A multi-agent cooperation management method and system for water affair data governance

Abstract

This invention relates to a multi-agent collaborative management method and system for water data governance, belonging to the field of water data governance technology. It includes: constructing a data asset management agent, a quality diagnosis agent, and a conversational analysis agent, and setting a standardized decision-making process; building a structured message queue; when water demand processing data is received, the data asset management agent analyzes the water demand processing data to generate water management data and sends it to the message queue; when the quality diagnosis agent listens for water management data in the message queue, it retrieves the water management data, analyzes it, generates governance strategy instructions, and sends them to the message queue; combining the conversational analysis agent and the governance strategy instructions, it determines governance adjustment instructions and executes the governance adjustment instructions. This invention improves the overall efficiency of water data governance and shortens the average problem resolution cycle.

Description

Technical Field

[0001] This invention relates to the field of water data governance technology, and in particular to a multi-agent collaborative management method and system for water data governance. Background Technology

[0002] Water data governance refers to a specialized data management process that uses standardized and systematic management methods to regulate and control the entire lifecycle of data generated in the water sector, from generation, collection, storage to application, ultimately achieving reliable data quality, efficient process collaboration, and tangible data value.

[0003] Currently, water data governance mainly adopts the traditional implementation model of manual processing plus rule engines. First, the data sources, data formats, and business logic of various water business systems are manually sorted out to clarify the scope and basic requirements of data governance. Then, based on the sorting results, technical personnel manually configure basic verification standards such as data integrity, accuracy, and consistency. Next, data probing tools are used to check for data anomalies, quality diagnostic tools are used to initially identify the types of problems, and lineage tracing tools are used to locate the data flow path. When the tools detect data anomalies, manual intervention is required to analyze the causes of the anomalies, formulate repair plans based on the experience of engineers, and manually execute data correction operations.

[0004] Because core processes such as data exploration, quality diagnosis, and lineage tracing rely on independent tools and lack effective collaboration mechanisms, the entire data chain from anomaly discovery to analysis and problem resolution is not well connected, resulting in low governance efficiency and a long average problem resolution cycle. Summary of the Invention

[0005] To improve the overall efficiency of water data governance and shorten the average problem resolution cycle, this invention provides a multi-agent collaborative management method and system for water data governance.

[0006] Firstly, the present invention provides a multi-agent collaborative management method for water data governance, employing the following technical solution: A multi-agent collaborative management method for water data governance includes: Construct intelligent agents for data asset management, quality diagnosis, and conversational analysis, and establish standardized decision-making processes; Build a structured message queue, which is used to share the input and output of the data asset management agent, the quality diagnosis agent, and the conversational analysis agent in real time; When water demand processing data is received, the data asset management intelligent agent analyzes the water demand processing data to generate water management data and sends it to the message queue; When the quality diagnostic agent listens to water management data from the message queue, it retrieves the water management data, analyzes it, generates governance strategy instructions, and sends them to the message queue. By combining conversational analytical agents with governance policy instructions, governance adjustment instructions are determined and executed.

[0007] By adopting the above technical solutions, and by constructing a data asset management intelligent agent, a quality diagnosis intelligent agent, and a conversational analysis intelligent agent, and setting standardized decision-making processes, and by using a structured message queue to achieve real-time sharing of input and output among the three intelligent agents, water demand processing data can be analyzed by the data asset management intelligent agent to generate water management data. This allows the quality diagnosis intelligent agent to quickly generate governance strategy instructions, and then the conversational analysis intelligent agent to determine the governance adjustment instructions. This achieves coordinated linkage of the governance process, significantly shortens the time spent on the entire chain from receiving and processing water data to executing governance adjustments, and improves the overall efficiency of water data governance.

[0008] Optionally, methods for generating governance adjustment instructions include: Determine whether user operation commands have been collected; If so, the conversational analysis agent retrieves governance policy instructions from the message queue and combines them with user operation instructions to generate user suggestion instructions and display information. The user suggestion instructions are then used as governance adjustment instructions, and the display information is output for presentation. If not, then the governance strategy directive will be treated as a governance adjustment directive.

[0009] By adopting the above technical solution, when user operation instructions are collected, the system combines the governance strategy instructions retrieved by the conversational analysis agent with the user operation instructions to generate user suggestion instructions and display information. If no user operation instructions are collected, the governance strategy instructions are used directly. This approach takes into account the user's need for autonomous decision-making, making the governance adjustment instructions more in line with actual usage scenarios, while also ensuring the continuity of the system's autonomous governance. This improves the adaptability and practicality of the governance adjustment instructions, and enhances the user's perception of the governance process by outputting display information.

[0010] Optional methods for setting up standardized decision-making processes include: Collect intelligent agent types and input information, including water demand processing data, governance strategy instructions, and user operation instructions; Determine the perception type, model type, action type, and observation baseline parameters based on the agent type; Input information is standardized based on the type of perception to form perception data; Based on the model type, the corresponding model is retrieved, and reasoning is performed on the perceived data to obtain the reasoning result; The system combines action type and reasoning results to generate and execute a decision plan, and collects execution detection parameters. The decision output results are determined and output by combining the execution detection parameters and the observation benchmark parameters. The decision output results include water management data, governance strategy instructions and display information.

[0011] By adopting the above technical solution, and by collecting agent type and input information, matching the corresponding perception type, model type, action type and observation benchmark parameters, we can achieve standardized processing of input information, accurate model reasoning, generation and execution of decision-making schemes, and output of results verification. This ensures that different agents follow a unified and standardized process when processing various types of input information, thereby reducing decision-making bias.

[0012] Optionally, methods for determining the decision output include: Retrieve execution detection results and execution time based on execution detection parameters; Retrieve observation baseline specifications and efficiency baseline values ​​based on observation baseline parameters; The test reference values ​​were determined by combining the test results with the observation benchmark specifications; An efficiency reference value is determined by combining the execution time value with the efficiency benchmark value; Determine whether the detection reference value and efficiency reference value are both less than the preset reference benchmark value; If so, the detection parameters will be used as the decision output. If not, then a comprehensive reference value is determined by combining the detection reference value and the efficiency reference value, and the comprehensive reference value is fed back to the model corresponding to the model type for re-inference.

