Multi-source sharing automation task collaborative optimization method and system of AI Agent

Abstract

The invention discloses a multi-source sharing automation task collaborative optimization method and system of an AI Agent, and relates to the related field of data processing, and the method comprises the steps: carrying out the task complexity calculation of a target task, and generating a decomposition instruction when a calculation result meets a preset threshold value; performing hierarchical disassembly on the target task to generate a sub-task set, and establishing sub-task dependence; performing self-inspection on the environment state, and establishing a task execution balance function according to a self-inspection result and a task demand; executing scheduling optimization of the subtask set, and establishing a multi-round scheduling optimization result; and performing busy state prediction on the environment, generating scheduling compensation according to a prediction result, performing optimization result screening through the scheduling compensation, and establishing an Agent decision scheme. The technical problems of low execution efficiency, non-uniform resource allocation and task failure rate increase in existing task cooperative execution are solved, and the technical effects of high task execution efficiency, high resource utilization rate and remarkable reduction of the task failure rate are achieved.

Description

Technical Field

[0001] This application relates to the field of data processing, and particularly to a multi-source shared automated task collaborative optimization method and system for AI Agent. Background Art

[0002] With the wide application of AI Agent in complex environments, how to efficiently, resource-optimally and stably collaborate to execute multiple tasks has become the key to improving the overall performance of the system. Currently, the main method to solve this problem is to assign tasks one by one to the Agent for processing through simple task assignment and sequential execution strategies. However, the current methods have problems such as low execution efficiency, uneven resource allocation, and increased task failure rate due to the lack of careful consideration of task complexity, effective management of dependencies between tasks, and adaptability to dynamic environmental changes.

[0003] In the current related technologies, there are technical problems such as low execution efficiency, uneven resource allocation, and increased task failure rate in the collaborative execution of multi-source shared automated tasks by AI Agent. Summary of the Invention

[0004] By providing a multi-source shared automated task collaborative optimization method and system for AI Agent, this application calculates the task complexity, disassembles the task into a sub-task set as needed and establishes dependencies, constructs a balance function based on environmental self-check and task requirements, uses the sub-task dependencies as constraints, performs scheduling optimization through the balance function, predicts the busy state of the environment and generates scheduling compensation, and screens the scheduling optimization results to form an Agent decision-making plan, achieving the technical effects of high task execution efficiency, high resource utilization rate, and significantly reduced task failure rate.

[0005] This application provides a multi-source shared automated task collaborative optimization method for AI Agent, including: calculating the task complexity of the target task, and generating a decomposition instruction when the task complexity calculation result meets a preset complexity threshold; hierarchically disassembling the target task using the decomposition instruction to generate a sub-task set and establishing sub-task dependencies; performing a state self-check on the environmental state, and establishing a task execution balance function based on the state self-check result and the task requirements of the target task, where the balance objectives of the task execution balance function include an execution efficiency objective, a resource utilization rate objective, and a task failure rate objective; using the sub-task dependencies as constraints, performing execution scheduling optimization of the sub-task set through the task execution balance function to establish multiple rounds of scheduling optimization results; predicting the busy state of the environment, generating scheduling compensation based on the busy state prediction result, and screening the multiple rounds of scheduling optimization results through the scheduling compensation to establish an Agent decision-making plan.

[0006] In a possible implementation, hierarchically decompose the target task using the decomposition instruction to generate a set of subtasks, and perform the following processing: Decompose the target task according to data flow, logical relationship, and resource occupancy respectively, control the task granularity through a multi-level task dependency graph, and establish a set of decomposed tasks; Calculate the task similarity of the set of decomposed tasks, and generate a set of subtasks according to the task similarity calculation result.

[0007] In a possible implementation, to establish subtask dependencies, perform the following processing: Analyze the input data of the tasks in the set of subtasks, and establish data flow dependencies according to the input data analysis result; Analyze the execution order of the set of subtasks, and establish timing dependencies according to the execution order analysis result; Analyze the tasks in the set of subtasks for sharing the same resources, and establish resource competition dependencies; Analyze the impact of the execution mode under parallel execution on the set of subtasks, and establish collaborative optimization dependencies; Generate subtask dependencies based on the data flow dependencies, the timing dependencies, the resource competition dependencies, and the collaborative optimization dependencies.

[0008] In a possible implementation, to establish a task execution balance function according to the status self-check result and the task requirements of the target task, perform the following processing: Obtain the historical database for collaborative task optimization, and construct a benchmark balance function based on the historical database. The benchmark balance function is configured with a benchmark weight factor; Use the status self-check result to perform adaptation analysis of execution efficiency, resource utilization rate, and task failure rate, and establish a first set of adjustment factors according to the adaptation analysis result; Use the task requirements to perform adaptation analysis of execution efficiency, resource utilization rate, and task failure rate, and establish a second set of adjustment factors according to the adaptation analysis result; After compensating the benchmark weight factor using the first set of adjustment factors and the second set of adjustment factors, establish a task execution balance function.

[0009] In a possible implementation, using the subtask dependencies as constraints, perform execution scheduling optimization of the set of subtasks through the task execution balance function, and establish multi-round scheduling optimization results. Perform the following processing: Determine the order constraint and execution precondition constraint of task execution according to the subtask dependencies; While maintaining the order constraint and the execution precondition constraint, perform adjustment of the execution strategy of the set of subtasks, and calculate the balance score through the task execution balance function; Update the execution strategy according to the balance score, perform iteration of the execution strategy adjustment, and establish multi-round scheduling optimization results.

[0010] In a possible implementation, updating the execution policy according to the balance score performs the following processing: performing itemized fitness calculation, where the itemized fitness calculation includes execution efficiency calculation, resource utilization rate calculation, and task failure rate calculation; establishing an update direction policy according to the itemized fitness calculation results; establishing an update step size policy according to the balance score, and performing an update search based on the update direction policy and the update step size policy to complete the policy adjustment iteration and establish the multi-round scheduling optimization result.

[0011] In a possible implementation, predicting the busy state of the environment and generating a scheduling compensation based on the busy state prediction result performs the following processing: performing identification of the main components of the predicted metrics to determine the key metrics; calling the real-time monitoring data and historical execution data, and performing regression analysis through the key metrics, the real-time monitoring data, and the historical execution data to generate the busy state prediction result; establishing a scheduling constraint according to the busy state prediction result, and using the scheduling constraint as the scheduling compensation for decision optimization.

[0012] In a possible implementation, after establishing the Agent decision plan, the following processing is performed: monitoring the execution of the Agent decision plan to establish an execution feedback; determining whether the deviation between the execution feedback and the Agent decision plan meets the expected deviation threshold; when the deviation between the execution feedback and the Agent decision plan meets the expected deviation threshold, generating an automatic correction instruction and performing overclock compensation for the Agent decision plan according to the automatic correction instruction.

[0013] In a possible implementation, determining whether the deviation between the execution feedback and the Agent decision plan meets the expected deviation threshold performs the following processing: when the deviation between the execution feedback and the Agent decision plan does not meet the expected deviation threshold, generating a deviation accumulation according to the deviation, and performing cumulative compensation for the Agent decision plan through the deviation accumulation.

