Multi-specialty collaborative agent construction method based on large scene model

Abstract

The invention belongs to the technical field of collaboration, and particularly relates to a multi-specialty collaboration agent construction method based on a scene large model, which comprises the following specific steps: S1, the scene large model: firstly collecting structured and unstructured data from multiple specialty fields, then training the scene large model according to the collected data, and then building the scene large model; and comprehensively evaluating the trained model, and after evaluation, carrying out iterative optimization on the model. In data engineering, structured and unstructured data in multiple professional fields are collected, cleaned, labeled and preprocessed, data formats and standards are unified, and data are stored and managed in a centralized manner by adopting a data warehouse and data lake technology; therefore, data barriers among professionals are broken through, and data which originally use different software and data formats and are independent from one another are communicated with one another.

Description

Technical Field

[0001] The present invention relates to the field of collaborative technology, and in particular to a method for constructing a multi-professional collaborative intelligent agent based on a large scene model. Background Art

[0002] Amidst the wave of digital transformation, industries are increasingly demanding intelligent and collaborative work models. This is especially true in areas like construction engineering, intelligent manufacturing, and healthcare. Projects often involve collaborative work across multiple disciplines, requiring the integration of diverse expertise and data resources for efficient decision-making and problem-solving. However, traditional collaboration models suffer from information silos, fragmented knowledge, and low collaborative efficiency, making them difficult to meet the business needs of complex scenarios. Specific challenges are as follows:

[0003] 1. Information silos: In the construction industry, disciplines such as building structure design, water supply and drainage design, and electrical design are often independently developed during the design phase, using different software and data formats. For example, architects use BIM software to create building models, while structural engineers use specialized structural analysis software for mechanical calculations. Since these data are not interoperable, design conflicts are prone to occur during later integration, leading to rework and project delays.

[0004] 2. Knowledge Dispersion: In healthcare, disease diagnosis and treatment plan development require comprehensive consideration of multiple disciplines, including clinical medicine, pharmacy, and laboratory medicine. When formulating treatment plans, doctors must consult extensive literature and case studies. This knowledge, however, is dispersed across diverse databases and resources, making it difficult to access and integrate. This can easily lead to missed or inaccurate information, impacting treatment outcomes.

[0005] 3. Collaboration efficiency issues: In the field of intelligent manufacturing, product development involves multiple disciplines, including mechanical design, electronic circuit design, and process planning. Traditional collaboration methods rely primarily on manual communication and meeting coordination, resulting in untimely and inaccurate information transmission. This leads to frequent design changes, extended R&D cycles, delayed product launches, and reduced market competitiveness.

[0006] Based on the above, a method for constructing a multi-professional collaborative intelligent agent based on a large scene model is invented. Summary of the Invention

[0007] To solve the above technical problems, according to one aspect of the present invention, the present invention provides the following technical solutions:

[0008] A method for constructing a multi-disciplinary collaborative intelligent agent based on a large scenario model includes the following specific steps:

[0009] S1, Scenario Big Model: First, collect structured and unstructured data from multiple professional fields, then train the scenario big model based on the collected data. After that, conduct a comprehensive evaluation of the trained model, and after the evaluation, iteratively optimize the model.

[0010] S2, agent construction and orchestration: First, build the professional knowledge map, mechanism model, AI model, and business model in sequence. Then, flexibly orchestrate professional agents according to business needs. Then, build multiple professional agents. After construction, multiple professional agents are combined and orchestrated to form a multi-professional agent.

[0011] S3, Intelligent Model and Digital Twin: First build the intelligent model, and then combine the output results of multi-professional intelligent agents with the digital twin.

[0012] As a preferred solution of the method for constructing a multi-professional collaborative intelligent agent based on a scene large model described in the present invention, the specific steps of S1 are as follows:

[0013] S11, Data Engineering: First, collect structured and unstructured data from multiple professional fields. Then, clean, label, and preprocess the data to remove noisy and duplicate data, unify data formats and standards, and build high-quality data sets. At the same time, use data warehouse and data lake technologies to achieve centralized data storage and management, providing data support for subsequent model training.

