A multi-agent interaction method based on a bridge intelligent maintenance large model

Abstract

The application discloses a kind of multi-agent interaction methods based on bridge wisdom maintenance big model, comprising: obtaining multidimensional bridge field dataset, training base big model to build initial bridge wisdom maintenance big model, fine-tuning is carried out to the initial bridge wisdom maintenance big model, and bridge wisdom maintenance big model is built;Bridge various standard document vector database is built;User demand is input into the bridge wisdom maintenance big model, and multi-agent interaction is carried out in combination with the bridge various standard document vector database.The present application realizes the intelligent management and collaborative work of bridge design, monitoring, maintenance, construction and the like whole life cycle, and improves the overall efficiency and quality control level of bridge engineering field.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of bridge engineering, and particularly relates to a multi-agent interaction method based on a bridge intelligent maintenance large model. BACKGROUND

[0002] With the vigorous development of bridge construction, the traditional bridge management mode relying on manual experience is at a loss when dealing with massive data, complex working conditions and multi-field collaboration needs. The existing intelligent monitoring and management technology lacks deep integration, the information of each link is isolated, and it is difficult to provide strong support for accurate decision-making and efficient execution, so innovative systematic solutions are needed to meet the intelligent needs of the whole life cycle of bridges. SUMMARY

[0003] To solve the above technical problems, the application provides a multi-agent interaction method based on a bridge intelligent maintenance large model, which realizes intelligent management and collaborative work of the whole life cycle of bridges such as bridge design, monitoring, maintenance and construction, and improves the overall efficiency and quality control level of the bridge engineering field.

[0004] To achieve the above purpose, the application provides a multi-agent interaction method based on a bridge intelligent maintenance large model, which comprises:

[0005] Obtain a multi-dimensional bridge field data set, train a base large model to construct an initial bridge intelligent maintenance large model, fine-tune the initial bridge intelligent maintenance large model, and construct a bridge intelligent maintenance large model;

[0006] Construct a bridge various standard document vector database;

[0007] Input user requirements into the bridge intelligent maintenance large model, and perform multi-agent interaction in combination with the bridge various standard document vector database.

[0008] Optionally, the multi-dimensional bridge field data set comprises standard specification data sets related to bridge detection, monitoring and maintenance; bridge detection, monitoring and maintenance report data sets; and various types of papers and books data sets of bridge detection, monitoring and maintenance research.

[0009] Optionally, fine-tuning the initial bridge intelligent maintenance large model comprises:

[0010] Based on the initial bridge intelligent maintenance large model and in combination with a bridge expert knowledge base, supervised fine-tuning is performed according to actual business scenarios and requirements of bridge maintenance, and the construction of the bridge intelligent maintenance large model is completed.

[0011] Optionally, constructing the bridge various standard document vector database comprises:

[0012] Collect standard documents related to each link of the bridge, screen the standard documents, and obtain screened standard documents;

[0013] Classify the screened standard documents according to different life cycle stages, professional fields and standard types of the bridge, and obtain classified standard documents;

[0014] Preprocess the classified standard documents to obtain preprocessed standard documents, perform text blocking on the preprocessed standard documents to obtain blocked standard documents;

[0015] Perform knowledge extraction on the blocked standard documents to obtain extracted key information, and perform semantic labeling on the extracted key information to obtain labeled key information;

[0016] Select a vector model according to data characteristics and application requirements and fine-tune it, train the fine-tuned vector model through the labeled key information, and obtain a trained vector model;

[0017] According to the trained vector model, the labeled key information is converted into corresponding knowledge vectors, a database structure for storing bridge standard knowledge vectors is designed, and the generated knowledge vectors are stored according to the designed database structure, completing the construction of the bridge various standard document vector database.

