Multi-agent data governance method and system based on negative feedback optimization

Abstract

The embodiment of the invention provides a multi-agent data governance method and system based on negative feedback optimization, and the method comprises the steps: carrying out the data governance of to-be-governed data based on a multi-agent system, obtaining a data governance result, and creating a task execution path map in a data governance process in real time, based on a preset evaluation index set and a dynamic threshold set, performing evaluation processing on the data governance result to obtain negative feedback data, querying a task execution path map according to the negative feedback data to obtain a task processing path, and performing evaluation processing on the task processing path to obtain one or more to-be-optimized agents, and performing matching processing based on the negative feedback data, the task type of the to-be-optimized agent and the optimization parameter matching library to obtain corresponding optimization data, performing optimization of the corresponding to-be-optimized agent according to the optimization data, and performing data management of the to-be-managed data again based on the optimized multi-agent system. According to the scheme, the intelligent agent can be optimized and iterated based on a negative feedback learning mechanism.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of electronic digital data processing, and in particular to a multi-agent data governance method and system based on negative feedback optimization. Background Art

[0002] With the deepening development of the big data era, data governance has become the core foundation of enterprise digital transformation and intelligent applications. Traditional data governance technologies generally use single algorithms or manual rule-driven workflow models, which are difficult to deal with complex quality defects (such as data redundancy, format errors, and semantic conflicts) in massive amounts of heterogeneous data. In particular, their adaptability and scalability face significant bottlenecks when faced with dynamically changing governance needs. In recent years, the combination of multi-agent systems and reinforcement learning has provided a new technical path for data governance. Through the collaborative decision-making of distributed agents, the system can handle multiple types of data governance tasks (such as data cleaning, classification, standardization, and quality assessment) in parallel, while reinforcement learning mechanisms empower agents to dynamically optimize strategies in an interactive environment.

[0003] Most related technologies rely on positive reward signals to drive optimization, and lack systematic mining and utilization of negative feedback data generated in the data governance process (such as error processing results and conflict operation records), resulting in relatively low data governance processing efficiency. Summary of the Invention The embodiments of the present application provide a multi-agent data governance method and system based on negative feedback optimization, which solves the problem in the prior art of lack of systematic mining and utilization of negative feedback data generated in the data governance process, resulting in relatively low data governance processing efficiency. It can optimize and iterate the agents based on the negative feedback learning mechanism, and perform data governance in an efficient and better manner.

[0004] In a first aspect, an embodiment of the present application provides a multi-agent data governance method based on negative feedback optimization, comprising: Performing data governance on the data to be governed based on a multi-agent system to obtain corresponding data governance results, and creating a task execution path map corresponding to the data governance process in real time, wherein the multi-agent system includes multiple agents that can independently execute data governance tasks; Evaluate the data governance results based on a preset evaluation indicator set and a dynamic threshold set to obtain negative feedback data, and the dynamic threshold set is updated every other evaluation process; According to the negative feedback data, the task execution path map is queried to obtain a corresponding task processing path, the task processing path is evaluated to obtain one or more agents to be optimized, and a matching process is performed based on the negative feedback data, the task type of the agent to be optimized, and an optimization parameter matching library to obtain corresponding optimization data; The corresponding intelligent agent to be optimized is optimized according to the optimization data, and data governance of the data to be governed is re-performed based on the optimized multi-agent system.

[0005] Optionally, the real-time creation of a task execution path map corresponding to the data governance process includes: Record the data flow and data processing volume of each agent executed in the data governance process in real time, generate a dedicated task node for each agent, and generate directed edges between each dedicated task node according to the data flow of each agent to obtain a graph of the task execution path to be adjusted; Each exclusive task node in the task execution path map to be adjusted is adjusted according to the data processing amount to obtain a task execution path map.

[0006] Optionally, adjusting each dedicated task node in the to-be-adjusted task execution path map according to the data processing amount to obtain the task execution path map includes: According to the data processing amount of each intelligent agent, an evaluation weight is allocated to the corresponding exclusive task node, and each evaluation weight is marked at a preset position of the corresponding exclusive task node to obtain a task execution path map.

[0007] Optionally, the preset evaluation indicator set includes an integrity indicator, a consistency indicator, and a timeliness indicator. The negative feedback data obtained by evaluating the data governance results based on the preset evaluation indicator set and the dynamic threshold set includes: Conduct integrity assessment, consistency assessment, and timeliness assessment on each item of result data in the data governance results to obtain corresponding integrity assessment values, consistency assessment values, and timeliness assessment values; The integrity assessment value, the consistency assessment value and the timeliness assessment value are respectively compared with the corresponding dynamic thresholds in the dynamic threshold set, and the result data with an assessment value less than the corresponding dynamic threshold is determined as negative feedback data.

[0008] Optionally, the negative feedback data includes error data and error type, the task execution path graph includes each directed-connected task node and a corresponding evaluation weight, querying the task execution path graph according to the negative feedback data to obtain a corresponding task processing path, and evaluating the task processing path to obtain one or more agents to be optimized, including: Querying the task node that outputs the error data in the task execution path map, and performing path backtracking in the task execution path map according to the task node that outputs the error data to obtain a task processing path; Determine a first task node in the task processing path whose node task type is associated with the error type, and screen out a second task node in the task processing path whose evaluation weight is greater than a preset weight, and determine the first task node and the second task node as nodes to be evaluated respectively; An evaluation process is performed based on the task execution logs and resource consumption monitoring information associated with each of the nodes to be evaluated to obtain one or more intelligent agents to be optimized.

[0009] Optionally, the step of performing evaluation processing based on the task execution logs and resource consumption monitoring information associated with each of the nodes to be evaluated to obtain one or more agents to be optimized includes: Performing anomaly detection on each record information in the task execution log to obtain a log detection result of each node to be evaluated; Determine corresponding resource consumption threshold information according to the node task type of each node to be evaluated, and compare the resource consumption monitoring information with the corresponding resource consumption threshold information to obtain a resource consumption detection result of each node to be evaluated; The intelligent agent corresponding to the node to be evaluated whose log detection result or resource consumption detection result is abnormal is determined as the intelligent agent to be optimized.