[0014] The present application also provides a multi-source shared automated task collaborative optimization system for an AI Agent, including: a task complexity calculation module, configured to calculate the task complexity of a target task, and generate a decomposition instruction when the task complexity calculation result meets a preset complexity threshold; a task decomposition module, configured to hierarchically disassemble the target task by using the decomposition instruction, generate a subtask set, and establish subtask dependencies; a status self-check and task execution balance function establishment module, configured to perform a status self-check on the environmental status, and establish a task execution balance function according to the status self-check result and the task requirements of the target task, where the balance objectives of the task execution balance function include an execution efficiency objective, a resource utilization rate objective, and a task failure rate objective; a subtask execution scheduling optimization module, configured to use the subtask dependencies as constraints, perform execution scheduling optimization on the subtask set through the task execution balance function, and establish multiple rounds of scheduling optimization results; an environmental busy status prediction and scheduling compensation module, configured to predict the busy status of the environment, generate scheduling compensation based on the busy status prediction result, and screen the multiple rounds of scheduling optimization results through the scheduling compensation to establish an Agent decision-making plan.

[0015] It is intended to first calculate the task complexity of a target task through the multi-source shared automated task collaborative optimization method and system for an AI Agent proposed in the present application. When the task complexity calculation result meets a preset complexity threshold, a decomposition instruction is generated. Then, the target task is hierarchically disassembled by using the decomposition instruction to generate a subtask set and establish subtask dependencies. Next, a status self-check is performed on the environmental status, and a task execution balance function is established according to the status self-check result and the task requirements of the target task. The balance objectives of the task execution balance function include an execution efficiency objective, a resource utilization rate objective, and a task failure rate objective. Then, using the subtask dependencies as constraints, the execution scheduling optimization of the subtask set is performed through the task execution balance function to establish multiple rounds of scheduling optimization results. Finally, the busy status of the environment is predicted, scheduling compensation is generated based on the busy status prediction result, and the multiple rounds of scheduling optimization results are screened through the scheduling compensation to establish an Agent decision-making plan. The technical effects of high efficiency in task execution, high utilization rate of resources, and significant reduction in the task failure rate are achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings of the embodiments of the present invention will be briefly introduced below. Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the operations in the front or below do not necessarily need to be executed precisely in sequence. On the contrary, according to the need, they can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several operations can be removed from these processes.

[0017] Figure 1 This is a schematic flowchart of the multi-source shared automated task collaborative optimization method for the AI Agent provided by the embodiments of the present application.

[0018] Figure 2 This is a schematic structural diagram of the multi-source shared automated task collaborative optimization system for the AI Agent provided by the embodiments of the present application.

[0019] Explanation of reference numerals: Task complexity calculation module 10, task decomposition module 20, state self-check and task execution balance function establishment module 30, sub-task execution scheduling optimization module 40, environment busy state prediction and scheduling compensation module 50. Detailed implementation manners

[0020] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the following specifically gives the detailed implementation manners of the present application.

[0021] In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0022] In the following description, "some embodiments" are involved, which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. The terms "first" and "second" involved are only used to distinguish similar objects and do not represent a specific order for the objects. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or modules not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.

[0023] The embodiments of the present application provide a multi-source shared automated task collaborative optimization method for the AI Agent, as Figure 1 shown, the method includes: Step S100, calculate the task complexity of the target task. When the calculation result of the task complexity meets the preset complexity threshold, generate a decomposition instruction.

[0024] Specifically, perform a structured analysis on the target task, that is, perform natural language processing (NLP) on the task description, and extract the main steps of the task, input and output requirements, required data sets or resources, etc. For example, NLP libraries (such as NLTK, SpaCy) can be used for text parsing, or specialized task parsing tools can be used to extract the key information of the task. From the parsed task, key features such as task scale (data processing volume, computing volume), required resources (CPU, memory, storage), task type (batch processing, real-time processing), task nesting level, etc. are extracted through statistical methods (such as counting, summing) or machine learning models (such as feature selectors). Using a predefined complexity calculation model, the extracted features are input into the model for calculation. The complexity calculation model is based on machine learning algorithms and is trained through historical task data, and can accurately evaluate the complexity of the task. Compare the calculated complexity with the preset complexity threshold. If the task complexity exceeds the threshold, generate a decomposition instruction.

[0025] Step S200, hierarchically decompose the target task using the decomposition instruction, generate a set of subtasks, and establish subtask dependencies.

[0026] Specifically, the hierarchical decomposition refers to decomposing a complex task into multiple relatively simple subtasks to form a hierarchical structure. According to the type and complexity of the target task, select a decomposition strategy, such as by functional module, data processing flow, time sequence, etc. According to the decomposition strategy, decompose the target task into multiple relatively independent subtasks, and each subtask has clear goals, inputs and outputs, and required resources. By drawing a task flow chart or using a task dependency graph (such as DAG, directed acyclic graph), analyze the dependency relationships between the subtasks, and determine their execution order, that is, which subtasks need to be completed first and which subtasks can be executed in parallel. Among them, subtask dependency refers to the sequence or dependency relationship that exists between subtasks.

[0027] In a possible implementation, the step of hierarchically disassembling the target task by using the disassembly instruction to generate a set of subtasks, step S200 further includes step S210, which disassembles the target task according to data flow, logical relationship, and resource occupation respectively, controls the task granularity through a multi-level task dependency graph, and establishes a set of disassembled tasks. Specifically, identify the input data source and output data target of the target task. Trace the flow path of data in the task, mark each data processing node, and each node represents a data processing step. Draw a data flow diagram to show the complete path of data from input to output and the data transfer relationship between each node. Among them, the data flow diagram is a graphical representation method for describing the data flow and processing in a system.

[0028] Determine the key steps and decision points in the target task. Analyze the dependency relationships between steps, including sequential dependencies, conditional dependencies, and parallel relationships, that is, determine which tasks must start after the previous task is completed, which tasks can be executed in parallel, and whether there are conditional branches (such as task selection based on specific conditions). Draw a logic flow chart to show the complete logical path of task execution, including conditional branches and loop structures. Among them, the logic flow chart is a graphical representation method for describing the sequential, conditional dependency, and parallel relationships between tasks.

[0029] Use a resource requirement assessment tool to evaluate the computing resources (such as CPU, memory) and storage resources required for the target task. Analyze the possible resource bottlenecks that may occur during task execution, such as memory overflow, disk I / O bottleneck, etc. According to the resource requirements and availability, disassemble the task into parts that can be independently or collaboratively executed on different resources.

[0030] After determining the data flow, logical relationship, and resource occupation, construct a multi-level task dependency graph. Each layer represents a different level of task granularity, gradually refining from the coarse-grained target task to the fine-grained executable subtasks. The dependency relationship is represented by an arrow, indicating the sequential execution order and dependency conditions between tasks. According to the multi-level task dependency graph, collect all the refined subtasks to form a set of disassembled tasks. This set contains all the subtasks required to execute the task and their dependency relationships.

