A method and device for automatically analyzing scenic area operation data based on multi-agent

Abstract

This invention discloses a multi-agent-based method and device for automatically analyzing scenic area operational data. The method includes the following steps: Step 1: Enhance task descriptions; Step 2: Generate control flows; and Step 3: Generate data flows. This invention uses multi-agent technology to automatically select APIs and generate workflows, enhancing task representation and avoiding weak semantic propagation. It also introduces a review mechanism to ensure correct output, saving costs and enabling rapid response to demand.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to a method and device for automatically analyzing scenic area operation data based on multiple intelligent agents. Background Art

[0002] In recent years, my country has accelerated the pace of construction of various types of parks and scenic spots, and related supporting facilities have been gradually improved. As a result, the demand for operation and management has also increased rapidly. For example, the relationship between the number of tourists and the revenue of a certain amusement project, the relationship between the number of tourists and a certain violation warning, etc.

[0003] According to the traditional garden information project management and development method, whenever new business needs arise, a new information system must be developed accordingly or a new functional module must be developed based on the original system. This method has the characteristics of long development cycle and slow demand response, which makes it difficult to meet the rapidly growing and dynamic new management needs of garden scenic area managers. Therefore, a method and device for automatically realizing scenic area operation data analysis based on multi-agent is proposed. Summary of the Invention

[0004] The present invention provides the following technical solutions:

[0005] A method for automatically analyzing scenic spot operation data based on multi-agents includes the following steps:

[0006] Step 1: Task description enhancement. In this stage, a garden smart management system is established. The garden smart management system includes a user terminal. The user terminal sets up a task placement port for users to upload tasks. Based on the task description uploaded by the user, the garden smart management system searches for relevant APIs through keyword matching, classification search, or semantic analysis. The retrieved API usage documents or instructions are used to improve and enhance the task requirement description.

[0007] Step 2: Control flow generation. In this stage, according to the enhanced task description, APIs are selected from the garden smart management system and new control nodes are gradually generated until a complete control flow is generated.

[0008] Step 3: Data flow generation: Use the intelligent agent based on the large model to perform iterative reasoning and establish a mapping relationship between the output of the upstream node and the input of the downstream node in the control flow, thereby establishing a complete data flow;

[0009] Utilizing multi-agent technology based on large models, according to the user's business needs description (covering comprehensive analysis, statistics, and other needs of various types of park operation and management data), the system automatically selects and assembles appropriate APIs from the garden intelligent management system and generates workflows that meet user needs. This can effectively reduce the establishment and implementation of new information technology projects in garden scenic areas, saving development costs, and quickly responding to user needs. During the implementation process, the semantic environment related to user needs is improved through task expression enhancement, avoiding the propagation of users' weak semantic expressions into control flow and data flow generation, so that the generated workflow can better meet user needs. During the control flow and data flow generation stage, an audit mechanism is introduced to verify and correct the interactive feedback of a pair of agents to generate correct output. The method and device for automatically implementing scenic area operation data analysis based on multi-agents are tested in combination with actual projects to verify its usability. The details are as follows:

[0010] The operating environment for the "Method and Device for Automatically Implementing Scenic Area Operation Data Analysis Based on Multi-Agents" is: Intel i9-10900X processor, NVIDIA GeForce RTX 3090 GPU, 128G memory, Python 3.6 and Pytorch 1.4.0.

[0011] Software Environment: This device is currently used in conjunction with our company's "Garden Smart Management System." The interface standards will be open to the public in the future to enable integration with other application systems. The "Garden Smart Management System" embeds KNIME, an open source workflow management system. Llama3-70B is used as the basic model to implement the relevant functions of multiple intelligent agents.

[0012] Device usage: Users can input business requirements in text form, such as: counting the relationship between the number of tourists and the revenue of a certain amusement project, counting the relevant information of a certain violation warning, etc. The device will automatically generate a workflow and execute the workflow to obtain relevant results.

