Task allocation method and device based on multiple agents, equipment and storage medium

Abstract

The invention relates to the technical field of task allocation, in particular to a multi-agent-based task allocation method and device, equipment and a storage medium. The multi-agent-based task allocation method comprises the following steps: generating heterogeneity parameters according to task allocation agents; performing task allocation according to preset system constraints and heterogeneity parameters to obtain an initial allocation task; generating a parameterized strategy set according to a preset strategy type and a preset task priority; performing bionic optimization on the initial allocation task according to the strategy type, the parameterized strategy set and a preset bionic agent to obtain an optimized allocation task; the initial allocation adaptability is improved through heterogeneity parameters, and resource mismatching is avoided; a high-priority task is guaranteed based on a parameterization strategy, and differential response is strengthened; problems are dynamically corrected, efficiency and stability are improved, and full-process support is provided for multi-agent complex tasks.

Description

Technical Field

[0001] This invention relates to the field of task allocation technology, and more specifically to a task allocation method, apparatus, device, and storage medium based on multi-agent systems. Background Technology

[0002] Existing multi-agent task allocation methods have significant shortcomings in complex dynamic scenarios: they lack consideration for individual differences among agents, often using homogeneous models to handle heterogeneous capabilities, resulting in poor adaptability between task allocation and actual capabilities, making it difficult to meet system resource and time constraints; they lack targeted task optimization design, and the flexibility of dynamic task adjustment is insufficient, that is, they lack an effective biomimetic optimization framework, making it difficult to support the efficient allocation of complex tasks. Summary of the Invention

[0003] In order to overcome the shortcomings of the prior art, the purpose of this invention is to provide a task allocation method, apparatus, device and storage medium based on multiple agents.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: This invention provides a multi-agent-based task allocation method, comprising: generating heterogeneity parameters according to the task allocation agent; allocating tasks according to preset system constraints and heterogeneity parameters to obtain initial allocated tasks; generating a parameterized policy set according to preset policy types and preset task priorities; and performing biomimetic optimization on the initial allocated tasks according to the policy types, parameterized policy set, and preset biomimetic agents to obtain optimized allocated tasks.

[0005] Furthermore, the step of performing biomimetic optimization on the initial allocation task based on the strategy type, the parameterized strategy set, and the preset biomimetic agent to obtain an optimized allocation task includes: optimizing the parameterized strategy set based on preset evaluation metrics and preset strategy gradients to obtain an optimized parameter strategy set; training a preset short-term memory neural network model based on preset time steps and preset historical observations to obtain a time series model; and performing biomimetic optimization on the initial allocation task based on the time series model, the optimized parameter strategy set, and the biomimetic agent to obtain an optimized allocation task.

[0006] Furthermore, the step of optimizing the parameterized strategy set according to the preset evaluation index and the preset policy gradient to obtain the optimized parameter strategy set includes: obtaining a first parameter strategy set and a second parameter strategy set from the parameterized strategy set according to the evaluation index; fine-tuning the first parameter strategy set according to the policy gradient to obtain the mutated parameter strategy set; and generating the optimized parameter strategy set according to the second parameter strategy set and the mutated parameter strategy set.

[0007] Furthermore, obtaining the first parameter strategy set and the second parameter strategy set from the parameterized strategy set based on the evaluation indicators includes: scoring the parameterized strategy set according to the evaluation indicators to obtain a low-performance strategy score set and a high-performance strategy score set; obtaining the first parameter strategy set from the parameterized strategy set based on the low-performance strategy score set; and obtaining the second parameter strategy set from the parameterized strategy set based on the high-performance strategy score set.

[0008] Furthermore, the step of performing biomimetic optimization on the initial allocation task based on the time series model, the optimization parameter policy set, and the biomimetic agent to obtain the optimized allocation task includes: converting the optimization parameter policy set to obtain a node feature set; constructing the node feature set based on a preset graph neural network model and a preset neighborhood agent to obtain a neighborhood relationship network; analyzing the neighborhood relationship network based on a preset link quality index to obtain local value estimation and potential policy vectors; and performing biomimetic optimization on the initial allocation task based on the biomimetic agent, local value estimation, time series model, and potential policy vectors to obtain the optimized allocation task.

[0009] Furthermore, the step of performing biomimetic optimization on the initial allocation task based on the biomimetic agent, local value estimation, time series model, and latent policy vector to obtain an optimized allocation task includes: performing feature analysis on the local value estimation and latent policy vector to obtain real-time observations and local estimated time steps; performing inference on the local estimated time steps based on the time series model to obtain a time dimension score; and performing biomimetic optimization on the initial allocation task based on the biomimetic agent, real-time observations, and time dimension score to obtain an optimized allocation task.

