An industrial body intelligent agent structure configuration generative agile design method, system and medium

Abstract

The application discloses an industrial body intelligent agent structure configuration generation type agile design method and system and a medium, comprising obtaining natural language requirement description, process constraints and site rules of the industrial body intelligent agent, forming a scene element set and constructing a scene graph; calculating a scene representation vector based on the scene graph, scoring the functions in the meta-function library with multiple labels, and outputting a meta-function set and its priority; inputting the scene element set and the meta-function set into an architecture element library and a graph matching model, automatically selecting components, and generating a candidate structure configuration; for the candidate structure configuration, multi-objective optimization and constraint solving are performed, and an optimal structure configuration and its architecture blueprint are output; the optimal structure configuration is verified and evaluated, and the verification result is written back to update its parameters. The application realizes the automation, standardization and agile design of the industrial body intelligent agent from the requirement description to the structure configuration, significantly shortens the cross-industry deployment cycle, and improves the scheme consistency, reusability and engineering landing performance.

Description

Technical Field

[0001] This invention relates to the field of robot system architecture design technology, and in particular to a generative agile design method, system and medium for industrial embodied intelligent agent structure configuration. Background Technology

[0002] Industrial embodied intelligent agents typically possess integrated characteristics of perception, cognition, planning, and execution, encompassing construction site safety inspection platforms, warehouse logistics robots, rebar tying equipment, industrial robotic arm collaborative operation platforms, and other integrated mobile operation systems. The structural configuration of such systems includes not only sensor and actuator configurations but also various engineering elements such as algorithm modules, computing platforms, communication links, safety interlocks, log traceability, and operation and maintenance interfaces. Therefore, its design directly determines whether the system can operate safely, stably, and cost-effectively in complex industrial environments.

[0003] In existing technologies, the structural design of industrial embodied intelligent agents mainly relies on domain experts manually analyzing requirements, selecting models based on experience, and repeatedly debugging. For the same task, when the working environment, process flow, or safety constraints change, it is often necessary to redefine the sensing methods, computing power configuration, interface relationships, and degradation strategies. This results in long design cycles, low knowledge reuse rates, and insufficient consistency in the solutions output by different designers, making it difficult to form a large-scale, standardized R&D model.

[0004] On the other hand, while existing model-driven or knowledge graph methods can describe system components, most lack the ability to automatically extract reusable meta-functions from natural language requirements, and also lack a mechanism for unified coupling modeling and solving of security, real-time performance, cost, robustness, and maintainability. Relying solely on fixed templates or single-scenario experience is insufficient to meet the requirements for rapid deployment and continuous iteration of embodied intelligent agents across multiple industries.

[0005] With the development of large language models and generative models, it has become possible to directly generate architectural descriptions from text. However, existing solutions generally suffer from problems such as unstable output content, unclear component selection criteria, disconnect from actual engineering interfaces, and lack of verification loops. The generated results often remain at the conceptual level and are difficult to directly form a deliverable structural configuration blueprint. Therefore, it is necessary to propose a generative agile design method for industrial embodied intelligent agent structural configuration that can connect "requirements description - meta-functions - architecture - verification loop". Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the existing technology. To achieve the above objectives, an agile design method, system and medium for the generative design of industrial embodied intelligent agent structure configuration is adopted to solve the problems mentioned in the background technology.

[0007] The technical solution provided in the first aspect is: a generative agile design method for industrial embodied intelligent agent structural configuration, including the following steps: Step S1: Obtain the natural language requirement description, process constraints and site rules of the industrial embodied intelligent agent, extract object elements, environmental elements, task elements, constraint elements and risk elements, form a scene element set and construct a scene diagram; Step S2: Calculate the scene representation vector based on the scene graph, perform multi-label scoring on the perception, localization, prediction, planning, control, safety and operation and maintenance functions in the meta-function library, and output the meta-function set and its priority. Step S3: Input the scene element set and meta-function set into the architecture element library and graph matching model, automatically select sensor, actuator, computing power, communication, security and operation and maintenance components, and generate at least one candidate structure configuration containing hierarchical modules, interface relationships, data flow and deployment relationships; Step S4: For the candidate structural configurations, perform multi-objective optimization and constraint solving based on security, real-time performance, cost, robustness and iteration cycle, and output the optimal structural configuration and its architecture blueprint. Step S5: Verify and evaluate the optimal structural configuration in a digital twin environment and a real industrial scenario. Write the verification results back to the meta-function library, architecture element library, and parameter model to update their parameters and form a design iteration closed loop.

