Multi-agent target allocation method suitable for complex task scene

Abstract

The invention relates to a multi-agent target allocation method suitable for a complex task scene, and belongs to the technical field of agent clusters. The method comprises the following steps: in a complex task scene, defining an agent cluster target allocation behavior, optimizing a target, establishing a mathematical model of an agent cluster target allocation problem, and defining a constraint; splitting an agent cluster target allocation problem into a multi-agent behavior design problem and a task allocation problem; the method comprises the following steps: aiming at a multi-agent behavior design problem, orderly segmenting a scene environment, and respectively searching an approximately optimal path in each segmented sub-region; aiming at a task distribution problem, target distribution gains brought by different task distribution strategies are estimated, and a task combination generating the maximum target distribution gain is selected for task distribution. According to the strategy method provided by the invention, the approximately optimal target distribution gain can be obtained, and the balance between the target distribution effect and the calculation efficiency can be effectively achieved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent agent cluster technology, and in particular to a multi-agent target allocation method suitable for complex task scenarios. Background Art

[0002] With the continuous advancement of modern science and technology, intelligent agent technology has been widely used in many fields including industry and rescue. In particular, with the development of intelligent agent technology, large-scale intelligent agent cluster systems have gradually become a reality. Compared with a single intelligent agent, an intelligent agent cluster can exert scale effects and complete various complex tasks more efficiently and comprehensively through internal collaboration and division of labor, showing great application prospects.

[0003] However, in practical applications, intelligent agent clusters often need to perform tasks in complex and dynamic environments and face the influence of various uncertain factors.

[0004] At present, there have been some preliminary research works on the control and target allocation of intelligent agent clusters. Some scholars have proposed a cluster framework based on task decomposition and dynamic allocation, which improves the efficiency of task completion by reasonably scheduling intelligent agents to perform different subtasks. Other works focus on the target allocation problem of clusters and design navigation control strategies based on threat assessment and risk avoidance, which improves the survivability of intelligent agent clusters to a certain extent.

[0005] However, existing research results still have shortcomings when dealing with large-scale tasks. On the one hand, most existing cluster control strategies are targeted at specific application scenarios, lack versatility and scalability, and are difficult to directly apply to large-scale intelligent agent clusters under different task requirements. On the other hand, although targeted target allocation methods have achieved certain results under specific threats, when dealing with complex and changeable confrontation environments, it is often difficult to balance the relationship between target allocation and task completion, which is not conducive to the overall effectiveness of the cluster.

[0006] In addition, as the scale of intelligent agent clusters continues to expand, the problem of information interaction and decision coordination within the cluster becomes increasingly prominent. How to quickly transmit and process massive amounts of perception data and control instructions within the cluster is a key issue related to real-time and reliability. At the same time, the task allocation and target allocation decisions of large-scale clusters involve high-dimensional, strongly coupled optimization problems. The existing centralized processing mode is difficult to meet actual needs in terms of computational efficiency and robustness.

[0007] Therefore, how to design a general, efficient, and scalable method for assigning targets to intelligent agent clusters based on the characteristics of large-scale tasks has become a key challenge that needs to be solved urgently. This method should not only balance the division of labor and collaboration with risk aversion at the task level, but also weigh computational complexity and optimality at the decision-making level. At the same time, it is also necessary to build a distributed information processing and control architecture for large-scale clusters to give full play to the overall advantages of the cluster and improve task execution efficiency and survivability. Summary of the invention

[0008] In view of the above analysis, the present invention aims to disclose a multi-agent target allocation method suitable for complex task scenarios; to solve the problem of achieving target allocation of agent clusters under the condition of controllable computational complexity when agents perform large-scale tasks in complex environments.

[0009] The present invention discloses a multi-agent target allocation method suitable for complex task scenarios, comprising:

[0010] Step S1: Under complex task scenarios, define the agent cluster target allocation behavior and optimization goals, establish a mathematical model for the agent cluster target allocation problem, and define constraints;

[0011] Step S2, splitting the intelligent agent cluster target allocation problem into a multi-agent behavior design problem and a task allocation problem; wherein the multi-agent behavior design problem is used to find a nearly optimal path; and the task allocation problem is used to find a task combination with the maximum target allocation gain;

[0012] Step S3: For the multi-agent behavior design problem, the scene environment is divided in an orderly manner, and the approximately optimal path is found in each divided sub-area; for the task allocation problem, the target allocation gain brought by different task allocation strategies is estimated, and the task combination that produces the maximum target allocation gain is selected for task allocation.