[0010] Optionally, the negative feedback data includes an error type of the error data, and the matching process based on the negative feedback data, the task type of the agent to be optimized, and the optimization parameter matching library to obtain the corresponding optimization data includes: Determine the corresponding matching optimization parameter type in the optimization parameter matching library according to the error type and the task type; In the case where there are multiple agents to be optimized, the task types of the multiple agents to be optimized are combined to obtain a task type combination, and the corresponding error type influence distribution information is queried according to the task type combination to obtain the influence of each task type on the error type; The parameter value of the optimization parameter type corresponding to the intelligent agent to be optimized is determined according to the influence degree, and each of the optimization parameter types and the parameter value is determined as the optimization data corresponding to the intelligent agent to be optimized.

[0011] In a second aspect, an embodiment of the present application further provides a multi-agent data governance system based on negative feedback optimization, comprising: A data governance module, configured to perform data governance on the data to be governed based on a multi-agent system and obtain corresponding data governance results, wherein the multi-agent system includes multiple agents that can independently perform data governance tasks; A graph creation module, used to create a task execution path graph corresponding to the data governance process in real time; An evaluation processing module, configured to evaluate the data governance results based on a preset evaluation indicator set and a dynamic threshold set to obtain negative feedback data, wherein the dynamic threshold set is updated every other evaluation process; A query module, configured to query the task execution path map according to the negative feedback data to obtain a corresponding task processing path; The evaluation processing module is further configured to evaluate the task processing path to obtain one or more intelligent agents to be optimized; A matching processing module, configured to perform matching processing based on the negative feedback data, the task type of the agent to be optimized, and an optimization parameter matching library to obtain corresponding optimization data; An optimization module, configured to optimize the corresponding intelligent agent to be optimized according to the optimization data; The data governance module is also used to re-perform data governance of the data to be governed based on the optimized multi-agent system.

[0012] In a third aspect, an embodiment of the present application further provides a multi-agent data governance device based on negative feedback optimization, the device comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the multi-agent data governance method based on negative feedback optimization described in the embodiment of the present application.

[0013] In a fourth aspect, an embodiment of the present application further provides a storage medium for storing computer-executable instructions, which, when executed by a computer processor, are used to execute the multi-agent data governance method based on negative feedback optimization described in an embodiment of the present application.

[0014] In an embodiment of the present application, data governance is performed on the data to be governed based on a multi-agent system to obtain corresponding data governance results, and a task execution path map corresponding to the data governance process is created in real time. The multi-agent system includes multiple agents that can independently execute data governance tasks. The data governance results are evaluated and processed based on a preset evaluation index set and a dynamic threshold set to obtain negative feedback data. The dynamic threshold set is updated every evaluation process. The task execution path map is queried based on the negative feedback data to obtain the corresponding task processing path. The task processing path is evaluated and processed to obtain one or more agents to be optimized. Based on the negative feedback data, the task type of the agent to be optimized and the optimization parameter matching library, a matching process is performed to obtain corresponding optimization data. The corresponding agent to be optimized is optimized according to the optimization data, and data governance of the data to be governed is re-performed based on the optimized multi-agent system. This solution solves the problem of the lack of systematic mining and utilization of negative feedback data generated in the data governance process in the prior art, which leads to relatively low data governance processing efficiency, by determining the negative feedback data in the data governance results and optimizing the multi-agent system based on the negative feedback data. It can optimize and iterate the agents based on the negative feedback learning mechanism to perform data governance in an efficient and better manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A flowchart of a multi-agent data governance method based on negative feedback optimization provided in an embodiment of the present application; Figure 2 A flowchart of a multi-agent data governance method based on negative feedback optimization, including creating a task execution path map, provided in an embodiment of the present application; Figure 3 A schematic diagram of a task execution path map to be adjusted provided in an embodiment of the present application; Figure 4 A schematic diagram of a task execution path map provided in an embodiment of the present application; Figure 5 A flowchart of a multi-agent data governance method based on negative feedback optimization for determining negative feedback data provided in an embodiment of the present application; Figure 6 A flowchart of a multi-agent data governance method based on negative feedback optimization for determining an agent to be optimized provided in an embodiment of the present application; Figure 7 A flowchart of a multi-agent data management method based on negative feedback optimization for determining optimization data of an agent to be optimized provided in an embodiment of the present application; Figure 8 A block diagram of the module structure of a multi-agent data governance system based on negative feedback optimization provided in an embodiment of the present application; Figure 9 A structural diagram of a multi-agent data governance device based on negative feedback optimization provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] The following is a further detailed description of the embodiments of the present application in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the embodiments of the present application, and are not intended to limit the embodiments of the present application. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions of the embodiments of the present application, rather than all structures.

[0017] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0018] The embodiment of this application provides a multi-agent data governance method based on negative feedback optimization, which can be applied to multi-agent data governance scenarios such as bank credit assessment, intelligent traffic scheduling, and enterprise-level data governance. The embodiment of this application provides a multi-agent data governance method based on negative feedback optimization, in which the execution entity of each step is the server.

[0019] Figure 1 A flowchart of a multi-agent data governance method based on negative feedback optimization provided in an embodiment of the present application is as follows: Figure 1 As shown, specifically including: Step S101: Perform data governance on the data to be governed based on the multi-agent system to obtain corresponding data governance results, and create a task execution path map corresponding to the data governance process in real time, wherein the multi-agent system includes multiple agents that can independently execute data governance tasks.

[0020] Among them, the multi-agent system is used to represent a system composed of multiple agents that can independently perform data governance tasks. These agents can complete tasks together through communication and coordinated cooperation. The multi-agent system can be used to perform data governance on the data to be governed to obtain data governance results. The data to be governed refers to the data currently waiting for data governance. In the process of data governance of the data to be governed, a corresponding task execution path map is created in real time. The task execution path map can be a map used to show the dependency relationship between the agents in the data governance process. The task execution path map can be composed of nodes and edges. In one embodiment, a way to create a task execution path map can be to record the data flow of each agent executed in the process of data governance in real time, and generate a dedicated task node for each agent, and generate a directed edge between each dedicated task node according to the data flow of each agent to obtain a task execution map path. By creating a task execution path map corresponding to the data governance process, the data governance process can be displayed more intuitively.

[0021] Step S102: Evaluate the data governance results based on the preset evaluation indicator set and the dynamic threshold set to obtain negative feedback data, wherein the dynamic threshold set is updated every other evaluation process.