[0013] Preferably, step one further comprises: TMAgent and TCAgent, wherein TMAgent checks the task description provided by the user, retrieves relevant APIs using a predefined API library or a real-time search function, reads the API usage instructions, comprehensively analyzes the correlation between the task and the API, summarizes the problems and improves the task description, TCAgent checks each improved task description and optimizes it accordingly, and TMAgent and TCAgent gradually extract information related to the user's task description from the API usage instructions through multiple rounds of discussions, eliminates semantic problems in the task description provided by the user, and provides a more detailed and context-related understanding of the task requirements;

[0014] Compared with previous methods, the iterative correction of the two intelligent agents improves the detailed semantic description of the task, avoids the propagation of weak semantic expressions to downstream tasks, and the subsequent generation process of control flow and data flow, and improves the ease of use of the generated workflow.

[0015] Preferably, the step 2 also includes: setting up two intelligent agents, APISelAgent and APIRevAgent, and when constructing the control flow, APISelAgent selects an API suitable for the current step operation from the candidate API set as a new node of the control flow based on the first step operation in the enhanced task description and the dependency relationship between the APIs recorded in the API usage instructions; then APIRevAgent reviews whether the selected API is suitable and provides feedback; APISelAgent decides whether to reselect the API based on the feedback; if it decides to reselect, the API selection is performed again and submitted to APIRevAgent for review; if it decides to keep the original selection or APIRevAgent confirms that the selected API is suitable, then once the API is confirmed to be selected, it enters the next process, updates the semantic environment (composed of task description, existing nodes in the control flow, dependencies between nodes, APIs bound to each node, API usage instructions, etc.), and repeats the operation until the entire process is constructed.

[0016] Preferably, the step three further comprises:

[0017] A1. Two large language model-based agents, the parameter selection agent (ParaSelAgent) and the parameter review agent (ParaRevAgent), work together to establish a correct mapping between upstream node outputs and downstream node inputs, ensuring the accuracy of parameter transfer between APIs. ParaSelAgent first identifies a list of candidate parameters based on the current semantic environment (i.e., the nodes in the control flow up to the current step, the dependencies between nodes, the APIs used by each node, and API usage instructions). It then maps each parameter in the candidate parameter list (the output of the upstream node) to the input of the downstream node.

[0018] Preferably, the step three further comprises:

[0019] A2. ParaSelAgent takes a parameter from the candidate parameter list and matches it with the input parameter of the next node API.

[0020] Preferably, the step three further comprises:

[0021] A3. ParaRevAgent reviews the matching results and provides feedback. ParaSelAgent then optimizes the matching based on the feedback. Regardless of whether the matching result is confirmed or rejected, the current semantic environment is updated before the next parameter matching begins. When the matching result is confirmed, the mapping relationship of the matching parameter is added to the current semantic environment. When the matching result is rejected, the current semantic environment records that the parameter has not been matched until all candidate parameters are processed.

[0022] Preferably, the step three further comprises:

[0023] A4. Enter the next process node and repeat the operations of A1-A3 until the process ends. By adding data flow to the control flow, a complete and executable workflow is formed.

[0024] A device for automatically analyzing scenic area operation data based on multiple agents includes: a task description enhancement module, a control flow generation module and a data flow generation module. The output end of the task description enhancement module is connected to the input end of the control flow generation module via data, and the output end of the control flow generation module is connected to the input end of the data flow generation module via data.

[0025] Preferably, the task description enhancement module is used to receive the task description provided by the user, and retrieve the relevant API in the garden smart management system, and improve the task requirement description according to the API usage instructions. The task description enhancement module includes TMAgent and TCAgent, wherein TMAgent is responsible for checking the task description, retrieving the API, reading the API usage instructions and comprehensively analyzing the correlation between the task and the API, while TCAgent is responsible for checking and optimizing the improved task description, gradually extracting relevant information and eliminating semantic problems through multiple rounds of discussion, and the control flow generation module selects the API from the garden smart management system according to the enhanced task description and gradually generates new control nodes until a complete control flow is generated. The control flow generation module includes APISelAgent and APIRevAgent, APISelAgent selects a suitable API as a new node of the control flow according to the dependency between the task description and the API, and APIRevAgent is responsible for reviewing the selected API and providing feedback to ensure the correct construction of the control flow.

