Distributed task allocation method, device and electronic device for heterogeneous intelligent agent cluster based on alliance formation game

Abstract

The present invention discloses a method, device, and electronic device for distributed task allocation in a heterogeneous agent cluster based on an alliance formation game. The method comprises: establishing a mathematical model for task allocation. Based on the task allocation scenario of a heterogeneous agent cluster, the method considers the specific resource requirements of the task and the differences in the agent's ability to perform the task, quantifies the heterogeneity, establishes a multi-objective function, and completes the modeling of the task allocation problem under heterogeneous conditions; establishes an alliance formation game model to transform the task allocation problem into an alliance partitioning problem; and solves and checks the problem based on maximum weighted matching and agent transfer alliance rules to achieve the final alliance result layer by layer. The method fully considers the heterogeneity of the agents and the resources required for the task, and maximizes the overlap between the resources required for the task and the resources carried by the agents in its alliance, thereby fully utilizing the cluster's effectiveness.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent agent task allocation, resource allocation and alliance formation game, and relates to a method, device and electronic equipment for distributed task allocation of a heterogeneous intelligent agent cluster based on alliance formation game. Background Art

[0002] Thanks to the division of labor and cooperation among intelligent agents, such as drones, spacecraft, and industrial robots, clusters of intelligent agents can deliver superior performance compared to single agents when executing tasks, exhibiting excellent adaptability and robustness. By designing a well-designed task allocation algorithm, tasks with varying requirements can be assigned to the most appropriate heterogeneous agents, improving cluster performance in terms of energy consumption and task utility.

[0003] Task allocation involves obtaining information such as the target's location and task type before an agent executes a task. This information is then used to calculate each agent's task allocation decision. Given the unique characteristics of task allocation in a cluster of agents, the task allocation algorithm must comprehensively consider the agent's own constraints, the environment's constraints, and the task's constraints. This information is then used to determine conflict-free solutions to maximize the benefits of task execution.

[0004] Numerous research results have proposed various task allocation models and algorithms for different scenarios. These approaches are primarily categorized by architecture into centralized and distributed approaches. As clusters continue to scale, centralized algorithms are increasingly becoming less adaptable to task allocation. Compared to centralized algorithms, distributed algorithms empower each agent with a degree of autonomy and decision-making power. Furthermore, distributed communication topologies require less bandwidth than centralized ones. This allows multi-agent systems to respond more quickly to dynamic flight environments and complex mission requirements. Consequently, they offer advantages such as greater adaptability to large-scale clusters, improved robustness, and rapid response.

[0005] Currently, distributed task allocation algorithms primarily include market-based approaches, distributed Markov decision processes, and game theory. In recent years, game theory has proven to be a promising approach for cooperation and negotiation among intelligent agents, leading to its widespread application in optimization problems. When applying game theory to distributed task allocation, given that participants prioritize maximizing their own self-interest, rather than the overall interests of the system, the agents themselves do not require complex decision-making systems or extensive global information.

[0006] There are currently many studies using coalition game algorithms to address the problem of allocating multiple tasks to multiple agents. The increasingly complex task requirements pose new challenges to the task allocation problem. Tasks that require multiple resource types cannot usually be completed by a single agent that lacks sufficient resource types and numbers, so these tasks require the collaborative execution of multiple agents. Existing algorithms usually lack a complete consideration of the heterogeneity of agents and tasks, ignore the processing of redundant resources, require a large number of iterations to converge to a stable solution, and have low solution efficiency. In response to the above-mentioned algorithmic defects, the present invention aims to improve the efficiency of task allocation and the real-time performance of the algorithm while adding a complete consideration of heterogeneous resources. Summary of the Invention

[0007] The purpose of the present invention is to overcome the defects of the prior art and provide a distributed task allocation method, device and electronic device for a cluster of heterogeneous intelligent agents based on alliance formation game, which can solve the shortcomings of existing task allocation models and algorithms, take into account the heterogeneity of the resources required by the intelligent agents and tasks, minimize the redundant resources of the intelligent agents in the effective alliance, and improve the solution efficiency while reducing the complexity of the problem.

[0008] In order to solve the above technical problems, the present invention adopts the following technical solutions.

[0009] The present invention provides a distributed task allocation method for a heterogeneous agent cluster based on a coalition formation game, comprising the following steps:

[0010] Step 1: Establish a mathematical model for task allocation:

[0011] Based on the task allocation scenario of heterogeneous agent clusters, the specific resource requirements of the tasks and the differences in the ability of agents to perform tasks are considered. The heterogeneity is quantified, and a multi-objective function is established to complete the modeling of the task allocation problem in heterogeneous conditions:

[0012] Among them, the optimization goals of the agent when performing task allocation are specifically:

[0013] maxr CS (1)

[0014]

[0015] Considering the overlap between tasks and heterogeneous resources within the alliance and the path cost of executing tasks, the agent benefit function is designed.