[0010] Furthermore, the step of generating a parameterized strategy set based on a preset strategy type and a preset task priority includes: allocating bandwidth to the initially assigned task based on the task priority and a preset semantic importance to obtain bandwidth allocation resources; obtaining a local perception summary and broadcasting the local perception summary based on the bandwidth allocation resources to obtain information interaction quality; generating strategy selection constraints based on the information interaction quality; and performing stratified sampling on a preset strategy pool based on the strategy type and strategy selection constraints to obtain a parameterized strategy set.

[0011] Furthermore, the multi-agent-based task allocation device includes: a parameter generation module for generating heterogeneous parameters based on the task allocation agent; a task allocation module for allocating tasks according to preset system constraints and heterogeneous parameters to obtain initial allocation tasks; a set generation module for generating a parameterized policy set according to preset policy types and preset task priorities; and a biomimetic optimization module for performing biomimetic optimization on the initial allocation tasks according to the policy type, the parameterized policy set, and preset biomimetic agents to obtain optimized allocation tasks.

[0012] Furthermore, the multi-agent-based task allocation device includes: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the multi-agent-based task allocation device to perform the various steps of the multi-agent-based task allocation method as described in any of the above descriptions.

[0013] Furthermore, a computer-readable storage medium stores instructions that, when executed by a processor, implement the steps of the multi-agent-based task allocation method as described in any one of the preceding descriptions.

[0014] In the technical solution of this invention, heterogeneous parameters are generated by the task allocation agent to replace the traditional homogeneous assumption. Combined with system constraints, precise matching between tasks and agent capabilities is achieved, thereby improving the adaptability of initial allocation, avoiding resource mismatch and illegal execution. A parameterized strategy set is constructed based on strategy type and task priority to prioritize the adaptation of high-priority tasks, enhance the system's differentiated response capability, and lay a solid foundation for optimization. Finally, relying on diversified strategies and biomimetic agent group logic, problems such as resource imbalance and insufficient adaptation in the initial allocation are dynamically corrected, improving task execution efficiency and dynamic adaptation capability, strengthening the system's anti-interference capability and stability in complex scenarios, comprehensively ensuring the quality of task completion and system operating efficiency, and providing flexible and reliable end-to-end support for the efficient advancement of complex tasks by multiple agents. Attached Figure Description

[0015] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a first flowchart of a multi-agent-based task allocation method provided in an embodiment of the present invention; Figure 2 This is a second flowchart of a multi-agent-based task allocation method provided in an embodiment of the present invention; Figure 3 This is a third flowchart of a multi-agent-based task allocation method provided in an embodiment of the present invention; Figure 4This is a fourth flowchart of a multi-agent-based task allocation method provided in an embodiment of the present invention; Figure 5 The fifth flowchart of the multi-agent-based task allocation method provided in the embodiments of the present invention; Figure 6 The sixth flowchart of the multi-agent-based task allocation method provided in the embodiments of the present invention; Figure 7 The seventh flowchart of the multi-agent-based task allocation method provided in the embodiments of the present invention; Figure 8 This is a schematic diagram of the structure of a multi-agent-based task allocation device provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the structure of a multi-agent-based task allocation device provided in an embodiment of the present invention. Detailed Implementation