[0018] Optionally, input the user demand into the bridge intelligent maintenance large model, and combine the bridge various standard document vector database for multi-agent interaction, including:

[0019] According to the user's demand, the multi-agent is expanded to form a strongly connected graph structure of each agent, the weight matrix coefficient of the edge is set, the multi-agent interaction path is designed for information interaction;

[0020] The bridge various standard document vector database and the corresponding agent are individually deeply connected, and the bidirectional dynamic update of the bridge various standard document vector database and the corresponding agent is further performed;

[0021] The multi-agent sends the multi-element information of the bridge to the bridge intelligent maintenance large model, intelligently guides and cooperates with the bridge intelligent maintenance large model, realizes the user demand, and feeds back the realization result to the user;

[0022] According to the decision-making ability of the bridge intelligent maintenance large model and the database specification, personalized solutions are provided for different users, and the multi-agent is regularly reviewed to realize multi-agent interaction.

[0023] Optionally, the users include owners, design engineers, detection engineers, monitoring engineers, maintenance engineers, construction engineers and related researchers in the field of bridges.

[0024] Optionally, the multiple intelligent agents include a cost accounting intelligent agent, a design assistance intelligent agent, a detection data analysis intelligent agent, a monitoring data intelligent agent, a maintenance planning intelligent agent, a construction scheduling intelligent agent and a research exploration intelligent agent.

[0025] Optionally, the two-way dynamic updating of the bridge various standard document vector database and the corresponding intelligent agent includes that when the knowledge vector of the bridge various standard document vector database is updated, the corresponding intelligent agent receives and updates the knowledge system in time through a push notification.

[0026] The new discoveries and new experiences generated by the intelligent agent in daily work are transmitted back to the bridge various standard document vector database after strict auditing and standardization.

[0027] Optionally, a system includes a memory, a processor and a multi-agent interaction program based on a bridge intelligent maintenance large model stored on the memory and executable on the processor, and the multi-agent interaction program based on the bridge intelligent maintenance large model implements the steps of the multi-agent interaction method based on the bridge intelligent maintenance large model when executed by the processor.

[0028] Optionally, a computer readable storage medium stores a multi-agent interaction program based on a bridge intelligent maintenance large model, and the multi-agent interaction program based on the bridge intelligent maintenance large model implements the steps of the multi-agent interaction method based on the bridge intelligent maintenance large model when executed by a processor.

[0029] The present application has the following technical effects: The multi-agent interaction method based on the bridge intelligent maintenance large model innovatively integrates multiple intelligent agents, fine-tuned large models and vector databases, deeply penetrates every link of the whole life cycle management of bridges, and makes the decision-making process like having a wise brain, so that the scientificity, accuracy and timeliness are greatly improved. The carefully constructed vector database is like a knowledge lighthouse, which guarantees real-time sharing and dynamic updating of knowledge, provides a solid knowledge foundation for bridge design, maintenance, construction and other links, and greatly optimizes the accuracy and reliability of each link. The customized intelligent agent is like a thoughtful housekeeper, which fully meets the personalized needs of multiple users and significantly improves the user experience, and injects continuous power into the efficient management and continuous technical innovation of bridge engineering. BRIEF DESCRIPTION OF DRAWINGS

[0030] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of this application, and their

[0031] Figure 1 A process schematic diagram for the bridge intelligent maintenance fine-tuning large model construction and training of the embodiments of the application;

[0032] Figure 2 A process schematic diagram for the bridge various standard document vector database construction of the embodiments of the application;

[0033] Figure 3 A process schematic diagram for a multi-agent interaction method based on the bridge intelligent maintenance large model of the embodiments of the application;

[0034] Figure 4 A schematic diagram of the overall architecture execution process of the bridge intelligent maintenance large model training and application of the embodiments of the application. DETAILED DESCRIPTION

[0035] It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0036] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in a different order than that shown.