[0008] As a further aspect of the present invention, the specific steps in step S1 include: Encode the requirement text using a pre-trained language model to obtain a global semantic vector of the requirement. ; The text fragments are classified to identify their element categories; based on at least one of the spatial, temporal, dependency, or risk propagation relationships between the elements, a scene graph is constructed with the elements as nodes and the relationships as edges. ; Among them, the node set O, E, T, C, and R represent object elements, environmental elements, task elements, constraint elements, and risk elements, respectively.

[0009] As a further aspect of the present invention: when classifying text fragments, the probability that each text fragment i belongs to element category c is calculated using a classification model. The calculation formula is as follows:

[0010] in, The semantic encoding of text fragment i. and For the classification model parameters corresponding to category c, u iterates through all preset element categories.

[0011] As a further aspect of the present invention: the generation of the target meta-function set in step S2 includes: calculating the scene representation vector based on the scene graph. ; Calculate the k-th candidate metafunction using a multi-label prediction model The recommended score is:

[0012] in, and For model parameters; at least the recommended score Candidate meta-functions exceeding a preset threshold are included in the target meta-function set; The generation of the target meta-function set also includes: based on the risk-related weights corresponding to the risk elements. The priority of computational meta-functions is as follows:

[0013] And priority is used as the basis for sorting or filtering.

[0014] As a further aspect of the present invention: the generation of candidate structural configurations in step S3 includes: The scene graph is fused with the target meta-function set to obtain a scene-meta-function fusion vector. ; The architectural element library is constructed as a heterogeneous graph. , where nodes The attribute vector includes at least one of the following: performance metric, interface type, power consumption, cost, and reliability; using the fused vector With the node attribute vector Calculate the matching score and use the matching score as the component corresponding to each function selection.

[0015] As a further aspect of the present invention: the candidate structural configuration Represented as

[0016] in, Represents a set of modules. Represents a collection of interfaces. Indicates data flow relationship, This represents a deployment matrix that describes the deployment locations of various functional modules on edge controllers, vehicle computing units, industrial gateways, cloud simulation platforms, or security controllers.

[0017] As a further aspect of the present invention: in step S4, a multi-objective optimization function J(A) is constructed, with functional satisfaction, security coverage, real-time performance score, and robustness score as gain terms, and cost and iteration cycle as penalty terms, while satisfying that the end-to-end latency is no greater than a threshold. Power consumption or computing power budget does not exceed the threshold. Safety coverage rate is not lower than the threshold. and the risk of downgrade is no greater than the threshold. The constraints.

[0018] As a further aspect of the present invention, step S5 specifically includes the following steps: The verification and evaluation adopts a combination of digital twins and real-world scenarios. The evaluation indicators include at least collision / intrusion response, security takeover success rate, end-to-end latency, critical link latency, performance degradation under dust / night / occlusion conditions, deployment cost, operation and maintenance efficiency, and iteration cycle. Based on the evaluation results, the priority of meta-functions, the posterior value of component reliability, and the structural configuration generation parameters are updated.