[0013] By adopting the above technical solution, the execution detection result and execution time value are determined by executing the detection parameters. The detection reference value and efficiency reference value are calculated by combining the observation benchmark and efficiency benchmark value. The execution detection parameters are only output when both are less than the preset reference benchmark value. Otherwise, the feedback is fed back to the corresponding model for re-inference. This achieves dual verification of the decision execution effect and efficiency, and improves the quality of the decision output result and the execution efficiency.

[0014] Optionally, the following can be included before executing the detection parameters as the decision output: Retrieve detection fields based on execution detection parameters; Retrieve baseline fields and field baseline values ​​based on observed baseline parameters; Determine the accuracy of the fields based on the detection fields and the benchmark fields; The actual number of fields is determined based on the fields being detected; Calculate the ratio between the actual number of fields and the baseline number of fields, and use this ratio as the field completeness. Determine data quality reference values ​​by combining field accuracy and field completeness; Determine whether the data quality reference value is less than the preset reference baseline value; If so, the detection parameters will be used as the decision output. If not, the data quality reference value will be fed back to the model corresponding to the model type for re-inference.

[0015] By adopting the above technical solution, the accuracy and completeness of the fields are calculated by retrieving the detection field and the benchmark field to determine the data quality reference value. The data quality reference value is only output when it is less than the preset reference benchmark value; otherwise, it is fed back to the model for re-inference, which further strengthens the data quality control of the decision output results.

[0016] Optionally, after determining the data quality reference values, the following may also be included: The source method is determined based on the execution detection parameters; The number of sources is determined based on the source method; Determine if the number of sources is only one; If yes, continue outputting data quality reference values; If not, retrieve the source field based on the source method; The consistency of fields is determined by combining the source field and the detection field; The data quality reference values ​​are updated and adjusted based on field consistency.

[0017] By adopting the above technical solution, the number of sources is determined according to the source method of the execution detection parameters. When there are multiple sources, the source fields are retrieved to calculate the field consistency and the data quality reference value is updated and adjusted. This makes the data quality reference value not only reflect the accuracy and completeness of the fields, but also reflect the field consistency of data from multiple sources, thereby improving the comprehensiveness and accuracy of the data quality reference value.

[0018] Optionally, after executing the governance adjustment instructions, the following may also be included: Decision paths for data asset management agents, quality diagnosis agents, and conversational analysis agents; Generate experience examples based on the decision-making path; The graph neural network algorithm is used to learn from experience instances and discover the optimal policy pattern. The pre-defined governance knowledge graph is updated based on the optimal strategy model.

[0019] By adopting the above technical solution, experience instances are generated by collecting the decision paths of three intelligent agents, and the optimal strategy pattern is learned and updated by graph neural network algorithm. This achieves the effective accumulation and reuse of water data governance experience, avoids the loss of governance experience of senior engineers, and allows the governance knowledge graph to be continuously enriched and improved, providing better knowledge support for the subsequent decision-making and reasoning of intelligent agents.

[0020] Optionally, updating the pre-defined governance knowledge graph based on the optimal strategy pattern includes: Inputting empirical examples into a pre-defined governance knowledge graph to retrieve similar examples; Retrieve similar strategies based on similar instances; Test according to similar strategies and collect data quality reference values ​​as similar reference values; The optimal initial strategy is retrieved based on the optimal strategy pattern; Test according to the optimal initial strategy and collect data quality reference values ​​as the optimal reference values; A relative selection strategy is determined by combining similar reference values ​​and the optimal reference value; The relative strategy pattern is determined based on the relative selection strategy, and the relative strategy pattern replaces the optimal strategy pattern.

[0021] By adopting the above technical solution, similar strategies are generated by retrieving similar instances. Similar reference values ​​and optimal reference values ​​are tested and collected with the optimal initial strategy, respectively. Then, the relative selection strategy is determined and the optimal strategy mode is replaced. This ensures that the updated optimal strategy mode has been verified by actual testing. Compared with the original mode, it is more in line with the needs of water data governance and improves the practicality and adaptability of the strategy mode in the governance knowledge graph.

[0022] Optional methods for determining the relative selection strategy include: Determine whether the similar reference value is consistent with the optimal reference value; If so, then collect similar human intervention rates based on similar strategies; The initial human intervention rate is collected based on the optimal initial strategy; Based on the comparison results between the initial human intervention rate and the similar human intervention rate, the human intervention selection strategy is determined and used as the relative selection strategy. If not, then based on the comparison results between similar reference values ​​and the optimal reference value, a reference selection strategy is determined, and the reference selection strategy is used as the relative selection strategy.

[0023] By adopting the above technical solution, and by comparing the similar reference value with the optimal reference value, the decision is made by either the rate of human intervention or the reference value. This achieves precision in strategy selection and further optimizes the effectiveness and efficiency of water data governance.

[0024] Secondly, this invention provides a multi-agent collaborative management system for water data governance, employing the following technical solution: A multi-agent collaborative management system for water data governance includes: The data acquisition module is used to collect user operation commands, agent type, input information, execution detection parameters, decision path, similar reference value, optimal reference value, similar human intervention rate, and initial human intervention rate. The memory stores a program for implementing a multi-agent collaborative management method for water data governance as described in any one of the first aspects; The processor loads and executes programs stored in memory.