[0037] As Figure 3 shown, the embodiments provide a multi-agent interaction method based on a bridge intelligent maintenance large model, comprising:

[0038] Obtaining a multi-dimensional bridge field data set, training a base large model to construct an initial bridge intelligent maintenance large model, fine-tuning the initial bridge intelligent maintenance large model, and constructing a bridge intelligent maintenance large model;

[0039] Constructing a bridge various standard document vector database;

[0040] Inputting user requirements into the bridge intelligent maintenance large model, and combining the bridge various standard document vector database for multi-agent interaction.

[0041] Further, the multi-dimensional bridge field data set includes bridge detection, monitoring and maintenance related standard specification data set; bridge detection, monitoring and maintenance report data set; bridge detection, monitoring and maintenance various types of frontier research papers and books data set; bridge design papers and books data.

[0042] Further, fine-tuning the initial bridge intelligent maintenance large model comprises:

[0043] Based on the initial bridge intelligent maintenance large model, combined with the bridge expert knowledge base, supervised fine-tuning is performed according to the actual business scenarios and needs of bridge maintenance, and the construction of the bridge intelligent maintenance large model is completed.

[0044] Specifically, as shown in Figure 1 Bridge intelligent maintenance large model construction and training steps are as follows:

[0045] Base model selection: Introduce a Chinese pre-trained large model, select a suitable large model as the base model according to project requirements, computing resources, and data characteristics, etc.

[0046] Data collection and knowledge base construction: Collect multi-dimensional bridge field data sets, including bridge detection, monitoring and maintenance related standard specification data sets, bridge detection, monitoring and maintenance report data sets from actual engineering, and papers and books data sets of various frontier researches on bridge detection, monitoring and maintenance, as well as papers and books data of bridge design. Integrate these rich materials and carefully construct the bridge knowledge base to lay a solid data foundation for subsequent retraining of the base large model.

[0047] Base model training: Use the constructed bridge knowledge base to train the selected base model. Before training, a series of key hyperparameters need to be set reasonably, such as the number of training rounds, which determines the frequency of model traversing data and deeply affects the learning degree of model; batch size, which is related to the sample size participating in operation in each iteration, and plays an important role in training stability and memory occupation; learning rate, which directly controls the step size of model parameter update, and too large or too small will hinder the model from converging to the optimal solution. After fine tuning and sufficient training, the preliminary bridge intelligent maintenance large model is output.

[0048] Supervised fine-tuning (SFT) starts: Since the previous steps focus on unsupervised learning, there is still room for improvement in the prediction accuracy of the model. At this time, it is crucial to introduce supervised fine-tuning. This stage mainly covers two key links: first, organize bridge experts to collaboratively organize professional and accurate bridge expert knowledge base, and inject deep domain knowledge into the model; second, carefully set fine-tuning instructions according to the actual business scenarios and needs of bridge maintenance, to promote the further precise alignment of the output of the large model with the actual application needs.

[0049] Model output and saving: After completing the supervised fine-tuning, the bridge intelligent maintenance fine-tuning large model is successfully output and saved for subsequent precise application in various aspects of bridge intelligent maintenance, and effectively helps to improve the quality and efficiency of bridge maintenance work.

[0050] Further, the bridge various types of standard document vector database construction includes:

[0051] Collecting standard documents related to each link of the bridge, screening the standard documents, and obtaining screened standard documents;

[0052] Classifying the screened standard documents according to different life cycle stages, professional fields and standard types of the bridge, and obtaining classified standard documents;

[0053] Preprocessing the classified standard documents to obtain preprocessed standard documents, and performing text blocking on the preprocessed standard documents to obtain blocked standard documents;

[0054] Performing knowledge extraction on the blocked standard documents to obtain extracted key information, and performing semantic labeling on the extracted key information to obtain labeled key information;

[0055] Selecting a vector model according to data characteristics and application requirements and fine-tuning the vector model, training the fine-tuned vector model according to the labeled key information, and obtaining a trained vector model;

[0056] According to the trained vector model, the labeled key information is converted into corresponding knowledge vectors, a database structure for storing bridge standard knowledge vectors is designed, and the generated knowledge vectors are stored according to the designed database structure, completing the construction of the bridge various types of standard document vector database.