[0016] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0017] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the task allocation method based on multi-agent systems in this invention includes: 101. Generate heterogeneity parameters for the agent based on task allocation; 102. Based on preset system constraints and heterogeneity parameters, tasks are allocated to obtain the initial task allocation; In this embodiment, the task allocation agent is a specific agent system that is responsible for the core scheduling of the entire task allocation process in a multi-agent system. Its core role is to coordinate the matching and connection between tasks and agent resources. The task allocation agent generates heterogeneity parameters that reflect the differences in agent capabilities and resource endowments, replacing the assumptions of the traditional homogeneous model. Then, based on preset system constraints (such as resource limits, time thresholds, and execution rules), it accurately matches tasks with heterogeneous agent capabilities and outputs initial allocation tasks that conform to basic specifications, thereby improving allocation adaptability from the source. 103. Generate a set of parameterized strategies based on preset strategy types and preset task priorities; In this embodiment, a parameterized strategy set is generated by combining preset strategy types and task priorities to achieve precise matching between strategy and task requirements. Priority is given to ensuring the strategy adaptability of high-priority tasks, and the appropriate strategy types are configured to effectively avoid strategy mismatch and resource waste. This enhances the practical value of the strategy set and the efficiency of task execution, strengthens the system's responsiveness to tasks of different priorities, lays a solid foundation for subsequent task allocation optimization, and helps tasks to proceed efficiently and orderly. 104. Based on the strategy type, parameterized strategy set, and preset bionic agent, perform bionic optimization on the initial task allocation to obtain an optimized task allocation; In this embodiment, relying on diversified parameterization strategies, the task scenarios and requirements are accurately matched. Combined with the biomimetic intelligent agent to simulate the logic of biological groups, the problems of resource imbalance and insufficient adaptation in the initial allocation are effectively solved. After optimization, not only is the task execution efficiency and dynamic adaptation capability improved, but the system's anti-interference and stability are also strengthened. It provides flexible and reliable support for the execution of complex tasks by multiple intelligent agents, ensuring efficient task progress in dynamic and complex scenarios and fully guaranteeing the quality of task completion and system operation performance. In this embodiment, heterogeneous parameters are generated by the task allocation agent to replace the traditional homogeneous assumption. Combined with system constraints, this achieves precise matching between tasks and agent capabilities, thereby improving the adaptability of initial allocation, avoiding resource mismatch and illegal execution. A parameterized strategy set is constructed based on strategy type and task priority to prioritize the adaptation of high-priority tasks, enhance the system's differentiated response capability, and lay a solid foundation for optimization. Finally, relying on diversified strategies and biomimetic agent group logic, problems such as resource imbalance and insufficient adaptation in the initial allocation are dynamically corrected, improving task execution efficiency and dynamic adaptation capability, strengthening the system's anti-interference ability and stability in complex scenarios, comprehensively ensuring the quality of task completion and system operating performance, and providing flexible and reliable end-to-end support for the efficient advancement of complex tasks by multiple agents.

[0018] Please see Figure 2 In the second embodiment of the task allocation method based on multi-agents in this invention, step 104 specifically includes: 201. Optimize the parameterized strategy set according to the preset evaluation index and preset strategy gradient to obtain the optimized parameter strategy set; In this embodiment, based on core evaluation indicators such as communication adaptability and environmental robustness, the parameterized strategy set is quantitatively scored, high-performance strategies and low-performance strategies are divided, optimization objects and high-quality basic strategies are identified, and for low-performance strategies, the focus is on performance shortcomings (such as insufficient communication adaptability) and key sensitive parameters (such as broadcast frequency), and the strategies are adjusted in a gradient direction to generate a set of variant strategies to avoid blind modification that may lead to failure. Finally, high-performance strategies and high-quality variant strategies are merged to form an optimized parameter strategy set that takes into account both stability and optimization potential. 202. Train the preset short-term memory neural network model according to the preset time steps and preset historical observations to obtain the time series model; In this embodiment, historical observations (such as past task execution efficiency, environmental dynamic changes, and other time-series data) are input in preset time steps to train the LSTM to learn the temporal correlation patterns of task execution (such as time-varying environmental complexity and agent performance fluctuations). The output is a time series model that can predict time-series trends, providing time-series decision support for optimization. 203. Based on the time series model, the optimization parameter strategy set, and the biomimetic agent, perform biomimetic optimization on the initial task allocation to obtain the optimized task allocation; In this embodiment, time series models, optimized parameter strategy sets, and biomimetic agents are integrated to accurately predict time-series risks. The initial task allocation is optimized by relying on high-quality strategies and biological group logic, which effectively solves problems such as insufficient task adaptation and resource imbalance, improves execution efficiency and dynamic adaptability, strengthens the system's anti-interference ability and execution stability, and provides reliable support for task execution in complex scenarios. In this embodiment, a selection strategy based on core indicators such as communication adaptability and environmental robustness is adopted. Combined with gradient-oriented fine-tuning to activate the potential of low-performance strategies, a strategy set that balances stability and optimization is formed, improving strategy adaptability and resource utilization. By learning historical time series patterns through the LSTM model, the system can accurately predict time series risks such as environmental fluctuations and performance fluctuations, providing forward-looking support for decision-making. Finally, by integrating high-quality strategies, time series prediction, and biomimetic agent group logic, the system can specifically solve problems such as insufficient task adaptability and resource imbalance in the initial allocation, improve task execution efficiency and dynamic adaptability, strengthen the system's anti-interference ability and adaptability to complex scenarios, provide efficient and reliable optimization support for the execution of complex tasks by multiple agents, and comprehensively ensure the quality of task completion and the stability of system operation.