[0031] Step S220: Calculate the task similarity of the split task set, and generate a subtask set according to the task similarity calculation result. Specifically, use the cosine similarity algorithm to compare the characteristics of tasks in the split task set (such as input data, processing steps, output results, etc.). Calculate the similarity scores between each pair of tasks. According to the similarity calculation result, merge highly similar tasks into a single subtask (the merger does not affect the independent execution and dependency relationship of the tasks). At the same time, ensure that each subtask maintains its unique processing logic and data flow. Finally, generate an optimized subtask set, where each subtask is independent and necessary, while maintaining the dependency relationship between tasks. This implementation method can more effectively allocate resources, reduce waiting time, and improve the overall execution efficiency by refining the task granularity and identifying independently executable subtasks.

[0032] In a possible implementation, the step of establishing subtask dependencies, step S200, further includes step S230: Analyze the input data of the tasks in the subtask set, and establish a data flow dependency according to the input data analysis result. Specifically, use a data management tool or script to automatically collect the input data required by each subtask, including but not limited to the source, format, size, access frequency, etc. of the data, to determine the data flow relationship between tasks. Based on the input data analysis result, determine the dependency relationship formed between different subtasks due to data transfer.

[0033] Step S240: Analyze the execution order of the subtask set, and establish a timing dependency according to the execution order analysis result. Specifically, based on the task logic and the overall requirements of the target task, use tools such as flowcharts, Gantt charts, or task dependency graphs to detail the execution order of the subtasks. Based on the execution order analysis, clarify the time sequence dependency between subtasks.

[0034] Step S250: Analyze the tasks in the subtask set for sharing the same resources, and establish a resource competition dependency. Specifically, use a resource monitoring tool or a custom script to detect the occupancy of key resources in the system in real time or regularly. Conduct a resource requirement analysis for the subtasks, including the amount of resources such as CPU, memory, and I / O required. Through comparative analysis, identify task pairs that may have resource competition. That is, the resource competition dependency refers to analyzing whether multiple subtasks may access or modify the same resource (such as memory, files, database connections, etc.) at the same time, thereby forming a resource competition relationship.

[0035] Step S260: Conduct an impact analysis on the execution mode during parallel execution of the subtask set, and establish collaborative optimization dependencies. Specifically, simulate the behaviors of different subtasks in a parallel execution environment, and analyze their impacts on aspects such as system performance, task completion time, and resource consumption. Use means such as performance testing and load testing to evaluate the effects of different execution strategies, and determine the optimal collaborative execution strategy.

[0036] Step S270: Generate subtask dependencies based on the data flow dependencies, the timing dependencies, the resource competition dependencies, and the collaborative optimization dependencies. Specifically, synthesize the analysis results of steps S230 to S260, and use graph theory or task scheduling algorithms to construct a complete subtask dependency graph that includes data flow dependencies, timing dependencies, resource competition dependencies, and collaborative optimization dependencies. This graph details various dependency relationships and execution constraints between subtasks, thereby providing an accurate dependency relationship model for subsequent task execution scheduling optimization, ensuring that tasks can be efficiently executed according to the predetermined logical order, resource allocation, and collaborative strategy. This implementation method ensures the accuracy, efficiency, and stability of task scheduling and execution by comprehensively and meticulously analyzing various dependency relationships between subtasks.

[0037] Step S300: Conduct a self-check on the environmental state, and establish a task execution balance function based on the self-check result and the task requirements of the target task. The balance objectives of the task execution balance function include an execution efficiency objective, a resource utilization rate objective, and a task failure rate objective.

[0038] Specifically, use system monitoring tools (such as Prometheus, Grafana) or monitoring services provided by cloud service providers to monitor key resources of the system in real time, such as CPU usage rate, memory occupancy, disk space, and network bandwidth. Based on the monitoring data, evaluate the overall state of the current environment, including the sufficiency of resources, the stability of the system, and potential bottlenecks. Analyze the resource requirements of the target task, such as the required number of CPU cores, memory size, and execution time window. Match the task requirements of the target task with the current environmental state to determine whether there are resource conflicts or potential risks. Based on the above evaluation results, construct a balance function that includes an execution efficiency objective, a resource utilization rate objective, and a task failure rate objective. This function is used to guide subsequent subtask scheduling. According to business requirements and environmental characteristics, assign reasonable weights to these three objectives.

[0039] In a possible implementation, establishing a task execution balance function according to the status self-check result and the task requirements of the target task, step S300 further includes step S310 of obtaining a historical database for collaborative task optimization, constructing a benchmark balance function according to the historical database, and the benchmark balance function is configured with benchmark weight factors. Specifically, extract historical data of past task executions from the database, including key indicators such as execution efficiency, resource utilization rate, and task failure rate. Clean, organize, and analyze this data to identify the performance under different task types, execution environments, and resource allocations. Based on the analysis results, construct a benchmark balance function that can reflect the balance relationship among execution efficiency, resource utilization rate, and task failure rate. The benchmark balance function is used to balance the three goals of execution efficiency, resource utilization rate, and task failure rate during task execution. Configure a set of benchmark weight factors for the benchmark balance function, and these factors are set according to the relative importance of each goal in the historical data.

[0040] Step S320, perform adaptation analysis of execution efficiency, resource utilization rate, and task failure rate using the status self-check result, and establish a first set of adjustment factors according to the adaptation analysis result. Specifically, perform a status self-check on the current execution environment to obtain real-time data on aspects such as system resources, network status, and hardware performance. Analyze the possible impacts of these status data on task execution efficiency, resource utilization rate, and task failure rate. According to the analysis results, calculate an adjustment factor for each goal in the benchmark balance function, and these factors reflect the impact of the current environment status on the goal weights.

[0041] Step S330, perform adaptation analysis of execution efficiency, resource utilization rate, and task failure rate using the task requirements, and establish a second set of adjustment factors according to the adaptation analysis result. Specifically, analyze the specific requirements of the target task, including task type, execution time requirement, resource requirement, etc. According to these requirements, evaluate their impacts on execution efficiency, resource utilization rate, and task failure rate. Calculate another adjustment factor for each goal in the benchmark balance function, and these factors reflect the impact of the task requirements on the goal weights.

[0042] Step S340: After compensating the reference weight factor using the first set of adjustment factors and the second set of adjustment factors, a task execution balance function is established. Specifically, the first set and the second set of adjustment factors are applied to the reference weight factor in the reference balance function for weight compensation. According to the compensated weight factor, the task execution balance function is reconstructed. This function is used in the subsequent task execution scheduling optimization process to guide the allocation and execution order of subtasks. This implementation method ensures that during the collaborative optimization process of multi-source shared automated tasks, the AI Agent can flexibly adjust the execution strategy according to different environments and task requirements, achieving efficient, stable, and reliable task execution by comprehensively considering the state of the current execution environment and task requirements.

