CN-122019098-A - Intelligent task scheduling and planning method

CN122019098ACN 122019098 ACN122019098 ACN 122019098ACN-122019098-A

Abstract

The application discloses an agent task scheduling planning method, and relates to the technical field of agent scheduling. The method comprises the steps of analyzing task instructions to generate atomic tasks capable of being independently executed, constructing sub-task dependency graphs, defining inter-task association constraint, determining real-time resource occupation states of agents to obtain resource state tensors, completing resource-task association anchoring and dependency priority ordering in combination with the sub-task dependency graphs to generate task priority sequences with resource constraint, carrying out dynamic capacity matching and predictive load balancing calculation on the sequences through a task-agent adaptation model, determining target execution agents of each atomic task, and carrying out time sequence scheduling arrangement and conflict resolution based on the target execution agents and the sub-task dependency graphs to generate a collaborative execution scheme. The method improves the rationality and the high efficiency of the agent task scheduling, reduces the risks of resource conflict and execution timeout, and is suitable for agent cluster cooperative scheduling in complex scenes.

Inventors

WANG JIAYING
LI HUAWEI
LI HAIYANG
FAN ZHICHAO
XU JINBO
WU GUANGPENG

Assignees

北京甲板智慧科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (10)

1. The intelligent agent task scheduling and planning method is characterized by comprising the following steps of: step 1, analyzing task instructions to generate an atomic task which can be independently executed so as to construct a subtask dependency graph; step 2, determining the real-time resource occupation state of the agent to obtain an agent resource state tensor, and performing resource-task association anchoring and dependency priority sorting processing based on the agent resource state tensor and the subtask dependency graph to generate a task priority sequence with resource constraint; step 3, performing dynamic capacity matching and predictive load balancing calculation processing on the task priority sequence with resource constraint through a task-agent adaptation model so as to determine target execution agents of each atomic task; and 4, performing time sequence scheduling and conflict resolution processing on the target execution agent and subtask dependency graphs of each atomic task to generate an agent task scheduling cooperative execution scheme.
2. The agent task scheduling and planning method according to claim 1, wherein step 1 includes: step 11, carrying out semantic deconstructing and core requirement extraction on the task instruction to generate a task core element set; step 12, performing granularity splitting and independent executability judgment on the task core element set to generate an atomic task capable of being independently executed; and 13, performing logic association analysis and dependency modeling on the atomic task to construct a subtask dependency graph.
3. The agent task scheduling and planning method according to claim 1, wherein step 2 includes: step 21, monitoring the calculation power, storage and communication links of each intelligent agent in real time to generate intelligent agent resource monitoring data; Step 22, tensor modeling is carried out on the intelligent agent resource monitoring data to obtain intelligent agent resource state tensor; Step 23, performing resource-task association anchoring based on the agent resource state tensor and the subtask dependency graph to generate a resource adaptation task set; And step 24, carrying out dependency priority sequencing on the resource adaptation task set to generate a task priority sequence with resource constraint.
4. The agent task scheduling and planning method according to claim 1, wherein step3 includes: Step 31, constructing a task-agent adaptation model, wherein the task-agent adaptation model is obtained by fusing a dynamic capacity evolution network, a load prediction feedback loop and a task type adaptation rule base; step 32, inputting a task priority sequence with resource constraint and an agent resource state tensor into a task-agent adaptation model, outputting an agent-task dynamic matching matrix through a dynamic capability evolution network, and carrying out matching verification by combining a task type adaptation rule base so as to determine a capability matching agent candidate set; And 33, based on the agent predicted load value output by the load prediction feedback loop, fusing the type weight of the task type adaptation rule base, carrying out dynamic load balancing calculation on the capability matching agent candidate set, and screening out target execution agents of each atomic task.
5. The agent task scheduling and planning method according to claim 4, wherein step 31 includes: step 311, collecting historical execution task data, real-time resource state tensors and task type adaptation records of each agent, extracting time sequence characteristics of the historical execution task data to obtain a capacity evolution trend vector, carrying out residual fusion on the capacity evolution trend vector and the real-time resource state tensors, and learning adaptation weights of different task types through an attention mechanism to generate a dynamic capacity evolution network.
6. The agent task scheduling and planning method according to claim 4, wherein the step 32 includes: Step 321, extracting type characteristics and labeling priority weights of atomic tasks in a task priority sequence with resource constraints, and querying a task type adaptation rule base to obtain intelligent capability adaptation dimensions and thresholds corresponding to the types so as to generate task characteristic vectors with weights and adaptation constraints; Step 322, inputting the task feature vector with weight and adaptation constraint into a dynamic capability evolution network, and performing association calculation with the intelligent agent dynamic capability feature stored in the network to generate an intelligent agent-task dynamic matching matrix containing time sequence adaptation coefficients; step 323, calling a task type adaptation rule base, performing adaptation threshold checking on the intelligent agent-task dynamic matching matrix, eliminating intelligent agents which do not meet the type adaptation rule, and screening intelligent agents with time sequence adaptation coefficients meeting standards to generate capability matching intelligent agent candidate sets.
7. The agent task scheduling and planning method according to claim 4, wherein the step 33 includes: Step 331, inputting real-time load values of the agents into a load prediction feedback loop, outputting predicted load values of the agents in a future preset period, and calculating load increment values after the execution of the tasks by combining resource demand characteristics of the atomic tasks and type load coefficients of a task type adaptation rule base; Step 332, calculating a load balancing coefficient of each intelligent agent based on the predicted load value, the load increment value and the type adaptation weight of the task type adaptation rule base, wherein the load balancing coefficient is obtained by fusing the deviation rate of the predicted load value and the load threshold, the task priority weight, the intelligent agent capability redundancy and the type adaptation weight; step 333, sorting the capacity matching agent candidate sets from high to low according to the load balancing coefficient, and distributing the first-order agent to each atomic task to determine the target execution agent of each atomic task.
8. The agent task scheduling and planning method according to claim 1, wherein step 4 includes: step 41, combining the subtask dependency graph with the target execution agent of each atomic task, and performing initial time sequence scheduling to generate an initial scheduling scheme; step 42, performing conflict detection on the initial scheduling scheme to generate task conflict items; And 43, resolving and optimizing task conflict items to generate an agent task scheduling collaborative execution scheme.
9. The agent task scheduling and planning method of claim 8, wherein step 41 comprises: step 411, determining the execution sequence of the atomic task according to the subtask dependency graph to generate a task time sequence; step 412, combining the task time sequence and the target execution agent, and allocating the task execution time period of each agent to generate a time period allocation table; step 413, integrating the task timing sequence and the time slot allocation table to generate an initial scheduling scheme.
10. The agent task scheduling and planning method of claim 8, wherein step 42 includes: Step 421, performing overlapping detection on task execution time periods of the same agent in the initial scheduling scheme to generate time period conflict items; Step 422, logically detecting the execution sequence of the atomic task with the dependency relationship in the initial scheduling scheme to generate a logical conflict item; Step 423, generating a task conflict item by integrating the period conflict item and the logic conflict item.