[0022] Among them, the preset evaluation indicator set can be a set of various indicators pre-set for evaluating data governance results. The dynamic threshold set is a set of standard dynamic thresholds corresponding to each evaluation indicator. The dynamic threshold set is updated every other evaluation process. The data governance results can be evaluated and processed by the preset evaluation indicator set and the dynamic threshold set to obtain negative feedback data. Negative feedback data is used to characterize the data that does not meet the standard conditions identified by evaluating the data governance results, such as error data, error type and other data. In one embodiment, the preset evaluation indicator set includes multiple evaluation indicators. A method for evaluating and processing data governance results can be to evaluate each result data in the data governance results according to the multiple evaluation indicators to obtain corresponding multiple evaluation values, compare each evaluation value with the corresponding dynamic threshold in the dynamic threshold set, and determine the result data with a preset number or more of the evaluation values ​​less than the corresponding dynamic threshold as negative feedback data. By evaluating and processing the data governance results through the evaluation indicator set and the dynamic threshold set, negative feedback data can be determined from multiple dimensions to improve its rationality and accuracy.

[0023] In one embodiment, a dynamic threshold set update method can be, after determining the negative feedback data, calculating the average evaluation value of each evaluation indicator of the non-negative feedback data in the multiple result data, comparing the average evaluation value of each evaluation indicator with the corresponding historical average evaluation value within a preset time, and when the difference between the average evaluation value and the corresponding historical average evaluation value is greater than a preset value, calculating the average of the average evaluation value and the corresponding historical average evaluation value, querying the corresponding update threshold based on the average value, and updating the corresponding dynamic threshold. By continuously updating the dynamic threshold, the accuracy of the evaluation process can be improved.

[0024] Step S103: query the task execution path map according to the negative feedback data to obtain the corresponding task processing path, evaluate the task processing path to obtain one or more intelligent agents to be optimized, and perform matching processing based on the negative feedback data, the task type of the intelligent agent to be optimized, and the optimization parameter matching library to obtain the corresponding optimization data.

[0025] Among them, the corresponding task processing path can be obtained by querying the task execution path map through the negative feedback data. The task processing path refers to the path passed when executing the subtask corresponding to the negative feedback data. One or more intelligent agents to be optimized can be obtained by evaluating the task processing path. The intelligent agent to be optimized can refer to an intelligent agent to be optimized with defects or problems. The task type refers to the type of specified task assigned to the intelligent agent, such as data collection, data cleaning, data integration and other task types. The corresponding optimization data can be obtained by matching the negative feedback data, the task type of the intelligent agent to be optimized and the optimization parameter matching library. The optimization parameter matching library can be a database that stores the corresponding optimization parameters under different situations. The optimization parameters can be the parameter types for the intelligent agent to be optimized, such as the learning rate, weight and other parameter types. Optimization data refers to the data used when the intelligent agent to be optimized is optimized. In one embodiment, the task execution path graph includes various directed connected task nodes. A method for determining the intelligent agent to be optimized may be to query the task node that outputs negative feedback data in the task execution path graph, perform path backtracing in the task execution path graph based on the task node that outputs negative feedback data to obtain the task processing path, perform error detection processing on the task execution log of each task node in the task processing path to obtain the detection result, and determine the intelligent agent corresponding to the task node with an abnormal detection result as the intelligent agent to be optimized.

[0026] In one embodiment, the negative feedback data includes the error type and error degree of the error data. One matching processing method may be to determine the corresponding matching optimization parameter type in the optimization parameter matching library according to the error type and task type, determine the preset numerical range in which the error degree falls, query the optimization parameter value of the optimization parameter type associated with the preset numerical range, and determine the optimization parameter type and optimization parameter value as the optimization data of the intelligent agent to be optimized.

[0027] Step S104: Optimize the corresponding agent to be optimized according to the optimization data, and re-manage the data to be managed based on the optimized multi-agent system.

[0028] After obtaining the optimized data of the agent to be optimized through matching, the optimized data can be used to optimize the corresponding agent to be optimized, and the optimized multi-agent system can re-govern the data to be governed until no negative feedback data is included in the data governance results. For example, if the optimized data of the agent to be optimized is to adjust the emotion threshold to 0.4, the current emotion threshold of the agent to be optimized is adjusted to 0.4, and the optimized multi-agent system can re-govern the data to be governed until no negative feedback data is included in the data governance results.

[0029] From the above, it can be seen that by performing data governance on the data to be governed based on a multi-agent system, the corresponding data governance results are obtained, and a task execution path map corresponding to the data governance process is created in real time. The multi-agent system includes multiple agents that can independently execute data governance tasks. The data governance results are evaluated and processed based on a preset evaluation index set and a dynamic threshold set to obtain negative feedback data. The dynamic threshold set is updated every time the evaluation process is performed. The task execution path map is queried based on the negative feedback data to obtain the corresponding task processing path. The task processing path is evaluated and processed to obtain one or more agents to be optimized. Based on the negative feedback data, the task type of the agent to be optimized and the optimization parameter matching library, the corresponding optimization data is obtained. The corresponding agent to be optimized is optimized according to the optimization data, and the data governance of the data to be governed is re-performed based on the optimized multi-agent system. This solution solves the problem of the lack of systematic mining and utilization of negative feedback data generated in the data governance process in the existing technology, which leads to relatively low data governance processing efficiency, by determining the negative feedback data in the data governance results and optimizing the multi-agent system based on the negative feedback data. It can optimize and iterate the agents based on the negative feedback learning mechanism to perform data governance in an efficient and better way.

[0030] Figure 2 A flowchart of a multi-agent data governance method based on negative feedback optimization including creating a task execution path map is provided in an embodiment of the present application, such as Figure 2As shown, specifically including: Step S201: Perform data governance on the data to be governed based on a multi-agent system to obtain corresponding data governance results, wherein the multi-agent system includes multiple agents that can independently execute data governance tasks.

[0031] Step S202: Record the data flow and data processing volume of each intelligent agent executed in the process of data governance in real time, and generate exclusive task nodes for each intelligent agent. Generate directed edges between each exclusive task node according to the data flow of each intelligent agent to obtain the execution path map of the task to be adjusted.