[0025] In summary, the present invention has at least one of the following beneficial technical effects: 1. By constructing a data asset management intelligent agent, a quality diagnosis intelligent agent, and a conversational analysis intelligent agent, and setting standardized decision-making processes, and using a structured message queue to achieve real-time sharing of input and output among the three intelligent agents, water demand processing data can be analyzed by the data asset management intelligent agent to generate water management data, which can then quickly trigger the quality diagnosis intelligent agent to generate governance strategy instructions. The conversational analysis intelligent agent can then determine the governance adjustment instructions, achieving coordinated linkage of the governance process. This significantly shortens the time consumed in the entire chain from receiving and processing water data to executing governance adjustments, and improves the overall efficiency of water data governance. 2. When user operation commands are collected, the system combines the governance strategy commands retrieved by the conversational analysis agent with the user operation commands to generate user suggestion commands and display information. If no user operation commands are collected, the governance strategy commands are used directly. This approach takes into account the user's need for autonomous decision-making, making the governance adjustment commands more in line with actual usage scenarios, while also ensuring the continuity of the system's autonomous governance. This improves the adaptability and practicality of the governance adjustment commands. At the same time, by outputting display information, the system enhances the user's perception of the governance process. 3. By collecting decision-making paths of three intelligent agents to generate experience instances, and by using graph neural network algorithms to learn and mine optimal strategy patterns and update the governance knowledge graph, the effective accumulation and reuse of water data governance experience is realized, avoiding the loss of governance experience of senior engineers, while continuously enriching and improving the governance knowledge graph, providing better knowledge support for subsequent decision-making and reasoning of intelligent agents. Attached Figure Description

[0026] Figure 1 This is a flowchart of a multi-agent collaborative management method for water data governance. Detailed Implementation

[0027] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0028] A multi-agent collaborative management method for water data governance is proposed. This method constructs three agents—data asset management, quality diagnosis, and conversational analysis—and establishes a standardized decision-making process. It relies on a structured message queue to achieve real-time sharing of agent input and output. Upon receiving water demand processing data, the data asset management agent generates water management data, triggering the quality diagnosis agent to produce governance strategy instructions. It flexibly generates governance adjustment instructions based on whether user operation instructions are present. The quality of the decision output is verified through field accuracy, completeness, and consistency. Furthermore, it generates experience instances by collecting agent decision paths and updates the governance knowledge graph by mining optimal strategy patterns using a graph neural network. This achieves collaborative linkage of the governance process, significantly shortening the entire time required from receiving and processing water data to executing governance adjustments, and improving the overall efficiency of water data governance.

[0029] Reference Figure 1 This invention discloses a multi-agent collaborative management method for water data governance, comprising: S100: Construct intelligent agents for data asset management, quality diagnosis, and conversational analysis, and set up standardized decision-making processes.

[0030] Among them, the data asset management intelligent agent refers to an intelligent module specifically responsible for asset management tasks such as water data exploration, registration, and catalog generation. The quality diagnosis intelligent agent refers to an intelligent module focusing on water data anomaly detection, root cause analysis, and the generation of governance strategy instructions. The conversational analysis intelligent agent refers to an intelligent module that supports natural language interaction, enabling data querying and result visualization. The data asset management intelligent agent, quality diagnosis intelligent agent, and conversational analysis intelligent agent are pre-built by the operator.

[0031] A standardized decision-making process refers to a unified and standardized "perception-processing-decision-output" operational flow set for data asset management agents, quality diagnosis agents, and conversational analysis agents to ensure consistency in decision-making. The standardized decision-making process is pre-set by the operator.

[0032] S101: Build a structured message queue.

[0033] The message queue is used to share the input and output of the data asset management agent, the quality diagnosis agent, and the conversational analysis agent in real time. The messages in the message queue have a fixed format (including message header, message body, and verification field), so as to standardize the storage and transmission of the input information and output results of the data asset management agent, the quality diagnosis agent, and the conversational analysis agent, and ensure that the data interaction between the agents is accurate, orderly, and parsable.

[0034] S102: When water demand processing data is received, the data asset management agent analyzes the water demand processing data to generate water management data and sends it to the message queue.

[0035] Water demand processing data refers to the raw data that triggers the water data governance process, covering various types of data that need to be governed in all water business scenarios (such as water quality monitoring data, equipment operating parameters, user water usage records, etc.). Water management data refers to the structured and standardized data products formed after analysis and processing by the data asset management intelligent agent. Water management data includes data catalogs, metadata, preliminary verification results, etc.

[0036] The data asset management intelligent agent interfaces with data sources such as water sensors, smart water meters, sewage treatment plant PLC systems, and user business processing platforms. When water demand processing data is received from various data sources, the data asset management intelligent agent analyzes and processes the water demand processing data according to a standardized decision-making process, thereby forming water management data and sending it to a message queue for subsequent use.

[0037] S103: When the quality diagnosis agent listens to water management data from the message queue, it retrieves the water management data, analyzes it, generates governance strategy instructions, and sends them to the message queue.

[0038] Among them, the governance strategy instructions are instructions used to guide the adjustment of water data governance. The governance strategy instructions include core information such as anomaly type identifier, root cause analysis conclusion, specific remediation operations, execution parameters and scope of application.

[0039] The quality diagnostic agent uses a pre-set queue listening mechanism to capture push notifications of water management data in the message queue in real time, automatically retrieves the data, and then analyzes and processes it according to a standardized decision-making process to form governance strategy instructions, which are then sent to the message queue for subsequent use.

[0040] S104: Combine the conversational analysis agent with the governance strategy instructions to determine the governance adjustment instructions and execute the governance adjustment instructions.

[0041] Among them, governance adjustment instructions refer to the instructions corresponding to the adjustments made to governance strategy instructions.

[0042] By combining and analyzing the conversational analytical agent with governance strategy instructions, governance adjustment instructions are determined and executed, thereby achieving coordinated linkage of the governance process. This significantly shortens the time consumed in the entire chain from receiving and processing water data to executing governance adjustments, and improves the overall efficiency of water data governance.

[0043] To further ensure the rationality of the governance adjustment directives, it is necessary to conduct further separate analysis and calculation of the directives, which will be explained in detail through the steps shown below.

[0044] The method for generating governance adjustment instructions includes the following steps: S200: Determine whether a user operation command has been collected. If yes, execute S201; if no, execute S202.

[0045] User operation instructions refer to various operation commands issued by users through the system's interactive terminal in response to water data governance needs. User operation instructions include governance and repair suggestions, manual intervention requests, data query requests, and other information.

[0046] By monitoring the system's interactive interface, it can be determined whether user operation commands have been collected and captured, and then whether the governance strategy commands need to be adjusted.

[0047] S201: The conversational analysis agent retrieves governance policy instructions from the message queue and combines them with user operation instructions to generate user suggestion instructions and display information. The user suggestion instructions are then used as governance adjustment instructions, and the display information is output for presentation.