[0016]

[0017] cost(u i ,t j )=d(u i ,t j ) (5)

[0018] Where T={t1,t2,...,t m} is the task set, m is the number of tasks; U={u1,u2,...,u n} is the set of agents, n is the number of agents; CS is the alliance structure, To perform the task t j The set of agent alliances; is the set of agent alliances that have not been assigned tasks; a and b are constant weights that determine the proportion of resource overlap and path cost in the benefits, respectively; I is the number of resources available to the agent in the alliance set corresponding to the task, which depends on the task itself and other members in the alliance; O refers to the agent's unused resources, that is, resources that cannot be used after joining the alliance; K and P are constant weights that determine their proportion in the value benefits, respectively; the path cost d refers to the Euclidean distance between the agent and the task;

[0019] The constraints in step 1 are:

[0020] Constraint 1: Execution mode constraints

[0021]

[0022] Indicates that the intersection of the alliance sets corresponding to different tasks is an empty set, that is, the agent performs at most one task;

[0023] Constraint 2: Task requirement constraints

[0024]

[0025] in, It is the resource type and number after quantifying the capabilities of the agent. There are l types of resources in total. It is the agent u i The number of k-th resources carried; It is the resource type and quantity after quantifying the task requirements. It is task t j The number of k-th resources required;

[0026] Step 2: Establish a coalition formation game model: The game model includes a set of agents, a set of tasks, and the agent's preference relationship for tasks; define the preference relationship and transform the task allocation problem into a coalition division problem; determine the game target solution and define the final Nash equilibrium solution;

[0027] Step 3: Solve and check based on maximum weighted matching and agent transfer alliance rules, and realize the final alliance structure layer by layer, including: assigning agents to tasks layer by layer based on the maximum weighted matching algorithm; designing agent transfer alliance rules, checking the assigned agents, updating the resources required for agents and tasks based on the inspection results, repeating the maximum weighted matching algorithm, and iterating in sequence until all agents are stable.

[0028] Specifically, the preference relationship is defined as:

[0029] Based on the definition of the agent's benefit function in step 1, when the agent u i Join the Alliance The benefits are greater than or equal to joining the alliance When the income in

[0030] Right now

[0031] Among them, the preference relationship between the agent and the task is for the benefit of a single agent, that is, the other agent members in the alliance are fixed.

[0032] The final Nash equilibrium solution is defined as:

[0033]

[0034] in, It is an alliance structure CS * Medium Agent u i The alliance to which you belong, Does not contain u i Any other set of alliances.

[0035] Furthermore, the step of assigning agents to tasks layer by layer based on maximum weighted matching specifically includes:

[0036] Step 3-11. Initialize the assignment result, mark all m tasks as unassigned, and the corresponding alliance structure is an empty set; all agents are not assigned tasks, at this time

[0037]

[0038] Step 3-12. When the alliance structure corresponding to each task is an empty set, calculate the benefit of each agent performing each task individually, and obtain the M*N dimensional benefit matrix V;

[0039] Step 3-13. Convert the payoff matrix into a weighted bipartite graph, initialize the top label, set the task vertex to the weight x[i] with the maximum payoff within its corresponding agent, and preset the agent vertex to 0; begin matching. The maximum weighted matching principle is that the sum of the vertex values ​​must be equal to the weight w[i][j] of the edges connected to it. Utilize the core idea of ​​the KM algorithm to continuously modify the top label and conduct in-depth exploration to find a feasible augmenting path. The algorithm vertex matching principle is:

[0040] x[i]+y[j]=w[i][j] (11)

[0041] Since the number of tasks is smaller than the number of agents, virtual nodes need to be added and their weights set to a very small value so that they do not affect the normal matching of other points. The final processing of the matching results is to delete the matching parts of the virtual nodes. At this time, each task is matched to one agent, which satisfies the maximum matching.

[0042] Step 3-14. Update the remaining resources required for the task based on the matching results.

[0043] Furthermore, the design of the agent transfer alliance rules and the inspection of the assigned agents specifically include:

[0044] Step 3-21. After the initial allocation results are obtained at the first level, the assigned agents are checked and their benefits are compared with other tasks after the required resources are updated. If and only if there is another task with the updated required resources where the benefits are still greater than the agent's benefits in the original alliance, the agent exits the original alliance structure and waits for a new match. Otherwise, the first-level allocation results are valid and the agent has reached a stable state.

