Heterogeneous agent hierarchical planning method for complex task decomposition

Abstract

The invention relates to the technical field of agent task planning and cooperation, discloses a heterogeneous agent hierarchical planning method for complex task decomposition, and belongs to the technical field of agent task planning and cooperation. According to the method, through a task decomposition and layering module, a sub-task distribution and scheduling module, a parallel execution and feedback module and a dynamic adjustment and optimization module, a heterogeneous agent system can decompose a complex task into multiple layers of sub-tasks, and parallel execution of the task is achieved through optimal distribution and scheduling. And the intelligent agent dynamically adjusts task allocation and scheduling according to task feedback to ensure efficient operation of the system. According to the method, the task decomposition and execution process is optimized based on a hierarchical planning structure, and the method is suitable for heterogeneous agent cooperation of complex task scenes such as a robot cluster and an unmanned aerial vehicle group.

Description

Technical Field

[0001] The present invention relates to the technical field of agent task planning and collaboration, and in particular to a heterogeneous agent hierarchical planning method for complex task decomposition. Background Art

[0002] The information provided in this section is only background information related to the present disclosure and is not necessarily prior art.

[0003] In a heterogeneous agent system, the collaboration and cooperation among agents are crucial for completing complex tasks. These agents may have different capabilities and resources, and reasonable task decomposition and allocation are particularly important. Traditional single-level task decomposition methods usually directly decompose complex tasks into several subtasks and allocate these subtasks to each agent. However, when facing multi-stage tasks, this method often has the following problems.

[0004] Single-level task decomposition easily leads to a lack of coherence in the execution process among subtasks. Due to the lack of context coordination, the execution of some subtasks may be affected by the progress of other subtasks, resulting in interruptions or delays in the overall process. This uncoordinated execution may cause some agents to be idle while waiting for other agents to complete related tasks, thus reducing the efficiency of the overall system.

[0005] The limitations of traditional methods make it difficult to perform tasks in parallel. In multi-stage tasks, some subtasks may depend on the results of other subtasks, and without effective strategies to manage these dependencies, the collaboration among agents will be affected, resulting in the inability to perform tasks efficiently in parallel.

[0006] Single-level task decomposition lacks flexibility and is difficult to adapt to dynamic environments and task requirements. In practical applications, environmental factors and task conditions often change, and traditional task decomposition methods cannot quickly adjust task allocation strategies, thus affecting the adaptability of the agent system in complex situations. Summary of the Invention

[0007] Aiming at the deficiencies of the prior art, the present invention provides a heterogeneous agent hierarchical planning method for complex task decomposition, enabling the agent system to efficiently complete complex multi-stage tasks through a multi-level task decomposition mechanism, realizing parallel task execution, and optimizing task allocation through a hierarchical planning structure to improve task collaboration efficiency.

[0008] The technical solution for achieving the object of the present invention is: A heterogeneous agent hierarchical planning method for complex task decomposition:

[0009] Step 1: Task Decomposition and Layering. Convert complex tasks into multiple smaller subtasks to form a manageable hierarchical structure for step-by-step execution and parallel processing.

[0010] Step 1-1: Task Decomposition. Decompose the complex task T into several subtasks to form a task decomposition set T1, T2, …, T n , in the task decomposition, the division of subtasks can be based on the functional characteristics of the task, and the decomposition method of task t is:

[0011] T = {T1, T2, … T i … T j …, T n},

[0012] where T i represents the subtask of the i-th layer, and the task decomposition process is carried out layer by layer through the method of recursive decomposition.

[0013] Furthermore, in Step 1-1, the decomposition of task T is specifically based on functional characteristics, and it can be divided into a perception subtask T1, a planning subtask T2, an execution subtask T3, a task allocation subtask T4, a coordination subtask T5, and an evaluation and feedback subtask T6, a total of 6 subtasks.

[0014] Among them, the perception subtask T1 includes the following subtasks: a data collection subtask responsible for collecting real-time data from sensors and other devices; a data preprocessing subtask responsible for performing data cleaning, denoising, and standardization processing; a target recognition and tracking subtask T 13 , responsible for identifying and tracking the state of the target; a situation assessment subtask T 14 , responsible for evaluating the combat environment and situation.

[0015] The planning subtask T2 includes the following subtasks: a task decomposition subtask responsible for converting high-level tasks into specific subtasks; a path planning subtask responsible for calculating the optimal path or action route; a resource scheduling subtask responsible for scheduling the resources required for the task; a situation prediction and adjustment subtask responsible for adjusting the task plan based on the feedback of the perception subtask.

[0016] The execution subtask T3 includes the following subtasks: a task execution subtask responsible for executing the task according to the instructions of the planning subtask; a task monitoring subtask responsible for real-time monitoring of the task progress and status; a fault recovery subtask responsible for handling abnormal situations during task execution.

[0017] The task assignment subtask T4 includes the following subtasks: the resource allocation subtask Responsible for task assignment according to task requirements, agent capabilities, and resource availability; the load balancing subtask Responsible for optimizing task assignment to balance the loads of each agent; the priority sorting subtask Responsible for sorting and assigning tasks according to task priorities.