[0019] The second aspect provides a technical solution as follows: a design system including a generative agile design method for industrial embodied intelligent agent structural configuration as described in any of the above, comprising: The requirement analysis and scenario modeling module is used to obtain the natural language requirement description, process constraints and site rules of industrial embodied intelligent agents, extract object elements, environmental elements, task elements, constraint elements and risk elements, form a scenario element set and construct a scenario graph; The metafunction extraction and recommendation module is used to calculate the scene representation vector based on the scene graph, perform multi-label scoring on the perception, positioning, prediction, planning, control, safety and operation and maintenance functions in the metafunction library, and output the metafunction set and its priority. The architecture mapping and configuration generation module is used to input the scene element set and meta-function set into the architecture element library and graph matching model, automatically select sensor, actuator, computing power, communication, security and operation and maintenance components, and generate at least one candidate structural configuration containing hierarchical modules, interface relationships, data flow and deployment relationships. The optimization solution and blueprint output module is used to perform multi-objective optimization and constraint solution based on security, real-time performance, cost, robustness and iteration cycle for the candidate structural configuration, and output the optimal structural configuration and its architecture blueprint. The verification, evaluation, and knowledge write-back module is used to verify and evaluate the optimal structural configuration in a digital twin environment and a real industrial scenario, and write the verification results back to the meta-function library, architecture element library, and parameter model to update their parameters.

[0020] The third aspect provides a technical solution as follows: a storage medium storing processor-executable instructions, which, when executed by a processor, are used to implement the method described in any of the above-mentioned aspects.

[0021] Compared with the prior art, the present invention has the following technical advantages: First, by establishing an intermediate expression layer that connects requirement descriptions to scenario diagrams and then to meta-functional sets, design knowledge previously scattered across documents, experience, and verbal communication is transformed into computable objects, thereby improving the standardization of design inputs from the source. Second, through an architecture element library and heterogeneous graph mapping mechanism, rapid selection of sensor, computing power, communication, security, and operation and maintenance components is achieved, significantly shortening the time for scheme reconstruction after requirement changes. Third, through multi-objective coupling constraint solving and digital twin verification closed loops, generative results no longer remain at the conceptual level but can output deliverable architecture blueprints with module relationships, interface relationships, and key parameters. Fourth, through continuous back-writing of verification data from real-world scenarios across multiple industries, a design knowledge system that becomes more accurate and stable with use is formed. Attached Figure Description

[0022] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings: Figure 1 This is a flowchart of the industrial embodied intelligent agent structure configuration generative agile design method according to an embodiment of the present invention; Figure 2 This is a hierarchical generation architecture diagram of the industrial embodied intelligent agent structure configuration according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the closed loop for multi-industry scenario verification, evaluation, and knowledge rewriting in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please refer to Figure 1 In this embodiment of the invention, a generative agile design method for industrial embodied intelligent agent structural configuration includes the following steps: Step S1: Requirement semantic parsing and scenario structured representation.

[0025] Step S11: Encoding the requirement text. This involves encoding the requirement description text. Input a pre-trained language model and a Transformer encoder to obtain a demand context representation. Then, the required global vector is obtained through pooling operations. Global demand vector Calculate according to formula (1): (1) Here, Emb(·) represents the word embedding function, Encoder(·) represents the multi-layer semantic encoder, and Pool(·) represents the pooling operator. Through this step, the requirements of free text can be transformed into a unified semantic vector that can be used for subsequent structured computation.

[0026] Step S12: Feature Classification and Slot Filling. Candidate text fragments are mapped to five semantic slots: object feature O, environment feature E, task feature T, constraint feature C, and risk feature R. The probability that candidate fragment i belongs to the c-th semantic slot is calculated. Calculate according to formula (2): (2) in, and These are the classification parameters for the corresponding categories. `u` iterates through the five categories of labels: O, E, T, C, and R. Fragments with confidence levels higher than a preset threshold are retained, and attribute fields are added, including target category, quantity, spatial location, process constraints, risk level, and timing requirements.

[0027] Step S13: Scene Graph Construction. Based on the extracted scene elements and their relationships, a unified scene graph is constructed. , where the set of nodes According to the definition in equation (3): (3) edge set This includes spatial adjacency relationships, temporal sequence relationships, control dependencies, and risk propagation relationships. Through scene diagram representation, heterogeneous information such as "workers, equipment, fences, dust, nighttime, inspections, alarms, emergency stops, and coordinated responses" can be incorporated into a unified model to support subsequent function extraction and architecture solving.

[0028] In this embodiment, the source of requirements can be R&D task books, bidding documents, process specifications, or natural language instructions input by on-site personnel. Through the above semantic parsing process, descriptions from different sources and at different granularities can be unified into standardized, structured input for a specific scenario, thereby reducing the impact of misunderstandings of requirements on subsequent design work.