[0048] User-suggested instructions refer to commands that adjust water management data based on user intent. Displayed information refers to the visual information generated after the management is implemented according to user-suggested instructions.

[0049] When user operation commands are collected, it indicates that the governance strategy commands need to be adjusted. Therefore, the conversational analysis agent retrieves the governance strategy commands from the message queue and combines them with the user operation commands. Then, it analyzes and processes the commands according to a standardized decision-making process to generate user suggestion commands and display information. The user suggestion commands are used as governance adjustment commands, and the output display information is shown. This takes into account the user's need for autonomous decision-making, makes the governance adjustment commands more in line with actual use scenarios, and improves the adaptability and practicality of the governance adjustment commands. At the same time, by outputting display information, the user's perception of the governance process is enhanced.

[0050] S202: Treat governance strategy directives as governance adjustment directives.

[0051] When no user operation commands are collected, it means that no adjustment to the governance strategy commands is needed at this time, so the governance strategy commands are treated as governance adjustment commands.

[0052] To further ensure the rationality of the standardized decision-making process, it is necessary to conduct further separate analysis and calculation of the standardized decision-making process, which will be explained in detail through the steps shown below.

[0053] The method for setting up a standardized decision-making process includes the following steps: S300: Collects agent type and input information.

[0054] Among them, the agent type refers to the type of agent that needs to make analysis and decisions. Agent types include data asset management agents, quality diagnosis agents, and conversational analysis agents.

[0055] Input information refers to the data currently input into the intelligent entity. Input information includes water demand processing data, governance strategy instructions, and user operation instructions.

[0056] S301: Determine the perception type, model type, action type, and observation baseline parameters based on the agent type.

[0057] Among these, perception type refers to the type of input information captured and standardized. Model type refers to the category of inference models adapted to the decision-making needs of the agent, providing algorithmic and model support for the agent to analyze perception data and generate decision results. Action type refers to the specific operation category that the agent needs to perform based on the inference results, clarifying the core execution action of the agent. Observation benchmark parameters refer to the quantitative benchmark indicators for judging the effectiveness of the agent's action execution.

[0058] Different agent types correspond to different perception types, model types, action types, and observation benchmark parameters. By inputting the agent type into a preset agent database, the perception type, model type, action type, and observation benchmark parameters are matched to facilitate subsequent use.

[0059] The agent database pre-stores a lookup table of different agent types and their corresponding perception types, model types, action types, and observation benchmark parameters. The agent database is obtained after pre-input by the operator.

[0060] For example, when the agent type is a data asset management agent, the corresponding perception type is water raw data capture and structured processing, the model type is data asset classification reasoning model, the action type is data exploration + registration + catalog generation, and the observation benchmark parameters are set as field non-empty rate ≥99% and record completeness rate ≥98%; when the agent type is a quality diagnosis agent, the corresponding perception type is water management data monitoring and anomaly feature extraction, the model type is data anomaly detection and root cause analysis model, the action type is governance strategy instruction generation, and the observation benchmark parameters are set as anomaly identification accuracy ≥95% and strategy generation time ≤5 minutes; when the agent type is a conversational analysis agent, the corresponding perception type is user instruction semantic parsing and queue data retrieval, the model type is NL2SQL conversion and visualization reasoning model, the action type is user suggestion instruction generation + information display output, and the observation benchmark parameters are set as instruction parsing accuracy ≥90% and visualization result generation time ≤3 minutes.

[0061] S302: Standardize the input information based on the perception type to form perception data.

[0062] Among them, perceived data refers to standardized data formed after the input information has undergone standardized processing of the corresponding perception type, resulting in data with a unified format, clear elements, and direct parsing and processing by the intelligent agent reasoning model.

[0063] By matching the perception type with the intelligent agent, corresponding standardized processing rules are formulated, and then the input information collected or retrieved by the intelligent agent is processed in a targeted manner. The core is to unify the data format, extract the core elements, verify the basic validity, and eliminate invalid and redundant information, ultimately forming perception data.

[0064] For example, the data asset management agent processes data-type input information for water demand, performs structured splitting and field completion according to the rule of "data object-indicator value-collection time", and transforms unformatted raw data into standardized perceptual data; the conversational analysis agent processes user operation command-type input information, extracts the core elements of "operation intention-processing object-execution requirements" through semantic parsing, and transforms them into text-type perceptual data that the model can recognize.

[0065] S303: Based on the model type, retrieve the corresponding model and perform inference on the perceived data to obtain the inference result.

[0066] Among them, the reasoning result refers to the decision-making conclusion obtained by the intelligent agent after retrieving the corresponding model to perform algorithmic reasoning and logical analysis on the perceived data.

[0067] The system pre-stores the association mapping relationship between model types and specific inference models. After the agent determines its own model type, it automatically retrieves the corresponding model and uses standardized perception data as input to analyze and deduce the perception data, and finally outputs the inference result.

[0068] S304: Combine action type and reasoning results to generate a decision plan and execute it, and collect execution detection parameters.

[0069] Among them, the decision-making scheme refers to the specific operational plan generated by the intelligent agent based on its matched action type and reasoning results, which is clear in steps, complete in elements, and directly implementable. Execution detection parameters are various quantitative monitoring data collected by the intelligent agent throughout the entire process of executing the decision-making scheme, used to verify the execution effect. Execution detection parameters include dimensions such as execution quality, efficiency, and completion rate.

[0070] By extracting the core decision conclusions, execution directions, and key requirements from the reasoning results, and combining them with the corresponding action type of the intelligent agent, the specific operation actions, execution objects, operation parameters, and execution processes are clarified. The two are then integrated to form a standardized decision scheme, which is then executed. Furthermore, through the system's built-in full-process monitoring nodes, operation data during the execution process and result data after the execution are collected in real time, forming multi-dimensional execution detection parameters.