[0013] Furthermore, in step S1, the agent cluster target allocation behavior defined refers to a number of agents performing tasks in the same area at the same time interval;

[0014] Defined Agent a i ∈A is the same as agent a j ∈A\{a i The target allocation behavior is expressed as follows:

[0015]

[0016] In the formula, δ represents the time required to execute the task; and Represents agent a i ,aj ∈A reaches the task point v k ,v h ∈V e The start time of the game; A represents the cluster of agents that perform goal allocation in the environment; V e Represents the set of all task points in the environment;

[0017] Using the Boolean variable α kh ={0,1} represents point v k ,v h Whether they are in the same area, α kh Equal to 1 means that the two mission points are in the same area, α kh A value of 0 means that the two mission points are not in the same area.

[0018] Further, in step S1, the optimization objective is defined as follows: the agent cluster A needs to execute a target allocation method that can maximize the target allocation utility;

[0019] Among them, using Π={π1,...,π i ,…,π n} represents the set of target allocation methods for each agent, π i represents the target allocation method of the ith agent, π i is a sequence {v0,e 01 ,v1,e 12 ,...,v k ,e kh ,v h}; e in the sequence kh Represents the task point v k and v h The edge between

[0020] In the optimization objective, agent a i ∈A and agent a j ∈A\{a i}At task point v k ∈V e The goal allocation behavior refers to the two appearing at the same task point in the same time period in their respective goal allocation methods. k ∈V e ;

[0021] Then, the final utility value of executing the target allocation method is:

[0022]

[0023] Furthermore, the target allocation problem of the intelligent agent cluster is to find a target allocation method with maximum utility for the intelligent agent cluster A in the environment of the undirected complete graph G. The mathematical model of the target allocation of the intelligent agent cluster established based on this is:

[0024]

[0025] In the formula, the decision variable Represents agent a i Whether it passes through the edge e kh , a i ∈A,e kh ∈E is a point v in the environment k 、v h The edge between; for any agent a i ∈A and point v in the environment k ∈V e , decision variables Represents agent a i Have you passed point v? k .

[0026] Furthermore, in step S1, the constraints of the mathematical model of the target allocation of the agent cluster are defined to include flow constraints, time constraints and target allocation behavior constraints; wherein,

[0027] Flow constraints are used to limit the feasibility of the target allocation method; flow constraints include:

[0028] a) The agent's path must be connected end to end;

[0029] b) After the agent completes the task at one point, it needs to move to the next point;

[0030] c) Each task point must be visited once and only once by the agent;

[0031] Time constraints are used to limit the time domain parameters that each agent participating in the target allocation must meet; time constraints include:

[0032] a) The time the agent takes to start from the starting point is greater than or equal to 0:

[0033] b) The total time taken by the agent to complete the task does not exceed t max :

[0034] c) The agent first reaches point v k Execute the task, then click v h Execute tasks;

[0035] Among them, t kh From the task point v k To the mission point v hThe time required for the journey, M is a sufficiently large constant; For agent a i Visit mission point v h time;

[0036] The target allocation row constraint is used to limit the number of times the agent can perform target allocation behaviors in two situations; the two situations include:

[0037] Case 1: For points v in the same area k and point v h , agent a i First reach point v in the area k Perform the task, during which agent a j Arrival point v in this area h Execute tasks, which constitute team goal allocation behavior;

[0038] Case 2: For points v in the same area k and point v h , agent a j First reach point v in the area h Perform the task, during which agent a i Arrival point v in this area k Executing tasks constitutes team goal allocation behavior.

[0039] Furthermore, in step S3, for the multi-agent behavior design problem, an approximate algorithm is used to reduce the problem scale by orderly dividing the environment, and a 4-approximate OP algorithm is executed in each sub-area to find an approximate optimal path. The results of each sub-area are integrated through a dynamic programming algorithm to form a global optimization strategy for the entire environment.

[0040] Furthermore, in step S3, the process of solving the multi-agent behavior design problem includes:

[0041] 1) The multi-agent behavior design problem is split into a weakly coupled directional motion sub-problem and a graph decomposition sub-problem;

[0042] 2) The environment is divided into sub-regions in an orderly manner through graph decomposition, and the 4-approximate OP algorithm is executed in each sub-region to find the approximate optimal path to solve the directional movement sub-problem in each sub-region;

[0043] 3) The approximate optimal path results of each sub-area are integrated through a dynamic programming algorithm to form a global optimization strategy for the entire environment.