[0032] Among them, in the process of data governance for the data to be governed, the data flow and data processing volume of each agent being executed are recorded in real time, and an exclusive task node for each agent is generated. The data flow direction can be the flow direction of the output data of the agent. For example, the output data of agent A is input to agent B as input data, then the data flow direction of agent A is agent B. The data processing volume refers to the amount of data processed by the agent when executing a task. The calculation formula for the data processing volume can be (basic data volume * data type coefficient) + model reasoning calculation volume + tool interaction data volume + collaborative communication data volume‌. The data flow direction can be used to generate directed edges between each exclusive task node to obtain a graph of the execution path of the task to be adjusted. For example, the agents executed in the data governance process are agent AG, and the data flow direction of agent A is obtained through recording to be agent B and agent C, the data flow direction of agent 2 is agent D, the data flow direction of agent C is agent E and agent F, and the data flow direction of agent E is agent G. The exclusive task nodes of each agent are connected through the corresponding data flow direction to obtain a graph of the execution path of the task to be adjusted, such as Figure 3 As shown, Figure 3 A schematic diagram of a task execution path map to be adjusted provided in an embodiment of the present application, AG corresponds to the exclusive task node of the intelligent agent AG, and each edge with an arrow (directed edge) represents the data flow direction of the intelligent agent.

[0033] Step S203: Adjust each dedicated task node in the task execution path map to be adjusted according to the data processing amount to obtain a task execution path map.

[0034] After generating the task execution path map to be adjusted, the data processing capacity of each agent executed during the data governance process can be used to adjust each dedicated task node in the task execution path map to be adjusted to obtain a task execution path map. Optionally, one adjustment method can be to assign an evaluation weight to the corresponding dedicated task node based on the data processing capacity of each agent, and mark each evaluation weight at a preset position of the corresponding dedicated task node to obtain a task execution path map, wherein the evaluation weight is used to represent the importance of the agent corresponding to the task node requiring evaluation processing. For example, the various intelligent agents executed in the data governance process are intelligent agent AG, and the data processing volume of intelligent agent AG is 80GB, 50GB, 75GB, 60GB, 40GB, 70GB, and 55GB, respectively. The larger the data processing volume, the larger the corresponding evaluation weight. Evaluation weights are assigned to corresponding exclusive task nodes according to the size of the data processing volume of each intelligent agent, and the evaluation weights of the exclusive task nodes of intelligent agent AG are 0.8, 0.5, 0.75, 0.6, 0.4, 0.7, and 0.55, respectively. Each evaluation weight is marked at the preset position of the corresponding exclusive task node to obtain a task execution path map, such as Figure 4 As shown, Figure 4 This is a schematic diagram of a task execution path map provided by an embodiment of the present application. 0.8 is the evaluation weight of task node A, 0.5 is the evaluation weight of task node B, 0.75 is the evaluation weight of task node C, 0.6 is the evaluation weight of task node D, 0.4 is the evaluation weight of task node E, 0.7 is the evaluation weight of task node F, and 0.55 is the evaluation weight of task node G.

[0035] In another embodiment, one adjustment method may be to determine the evaluation level of the corresponding exclusive task node based on the data processing volume of each intelligent agent, and adjust the color of the exclusive task node to the associated color of the corresponding evaluation level to obtain the task execution path map.

[0036] Step S204: Evaluate the data governance results based on the preset evaluation indicator set and the dynamic threshold set to obtain negative feedback data, wherein the dynamic threshold set is updated every other evaluation process.

[0037] Step S205: query the task execution path map based on the negative feedback data to obtain the corresponding task processing path, evaluate the task processing path to obtain one or more intelligent agents to be optimized, and perform matching processing based on the negative feedback data, the task type of the intelligent agent to be optimized, and the optimization parameter matching library to obtain the corresponding optimization data.

[0038] Step S206: Optimize the corresponding agent to be optimized according to the optimization data, and re-perform data management of the data to be managed based on the optimized multi-agent system.

[0039] As can be seen from the above, by recording the data flow and data processing volume of each agent executed during the data governance process in real time and generating dedicated task nodes for each agent, a graph of the task execution paths to be adjusted is generated based on the directed edges between each dedicated task node according to the data flow of each agent. Each dedicated task node in the graph of the task execution paths to be adjusted is then adjusted based on the data processing volume to obtain a task execution path graph. This solution generates a task execution path graph by using the data flow and data processing volume of each agent executed during the data governance process, which can more intuitively display the data governance process and the evaluation importance of each agent.

[0040] Figure 5 A flowchart of a multi-agent data governance method based on negative feedback optimization for determining negative feedback data provided in an embodiment of the present application is shown as follows: Figure 5 As shown, specifically including: Step S301: Perform data governance on the data to be governed based on the multi-agent system to obtain corresponding data governance results, and create a task execution path map corresponding to the data governance process in real time, wherein the multi-agent system includes multiple agents that can independently execute data governance tasks.

[0041] Step S302: Perform integrity assessment, consistency assessment, and timeliness assessment on each item of result data in the data governance results to obtain corresponding integrity assessment values, consistency assessment values, and timeliness assessment values.

[0042] Among them, the preset evaluation index set includes integrity index, consistency index and timeliness index. Through the integrity index, consistency index and timeliness index, each result data in the data governance result can be evaluated to obtain the corresponding integrity evaluation value, consistency evaluation value and timeliness evaluation value. The integrity evaluation value is used to characterize the completeness of the result data, and its evaluation method can be (actual data volume / expected data volume)*100%. The consistency evaluation value can be the degree of consistency between the result data and the expected result. The timeliness evaluation value can be the degree of proximity between the actual generation time of the data and the expected time. For example, the result data in the data governance result are result data 1, result data 2 and result data 3, and the integrity evaluation, consistency evaluation and timeliness evaluation are performed on result data 1, result data 2 and result data 3 respectively. The integrity evaluation value of result data 1 is 98, the consistency evaluation value is 96, and the timeliness evaluation value is 100. The integrity evaluation value of result data 2 is 99, the consistency evaluation value is 97, and the timeliness evaluation value is 95. The integrity evaluation value of result data 3 is 100, the consistency evaluation value is 100, and the timeliness evaluation value is 75.

[0043] Step S303: Compare the integrity assessment value, consistency assessment value, and timeliness assessment value with the corresponding dynamic thresholds in the dynamic threshold set respectively, and determine the result data with an assessment value less than the corresponding dynamic threshold as negative feedback data, wherein the dynamic threshold set is updated after each assessment process.