[0071] For example, the action type of the data asset management intelligent agent is data exploration + registration. The inference result is "Register the pipeline network data of Area A according to the drainage link". Based on this, a decision plan is generated to "Perform field verification on the pressure / flow data of the pipeline network of Area A, fill in the missing collection time, and complete the asset registration according to the drainage-pipeline network operation category". After execution, the field completion rate, registration completion time, and data classification accuracy are used as execution detection parameters. The action type of the quality diagnosis intelligent agent is the generation of treatment strategy instructions. The inference result is "The root cause of COD exceeding the standard in the influent is insufficient coagulant addition". A decision plan is generated to "Generate COD exceeding the standard treatment strategy instructions, clarify that the coagulant addition amount is adjusted to 8kg / hour, and improve the core elements such as anomaly type and repair operation". After execution, the time for instruction generation, the completeness of strategy elements, and the accuracy of anomaly root cause labeling are used as execution detection parameters.

[0072] S305: Combine the execution detection parameters with the observation benchmark parameters to determine the decision output result and output it.

[0073] The decision output refers to the compliant governance results ultimately output by the intelligent agent. The decision output includes water management data, governance strategy instructions, and displayed information.

[0074] By combining and analyzing the execution detection parameters with the observation benchmark parameters, the decision output results are determined and output, so that each intelligent agent follows a unified and standardized process when processing various types of input information, thereby reducing decision bias.

[0075] To further ensure the rationality of the decision output, it is necessary to perform further separate analysis and calculation on the decision output, which will be explained in detail through the steps shown below.

[0076] The method for determining the decision output includes the following steps: S400: Retrieves execution detection results and execution time based on execution detection parameters.

[0077] Among them, the execution detection results are the outcome data for each dimension after execution is completed. The execution time value refers to the quantitative indicator representing the total time taken from the initiation of the decision-making scheme to its completion. The execution detection parameters include the execution detection results and the execution time value.

[0078] The detection results and execution time can be retrieved by executing the detection parameters, which is convenient for subsequent use.

[0079] S401: Retrieve observation baseline specifications and efficiency baseline values ​​based on observation baseline parameters.

[0080] Among them, the observation benchmark specification refers to the qualitative judgment rules and standard descriptions for determining whether the execution test results meet the standards, clarifying the judgment logic and the basis for defining the scope of compliance for each dimension of test parameters. The efficiency benchmark value refers to the quantitative threshold used to measure the execution efficiency of the decision-making plan. The observation benchmark parameters include the observation benchmark specification and the efficiency benchmark value.

[0081] By observing the benchmark parameters, the benchmark specifications and efficiency benchmark values ​​can be retrieved for convenient subsequent use.

[0082] S402: Determine the test reference value by combining the test results with the observation benchmark specifications.

[0083] Among them, the detection reference value refers to the quantitative indicator that characterizes whether the test results meet the standard requirements.

[0084] The test results are evaluated to determine whether they meet the observation benchmark. For example, items that meet the standard are scored as 0 points, minor deviations are scored as fixed points, and serious deviations are scored as higher points. Finally, the test reference value is calculated by combining the scores of each test result through weighted average or direct summation. The lower the value, the better the execution quality meets the standard requirements, which is convenient for subsequent use.

[0085] S403: Determine the efficiency reference value by combining the execution time value and the efficiency benchmark value.

[0086] Among them, the efficiency reference value refers to a quantitative indicator that represents whether the execution time value meets the standard.

[0087] By comparing the execution time with an efficiency benchmark, when the execution time is less than the benchmark, 0 is used as the efficiency reference value. When the execution time is not less than the benchmark, the difference between the execution time and the benchmark is calculated, and then the product of this difference and a preset efficiency reference coefficient is used as the efficiency reference value for subsequent use. The larger the difference between the execution time and the benchmark, the larger the efficiency reference value.

[0088] The efficiency reference coefficient is a coefficient used to convert the difference between the execution time value and the efficiency benchmark value into an efficiency reference value. The efficiency reference coefficient is preset by the operator according to actual needs.

[0089] S404: Determine whether the detection reference value and efficiency reference value are both less than the preset reference baseline value. If yes, proceed to S405; if no, proceed to S406.

[0090] The reference benchmark value refers to a pre-set benchmark value used to determine whether the test reference value and efficiency reference value meet the standards. The reference benchmark value is obtained after being pre-input by the operator.

[0091] By judging whether the detection reference value and efficiency reference value are both less than the preset reference benchmark value, it can be determined whether re-inference is needed.

[0092] S405: The execution detection parameters are used as the decision output.

[0093] When both the detection reference value and the efficiency reference value are less than the preset reference benchmark value, it means that no re-inference is needed at this time, so the execution detection parameters will be used as the decision output result.

[0094] S406: Combine the detection reference value and the efficiency reference value to determine the comprehensive reference value, and feed the comprehensive reference value back to the model corresponding to the model type for re-inference.

[0095] Among them, the comprehensive reference value refers to a quantitative indicator that comprehensively represents the overall degree of achievement of the quality and efficiency of the decision-making scheme.

[0096] When both the detection reference value and the efficiency reference value are less than the preset reference benchmark value, it indicates that re-inference is required. Therefore, the detection reference value and the efficiency reference value are weighted and calculated, and the calculation result is used as the comprehensive reference value. The comprehensive reference value is then fed back to the model corresponding to the model type for re-inference, thereby improving the accuracy of inference.

[0097] The weighted calculation coefficients are preset by the operator according to actual needs.

[0098] To further ensure the rationality of using the execution detection parameters as the decision output, it is necessary to perform further separate analysis and calculation before using the execution detection parameters as the decision output, which will be explained in detail through the steps shown below.

[0099] The following steps are included before the detection parameters are used as the decision output: S500: Retrieves detection fields based on execution detection parameters.

[0100] Among them, the detection fields refer to the specific attribute fields that correspond one-to-one with the execution detection parameters and characterize each dimension of the decision-making scheme execution. Execution detection parameters include the detection fields.

[0101] The detection fields are retrieved by executing the detection parameters, which facilitates subsequent use.

[0102] S501: Retrieve baseline fields and field baseline values ​​based on observation baseline parameters.

[0103] In this context, a baseline field refers to an attribute field that corresponds one-to-one with the observed baseline parameters and represents a dimension for evaluating the execution effect. A field baseline value refers to the numerical value corresponding to the baseline field. Observed baseline parameters include both baseline fields and field baseline values.

[0104] By observing the benchmark parameters, the benchmark field and its benchmark value can be retrieved for convenient subsequent use.