[0044] Furthermore, the graph decomposition process includes:

[0045] (1) constructing an auxiliary graph based on the data model in step S1;

[0046] (2) selecting a landmark point in the auxiliary map;

[0047] (3) Using the selected landmarks, a graph decomposition operation is performed to divide the environment into sub-regions in an orderly manner.

[0048] Furthermore, in step S3, for the task allocation problem, a heuristic algorithm of task graph merging is adopted, and the minimum spanning tree algorithm is used to estimate the target allocation gains that may be brought about by different task allocation strategies, and the task combination that produces the maximum target allocation gain is selected through a greedy algorithm for task allocation;

[0049] Among them, the task graph refers to the task graph assigned to an agent a i ∈A and its corresponding task execution strategy. The minimum spanning tree algorithm MST is used to approximate the benefits of the task graph, and the task combination that produces the maximum target allocation gain is selected through a greedy algorithm for task allocation.

[0050] Furthermore, the specific steps of task graph merging are:

[0051] 1) Construct an initial task graph set and initialize the corresponding tree structure and data structure; the tree structure is used to store the task graph merging process;

[0052] 2) Iteratively merge the task graph to form a tree structure until all tasks are merged into the same task graph;

[0053] 3) calculating a set of task graphs with positive gains in terms of target allocation gain, and extracting a set of task graphs of the merged degree from the formed tree structure;

[0054] 4) Construct a task allocation strategy from the extracted task graph set; delete all task graphs that cannot produce positive target allocation gain; the remaining task graphs are assigned to the corresponding agents and enter the task execution process.

[0055] The multi-agent target allocation method disclosed in the present invention, which is applicable to complex task scenarios, can achieve the following beneficial effects:

[0056] (1) Realizing effective target allocation of intelligent agent clusters: By solving the multi-agent behavior design problem and the task allocation problem respectively, the present invention provides an effective strategy that enables intelligent agent clusters to achieve effective target allocation behavior when processing large-scale tasks in an adversarial environment, thereby improving the security and adversarial capability of the entire system.

[0057] (2) Controllable computational complexity: By adopting an approximate algorithm and a heuristic algorithm based on greedy thinking, the present invention effectively controls the computational complexity and ensures the feasibility of application in large-scale tasks. At the same time, it achieves a balance between efficiency and effect while ensuring the target allocation effect.

[0058] (3) Efficient task allocation and target allocation method optimization: By introducing the minimum spanning tree algorithm to estimate the target allocation gain and combining it with the greedy algorithm for task allocation, the present invention not only achieves efficient task allocation, but also optimizes the target allocation method through ordered segmentation and dynamic programming algorithms, thereby improving the safety and efficiency of task execution. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.

[0060] Figure 1 It is a flow chart of a multi-agent target allocation method applicable to complex task scenarios in an embodiment of the present invention;

[0061] Figure 2 A schematic diagram of multi-agent target allocation behavior in an embodiment of the present invention;

[0062] Figure 3 Schematic diagram of target allocation behavior determination for a multi-agent team in an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used to illustrate the principles of the present invention together with the embodiments of the present invention.

[0064] One embodiment of the present invention discloses a multi-agent target allocation method suitable for complex task scenarios, such as Figure 1 As shown, the following steps are included:

[0065] Step S1: Under complex task scenarios, define the agent cluster target allocation behavior and optimization goals, establish a mathematical model for the agent cluster target allocation problem, and define constraints;

[0066] Step S2, splitting the intelligent agent cluster target allocation problem into a multi-agent behavior design problem and a task allocation problem; wherein the multi-agent behavior design problem is used to find a nearly optimal path; and the task allocation problem is used to find a task combination with the maximum target allocation gain;

[0067] Step S3: For the multi-agent behavior design problem, the scene environment is divided in an orderly manner, and the approximately optimal path is found in each divided sub-area; for the task allocation problem, the target allocation gain brought by different task allocation strategies is estimated, and the task combination that produces the maximum target allocation gain is selected for task allocation.

[0068] The method of this embodiment can achieve a nearly optimal target allocation gain, effectively achieving a balance between the target allocation effect and the computational efficiency.