[0026] Preferably, the data flow generation module uses an intelligent agent based on a large model to perform iterative reasoning, establishes a mapping relationship between the output of the upstream node and the input of the downstream node in the control flow, and thus establishes a complete data flow. The data flow generation module includes ParaSelAgent and ParaRevAgent, which are used to identify candidate parameter lists, establish parameter mappings, perform matching audits and optimizations, and ensure the accuracy of parameter transfer between APIs.

[0027] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0028] 1: This invention uses multi-agent technology based on a large model to automatically select and assemble appropriate APIs from the garden intelligent management system based on the user's business needs description, and generates a workflow that meets the user's needs. This can effectively reduce the project establishment and implementation work of new information technology projects in garden scenic areas, save development costs, and quickly respond to user needs.

[0029] 2: This invention improves the semantic environment related to user needs by enhancing task expressions, avoiding the propagation of users' weak semantic expressions into control flow and data flow generation, so that the generated workflow can better meet user needs;

[0030] 3: This invention introduces a review mechanism in the control flow and data flow generation stages, and uses interactive feedback from a pair of intelligent agents to perform verification and correction to generate correct output. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of a multi-agent-based workflow generation device of the present invention.

[0032] Figure 2 This is an example diagram of the enhanced task description of the present invention.

[0033] Figure 3 This is an example diagram of control flow generation according to the present invention. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0035] See also Figure 1 、 Figure 2 and Figure 3 The present invention provides a method for automatically analyzing scenic spot operation data based on multi-agent, comprising the following steps:

[0036] Step 1: Task description enhancement. In this stage, a garden smart management system is established. The garden smart management system includes a user terminal. The user terminal sets up a task placement port for users to upload tasks. Based on the task description uploaded by the user, the garden smart management system searches for relevant APIs through keyword matching, classification search, or semantic analysis. The retrieved API usage documents or instructions are used to improve and enhance the task requirement description.

[0037] The intelligent garden management system is a comprehensive management system that integrates modern science and technology. It applies advanced technologies such as big data, cloud computing, the Internet of Things, the Internet, and information intelligence to urban garden management, including the following aspects:

[0038] The garden intelligent management system mainly consists of multiple functional modules, including but not limited to:

[0039] Automatically manage daily maintenance tasks such as watering, fertilizing, and pruning of garden plants to reduce labor costs and improve work efficiency.

[0040] Unified management and maintenance of facilities, equipment and other assets within the gardens to ensure the integrity and effective use of the assets.

[0041] By integrating sensors and other hardware equipment, the environmental parameters in the garden, such as temperature, humidity, light, soil moisture, PM2.5 concentration, noise level, etc., are monitored in real time, providing a scientific basis for garden management.

[0042] Monitor and manage energy-consuming equipment in the garden to achieve energy conservation and emission reduction.

[0043] Through electronic processing procedures, the comprehensive electronic and efficient management of garden patrol and inspection work can be achieved.

[0044] Features include:

[0045] By installing sensors and cameras, the environmental parameters and plant growth conditions in the garden can be monitored in real time.

[0046] Once an abnormality is detected, the system will immediately issue an early warning and notify relevant personnel via mobile application or text message.

[0047] Automatically adjust the frequency and amount of irrigation and fertilization based on information such as soil moisture, plant type and growth stage.

[0048] Achieve precise irrigation and fertilization, saving water resources and ensuring the healthy growth of plants.

[0049] Positioning, displaying and tracking the trajectories of garden management personnel, vehicles and mobile equipment.

[0050] Once the movement range is exceeded, the system will automatically alarm and push the information to the relevant operating personnel.

[0051] Based on the AIOT platform, it supports the access of massive devices and enables all smart facilities in the garden to be managed in a unified network.

[0052] Get real-time insights into the operational status and early warning information of smart facilities to ensure they are always in service.

[0053] Step 2: Control flow generation. In this stage, according to the enhanced task description, APIs are selected from the garden smart management system and new control nodes are gradually generated until a complete control flow is generated.