[0043] The following is a specific example: The AI Agent needs to execute a complex data analysis task, which involves multiple subtasks, including data cleaning, feature extraction, model training, and result evaluation. The goal of the task is to complete the task with the highest execution efficiency and the lowest failure rate under limited computing resources. Extract historical execution data of similar data analysis tasks in the past year from the database. The data includes task type, execution environment configuration, resource usage, execution time, and success / failure status. Based on the historical data, use the NSGA-II algorithm to construct a benchmark balance function, which is used to balance the execution efficiency (measured by task completion time), resource utilization rate (measured by CPU and memory usage), and task failure rate (measured by the task success / failure ratio). The initial benchmark weight factors are set as follows: the execution efficiency is 0.5, the resource utilization rate is 0.3, and the task failure rate is 0.2. The AI Agent performs a self-check on the current execution environment status to obtain real-time data on CPU usage, memory usage, and network bandwidth. Analyze the real-time data and find that the CPU usage is relatively high and the memory usage is relatively low. Predict that in the current state, more attention should be paid to the optimized use of memory for the goal of improving resource utilization. According to the adaptation analysis results, calculate an adjustment factor for the resource utilization rate target in the benchmark balance function (increase from 0.3 to 0.4). The adjustment factors for the execution efficiency and task failure rate remain unchanged (0.5 and 0.2 respectively, but since the sum of the weights needs to remain 1, the weight of the execution efficiency will be correspondingly reduced to 0.4 to keep the sum of the weights at 1). Analyze the requirements of the current data analysis task and find that the task has strict requirements for the execution time and needs to be completed within 24 hours. At the same time, the task has high resource requirements, but the failure rate must be kept at an extremely low level. According to the task requirements, predict that a higher weight needs to be given to the execution efficiency to ensure that the task is completed on time. At the same time, since the task has strict requirements for the failure rate, the weight of the task failure rate should also be appropriately increased. Calculate an adjustment factor for the execution efficiency target (increase from 0.4 to 0.5), and calculate an adjustment factor for the task failure rate target (increase from 0.2 to 0.3). Since the sum of the weights needs to remain 1, the weight of the resource utilization rate is correspondingly reduced to 0.2. Apply the first group and the second group of adjustment factors to the benchmark weight factors in the benchmark balance function for compensation. The compensated weight factors are: the execution efficiency is 0.5, the resource utilization rate is 0.2, and the task failure rate is 0.3. According to the compensated weight factors, reconstruct the task execution balance function for use in the subsequent task execution scheduling optimization process to guide the allocation and execution order of subtasks.

[0044] Step S400, with the subtask dependencies as constraints, perform execution scheduling optimization on the subtask set through the task execution balance function to establish multiple rounds of scheduling optimization results.

[0045] Specifically, according to the task execution balance function, scheduling strategies are selected, such as greedy algorithms, genetic algorithms, particle swarm optimization algorithms, etc. The sub-task dependencies are used as constraints in the scheduling process, and constraint handling logic is embedded in the scheduling algorithm to ensure that the generated scheduling plan does not violate the sub-task dependencies. Through multiple iterations, the execution order and resource allocation of sub-tasks are continuously adjusted to optimize the balance function value. After each iteration, a new scheduling plan is generated, and its corresponding balance function value is calculated. The results of each round of scheduling optimization are recorded, including the execution order of sub-tasks, the resource allocation plan, and the corresponding balance function value.

[0046] In a possible implementation, taking the sub-task dependencies as constraints, performing optimization for the execution scheduling of the sub-task set through the task execution balance function, and establishing multiple rounds of scheduling optimization results. Step S400 further includes step S410 of determining the order constraints and execution preconditions for task execution according to the sub-task dependencies. Specifically, the sub-task dependencies are parsed, including data flow dependencies, timing dependencies, resource competition dependencies, and collaborative optimization dependencies, etc. Based on the parsed sub-task dependencies, the order of task execution is determined. For example, if the output of task A is the input of task B, then task B must be executed after task A. In addition to the order constraints, the system also needs to analyze the execution preconditions, which include pre-requirements in aspects such as resource preparation, environment configuration, and data availability. For example, if task C requires specific computing resources that can only be used after being released by task D, then the execution of task C needs to wait for the completion of task D.

[0047] Step S420, while maintaining the order constraints and execution preconditions, perform adjustment of the execution strategy for the sub-task set, and calculate the balance score through the task execution balance function. Specifically, the system first sets an initial execution strategy, which is a simple scheduling plan based on sub-task dependencies, including task allocation, execution order, resource usage, etc. While maintaining the order constraints and execution preconditions, the system starts to iteratively adjust the execution strategy, including task parallelization, resource reallocation, and fine-tuning of the execution order, etc. For each execution strategy, the system uses the task execution balance function to calculate its balance score, which reflects the trade-off among execution efficiency, resource utilization rate, and task failure rate.

[0048] Step S430: Update the execution policy according to the balance score, adjust and iterate the execution policy, and establish the multi-round scheduling optimization results. Specifically, in each iteration, the system evaluates the balance score of the current execution policy and compares it with the previous optimal score. If the current score is better, the optimal execution policy is updated. The iteration process continues until the termination condition is met, such as reaching the preset number of iterations, the improvement of the balance score is less than a certain threshold, or the optimal solution that satisfies all constraint conditions is found. After the iteration terminates, the system saves the optimal execution policies generated during all iterations to form the multi-round scheduling optimization results, which can be used for subsequent analysis, decision-making, or execution. By strictly following the sequential constraints and execution pre-constraints, the system can ensure that tasks are executed in the correct order and conditions, avoiding conflicts and data inconsistencies between tasks. By iteratively adjusting the execution policy and calculating the balance score, the system can find the best balance point among multiple objectives, thereby optimizing the utilization of resources and the execution efficiency of tasks. The multi-round scheduling optimization results provide multiple possible execution plans, enabling the system to flexibly select the most suitable execution policy according to different environments and task requirements.

[0049] In a possible implementation manner, for the step of updating the execution policy according to the balance score in step S430, it further includes step S431: perform itemized fitness calculation, and the itemized fitness calculation includes execution efficiency calculation, resource utilization rate calculation, and task failure rate calculation. Specifically, collect data such as the execution time and waiting time of each subtask under the current execution policy. Based on the collected data, calculate the execution efficiency of each subtask, which is measured by the number of tasks completed per unit time. Aggregate the execution efficiencies of each subtask to obtain the execution efficiency of the entire task set.

[0050] Real-time monitor the usage of key resources such as CPU, memory, disk, and network. According to the monitoring data, calculate the utilization rate of each resource, which is measured by the ratio of the resource usage amount to the total resource amount. Synthesize the utilization rates of various resources to obtain a comprehensive evaluation result of the resource utilization rate.

[0051] Record the failure events that occur during the execution of each subtask, including the number of failures, failure reasons, etc. Based on the recorded data, calculate the failure rate of each subtask, which is measured by the ratio of the number of failures to the total number of executions. Aggregate the failure rates of each subtask to obtain the failure rate of the entire task set.