[0057] Specifically, as shown in Figure 2 Bridge various types of standard document vector database construction steps are explained as follows:

[0058] Standard document collection and arrangement includes: comprehensive search: using professional databases, industry association websites, government standard publishing platforms and other resources, widely collecting national standards, industry standards, local standards and international standard documents related to bridge design, construction, detection, maintenance and other aspects. Document screening: screening the collected mass of standard documents, removing outdated, repetitive or irrelevant content. According to the latest regulations and policies and industry trends, keep the current effective standard documents to ensure the timeliness and accuracy of the data. Classification and archiving: according to the different life cycle stages (design, construction, detection, maintenance) of the bridge, professional fields (structure, material, electromechanical, disaster prevention, etc.) and standard types (specification, procedure, guidebook, etc.), the screened standard documents are classified in detail, and a clear directory structure is established to facilitate subsequent processing.

[0059] Text preprocessing includes: format conversion: standard documents in different formats are uniformly converted into plain text format for subsequent text analysis. Clean up noise: remove special characters, garbled characters, extra spaces, headers and footers, and non-text information in the text to make the text content concise and clear, which is beneficial for analysis. Word segmentation processing: use word segmentation tools in natural language processing to cut continuous text into individual words or terms according to certain rules, which is convenient for subsequent vector model construction.

[0060] Knowledge extraction and labeling includes: key information extraction: extract core knowledge elements from preprocessed text, including but not limited to bridge technical parameters, material requirements, construction technology, detection indicators, and maintenance points. Develop detailed extraction rules and combine manual inspection to ensure the accuracy of the extracted information. Semantic annotation: annotate the extracted key information semantically to clarify its belonging to the knowledge domain, its position in the bridge life cycle, and its relationship with other information.

[0061] Vector model construction includes: selecting a vector model: according to the characteristics of the data and the application requirements, select a suitable vector model for fine-tuning. Model training: use the labeled key information as training data to input the selected vector model for training. Model evaluation: use the test set (a part of the labeled key information) to evaluate the trained vector model.

[0062] Knowledge vector database creation and storage includes: vector generation: use the trained vector model to convert all extracted and labeled bridge standard key information into corresponding knowledge vectors. Each vector represents a specific bridge knowledge element, and its dimension and value reflect the position and characteristics of the element in the semantic space learned by the model. Database design: design a database structure specifically for storing bridge standard knowledge vectors, including table name, field name, data type, etc. Considering the subsequent query, retrieval and update requirements, reasonably plan the index and relationship mode of the database. Data storage: store the generated knowledge vectors according to the designed database structure to ensure data integrity and security.

[0063] Database maintenance and update includes: regular monitoring: establish a regular monitoring mechanism to monitor the update and release of industry standards, the development of bridge technology and feedback information in actual application, and timely discover the opportunity to update the database. Data update: when new standard documents or knowledge content that needs to be corrected are found, follow the above steps to operate a round of update, and store the updated knowledge vectors in the database to ensure that the database always maintains the latest and most accurate state.

[0064] Further, the user includes the owner, design engineer, detection engineer, monitoring engineer, maintenance engineer, construction engineer and related researchers in the field of bridge.

[0065] Further, the multi-agent includes a cost accounting agent, a design assistance agent, a detection data analysis agent, a monitoring data agent, a maintenance planning agent, a construction scheduling agent, and a research exploration agent.

[0066] Further, the user demand is input into the bridge intelligent maintenance large model, and multi-agent interaction is performed in combination with the bridge various standard document vector database.

[0067] According to the user's demand, the multi-agent is expanded, a strongly connected graph structure of each agent is formed, a weight matrix coefficient of an edge is set, a multi-agent interaction path is designed, and information interaction is performed;

[0068] The bridge various standard document vector database is individually and deeply connected with the corresponding agent, and bidirectional dynamic updating of the bridge various standard document vector database and the corresponding agent is further performed.