[0105] S502: Determine the field accuracy based on the detection field and the baseline field.

[0106] Field accuracy refers to the quantitative parameter that indicates a consistent match between the detected field and the benchmark field.

[0107] By retrieving the collective parameter values ​​corresponding to the detection field and the baseline field respectively, and analyzing whether they meet the requirements, if the requirements are met, the ratio between the two parameters is calculated as the individual accuracy; if the requirements are not met, the individual accuracy is directly assigned to 0. Then, the average value of the individual accuracy of each detection field is calculated as the field accuracy for convenient subsequent use.

[0108] S503: Determine the actual number of fields based on the detected fields.

[0109] The actual number of fields refers to the number of values ​​corresponding to the detected fields.

[0110] The number of items is counted by detecting the field, and the count result is used as the actual field quantity for convenient subsequent use.

[0111] S504: Calculate the ratio between the actual number of fields and the baseline number of fields and use it as the field completeness.

[0112] Field completeness refers to the ratio between the actual number of fields and the baseline number of fields.

[0113] The ratio between the actual number of fields and the baseline number of fields is calculated, and the calculation result is used as the field completeness for convenient subsequent use.

[0114] S505: Determine data quality reference values ​​by combining field accuracy and field completeness.

[0115] Among them, the data quality reference value refers to a quantitative parameter that reflects the overall degree of compliance of data quality after the implementation of the decision-making plan.

[0116] By weighting the accuracy and completeness of fields, and using the result as a data quality reference value, subsequent use is facilitated. The weighting coefficients are preset by the operator according to actual needs.

[0117] S506: Determine whether the data quality reference value is less than the preset reference baseline value. If yes, proceed to S507; if no, proceed to S508.

[0118] Specifically, the system determines whether re-inference is needed by judging whether the data quality reference value is less than the preset reference benchmark value.

[0119] S507: Use the execution detection parameters as the decision output.

[0120] When the data quality reference value is less than the preset reference benchmark value, it means that no re-inference is needed at this time, so the execution detection parameters will be used as the decision output result.

[0121] S508: Feed back the data quality reference value to the model corresponding to the model type for re-inference.

[0122] When the data quality reference value is not less than the preset reference benchmark value, it indicates that re-inference is required. Therefore, the data quality reference value is fed back to the model corresponding to the model type for re-inference, thereby improving the accuracy of inference.

[0123] To further ensure the rationality of the determined data quality reference values, further separate analysis and calculation are required after the data quality reference values ​​are determined. The specific steps are explained in detail below.

[0124] After determining the data quality reference values, the following steps are also included: S600: Determine the source method based on the execution detection parameters.

[0125] The source method refers to the way the detection parameters are collected and the channel through which the data is generated.

[0126] The source method is retrieved by executing the detection parameters, which facilitates subsequent use.

[0127] S601: Determine the number of sources based on the source method.

[0128] The number of sources refers to the number of ways of sourcing.

[0129] By counting the source methods and using the count result as the number of sources, it is convenient to use them later.

[0130] S602: Determine if there is only one source. If yes, proceed to S603; otherwise, proceed to S604.

[0131] Specifically, by determining whether there is only one source, it can be determined whether the data quality reference value needs to be adjusted.

[0132] S603: Continue outputting data quality reference values.

[0133] When there is only one source, it means that there is no need to adjust the data quality reference value, so the data quality reference value continues to be output.

[0134] S604: Retrieve source field based on source method.

[0135] The source field refers to the specific attribute field corresponding to the source method.

[0136] When there is more than one source, it indicates that the data quality reference value needs to be adjusted. Therefore, the source field is retrieved through the source method to facilitate subsequent retrieval.

[0137] S605: Determine the field consistency by combining the source field and the detection field.

[0138] Among them, field consistency refers to the quantitative parameter corresponding to the consistency between the source field and the detection field.

[0139] By comparing the source fields and the detection fields one by one, and calculating the ratio between the number of successful comparisons and the total number of detection fields, the ratio is used as the field consistency for convenient subsequent use.

[0140] S606: Update and adjust data quality reference values ​​based on field consistency.

[0141] This method improves the accuracy of data quality reference values ​​by weighting field consistency with field accuracy and field completeness, and then using the calculation results to update and replace the data quality reference values. The weighting coefficients are preset by the operator according to actual needs.

[0142] To further ensure the rationality of implementing governance adjustment directives, a more detailed separate analysis and calculation is required after the implementation of these directives, which will be explained in detail through the steps shown below.

[0143] The following steps are included after the governance adjustment directive is executed: S700: Decision paths for data asset management agents, quality diagnosis agents, and conversational analysis agents.

[0144] The decision path refers to the analysis path taken by each agent during the analysis. The decision path includes the following steps: input information, model reasoning, solution execution, parameter detection, multi-dimensional verification, and output of results that meet the criteria or re-reasoning if the criteria are not met.

[0145] The decision-making paths of the data asset management agent, quality diagnosis agent, and conversational analysis agent are retrieved and collected to facilitate subsequent use.

[0146] S701: Generate experience instances based on decision paths.

[0147] Among them, experience examples refer to actual cases that use historical governance situations as experience.

[0148] The decision path retrieves corresponding water demand processing data, governance strategy instructions, user operation instructions, governance adjustment instructions, and displayed information, and combines them to form a case study, which serves as an experience example for future use.

[0149] S702: Learn from experience instances and discover the optimal policy pattern based on graph neural network algorithms.

[0150] The optimal strategy pattern refers to the set of optimal standardized strategies selected.

[0151] By structurally representing experience instances as decision knowledge graphs, and then using graph neural networks (GCNs) to learn the general decision rules and association patterns contained in the graphs, the optimal policy pattern is automatically mined and verified from the trained model for convenient subsequent use.

[0152] S703: Update the preset governance knowledge graph based on the optimal strategy pattern.

[0153] In this context, the governance knowledge graph refers to the decision-making knowledge graph corresponding to different pre-stored experience instances. The governance knowledge graph is pre-defined by the operator.