[0018] The cooperation subtask T5 includes the following subtasks: the information sharing subtask Responsible for ensuring information sharing and circulation among agents; the task coordination subtask Responsible for coordinating the task execution of different agents; the cooperation feedback subtask T 53 , responsible for collecting the execution status and adjusting the cooperation strategy.

[0019] The evaluation and feedback subtask T6 includes the following subtasks: the performance evaluation subtask Responsible for evaluating the task execution effect; the feedback analysis subtask Responsible for analyzing the problems and bottlenecks in the execution; the optimization and adjustment subtask Responsible for adjusting task scheduling and resource allocation based on the feedback.

[0020] Steps 1-2: Hierarchical structure construction. Construct the hierarchical structure of tasks according to the task dependencies. The subtasks in each layer have specific dependencies with the tasks in other layers, and the modeling of the dependencies is represented by a matrix.

[0021] Set the dependency matrix of the tasks as D, where D ij represents whether task T i depends on task T j . If T i depends on T j , then D ij =1, otherwise D ij =0; D ii =0, i = 1, 2, 3,..., n;

[0022] The task hierarchical structure is:

[0023] L = {L1, L2,…, L i …, L n};

[0024] where L i represents the task set of the i-th layer, satisfying the sorting relationship of task T i in the hierarchical structure.

[0025] Step 1-3: Phase Dependency Analysis. The dependency analysis of tasks provides a basis for the scheduling of subsequent tasks. According to the dependency relationships between tasks, a task dependency matrix D is formed to represent the sequential execution constraints between tasks.

[0026] The dependency matrix D can be constructed in the following way:

[0027]

[0028] where D ij represents the relationship that task T i depends on task T j .

[0029] Step 2: Subtask Allocation and Scheduling. Allocate tasks at each layer to agents, and optimize the scheduling and resource allocation of subtasks to achieve parallel execution.

[0030] Step 2-1: Subtask Allocation Model. The system allocates tasks according to the capabilities of agents, task requirements, and their resource availability. Each subtask T i needs to be allocated to a suitable agent A j .

[0031] Establish a task allocation function based on the skill matching degree, availability, and current load of agents. The allocation function P ij represents the probability or weight that task T i is allocated to agent A j , and the calculation method is as follows:

[0032] P ij = w1·skill(A j ,T i ) + w2·availability(A j ) + w3·load(A j );

[0033] where skill(A j ,T i ) represents the skill matching degree of agent A j executing task T i , availability(A j ) represents the availability of agent A j , load(A j ) represents the current task load of agent A j , and w1, w2, w3 are the weights of each factor, which are determined by tuning.

[0034] Step 2-2: Scheduling Optimization. After the subtask allocation is completed, the goal of scheduling optimization is to minimize the total execution time. Assume that all tasks have a completion time function completion_time(T i ), then the goal of scheduling optimization is:

[0035]

[0036] During the scheduling process, the dependency relationship between tasks needs to be considered to ensure that the task execution order meets the prerequisite conditions. The scheduling constraints can be represented by the following inequalities:

[0037] completion_time(T i ) ≥ completion_time(T j ) if D ij == 1;

[0038] It means that the completion time of task T i cannot be earlier than that of task T j , that is, task T j is completed before T i .

[0039] Step 3: Parallel Execution and Feedback. After the allocation and scheduling are completed, the agent executes the subtasks in parallel according to the task hierarchy and performs state feedback during the execution process.

[0040] Step 3-1: Parallel Execution. After the scheduling optimization is completed, each agent executes its respective subtasks in parallel according to the task allocation result. Assume that the execution time of each task is Then the execution time of all tasks during parallel execution can be determined by the maximum value:

[0041]

[0042] Step 3-2: State Feedback. During the task execution process, the agent will provide real-time feedback on the task execution status, including information such as task progress, task completion, time consumption, computing resource consumption, task execution exception handling, environmental change feedback, and resource consumption information. The feedback status of each task can be represented as a vector:

[0043] F = {f1, f2, …, f n};

[0044] where f i represents the feedback status of task T i .

[0045] Step 3-3: Task Completion Confirmation. This step is crucial to ensure that the task is completed according to the predefined goals and standards. The system confirms the task completion status based on the feedback data and activates subsequent tasks according to the task dependencies. Let the completion status of task T i be C(T i ). Then the task completion confirmation can be expressed as:

[0046]

[0047] Task completion confirmation is achieved by continuously tracking the three types of information in the execution status feedback: "task progress, task completion status, and task execution exception handling".

[0048] Task progress judgment: If the task progress reaches 100%, then the corresponding f i = completed.

[0049] Task completion status judgment: The execution of a task not only considers whether the progress reaches 100%, but also whether the task meets the predefined goals and whether all subtasks or operations in the plan are successfully completed. If all subtasks are completed and meet the predefined goals, then the task completion status reaches 100%, and the feedback f i = completed.

[0050] Task execution exception handling: During the task execution, the task may not be completed, and then the exception situation is feedback. If the task cannot continue to execute due to resource problems, the feedback f i = failed; if the task times out, the feedback f i = timeout; if the task becomes unexecutable due to environmental changes, the feedback f i = interrupted.