[0014] S12, Model Training: Based on the preprocessed dataset, a deep learning algorithm is used to train the scenario-based large model. During the training process, transfer learning and reinforcement learning techniques are combined, using the existing general large model parameters as initial weights. By fine-tuning the model on specific scenario data, the model can better adapt to business needs. At the same time, a reinforcement learning mechanism is introduced to dynamically adjust the model parameters based on feedback from the model in actual applications to improve the model's performance and accuracy.

[0015] S13, Model Evaluation: First, establish a scientific and reasonable evaluation index system, then conduct a comprehensive evaluation of the trained model, using cross-validation and leave-one-out evaluation methods to ensure the reliability and effectiveness of the evaluation results. This will help identify problems and deficiencies in the model and provide a basis for model optimization.

[0016] S14, iterative correction: First, iteratively optimize the model based on the model evaluation results, then adjust the model structure and parameters, increase or decrease training data, improve the training algorithm, and continuously improve the model's performance and generalization capabilities. Through continuous iterative correction, the large scenario model can better adapt to changes in different scenarios and business needs.

[0017] As a preferred solution of the method for constructing a multi-professional collaborative intelligent agent based on a large scene model described in the present invention, the structured and unstructured data in S11 include design drawings, experimental data, case reports, and literature.

[0018] As a preferred solution of the method for constructing a multi-professional collaborative intelligent agent based on a large scene model described in the present invention, the evaluation index system in S13 includes accuracy, recall rate, F1 value, and mean square error.

[0019] As a preferred solution of the method for constructing a multi-professional collaborative intelligent agent based on a large scene model described in the present invention, the specific steps of S2 are as follows:

[0020] S21, Professional Knowledge and Model: Build professional knowledge maps, mechanism models, AI models, and business models;

[0021] S22, Agent Orchestration Tool: Develop an agent orchestration tool that provides a visual interface and allows users to flexibly orchestrate specialized agents based on business needs. Users can define interactions and data flows between agents by dragging and dropping, connecting, and enabling collaborative work among multiple specialized agents. The agent orchestration tool also supports simulation and verification of orchestration plans, identifying potential issues in advance and optimizing them.

[0022] S23, Professional Agents: Build multiple professional agents based on the needs of different professional fields. Each professional agent integrates corresponding professional knowledge, models, and algorithms, can independently complete tasks in its own professional field, and collaborate with other professional agents through agent orchestration tools.

[0023] S24, multi-professional agent: Multiple professional agents are combined and orchestrated through agent orchestration tools to form a multi-professional agent. The multi-professional agent can integrate the capabilities of various professional agents, achieve cross-professional collaboration, and solve business problems in complex scenarios.

[0024] As a preferred solution of the method for constructing a multi-professional collaborative intelligent agent based on a scene large model described in the present invention, the specific steps of S21 are as follows:

[0025] S211, Knowledge Engineering: Organize and model knowledge in various professional fields to construct a professional knowledge graph. Through knowledge extraction and knowledge fusion technologies, dispersed professional knowledge is integrated into the knowledge graph to achieve structured and semantic representation of knowledge. The knowledge graph can clearly display the relationships between professional knowledge and provide rich knowledge support for intelligent agents.

[0026] S212, Mechanism Model: Build a mechanism model based on mathematical equations and physical laws, based on the physical, chemical, and biological principles of each professional field. The mechanism model can accurately describe the inherent laws of the professional field and provide a scientific basis for the intelligent agent's decision-making;

[0027] S213, Professional AI Model: Based on the data and business needs of various professional fields, professional AI models are trained. Professional AI models can leverage the capabilities of large scene models and combine the characteristics of professional fields to achieve intelligent processing of professional tasks.

[0028] S214, Business Model: Establish a business model based on business processes and requirements to describe the logical relationships and data flows between business activities. The business model can combine professional knowledge and models with actual business to achieve automation and intelligence of business processes.

[0029] As a preferred solution of the method for constructing a multi-professional collaborative intelligent agent based on a large scene model described in the present invention, the specific steps of S3 are as follows:

[0030] S31, building intelligent models;

[0031] S32 combines the output results of multi-professional intelligent agents with digital twins. Digital twins are digital mappings of physical entities or business processes that can reflect the status and behavior of physical entities in real time. By applying the decision-making results of multi-professional intelligent agents to digital twins for simulation and verification, accurate predictions and optimization suggestions are provided for actual business.