[0154] By adjusting and replacing similar experience instances in the preset governance knowledge graph according to the optimal strategy pattern, the adjusted governance knowledge graph is output as a new governance knowledge graph, which facilitates its subsequent use.

[0155] To further ensure the rationality of updating the preset governance knowledge graph based on the optimal strategy pattern, it is necessary to perform a further separate analysis and calculation on updating the preset governance knowledge graph based on the optimal strategy pattern. The specific steps are explained in detail below.

[0156] Updating the pre-defined governance knowledge graph based on the optimal strategy model includes the following steps: S800: Input experience instances into a pre-defined governance knowledge graph to retrieve similar instances.

[0157] Similar instances refer to instances selected that are similar to experience instances in terms of decision-making logic, scenario characteristics, indicator range, and execution effect.

[0158] By inputting experience examples into a pre-defined governance knowledge graph, and matching them one by one according to decision-making logic, scenario characteristics, indicator range, and execution effect to obtain a matching degree, the matching degrees are then weighted to calculate a comprehensive similarity. Finally, the instance with the highest similarity is selected as the similar instance for subsequent use. The specific weights of the weighted calculation are preset by the operator according to actual needs.

[0159] S801: Retrieve similar strategies based on similar instances.

[0160] Among them, the similar strategy refers to the strategy corresponding to the governance adjustment instructions in similar instances.

[0161] By retrieving the corresponding strategy from similar instances and using it as a similar strategy, it is convenient to use it later.

[0162] S802: Conduct tests based on similarity strategies and collect data quality reference values ​​as similarity reference values.

[0163] The similarity reference value refers to the data quality reference value corresponding to testing according to a similar strategy.

[0164] By employing similar strategies for analysis and testing, and collecting similar reference values, it becomes easier to use them in the future.

[0165] S803: Retrieves the corresponding optimal initial policy based on the optimal policy pattern.

[0166] Among them, the optimal initial strategy refers to the strategy corresponding to the governance adjustment instructions under the optimal strategy mode.

[0167] The optimal initial strategy is retrieved from the pre-defined governance knowledge graph using the optimal strategy pattern, making it convenient for subsequent use.

[0168] S804: Test according to the optimal initial strategy and collect data quality reference values ​​as the optimal reference values.

[0169] The optimal reference value refers to the data quality reference value corresponding to the test conducted according to the optimal initial strategy.

[0170] By employing the optimal initial strategy for analysis and testing, and collecting the optimal reference value, it is convenient for subsequent use.

[0171] S805: Determine the relative selection strategy by combining similar reference values ​​and the optimal reference value.

[0172] Among them, relative selection strategy refers to the strategy corresponding to the selected strategy.

[0173] By combining and analyzing similar reference values ​​with optimal reference values, a relative selection strategy is obtained, which facilitates subsequent use.

[0174] S806: Determine the relative strategy pattern based on the relative selection strategy, and replace the optimal strategy pattern with the relative strategy pattern.

[0175] Among them, the relative strategy pattern refers to the standardized set of strategies corresponding to the relative selection strategy.

[0176] By using the standardized set of policies corresponding to the relative selection policy as the relative policy pattern, and then replacing the optimal policy pattern with the relative policy pattern, the accuracy of the obtained optimal policy pattern is improved.

[0177] To further ensure the rationality of the relative selection strategy, it is necessary to perform a more detailed separate analysis and calculation of the relative selection strategy, which will be explained in detail through the steps shown below.

[0178] The method for determining the relative selection strategy includes the following steps: S900: Determine whether the similar reference value is consistent with the optimal reference value. If yes, proceed to S901; if no, proceed to S904.

[0179] In this process, the consistency between similar reference values ​​and the optimal reference value is used to determine whether further analysis is needed.

[0180] S901: Collect similar manual intervention rates based on similar strategies.

[0181] Among them, the similarity rate of human intervention refers to a quantitative parameter of the degree of intervention by users under similar strategies.

[0182] When the similar reference value is consistent with the optimal reference value, it indicates that further analysis is needed. Therefore, we retrieve the number of user operation command inputs under the similar strategy and use it as the number of manual inputs. Then, we retrieve the number of governance strategy command generation under the similar strategy and use it as the number of automatic governances. We then calculate the sum between the number of manual inputs and the number of automatic governances and use it as the total number of governances. Finally, we calculate the ratio between the number of manual inputs and the total number of governances and use it as the similar manual intervention rate for subsequent use.

[0183] S902: Collect the initial human intervention rate based on the optimal initial strategy.

[0184] The initial human intervention rate refers to a quantitative parameter of the degree of user intervention under the optimal initial strategy.

[0185] By collecting and retrieving the number of manual inputs and automatic governance times corresponding to the optimal initial strategy, the initial manual intervention rate is calculated for convenient subsequent use.

[0186] S903: Based on the comparison results between the initial human intervention rate and the similar human intervention rate, determine the human intervention selection strategy and use the human intervention selection strategy as the relative selection strategy.

[0187] Among them, the strategy for selecting manual intervention refers to the strategy selected based on the situation of manual intervention.

[0188] By comparing the initial human intervention rate with the similar human intervention rate, when the initial human intervention rate is greater than the similar human intervention rate, the similar strategy is selected as the human intervention selection strategy. When the initial human intervention rate is not greater than the similar human intervention rate, the optimal initial strategy is selected as the human intervention selection strategy. The human intervention selection strategy is then used as the relative selection strategy, thereby improving the accuracy of the obtained relative selection strategy.

[0189] S904: Based on the comparison results between similar reference values ​​and the optimal reference value, determine the reference selection strategy and use the reference selection strategy as the relative selection strategy.

[0190] Among them, the reference selection strategy refers to the strategy of selecting based on similar reference values.

[0191] When the similar reference value is inconsistent with the optimal reference value, it means that no further analysis is needed. Therefore, we judge the size between the similar reference value and the optimal reference value. When the similar reference value is less than the optimal reference value, we select the similar strategy as the reference selection strategy. When the similar reference value is not less than the optimal reference value, we select the optimal initial strategy as the reference selection strategy. Then, we use the reference selection strategy as the relative selection strategy to improve the accuracy of the obtained relative selection strategy.