[0051] Step 3-4: Task Activation or Adjustment. When the current subtask is completed without exceptions, the system determines according to the dependency matrix D whether there are other tasks T j depending on the current subtask T i . If D ij = 1 and f i = completed, then task T j is activated.

[0052] When a task fails, times out, etc., it may block subsequent tasks that depend on it. The system needs to adjust or reschedule the execution of the dependent tasks.

[0053] Task blockage confirmation: If task T i is not completed or fails, and the task T i depending on task T jwill be blocked until task T i is successfully completed.

[0054] Rescheduling and task adjustment: The system will adjust task T j according to the feedback information, including adjusting the priority and reallocating resources.

[0055] Step 4: Dynamic adjustment and optimization. During the task execution, the system dynamically adjusts task allocation and scheduling based on the status feedback to improve the system response efficiency.

[0056] Step 4-1: Task adjustment. During the task execution, if an exception occurs or the task dependency changes, the system needs to reallocate tasks or adjust task priorities. Task adjustment can be carried out through the following formula:

[0057]

[0058] where Adjust(T i , D, F) represents adjusting task T i according to the current task status F and dependency D.

[0059] Step 4-2: Allocation optimization. Dynamically adjust task allocation according to the feedback data to optimize resource utilization. Assume that the load situation of each agent is represented by L(A j ). The new task allocation scheme can be optimized through the following formula:

[0060]

[0061] where represents the ratio of the load of agent A j to its maximum load capacity. The values of the weight coefficients w1, w2, and w3 in Step 4-2 are the same as those in Step 2.

[0062] Step 4-3: System optimization. During the task execution, the core goal of system optimization is to minimize the total execution time, balance the load of each agent, improve resource utilization, and ensure that the dependencies between tasks are satisfied. To achieve this goal, system optimization comprehensively considers four aspects: execution time, load balance, task dependencies, and resource utilization. The specific optimization goals and methods are as follows:

[0063] Minimization of total execution time: The main goal of the system is to minimize the total execution time of all tasks, that is, to ensure that tasks are completed in the shortest time. The total execution time T total is defined as the maximum time when all tasks are executed, and the optimization goal is to minimize this value:

[0064]

[0065] where \(I\) k represents the task set assigned to the \(k\)-th agent, and execution_time(\(T\) i ) represents the execution time of task \(T\) i . By adjusting the task assignment and scheduling order, the overall maximum execution time is reduced.

[0066] Load balancing optimization: The load balancing of agents is another optimization goal, aiming to ensure the workload balance of each agent to avoid resource waste or low execution efficiency caused by some agents being overloaded. The load of each agent \(A\) j load(\(A\) j ) represents the sum of the execution times of the tasks it executes. The system needs to minimize the variance of the loads of each agent. The goal of load balancing optimization can be expressed as:

[0067]

[0068] where \(m\) is the number of agents, and load(\(A\) j ) is the load of the \(j\)-th agent. The optimization process makes the workloads of each agent as close as possible by reducing the variance between the loads.

[0069] Task dependency constraint: When executing tasks, the dependency relationship between tasks must be considered to ensure that prerequisite tasks are completed before subsequent tasks. Suppose task \(T\) i depends on task \(T\) j , that is, \(D\) ij = 1, then the completion time of task \(T\) i cannot be earlier than the completion time of task \(T\) j . The dependency constraint can be expressed as:

[0070]

[0071] where completion_time(\(T\) i ) represents the completion time of task \(T\) i , duration(\(T_j\)) represents the execution time of task \(T\) j , and \(D\) ij represents whether task \(T\) i depends on task \(T\) j . This constraint ensures that the execution order of tasks conforms to the dependency relationship.

[0072] Maximizing resource utilization, to improve the overall efficiency of the system, the resource utilization of agents also needs to be maximized. The resource utilization of agents can be improved by reducing the idle time. The idle time \(T\) idle (\(A\) j) Represents Agent A j The idle duration of Agent A during task execution. The optimization objective is to minimize the total idle time:

[0073]

[0074] Where T idle (A j ) is the idle time of Agent A j . This objective ensures that the execution ability of the agent is fully utilized, thereby improving the resource utilization efficiency.

[0075] Combining the optimization objectives, considering the above optimization objectives, the comprehensive optimization problem of the system can be expressed as:

[0076]

[0077] Where: T idle is the total execution time, load(A j ) is the load of Agent A j , and T idle (A j ) is the idle time of Agent A j .

[0078] Compared with the prior art, the present invention has the following beneficial effects:

[0079] Through the hierarchical planning method of heterogeneous agents, the present invention realizes the efficient decomposition and collaborative execution of complex tasks, can adapt to dynamic and changeable task scenarios, and significantly improves the task response speed and execution efficiency. Compared with the traditional manual scheduling and single-task allocation methods, the present invention combines task hierarchical decomposition and optimization scheduling mechanisms, effectively improving the resource utilization rate and task completion accuracy. By introducing a multi-objective optimization model and task dependency constraint analysis, it ensures that task scheduling can still be efficiently completed under complex dependency conditions, reducing the system operation cost. In addition, the real-time feedback and dynamic adjustment mechanism can quickly respond to unexpected situations during the execution process, enhancing the adaptability and robustness of the system. The present invention is applicable to various complex task scenarios, such as unmanned system cooperation and tactical operation planning, and has significant application value and promotion potential.