[0069] The multi-agent sends the multi-element information of the bridge to the bridge intelligent maintenance large model, intelligently guides and cooperates with the bridge intelligent maintenance large model, realizes the user demand, and feeds back the realization result to the user.

[0070] According to the decision-making power of the bridge intelligent maintenance large model and the database specification, individualized solutions are provided for different users, the multi-agent is regularly reviewed, and multi-agent interaction is realized.

[0071] Specifically, as shown in Figures 3-4 The multi-agent interaction method based on the bridge intelligent maintenance large model is explained as follows:

[0072] (1) Agent system construction:

[0073] Precise adaptation of user demand to agent expansion:

[0074] For the owner: the cost accounting agent assists the owner in overall planning of the bridge project whole process, including progress control, fund allocation, contract management, etc., provides overall project planning suggestions according to the key requirements of the owner to the project budget, construction period, etc., and the feedback of other agents.

[0075] For design engineers: the design assistance agent not only provides innovative design scheme suggestions according to the bridge design specification, user demand and actual environmental conditions, but also can interact with design engineers in real time, understand their design ideas, provide similar successful case references and structure optimization direction suggestions based on the bridge standard knowledge vector database, and assist in design innovation and compliance guarantee.

[0076] For detection engineers: Detection data analysis agents assist detection engineers in processing massive field detection data, quickly filter out abnormal data using detection index standards in the database, and preliminarily judge disease types and potential risk points to provide direction for subsequent accurate detection.

[0077] For monitoring engineers: Monitoring data agents focus on integrating real-time monitoring data from various sensors, generating monitoring reports based on monitoring frequency standards and early warning thresholds in the database, and presenting bridge structure dynamic changes in a visual manner. When abnormal fluctuations occur, timely warnings are sent to relevant personnel.

[0078] For maintenance engineers: Maintenance planning agents combine maintenance processes, cycle standards, and real-time bridge state data to develop detailed maintenance plans for maintenance engineers, including daily maintenance task arrangements, maintenance material preparation lists, and maintenance personnel scheduling schemes, ensuring that bridge maintenance work is orderly and efficient.

[0079] For construction engineers: Construction scheduling agents are responsible for allocating resources such as manpower and materials based on construction processes and progress requirements, optimizing construction processes based on database knowledge, avoiding construction conflicts, and improving construction efficiency.

[0080] For researchers: Research exploration agents collect, organize, and analyze the latest scientific research achievements in the bridge field, and based on the research direction of researchers, they can mine relevant knowledge from the database to provide data support for experimental design and theoretical verification.

[0081] Design of multi-agent interaction path: Initialize the above 7 agent nodes to form a strongly connected graph structure, appropriately set the weight matrix coefficients of the edges, and use the shortest path algorithm to achieve information exchange during the interaction process to achieve efficient information exchange.

[0082] (2) Vector database access and data sharing

[0083] Personalized deep connection and application:

[0084] Owner: Cost accounting agents read different stage cost standards, project cycle reference cases, and other data from the database to assist owners in developing reasonable project goals and budgets. At the same time, based on data feedback during project execution, dynamically update project expected benefit evaluation.

[0085] Design engineer: Design assistance agents deeply mine knowledge such as structural mechanics principles and material performance parameters, combine user needs and environmental conditions, consider factors such as construction convenience and maintenance costs in subsequent stages during the design phase, and provide innovative and feasible design solutions. They can also transmit experience data such as mechanical performance test results of new structures back to the database.

[0086] Detection engineer: The detection data analysis agent processes and analyzes the collected data according to various detection indicators and detection method standards in the database, accurately identifies disease characteristics, and uploads newly discovered disease phenomena and characteristic data to enrich the disease sample library in the database.

[0087] Monitoring engineer: The monitoring data agent strictly follows the monitoring frequency and data transmission protocol standards in the database to collect, transmit, and store bridge monitoring data in real time. It uses early warning thresholds and other knowledge to promptly identify abnormal bridge conditions and provide feedback, and updates long-term monitoring data such as seasonal bridge deformation patterns to the database.