[0019] Please see Figure 3 In the third embodiment of the task allocation method based on multi-agents in this invention, step 201 specifically includes: 301. Obtain the first parameter strategy set and the second parameter strategy set from the parameterized strategy set based on the evaluation index; In this embodiment, the parameterized strategy set is analyzed using communication adaptability and environmental robustness as core evaluation indicators. A weighted quantitative score is generated, and the parameter strategy set is divided according to a clear threshold to improve the accuracy of strategy identification, optimize the overall adaptability of strategy selection, reduce execution risk, improve system resource utilization and robustness, and provide reliable support for complex tasks of multi-agent systems. 302. Fine-tune the first parameter policy set according to the policy gradient to obtain the mutated parameter policy set; In this embodiment, the strategy gradient is a two-dimensional optimization approach centered on performance shortcomings and parameter sensitivity. First, it accurately identifies the root causes of low performance in the first set of parameter strategies (low-performance strategies) (e.g., insufficient communication adaptability, weak environmental robustness). Then, it selects the key parameters that have the most significant impact on this shortcoming (e.g., insufficient communication adaptability corresponds to broadcast frequency and data compression ratio, while weak robustness corresponds to iteration step size and exploration rate). This forms a clear gradient between the problem and the core parameters. Fine-tuning logic revolves around this gradient, focusing on the shortcomings to determine the direction of parameter adjustments. The adjustment range is gradually controlled according to the gradient, avoiding blind modifications that could lead to strategy failure, while precisely optimizing inefficient aspects. Ultimately, this activates the potential value of low-performance strategies, generating a set of variable parameter strategies with optimization potential and strong controllability, laying the foundation for subsequent integration with high-performance strategies. 303. Generate an optimization parameter strategy set based on the second parameter strategy set and the mutation parameter strategy set; In this embodiment, a quantitative evaluation system centered on communication adaptability and environmental robustness accurately divides high-performance and low-performance strategy sets according to clear thresholds, improving the accuracy of strategy identification and selection adaptability, and effectively reducing execution risks. For low-performance strategies, a two-dimensional strategy gradient-oriented fine-tuning based on performance shortcomings and parameter sensitivity is used to accurately optimize inefficient links, activate the potential value of low-performance strategies, avoid resource waste, and improve strategy resource utilization. Finally, the optimized parameter strategy set is formed by integrating high-performance strategies and high-quality variant strategies. It retains the core advantages of stability and efficiency while taking into account system stability and dynamic adaptability, improving the efficiency and resource utilization of multi-agent tasks, and providing reliable and efficient strategy support for task optimization in complex scenarios.

[0020] Please see Figure 4 In the fourth embodiment of the task allocation method based on multi-agents in this invention, step 301 specifically includes: 401. Score the parameterized strategy set according to the evaluation indicators to obtain the low-performance strategy score set and the high-performance strategy score set; 402. Obtain the first parameter strategy set from the parameterized strategy set based on the low-performance strategy score set; 403. Obtain the second parameter strategy set from the parameterized strategy set based on the high-performance strategy score set; In this embodiment, the evaluation indicators include communication adaptability (the degree of matching between the strategy and the current bandwidth and interaction quality) and environmental robustness (stability in the face of dynamic environments), with weights set to 0.7 and 0.3 respectively. A weighted summation method is used (total score = 0.7 × adaptability + 0.3 × robustness) to generate individual strategy scores ranging from 0 to 100. The parameterized strategy set is divided according to the scores of each strategy and the high-performance threshold (≥60 points) to obtain a low-performance strategy score set, and the parameterized strategy set is divided according to the low-performance threshold (<60 points) to obtain a low-performance strategy score set. In this embodiment, a weighted scoring model is constructed using evaluation metrics. This model highlights the core requirements for adapting strategies to bandwidth and interaction quality while also considering the stability of dynamic environments. It quantifies scores and defines thresholds to improve the accuracy of identifying high-performance and low-performance strategies. The second parameter strategy set can be directly reused, reducing the cost of repeated sampling and parameter debugging, and improving task optimization efficiency. The first parameter strategy set (low-performance) clearly targets the improvement objects, avoiding the occupation of system resources by inferior strategies. It optimizes the accuracy and adaptability of overall strategy selection, reduces the execution risks caused by communication incompatibility and poor environmental adaptability, improves system resource utilization and robustness, and provides efficient and reliable strategy support for the optimization of complex tasks by multiple agents.