[0052] Step S432: Establish an update direction strategy based on the sub-item fitness calculation results. Specifically, comprehensively analyze the sub-item fitness such as execution efficiency, resource utilization rate, task failure rate, etc., to identify the advantages and disadvantages of the current execution strategy. Based on the analysis results, determine the optimization objectives, such as improving execution efficiency, reducing resource utilization rate, decreasing task failure rate, etc. According to the optimization objectives, formulate an updated direction strategy, including adjusting the task execution order, optimizing resource allocation, improving the task execution method, etc.

[0053] Step S433: Establish an update step size strategy based on the balance score, and perform an update search based on the update direction strategy and the update step size strategy to complete the policy adjustment iteration and establish the multi-round scheduling optimization results. Specifically, calculate the change in the balance score between the current execution strategy and the previous execution strategy to evaluate the effect of the policy adjustment. Based on the change in the balance score, determine the update step size, that is, the magnitude or range of change for each policy adjustment. Under the guidance of the update direction strategy and the update step size, perform an update search, that is, try different policy adjustment schemes and calculate their balance scores. Compare the balance scores of different policy adjustment schemes and select the scheme with the highest score as the execution strategy for the next round of iteration. Repeat steps S431 - S433 until the iteration termination condition is reached (such as the balance score no longer improves significantly, the number of iterations reaches the preset upper limit, etc.). This implementation method is based on the analysis results of the sub-item fitness, can determine the optimization objectives and directions, and provides clear guidance for policy adjustment. By calculating the change in the balance score and determining the update step size, it can control the complexity of the adjustment and the stability of the system while ensuring the effect of the policy adjustment.

[0054] Step S500: Predict the busy state of the environment, generate a scheduling compensation based on the busy state prediction result, and screen the multi-round scheduling optimization results through the scheduling compensation to establish an Agent decision-making scheme.

[0055] Specifically, historical data is used to train a model to predict the busy state of the environment in the future for a period of time, including predicting key indicators such as resource utilization rate and system response time. According to the prediction results of the busy state, a scheduling compensation strategy is generated, such as adjusting the execution time of subtasks, increasing resource allocation, etc., to cope with possible resource bottlenecks or system busyness in the future. The scheduling compensation is applied to the results of multi-round scheduling optimization. According to the task execution balance function value and the predicted future environment state, the scheduling plan that best conforms to the current and future environment states is selected. The selected optimal scheduling plan is used as the final Agent decision-making plan, and the decision-making plan is output to the AI Agent in a format such as a task list, Gantt chart or other easy-to-understand format to guide it to execute the target task. The embodiments of the present application adopt technical means such as calculating task complexity, disassembling tasks into a subtask set as needed and establishing dependencies, constructing a balance function based on environmental self-check and task requirements, performing scheduling optimization through the balance function, predicting the busy state of the environment and generating scheduling compensation, and screening the scheduling optimization results to form an Agent decision-making plan, achieving the technical effects of high efficiency of task execution, high utilization rate of resources, and significant reduction in task failure rate.

[0056] In a possible implementation manner, for the prediction of the busy state of the environment and the generation of scheduling compensation based on the prediction results of the busy state, step S500 further includes step S510 of performing prediction index principal component identification to determine key indicators. Specifically, various possible busy state-related index data are collected from the environmental monitoring system, including CPU utilization rate, memory occupancy rate, network bandwidth occupancy, disk I / O speed, etc. The collected data is cleaned to remove outliers and missing values to ensure the accuracy and integrity of the data. Using principal component analysis technology, a small number of principal components are extracted from multiple related indicators, and these principal components can maximize the reflection of the information of the original data. Through principal component analysis (PCA), it is determined which indicators are the most critical for predicting the busy state.

[0057] In step S520, real-time monitoring data and historical execution data are called, and regression analysis is performed through the key indicators, the real-time monitoring data, and the historical execution data to generate prediction results of the busy state. Specifically, the key indicators, real-time monitoring data, and historical execution data are integrated to form a complete data set. According to the needs of regression analysis, feature extraction and transformation are performed on the data to generate input features suitable for the regression model. According to the characteristics of the data and the prediction requirements, a regression model is selected, such as linear regression, polynomial regression, decision tree regression, etc. The regression model is trained using historical data, and the model parameters are adjusted to minimize the prediction error. The real-time monitoring data is input into the trained regression model to obtain the prediction results of the busy state.

[0058] Step S530, establish scheduling constraints based on the predicted busy state, and use the scheduling constraints as scheduling compensation for decision optimization. Specifically, according to the predicted busy state, set scheduling constraint conditions, such as restricting the execution priority of tasks, resource allocation, etc. under different degrees of busyness. On the basis of the existing scheduling strategy, introduce a scheduling compensation mechanism, and adjust the task execution plan according to the prediction results and scheduling constraints to optimize the overall performance and resource utilization rate. After considering the scheduling constraints and scheduling compensation, re-evaluate the task execution plan and select the optimal execution strategy to ensure that while meeting the task requirements, the execution efficiency and resource utilization rate are maximized. This implementation method accurately predicts the busy state of the environment through technical means such as principal component analysis and regression analysis, providing a basis for scheduling decisions.

[0059] In a possible implementation, after establishing the Agent decision-making scheme, the method further includes step S600 of performing execution monitoring on the Agent decision-making scheme and establishing execution feedback. Specifically, through a monitoring module embedded in the AI Agent, key index data of the Agent during the execution of the decision-making scheme are collected and recorded in real time, such as task execution progress, resource usage (CPU usage rate, memory occupancy, network bandwidth, etc.), task execution efficiency, task failure rate, etc. Compare and analyze the collected key index data with the expected goals in the decision-making scheme to form an execution feedback report, which details the deviations between the actual execution situation and the expected goals, as well as any abnormal situations or potential problems.

[0060] Step S700, determine whether the deviation between the execution feedback and the Agent decision-making scheme meets the expected deviation threshold. Specifically, use statistical methods, such as mean deviation, to calculate the deviation degree between the actual execution situation and the expected goals of the decision-making scheme. Compare the calculated deviation degree with the preset expected deviation threshold. The expected deviation threshold is the maximum acceptable deviation degree between the actual execution situation and the expected goals, which is comprehensively set according to factors such as historical data, task requirements, and environmental status, and is used to determine whether the actual execution situation is within the acceptable range.

[0061] Step S800: When the deviation between the execution feedback and the Agent decision-making scheme meets the expected deviation threshold, an automatic correction instruction is generated, and overclocking compensation for the Agent decision-making scheme is performed according to the automatic correction instruction. Specifically, when it is determined that the deviation between the actual execution situation and the decision-making scheme exceeds the expected threshold, the AI Agent automatically triggers a correction mechanism to generate a correction instruction. The correction instruction includes adjusting the task execution order, optimizing resource allocation, increasing or decreasing computing resources, etc. According to the correction instruction, overclocking compensation is performed on the decision-making scheme, that is, by means of increasing computing resources, optimizing algorithms, adjusting task priorities, etc., to accelerate the task execution progress or improve the task execution efficiency, so as to narrow the deviation between the actual execution situation and the expected target. Through this implementation method and the real-time monitoring and feedback mechanism, the AI Agent can timely detect and correct the deviations in the execution process, ensuring that the task can be successfully completed according to the expected target. At the same time, the overclocking compensation mechanism can provide additional computing resources or optimization means when necessary to cope with unexpected situations or improve the task execution efficiency. This not only improves the decision-making accuracy and execution efficiency of the AI Agent, but also enhances its ability to adapt to complex and changeable environments.

