20240828001" and the node identifier is "Node05", the evidence storage number is set as "T20240828001-Node05-123456". The information types are respectively labeled as task distribution request, target industrial intelligent agent information, distribution instruction, and task execution status information. The specific content must retain the integrity of the original information. For example, the task execution status information must record the specific values ​​such as the average CPU utilization and total latency. The anomaly handling record must retain the details such as the time of the anomaly and the response measures. The timestamp is accurate to the second and is consistent with the information generation / acquisition time.

[0077] The associated identifier is filled with the unique number of the corresponding task, ensuring that the four types of information are associated through the task number to form a complete process data chain; the standardized evidence information is verified through the consensus mechanism of the distributed ledger (such as the PBFT consensus algorithm). Multiple nodes in the ledger (at least 3 or more) verify the authenticity of the information, such as verifying whether the digital signature of the distribution instruction matches the initiating platform, whether the task execution status information is consistent with the actual performance of the target industrial intelligent agent, and its completeness, such as checking whether any core data in the task execution status information is missing. After all participating nodes have reported that the verification has passed, the evidence information is confirmed to meet the ledger storage requirements; the verified information is then... The evidence storage information is packaged and stored according to the block structure of the distributed ledger. Each block contains a block header, including the hash value of the previous block, the hash value of the current block, the block generation time, and the block body, which contains the four types of evidence storage information to be stored. The hash value of the current block is generated by calculating the block body information using the SHA-256 algorithm to ensure that the information cannot be tampered with after storage. At the same time, the block is synchronized to all nodes of the distributed ledger to realize multi-node backup of the evidence storage information, avoiding the loss of evidence storage due to the failure of a single node. Feedback information on the completion of evidence storage is generated, including the evidence storage number, block height, i.e., the position of the block containing the evidence storage in the ledger, and query address, and is fed back to the task requester and the target large model service platform.

[0078] like Figure 2 As shown, embodiments of the present invention also provide a trusted industrial intelligent agent task distribution system, comprising:

[0079] The registration module is used to receive registration requests sent by industrial intelligent agent developers, store the metadata information of industrial intelligent agents in the distributed ledger based on the registration requests, and receive task distribution requests sent by task requesters based on the metadata information, and match candidate industrial intelligent agents and unique identifiers that meet the functional requirements in the distributed ledger.

[0080] The screening module is used to obtain the historical performance data of each candidate industrial intelligent agent based on a unique identifier as an evaluation benchmark to construct a trust evaluation space. Five evaluation nodes are set in the trust evaluation space, and a dynamic trust trajectory is generated based on the five evaluation nodes. The credibility correction parameter is calculated based on the dynamic trust trajectory. The credibility level is dynamically adjusted through the credibility correction parameter, and distributable intelligent agents are selected from the candidate industrial intelligent agents based on the adjusted credibility level and the platform credibility threshold corresponding to the identifier of the target large model service platform.

[0081] The scheduling module is used to determine the target industrial intelligent agent from the distributable intelligent agents based on a preset scheduling strategy, generate distribution instructions, and send them to the target large model service platform.

[0082] The update module is used to perform multi-dimensional analysis on the status information of the execution and distribution instructions returned by the target large model service platform to generate behavior adjustment coefficients, update the behavior compliance data of the target industrial intelligent agent based on the behavior adjustment coefficients, and recalculate the credibility level of the target industrial intelligent agent.

[0083] The evidence storage module is used to record task distribution requests, target industrial intelligent agents, distribution instructions, and task execution status information into a distributed ledger to form task distribution and execution evidence storage.

[0084] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0085] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0086] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0087] The technical solutions (including task distribution methods, system modules, etc.) involved in this patent are merely technical innovations and scheme descriptions, aiming to provide technical ideas for intelligent agent task management in industrial scenarios. They do not constitute direct operational guidance or quality commitments for any specific industrial production activities. Actual applications require further verification and adaptation in conjunction with specific industrial scenarios, equipment parameters, and safety specifications. The technologies mentioned in the solution, such as "distributed ledger evidence storage" and "dynamic trust assessment," rely on legal and compliant data sources for data storage and processing. Any privacy leaks or legal liabilities resulting from users failing to obtain authorization from data subjects, using illegal data, or violating data security regulations shall be borne by the users themselves and are unrelated to the technical solutions of this patent. This patent does not involve the collection, storage, or transmission design of any real user privacy information. The "metadata" and "status information" in the solution are technical parameter data for the operation of industrial intelligent agent equipment (not personal privacy data). If users use the solution for personal privacy data processing without authorization in actual applications, the resulting privacy infringement risks and legal consequences shall be borne independently by the users. The effectiveness of this patented technical solution (such as the reliability and traceability of task execution) is based on ideal technical environment and standard parameter assumptions. In actual application, it may be affected by factors such as hardware performance, network environment, and complexity of industrial scenarios, resulting in different effects. This patent is not responsible for the application results caused by non-technical factors.