[0080] The multi-level task decomposition and planning mechanism can structure complex tasks into hierarchical subtasks, thereby realizing the efficient division of labor and collaborative execution of the agent system. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] The following further specifically describes the present invention in conjunction with the drawings and specific embodiments, and the above and / or other advantages of the present invention will become clearer.

[0082] Figure 1 Is the overall flowchart of this solution.

[0083] Figure 2 This is the task decomposition and hierarchical flow chart of this solution.

[0084] Figure 3 This is the multi-task dynamic adjustment and optimization of this solution. Detailed implementation manners

[0085] As Figure 1 shown, a heterogeneous agent hierarchical planning method for complex task decomposition includes the following steps:

[0086] Step 1: Task decomposition and layering, converting a complex task into multiple smaller subtasks to form a manageable hierarchical structure for step-by-step execution and parallel processing.

[0087] Step 2: Subtask assignment and scheduling, assigning the tasks at each layer obtained in Step 1 to agents, optimizing the scheduling and resource allocation of subtasks, and achieving parallel execution.

[0088] Step 3: Parallel execution and feedback, after the assignment and scheduling are completed, the agents execute the subtasks in parallel according to the task hierarchical structure obtained in Step 2 and perform status feedback during the execution process.

[0089] Step 4: Dynamic adjustment and optimization, during the task execution process, the system dynamically adjusts the task assignment and scheduling according to the status feedback.

[0090] As Figure 2 shown, the task decomposition and layering method in Step 1 specifically includes:

[0091] Step 1-1: Task decomposition, decomposing the complex task T into several subtasks to form a task decomposition set T1, T2, …, T n , in the task decomposition, the division of subtasks can be based on the functional characteristics of the task, and the decomposition method of task T is:

[0092] T = {T1, T2, … T i … T j …, T n},

[0093] where T i represents the subtask at the i-th layer, and the task decomposition process is carried out layer by layer through a recursive decomposition method.

[0094] Step 1-2: Hierarchical structure construction, constructing the hierarchical structure of tasks according to the task dependency relationship, where the subtasks in each layer have specific dependency relationships with the tasks in other layers, and the modeling of the dependency relationship is represented by a matrix.

[0095] Set the dependency relationship matrix of the task as D, where Dij Denote task T i Whether it depends on task T j , if T i depends on T j , then D ij = 1, otherwise D ij = 0; D ii = 0, i = 1, 2, 3,..., n;

[0096] The task hierarchy is:

[0097] L = {L1, L2,…, L i …, L n};

[0098] Among them, L i represents the task set of the i-th layer, satisfying the sorting relationship of task T i in the hierarchy.

[0099] Steps 1 - 3: Phase dependency analysis. The dependency analysis of tasks provides a basis for the scheduling of subsequent tasks. According to the dependency relationship between tasks, a task dependency matrix D is formed to represent the sequential execution constraints between tasks;

[0100] The dependency matrix D can be constructed in the following way:

[0101]

[0102] Among them, D ij represents the relationship that task T i depends on task T j .

[0103] When decomposing tasks based on functional characteristics, taking the decomposition of typical information system tasks as an example, the decomposition of task T in step 1 - 1 is specifically to decompose tasks based on functional characteristics, which can be divided into a perception subtask T1, a planning subtask T2, an execution subtask T3, a task allocation subtask T4, a collaboration subtask T5, and an evaluation and feedback subtask T6, a total of 6 subtasks.

[0104] Among them, the perception subtask T1 includes the following subtasks: a data collection subtask responsible for collecting real-time data from sensors and other devices; a data preprocessing subtask responsible for performing data cleaning, denoising, and standardization processing; a target recognition and tracking subtask T 13 , responsible for identifying and tracking the state of the target; a situation assessment subtask T 14 , responsible for assessing the combat environment and situation.

[0105] The planning subtask T2 includes the following subtasks: a task decomposition subtask Responsible for converting high-level tasks into specific subtasks; path planning subtask Responsible for calculating the optimal path or action route; resource scheduling subtask Responsible for scheduling the resources required for the task; situation prediction and adjustment subtask Responsible for adjusting the task plan based on the feedback from the perception subtask.

[0106] The execution subtask T3 includes the following subtasks: task execution subtask Responsible for executing the task according to the instructions of the planning subtask; task monitoring subtask Responsible for monitoring the task progress and status in real time; fault recovery subtask Responsible for handling abnormal situations during task execution.

[0107] The task allocation subtask T4 includes the following subtasks: resource allocation subtask Responsible for allocating tasks according to task requirements, agent capabilities and resource availability; load balancing subtask Responsible for optimizing task allocation to balance the load of each agent; priority sorting subtask Responsible for sorting and allocating tasks according to task priorities.