[0021] Please see Figure 5 In the fifth embodiment of the task allocation method based on multi-agents in this invention, step 203 specifically includes: 501. Convert the type of the optimization parameter strategy set to obtain the node feature set; In this embodiment, the various strategies in the optimization parameter strategy set are transformed into structured node features that can be processed by graph neural networks. Each type of optimization strategy (such as rule-based strategy, heuristic strategy, and high communication dependency strategy) is decomposed according to a preset dimension, and core attributes (such as strategy communication requirements, resource consumption threshold, execution efficiency benchmark, and adapted task type) are extracted. The decomposed attributes are transformed into quantifiable feature vectors (such as communication requirements being quantified with a score of 0-10, and resource consumption being represented by a percentage), forming node features corresponding to a single strategy. The node features corresponding to all strategies are summarized to form a unified dimension and standardized node feature set, providing standardized input for subsequent network construction. 502. Based on a preset graph neural network model and a preset neighborhood agent, a node feature set is constructed to obtain a neighborhood relationship network; In this embodiment, based on the association modeling capability of the graph neural network (GNN) model and combined with the interaction relationships of neighborhood agents, isolated node features are connected into a neighborhood relationship network with logical connections, realizing an upgrade from the perspective of individual strategies to group relationships. The construction logic is as follows: First, the preset scope of neighborhood agents (such as groups of agents with close geographical distance, close task cooperation, and resource sharing) is defined to delineate the boundaries of network construction. Then, the strategies with concentrated node features are used as nodes, and the interaction relationships between neighborhood agents (such as resource scheduling association and task connection logic) are used as edges to construct the neighborhood relationship network through the graph neural network model. 503. Analyze the neighborhood relationship network based on the preset link quality indicators to obtain local value estimation and potential strategy vectors; In this embodiment, link quality is used as a benchmark to analyze the value of each node and link in the neighborhood relationship network, extract the basis for local optimization and potential strategy directions, and quantify and score each interactive link in the network based on link quality indicators (such as link communication stability, data transmission efficiency, and resource consumption balance). Combining node characteristics and link quality scores, the execution value of a single strategy in the local neighborhood environment (i.e., local value estimation) is calculated, reflecting the actual utility of the strategy in the current association scenario. Link and node combinations with high scores but not yet activated are identified in the network, and the corresponding strategy directions and parameter features are extracted to form a potential strategy vector, representing the optimization potential that has not been activated. 504. Based on the biomimetic agent, local value estimation, time series model and potential policy vector, perform biomimetic optimization on the initial assignment task to obtain the optimized assignment task; In this embodiment, the initial task allocation is optimized by integrating biomimetic agents, local value estimation, time series models and potential policy vectors, thereby activating the potential of high-quality policies. The task allocation is optimized by simulating the logic of biological groups through biomimetic agents, which solves problems such as insufficient adaptation of the initial scheme and resource imbalance, improves the adaptability and efficiency of task execution, and strengthens the dynamic response capability and execution stability of the system. In this embodiment, the optimization strategy is transformed into unified node features to provide accurate input for subsequent analysis. A neighborhood relationship network is built using GNN to connect the interaction links between the strategy and the agent, realizing an upgrade from the perspective of individuals to groups. By combining link quality indicators, the value of local strategies and the vector of inactive potential are accurately extracted to clarify the optimization direction. Finally, multiple factors are integrated to optimize the task through a biomimetic agent, solving problems such as poor initial allocation adaptation and resource imbalance, activating the potential of high-quality strategies, improving the adaptability and efficiency of task execution, and strengthening the dynamic response and stability of the system in complex scenarios.