[0045] Step 3-22. Assign the agents that have reached a stable state to their respective alliances based on the inspection results, and update the remaining unassigned agent set. and the remaining resource types and quantities required for each task, to obtain a new payoff matrix, and repeat the maximum weighted matching using steps 3-13;

[0046] Step 3-23. Check the benefits after updating the remaining resources only for the newly matched agent, and repeat step 3-21;

[0047] Step 3-24. Algorithm convergence condition: Each agent reaches a stable state, that is, after inspection, it will not increase its own benefits by changing its alliance, and the algorithm ends; otherwise, return to step 3-22.

[0048] The present invention provides a distributed task allocation device for a heterogeneous agent cluster based on a coalition formation game, comprising:

[0049] The task allocation model building module is used to: Based on the task allocation scenario of heterogeneous agent clusters, consider the specific resource requirements of the tasks and the differences in the ability of the agents to perform tasks, quantify the heterogeneity, establish a multi-objective function, and complete the modeling of the task allocation problem in heterogeneous conditions:

[0050] Among them, the optimization goals of the agent when performing task allocation are specifically:

[0051] maxr CS (1)

[0052]

[0053] Considering the overlap between tasks and heterogeneous resources within the alliance and the path cost of executing tasks, the agent benefit function is designed.

[0054]

[0055] cost(u i ,t j )=d(u i ,t j ) (5)

[0056] Where T={t1,t2,...,t m} is the task set, m is the number of tasks; U = {u 1, u2,...,u n} is the set of agents, n is the number of agents; CS is the alliance structure, To perform the task t j The set of agent alliances; is the set of agent alliances that have not been assigned tasks; a and b are constant weights that determine the proportion of resource overlap and path cost in the benefits, respectively; I is the number of resources available to the agent in the alliance set corresponding to the task, which depends on the task itself and other members in the alliance; O refers to the agent's unused resources, that is, resources that cannot be used after joining the alliance; K and P are constant weights that determine their proportion in the value benefits, respectively; the path cost d refers to the Euclidean distance between the agent and the task;

[0057] The game model building module is used to establish a coalition formation game model, define preference relationships, and transform the task allocation problem into a coalition division problem; determine the game target solution and define the final Nash equilibrium solution;

[0058] The final alliance module is used to solve and check based on maximum weighted matching and agent transfer alliance rules, and realize the final alliance structure layer by layer, including: assigning agents to tasks layer by layer based on the maximum weighted matching algorithm; designing agent transfer alliance rules, checking the assigned agents, updating the resources required for agents and tasks based on the inspection results, repeating the maximum weighted matching algorithm, and iterating in sequence until all agents are stable.

[0059] An electronic device of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the distributed task allocation method for heterogeneous intelligent agent clusters based on alliance formation game as described above is implemented.

[0060] A non-transitory computer-readable storage medium of the present invention stores a computer program thereon, which, when executed by a processor, implements the above-mentioned method for distributed task allocation of heterogeneous intelligent agent clusters based on alliance formation game.

[0061] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0062] 1. Comprehensively consider heterogeneous conditions such as task requirements and agent resources, and quantify heterogeneous characteristics into resource types and quantities. A distributed algorithm based on the alliance formation game model is proposed, in which multiple agents can coordinate resources to complete the same task. At the same time, redundant resources within the alliance are taken into consideration to reduce resource waste within the alliance and fully utilize the effectiveness of agents.

[0063] 2. No global information interaction is required. The agent only needs to exchange updated resource information with the agent that has completed the matching in the previous layer, avoiding mutual communication between all agents, improving timeliness and fault tolerance to communication failures.

[0064] 3. Compared with other distributed algorithms, this algorithm reduces the complexity of the problem, requires fewer iterations to converge to a better stable solution, and has strong scalability. Compared with random initialization followed by iteration, the algorithm's optimization is more guided. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a flow chart of a method for distributed task allocation of heterogeneous intelligent agent clusters based on alliance formation game of the present invention.

[0066] Figure 2 This is an overall flow chart of the alliance formation game algorithm according to an embodiment of the present invention.

[0067] Figure 3 This is a comparison chart of the attenuation of the benefit function of an embodiment method of the present invention and the enumeration method.

[0068] Figure 4 This is a comparison chart of the average running time of an embodiment method of the present invention relative to the enumeration method under different agent-task ratios.