[0108] The cooperation subtask T5 includes the following subtasks: information sharing subtask Responsible for ensuring information sharing and circulation among agents; task coordination subtask Responsible for coordinating the task execution of different agents; cooperation feedback subtask T 53 , responsible for collecting the execution status and adjusting the cooperation strategy.

[0109] The evaluation and feedback subtask T6 includes the following subtasks: performance evaluation subtask Responsible for evaluating the task execution effect; feedback analysis subtask Responsible for analyzing the problems and bottlenecks during execution; optimization and adjustment subtask Responsible for adjusting task scheduling and resource allocation based on feedback.

[0110] The subtask allocation and scheduling method in step 2 specifically includes:[[]]

[0111] Step 2-1: Subtask allocation model. The system allocates tasks according to the capabilities of agents, task requirements and their resource availability. Each subtask T i needs to be allocated to a suitable agent A j .

[0112] Establish a task allocation function based on the skill matching degree, availability and current load of agents. The allocation function P ij represents that task T i is allocated to agent Aj The probability or weight is calculated as follows:

[0113] P ij = w1·skill(A j ,T i ) + w2·availability(A j ) + w3·load(A j );

[0114] Wherein, skill(A j ,T i ) represents the skill matching degree of agent A j executing task T i ), availability(A j ) represents the availability of agent A j ), load(A j ) represents the current task load of agent A j ; w1, w2, and w3 are the weights of each factor, which are determined by tuning;

[0115] Step 2-2: Scheduling optimization. After the subtask allocation is completed, the goal of scheduling optimization is to minimize the total execution time. Assuming that all tasks have a completion time function completion_time(T i ), then the goal of scheduling optimization is:

[0116]

[0117] During the scheduling process, the dependency relationship between tasks needs to be considered to ensure that the task execution order meets the prerequisite conditions. The scheduling constraints can be represented by the following inequality:

[0118] completion_time(T i ) ≥ completion_time(T j ) if D ij = 1;

[0119] represents that the completion time of task T i cannot be earlier than that of task T j , that is, task T j is completed before T i .

[0120] The parallel execution and feedback method in step 3 specifically includes:

[0121] Step 3-1: Parallel execution. After the scheduling optimization is completed, each agent executes its respective subtasks in parallel according to the task allocation result. Assuming that the execution time of each task is When executed in parallel, the execution time of all tasks can be determined by the maximum value:

[0122]

[0123] Step 3-2: State feedback. During the task execution, the intelligent agent will provide real-time feedback on the task execution status, including information such as task progress, task completion, time consumption, computing resource consumption, task execution exception handling, environmental change feedback, and resource consumption information. The feedback status of each task can be represented as a vector:

[0124] F = {f1, f2, …, f n};

[0125] where f i represents the feedback status of task T i .

[0126] Step 3-3: Task completion confirmation. This step is crucial to ensure that the task is completed according to the predetermined goals and standards. The system will confirm the task completion status based on the feedback data and activate subsequent tasks according to the task dependencies. Let the completion status of task T i be C(T i ), then the task completion confirmation can be expressed as:

[0127]

[0128] Task completion confirmation is achieved by continuously tracking the three types of information in the execution status feedback: "task progress", "task completion", and "task execution exception handling".

[0129] Task progress judgment: If the task progress reaches 100%, then the corresponding f i = completed.

[0130] Task completion judgment: The execution of a task not only considers whether the progress reaches 100%, but also whether the task meets the predetermined goals and whether all subtasks or operations in the plan have been successfully completed. If all subtasks have been completed and meet the predetermined goals, then the task completion reaches 100%, and the feedback f i = completed.

[0131] Task execution exception handling: During the task execution, the task may not be completed, and then the exception situation will be feedback. If the task cannot continue to execute due to resource problems, the feedback f i = failed; if the task times out, the feedback f i = timeout; if the task cannot be executed due to environmental changes, the feedback f i = interrupted.

[0132] Step 3-4: Task Activation or Adjustment

[0133] If the current subtask has been completed without any exceptions, the system determines whether there are other tasks T j that depend on the current subtask T i . If D ij = 1 and f i = completed, then task T is activated j .

[0134] When a task fails, times out, etc., it may block subsequent tasks that depend on it. The system needs to adjust or reschedule the execution of the dependent tasks

[0135] Task Blocking Confirmation: If task T i is not completed or fails, the task T i that depends on task T j will be blocked until task T i is successfully completed

[0136] Rescheduling and Task Adjustment: The system will adjust task T j according to the feedback information, and the means include adjusting the priority and reallocating resources

[0137] As Figure 3 shown, the dynamic adjustment and optimization method in step 4 specifically includes

[0138] Step 4-1: Task Adjustment. During the execution of a task, if an exception occurs or the task dependency relationship changes, the system needs to reallocate tasks or adjust task priorities. Task adjustment can be performed through the following formula

[0139]

[0140] where Adjust(T i , D, F) represents adjusting task T i according to the current task status F and the dependency relationship d

[0141] Step 4-2: Allocation Optimization. Dynamically adjust task allocation according to the feedback data to optimize resource usage. Assume that the load situation of each agent is represented by L(A j ), and the new task allocation scheme can be optimized through the following formula

[0142]

[0143] where represents agent A jThe ratio of the load to its maximum load capacity. The values of the weight coefficients w1, w2, and w3 in step 4-2 are the same as those in step 2.