[0062] In a possible implementation manner, determining whether the deviation between the execution feedback and the Agent decision-making scheme meets the expected deviation threshold further includes step S900: When the deviation between the execution feedback and the Agent decision-making scheme does not meet the expected deviation threshold, a deviation accumulation is generated according to the deviation, and cumulative compensation for the Agent decision-making scheme is performed through the deviation accumulation.

[0063] Specifically, when the deviation between the execution feedback and the Agent decision-making scheme does not meet the expected deviation threshold, by comparing the difference between the actual execution result and the expected execution result, the deviation degree between the execution feedback and the Agent decision-making scheme is quantified. The difference can be measured by various indicators, such as execution efficiency, resource utilization rate, task completion degree, etc. For each calculated deviation degree, it is recorded in the deviation accumulation. The deviation accumulation can be a simple numerical accumulation or a more complex statistical model for tracking and analyzing the deviation trend. The deviation causes are deeply analyzed, including checking for abnormalities in the task execution process, resource allocation problems, environmental changes, etc. Based on the deviation accumulation and the analysis results, targeted cumulative compensation strategies are formulated, including adjusting the task execution order, optimizing resource allocation, updating the task execution strategy, etc. The formulated cumulative compensation strategies are applied to the Agent decision-making scheme to correct the previous deviations and reduce the deviations in future executions. Through this implementation method, by calculating the deviation degree, generating the deviation accumulation, analyzing the deviation causes, and formulating the cumulative compensation strategy, it is ensured that the Agent decision-making scheme can continuously maintain high efficiency and accuracy in actual execution, improving the stability and reliability of task execution.

[0064] The following is a specific example: Deploy the Prometheus + Grafana monitoring system to collect the following metrics in real time: task execution status, resource usage, and error logs. Among them, the task execution status includes the start / end timestamps of subtasks, the elapsed time, and the success / failure flag; the resource usage includes the CPU occupancy rate (%), memory occupancy (GB), disk I / O (MB / s), and network bandwidth (Mbps); the error logs include captured Python try-except exceptions and system-level errors (such as MemoryError). The data collection frequency is to sample resource metrics once per second and record immediately when the task status changes. The collected data is shown in Table 1.

[0065] Table 1: Execution Monitoring Data Sub-task Expected Duration Actual Duration CPU Peak Memory Peak Result Status Data Cleaning 60s 82s 75% 3.2GB Success Feature Extraction 120s 98s 60% 2.1GB Failure Model Training 300s 285s 90% 4.5GB Success The execution efficiency deviation = (actual elapsed time - expected elapsed time) / expected elapsed time * 100%, and a single task deviation > 20% triggers correction. The resource utilization deviation = (actual resource utilization rate - expected resource utilization rate) / expected resource utilization rate * 100%, and CPU > 85% or memory > 80% continuously for 1 minute triggers an alarm.

[0066] Examples of data calculation are as follows: For the data cleaning subtask, the deviation rate = (82s - 60s) / 60s * 100% = 36.67% > 20%, triggering correction. For the feature extraction subtask, it fails due to "ValueError: Incorrect input data format" and is directly marked as an exception.

[0067] The correction strategy library is shown in Table 2.

[0068] Table 2: Correction Strategy Library Deviation Type Correction Action Parameter Configuration Low Execution Efficiency Increase the Number of Parallel Threads Maximum Thread Number ≤ System Core Number × 2 High Resource Utilization Dynamically Adjust Resource Quota Increase CPU Quota by 20% and Memory Quota by 15% Task Failure Retry Mechanism + Data Verification Preprocessing Maximum Retry Times = 3, Verification Log Analysis Examples of correction execution are as follows: For data cleaning timeout, the automatic correction instruction is "python data_cleaner.py --threads = 4" (original number of threads = 2), and an additional 1.5GB of memory is allocated (original 3.2GB → 4.7GB).

[0069] For feature extraction failure, the automatic correction instruction is "preprocess_data.py --validate = True" (new data validation step), and the priority of this subtask is increased from level 5 to level 1.

[0070] The implementation logic of the cumulative compensation mechanism is as follows: The sliding window of 5 task execution records is the deviation statistical period. If Σ(efficiency deviation rate) > 15%, the execution efficiency weight is reduced by 10%. If Σ(resource peak) > threshold × 3, the resource utilization weight is increased by 15%. The recent 5 task execution records are shown in Table 3.

[0071] Table 3: Record of the Last 5 Task Executions Sub-task Expected Duration Actual Duration Efficiency Deviation Rate CPU Peak Memory Peak Result Status Data Cleaning ① 60s 78s +30% 75% 3.2GB Success Feature Extraction ① 120s 102s -15% 60% 2.1GB Success Model Training ① 300s 285s -5% 90% 4.5GB Success Data Cleaning ② 60s 85s +41.67% 80% 3.5GB Success Feature Extraction ② 120s 114s -5% 65% 2.3GB Success The calculation example of the cumulative deviation compensation is as follows: The cumulative efficiency deviation = Σ (efficiency deviation rate) = 30% - 15% - 5% + 41.67% - 5% = 56.67%. Since 56.67% > 15%, the execution efficiency weight is reduced by 10%. Calculate the peak resource deviation of each task (only CPU is counted in the example, and the expected CPU threshold is 60%): Data cleaning ①: 75% - 60% = +15%; Feature extraction ①: 60% - 60% = 0%; Model training ①: 90% - 60% = +30%; Data cleaning ②: 80% - 60% = +20%; Feature extraction ②: 65% - 60% = +5%. The cumulative resource utilization rate = Σ (peak resource deviation) = 15% + 0% + 30% + 20% + 5% = 70%. The allowable range of the resource deviation for a single task is ±10%, that is, ∑ (peak resource deviation) > (10% × 3) is the trigger condition. Since 70% > 30%, compensation is triggered, and the resource utilization weight is increased by 15%. The weight adjustment after compensation is shown in Table 4.

[0072] Table 4: Weight Adjustment after Compensation Target Original Weight Weight after Cumulative Compensation Execution Efficiency 0.5 0.5-0.1=0.4 Resource Utilization 0.2 0.2+0.15=0.35 Task Failure Rate 0.3 0.25 The resource allocation priority is adjusted as follows: Double the default memory limit for model training, enable dynamic overclocking mode for data processing (upper limit 95%), and establish a resource reservation pool (accounting for 10% of the total resources) to prevent sudden contention. The compensation effect is as follows: The resource utilization rate drops from an average of 72% to 63% (a decrease of 12.5%), and the average task completion time is shortened by 18% (from 12 minutes and 34 seconds to 10 minutes and 21 seconds).