[0069] Figure 5 A graph showing the attenuation of the profit function relative to the enumeration method at different agent-task ratios for an embodiment of the present invention. DETAILED DESCRIPTION

[0070] The present invention provides a method for distributed task allocation for heterogeneous agent clusters based on alliance formation game, comprising: establishing a mathematical model for task allocation, taking into account the specific resource requirements of the task and the differences in the ability of other agents to perform the task based on the task allocation scenario of the heterogeneous agent cluster, quantifying the heterogeneity, establishing a multi-objective function, and completing the modeling of the task allocation problem under heterogeneous conditions. The innovation of the present invention lies in establishing an alliance formation game model, transforming the task allocation problem into an alliance partitioning problem; secondly, solving and checking based on maximum weighted matching and agent transfer alliance rules, and finally realizing the final alliance structure layer by layer. By utilizing the distributed task allocation method for heterogeneous agent clusters, the number of iterations required for allocation is reduced, the real-time performance of task allocation is improved, and the heterogeneity of the resources required by agents and tasks is fully considered, so that the agent clusters are allocated to their respective appropriate alliances to exert their effectiveness.

[0071] This method eliminates the need for global information exchange; agents only need to exchange updated resource information with agents that have completed matching in the previous layer. This avoids inter-agent communication, improving timeliness and fault tolerance to communication failures. Compared to other distributed algorithms, this algorithm reduces problem complexity, requires fewer iterations to converge to a stable solution, and offers strong scalability. Compared to iterations following random initialization, this algorithm provides a more guided optimization approach.

[0072] Consider the task allocation problem of assigning m tasks to n heterogeneous agents. Each agent participates in the execution of at most one task, and each task needs to be performed collaboratively by multiple agents. The task set number is T = {t1, t2, ..., t m}, the agent set number is U = {u1,u2,...,u n The goal of the task allocation algorithm is to obtain a coalition structure CS that can give full play to the effectiveness of the agent cluster, reduce resource waste and lower path costs.

[0073] Agents are heterogeneous in terms of type, resources, etc., and they perform the same task t j can provide different performances, and this heterogeneity can be quantified as the different values ​​of the types and numbers of resources carried by the agents. Assuming there are l types of resources, It is the agent u i The number of resources of type k carried. Represents the specific performance of the agent, that is, the type and number of resources it carries. Similarly, the type and number of resources required for the task are recorded as It is task t j The number of k-th resources required.

[0074] like Figure 1 As shown in FIG, a distributed task allocation method for a heterogeneous agent cluster based on a coalition formation game of the present invention is used to solve the overall structure of the agent task allocation problem. With respect to the above task scenario description, the method of the embodiment of the present invention includes the following steps:

[0075] Step 1: Based on the task allocation scenario of heterogeneous intelligent agent clusters, establish a task allocation mathematical model;

[0076] Step 2: Establish a coalition formation game model, which includes a set of agents, a set of tasks, and task preference relationships;

[0077] Step 3: Solve and check based on maximum weighted matching and agent transfer alliance rules, and realize the final alliance structure layer by layer.

[0078] In step 1, the objective function and constraints of the agent when performing task allocation are specifically as follows:

[0079]

[0080] Considering the task allocation optimization goal, design the agent benefit function

[0081]

[0082] cost(u i ,t j )=d(u i ,t j ) (5)

[0083] To perform the task t j A set of agent alliances, where is the set of unassigned agent alliances. a and b are constant weights that determine the weight of resource overlap and path cost in the payoff, respectively. I is the number of resources available to the agent in the alliance corresponding to the task, which depends on the task itself and the other members of the alliance. O refers to the agent's unused resources, that is, resources that cannot be used after joining the alliance. Similarly, K and P are constant weights that determine their respective weights in the value payoff. The path cost d is the Euclidean distance between the agent and the task.

[0084] The specific constraints when allocating tasks are:

[0085] Constraint 1: Execution mode constraints

[0086]

[0087] Indicates that the intersection of the alliance sets corresponding to different tasks is an empty set, that is, the agent can perform at most one task.

[0088] Constraint 2: Task requirement constraints

[0089]

[0090] According to the above objective function and constraints, the optimization model is established as follows:

[0091]

[0092] and

[0093]

[0094] In step 2, a game model is established to transform the task planning problem into a coalition partitioning problem. The model includes a set of intelligent agents, a set of tasks, and the preference relationship between intelligent agents and tasks, G = (U, T, λ). The goal of the game is to obtain a stable coalition structure.