[0069] Specifically, in step S1, in a complex task scenario, an undirected complete graph G(V e , E) represents the scene environment, V e represents the set of all task points in the environment, E is the edge set between task points in the environment; A={a1,...,a i ,...,a n} represents the cluster of agents that perform goal assignment in the environment;

[0070] Specifically, in step S1, the agent cluster target allocation behavior defined refers to a number of agents performing tasks in the same area at the same time interval; Figure 2 shown.

[0071] In this embodiment, the defined agent a i ∈A is the same as agent a j ∈A\{a i The target allocation behavior is expressed as follows:

[0072]

[0073] In the formula, δ represents the time required to execute the task; and Represents agent a i ,a j ∈A reaches the task point v k ∈V e The start time of kh ={0,1} represents point v k ,v h Whether they are in the same area, α kh Equal to 1 means that the two mission points are in the same area, α kh A value of 0 means that the two mission points are not in the same area.

[0074] If the agent a i ∈A in agent a j ∈A execution point v h ∈V, the execution point v in the same areak ∈V, then the agent a is called i ,a j The target allocation behavior is performed, that is, Otherwise 0.

[0075] Specifically, in step S1, the optimization goal is defined as follows: the agent cluster A needs to execute a target allocation method that can maximize the target allocation utility;

[0076] Among them, using Π={π1,...,π i ,…,π n} represents the set of target allocation methods for each agent, π i represents the target allocation method of the ith agent, π i is a sequence {v0,e 01 ,v1,e 12 ,...,v k ,e kh ,v h}; e in the sequence kh Represents the task point v k and v h The edge between

[0077] In the optimization objective, agent a i ∈A and agent a j ∈A\{a i}At task point v k ∈V e The goal allocation behavior refers to the two appearing at the same task point in the same time period in their respective goal allocation methods. k ∈V e ;

[0078] Then, the final utility value of executing the target allocation method is:

[0079]

[0080] Specifically, in step S1, the agent cluster target allocation problem is to find a target allocation method with maximum utility for the agent cluster A in the environment of the undirected complete graph G. The mathematical model of agent cluster target allocation established based on this is:

[0081]

[0082] In the formula, the decision variable Represents agent a i Whether it passes through the edge e kh , a i ∈A,e kh∈E is a point v in the environment k 、v h The edge between; for any agent a i ∈A and point v in the environment k ∈V e , decision variables Represents agent a i Have you passed point v? k .

[0083] Specifically, in step S1, the constraints of the mathematical model of the target allocation of the agent cluster are defined to include flow constraints, time constraints, and target allocation behavior constraints;

[0084] in,

[0085] Flow constraints are used to limit the feasibility of the target allocation method; flow constraints include:

[0086] a) The agent's path must be connected end to end; as shown below:

[0087]

[0088]

[0089] b) After the agent completes the task at one point, it needs to move to the next point; as shown below:

[0090]

[0091] c) Each task point must be visited once and only once by the agent; as shown below:

[0092]

[0093] Time constraints are used to limit the time domain parameters that each agent participating in the target allocation must meet; time constraints include:

[0094] a) The time the agent takes to start from the starting point is greater than or equal to 0:

[0095]

[0096] in, For agent a i The time from the starting point v0;

[0097] b) The total time taken by the agent to complete the task does not exceed t max :

[0098]

[0099] in, For agent ai Visit mission point v k Time, t k0 From the task point v k The time required to reach the starting point v0;

[0100] c) The agent is at point v k Execute the task, then click v h To perform the task, you must:

[0101]

[0102] Among them, t kh From the task point v k To the mission point v h The time required for the journey, M is a sufficiently large constant; For agent a i Visit mission point v h time;

[0103] The target allocation row constraint is used to limit the number of times the agent can perform target allocation behaviors; Figure 3 As shown, the target allocation behavior includes two cases:

[0104] Case 1: Figure 3 As shown in (a), for points v in the same area k and point v h , agent a i First reach point v in the area k Perform the task, during which agent a j Arrival point v in this area h Execute tasks, which constitute team goal allocation behavior;

[0105] Case 2: Figure 3 As shown in (b), for points v in the same area k and point v h , agent a j First reach point v in the area h Perform the task, during which agent a i Arrival point v in this area k Executing tasks constitutes team goal allocation behavior.