[0088] Maintenance engineer: The maintenance planning agent uses maintenance technology, cycle standards, and material performance database knowledge to develop detailed maintenance plans based on real-time bridge conditions. It also updates optimization schemes from maintenance practices, such as maintenance cycle adjustments in special environments, to the database.

[0089] Construction engineer: The construction scheduling agent arranges human and material resources at the construction site based on the construction technology sequence and resource allocation standards in the database. The construction quality monitoring agent supervises quality in real time against construction standards. Both agents update innovative construction technologies and quality problem solutions to the database for knowledge iteration.

[0090] Researcher: The research and exploration agent uses the database as a foundation to explore breakthroughs in cutting-edge technologies. It converts new theories, technologies, and methods from scientific research achievements, such as the application of new materials in bridges, into updateable data and updates them to the database.

[0091] Two-way dynamic update: When the database has new knowledge vector updates, such as revised industry standards or new scientific research achievements, each agent receives and updates its knowledge system through push notifications. New discoveries and experiences generated by agents in their daily work are rigorously reviewed and standardized before being transmitted back to the database, ensuring that the database always reflects the latest knowledge and practical experience in the bridge field.

[0092] (3) Collaborative interaction based on fine-tuning large models

[0093] Multi-source information aggregation driven by diverse user needs:

[0094] The monitoring data agent transmits the real-time monitoring data of the bridge to the bridge intelligent maintenance big model; the design assistant agent pushes the design draft, creative idea and designer feedback; the cost accounting agent reports the cost estimate details; the research and exploration agent shares the frontier technology trends. At the same time, the user-oriented agent also transmits the user's specific demand information, such as the owner's schedule compression requirement, the design engineer's structure optimization goal, the detection engineer's disease suspect, the monitoring engineer's early warning analysis requirement, the maintenance engineer's maintenance problem, the construction engineer's construction obstruction, and the research personnel's scientific research confusion, to the big model, requests integration and analysis, and excavates potential problems and optimization direction.

[0095] Intelligent guidance and cooperation:

[0096] After receiving multiple information, the bridge intelligent maintenance fine-tuning big model first starts the intelligent diagnosis process. It instructs the monitoring data agent to combine the structural parameters, material characteristics, disease cases and other knowledge in the bridge standard knowledge vector database to deeply analyze the current health status of the bridge.

[0097] According to the diagnosis result, the bridge intelligent maintenance big model accurately guides the decision-making agent to act. For the maintenance problem proposed by the maintenance engineer, such as frequent rust problem in a specific part of the bridge, the bridge intelligent maintenance big model prompts the decision-making agent to retrieve relevant maintenance technology, material selection standard from the database, and develop detailed maintenance strategies including targeted corrosion prevention measures, optimized maintenance period, and required maintenance material list.

[0098] If the owner expects to shorten the construction period, the bridge intelligent maintenance big model coordinates the construction scheduling agent to re-examine the construction process, combines the efficient construction technology cases in the database, breaks through the limitations of the conventional construction sequence, and proposes a reasonable work implementation plan. At the same time, it instructs the design assistant agent to quickly evaluate the impact of design changes on structural safety, ensures that the design is adjusted to meet the safety requirements under the condition of adjusting the design to adapt to the tight construction period, and requires the maintenance plan to be adjusted accordingly to ensure the continuous health of the bridge during the rush period.

[0099] In terms of cost control, as soon as the strategy changes, the bridge intelligent maintenance big model immediately coordinates the cost accounting agent to re-evaluate. The cost accounting agent finely calculates the cost changes according to the material price fluctuation data, labor cost standard and resources required by the new construction or maintenance scheme, and feeds back the cost increment or saving space to the bridge intelligent maintenance big model, such as the use of new anticorrosive materials which increases the initial investment but reduces the long-term maintenance cost, helping the decision-making layer to make the best economic choice.