[0095] The definition of preference relationship is as follows:

[0096] When the agent u i Join the Alliance The benefits are greater than or equal to joining the alliance When the income in

[0097] Right now

[0098] Among them, the preference relationship between the agent and the task is for the benefit of a single agent, that is, the benefits of other member agents in the alliance are fixed.

[0099] Determine the target solution of the game and define the final Nash equilibrium solution.

[0100] The final Nash equilibrium solution is defined as:

[0101]

[0102] in, It is an alliance structure CS * Medium Agent u i The alliance to which you belong, Does not contain u i . Agents constantly switch between alliances to improve their own benefits. When no agent can improve its benefits by adjusting its alliance independently, a stable alliance structure is achieved. The goal of the alliance formation game is to obtain a stable alliance structure, which is consistent with the goal of task allocation. Therefore, the solution obtained through this game is the solution to the task allocation problem.

[0103] In step 3, the solution and check are performed based on the maximum weighted matching and agent transfer alliance rules, and the final alliance structure is realized layer by layer. The specific steps are:

[0104] Step 3-1: Initialize the assignment result, mark all m tasks as unassigned, and the corresponding alliance structure is an empty set; all agents are not assigned tasks, at this time

[0105] in, is the set of coalitions of agents that have not been assigned tasks, and U is the set of all agents.

[0106] Assuming the coalition structure corresponding to each task is an empty set, calculate the payoff of each agent performing each task individually, resulting in an M*N-dimensional payoff matrix V. Convert the payoff matrix into a weighted bipartite graph, initialize the top index, and set the task vertex to the weight x[i] with the maximum payoff within its corresponding agent. The agent vertex is preset to 0. Matching begins, following the maximum weighted matching principle, which states that the sum of the vertex values ​​must equal the weight of the edges connecting them. Leveraging the core concept of the KM algorithm, we continuously modify the top index and conduct deep exploration to find feasible augmenting paths.

[0107] Step 3-2: Design the agent transfer alliance rules, check the assigned agents, update the resources required for the agents and tasks based on the inspection results, repeat the maximum weighted matching algorithm, and iterate in sequence until all agents are stable, completing the agent task allocation.

[0108] After the initial allocation results are obtained at the first level, the assigned agents are checked and their benefits are compared with other tasks after the required resources are updated. If and only if there is another task with the updated required resources that still has a greater benefit than the agent's benefit in the original alliance, the agent will exit the original alliance structure and wait for a new match. Otherwise, the first-level allocation results are valid and the agent has reached a stable state. The switching rules are as follows:

[0109]

[0110] According to the inspection results, the agents that have reached a stable state are assigned to their alliances, and the set of remaining unassigned agents is updated. The remaining resource types and quantities for each task are combined to obtain a new payoff matrix, and the maximum weighted matching is repeated. Each time, only the payoff after updating the remaining resources is checked for the newly matched agents.

[0111] Algorithm convergence condition: Each agent reaches a stable state, that is, after inspection, it will not increase its own benefits by changing its alliance, and the algorithm ends.

[0112] The distributed task allocation method for heterogeneous agent clusters based on alliance formation game provided by the present invention can reduce the complexity of the problem while fully utilizing the cluster efficiency when applied to specific examples, including:

[0113] Consider a scenario consisting of three tasks, each comprised of subtasks of different types: reconnaissance, detection, and attack. The agent cluster comprises nine agents with heterogeneous capabilities, each capable of performing different functions in reconnaissance, detection, and attack. The coordinates of the agents and tasks are randomly generated within a 100×100 space. Subtask types and requirements can be quantified using constants, with different subtask types represented by A, B, and C. The specific parameters of the tasks and agents are shown in Tables 1 and 2.

[0114] Table 1. Task parameters

[0115] Task Number Coordinate information Required Category A resources Required Category B resources Required Category C resources <![CDATA[t1]]> (11.90，49.84) 3 4 5 <![CDATA[t2]]> (95.97，34.04) 2 5 6 <![CDATA[t3]]> (58.53，22.38) 6 3 4

[0116] Table 2. Agent parameters

[0117] Agent number Coordinate information Carrying Class A resources Carrying Class B resources Carrying Class C resources <![CDATA[u1]]> (75.13，225.51) 2 1 5 <![CDATA[u2]]> (50.60，69.91) 3 1 2 <![CDATA[u3]]> (89.09，95.93) 2 3 4 <![CDATA[u4]]> (54.72，13.86) 0 2 2 <![CDATA[u5]]> (14.93，25.75) 4 1 4 <![CDATA[u6]]> (84.07，25.43) 0 2 0 <![CDATA[u7]]> (81.43，24.35) 6 4 0 <![CDATA[u8]]> (92.93，35.00) 1 1 3 <![CDATA[u9]]> (19.66，25.11) 0 3 2