[0106] The target allocation row constraint is expressed mathematically as follows:

[0107]

[0108] Specifically, in step S2, in large-scale scenarios, the number of tasks that each agent needs to perform is huge, which increases the time cost of obtaining an exact solution. The agent cluster target allocation problem is divided into a multi-agent behavior design problem and a task allocation problem; the multi-agent behavior design problem is used to find an approximately optimal path; the task allocation problem is used to find the task combination with the maximum target allocation gain.

[0109] Specifically, in step S3, for the multi-agent behavior design problem, an approximate algorithm is used to reduce the problem scale by orderly dividing the environment, and a 4-approximate OP algorithm is executed in each sub-area to find an approximate optimal path. The results of each sub-area are integrated through a dynamic programming algorithm to form a global optimization strategy for the entire environment.

[0110] 4-Approximate OP Algorithm See the algorithm disclosed by Blum et al. in the article “Approximation Algorithms for Orienteering and Discounted-Reward TSP”.

[0111] In large-scale scenarios, each agent needs to perform a huge number of tasks, which increases the time cost of obtaining an exact solution. In order to balance the efficiency of the solution and the effect of the solution,

[0112] In this embodiment, the process of solving the multi-agent behavior design problem includes:

[0113] 1) The multi-agent behavior design problem is split into a weakly coupled directional motion sub-problem and a graph decomposition sub-problem;

[0114] 2) The environment is divided into sub-regions in an orderly manner through graph decomposition, and the 4-approximate OP algorithm is executed in each sub-region to find the approximate optimal path to solve the directional movement sub-problem in each sub-region;

[0115] The specific graph decomposition process includes:

[0116] (1) constructing an auxiliary graph based on the data model in step S1;

[0117] (2) selecting a landmark point in the auxiliary map;

[0118] (3) Using the selected landmarks, a graph decomposition operation is performed to divide the environment into sub-regions in an orderly manner.

[0119] 3) The approximate optimal path results of each sub-area are integrated through a dynamic programming algorithm to form a global optimization strategy for the entire environment.

[0120] In the directional movement subproblem, the agent visits different nodes v in the environment i∈V obtains different benefits The agent needs to spend a total time t max By visiting nodes within the time range, the benefits obtained by the system are maximized; since the benefits obtained by carrying out the target allocation behavior will change dynamically over time; because the number of agents that visit the same point in different time periods is different, the benefits obtained by visiting the same node in different time periods are different; therefore, in this embodiment, the graph is divided by graph decomposition, and the 4-approximate OP algorithm is used on the divided graph to solve the directional motion sub-problem in each sub-graph;

[0121] More specifically, for the problem of multi-agent behavior design, the environment is divided in order according to preset rules, and in each divided sub-area, a near-optimal path is found, including:

[0122] 1) In the initial stage of the graph decomposition process, an auxiliary graph G is constructed based on the data model in step S1 A = <V A ,E A >, the point in the auxiliary graph contains a triple <r i ,[s i ,e i ]>, where r i For agent a i In the time period [s i ,e i ]Execute node The target allocation gain obtained by the task at Represents a node in the auxiliary graph and connection relationship;

[0123] The specific construction process is as follows:

[0124] a. Setting points on the auxiliary graph: adding fictitious nodes to the original mathematical model, wherein the fictitious nodes are the same points with different benefits at different access times in the original mathematical model point set, and multiple virtual points with different benefits according to the corresponding time periods;

[0125] The points in the auxiliary graph contain a triple <r i ,[s i ,e i ]>, where r i For agent a i In the time period [s i ,e i ] The target allocation gain obtained by the task at the location;

[0126] b. Set the edges on the auxiliary graph: In the constructed auxiliary graph G A = <V A , E A >, each edge represents the connection relationship between nodes and in the auxiliary graph. If the access time setting between any two nodes and is the same as the access time setting in the original mathematical model; it should be noted that since fictional nodes are added in the auxiliary graph and these nodes are at the same location, the movement time between two nodes is

[0127] 2) Select a landmark point in the auxiliary graph G A ; The landmark point n i is for any point n j ∈O in the optimal path O, and in the auxiliary graph, it satisfies the actual access time t j ≤ t i or the access deadline e j ≤ e i ; In the form of arranging in ascending order of the access deadline, the landmark points in the auxiliary graph are represented as

[0128] The constructed auxiliary graph will be decomposed into a set of sets of point sets {V i} that satisfy the actual access time limit and the access deadline limit by using the selected landmark points;

[0129] The actual access time limit is: For any real number pair <i, j> where i < j, then the nodes in the point set V i are accessed before the nodes in the point set V j ;

[0130] The access deadline limit is: In any point set V i , the point with the minimum access deadline is accessed latest.