Description

Intelligent task scheduling and planning method Technical Field The application relates to the technical field of agent scheduling, in particular to an agent task scheduling planning method. Background Under the background of rapid development of artificial intelligence and distributed computing technology, an intelligent system is widely applied to a plurality of fields such as cloud computing, industrial automation, automatic driving and the like, and is used for carrying out and cooperating complex compound tasks. Such composite tasks generally comprise a plurality of associated sub-tasks, and have strict requirements on execution timing, resource allocation and reliability, so that efficient task scheduling planning becomes a core for ensuring stable operation of the intelligent agent system. In the prior art, a typical agent task scheduling scheme simply disassembles task instructions, generates independent task units, and then performs task allocation and timing arrangement based on static resource states. The scheme includes the steps of firstly carrying out basic analysis on task instructions, extracting core task elements and splitting the core task elements into executable units, then collecting real-time resource data of an intelligent agent, establishing a matching relation between resources and tasks, then distributing an execution main body according to task priorities and intelligent agent capabilities, and finally generating and executing a scheduling scheme through time sequence sequencing and conflict detection. However, the prior art has obvious technical defects in practical application that the prior art only carries out simple disassembly and initial modeling on the task, does not carry out effective verification on the closed loop and the integrity of the dependency, is easy to cause deadlock of the task due to cyclic dependence, is difficult to find implicit dependence conflict, carries out static matching on the basis of a real-time state, cannot predict a peak of future resource demand, causes overtime execution of a high-priority task due to resource competition, has low resource utilization rate, and lacks a dynamic adjustment mechanism after the scheduling scheme is generated, cannot respond quickly when the task is executed abnormally, needs manual intervention to re-plan, and seriously affects the task completion efficiency and the system stability. Disclosure of Invention In order to solve the above technical problems, the present application provides an agent task scheduling planning method, so as to at least alleviate the above technical problems. The technical scheme provided by the embodiment of the application is as follows: A scheduling planning method for an agent task comprises the steps of 1, analyzing task instructions to generate an atomic task capable of being independently executed to construct a subtask dependency graph, 2, determining real-time resource occupation states of the agent to obtain an agent resource state tensor, performing resource-task association anchoring and dependency priority sequencing processing based on the agent resource state tensor and the subtask dependency graph to generate a task priority sequence with resource constraint, 3, performing dynamic capacity matching and predictive load balancing calculation processing on the task priority sequence with the resource constraint through a task-agent adaptation model to determine target execution agents of all atomic tasks, and 4, performing time sequence scheduling and conflict resolution processing on the target execution agents and the subtask dependency graph of all the atomic tasks to generate an agent task scheduling collaborative execution scheme. Aiming at the technical defects of dependency verification deficiency, resource allocation staticization, insufficient scheduling adaptability and the like in the prior art, the intelligent agent task scheduling planning method provided by the application realizes the improvement of the high efficiency and the robustness of task scheduling through multi-link collaborative design, and has the following specific technical advantages: In the task analysis and dependency modeling link, the method generates an atomic task which can be independently executed and constructs a subtask dependency graph through a complete flow of semantic deconstructing, granularity splitting and dependency modeling. Compared with a simple task disassembly mode in the prior art, the method can systematically comb the association logic among atomic tasks by executing the pre-condition extraction, the dependency pair matching and the mapping modeling, clearly present the sequential constraint relation of task execution, provide accurate dependency basis for subsequent scheduling, and effectively reduce the execution confusion caused by fuzzy dependency relation. Meanwhile, through fine processing such as word segmentation, syntax analysis and redundancy elimination, the