[0088] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A trusted industrial agent task distribution method, characterized in that, The method comprises: Step 1, receiving a registration request sent by an industrial agent developer, storing metadata information of the industrial agent into a distributed ledger based on the registration request; based on the metadata information, receiving a task distribution request sent by a task requester, and matching candidate industrial agents and unique identifiers that meet the functional requirements in the distributed ledger; Step 2, based on the unique identifier, obtaining the historical performance data of each candidate industrial agent as an evaluation benchmark point to construct a trust evaluation space; five evaluation nodes are set in the trust evaluation space, a dynamic trust trajectory is generated according to the five evaluation nodes, and a trust correction parameter is calculated based on the dynamic trust trajectory; the trust level is dynamically adjusted through the trust correction parameter, and the platform trust threshold corresponding to the target large model service platform is compared based on the adjusted trust level and the identification, to screen the distributable agents from the candidate industrial agents, including: Step 200, according to the unique identifier of each candidate industrial agent, obtaining the historical performance data of each candidate industrial agent in the latest 100 times of task execution from the historical task execution record library, wherein the historical performance data includes task success rate, task execution delay and resource utilization rate; Step 201, taking the historical performance data as an evaluation benchmark point to construct a two-dimensional trust evaluation space, wherein the horizontal axis represents the time dimension and the vertical axis represents the task success rate index dimension; Step 202, five evaluation nodes are set in the two-dimensional trust evaluation space, wherein the first evaluation node is located at the center position of the high performance stable area with task success rate greater than or equal to 95%, the second evaluation node is located at the center position of the performance fluctuation area with task success rate between 85% and 94%, the third evaluation node is located at the center position of the low performance area with task success rate less than 85%, the fourth evaluation node is located at the internal critical position of the trust boundary with task success rate of 90%, and the fifth evaluation node is located at the external critical position of the trust boundary with task success rate of 80%; Step 203, a dynamic trust trajectory is generated according to the continuous change relationship of the five evaluation nodes in the time dimension; Step 204, a trust correction parameter is calculated based on the curvature change rate of the dynamic trust trajectory, wherein the curvature change rate is obtained by the ratio of the tangent angle change amount to the time change amount of the trajectory curve; Step 205, the initial trust level of each candidate industrial agent is determined based on the historical performance data, the trust correction parameter is weighted and fused with the initial trust level to obtain the dynamically adjusted trust level; Step 206, compare the dynamically adjusted trust level of each candidate industrial agent with the platform trust threshold of the target large model service platform, and select the industrial agents with trust level value greater than or equal to the platform trust threshold value to form a distributable agent set; Step 3, determining a target industrial agent from the distributable agents based on a preset scheduling strategy, generating a distribution instruction, and sending it to the target large model service platform; Step 4: Perform multi-dimensional analysis on the state information of the execution distribution instruction returned by the target large model service platform to generate a behavior adjustment coefficient, update the behavior compliance data of the target industrial agent based on the behavior adjustment coefficient, and recalculate the credibility level of the target industrial agent; Step 5: Record the task distribution request, target industrial agent, distribution instruction, and task execution state information to the distributed ledger to form a task distribution and execution record.

2. The trusted industrial agent task distribution method of claim 1, wherein, The step 1 includes: Step 100: Receive a registration request sent by an industrial agent developer, the registration request containing a unique identifier of the industrial agent, function description information, and identity verification credentials; Step 101: Verify the validity of the identity verification credentials, and if verified, extract the unique identifier and function description information as metadata information, and add the metadata information to the agent registration block of the distributed ledger; Step 102: Based on the metadata in the agent registration block, receive a task distribution request sent by a task requester, parse the task distribution request to obtain task description, function requirements of the required industrial agent, and identification of the target large model service platform; Step 103: Compare the function requirements with the function description information of all registered industrial agents in the distributed ledger, and select industrial agents whose function description information contains all necessary function items in the function requirements as candidate industrial agents, and obtain the unique identifiers of each candidate industrial agent.