[0032] Compared with existing technologies:

[0033] 1. Addressing the problem of information silos: The data engineering approach of this invention collects structured and unstructured data from multiple disciplines, cleans, annotates, and preprocesses it, unifying data formats and standards. Data warehouse and data lake technologies are then used to centrally store and manage the data. This breaks down data barriers between disciplines, enabling interoperability of previously independent data generated using different software and data formats. For example, in the field of construction engineering, data from disciplines such as architectural design and structural analysis can be effectively integrated after processing, avoiding design conflicts caused by data interoperability, reducing rework and project delays. This resolves the problem of information silos at the data level and lays the data foundation for collaborative work among multi-disciplinary intelligent entities.

[0034] 2. Addressing the problem of knowledge dispersion: In the knowledge engineering of this invention, by combing through various professional knowledge to construct a knowledge graph, knowledge extraction and fusion technologies are used to structure and semantically integrate professional knowledge scattered across different databases and materials. Taking the medical and health field as an example, multiple professional knowledge such as clinical medicine and pharmacy are integrated into the knowledge graph to clearly present the relationships between knowledge. When doctors formulate treatment plans, the intelligent agent can rely on the knowledge graph to quickly obtain and integrate multiple professional knowledge, avoiding omissions or inaccuracies in information, providing comprehensive knowledge support for accurate diagnosis and treatment, and effectively solving the problem of knowledge dispersion.

[0035] 3. Regarding the issue of collaborative efficiency: The intelligent agent orchestration tool of the present invention can provide a visual operation interface, and users can flexibly orchestrate professional intelligent agents according to business needs, define interactive relationships and data flows; in the field of intelligent manufacturing, product research and development involves multi-professional collaboration. Through the intelligent agent orchestration tool, professional intelligent agents such as mechanical design and electronic circuit design can collaborate efficiently without relying on traditional inefficient manual communication and meeting coordination; in addition, the orchestration tool supports solution simulation verification, discovers potential problems in advance and optimizes them, reduces the number of design changes, shortens the R&D cycle, enables products to be launched faster, significantly improves the efficiency of multi-professional collaboration, and enhances the company's market competitiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of the overall framework of the present invention. DETAILED DESCRIPTION

[0037] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0038] This invention provides a method for constructing a multi-professional collaborative intelligent agent based on a large scene model. Figure 1 , including the following specific steps:

[0039] S1, Scenario Big Model: First, collect structured and unstructured data from multiple professional fields, then train the scenario big model based on the collected data. After that, conduct a comprehensive evaluation of the trained model, and after the evaluation, iteratively optimize the model.

[0040] The specific steps of S1 are as follows:

[0041] S11, Data Engineering: First, structured and unstructured data from multiple professional fields are collected. Then, the data is cleaned, labeled, and preprocessed to remove noisy and duplicate data, unify data formats and standards, and build high-quality data sets. At the same time, data warehouse and data lake technologies are used to achieve centralized data storage and management to provide data support for subsequent model training. Structured and unstructured data include design drawings, experimental data, case reports, and literature.

[0042] S12, Model Training: Based on the preprocessed dataset, a deep learning algorithm is used to train the scenario-based large model. During the training process, transfer learning and reinforcement learning techniques are combined, using the existing general large model parameters as initial weights. By fine-tuning the model on specific scenario data, the model can better adapt to business needs. At the same time, a reinforcement learning mechanism is introduced to dynamically adjust the model parameters based on feedback from the model in actual applications to improve the model's performance and accuracy.

[0043] S13, Model Evaluation: First, establish a scientific and reasonable evaluation index system, then conduct a comprehensive evaluation of the trained model, using cross-validation and leave-one-out evaluation methods to ensure the reliability and effectiveness of the evaluation results. This will help identify problems and deficiencies in the model and provide a basis for model optimization. The evaluation index system includes accuracy, recall, F1 value, and mean squared error.

[0044] S14, Iterative Revision: First, iteratively optimize the model based on the model evaluation results. Then, adjust the model structure and parameters, increase or decrease training data, and improve the training algorithm to continuously improve the model's performance and generalization capabilities. Through continuous iterative revision, the scenario model can better adapt to changes in different scenarios and business needs.

[0045] S2, agent construction and orchestration: First, build the professional knowledge map, mechanism model, AI model, and business model in sequence. Then, flexibly orchestrate professional agents according to business needs. Then, build multiple professional agents. After construction, multiple professional agents are combined and orchestrated to form a multi-professional agent.