[0054] Step 3: Data flow generation: Use the intelligent agent based on the large model to perform iterative reasoning and establish a mapping relationship between the output of the upstream node and the input of the downstream node in the control flow, thereby establishing a complete data flow;

[0055] Utilizing multi-agent technology based on large models, according to the user's business needs description (covering comprehensive analysis, statistics, and other needs of various types of park operation and management data), the system automatically selects and assembles appropriate APIs from the garden intelligent management system and generates workflows that meet user needs. This can effectively reduce the establishment and implementation of new information technology projects in garden scenic areas, saving development costs, and quickly responding to user needs. During the implementation process, the semantic environment related to user needs is improved through task expression enhancement, avoiding the propagation of users' weak semantic expressions into control flow and data flow generation, so that the generated workflow can better meet user needs. During the control flow and data flow generation stage, an audit mechanism is introduced to verify and correct the interactive feedback of a pair of agents to generate correct output. The method and device for automatically implementing scenic area operation data analysis based on multi-agents are tested in combination with actual projects to verify its usability. The details are as follows:

[0056] The operating environment for the "Method and Device for Automatically Implementing Scenic Area Operation Data Analysis Based on Multi-Agents" is: Intel i9-10900X processor, NVIDIA GeForce RTX 3090 GPU, 128G memory, Python 3.6 and Pytorch 1.4.0.

[0057] Software Environment: This device is currently used in conjunction with our company's "Garden Smart Management System." The interface standards will be open to the public in the future to enable integration with other application systems. The "Garden Smart Management System" embeds KNIME, an open source workflow management system. Llama3-70B is used as the basic model to implement the relevant functions of multiple intelligent agents.

[0058] Device usage: Users can input business requirements in text form, such as: counting the relationship between the number of tourists and the revenue of a certain amusement project, counting the relevant information of a certain violation warning, etc. The device will automatically generate a workflow and execute the workflow to obtain relevant results.

[0059] See also Figure 1 、 Figure 2 and Figure 3 ,Step 1 also includes: TMAgent and TCAgent, where TMAgent checks the task description provided by the user, retrieves relevant APIs using a predefined API library or real-time search function, reads the API usage instructions, comprehensively analyzes the correlation between the task and the API, summarizes the problems and improves the task description, TCAgent checks each improved task description and optimizes it accordingly, and TMAgent and TCAgent gradually extract information related to the user's task description from the API usage instructions through multiple rounds of discussion, eliminates semantic problems in the task description provided by the user, and provides a more detailed and context-related understanding of the task requirements;

[0060] Compared with previous methods, the iterative correction of the two intelligent agents improves the detailed semantic description of the task, avoids the propagation of weak semantic expressions to downstream tasks, and the subsequent generation process of control flow and data flow, and improves the ease of use of the generated workflow.

[0061] See also Figure 1 、 Figure 2 and Figure 3 , step 2 also includes: setting up two intelligent agents, APISelAgent and APIRevAgent. When building the control flow, APISelAgent selects an API suitable for the current step operation from the candidate API set as the new node of the control flow based on the first step operation in the enhanced task description and the dependency relationship between APIs recorded in the API usage instructions; then APIRevAgent reviews whether the selected API is suitable and provides feedback; APISelAgent decides whether to reselect the API based on the feedback; if it decides to reselect, it selects the API again and submits it to APIRevAgent for review; if it decides to keep the original selection or APIRevAgent confirms that the selected API is suitable, then once the API is confirmed to be selected, it enters the next process and updates the semantic environment (composed of task description, existing nodes in the control flow, dependency relationships between nodes, APIs bound to each node, API usage instructions, etc.), and repeats the operation until the entire process is constructed.

[0062] See also Figure 1 、 Figure 2 and Figure 3 , step three also includes:

[0063] A1. Two large language model-based agents, the parameter selection agent (denoted as ParaSelAgent) and the parameter review agent (denoted as ParaRevAgent), work together to establish a correct mapping relationship between the output of the upstream node and the input of the downstream node, and to ensure the accuracy of parameter transfer between APIs. ParaSelAgent first identifies a list of candidate parameters based on the current semantic environment (i.e., the nodes included in the control flow up to the current step, the dependencies between nodes, the APIs used by each node, and API usage instructions). It then establishes a mapping between each parameter in the candidate parameter list (the output of the upstream node) and the input of the downstream node.