[0131] The landmark point is for any point n j ∈O in the optimal path, and in the auxiliary graph, it satisfies the actual access time t j ≤ t i or the access deadline e j ≤ e i ; In the form of arranging in ascending order of the access deadline, the landmark points in the auxiliary graph can be represented as i ;

[0132] 3) Using the selected landmarks, perform graph decomposition and execute the 4-approximate OP algorithm in the decomposed sub-regions to find the approximate optimal path O;

[0133] Specifically, for any pair of landmarks On the top, build a set operation function This function returns all the optimal paths O whose actual access time is in the interval And the access deadline is in the interval to ensure that all point sets obtained after set operations are within the deadline Previously visited one of the nodes The last one visited by the agent in the optimal path O;

[0134] 4) Based on the dynamic programming algorithm, the set state transfer function is continuously iterated, the calculation results of the directional motion subproblem and the graph decomposition subproblem are coupled to realize the design goal allocation of the intelligent agent;

[0135] Specifically, two profit functions are defined: r(j,k,g,h,t) and R(k,t);

[0136] Among them, the function r(j,k,g,h,t) is the profit function of the 4-approximate OP algorithm to find the approximate optimal path; it means that in a time period t, the access deadline is in the interval [e j ,e k )’s optimal path to obtain the benefits;

[0137] Among them, g,h are the deadlines in the deadline interval [e j ,e k ) The set of points V visited in jk points in the auxiliary graph; j, k are points in the auxiliary graph V A point in; t is the time t≤e j ;

[0138] Function R(k,t) represents a path whose access deadline is less than e in time period t. k The benefit that can be obtained from the optimal path

[0139] The specific form of the state transition function set based on the two profit functions r(j,k,g,h,t) and R(k,t) is as follows:

[0140] R(k,t)=max{R(j,t′)+r(j,k,g,h,tt′)};

[0141] where t′∈[0,t].

[0142] Through continuous iterations, the computational results of the directional motion subproblem are coupled with those of the graph decomposition subproblem.

[0143] A more specific sub-problem merging algorithm is as follows:

[0144]

[0145]

[0146] Specifically, in step S3, for the task allocation problem, a heuristic algorithm of task graph merging is adopted, and the minimum spanning tree algorithm is used to estimate the target allocation gains that may be brought about by different task allocation strategies, and the task combination that produces the maximum target allocation gain is selected through a greedy algorithm for task allocation;

[0147] Among them, the task graph refers to the task graph assigned to an agent a i ∈A and its corresponding task execution strategy. The minimum spanning tree algorithm MST is used to approximate the benefits of the task graph, and the task combination that produces the maximum target allocation gain is selected through a greedy algorithm for task allocation.

[0148] During the execution of the task graph merging algorithm, it is necessary to calculate the target allocation gain attribute corresponding to the new task graph after the two task graphs are merged. The precise target allocation gain can be expressed as:

[0149]

[0150] Task Map The target distribution income, is the target allocation benefit after merging the two task graphs.

[0151] In the actual operation process, it is very time-consuming to calculate the above formula to obtain accurate benefits, and it is NP-hard (non-deterministic polynomial hard) to supplement the optimal task execution strategy in the merged task graph. Therefore, it is necessary to approximate the benefit attributes of the task graph to cope with the situation where there are hundreds or thousands of tasks in reality. In order to speed up the calculation process, the minimum spanning tree algorithm MST can be used to approximate the benefit attributes of the task graph. Using the above approximation method, the target allocation gain brought by the task graph merger can be approximately expressed in this section as:

[0152]

[0153] They are the task graphs approximated by the minimum spanning tree algorithm MST The target distribution income, It is the target allocation benefit after merging the two task graphs approximated by the minimum spanning tree algorithm MST.