[0118] Simulation results:

[0119] First, we use a simpler case to verify the effectiveness of the proposed method. The constant weights in the reward function are set as follows: First, K = 1. The setting of P in this case is to ensure that the reward is positive when there are available resources after the agent joins the alliance through multiple tests, and finally obtain P = 0.1. For the path cost, after weighing the above benefits, in order to ensure that the reward function for the agent to perform tasks after joining the alliance is non-negative, the reward function in this example is adjusted to be defined as

[0120]

[0121] Among them, μ is a constant less than 0. At this time, the profit is always greater than 0. Secondly, when val(u i ,t j ) is 0, the adjusted Less than 0, meaning that when the agent u i Join the task j When an alliance cannot contribute resources, the benefits of moving to that alliance are less than in The income in .

[0122] Based on the above simulation parameter settings, the task allocation obtained by running the algorithm is shown in Table 3.

[0123] Table 3. Task allocation results

[0124] Task Number Agent number (in order of assignment) Number of redundant resources within the alliance Total alliance revenue <![CDATA[t1]]> <![CDATA[u5,u9]]> 2 12.44 <![CDATA[t2]]> <![CDATA[u3,u4]]> 0 13.23 <![CDATA[t3]]> <![CDATA[u7,u1]]> 5 13.22

[0125] Five runs were run, with an average simulation time of times = 0.21. The above results show that the resources within each coalition are sufficient to meet the task requirements; after full allocation, agent u2 does not need to participate in the task coalition, thus avoiding resource waste. Based on the above simulation parameter settings, the enumeration method is used to verify these results.

[0126] Table 4. Optimal task allocation results of enumeration method

[0127] Task Number Agent number Number of redundant resources within the alliance Total alliance revenue <![CDATA[t1]]> <![CDATA[u5,u9]]> 2 12.44 <![CDATA[t2]]> <![CDATA[u3,u4]]> 0 13.23 <![CDATA[t3]]> <![CDATA[u1,u7]]> 5 13.22

[0128] Five runs were run, with an average simulation time of times = 6.47. This demonstrates that the proposed distributed task allocation method for heterogeneous drone swarms based on coalition formation game achieves optimal task allocation while reducing the number of iterations and significantly improving the real-time performance of task allocation. Furthermore, the type and number of resources contributed by each agent in the coalition are known, eliminating the need for further coordination and facilitating task reallocation in the event of a subsequent failure.

[0129] Algorithm Performance Verification: To verify the algorithm's effectiveness and convergence, maintaining the same simulation parameters as above, we compare the total alliance revenue and average runtime achieved by using a distributed task allocation method for a heterogeneous agent cluster based on a coalition formation game with the enumeration method under varying numbers of tasks and agents. In each scenario, the number of agents exceeds the number of tasks, the total resources meet the total number of tasks, and the locations of agents and tasks are randomly distributed. Each scenario is repeated multiple times, and the average value is taken.

[0130] The method of the present invention is compared with the enumeration method in terms of the total alliance revenue. The scenarios are set with the number of tasks m∈{1,2,3,4,5,6} and the corresponding number of agents n∈{3,6,9,10,12,14}. The number of agents and task resources is randomly generated. The total number of agent resources is greater than the number of task resources, and the number of single agent resources is less than the number of single task resources.

[0131] The simulation results of the objective function attenuation of the method of the present invention compared with the enumeration method are as follows: Figure 3 As shown, the runtime comparison of the present method with the enumeration method can be seen in the average runtime in the above example. When the number of agents and tasks is small, the present method can achieve the same benefits as the enumeration method, while running in a much shorter time. As the number of agents and tasks increases, the present method still provides good performance in terms of benefits, while avoiding the explosion in the number of possible combinations caused by the increase in the number of agents and tasks, and still providing good real-time performance.

[0132] The runtime of the proposed method is related to the ratio of the number of agents to tasks. Different agent-to-task ratios were used to compare runtime and benefits using the enumeration method. The number of tasks was fixed at m = 3, and the number of agents n∈{5, 6, 7, 8, 9, 10} was chosen. The number of task resources was fixed, while the total number of agent resources was similar, showing a probabilistic distribution.