[0064] Based on a large language model, ParaSelAgent can automatically identify candidate parameter lists according to the current semantic environment and establish mapping relationships with the input parameters of downstream nodes; this greatly reduces manual intervention and improves work efficiency. Through deep learning and big data analysis, ParaSelAgent can more accurately understand the semantic environment, thereby identifying the most appropriate candidate parameters and establishing accurate mapping relationships; this reduces the risk of parameter transmission errors. ParaSelAgent can dynamically adjust the mapping strategy as the semantic environment changes to ensure the accuracy and applicability of the mapping relationship; this enhances the flexibility and adaptability of the system.

[0065] See also Figure 1 、 Figure 2 and Figure 3 , step three also includes:

[0066] A2. ParaSelAgent takes a parameter from the candidate parameter list and matches it with the input parameter of the next node API;

[0067] Through automated matching, the time spent on manual parameter search and matching is reduced, improving overall work efficiency. ParaSelAgent can ensure the accuracy and applicability of matching parameters based on semantic understanding and context analysis; this reduces the possibility of parameter passing errors. As the API set is updated and expanded, ParaSelAgent can automatically adapt to new APIs and parameters to ensure the continuity and accuracy of matching.

[0068] See also Figure 1 、 Figure 2 and Figure 3 , step three also includes:

[0069] A3. ParaRevAgent reviews the matching results and provides feedback. ParaSelAgent then optimizes the matching based on the feedback. Regardless of whether the matching result is confirmed or rejected, the current semantic environment is updated and the next parameter matching begins. When the matching result is confirmed, the mapping relationship of the matching parameter is added to the current semantic environment. When the matching result is rejected, the current semantic environment records that the parameter is still not matched until all candidate parameters are processed.

[0070] In terms of parameter review and optimization, ParaRevAgent conducts paired review of matching results, and ParaSelAgent optimizes matching based on feedback. ParaRevAgent's review process provides a double check for matching results, ensuring their accuracy and reliability. This reduces the risk of system failure or data loss due to parameter transfer errors. Through ParaRevAgent's feedback, ParaSelAgent can optimize matching results to improve matching accuracy and efficiency. This feedback and optimization mechanism enables the system to continuously improve and perfect. Through parameter review and optimization, the system can promptly detect and correct potential parameter transfer errors, thereby improving system stability and reliability. This ensures the stability and accuracy of the system during long-term operation.

[0071] See also Figure 1 、 Figure 2 and Figure 3 , step three also includes:

[0072] A4. Enter the next process node and repeat operations A1-A3 until the process ends. By adding data flow to the control flow, a complete and executable workflow is formed.

[0073] Step A4 realizes the automation and intelligence of the process by introducing an intelligent agent based on a large language model. The intelligent agent can automatically enter the next process node according to the information of the control flow and data flow, and repeat operations such as parameter selection, matching, review and optimization until the end of the process. This automated and intelligent processing method greatly reduces manual intervention and improves work efficiency. By adding data flow to the control flow, a complete and executable workflow is formed. This means that every step of the entire process from input to output is clearly defined and executed, ensuring the integrity and executability of the workflow, which helps to reduce errors and failures caused by incomplete processes or inconsistent execution. The intelligent agent based on the large language model has strong semantic understanding and context analysis capabilities, which This enables them to flexibly adjust workflows and parameter mapping strategies according to different semantic environments and task requirements, thereby demonstrating greater flexibility and adaptability when handling processes of different types and sizes. The parameter review agent (ParaRevAgent) performs paired review of matching results and provides feedback. This review mechanism helps to promptly detect and correct potential parameter transfer errors, thereby improving the stability and reliability of the entire process. Due to the use of an agent based on a large language model, system maintenance and upgrades become more convenient. The agent can continuously optimize its performance through continuous learning and training to adapt to new task requirements and semantic environments, which reduces the cost of system maintenance and upgrades and improves the scalability and sustainability of the system.

[0074] A device for automatically analyzing scenic area operation data based on multiple intelligent agents includes: a task description enhancement module, a control flow generation module and a data flow generation module. The output end of the task description enhancement module is connected to the input end of the control flow generation module through data, and the output end of the control flow generation module is connected to the input end of the data flow generation module through data.