[0100] In the face of technical problems such as difficult diseases detected by engineers, the bridge intelligent maintenance large model allocates research and exploration of agent joint detection data analysis agent to dig into the root cause. Research and exploration of agents retrieve cutting-edge research from databases, similar disease cracking cases, combined with accurate data provided by detection data analysis agents, use advanced analysis techniques such as disease feature clustering analysis based on big data, numerical models simulating disease development, and try to find innovative solutions.

[0101] At the same time, the design assistance agent optimizes subsequent design based on research results to prevent similar problems from occurring again, and improves the quality of the bridge from the root.

[0102] Execution and feedback loop:

[0103] Execution: The execution user implements the strategy, and real-time feedback of repair, construction progress, and obstacles encountered. Interact with the system, and the agent optimizes the work accordingly, such as the design assistance agent fine-tuning the design based on construction feedback, the cost accounting agent updating the cost based on the progress, and the bridge intelligent maintenance large model continuously optimizing subsequent instructions based on feedback. Feedback: Each user-oriented agent will also feedback the execution results and optimization suggestions to the corresponding user in an intuitive and easy-to-understand way, such as generating a project progress report for the owner, providing a design optimization report for the design engineer, compiling a disease diagnosis report for the detection engineer, outputting a monitoring analysis report for the monitoring engineer, providing a maintenance summary report for the maintenance engineer, compiling a construction review report for the construction engineer, and presenting a research progress report for the research personnel, forming a complete user interaction loop.

[0104] (4) Interaction system optimization

[0105] Customized solution optimization: With the decision-making power of the bridge intelligent maintenance large model and the database specification, weigh the pros and cons based on urgency, importance, feasibility, etc. to provide customized solutions for different users.

[0106] Continuous system iteration optimization: Use reinforcement learning methods to regularly review the multi-agent interaction system, adjust the agent division, communication protocol, and collaboration mode with the large model based on feedback such as project completion quality, cost control effectiveness, technical innovation results, and user satisfaction, and comprehensively improve the level of bridge life cycle management, and continuously adapt to the dynamic needs of various users in the bridge field.

[0107] The embodiment also provides a system, which comprises a memory, a processor, and a multi-agent interaction program based on a bridge intelligent maintenance large model stored on the memory and executable on the processor. When the multi-agent interaction program based on the bridge intelligent maintenance large model is executed by the processor, the steps of the multi-agent interaction method based on the bridge intelligent maintenance large model are implemented.

[0108] The embodiment also provides a computer readable storage medium storing a multi-agent interaction program based on the bridge intelligent maintenance large model, and the multi-agent interaction program based on the bridge intelligent maintenance large model, when executed by a processor, implements the steps of the multi-agent interaction method based on the bridge intelligent maintenance large model.

[0109] The application discloses a multi-agent interaction method based on a bridge intelligent maintenance large model, which innovatively fuses a multi-agent, a fine-tuned large model and a vector database, deeply penetrates each link of the whole life cycle management of a bridge, and makes the decision process like having a wise brain, so that the scientificity, accuracy and timeliness are greatly improved. The carefully constructed vector database is like a knowledge lighthouse, guarantees real-time sharing and dynamic updating of knowledge, provides a solid knowledge foundation for the operation of each link of bridge design, maintenance and construction, and greatly optimizes the accuracy and reliability of each link. The customized agent is like a thoughtful housekeeper, which fully meets the personalized needs of multi-element users, significantly improves the user experience, and injects continuous power into the efficient management and continuous technical innovation of bridge engineering.

[0110] The above is only a preferred specific embodiment of the application, but the protection scope of the application is not limited to this, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the application, which should be covered in the protection scope of the application. Therefore, the protection scope of the application should be subject to the protection scope of the claims.