3. The trusted industrial agent task distribution method of claim 2, wherein, The step 3 includes: Step 300: Obtain the current credibility level and real-time performance indicators of all industrial agents in the distributable agent set; Step 301: According to the preset credibility priority scheduling strategy, sort the industrial agents in the distributable agent set from high to low according to the credibility level to obtain the sorted distributable agent set; Step 302: Select the industrial agent with the highest credibility level from the sorted distributable agent set as the target industrial agent; Step 303: Based on the unique identifier of the target industrial agent and the task description in the task distribution request, generate a structured distribution instruction; Step 304: Send the structured distribution instruction to the target large model service platform specified in the task distribution request.

4. The trusted industrial agent task distribution method of claim 3, wherein, The step 301 includes: Step 301a: Extract the credibility level value of each industrial agent in the distributable agent set after dynamic adjustment; arrange all credibility level values in descending order to generate a preliminary sorting sequence; Step 301b: For industrial agents with the same credibility level value in the preliminary sorting sequence, obtain the recorded historical task success rate, and arrange the industrial agents with the same credibility level value in descending order according to the historical task success rate, with the industrial agent with the higher historical task success rate arranged in front, to generate a final ordered distributable agent set.

5. The trusted industrial agent task distribution method of claim 4, wherein, The step 4 includes: Step 400: Receive the state information of the target industrial agent executing the task returned by the target large model service platform, the state information including resource occupancy rate data, task execution delay data, result accuracy rate data, and abnormal processing record data; Step 401, analyzing the resource occupancy data to obtain a deviation value of the resource occupancy from a preset resource threshold; analyzing the task execution delay data to obtain a compliance degree of the task execution delay from a preset delay standard; analyzing the result accuracy data to obtain an achievement degree of the result accuracy from a preset accuracy benchmark; analyzing the abnormal processing record data to obtain a matching degree of the abnormal processing from a preset abnormal processing specification; Step 402, generating a behavior adjustment coefficient through a weighted calculation method based on the deviation value, the compliance degree, the achievement degree and the matching degree; fusing the behavior adjustment coefficient with original behavior compliance data of the target industrial agent to obtain an updated behavior compliance data set; Step 403, according to the updated behavior compliance data set, combining the preset weight to comprehensively calculate each compliance index, and re-determining the credibility level of the target industrial agent.

6. The trusted industrial agent task distribution method of claim 5, wherein, The step 402 comprises: Step 402a, weighting and fusing the four parameters of the deviation value, the compliance degree, the achievement degree and the matching degree with preset weight coefficients respectively to generate a behavior adjustment coefficient in the form of a scalar; Step 402b, obtaining original behavior compliance data of the target industrial agent, fusing the behavior adjustment coefficient with each index item in the original behavior compliance data in turn to obtain a group of updated preliminary index values; Step 402c, normalizing the updated preliminary index values to constrain all index values in the closed interval of 0 to 1, and finally generating an updated behavior compliance data set.

7. A trusted industrial agent task distribution system, the system implementing the method of any one of claims 1 to 6, characterized in that, Comprise: The registration module is used for receiving a registration request sent by an industrial agent developer, and storing metadata information of the industrial agent into a distributed ledger based on the registration request; Based on the metadata information, receiving a task distribution request sent by a task requester, and matching candidate industrial agents and unique identifiers that meet the functional requirements in the distributed ledger; The screening module is used for obtaining historical performance data of each candidate industrial agent as an evaluation benchmark based on the unique identifier, so as to construct a trust evaluation space; Five evaluation nodes are set in the trust evaluation space, a dynamic trust trajectory is generated according to the five evaluation nodes, and a credibility correction parameter is calculated based on the dynamic trust trajectory; the credibility level is dynamically adjusted through the credibility correction parameter, and based on the adjusted credibility level and the platform credibility threshold corresponding to the identifier of the target large model service platform, a distributable agent is screened from the candidate industrial agents; The scheduling module is used for determining a target industrial agent from the distributable agents based on a preset scheduling strategy, generating a distribution instruction, and sending it to the target large model service platform; The updating module is used for generating a behavior adjustment coefficient by multi-dimensional analysis on state information of the execution of the distribution instruction returned by the target large model service platform, updating behavior compliance data of the target industrial agent based on the behavior adjustment coefficient, and re-calculating the credibility level of the target industrial agent; The evidence storage module is used for recording the task distribution request, the target industrial agent, the distribution instruction and the task execution state information to the distributed ledger to form a task distribution and execution evidence.

8. A computing device, comprising: Comprise: One or more processors; a memory device for storing one or more programs that, when executed by the one or more processors, cause the one or more processors to implement a method recited in any of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a program, which, when executed by a processor, implements a method recited in any of claims 1-6.

Images