[0075] See also Figure 1 、 Figure 2 and Figure 3 ,The task description enhancement module is used to receive the task description provided by the ,user, and retrieve the relevant API in the garden smart ,management system, and improve the task requirement description according to the ,API usage instructions. The task description enhancement module includes TMAgent and TCAgent. TMAgent ,is responsible for checking the task description, retrieving the API, reading the API usage instructions and ,comprehensively analyzing the correlation between the task and the API, while ,TCAgent is responsible for checking and optimizing the improved task ,description, gradually extracting relevant information and eliminating semantic ,problems through multiple rounds of discussion;

[0076] Through the collaborative work of TMAgent and TCAgent, it can ensure that the task description is highly matched with the API in the garden smart management system, reducing misunderstandings and ambiguities. TMAgent can retrieve the API usage instructions and comprehensively analyze the correlation between the task and the API, thereby improving the task requirement description and improving the accuracy and efficiency of task execution. The task description enhancement module gradually extracts relevant information and eliminates semantic problems through multiple rounds of discussion, allowing users to understand the task requirements more clearly and reduce the difficulty of operation. The optimized task description helps users find the API that suits them more quickly, improve user satisfaction and loyalty, and optimize user experience. The task description enhancement module can automatically identify and process key Information, improve the automation and intelligence level of the system. Through continuous learning and optimization, the task description enhancement module can better adapt to changes in user needs, improve the overall performance of the system, and enhance the intelligence level of the system. The control flow generation module selects APIs from the garden smart management system according to the enhanced task description and gradually generates new control nodes until a complete control flow is generated. The control flow generation module includes APISelAgent and APIRevAgent. APISelAgent selects appropriate APIs as new nodes of the control flow according to the dependency between the task description and the API. APIRevAgent is responsible for reviewing the selected APIs and providing feedback to ensure the correct construction of the control flow.

[0077] APISelAgent can select the appropriate API as the new node of the control flow according to the dependency between the task description and the API, reducing the tedious process of manual selection and configuration. APIRevAgent is responsible for reviewing the selected API and providing feedback to ensure the correct construction of the control flow, reduce the error rate, and simplify the control flow construction process. The control flow generation module supports dynamic addition and deletion of control nodes, so that the control flow can be flexibly adjusted according to different task requirements. Through integration with the garden smart management system, the control flow generation module can automatically obtain the latest API information, ensure the scalability and compatibility of the control flow, and improve the flexibility and scalability of the control flow. The optimized control flow can reduce unnecessary waiting time and resource consumption, and improve the execution efficiency of the system. Through reasonable control flow design, the system load can be balanced, the overall performance and stability can be improved, and the system execution efficiency can be improved.

[0078] See also Figure 1 、 Figure 2 and Figure 3,The data flow generation module uses an agent based on a large model to perform iterative reasoning, ,and establishes a mapping relationship between the output of upstream nodes and the input,of downstream nodes in the control flow, thereby establishing a complete data flow.,The data flow generation module includes ParaSelAgent and ParaRevAgent, which are used to identify,the candidate parameter list, establish parameter mapping, perform matching,audit and optimization, and ensure the accuracy of parameter transfer,between APIs;

[0079] The data flow generation module uses an intelligent agent based on a large model to perform iterative reasoning, establish a mapping relationship between upstream node output and downstream node input, and ensure accurate data transmission. ParaSelAgent and ParaRevAgent are responsible for identifying candidate parameter lists, establishing parameter mapping, and performing matching review and optimization to further reduce the risk of data transmission errors. The data flow generation module can automatically identify and process key information in the data flow, reducing the tedious process of manual intervention and operation. Through integration with the garden smart management system, the data flow generation module can automatically obtain and process relevant data, improve the degree of automation of data processing, and optimize the data flow to reduce the risk of data loss and damage, improve the reliability and stability of the system, and through reasonable parameter mapping and matching review mechanisms, it can ensure the smooth transmission and processing of data flow in the system, reducing the risk of system crash and failure.