[0029] Step S2: Extraction, summarization, and recommendation generation of meta-functions.

[0030] Step S21: Scene Graph Encoding and Metafunction Scoring. The scene graph... The input graph encoder obtains the scene representation vector. And perform multi-label scoring on K types of metafunctions in the metafunction library. The k-th type of metafunction... Recommended score Calculate according to formula (4): (4) in, and These are the parameters corresponding to the k-th type of meta-function. The meta-function library covers functions in perception, positioning, prediction, planning, control, and security operations, such as detection, segmentation, tracking, GNSS / IMU fusion, SLAM, risk trend assessment, obstacle avoidance, trajectory tracking, emergency stop, alarm, auditing, and OTA.

[0031] Step S22: Priority Calculation. Considering the varying importance of the same function across different risk scenarios, risk-related weights corresponding to the risk elements are introduced. Priority of meta-functions The calculation is performed as shown in equation (5): (5) By calculating priorities as described above, meta-functions related to critical risk links can receive higher weights in the recommendation results. For example, in the construction site guardian scenario, meta-functions such as fence constraints, boundary crossing detection, risk alarms, and coordinated response have higher priorities; while in the industrial robotic arm assembly scenario, force-controlled assembly, work trajectory constraints, and safety protection have higher priorities.

[0032] Step S23: Comparison and expansion of meta-functional prototypes. When the scene representation vector When the similarity to existing prototypes in the meta-function prototype library is below a threshold, the generation of new meta-function combinations is triggered to avoid insufficient function coverage due to the novelty of the scenario. This mechanism enables the meta-function library to maintain stable reuse while continuously expanding with the introduction of new industry scenarios.

[0033] Step S3: Architectural element mapping and candidate structure configuration generation.

[0034] Step S31: Construct a heterogeneous diagram of architectural elements. Organize architectural elements such as sensors, actuators, computing platforms, communication links, security interlock modules, and operation and maintenance components into a heterogeneous diagram. Any architecture node attribute vector It should include at least performance metrics, interface type, power consumption, cost, reliability, and environmental adaptability level.

[0035] Step S32, Scene-Meta-Function Fusion Representation. The scene representation vector... Metafunctional representation vector Fusion yields a scene-meta-function fusion vector. According to equation (6): (6) The symbol || represents the vector concatenation operation. This fusion representation can simultaneously preserve both scenario constraints and functional requirements, providing a unified semantic basis for subsequent architectural element selection.

[0036] Step S33: Architectural Element Selection and Scoring. Based on the fusion vector. With node attribute vector Calculate node matching score As shown in equation (7): (7) Furthermore, the matching score is normalized to obtain the probability of a node being selected. As shown in equation (8): (8) in, , and To match model parameters, this step automatically sorts and combines components such as cameras, LiDAR, millimeter-wave radar, UWB / RTK, industrial Ethernet, 5G modules, edge computing units, security PLCs, log gateways, and remote maintenance interfaces.

[0037] Step S34: Candidate Structure Configuration Generation. Using the generative model and selection results, multiple candidate structure configurations are output. Each candidate structure configuration... It can be expressed as equation (9): (9) in, Represents a set of modules. Represents a collection of interfaces. Indicates data flow relationship, This represents the deployment matrix. The deployment matrix describes the deployment locations of functional modules such as sensing, planning, control, safety interlocking, and operation and maintenance on edge controllers, industrial gateways, cloud simulation platforms, and safety controllers. The generated candidate architecture also outputs a module diagram, interface list, key parameters, takeover rules, emergency stop rules, and degradation strategies to ensure the results are engineering deliverable.

[0038] Step S4: Solving and optimizing the structural configuration under multi-factor coupling constraints.

[0039] Step S41, Real-time and Cost Modeling. For any candidate structural configuration... Its end-to-end delay Calculate according to formula (10): (10) in, Indicates module processing latency. This represents the link transmission delay. Correspondingly, it represents the total cost of the structural configuration. Calculate according to formula (11): (11) in, Indicates module cost. This indicates the link or deployment cost. This step allows for a unified quantification of latency and cost across different architectures.