[0154] The specific steps for task graph merging are:

[0155] 1) Construct an initial task graph set and initialize the corresponding tree structure and data structure; the tree structure is used to store the task graph merging process; in the algorithm practice process, the structure is constructed using pointers;

[0156] 2) Iteratively merge the task graphs to form a tree structure until all tasks are merged into the same task graph; during the merging process, the corresponding tree structure will be updated in real time due to changes in target allocation gains, changes in task graph attributes, and other changes;

[0157] 3) calculating a set of task graphs with positive gains in terms of target allocation gain, and extracting a set of task graphs of the merged degree from the formed tree structure;

[0158] 4) Construct a task allocation strategy from the extracted task graph set; delete all task graphs that cannot produce positive target allocation gain; the remaining task graphs are assigned to the corresponding agents and enter the task execution process.

[0159] Specifically, the task graph merging algorithm is as follows:

[0160]

[0161]

[0162] In summary, the multi-agent target allocation method applicable to complex task scenarios according to the embodiment of the present invention can achieve the following beneficial effects:

[0163] (1) Realizing effective target allocation of intelligent agent clusters: By solving the multi-agent behavior design problem and the task allocation problem respectively, the present invention provides an effective strategy that enables intelligent agent clusters to achieve effective target allocation behavior when processing large-scale tasks in an adversarial environment, thereby improving the security and adversarial capability of the entire system.

[0164] (2) Controllable computational complexity: By adopting an approximate algorithm and a heuristic algorithm based on greedy thinking, the present invention effectively controls the computational complexity and ensures the feasibility of application in large-scale tasks. At the same time, it achieves a balance between efficiency and effect while ensuring the target allocation effect.

[0165] (3) Efficient task allocation and target allocation method optimization: By introducing the minimum spanning tree algorithm to estimate the target allocation gain and combining it with the greedy algorithm for task allocation, the present invention not only achieves efficient task allocation, but also optimizes the target allocation method through ordered segmentation and dynamic programming algorithms, thereby improving the safety and efficiency of task execution.

[0166] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A multi-agent target allocation method suitable for complex task scenarios, characterized in that: include: Step S1: Under complex task scenarios, define the agent cluster target allocation behavior and optimization goals, establish a mathematical model for the agent cluster target allocation problem, and define constraints; Step S2, splitting the intelligent agent cluster target allocation problem into a multi-agent behavior design problem and a task allocation problem; wherein the multi-agent behavior design problem is used to find a nearly optimal path; and the task allocation problem is used to find a task combination with the maximum target allocation gain; Step S3: For the multi-agent behavior design problem, the scene environment is divided in an orderly manner, and the approximately optimal path is found in each divided sub-area; for the task allocation problem, the target allocation gain brought by different task allocation strategies is estimated, and the task combination that produces the maximum target allocation gain is selected for task allocation.

2. The multi-agent target allocation method suitable for complex task scenarios according to claim 1 is characterized in that: In step S1, the agent cluster target allocation behavior defined refers to a number of agents performing tasks in the same area at the same time interval; Defined Agent a i ∈A is the same as agent a j ∈A\{a i The target allocation behavior is expressed as follows: In the formula, δ represents the time required to execute the task; and Represents agent a i ,a j ∈A reaches the task point v k ,v h ∈V e The start time of the game; A represents the cluster of agents that perform goal allocation in the environment; V e Represents the set of all task points in the environment; Using the Boolean variable α kh ={0,1} represents point v k ,v h Whether they are in the same area, α kh Equal to 1 means that the two mission points are in the same area, α kh A value of 0 means that the two mission points are not in the same area.

3. The multi-agent target allocation method suitable for complex task scenarios according to claim 2 is characterized in that: In step S1, the optimization goal is defined as follows: the agent cluster A needs to execute a target allocation method that can maximize the target allocation utility; Among them, using Π={π1,...,π i ,…,π n } represents the set of target allocation methods for each agent, π i represents the target allocation method of the ith agent, π i is a sequence {v0,e 01 ,v1,e 12 ,...,v k ,e kh ,v h }; e in the sequence kh Represents the task point v k and v h The edge between In the optimization objective, agent a i ∈A and agent a j ∈A\{a i }At task point v k ∈V e The goal allocation behavior refers to the two appearing at the same task point in the same time period in their respective goal allocation methods. k ∈V e ; Then, the final utility value of executing the target allocation method is:

4. The multi-agent target allocation method suitable for complex task scenarios according to claim 3 is characterized in that: The problem of target allocation for agent clusters is to find a target allocation method with maximum utility for agent cluster A in the environment of undirected complete graph G. The mathematical model of target allocation for agent clusters established based on this is: In the formula, the decision variable Represents agent a i Whether it passes through the edge e kh , a i ∈A,e kh ∈E is a point v in the environment k 、v h The edge between; for any agent a i ∈A and point v in the environment k ∈V e , decision variables Represents agent a i Have you passed point v? k .