[0080] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0081] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for automatically analyzing scenic spot operation data based on multi-agent, characterized in that: The steps include: Step 1: Enhance task description Establishing a garden smart management system, which includes a user terminal. The user terminal sets up a task placement port for users to upload tasks, and searches for relevant APIs in the garden smart management system through keyword matching, classification search, or semantic analysis based on the task description uploaded by the user, and improves and enhances the task requirement description based on the retrieved API usage documents or instructions; Step 2: Control flow generation According to the enhanced task description, select API from the garden smart management system and gradually generate new control nodes until a complete control flow is generated; Step 3: Data stream generation Using large-model-based agents for iterative reasoning, we establish a mapping relationship between the output of upstream nodes and the input of downstream nodes in the control flow, thus establishing a complete data flow. The step one further includes: TMAgent and TCAgent, TMAgent checking the task description provided by the user, using a predefined API library or a real-time search function to retrieve relevant APIs, reading API usage instructions, comprehensively analyzing the correlation between the task and the API, summarizing problems and improving the task description, TCAgent checking each improved task description and performing corresponding optimizations, TMAgent and TCAgent gradually extracting information related to the user's task description from the API usage instructions through multiple rounds of discussions, and eliminating semantic problems in the task description provided by the user; The step 2 further includes: setting up two intelligent agents, APISelAgent and APIRevAgent, and when constructing the control flow, APISelAgent selects an API suitable for the current step operation from the candidate API set as a new node of the control flow based on the first step operation in the enhanced task description and the dependency relationship between the APIs recorded in the API usage instructions; then APIRevAgent reviews whether the selected API is suitable and provides feedback; APISelAgent decides whether to reselect the API based on the feedback; if it decides to reselect, it selects the API again and submits it to APIRevAgent for review; if it decides to keep the original selection or APIRevAgent confirms that the selected API is suitable, then once the API is confirmed to be selected, it enters the next process, updates the semantic environment, and repeats the operation until the entire process is constructed; The step three also includes: A1, ParaSelAgent and ParaRevAgent. In order to establish a correct mapping relationship between the output of the upstream node and the input of the downstream node and ensure the accuracy of parameter transfer between APIs, ParaSelAgent first identifies the candidate parameter list according to the current semantic environment, and then establishes a mapping between each parameter in the candidate parameter list and the input of the downstream node.

2. The method for automatically analyzing scenic area operation data based on multi-agent according to claim 1, characterized in that: The step three also includes: A2. ParaSelAgent takes a parameter from the candidate parameter list and matches it with the input parameter of the next node API.

3. The method for automatically analyzing scenic area operation data based on multi-agent according to claim 2, characterized in that: The step three also includes: A3. ParaRevAgent reviews the matching results and provides feedback. ParaSelAgent then optimizes the matching based on the feedback. Regardless of whether the matching result is confirmed or rejected, the current semantic environment is updated and the next parameter matching begins, until all candidate parameters are processed.

4. The method for automatically analyzing scenic spot operation data based on multi-agent according to claim 3 is characterized in that: The step three also includes: A4. Enter the next process node and repeat the operations of A1-A3 until the process ends. By adding data flow to the control flow, a complete executable workflow is formed.

5. A device for automatically analyzing scenic area operation data based on multi-agents, comprising a method for automatically analyzing scenic area operation data based on multi-agents according to any one of claims 1 to 4, characterized in that: include: A task description enhancement module, a control flow generation module and a data flow generation module, wherein the output end of the task description enhancement module is connected to the input end of the control flow generation module via data, and the output end of the control flow generation module is connected to the input end of the data flow generation module via data.

6. The device for automatically analyzing scenic area operation data based on multi-agent according to claim 5, characterized in that: The task description enhancement module receives the task description provided by the user, retrieves the relevant API in the garden smart management system, and improves the task requirement description according to the API usage instructions. The control flow generation module selects the API from the garden smart management system based on the enhanced task description and gradually generates new control nodes until a complete control flow is generated.

7. The device for automatically analyzing scenic area operation data based on multiple agents according to claim 6, characterized in that: The data flow generation module uses an intelligent agent based on a large model to perform iterative reasoning, establishes a mapping relationship between the output of the upstream node and the input of the downstream node in the control flow, and thus establishes a complete data flow.

Images