[0044] Among them, each dynamic threshold value in the dynamic threshold set is the standard value corresponding to the integrity assessment value, consistency assessment value and timeliness assessment value. Negative feedback data can be determined by comparing the dynamic threshold values ​​in the dynamic threshold set with the integrity assessment value, consistency assessment value and timeliness assessment value of each result data. For example, the integrity threshold value in the dynamic threshold set is 95, the consistency threshold value is 95, and the timeliness threshold value is 90. The integrity assessment value of result data 1 in the data governance result is 98, the consistency assessment value is 96, and the timeliness assessment value is 100. The integrity assessment value of result data 2 is 99, the consistency assessment value is 97, and the timeliness assessment value is 95. The integrity assessment value of result data 3 is 100, the consistency assessment value is 100, and the timeliness assessment value is 75. Among them, if the timeliness assessment value of result data 3 is less than the timeliness threshold, result data 3 is determined as negative feedback data.

[0045] Step S304: query the task execution path map based on the negative feedback data to obtain the corresponding task processing path, evaluate the task processing path to obtain one or more intelligent agents to be optimized, and perform matching processing based on the negative feedback data, the task type of the intelligent agent to be optimized, and the optimization parameter matching library to obtain the corresponding optimization data.

[0046] Step S305: Optimize the corresponding agent to be optimized according to the optimization data, and re-manage the data to be managed based on the optimized multi-agent system.

[0047] As can be seen from the above, by performing integrity assessment, consistency assessment, and timeliness assessment on each result data in the data governance results, the corresponding integrity assessment value, consistency assessment value, and timeliness assessment value are obtained. The integrity assessment value, consistency assessment value, and timeliness assessment value are respectively compared with the corresponding dynamic threshold value in the dynamic threshold set, and the result data with an assessment value less than the corresponding dynamic threshold value is determined as negative feedback data. This solution evaluates and processes data governance results through multiple evaluation indicators and corresponding dynamic thresholds. It can determine negative feedback data from multiple dimensions and improve its rationality and accuracy.

[0048] Figure 6 A flowchart of a multi-agent data governance method based on negative feedback optimization for determining an agent to be optimized is provided in an embodiment of the present application, such as Figure 6 As shown, specifically including: Step S401: Perform data governance on the data to be governed based on the multi-agent system to obtain corresponding data governance results, and create a task execution path map corresponding to the data governance process in real time, wherein the multi-agent system includes multiple agents that can independently execute data governance tasks.

[0049] Step S402: Evaluate the data governance results based on the preset evaluation indicator set and the dynamic threshold set to obtain negative feedback data, wherein the dynamic threshold set is updated every other evaluation process.

[0050] Step S403: query the task execution path map for the task node that outputs the error data, and perform path backtracking in the task execution path map based on the task node that outputs the error data to obtain the task processing path.

[0051] Among them, negative feedback data includes error data and error types. Error data is the erroneous data in the data governance results, and error types are the error classification of erroneous data, such as integrity errors, timeliness errors and other error types. The task execution path graph includes each directed connected task node and the corresponding evaluation weight. The evaluation weight can be the importance of the intelligent agent corresponding to the task node that needs to be evaluated. By querying the task execution path graph, the task node that outputs the error data can be obtained. By tracing back the path in the task execution path graph through the task node that outputs the error data, the task processing path of the error data can be obtained. Exemplarily, the task node that outputs the error data is queried as task node G in the task execution path graph. The task execution path graph is as follows. Figure 4 As shown, the task processing path is obtained by backtracing the task node G in the task execution path graph as A→C→E→G.

[0052] Step S404: determine the first task node in the task processing path whose node task type is associated with the error type, and filter out the second task node in the task processing path whose evaluation weight is greater than the preset weight, and determine the first task node and the second task node as nodes to be evaluated respectively, and perform evaluation processing based on the task execution logs and resource consumption monitoring information associated with each node to be evaluated to obtain one or more intelligent entities to be optimized.

[0053] Among them, the node task type is used to characterize the classification of tasks performed by the agent corresponding to the task node. The node task type and evaluation weight of each task node can be used to determine the node to be evaluated. The task execution log and resource consumption monitoring information associated with each node to be evaluated can be used to perform evaluation processing to obtain one or more agents to be optimized. The task execution log refers to a detailed record file of the entire process when the agent performs a data governance task. Resource consumption monitoring information can refer to real-time monitoring and recording data on the usage of various computing resources during the agent's task execution, such as CPU utilization, memory consumption rate and other information. Optionally, a method for determining the agent to be optimized can be to perform anomaly detection on each record information in the task execution log to obtain the log detection results of each node to be evaluated, determine the corresponding resource consumption threshold information according to the node task type of each node to be evaluated, compare the resource consumption monitoring information with the corresponding resource consumption threshold information to obtain the resource consumption detection results of each node to be evaluated, and determine the agent corresponding to the node to be evaluated whose log detection result or resource consumption detection result is abnormal as the agent to be optimized. For example, the error type of the erroneous data is a consistency error type, the node task type associated with the consistency error type is data cleaning, the task node that performs the data cleaning task is task node D, the task processing path is A→B→C→D→E, the corresponding evaluation weights are 0.8, 0.75, 0.4, 0.55, and 0.3, respectively, the preset weight is 0.6, and the task nodes corresponding to the evaluation weights greater than the preset weights are task nodes A and task node B. Then task nodes A, B, and D are the nodes to be evaluated, the node task type of task node A is data cleaning, and the corresponding resource consumption is The threshold information is the CPU utilization threshold of 50%, and the memory consumption threshold of 40%. The anomaly detection result of the various record information of task node A is no anomaly. The resource consumption monitoring information of task node A is CPU utilization of 60%, and memory consumption of 45%. The CPU utilization and memory consumption of task node A are both greater than the corresponding thresholds, that is, the resource consumption detection result is abnormal. The intelligent agent corresponding to task node A is determined as the intelligent agent to be optimized. In the same way, the log detection results and resource consumption detection results of task node B and task node D are detected to be normal.

[0054] In another embodiment, a method for determining the intelligent agent to be optimized may be to perform anomaly detection on each record information in the task execution log to obtain the log detection result of each node to be evaluated, compare the resource consumption monitoring information of each node to be evaluated with the preset resource consumption threshold information to obtain the resource consumption detection result of each node to be evaluated, and determine the intelligent agent corresponding to the node to be evaluated whose log detection result or resource consumption detection result is abnormal as the intelligent agent to be optimized.