[0046] The specific steps of S2 are as follows:

[0047] S21, Professional Knowledge and Model: Build professional knowledge maps, mechanism models, AI models, and business models;

[0048] The specific steps of S21 are as follows:

[0049] S211, Knowledge Engineering: Organize and model knowledge in various professional fields to construct a professional knowledge graph. Through knowledge extraction and knowledge fusion technologies, dispersed professional knowledge is integrated into the knowledge graph to achieve structured and semantic representation of knowledge. The knowledge graph can clearly display the relationships between professional knowledge and provide rich knowledge support for intelligent agents.

[0050] S212, Mechanism Model: Build a mechanism model based on mathematical equations and physical laws, based on the physical, chemical, and biological principles of each professional field. The mechanism model can accurately describe the inherent laws of the professional field and provide a scientific basis for the intelligent agent's decision-making;

[0051] S213, Professional AI Model: Based on the data and business needs of various professional fields, professional AI models are trained. Professional AI models can leverage the capabilities of large scene models and combine the characteristics of professional fields to achieve intelligent processing of professional tasks.

[0052] S214, Business Model: Build a business model based on business processes and requirements to describe the logical relationships and data flows between business activities. The business model can combine professional knowledge and models with actual business operations to achieve automation and intelligentization of business processes.

[0053] S22, Agent Orchestration Tool: Develop an agent orchestration tool that provides a visual interface and allows users to flexibly orchestrate specialized agents based on business needs. Users can define interactions and data flows between agents by dragging and dropping, connecting, and enabling collaborative work among multiple specialized agents. The agent orchestration tool also supports simulation and verification of orchestration plans, identifying potential issues in advance and optimizing them.

[0054] S23, Professional Agents: Build multiple professional agents based on the needs of different professional fields. Each professional agent integrates corresponding professional knowledge, models, and algorithms, can independently complete tasks in its own professional field, and collaborate with other professional agents through agent orchestration tools.

[0055] S24, Multi-disciplinary Agent: Multiple specialized agents are combined and orchestrated using agent orchestration tools to form a multi-disciplinary agent. This agent integrates the capabilities of each specialized agent, enabling cross-disciplinary collaboration and solving business problems in complex scenarios.

[0056] S3, Intelligent Model and Digital Twin: First build the intelligent model, then combine the output of multi-disciplinary intelligent agents with the digital twin;

[0057] The specific steps of S3 are as follows:

[0058] S31, building intelligent models;

[0059] S32 combines the output results of multi-professional intelligent agents with digital twins. Digital twins are digital mappings of physical entities or business processes that can reflect the status and behavior of physical entities in real time. By applying the decision-making results of multi-professional intelligent agents to digital twins for simulation and verification, accurate predictions and optimization suggestions are provided for actual business.

[0060] Although the present invention has been described above with reference to embodiments, various modifications may be made thereto and equivalent components may be substituted without departing from the scope of the present invention. In particular, as long as there are no structural conflicts, the various features of the embodiments disclosed herein may be combined with each other in any manner, and the omission of an exhaustive description of such combinations in this specification is solely for the sake of space and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for constructing a multi-professional collaborative intelligent agent based on a large scene model, characterized in that: The specific steps are as follows: S1, Scenario Big Model: First, collect structured and unstructured data from multiple professional fields, then train the scenario big model based on the collected data. After that, conduct a comprehensive evaluation of the trained model, and after the evaluation, iteratively optimize the model. S2, agent construction and orchestration: First, build the professional knowledge map, mechanism model, AI model, and business model in sequence. Then, flexibly orchestrate professional agents according to business needs. Then, build multiple professional agents. After construction, multiple professional agents are combined and orchestrated to form a multi-professional agent. S3, Intelligent Model and Digital Twin: First build the intelligent model, and then combine the output results of multi-professional intelligent agents with the digital twin.