[0144] Step 4-3: System optimization. During the task execution, the core objective of system optimization is to minimize the total execution time, balance the loads of each agent, improve resource utilization, and ensure that the dependencies between tasks are satisfied. To achieve this goal, system optimization comprehensively considers four aspects: execution time, load balance, task dependencies, and resource utilization. The specific optimization objectives and methods are as follows:

[0145] Minimization of the total execution time: The main goal of the system is to minimize the total execution time of all tasks, that is, to ensure that tasks are completed in the shortest possible time. The total execution time T total is defined as the maximum time when all tasks are executed, and the optimization objective is to minimize this value:

[0146]

[0147] where I k represents the task set assigned to the k-th agent, and execution_time(T i ) represents the execution time of task T i . By adjusting the task assignment and scheduling order, the overall maximum execution time is reduced.

[0148] Load balance optimization: The load balance of agents is another optimization objective, aiming to ensure that the workloads of each agent are balanced to avoid resource waste or low execution efficiency caused by some agents being overloaded. The load load(A j ) of each agent A j represents the sum of the execution times of the tasks it executes. The system needs to minimize the variance of the loads of each agent. The objective of load balance optimization can be expressed as:

[0149]

[0150] where m is the number of agents, and load(A j ) is the load of the j-th agent. The optimization process makes the workloads of each agent as close as possible by reducing the variance between the loads.

[0151] Task dependency constraints: When executing tasks, the dependencies between tasks must be considered to ensure that prerequisite tasks are completed before subsequent tasks. Suppose task T i depends on task T j , that is, D ij = 1, then the completion time of task T i cannot be earlier than the completion time of task T jCompletion time. The dependency constraint can be expressed as:

[0152]

[0153] where completion_time(T i ) represents the completion time of task T i , duration(T j ) represents the execution time of task T j , and D ij represents whether task T i depends on task T j . This constraint ensures that the execution order of tasks conforms to the dependency relationship.

[0154] Maximize resource utilization. To improve the overall efficiency of the system, the resource utilization of the agent also needs to be maximized. The resource utilization of the agent can be improved by reducing idle time. The idle time T idle (A j ) represents the idle duration of agent A j during task execution. The optimization goal is to minimize the total idle time:

[0155]

[0156] where T idle (A j ) is the idle time of agent A j . This goal ensures that the execution ability of the agent is fully utilized, thereby improving the resource usage efficiency.

[0157] Comprehensive optimization goal. Combining the above optimization goals, the comprehensive optimization problem of the system can be expressed as:

[0158]

[0159] where: T idle is the total execution time, load(A j ) is the load of agent A j , and T idle (A j ) is the idle time of agent A j .

[0160] The present invention provides a heterogeneous agent hierarchical planning method for complex task decomposition. There are many methods and ways to specifically implement this technical solution. The above description is only a preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art of this technology, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention. Each component not clearly defined in this embodiment can be implemented by using existing technologies.

Claims

1. A hierarchical planning method for heterogeneous agents oriented to complex task decomposition, characterized in that, It includes the following steps: Step 1: Task decomposition and layering. Convert complex tasks into multiple smaller subtasks to form a manageable hierarchical structure for step-by-step execution and parallel processing; Step 2: Subtask allocation and scheduling. Allocate the tasks at each layer obtained in Step 1 to agents, optimize the scheduling and resource allocation of subtasks, and achieve parallel execution; Step 3: Parallel execution and feedback. After the allocation and scheduling are completed, the agents execute the subtasks in parallel according to the task hierarchical structure obtained in Step 2 and perform status feedback during the execution process; Step 4: Dynamic adjustment and optimization. During the task execution process, the system dynamically adjusts the task allocation and scheduling according to the status feedback.