[0055] Step S405: Based on the negative feedback data, the task type of the agent to be optimized and the optimization parameter matching library, matching processing is performed to obtain the corresponding optimization data, the corresponding agent to be optimized is optimized according to the optimization data, and data governance of the data to be governed is re-performed based on the optimized multi-agent system.

[0056] As can be seen from the above, the task nodes that output error data in the task execution path map are queried, and the task processing path is obtained by backtracing the path in the task execution path map based on the task nodes that output error data. The first task node in the task processing path is determined to be associated with the node task type and the error type, and the second task node in the task processing path is screened out with an evaluation weight greater than a preset weight. The first task node and the second task node are respectively determined as nodes to be evaluated, and evaluation and processing are performed based on the task execution logs and resource consumption monitoring information associated with each node to be evaluated to obtain one or more intelligent entities to be optimized. This solution obtains the corresponding execution path by querying the task execution path map through error data, determines the task nodes to be evaluated in the execution path, performs log and resource consumption detection on the task nodes to be evaluated to determine the intelligent entity to be optimized, thereby improving the rationality and efficiency of determining the intelligent entity to be optimized.

[0057] Figure 7 A flowchart of a multi-agent data management method based on negative feedback optimization for determining optimization data of an agent to be optimized is provided in an embodiment of the present application, such as Figure 7 As shown, specifically including: Step S501: Perform data governance on the data to be governed based on the multi-agent system to obtain corresponding data governance results, and create a task execution path map corresponding to the data governance process in real time, wherein the multi-agent system includes multiple agents that can independently execute data governance tasks.

[0058] Step S502: Evaluate the data governance results based on the preset evaluation indicator set and the dynamic threshold set to obtain negative feedback data, wherein the dynamic threshold set is updated every other evaluation process.

[0059] Step S503: query the task execution path map based on the negative feedback data to obtain the corresponding task processing path, evaluate the task processing path to obtain one or more intelligent agents to be optimized, and determine the corresponding matching optimization parameter type in the optimization parameter matching library based on the error type and task type.

[0060] The negative feedback data includes the error type of the error data. The error type and the task type of the agent to be optimized can be used to determine the corresponding optimization parameter type in the optimization parameter matching library. For example, if the error type is a consistency error type and the task type of the agent to be optimized is data cleaning, the optimization parameter matching library is searched for optimization parameter types that match the consistency error type and data cleaning. The query result is the learning rate in the model training parameters.

[0061] Step S504: When there are multiple intelligent agents to be optimized, the task types of the multiple intelligent agents to be optimized are combined to obtain a task type combination. According to the task type combination, the corresponding error type influence distribution information is queried to obtain the influence of each task type on the error type. According to the influence, the parameter value of the optimization parameter type corresponding to the intelligent agent to be optimized is determined, and each optimization parameter type and parameter value are determined as the optimization data of the corresponding intelligent agent to be optimized.

[0062] In the case where there are multiple intelligent agents to be optimized, the task types of the multiple intelligent agents to be optimized are combined to obtain a task type combination. By querying the corresponding error type impact distribution information through the task type combination, the impact of each task type on the error type can be obtained. The error type impact distribution information can be the distribution information of the impact of each task type on the error type in a pre-set task type combination. The impact can be used to determine the parameter value of the optimization parameter type of each intelligent agent to be optimized, and the optimization parameter type and the parameter value can be determined as the optimization data corresponding to the intelligent agent to be optimized. Exemplarily, the task type of the agent 1 to be optimized is data cleaning, the optimization parameter type is learning rate, the task type of the agent 2 to be optimized is data integration, and the optimization parameter type is online learning frequency. The task type combination is (data cleaning, data integration), the error type is consistency error, and the corresponding consistency impact distribution information queried is data cleaning 40%, data integration 60%, when the impact of data cleaning on consistency is 40%, the corresponding pre-set learning rate value is 0.01, and when the impact of data integration on consistency is 60%, the corresponding pre-set online learning frequency value is batch updated after accumulating 500 samples. The learning rate of 0.01 is determined as the optimization data of the agent 1 to be optimized, and the online learning frequency of batch updated after accumulating 500 samples is determined as the optimization data of the agent 2 to be optimized.

[0063] Step S505: Optimize the corresponding agent to be optimized according to the optimization data, and re-perform data management of the data to be managed based on the optimized multi-agent system.

[0064] As can be seen from the above, by determining the corresponding matching optimization parameter type in the optimization parameter matching library based on the error type and task type, when there are multiple intelligent agents to be optimized, the task types of the multiple intelligent agents to be optimized are combined to obtain a task type combination. The corresponding error type influence distribution information is queried based on the task type combination to obtain the influence of each task type on the error type. The parameter value of the optimization parameter type corresponding to the intelligent agent to be optimized is determined based on the influence, and each optimization parameter type and parameter value are determined as the optimization data corresponding to the intelligent agent to be optimized. This solution determines the corresponding matching optimization parameters in the database based on the error type and the intelligent agent's task type, and determines the value of the optimization parameter based on the influence of each task type on the error type, which can ensure the rationality and accuracy of the optimization data.

[0065] Figure 8 This is a module structure diagram of a multi-agent data governance system based on negative feedback optimization provided in an embodiment of the present application. The system is used to execute a multi-agent data governance method based on negative feedback optimization provided in the above embodiment, and has the corresponding functional modules and beneficial effects of the execution method. Figure 8 As shown, the system specifically includes: A data governance module 101 is configured to perform data governance on the data to be governed based on a multi-agent system to obtain corresponding data governance results, wherein the multi-agent system includes multiple agents that can independently perform data governance tasks; A graph creation module 102 is used to create a task execution path graph corresponding to the data governance process in real time; An evaluation processing module 103 is configured to evaluate the data governance results based on a preset evaluation indicator set and a dynamic threshold set to obtain negative feedback data, wherein the dynamic threshold set is updated every other evaluation process; A query module 104 is configured to query the task execution path map according to the negative feedback data to obtain a corresponding task processing path; The evaluation processing module 103 is further configured to evaluate the task processing path to obtain one or more agents to be optimized; A matching processing module 105 is configured to perform matching processing based on the negative feedback data, the task type of the agent to be optimized, and an optimization parameter matching library to obtain corresponding optimization data; An optimization module 106 is configured to optimize the corresponding agent to be optimized according to the optimization data; The data governance module 101 is also used to re-perform data governance of the data to be governed based on the optimized multi-agent system.