[0073] In the above text, reference is made to Figure 1 The multi-source shared automated task collaborative optimization method of the AI Agent according to the embodiments of the present invention is described in detail. Next, reference will be made to Figure 2 Describe the multi-source shared automated task collaborative optimization system of the AI Agent according to the embodiments of the present invention.

[0074] The multi-source shared automated task collaborative optimization system of the AI Agent according to the embodiments of the present invention is used to solve the technical problems of low execution efficiency, uneven resource allocation, and rising task failure rate existing in the prior art, and achieve the technical effects of high efficiency in task execution, high utilization rate of resources, and significant reduction in task failure rate. The multi-source shared automated task collaborative optimization system of the AI Agent includes: a task complexity calculation module 10, a task decomposition module 20, a state self-check and task execution balance function establishment module 30, a sub-task execution scheduling optimization module 40, and an environment busy state prediction and scheduling compensation module 50.

[0075] The task complexity calculation module 10 is used to calculate the task complexity of the target task. When the task complexity calculation result meets the preset complexity threshold, a decomposition instruction is generated. The task decomposition module 20 is used to hierarchically disassemble the target task by using the decomposition instruction, generate a subtask set, and establish subtask dependencies. The status self-check and task execution balance function establishment module 30 is used to perform status self-check on the environmental status, and establish a task execution balance function according to the status self-check result and the task requirements of the target task. The balance objectives of the task execution balance function include an execution efficiency objective, a resource utilization rate objective, and a task failure rate objective. The subtask execution scheduling optimization module 40 is used to use the subtask dependencies as constraints, perform execution scheduling optimization on the subtask set through the task execution balance function, and establish multi-round scheduling optimization results. The environmental busy status prediction and scheduling compensation module 50 is used to predict the busy status of the environment, generate scheduling compensation based on the busy status prediction result, screen the multi-round scheduling optimization results through the scheduling compensation, and establish an Agent decision-making plan.

[0076] Next, the specific configuration of the task decomposition module 20 will be described in detail. As described above, the target task is hierarchically disassembled by using the decomposition instruction to generate a subtask set. The task decomposition module 20 may further include: a split task set establishment unit for disassembling the target task according to data flow, logical relationship, and resource occupation respectively, controlling the task granularity through a multi-layer task dependency graph, and establishing a split task set; a task similarity calculation unit for calculating the task similarity of the split task set and generating a subtask set according to the task similarity calculation result.

[0077] Among them, for establishing the subtask dependencies, the task decomposition module 20 may further include: a data flow dependency establishment unit for analyzing the input data of the tasks in the subtask set and establishing a data flow dependency according to the input data analysis result; a timing dependency establishment unit for analyzing the execution order of the subtask set and establishing a timing dependency according to the execution order analysis result; a resource competition dependency establishment unit for analyzing the tasks in the subtask set that share the same resources and establishing a resource competition dependency; a collaborative optimization dependency establishment unit for analyzing the influence of the execution mode under parallel execution on the subtask set and establishing a collaborative optimization dependency; a subtask dependency generation unit for generating subtask dependencies based on the data flow dependency, the timing dependency, the resource competition dependency, and the collaborative optimization dependency.

[0078] Next, the specific configuration of the status self-check and task execution balance function establishment module 30 will be described in detail. As described above, a task execution balance function is established based on the status self-check result and the task requirements of the target task. The status self-check and task execution balance function establishment module 30 may further include: a benchmark balance function construction unit for obtaining a historical database of collaborative task optimization, constructing a benchmark balance function according to the historical database, and the benchmark balance function is configured with a benchmark weight factor; a first set of adjustment factor establishment units for performing adaptation analysis of execution efficiency, resource utilization rate, and task failure rate using the status self-check result, and establishing a first set of adjustment factors according to the adaptation analysis result; a second set of adjustment factor establishment units for performing adaptation analysis of execution efficiency, resource utilization rate, and task failure rate using the task requirements, and establishing a second set of adjustment factors according to the adaptation analysis result; a task execution balance function establishment unit for compensating the benchmark weight factor using the first set of adjustment factors and the second set of adjustment factors, and establishing a task execution balance function.

[0079] Next, the specific configuration of the subtask execution scheduling optimization module 40 will be described in detail. As described above, with the subtask dependency as a constraint, the execution scheduling of the subtask set is optimized through the task execution balance function to establish multi-round scheduling optimization results. The subtask execution scheduling optimization module 40 may further include: a constraint determination unit for determining the order constraint and execution precondition constraint of task execution according to the subtask dependency; an execution strategy adjustment unit for performing the execution strategy adjustment of the subtask set while maintaining the order constraint and execution precondition constraint, and calculating a balance score through the task execution balance function; a multi-round scheduling optimization result establishment unit for updating the execution strategy according to the balance score, iterating the execution strategy adjustment, and establishing multi-round scheduling optimization results.

[0080] Among them, for the multi-round scheduling optimization result establishment unit that updates the execution strategy according to the balance score, it may further include: a sub-item fitness calculation subunit for performing sub-item fitness calculation, and the sub-item fitness calculation includes execution efficiency calculation, resource utilization rate calculation, and task failure rate calculation; an update direction strategy establishment subunit for establishing an update direction strategy according to the sub-item fitness calculation result; an update search subunit for establishing an update step size strategy according to the balance score, and performing an update search based on the update direction strategy and the update step size strategy to complete the strategy adjustment iteration and establish multi-round scheduling optimization results.

[0081] Next, the specific configuration of the environmental busy state prediction and scheduling compensation module 50 will be described in detail. As described above, the busy state of the environment is predicted, and scheduling compensation is generated based on the busy state prediction result. The environmental busy state prediction and scheduling compensation module 50 may further include: a key index determination unit for performing identification of the main components of the metrics for prediction and determining the key metrics; a regression analysis unit for calling real-time monitoring data and historical execution data, and performing regression analysis through the key metrics, the real-time monitoring data, and the historical execution data to generate a busy state prediction result; and a scheduling constraint establishment unit for establishing scheduling constraints based on the busy state prediction result and using the scheduling constraints as scheduling compensation for decision optimization.

[0082] Among them, after the Agent decision-making scheme is established, the system may further include: an execution monitoring module for monitoring the execution of the Agent decision-making scheme and establishing an execution feedback; a judgment module for judging whether the deviation between the execution feedback and the Agent decision-making scheme meets the expected deviation threshold; and an overclocking compensation module for generating an automatic correction instruction when the deviation between the execution feedback and the Agent decision-making scheme meets the expected deviation threshold and performing overclocking compensation for the Agent decision-making scheme according to the automatic correction instruction.

[0083] Among them, for judging whether the deviation between the execution feedback and the Agent decision-making scheme meets the expected deviation threshold, the system may further include: an accumulation compensation module for generating a deviation accumulation based on the deviation when the deviation between the execution feedback and the Agent decision-making scheme does not meet the expected deviation threshold, and performing accumulation compensation for the Agent decision-making scheme through the deviation accumulation.