[0133] The simulation results of the comparison of the average running time and benefits of the method of the present invention with the enumeration method are as follows: Figure 4 and Figure 5 As shown in the figure, it can be seen that when the number of tasks m=3 is fixed and the number of agents n>8, the running time of the method of the present invention is less than 0.1 of the running time of the enumeration method, and at the same time it can obtain a good benefit, reaching more than 0.95 of the global optimality. For the task allocation problem of assigning m tasks to n agents, the algorithm complexity of the enumeration method is O(m×2 n ×n), which is an exponential complexity. For a larger number of agents, the amount of calculation will increase dramatically. The complexity of the method of the present invention is O(m×n 2 ), reducing the complexity of the problem.

[0134] The above simulations show that the distributed task allocation method for heterogeneous agent clusters based on alliance formation game of the present invention can obtain a better task allocation plan while ensuring low complexity, and at the same time greatly improves the real-time performance of task allocation. It can comprehensively consider the heterogeneous characteristics of agents and resources required for tasks and maximize the overlap between the resources required for tasks and the resources carried by agents in their alliances, thereby giving full play to the effectiveness of agent clusters.

Claims

1. A distributed task allocation method for heterogeneous agent clusters based on alliance formation game, characterized by: The following steps are involved: Step 1: Establish a mathematical model for task allocation: Based on the task allocation scenario of heterogeneous agent clusters, the specific resource requirements of the tasks and the differences in the ability of agents to perform tasks are considered. The heterogeneity is quantified, and a multi-objective function is established to complete the modeling of the task allocation problem in heterogeneous conditions: Among them, the optimization goals of the agent when performing task allocation are specifically: maxr CS (1) Considering the overlap between tasks and heterogeneous resources within the alliance and the path cost of executing tasks, the agent benefit function is designed. cost(u i ,t j )=d(u i ,t j ) (5) Where T={t1,t2,...,t m } is the task set, m is the number of tasks; U = {u 1, u2,...,u n } is the set of agents, n is the number of agents; CS is the alliance structure, To perform the task t j The set of agent alliances; is the set of agent alliances that have not been assigned tasks; a and b are constant weights that determine the proportion of resource overlap and path cost in the benefits, respectively; I is the number of resources available to the agent in the alliance set corresponding to the task, which depends on the task itself and other members in the alliance; O refers to the agent's unused resources, that is, resources that cannot be used after joining the alliance; K and P are constant weights that determine their proportion in the value benefits, respectively; the path cost d refers to the Euclidean distance between the agent and the task; The constraints in step 1 are: Constraint 1: Execution mode constraints Indicates that the intersection of the alliance sets corresponding to different tasks is an empty set, that is, the agent performs at most one task; Constraint 2: Task requirement constraints in, It is the resource type and number after quantifying the capabilities of the agent. There are l types of resources in total. is the agent u i The number of k-th resources carried; It is the resource type and quantity after quantifying the task requirements. It is task t j The number of k-th resources required; Step 2: Establish a coalition formation game model: The game model includes a set of agents, a set of tasks, and the agent's preference relationship for tasks; define the preference relationship and transform the task allocation problem into a coalition division problem; determine the game target solution and define the final Nash equilibrium solution; Step 3: Solve and check based on maximum weighted matching and agent transfer alliance rules, and realize the final alliance structure layer by layer, including: assigning agents to tasks layer by layer based on the maximum weighted matching algorithm; designing agent transfer alliance rules, checking the assigned agents, updating the resources required for agents and tasks based on the inspection results, repeating the maximum weighted matching algorithm, and iterating in sequence until all agents are stable.

2. A distributed task allocation method for heterogeneous agent clusters based on alliance formation game according to claim 1, characterized in that: The preference relationship is defined as: Based on the definition of the agent's benefit function in step 1, when the agent u i Join the Alliance The benefits are greater than or equal to joining the alliance When the income in Right now Among them, the preference relationship between the agent and the task is for the benefit of a single agent, that is, the other agent members in the alliance are fixed.

3. The method for distributed task allocation of heterogeneous agent clusters based on alliance formation game according to claim 1, characterized in that: The final Nash equilibrium solution is defined as: in, It is an alliance structure CS * Medium Agent u i The alliance to which you belong, Does not contain u i Any other set of alliances.