[0066] According to the above scheme, by performing data governance on the data to be governed based on a multi-agent system, the corresponding data governance results are obtained, and a task execution path map corresponding to the data governance process is created in real time. The multi-agent system includes multiple agents that can independently execute data governance tasks. The data governance results are evaluated and processed based on a preset evaluation index set and a dynamic threshold set to obtain negative feedback data. The dynamic threshold set is updated every time the evaluation process is performed. The task execution path map is queried based on the negative feedback data to obtain the corresponding task processing path. The task processing path is evaluated and processed to obtain one or more agents to be optimized. Based on the negative feedback data, the task type of the agent to be optimized and the optimization parameter matching library, the corresponding optimization data is obtained. The corresponding agent to be optimized is optimized according to the optimization data, and the data governance of the data to be governed is re-performed based on the optimized multi-agent system. This scheme solves the problem of the lack of systematic mining and utilization of negative feedback data generated in the data governance process in the existing technology, which leads to relatively low data governance processing efficiency, by determining the negative feedback data in the data governance results and optimizing the multi-agent system based on the negative feedback data. It can optimize and iterate the agents based on the negative feedback learning mechanism to perform data governance in an efficient and better way.

[0067] In a possible embodiment, the map creation module 102 is specifically configured to: Record the data flow and data processing volume of each agent executed in the data governance process in real time, generate a dedicated task node for each agent, and generate directed edges between each dedicated task node according to the data flow of each agent to obtain a graph of the task execution path to be adjusted; Each exclusive task node in the task execution path map to be adjusted is adjusted according to the data processing amount to obtain a task execution path map.

[0068] In a possible embodiment, the atlas creation module 102 is further configured to: According to the data processing amount of each intelligent agent, an evaluation weight is allocated to the corresponding exclusive task node, and each evaluation weight is marked at a preset position of the corresponding exclusive task node to obtain a task execution path map.

[0069] In a possible embodiment, the evaluation processing module 103 is specifically configured to: Conduct integrity assessment, consistency assessment, and timeliness assessment on each item of result data in the data governance results to obtain corresponding integrity assessment values, consistency assessment values, and timeliness assessment values; The integrity assessment value, the consistency assessment value and the timeliness assessment value are respectively compared with the corresponding dynamic thresholds in the dynamic threshold set, and the result data with an assessment value less than the corresponding dynamic threshold is determined as negative feedback data.

[0070] In a possible embodiment, the evaluation processing module 103 is further configured to: Querying the task node that outputs the error data in the task execution path map, and performing path backtracking in the task execution path map according to the task node that outputs the error data to obtain a task processing path; Determine a first task node in the task processing path whose node task type is associated with the error type, and screen out a second task node in the task processing path whose evaluation weight is greater than a preset weight, and determine the first task node and the second task node as nodes to be evaluated respectively; An evaluation process is performed based on the task execution logs and resource consumption monitoring information associated with each of the nodes to be evaluated to obtain one or more intelligent agents to be optimized.

[0071] In a possible embodiment, the evaluation processing module 103 is further configured to: Performing anomaly detection on each record information in the task execution log to obtain a log detection result of each node to be evaluated; Determine corresponding resource consumption threshold information according to the node task type of each node to be evaluated, and compare the resource consumption monitoring information with the corresponding resource consumption threshold information to obtain a resource consumption detection result of each node to be evaluated; The intelligent agent corresponding to the node to be evaluated whose log detection result or resource consumption detection result is abnormal is determined as the intelligent agent to be optimized.

[0072] In a possible embodiment, the matching processing module 105 is specifically configured to: Determine the corresponding matching optimization parameter type in the optimization parameter matching library according to the error type and the task type; In the case where there are multiple agents to be optimized, the task types of the multiple agents to be optimized are combined to obtain a task type combination, and the corresponding error type influence distribution information is queried according to the task type combination to obtain the influence of each task type on the error type; Determine the parameter value of the optimization parameter type corresponding to the agent to be optimized based on the influence, and determine each of the optimization parameter types and the parameter value as the optimization data corresponding to the agent to be optimized.

[0073] Figure 9A schematic diagram of the structure of a multi-agent data management device based on negative feedback optimization provided in an embodiment of the present application is shown as follows: Figure 9 As shown, the device includes a processor 201, a memory 202, an input device 203 and an output device 204; the number of processors 201 in the device can be one or more. Figure 9 In the embodiment, a processor 201 is used as an example; the processor 201, the memory 202, the input device 203 and the output device 204 in the device can be connected by a bus or other means. Figure 9 The example of the connection via bus is taken as an example. The memory 202, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as program instructions or modules corresponding to a multi-agent data governance method based on negative feedback optimization in an embodiment of the present application. The processor 201 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory 202, that is, realizes the above-mentioned multi-agent data governance method based on negative feedback optimization. The input device 203 can be used to receive input digital or character information, and generate key signal input related to user settings and function control of the device. The output device 204 may include a display device such as a display screen.

[0074] The present application also provides a storage medium containing computer-executable instructions. When executed by a computer processor, the computer-executable instructions are used to perform a multi-agent data governance method based on negative feedback optimization, the method comprising: Performing data governance on the data to be governed based on a multi-agent system to obtain corresponding data governance results, and creating a task execution path map corresponding to the data governance process in real time, wherein the multi-agent system includes multiple agents that can independently execute data governance tasks; Evaluate the data governance results based on a preset evaluation indicator set and a dynamic threshold set to obtain negative feedback data, and the dynamic threshold set is updated every other evaluation process; According to the negative feedback data, the task execution path map is queried to obtain a corresponding task processing path, the task processing path is evaluated to obtain one or more agents to be optimized, and a matching process is performed based on the negative feedback data, the task type of the agent to be optimized, and an optimization parameter matching library to obtain corresponding optimization data; The corresponding intelligent agent to be optimized is optimized according to the optimization data, and data governance of the data to be governed is re-performed based on the optimized multi-agent system.

[0075] It is worth noting that in the above-mentioned embodiment of the multi-agent data governance method system based on negative feedback optimization, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the protection scope of the embodiments of this application.

[0076] Note that the above are only preferred embodiments of the present application and the technical principles used. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present application. Therefore, although the embodiments of the present application are described in more detail through the above embodiments, the embodiments of the present application are not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the embodiments of the present application, and the scope of the embodiments of the present application is determined by the scope of the appended claims.