[0040] Step S42, Security and Robustness Modeling. This involves modeling the set of risk use cases. Introducing security coverage Its definition is shown in equation (12): (12) in, The weight of risk use case r is represented. This is an indicator function. It takes a value of 1 when the candidate configuration can cover the corresponding risk use case, and a value of 0 otherwise. Further, a real-time score... Normalization can be performed using equation (13): (13) Robustness scoring The calculation is based on the performance degradation under extreme conditions, as shown in equation (14): (14) in, This indicates the amount of performance degradation compared to normal conditions under extreme conditions such as dust, nighttime, obstruction, and high reflectivity. This represents the baseline performance under normal conditions.

[0041] Step S43: Multi-objective optimization solution. Construct the comprehensive utility function. As shown in equation (15): (15) in, Indicates the degree of functional satisfaction. This represents the iteration period of the proposed scheme, with a1 to a6 being adjustable weights. The final optimal structural configuration is A. According to equation (16), we can obtain: , (16) in, Let B(A) represent the set of candidate architecture configurations, and let B(A) represent the computing power or power consumption budget. This indicates the risk of degradation. Through the above constraint solution process, an optimal structural configuration scheme that balances cost, robustness, and iteration cycle can be output while meeting the safety and real-time requirements of industrial sites.

[0042] Step S44: Output the architecture blueprint. For the obtained optimal structural configuration A... The output includes a layered module diagram, interface list, key parameter table, deployment matrix, communication topology, fault takeover strategy, and emergency stop / degradation link description, so that R&D personnel, testers, and operations personnel can share the same structural configuration blueprint.

[0043] Step S5: Verification, evaluation, and data-driven iterative closed loop.

[0044] Step S51: Construction of verification evaluation indicators. A comprehensive verification score V(A) is established based on security, real-time performance, robustness, and engineering feasibility, as shown in Equation (17): (17) in, This indicates the score for safety indicators. This indicates the score for the real-time performance indicator. Indicates the robustness index score , represents the engineering metric score, with b1 to b4 being the corresponding weights. Safety metrics include at least collision / intrusion response, takeover success rate, and anomaly handling accuracy; real-time metrics include at least end-to-end latency and critical link latency; robustness metrics include at least performance degradation under dust, nighttime, and occlusion conditions; and engineering metrics include at least deployment cost, operational efficiency, and iteration cycle.

[0045] Step S52, Knowledge Base Update. The reliability posterior values ​​of each architectural component are updated based on the verification results, as shown in equation (18): (18) in, This represents the component reliability estimate at time t. This represents the reliability observation value obtained based on the latest verification results. This represents the smoothing coefficient. Meanwhile, the generated parameters can be iteratively updated based on the feedback loss, as shown in equation (19): (19) in, Indicates the learning rate. This represents the feedback loss function constructed based on verification errors, missed risks, and deployment costs. Through the aforementioned write-back process, the meta-function library and architectural element library can continuously evolve, thereby improving the hit rate and accuracy of subsequent design tasks.

[0046] Step S53: Multi-industry scenario verification. The verification scenarios may include construction site guardians, rebar tying robots, industrial robotic arm operation platforms, and warehouse logistics robots, etc. Different scenarios share the same design framework, but their object elements, task elements, environmental conditions, and risk distributions are different. Therefore, a unified method can be used to complete the differentiated structural configuration design.

[0047] Taking the construction site safety monitoring scenario as an example, the requirements include worker identification, hazardous area fencing, nighttime boundary crossing alarms, linked audio-visual prompts, and remote log auditing; environmental factors include dust, insufficient nighttime illumination, and obstruction by large machinery; and constraints include alarm real-time performance, deployment cost, and power supply conditions at the construction site. Through steps S1 to S4, candidate structural configurations including a visual perception unit, a laser ranging unit, an edge computing box, a safety relay, a wireless communication module, and a cloud-based operation and maintenance interface can be automatically generated, and the optimal solution will be output while meeting the constraints of alarm latency and risk coverage.