[0022] Please see Figure 6 In the sixth embodiment of the task allocation method based on multi-agents in this invention, step 504 specifically includes: 601. Perform feature analysis on local value estimation and potential policy vectors to obtain real-time observations and local estimation time steps; In this embodiment, the logic for extracting real-time observations involves decomposing and quantifying the potential policy vector to extract real-time state information, such as the local environment (neighborhood agent load, communication link quality), task dynamics (progress deviation, remaining resources), and policy adaptation status (current policy execution bottleneck), forming an instantaneous state snapshot. The logic for determining the local estimation time step involves extracting the time series markers corresponding to the local value estimation, clarifying the time node of the current local state (such as the Nth cycle after task start, the time node of key processes), and providing accurate time coordinates for time series reasoning. 602. Based on the time series model, reason about the local estimated time steps to obtain the time dimension score; In this embodiment, the time series model, based on the temporal knowledge formed by historical training, analyzes the potential risks (such as communication congestion or insufficient computing power of the agent) and optimization opportunities (such as low environmental interference and high resource redundancy) of the current time step. The time adaptability is presented intuitively by outputting a quantitative score (time dimension score, such as 0-100 points). A high score indicates that the current time step is highly adaptable (low risk, high return), while a low score indicates that there are potential time-series risks (which need to be avoided or adjusted). 603. Based on the bionic intelligent agent, real-time observations, and time-dimensional scores, perform bionic optimization on the initial task assignment to obtain an optimized task assignment; In this embodiment, the biomimetic optimization logic inputs real-time observations (current static state) and time-dimensional scores (future dynamic predictions) into the biomimetic agent system to construct multi-dimensional decision-making criteria. The biomimetic agent simulates the logic of a biological group, comprehensively evaluates the execution efficiency of a single agent and the benefits of multi-agent collaboration, identifies problems in the initial allocation (such as uneven local load, poor timing adaptability, and unactivated strategy potential), and makes targeted adjustments to task allocation details, such as reallocating agent division of labor, switching to potentially high-quality strategies, and optimizing task execution timing (avoiding low-scoring periods). Finally, it outputs optimized task allocation that balances real-time adaptability and long-term stability. In this embodiment, relying on biomimetic intelligent agents to simulate the logic of biological groups, the efficiency of single agents and the benefits of multi-agent collaboration are comprehensively evaluated. The division of labor among agents, strategy selection, and task execution timing are adjusted in a targeted manner. This effectively solves problems such as uneven local load, poor timing adaptability, and unactivated strategy potential in the initial allocation. Ultimately, it achieves a balance between real-time adaptation and long-term stability in task allocation, improves the accuracy and foresight of decision-making, strengthens the dynamic adaptability and anti-interference capability of the multi-agent system, optimizes resource utilization and task completion quality, and provides efficient and reliable guarantee for multi-agent task execution in complex dynamic scenarios.

[0023] Please see Figure 7 In the seventh embodiment of the task allocation method based on multi-agents in this invention, step 103 specifically includes: 701. Allocate bandwidth to the initially assigned tasks based on task priority and preset semantic importance to obtain bandwidth allocation resources; In this embodiment, task priority is determined by quantifying task urgency (e.g., "survivor location" has a higher priority than "material transportation" in disaster relief) to determine the basic weight for bandwidth allocation; semantic importance is determined by characterizing the relevance of the task to the global objective (e.g., "core component assembly" has a higher semantic relevance than "product production" than "environmental monitoring") to adjust the bandwidth allocation ratio; the bandwidth allocation rule allocates more bandwidth resources to high-priority and high-semantic-importance tasks (e.g., 30%-50% of the total bandwidth), while low-priority tasks are allocated basic bandwidth to ensure that resources are configured on demand, providing a resource foundation for subsequent perception summary broadcasting and avoiding data transmission delays and packet loss due to insufficient bandwidth for critical tasks; 702. Obtain the local perception summary and broadcast the local perception summary according to the bandwidth allocation resources to obtain the information interaction quality; In this embodiment, the local perception summary mainly includes three types of key information: first, environmental perception information (such as the location information of surrounding obstacles, the status of neighboring intelligent agents, environmental noise, etc.); second, system status information (such as remaining energy, computing power usage, hardware fault warnings, etc.); and third, historical task execution information (such as task completion progress, resource consumption deviation, execution anomalies, etc.). The broadcast logic first generates a simplified local perception summary containing environmental, system, and task information; then, it allocates resources according to bandwidth for transmission. High-bandwidth tasks transmit the complete summary and broadcast at high frequency (e.g., once every 200ms), using lightweight compression; low-bandwidth tasks transmit core fields and broadcast at low frequency (e.g., once every 2000ms), using deep compression. Transmission conflicts are avoided through time-division multiplexing and priority preemption. Finally, the quality of information interaction is quantified by data integrity, latency, and confidence. 703. Select constraints based on the information interaction quality generation strategy; In this embodiment, the generation logic of strategy selection constraints is based on the quality of information interaction (quantified as 0-100 points). First, the matching relationship between strategy type and communication dependency is clarified: high communication dependency strategies (such as game theory and reinforcement learning strategies) require high-quality communication support (≥80 points), medium dependency strategies (such as hybrid collaborative strategies) are suitable for medium communication (≥60 points), and low dependency strategies (such as rule-based and heuristic strategies) can adapt to low-quality communication (≥50 points). Then, constraints are generated according to the quality range: high-quality communication (≥80 points) allows all types of strategies to participate, and the proportion of high-dependency strategies is ≤40%; medium communication (60-79 points) disables high-dependency strategies, and the proportion of medium-dependency strategies is ≤50%; low-quality communication (<60 points) only allows low-dependency strategies. Finally, a set of rules of "disabled, restricted, and allowed" is formed to ensure that strategy selection is adapted to actual communication conditions and to provide clear boundaries for subsequent sampling. 704. Perform stratified sampling on the preset policy pool according to the policy type and policy selection constraints to obtain a set of parameterized policies; In this embodiment, the strategy pool is first divided into multiple layers according to type: rule-based, heuristic, and high communication dependency. Each layer contains 3-5 strategies to ensure diversity. Then, the layers are selected based on the information interaction quality constraints: high-quality communication (≥80 points) opens all layers, medium (60-79 points) removes the high communication dependency layer, and poor (<60 points) retains only the rule and heuristic layers. Within each layer, sampling is weighted by historical performance, increasing the probability of selecting the top 30% of strategies by 50%, while controlling the proportion (e.g., the proportion of high dependency layers is ≤40% in the case of high-quality communication). Finally, specific parameters are assigned to form a set, taking into account both communication adaptability and optimization space. In this embodiment, bandwidth allocation combines task priority and semantic importance to achieve on-demand resource configuration. High-value tasks receive sufficient bandwidth, avoiding transmission delays and packet loss of critical information and improving resource utilization. Locally-aware summary broadcasting adapts to bandwidth resources, enabling high-bandwidth tasks to efficiently transmit complete information and low-bandwidth tasks to streamline core content. Combined with conflict avoidance mechanisms, this improves information integrity and keeps latency within thresholds. Policy selection constraints are dynamically adjusted based on interaction quality to ensure policy compatibility with communication conditions and eliminate infeasible policies. Hierarchical sampling balances policy diversity and performance, improving the suitability of the generated parameterized policy set and providing efficient support for complex task optimization.