2. The heterogeneous agent hierarchical planning method for complex task decomposition according to claim 1, wherein The method of task decomposition and layering in Step 1 specifically includes: Step 1-1: Task decomposition. Decompose the complex task T into several subtasks to form a task decomposition set T1, T2, …, T n , in the task decomposition, the division of subtasks can be based on the functional characteristics of the task. The decomposition method of task T is as follows: T = {T1, T2, … T i … T j …, T n}, Among them, T i represents the subtask of the i-th layer, and the task decomposition process is carried out layer by layer through a recursive decomposition method; Step 1-2: Hierarchical structure construction. Construct the hierarchical structure of tasks according to the task dependency relationship. The subtasks in each layer have specific dependency relationships with the tasks in other layers, and the modeling of the dependency relationship is represented by a matrix; Set the dependency matrix of the tasks as D, where D ij represents whether task T i depends on task T j . If T i depends on T j , then D ij = 1, otherwise D ij = 0; D ii = 0, i = 1, 2, 3,..., n; The task hierarchical structure is: L = {L1, L2, …, L i …, L n}; Among them, L i represents the task set of the i-th layer, satisfying the task T i 's sorting relationship in the hierarchical structure; Step 1-3: Phase dependency analysis. The dependency analysis of tasks provides a basis for the subsequent task scheduling. According to the dependency relationship between tasks, a task dependency matrix D is formed to represent the sequential execution constraints between tasks; The dependency matrix D can be constructed in the following way: Among them, D ij represents that task T i depends on task T j in the relationship of 3. The heterogeneous agent hierarchical planning method for complex task decomposition according to claim 2, wherein The decomposition of task T in Step 1-1 is specifically to decompose the task based on functional characteristics, which can be divided into 6 subtasks: a perception subtask T1, a planning subtask T2, an execution subtask T3, a task allocation subtask T4, a collaboration subtask T5, and an evaluation and feedback subtask T6; Among them, the perception subtask T1 includes the following subtasks: the data collection subtask responsible for collecting real-time data from sensors and other devices; the data preprocessing subtask responsible for performing data cleaning, denoising, and normalization; the target recognition and tracking subtask T 13 , responsible for identifying and tracking the state of the target; the situation assessment subtask T 14 , responsible for evaluating the combat environment and situation; The planning subtask T2 includes the following subtasks: task decomposition subtask Responsible for transforming high-level tasks into specific subtasks; path planning subtask Responsible for calculating the optimal path or action route; resource scheduling subtask Responsible for scheduling the resources required for the task; situation prediction and adjustment subtask Responsible for adjusting the task plan based on the feedback from the perception subtask; The execution of subtask T3 includes the following subtasks: task execution subtask Responsible for executing tasks according to the instructions of the planning subtask; task monitoring subtask Responsible for monitoring the task progress and status in real time; fault recovery subtask Responsible for handling abnormal situations during task execution; The task assignment subtask T4 includes the following subtasks: the resource allocation subtask Responsible for allocating tasks according to task requirements, agent capabilities, and resource availability; the load balancing subtask Responsible for optimizing task allocation to balance the loads of each agent; the priority sorting subtask Responsible for sorting and allocating tasks according to task priorities; The collaborative subtask T5 includes the following subtasks: the information sharing subtask responsible for ensuring information sharing and circulation among agents; the task coordination subtask responsible for coordinating the task execution of different agents; the collaborative feedback subtask T 53 , responsible for collecting the execution status and adjusting the collaborative strategy; The evaluation and feedback subtask T6 includes the following subtasks: the performance evaluation subtask Responsible for evaluating the task execution effect; the feedback analysis subtask Responsible for analyzing the problems and bottlenecks in execution; the optimization and adjustment subtask Responsible for adjusting task scheduling and resource allocation based on feedback.

4. The heterogeneous agent hierarchical planning method for complex task decomposition according to claim 3, wherein The method of subtask allocation and scheduling in Step 2 specifically includes: Step 2-1: Subtask Allocation Model. The system allocates tasks according to the capabilities, task requirements, and resource availability of the agents, and each subtask T i needs to be allocated to a suitable agent A j ; Establish a task allocation function based on the skill matching degree, availability, and current load of the agent. The allocation function P ij represents the task T i assigned to the agent A j The probability or weight, and the calculation method is as follows: P ij = w1·skill(A j , T i ) + w2·availability(A j ) + w3·load(A j ); Among them, skill(A j ,T i ) represents the skill matching degree of agent A j executing task T i , availability(A j ) represents the availability of agent A j , load(A j ) represents the current task load of agent A j , w1, w2, w3 are the weights of each factor, which are determined by tuning; Step 2-2: Scheduling optimization. After the subtask allocation is completed, the goal of scheduling optimization is to minimize the total execution time. Assume that all tasks have a completion time function completion_time(T i ). Then the goal of scheduling optimization is: During the scheduling process, the dependency relationship between tasks needs to be considered to ensure that the task execution order meets the preconditions. The scheduling constraints can be represented by the following inequality: completion_time(T i )≥completion_time(T j )ifD ij =1; Indicates that the completion time of task T i cannot be earlier than that of task T j , that is, task T j is completed before T i .

5. A heterogeneous agent hierarchical planning method for complex task decomposition according to claim 4, characterized in that, The method of parallel execution and feedback in Step 3 specifically includes: Step 3-1: Execute in parallel. After the scheduling optimization is completed, each agent executes its respective subtasks in parallel according to the task allocation result. Assume that the execution time of each task is Then, when executing in parallel, the execution time of all tasks can be determined by the maximum value: Step 3-2: Status feedback. During the task execution process, the agent will real-time feedback the execution status of the task. The feedback status information includes task progress, task completion, time consumption, computing resource consumption, task execution exception handling, environmental change feedback, and resource consumption. The feedback status of each task can be represented as a vector: F = {f1, f2, …, f n}; where f i represents the feedback status of task T i ; Step 3-3: Task Completion Confirmation. The system confirms the task completion status based on the feedback data and activates subsequent tasks according to the task dependencies; assume the completion status of task T i is C(T i ), then the task completion confirmation can be expressed as: The system determines whether to activate the subsequent task T i based on the completion status of task T j ; Step 3-4: Task activation. If the current subtask has been completed without any exceptions, the system determines, based on the dependency matrix D, whether there is any other task T j that depends on the current subtask T i . If D ij = 1 and f i = completed, then task T is activated j .