Claims

1. A multi-agent interaction method based on a bridge intelligent maintenance large model, characterized in that, The method comprises the following steps: acquiring a multi-dimensional bridge field data set, training a base large model to construct an initial bridge intelligent maintenance large model, fine-tuning the initial bridge intelligent maintenance large model to construct a bridge intelligent maintenance large model; constructing a bridge various standard document vector database; inputting user requirements into the bridge intelligent maintenance large model and combining the bridge various standard document vector database for multi-agent interaction; the multi-dimensional bridge field data set comprises bridge detection, monitoring and maintenance related standard specification data set, bridge detection, monitoring and maintenance report data set, bridge detection, monitoring and maintenance various types of frontier research papers and books data set, and bridge design papers and books data set; the fine-tuning of the initial bridge intelligent maintenance large model comprises: based on the initial bridge intelligent maintenance large model, combining a bridge expert knowledge base, and performing supervised fine-tuning according to bridge maintenance actual business scenarios and requirements to complete the construction of the bridge intelligent maintenance large model; constructing the bridge various standard document vector database comprises: collecting standard documents related to each link of the bridge, screening the standard documents to obtain screened standard documents; classifying the screened standard documents according to different life cycle stages, professional fields and standard types of the bridge to obtain classified standard documents; preprocessing the classified standard documents to obtain preprocessed standard documents, and performing text blocking on the preprocessed standard documents to obtain blocked standard documents; performing knowledge extraction on the blocked standard documents to obtain extracted key information, and performing semantic annotation on the extracted key information to obtain annotated key information; selecting a vector model according to data characteristics and application requirements, training the annotated key information on the fine-tuned vector model to obtain a trained vector model; according to the trained vector model, converting the annotated key information into corresponding knowledge vectors, designing a database structure for storing bridge standard knowledge vectors, storing the generated knowledge vectors according to the designed database structure, and completing the construction of the bridge various standard document vector database; inputting user requirements into the bridge intelligent maintenance large model and combining the bridge various standard document vector database for multi-agent interaction comprises: according to the user's requirements, expanding the multi-agent to form a strongly connected graph structure of each agent, setting the weight matrix coefficient of the edge, designing a multi-agent interaction path for information interaction; the bridge various standard document vector database and the corresponding agent are individually deeply connected, and the bidirectional dynamic update of the bridge various standard document vector database and the corresponding agent is further performed; the multi-agent sends the multi-element information of the bridge to the bridge intelligent maintenance large model, intelligently guides and cooperates with the bridge intelligent maintenance large model, realizes the user's requirements, and feeds back the implementation results to the user; according to the decision-making ability of the bridge intelligent maintenance large model and the database specification, individualized solutions are provided for different users, the multi-agent is periodically reviewed, and the multi-agent interaction is realized. The users include owners, design engineers, detection engineers, monitoring engineers, maintenance engineers, construction engineers and related researchers in the field of bridges; The multi-agent includes a cost accounting agent, a design assistance agent, a detection data analysis agent, a monitoring data agent, a maintenance planning agent, a construction scheduling agent and a research exploration agent; The two-way dynamic update of the bridge various types of standard document vector database and the corresponding agent includes that when the knowledge vector of the bridge various types of standard document vector database is updated, the corresponding agent receives and updates the knowledge system in time through a push notification; The new discoveries and new experiences generated by the agent in daily work are returned to the bridge various types of standard document vector database after strict auditing and standardized processing.

2. A system, characterized by The system includes a memory, a processor and a multi-agent interaction program based on a bridge intelligent maintenance large model stored on the memory and executable on the processor, and the multi-agent interaction program based on the bridge intelligent maintenance large model implements the steps of the multi-agent interaction method based on the bridge intelligent maintenance large model of claim 1 when executed by the processor.

3. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a multi-agent interaction program based on a bridge intelligent maintenance large model, and the multi-agent interaction program based on the bridge intelligent maintenance large model implements the steps of the multi-agent interaction method based on the bridge intelligent maintenance large model of claim 1 when executed by the processor.

Images