[0024] The multi-agent task allocation method in the embodiments of the present invention has been described above. The multi-agent task allocation device in the embodiments of the present invention is described below. Please refer to [link to relevant documentation]. Figure 8 One embodiment of the task allocation device based on multi-agent technology in this invention includes: Parameter generation module 1 is used to generate heterogeneous parameters based on the task-assigned agent; Task allocation module 2 is used to allocate tasks according to preset system constraints and heterogeneity parameters to obtain the initial allocated tasks; The set generation module 3 is used to generate a set of parameterized strategies based on preset strategy types and preset task priorities; The biomimetic optimization module 4 is used to perform biomimetic optimization on the initial task allocation based on the strategy type, parameterized strategy set, and preset biomimetic agent to obtain an optimized task allocation. In this embodiment, heterogeneous parameters are generated by the task allocation agent to replace the traditional homogeneous assumption. Combined with system constraints, this achieves precise matching between tasks and agent capabilities, thereby improving the adaptability of initial allocation, avoiding resource mismatch and illegal execution. A parameterized strategy set is constructed based on strategy type and task priority to prioritize the adaptation of high-priority tasks, enhance the system's differentiated response capability, and lay a solid foundation for optimization. Finally, relying on diversified strategies and biomimetic agent group logic, problems such as resource imbalance and insufficient adaptation in the initial allocation are dynamically corrected, improving task execution efficiency and dynamic adaptation capability, strengthening the system's anti-interference ability and stability in complex scenarios, comprehensively ensuring the quality of task completion and system operating performance, and providing flexible and reliable end-to-end support for the efficient advancement of complex tasks by multiple agents.

[0025] Figure 9 This is a schematic diagram of the structure of a multi-agent-based task allocation device 900 provided in an embodiment of the present invention. The multi-agent-based task allocation device 900 can vary significantly due to different configurations or performance characteristics. It may include one or more processors (CPUs) 910 (e.g., one or more processors) and a memory 920, and one or more storage media 930 (e.g., one or more mass storage devices) for storing application programs 933 or data 932. The memory 920 and storage media 930 can be temporary or persistent storage. The program stored in the storage media 930 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the multi-agent-based task allocation device 900. Furthermore, the processor 910 may be configured to communicate with the storage media 930 and execute the series of instruction operations in the storage media 930 on the multi-agent-based task allocation device 900 to implement the steps of the multi-agent-based task allocation method provided in the above-described method embodiments.

[0026] The multi-agent task allocation device 900 may also include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input / output interfaces 960, and / or one or more operating devices 931, such as Windows Server, MacOSX, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand. Figure 9 The structure of the multi-agent-based task allocation device 900 shown does not constitute a limitation on the multi-agent-based task allocation device 900, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0027] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of a multi-agent task allocation method.