2. The method for constructing a multi-professional collaborative agent based on a large scene model according to claim 1, characterized in that: The specific steps of S1 are as follows: S11, Data Engineering: First, collect structured and unstructured data from multiple professional fields. Then, clean, label, and preprocess the data to remove noisy and duplicate data, unify data formats and standards, and build high-quality data sets. At the same time, use data warehouse and data lake technologies to achieve centralized data storage and management, providing data support for subsequent model training. S12, Model Training: Based on the preprocessed dataset, a deep learning algorithm is used to train the scenario-based large model. During the training process, transfer learning and reinforcement learning techniques are combined, using the existing general large model parameters as initial weights. By fine-tuning the model on specific scenario data, the model can better adapt to business needs. At the same time, a reinforcement learning mechanism is introduced to dynamically adjust the model parameters based on feedback from the model in actual applications to improve the model's performance and accuracy. S13, Model Evaluation: First, establish a scientific and reasonable evaluation index system, then conduct a comprehensive evaluation of the trained model, using cross-validation and leave-one-out evaluation methods to ensure the reliability and effectiveness of the evaluation results. This will help identify problems and deficiencies in the model and provide a basis for model optimization. S14, iterative correction: First, iteratively optimize the model based on the model evaluation results, then adjust the model structure and parameters, increase or decrease training data, improve the training algorithm, and continuously improve the model's performance and generalization capabilities. Through continuous iterative correction, the large scenario model can better adapt to changes in different scenarios and business needs.

3. The method for constructing a multi-professional collaborative agent based on a large scenario model according to claim 2 is characterized in that: The structured and unstructured data in S11 include design drawings, experimental data, case reports, and literature.

4. The method for constructing a multi-professional collaborative agent based on a large scene model according to claim 2 is characterized in that: The evaluation index system in S13 includes accuracy, recall, F1 value, and mean square error.

5. The method for constructing a multi-professional collaborative agent based on a large scene model according to claim 1 is characterized in that: The specific steps of S2 are as follows: S21, Professional Knowledge and Model: Build professional knowledge maps, mechanism models, AI models, and business models; S22, Agent Orchestration Tool: Develop an agent orchestration tool that provides a visual interface and allows users to flexibly orchestrate specialized agents based on business needs. Users can define interactions and data flows between agents by dragging and dropping, connecting, and enabling collaborative work among multiple specialized agents. The agent orchestration tool also supports simulation and verification of orchestration plans, identifying potential issues in advance and optimizing them. S23, Professional Agents: Build multiple professional agents based on the needs of different professional fields. Each professional agent integrates corresponding professional knowledge, models, and algorithms, can independently complete tasks in its own professional field, and collaborate with other professional agents through agent orchestration tools. S24, multi-professional agent: Multiple professional agents are combined and orchestrated through agent orchestration tools to form a multi-professional agent. The multi-professional agent can integrate the capabilities of various professional agents, achieve cross-professional collaboration, and solve business problems in complex scenarios.

6. The method for constructing a multi-professional collaborative agent based on a large scenario model according to claim 5 is characterized in that: The specific steps of S21 are as follows: S211, Knowledge Engineering: Organize and model knowledge in various professional fields to construct a professional knowledge graph. Through knowledge extraction and knowledge fusion technologies, dispersed professional knowledge is integrated into the knowledge graph to achieve structured and semantic representation of knowledge. The knowledge graph can clearly display the relationships between professional knowledge and provide rich knowledge support for intelligent agents. S212, Mechanism Model: Build a mechanism model based on mathematical equations and physical laws, based on the physical, chemical, and biological principles of each professional field. The mechanism model can accurately describe the inherent laws of the professional field and provide a scientific basis for the intelligent agent's decision-making; S213, Professional AI Model: Based on the data and business needs of various professional fields, professional AI models are trained. Professional AI models can leverage the capabilities of large scene models and combine the characteristics of professional fields to achieve intelligent processing of professional tasks. S214, Business Model: Establish a business model based on business processes and requirements to describe the logical relationships and data flows between business activities. The business model can combine professional knowledge and models with actual business to achieve automation and intelligence of business processes.

7. The method for constructing a multi-professional collaborative intelligent agent based on a large scene model according to claim 1 is characterized in that: The specific steps of S3 are as follows: S31, building intelligent models; S32 combines the output results of multi-professional intelligent agents with digital twins. Digital twins are digital mappings of physical entities or business processes that can reflect the status and behavior of physical entities in real time. By applying the decision-making results of multi-professional intelligent agents to digital twins for simulation and verification, accurate predictions and optimization suggestions are provided for actual business.

Images