[0048] Taking a warehouse logistics robot scenario as an example, the requirements description includes cargo recognition, path planning, pick-and-place operations, and collaborative obstacle avoidance, while environmental factors focus on aisle congestion, shelf obstruction, and wireless channel contention. Using the same method, a structural configuration centered on multi-sensor fusion, map localization, task scheduling, and collaborative obstacle avoidance can be generated. Different computing platforms and communication topologies can then be selected based on the throughput requirements, reliability requirements, and operation and maintenance mode of the warehouse system.

[0049] For industrial robotic arm assembly and rebar binding scenarios, this invention can also unify and abstract capabilities such as force control assembly, trajectory planning, docking and positioning, safety protection, process auditing and remote upgrade through the meta-functional intermediate layer, and then use architectural element mapping and multi-objective optimization to output structural configuration solutions that meet the requirements of high precision, high reliability and high safety, thereby reducing the cost of repetitive design when migrating across industries.

[0050] The second technical solution is a generative agile design system for industrial embodied intelligent agent structure configuration, comprising a requirement semantic parsing and scenario modeling module, a meta-function extraction and recommendation module, an architectural element mapping and candidate configuration generation module, a multi-objective constraint solving and architectural blueprint output module, and a verification evaluation and knowledge base write-back module. The requirement semantic parsing and scenario modeling module is used to complete requirement text encoding, element classification, and scenario graph construction; the meta-function extraction and recommendation module is used to complete multi-label scoring of meta-functions, priority ranking, and generation of new meta-function combinations; the architectural element mapping and candidate configuration generation module is used to complete automatic component selection, candidate structure configuration generation, and interface relationship orchestration; the multi-objective constraint solving and architectural blueprint output module is used to complete optimal structure configuration solving and blueprint output; and the verification evaluation and knowledge base write-back module is used to perform digital twin verification, real-world scenario verification, and knowledge updates.

[0051] The third aspect of the technical solution is a storage medium storing processor-executable instructions, which, when executed by the processor, are used to implement the above-mentioned industrial embodied intelligent agent structure configuration generative agile design method.

[0052] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention. The scope of the invention is defined by the appended claims and their equivalents, all of which should be included within the scope of protection of the invention.

Claims

1. A generative agile design method for the structural configuration of an industrial embodied intelligent agent, characterized in that, Includes the following steps: Step S1: Obtain the natural language requirement description, process constraints and site rules of the industrial embodied intelligent agent, extract object elements, environmental elements, task elements, constraint elements and risk elements, form a scene element set and construct a scene diagram; Step S2: Calculate the scene representation vector based on the scene graph, perform multi-label scoring on the perception, localization, prediction, planning, control, safety and operation and maintenance functions in the meta-function library, and output the meta-function set and its priority. Step S3: Input the scene element set and meta-function set into the architecture element library and graph matching model, automatically select sensor, actuator, computing power, communication, security and operation and maintenance components, and generate at least one candidate structure configuration containing hierarchical modules, interface relationships, data flow and deployment relationships; Step S4: For the candidate structural configurations, perform multi-objective optimization and constraint solving based on security, real-time performance, cost, robustness and iteration cycle, and output the optimal structural configuration and its architecture blueprint. Step S5: Verify and evaluate the optimal structural configuration in a digital twin environment and a real industrial scenario. Write the verification results back to the meta-function library, architecture element library, and parameter model to update their parameters and form a design iteration closed loop.

2. The agile design method for the generative configuration of industrial embodied intelligent agents according to claim 1, characterized in that, The specific steps in step S1 include: Encode the requirement text using a pre-trained language model to obtain a global semantic vector of the requirement. ; The text fragments are classified to identify their element categories; based on at least one of the spatial, temporal, dependency, or risk propagation relationships between the elements, a scene graph is constructed with the elements as nodes and the relationships as edges. ; Among them, the node set O, E, T, C, and R represent object elements, environmental elements, task elements, constraint elements, and risk elements, respectively.

3. The agile design method for the generative configuration of industrial embodied intelligent agents according to claim 2, characterized in that, When classifying text fragments, the probability that each text fragment i belongs to feature category c is calculated using a classification model. The calculation formula is as follows: in, The semantic encoding of text fragment i. and For the classification model parameters corresponding to category c, u iterates through all preset element categories.