[0028] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the device or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0029] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0030] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for task allocation based on multi-agent, characterized in that, The method comprises the following steps: generating a heterogeneity parameter according to a task allocation agent; performing task allocation according to preset system constraints and the heterogeneity parameter to obtain an initial allocation task; generating a parameterized strategy set according to a preset strategy type and a preset task priority; performing bionic optimization on the initial allocation task according to the strategy type, the parameterized strategy set and a preset bionic agent to obtain an optimized allocation task.

2. The multi-agent based task allocation method of claim 1, wherein, The bionic optimization on the initial allocation task according to the strategy type, the parameterized strategy set and the preset bionic agent to obtain the optimized allocation task comprises the following steps: optimizing the parameterized strategy set according to a preset evaluation index and a preset strategy gradient to obtain an optimized parameter strategy set; training a preset short-term memory neural network model according to a preset time step and a preset historical observation value to obtain a time series model; performing bionic optimization on the initial allocation task according to the time series model, the optimized parameter strategy set and the bionic agent to obtain the optimized allocation task.

3. The multi-agent based task allocation method of claim 2, wherein, The optimization of the parameterized strategy set according to the preset evaluation index and the preset strategy gradient to obtain the optimized parameter strategy set comprises the following steps: obtaining a first parameter strategy set and a second parameter strategy set from the parameterized strategy set according to the evaluation index; fine-tuning the first parameter strategy set according to the strategy gradient to obtain a variation parameter strategy set; generating the optimized parameter strategy set according to the second parameter strategy set and the variation parameter strategy set.

4. The multi-agent based task allocation method of claim 3, wherein, The obtaining of the first parameter strategy set and the second parameter strategy set from the parameterized strategy set according to the evaluation index comprises the following steps: scoring the parameterized strategy set according to the evaluation index to obtain a low-performance strategy score set and a high-performance strategy score set; obtaining the first parameter strategy set from the parameterized strategy set according to the low-performance strategy score set; obtaining the second parameter strategy set from the parameterized strategy set according to the high-performance strategy score set.

5. The multi-agent based task allocation method of claim 2, wherein, The bionic optimization on the initial allocation task according to the time series model, the optimized parameter strategy set and the bionic agent to obtain the optimized allocation task comprises the following steps: type conversion is performed on the optimized parameter strategy set to obtain a node feature set; the node feature set is constructed based on a preset graph neural network model and a preset neighborhood agent to obtain a neighborhood relationship network; the neighborhood relationship network is analyzed according to a preset link quality index to obtain a local value estimate and a potential strategy vector; the initial allocation task is bionically optimized according to the bionic agent, the local value estimate, the time series model and the potential strategy vector to obtain the optimized allocation task.

6. The multi-agent based task allocation method of claim 5, wherein, The bionic optimization on the initial allocation task according to the bionic agent, the local value estimate, the time series model and the potential strategy vector to obtain the optimized allocation task comprises the following steps: feature analysis is performed on the local value estimate and the potential strategy vector to obtain a real-time observation value and a local estimation time step; the local estimation time step is inferred according to the time series model to obtain a time dimension score; the initial allocation task is bionically optimized according to the bionic agent, the real-time observation value and the time dimension score to obtain the optimized allocation task.

7. The multi-agent based task allocation method of claim 1, wherein, The generating the parameterized strategy set according to the preset strategy type and the preset task priority includes: performing bandwidth allocation on the initial allocation task according to the task priority and the preset semantic importance to obtain a bandwidth allocation resource; obtaining a local perception abstract, and performing broadcasting on the local perception abstract according to the bandwidth allocation resource to obtain an information interaction quality; generating a strategy selection constraint according to the information interaction quality; performing hierarchical sampling on a preset strategy pool according to the strategy type and the strategy selection constraint to obtain the parameterized strategy set.

8. A multi-agent based task allocation apparatus, characterized by, The method comprises the steps of: a parameter generating module configured to generate a heterogeneity parameter according to a task allocation agent; a task allocation module configured to perform task allocation according to a preset system constraint and the heterogeneity parameter to obtain an initial allocation task; a set generating module configured to generate a parameterized strategy set according to a preset strategy type and a preset task priority; a bionic optimization module configured to perform bionic optimization on the initial allocation task according to the strategy type, the parameterized strategy set and a preset bionic agent to obtain an optimized allocation task.

9. A multi-agent based task allocation device, characterized by The method comprises the steps of: a memory and at least one processor, wherein the memory stores instructions; at least one processor calls the instructions in the memory to enable the multi-agent based task allocation device to perform each step of the multi-agent based task allocation method according to any one of claims 1-7.

10. A computer-readable storage medium having stored thereon instructions, the instructions comprising, The instructions are executed by the processor to implement each step of the multi-agent based task allocation method according to any one of claims 1-7.

Images