[0084] The multi-source shared automated task collaborative optimization system of the AI Agent provided by the embodiments of the present invention can execute the multi-source shared automated task collaborative optimization method of the AI Agent provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0085] Although various references are made to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or the server. The included units and modules are only divided according to functional logic, but are not limited to the above 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 mutual distinction and do not limit the protection scope of the present invention.

[0086] The above specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present application shall be included within the protection scope of the present application. In some cases, the actions or steps recited in the present application can be executed in an order different from that in the embodiments and still achieve the desired results. Additionally, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A multi-source sharing automated task collaborative optimization method for an AI agent, characterized in that The method includes: Calculating the task complexity of the target task. When the calculation result of the task complexity meets the preset complexity threshold, a decomposition instruction is generated; Using the decomposition instruction to hierarchically decompose the target task, generating a subtask set, and establishing subtask dependencies; Performing a status self-check on the environmental state, and establishing a task execution balance function according to the status self-check result and the task requirements of the target task. The balance objectives of the task execution balance function include an execution efficiency objective, a resource utilization rate objective, and a task failure rate objective; Taking the subtask dependencies as constraints, performing an execution scheduling optimization of the subtask set through the task execution balance function, and establishing a multi-round scheduling optimization result; Predicting the busy state of the environment, generating a scheduling compensation based on the busy state prediction result, and screening the multi-round scheduling optimization result through the scheduling compensation to establish an Agent decision-making scheme.

2. The multi-source sharing automated task collaborative optimization method of the AI Agent according to claim 1, wherein The using the decomposition instruction to hierarchically decompose the target task and generating a subtask set includes: Decomposing the target task according to data flow, logical relationship, and resource occupancy respectively, controlling the task granularity through a multi-layer task dependency graph, and establishing a set of split tasks; Calculating the task similarity of the set of split tasks, and generating a subtask set according to the calculation result of the task similarity.

3. The multi-source sharing automated task collaborative optimization method of the AI Agent according to claim 2, wherein, The establishing subtask dependencies includes: Performing input data analysis of the tasks in the subtask set, and establishing a data flow dependency according to the input data analysis result; Performing an execution order analysis of the subtask set, and establishing a timing dependency according to the execution order analysis result; Performing an analysis of tasks in the subtask set sharing the same resources, and establishing a resource competition dependency; Performing an analysis of the impact of the execution mode under parallel execution on the subtask set, and establishing a collaborative optimization dependency; Generating subtask dependencies based on the data flow dependency, the timing dependency, the resource competition dependency, and the collaborative optimization dependency.

4. The multi-source sharing automated task collaborative optimization method for an AI Agent according to claim 1, wherein The establishing a task execution balance function according to the status self-check result and the task requirements of the target task includes: Obtaining a historical database for collaborative task optimization, constructing a benchmark balance function according to the historical database, and the benchmark balance function is configured with a benchmark weight factor; Using the status self-check result to perform an adaptation analysis of execution efficiency, resource utilization rate, and task failure rate, and establishing a first set of adjustment factors according to the adaptation analysis result; Using the task requirements to perform an adaptation analysis of execution efficiency, resource utilization rate, and task failure rate, and establishing a second set of adjustment factors according to the adaptation analysis result; After compensating the benchmark weight factor by using the first set of adjustment factors and the second set of adjustment factors, a task execution balance function is established.

5. The multi-source sharing automated task collaborative optimization method of the AI Agent according to claim 1, characterized in that The taking the subtask dependencies as constraints and performing an execution scheduling optimization of the subtask set through the task execution balance function to establish a multi-round scheduling optimization result includes: Determining the order constraint and execution precondition constraint of task execution according to the subtask dependencies; While maintaining the order constraint and execution precondition constraint, performing an adjustment of the execution strategy of the subtask set, and calculating a balance score through the task execution balance function; Update the execution policy according to the balance score, adjust and iterate the execution policy, and establish the multi-round scheduling optimization results.

6. The multi-source sharing automated task collaborative optimization method of the AI Agent according to claim 5, characterized in that, The updating of the execution policy according to the balance score includes: Perform sub-item fitness calculation, where the sub-item fitness calculation includes execution efficiency calculation, resource utilization rate calculation, and task failure rate calculation; Establish an update direction policy based on the sub-item fitness calculation results; Establish an update step size policy according to the balance score, and perform update search based on the update direction policy and the update step size policy to complete the policy adjustment iteration and establish the multi-round scheduling optimization results.

7. The multi-source sharing automated task collaborative optimization method of the AI Agent according to claim 1, characterized in that, Predict the busy state of the environment and generate scheduling compensation based on the busy state prediction results, including: Perform the identification of the main components of the prediction indicators to determine the key indicators; Call the real-time monitoring data and historical execution data, and perform regression analysis through the key indicators, the real-time monitoring data, and the historical execution data to generate the busy state prediction results; Establish scheduling constraints according to the busy state prediction results, and use the scheduling constraints as scheduling compensation for decision optimization.

8. The multi-source shared automated task collaborative optimization method for the AI Agent according to claim 1, wherein After establishing the Agent decision-making plan, it includes: Monitor the execution of the Agent decision-making plan and establish execution feedback; Judge whether the deviation between the execution feedback and the Agent decision-making plan meets the expected deviation threshold; When the deviation between the execution feedback and the Agent decision-making plan meets the expected deviation threshold, generate an automatic correction instruction and perform overclock compensation for the Agent decision-making plan according to the automatic correction instruction.

9. The multi-source sharing automated task collaborative optimization method for an AI Agent according to claim 8, characterized in that, The judgment of whether the deviation between the execution feedback and the Agent decision-making plan meets the expected deviation threshold includes: When the deviation between the execution feedback and the Agent decision-making plan does not meet the expected deviation threshold, generate a deviation accumulation according to the deviation, and perform cumulative compensation for the Agent decision-making plan through the deviation accumulation.

10. A multi-source shared automated task collaborative optimization system for an AI Agent, characterized in that, The system is used to implement the multi-source sharing automated task collaborative optimization method of the AI Agent according to any one of claims 1-9. The system includes: A task complexity calculation module for calculating the task complexity of the target task. When the task complexity calculation result meets the preset complexity threshold, a decomposition instruction is generated; A task decomposition module for hierarchically disassembling the target task using the decomposition instruction to generate a sub-task set and establish sub-task dependencies; A state self-check and task execution balance function establishment module for performing state self-check on the environmental state, and establishing a task execution balance function according to the state self-check results and the task requirements of the target task. The balance objectives of the task execution balance function include an execution efficiency objective, a resource utilization rate objective, and a task failure rate objective; A sub-task execution scheduling optimization module for performing execution scheduling optimization of the sub-task set through the task execution balance function with the sub-task dependencies as constraints, and establishing multi-round scheduling optimization results; An environmental busy state prediction and scheduling compensation module for predicting the busy state of the environment, generating scheduling compensation based on the busy state prediction results, screening the multi-round scheduling optimization results through the scheduling compensation, and establishing an Agent decision-making plan.

Images