4. The method for distributed task allocation of heterogeneous agent clusters based on alliance formation game according to claim 1 is characterized in that: The method of assigning agents to tasks layer by layer based on maximum weighted matching specifically includes: Step 3-11. Initialize the assignment result, mark all m tasks as unassigned, and the corresponding alliance structure is an empty set; all agents are not assigned tasks, at this time Step 3-12. When the alliance structure corresponding to each task is an empty set, calculate the benefit of each agent performing each task individually, and obtain the M*N dimensional benefit matrix V; Step 3-13. Convert the payoff matrix into a weighted bipartite graph, initialize the top label, set the task vertex to the weight x[i] with the maximum payoff within its corresponding agent, and preset the agent vertex to 0; begin matching. The maximum weighted matching principle is that the sum of the vertex values ​​must be equal to the weight w[i][j] of the edges connected to it. Utilize the core idea of ​​the KM algorithm to continuously modify the top label and conduct in-depth exploration to find a feasible augmenting path. The algorithm vertex matching principle is: x[i]+y[j]=w[i][j] (11) Since the number of tasks is smaller than the number of agents, virtual nodes need to be added and their weights set to a very small value so that they do not affect the normal matching of other points. The final processing of the matching results is to delete the matching parts of the virtual nodes. At this time, each task is matched to one agent, which satisfies the maximum matching. Step 3-14. Update the remaining resources required for the task based on the matching results.

5. The method for distributed task allocation of heterogeneous agent clusters based on alliance formation game according to claim 1, characterized in that: The design of the agent transfer alliance rules checks the assigned agents, specifically including: Step 3-21. After the initial allocation results are obtained at the first level, the assigned agents are checked and their benefits are compared with other tasks after the required resources are updated. If and only if there is another task with the updated required resources where the benefits are still greater than the agent's benefits in the original alliance, the agent exits the original alliance structure and waits for a new match. Otherwise, the first-level allocation results are valid and the agent has reached a stable state. Step 3-22. Assign the agents that have reached a stable state to their respective alliances based on the inspection results, and update the remaining unassigned agent set. and the remaining resource types and quantities required for each task, to obtain a new payoff matrix, and repeat the maximum weighted matching using steps 3-13; Step 3-23. Check the benefits after updating the remaining resources only for the newly matched agent, and repeat step 3-21; Step 3-24. Algorithm convergence condition: Each agent reaches a stable state, that is, after inspection, it will not increase its own benefits by changing its alliance, and the algorithm ends; otherwise, return to step 3-22.

6. A distributed task allocation device for heterogeneous agent clusters based on alliance formation game, characterized in that: include: The task allocation model building module is used to: Based on the task allocation scenario of heterogeneous agent clusters, consider the specific resource requirements of the tasks and the differences in the ability of the agents to perform tasks, quantify the heterogeneity, establish a multi-objective function, and complete the modeling of the task allocation problem in heterogeneous conditions: Among them, the optimization goals of the agent when performing task allocation are specifically: maxr CS (1) Considering the overlap between tasks and heterogeneous resources within the alliance and the path cost of executing tasks, the agent benefit function is designed. cost(u i ,t j )=d(u i ,t j ) (5) Where T={t1,t2,...,t m } is the task set, m is the number of tasks; U = {u 1, u2,...,u n } is the set of agents, n is the number of agents; CS is the alliance structure, To perform the task t j The set of agent alliances; is the set of agent alliances that have not been assigned tasks; a and b are constant weights that determine the proportion of resource overlap and path cost in the benefits, respectively; I is the number of resources available to the agent in the alliance set corresponding to the task, which depends on the task itself and other members in the alliance; O refers to the agent's unused resources, that is, resources that cannot be used after joining the alliance; K and P are constant weights that determine their proportion in the value benefits, respectively; the path cost d refers to the Euclidean distance between the agent and the task; The constraints are: Constraint 1: Execution mode constraints Indicates that the intersection of the alliance sets corresponding to different tasks is an empty set, that is, the agent performs at most one task; Constraint 2: Task requirement constraints in, It is the resource type and number after quantifying the capabilities of the agent. There are l types of resources in total. is the agent u i The number of k-th resources carried; It is the resource type and quantity after quantifying the task requirements. It is task t j The number of k-th resources required; The game model building module is used to establish a coalition formation game model, define preference relationships, and transform the task allocation problem into a coalition division problem; determine the game target solution and define the final Nash equilibrium solution; The final alliance module is used to solve and check based on maximum weighted matching and agent transfer alliance rules, and realize the final alliance structure layer by layer, including: assigning agents to tasks layer by layer based on the maximum weighted matching algorithm; designing agent transfer alliance rules, checking the assigned agents, updating the resources required for agents and tasks based on the inspection results, repeating the maximum weighted matching algorithm, and iterating in sequence until all agents are stable.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for distributed task allocation of a heterogeneous agent cluster based on an alliance formation game as described in any one of claims 1 to 5 is implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for distributed task allocation of a heterogeneous agent cluster based on a coalition formation game as described in any one of claims 1 to 5 is implemented.

Images