4. The agile design method for the generative configuration of industrial embodied intelligent agents according to claim 1, characterized in that, The step S2 of generating the target meta-function set includes: calculating the scene representation vector based on the scene graph. ; Calculate the k-th candidate metafunction using a multi-label prediction model The recommended score is: in, and For model parameters; at least the recommended score Candidate meta-functions exceeding a preset threshold are included in the target meta-function set; The generation of the target meta-function set also includes: based on the risk-related weights corresponding to the risk elements. The priority of computational meta-functions is as follows: And priority is used as the basis for sorting or filtering.

5. The agile design method for the generative configuration of industrial embodied intelligent agents according to claim 1, characterized in that, The generation of candidate structural configurations in step S3 includes: The scene graph is fused with the target meta-function set to obtain a scene-meta-function fusion vector. ; The architectural element library is constructed as a heterogeneous graph. , where nodes The attribute vector includes at least one of the following: performance metric, interface type, power consumption, cost, and reliability; using the fused vector With the node attribute vector Calculate the matching score and use the matching score as the component corresponding to each function selection.

6. The agile design method for the generative configuration of industrial embodied intelligent agents according to claim 5, characterized in that, The candidate structural configuration Represented as in, Represents a set of modules. Represents a collection of interfaces. Indicates data flow relationship, This represents a deployment matrix that describes the deployment locations of various functional modules on edge controllers, vehicle computing units, industrial gateways, cloud simulation platforms, or security controllers.

7. The agile design method for the generative configuration of industrial embodied intelligent agents according to claim 1, characterized in that, In step S4, a multi-objective optimization function J(A) is constructed, with functional satisfaction, security coverage, real-time performance score, and robustness score as gain terms, and cost and iteration cycle as penalty terms, while ensuring that the end-to-end latency does not exceed a threshold. Power consumption or computing power budget does not exceed the threshold. Safety coverage rate is not lower than the threshold. and the risk of downgrade is no greater than the threshold. The constraints.

8. The agile design method for the generative configuration of industrial embodied intelligent agents according to claim 1, characterized in that, The specific steps in step S5 include: The verification and evaluation adopts a combination of digital twins and real-world scenarios. The evaluation indicators include at least collision / intrusion response, security takeover success rate, end-to-end latency, critical link latency, performance degradation under dust / night / occlusion conditions, deployment cost, operation and maintenance efficiency, and iteration cycle. Based on the evaluation results, the priority of meta-functions, the posterior value of component reliability, and the structural configuration generation parameters are updated.

9. A design system comprising a generative agile design method for industrial embodied intelligent agent structural configuration as described in any one of claims 1 to 8, characterized in that, include: The requirement analysis and scenario modeling module is used to obtain the natural language requirement description, process constraints and site rules of industrial embodied intelligent agents, extract object elements, environmental elements, task elements, constraint elements and risk elements, form a scenario element set and construct a scenario graph; The metafunction extraction and recommendation module is used to calculate the scene representation vector based on the scene graph, perform multi-label scoring on the perception, positioning, prediction, planning, control, safety and operation and maintenance functions in the metafunction library, and output the metafunction set and its priority. The architecture mapping and configuration generation module is used to input the scene element set and meta-function set into the architecture element library and graph matching model, automatically select sensor, actuator, computing power, communication, security and operation and maintenance components, and generate at least one candidate structural configuration containing hierarchical modules, interface relationships, data flow and deployment relationships. The optimization solution and blueprint output module is used to perform multi-objective optimization and constraint solution based on security, real-time performance, cost, robustness and iteration cycle for the candidate structural configuration, and output the optimal structural configuration and its architecture blueprint. The verification, evaluation, and knowledge write-back module is used to verify and evaluate the optimal structural configuration in a digital twin environment and a real industrial scenario, and write the verification results back to the meta-function library, architecture element library, and parameter model to update their parameters.

10. A storage medium storing processor-executable instructions, characterized in that, The processor-executable instructions, when executed by the processor, are used to implement the method as described in any one of claims 1 to 8.

Images