6. The heterogeneous agent hierarchical planning method for complex task decomposition according to claim 5, wherein Step 3-3 determines the completion status of task T i Specifically includes: By continuously tracking the three types of information of "task progress, task completion, and task execution exception handling" in the execution status feedback, the task completion confirmation is comprehensively realized; Task progress judgment: If the task progress reaches 100%, then the corresponding f i = completed; Task Completion Judgment: The execution of a task is not only considered by whether the progress reaches 100%, but also by whether the task meets the predefined goals and whether all subtasks or operations in the plan have been successfully completed. If all subtasks have been completed and meet the predefined goals, the task completion reaches 100%, and the feedback f i = completed; Task execution exception handling: During the task execution, the task may not be completed, then an exception situation is reported. If the task cannot continue to execute due to resource problems, report f i = failed; If the task times out, report f i = timeout; If the task becomes unexecutable due to environmental changes, report f i = interrupted.

7. A heterogeneous agent hierarchical planning method for complex task decomposition according to claim 6, characterized in that, In Step 3, when a problem occurs in a certain task and may block the subsequent tasks that depend on this task, the system needs to adjust or reschedule the execution of the dependent tasks. The steps include; Task Blocking Confirmation: If task T i is not completed or fails, tasks T i that depend on task T j will be blocked until task T i is successfully completed; Rescheduling and Task Adjustment: The system adjusts task T based on the feedback information. j The adjustment measures include adjusting the priority and reallocating resources.

8. A heterogeneous agent hierarchical planning method for complex task decomposition according to claim 7, characterized in that The method of dynamic adjustment and optimization in Step 4 specifically includes: Step 4-1: Task adjustment. During the task execution process, if an exception occurs or the task dependency relationship changes, the system reallocates tasks or adjusts the task priorities. The task adjustment can be carried out through the following formula: where Adjust(T i , D, F) means adjusting task T i according to the current task status F and the dependency relationship D; Step 4-2: Allocation optimization, dynamically adjust task allocation according to feedback data to optimize resource utilization; assume that the load situation of each agent is represented by the maximum load capacity L(A j ), and the new task allocation scheme can be optimized by the following formula: wherein represents the ratio of the load of Agent A j to its maximum load capacity; Step 4-3: System optimization. During the task execution process, the core goal of system optimization is to minimize the total execution time, balance the load of each agent, improve resource utilization, and ensure that the dependency relationship between tasks is satisfied.

9. A heterogeneous agent hierarchical planning method for complex task decomposition according to claim 8, characterized in that, The values of the weight coefficients w1, w2, and w3 in Step 4-2 are the same as those in Step 2.

10. A heterogeneous agent hierarchical planning method for complex task decomposition according to claim 9, characterized in that The system optimization in Step 4-3 specifically includes: The system optimization comprehensively considers the execution time, load balancing, task dependency relationships, and resource utilization. The specific optimization objectives and methods are as follows: Minimization of the total execution time: Set minimizing the total execution time of all tasks as the main goal of the system, and the total execution time T total is defined as the maximum time when all tasks are completed, and the optimization goal is to minimize this value: Among them, I k represents the task set assigned to the k-th agent, and execution_time(T i ) represents the execution time of task T i ; by adjusting the assignment and scheduling order of tasks, the overall maximum execution time is reduced; Load balancing optimization: The load balancing of agents ensures that the workloads of each agent are balanced. The load of each agent A j is denoted as load(A j ), which represents the total execution time of the tasks it executes. The system needs to minimize the variance of the loads of each agent. The goal of load balancing optimization can be expressed as: where m is the number of agents, and load(A j ) is the load of the j-th agent. The optimization process aims to minimize the variance between the loads, making the workloads of all agents as close as possible. Task Dependency Constraint: When executing tasks, consider the dependencies between tasks to ensure that prerequisite tasks are completed before subsequent tasks; assume task T i depends on task T j , that is, D ij = 1, then the completion time of task T i cannot be earlier than the completion time of task T j . The dependency constraint can be expressed as: where completion_time(T i ) represents the completion time of task T i , duration(T j ) represents the execution time of task T j , D ij represents whether task T i depends on task T j , and this constraint ensures that the execution order of tasks conforms to the dependency relationship; Maximize resource utilization. The resource utilization of the agent is improved by reducing idle time; the idle time T of the agent idle (A j ) represents the idle duration of agent A j when performing tasks. The optimization goal is to minimize the total idle time: where T idle (A j ) is the idle time of agent A j ; Comprehensive optimization objective. Considering the above optimization objectives, the comprehensive optimization problem of the system can be expressed as: Where: T idle is the total execution time, load(A j ) is the load of agent A j , and T idle (A j ) is the idle time of agent A j .

Images