Claims

1. A multi-agent data governance method based on negative feedback optimization, characterized by: The method comprises: Performing data governance on the data to be governed based on a multi-agent system to obtain corresponding data governance results, and creating a task execution path map corresponding to the data governance process in real time, wherein the multi-agent system includes multiple agents that can independently execute data governance tasks; Evaluate the data governance results based on a preset evaluation indicator set and a dynamic threshold set to obtain negative feedback data, and the dynamic threshold set is updated every other evaluation process; According to the negative feedback data, the task execution path map is queried to obtain a corresponding task processing path, the task processing path is evaluated to obtain one or more agents to be optimized, and a matching process is performed based on the negative feedback data, the task type of the agent to be optimized, and an optimization parameter matching library to obtain corresponding optimization data; The corresponding intelligent agent to be optimized is optimized according to the optimization data, and data management of the data to be managed is re-performed based on the optimized multi-agent system.

2. The multi-agent data governance method based on negative feedback optimization according to claim 1 is characterized in that: The task execution path map corresponding to the process of creating the data governance in real time includes: Record the data flow and data processing volume of each agent executed in the data governance process in real time, generate a dedicated task node for each agent, and generate directed edges between each dedicated task node according to the data flow of each agent to obtain a graph of the task execution path to be adjusted; Each exclusive task node in the task execution path map to be adjusted is adjusted according to the data processing amount to obtain a task execution path map.

3. The multi-agent data governance method based on negative feedback optimization according to claim 2 is characterized in that: The step of adjusting each dedicated task node in the to-be-adjusted task execution path map according to the data processing amount to obtain the task execution path map includes: According to the data processing amount of each intelligent agent, an evaluation weight is allocated to the corresponding exclusive task node, and each evaluation weight is marked at a preset position of the corresponding exclusive task node to obtain a task execution path map.

4. The multi-agent data governance method based on negative feedback optimization according to any one of claims 1 to 3, characterized in that: The preset evaluation indicator set includes an integrity indicator, a consistency indicator, and a timeliness indicator. The negative feedback data obtained by evaluating the data governance results based on the preset evaluation indicator set and the dynamic threshold set includes: Conduct integrity assessment, consistency assessment, and timeliness assessment on each item of result data in the data governance results to obtain corresponding integrity assessment values, consistency assessment values, and timeliness assessment values; The integrity assessment value, the consistency assessment value and the timeliness assessment value are respectively compared with the corresponding dynamic thresholds in the dynamic threshold set, and the result data with an assessment value less than the corresponding dynamic threshold is determined as negative feedback data.

5. The multi-agent data governance method based on negative feedback optimization according to any one of claims 1 to 3, characterized in that: The negative feedback data includes error data and error types, the task execution path graph includes each directed-connected task node and a corresponding evaluation weight, querying the task execution path graph according to the negative feedback data to obtain a corresponding task processing path, and evaluating the task processing path to obtain one or more agents to be optimized, including: Querying the task node that outputs the error data in the task execution path map, and performing path backtracking in the task execution path map according to the task node that outputs the error data to obtain a task processing path; Determine a first task node in the task processing path whose node task type is associated with the error type, and screen out a second task node in the task processing path whose evaluation weight is greater than a preset weight, and determine the first task node and the second task node as nodes to be evaluated respectively; An evaluation process is performed based on the task execution logs and resource consumption monitoring information associated with each of the nodes to be evaluated to obtain one or more intelligent agents to be optimized.

6. The multi-agent data governance method based on negative feedback optimization according to claim 5 is characterized in that: The step of performing evaluation processing based on the task execution logs and resource consumption monitoring information associated with each of the nodes to be evaluated to obtain one or more agents to be optimized includes: Performing anomaly detection on each record information in the task execution log to obtain a log detection result of each node to be evaluated; Determine corresponding resource consumption threshold information according to the node task type of each node to be evaluated, and compare the resource consumption monitoring information with the corresponding resource consumption threshold information to obtain a resource consumption detection result of each node to be evaluated; The intelligent agent corresponding to the node to be evaluated whose log detection result or resource consumption detection result is abnormal is determined as the intelligent agent to be optimized.

7. The multi-agent data governance method based on negative feedback optimization according to any one of claims 1 to 3, characterized in that: The negative feedback data includes an error type of the error data, and the matching process based on the negative feedback data, the task type of the agent to be optimized, and the optimization parameter matching library to obtain the corresponding optimization data includes: Determine the corresponding matching optimization parameter type in the optimization parameter matching library according to the error type and the task type; In the case where there are multiple agents to be optimized, the task types of the multiple agents to be optimized are combined to obtain a task type combination, and the corresponding error type influence distribution information is queried according to the task type combination to obtain the influence of each task type on the error type; The parameter value of the optimization parameter type corresponding to the intelligent agent to be optimized is determined according to the influence degree, and each of the optimization parameter types and the parameter value is determined as the optimization data corresponding to the intelligent agent to be optimized.

8. A multi-agent data governance system based on negative feedback optimization, characterized by: include: A data governance module, configured to perform data governance on the data to be governed based on a multi-agent system and obtain corresponding data governance results, wherein the multi-agent system includes multiple agents that can independently perform data governance tasks; A graph creation module, used to create a task execution path graph corresponding to the data governance process in real time; An evaluation processing module, configured to evaluate the data governance results based on a preset evaluation indicator set and a dynamic threshold set to obtain negative feedback data, wherein the dynamic threshold set is updated every other evaluation process; A query module, configured to query the task execution path map according to the negative feedback data to obtain a corresponding task processing path; The evaluation processing module is further configured to evaluate the task processing path to obtain one or more intelligent agents to be optimized; A matching processing module, configured to perform matching processing based on the negative feedback data, the task type of the agent to be optimized, and an optimization parameter matching library to obtain corresponding optimization data; An optimization module, configured to optimize the corresponding intelligent agent to be optimized according to the optimization data; The data governance module is also used to re-perform data governance of the data to be governed based on the optimized multi-agent system.

9. A multi-agent data management device based on negative feedback optimization, the device comprising: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, so that the one or more processors implement the multi-agent data governance method based on negative feedback optimization as described in any one of claims 1-7.

10. A storage medium storing computer-executable instructions, which, when executed by a computer processor, are used to execute the multi-agent data governance method based on negative feedback optimization as described in any one of claims 1 to 7.

Images