[0192] Based on the same inventive concept, embodiments of the present invention provide a multi-agent collaborative management system for water data governance, comprising: The data acquisition module is used to collect user operation commands, agent type, input information, execution detection parameters, decision path, similar reference value, optimal reference value, similar human intervention rate, and initial human intervention rate. The memory stores a program for implementing a multi-agent collaborative management method for water data governance as described above. The processor loads and executes programs stored in memory.

[0193] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0194] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A multi-agent collaborative management method for water data governance, characterized in that, include: Construct intelligent agents for data asset management, quality diagnosis, and conversational analysis, and establish standardized decision-making processes; Build a structured message queue, which is used to share the input and output of the data asset management agent, the quality diagnosis agent, and the conversational analysis agent in real time; When water demand processing data is received, the data asset management intelligent agent analyzes the water demand processing data to generate water management data and sends it to the message queue; When the quality diagnostic agent listens to water management data from the message queue, it retrieves the water management data, analyzes it, generates governance strategy instructions, and sends them to the message queue. By combining conversational analytical agents with governance policy instructions, governance adjustment instructions are determined and executed.

2. The multi-agent collaborative management method for water data governance according to claim 1, characterized in that, The methods for generating governance adjustment instructions include: Determine whether user operation commands have been collected; If so, the conversational analysis agent retrieves governance policy instructions from the message queue and combines them with user operation instructions to generate user suggestion instructions and display information. The user suggestion instructions are then used as governance adjustment instructions, and the display information is output for presentation. If not, then the governance strategy directive will be treated as a governance adjustment directive.

3. The multi-agent collaborative management method for water data governance according to claim 2, characterized in that, Methods for establishing standardized decision-making processes include: Collect intelligent agent types and input information, including water demand processing data, governance strategy instructions, and user operation instructions; Determine the perception type, model type, action type, and observation baseline parameters based on the agent type; Input information is standardized based on the type of perception to form perception data; Based on the model type, the corresponding model is retrieved, and reasoning is performed on the perceived data to obtain the reasoning result; The system combines action type and reasoning results to generate and execute a decision plan, and collects execution detection parameters. The decision output results are determined and output by combining the execution detection parameters and the observation benchmark parameters. The decision output results include water management data, governance strategy instructions and display information.

4. The multi-agent collaborative management method for water data governance according to claim 3, characterized in that, Methods for determining decision output include: Retrieve execution detection results and execution time based on execution detection parameters; Retrieve observation baseline specifications and efficiency baseline values ​​based on observation baseline parameters; The test reference values ​​were determined by combining the test results with the observation benchmark specifications; An efficiency reference value is determined by combining the execution time value with the efficiency benchmark value; Determine whether the detection reference value and efficiency reference value are both less than the preset reference benchmark value; If so, the detection parameters will be used as the decision output. If not, then a comprehensive reference value is determined by combining the detection reference value and the efficiency reference value, and the comprehensive reference value is fed back to the model corresponding to the model type for re-inference.

5. A multi-agent collaborative management method for water data governance according to claim 4, characterized in that, Before using the detection parameters as the decision output, the following steps are also included: Retrieve detection fields based on execution detection parameters; Retrieve baseline fields and field baseline values ​​based on observed baseline parameters; Determine the accuracy of the fields based on the detection fields and the benchmark fields; The actual number of fields is determined based on the fields being detected; Calculate the ratio between the actual number of fields and the baseline number of fields, and use this ratio as the field completeness. Determine data quality reference values ​​by combining field accuracy and field completeness; Determine whether the data quality reference value is less than the preset reference baseline value; If so, the detection parameters will be used as the decision output. If not, the data quality reference value will be fed back to the model corresponding to the model type for re-inference.

6. A multi-agent collaborative management method for water data governance according to claim 5, characterized in that, After determining the data quality reference values, the following also includes: The source method is determined based on the execution detection parameters; The number of sources is determined based on the source method; Determine if the number of sources is only one; If yes, continue outputting data quality reference values; If not, retrieve the source field based on the source method; The consistency of fields is determined by combining the source field and the detection field; The data quality reference values ​​are updated and adjusted based on field consistency.

7. A multi-agent collaborative management method for water data governance according to claim 5, characterized in that, Following the execution of the governance adjustment directive, the following also applies: Decision paths for data asset management agents, quality diagnosis agents, and conversational analysis agents; Generate experience examples based on the decision-making path; The graph neural network algorithm is used to learn from experience instances and discover the optimal policy pattern. The pre-defined governance knowledge graph is updated based on the optimal strategy model.

8. A multi-agent collaborative management method for water data governance according to claim 7, characterized in that, Updating the pre-defined governance knowledge graph based on the optimal strategy model includes: Inputting empirical examples into a pre-defined governance knowledge graph to retrieve similar examples; Retrieve similar strategies based on similar instances; Test according to similar strategies and collect data quality reference values ​​as similar reference values; The optimal initial strategy is retrieved based on the optimal strategy pattern; Test according to the optimal initial strategy and collect data quality reference values ​​as the optimal reference values; A relative selection strategy is determined by combining similar reference values ​​and the optimal reference value; The relative strategy pattern is determined based on the relative selection strategy, and the relative strategy pattern replaces the optimal strategy pattern.

9. A multi-agent collaborative management method for water data governance according to claim 8, characterized in that, Methods for determining relative selection strategies include: Determine whether the similar reference value is consistent with the optimal reference value; If so, then collect similar human intervention rates based on similar strategies; The initial human intervention rate is collected based on the optimal initial strategy; Based on the comparison results between the initial human intervention rate and the similar human intervention rate, the human intervention selection strategy is determined and used as the relative selection strategy. If not, then based on the comparison results between similar reference values ​​and the optimal reference value, a reference selection strategy is determined, and the reference selection strategy is used as the relative selection strategy.

10. A multi-agent collaborative management system for water resources data governance, characterized in that, include: The data acquisition module is used to collect user operation commands, agent type, input information, execution detection parameters, decision path, similar reference value, optimal reference value, similar human intervention rate, and initial human intervention rate. The memory stores a program for implementing a multi-agent collaborative management method for water data governance as described in any one of claims 1 to 9; The processor loads and executes programs stored in memory.

Images