5. The multi-agent target allocation method applicable to complex task scenarios according to claim 4 is characterized in that: In step S1, the constraints of the mathematical model of the target allocation of the agent cluster are defined, including flow constraints, time constraints and target allocation behavior constraints; wherein, Flow constraints are used to limit the feasibility of the target allocation method; flow constraints include: a) The agent's path must be connected end to end; b) After the agent completes the task at one point, it needs to move to the next point; c) Each task point must be visited once and only once by the agent; Time constraints are used to limit the time domain parameters that each agent participating in the target allocation must meet; time constraints include: a) The time the agent takes to start from the starting point is greater than or equal to 0: b) The total time taken by the agent to complete the task does not exceed t max : c) The agent is at point v k Execute the task, then click v h Execute tasks; Among them, t kh From the task point v k To the mission point v h The time required for the journey, M is a sufficiently large constant; For agent a i Visit mission point v h time; The target allocation row constraint is used to limit the number of times the agent can perform target allocation behaviors in two situations; the two situations include: Case 1: For points v in the same area k and point v h , agent a i First reach point v in the area k Perform the task, during which agent a j Arrival point v in this area h Execute tasks, which constitute team goal allocation behavior; Case 2: For points v in the same area k and point v h , agent a j First reach point v in the area h Perform the task, during which agent a i Arrival point v in this area k Executing tasks constitutes team goal allocation behavior.

6. The multi-agent target allocation method suitable for complex task scenarios according to claim 4 is characterized in that: In step S3, for the multi-agent behavior design problem, an approximate algorithm is used to reduce the problem scale by orderly dividing the environment. The 4-approximate OP algorithm is executed in each sub-area to find the approximate optimal path. The results of each sub-area are integrated through the dynamic programming algorithm to form a global optimization strategy for the entire environment.

7. The multi-agent target allocation method applicable to complex task scenarios according to claim 6 is characterized in that: In step S3, the process of solving the multi-agent behavior design problem includes: 1) The multi-agent behavior design problem is split into a weakly coupled directional motion sub-problem and a graph decomposition sub-problem; 2) The environment is divided into sub-regions in an orderly manner through graph decomposition, and the 4-approximate OP algorithm is executed in each sub-region to find the approximate optimal path to solve the directional movement sub-problem in each sub-region; 3) The approximate optimal path results of each sub-area are integrated through a dynamic programming algorithm to form a global optimization strategy for the entire environment.

8. The multi-agent target allocation method applicable to complex task scenarios according to claim 7 is characterized in that: The graph decomposition process includes: (1) constructing an auxiliary graph based on the data model in step S1; (2) selecting a landmark point in the auxiliary map; (3) Using the selected landmarks, a graph decomposition operation is performed to divide the environment into sub-regions in an orderly manner.

9. The multi-agent target allocation method applicable to complex task scenarios according to claim 8, characterized in that: In step S3, for the task allocation problem, a heuristic algorithm of task graph merging is adopted, and the minimum spanning tree algorithm is used to estimate the target allocation gains that may be brought about by different task allocation strategies. The task combination that produces the maximum target allocation gain is selected through a greedy algorithm for task allocation; Among them, the task graph refers to the task graph assigned to an agent a i ∈A and its corresponding task execution strategy. The minimum spanning tree algorithm MST is used to approximate the benefits of the task graph, and the task combination that produces the maximum target allocation gain is selected through a greedy algorithm for task allocation.

10. The multi-agent target allocation method applicable to complex task scenarios according to claim 9, characterized in that: The specific steps for task graph merging are: 1) Construct an initial task graph set and initialize the corresponding tree structure and data structure; the tree structure is used to store the task graph merging process; 2) Iteratively merge the task graph to form a tree structure until all tasks are merged into the same task graph; 3) calculating a set of task graphs with positive gains in terms of target allocation gain, and extracting a set of task graphs of the merged degree from the formed tree structure; 4) Construct a task allocation strategy from the extracted task graph set; delete all task graphs that cannot produce positive target allocation gain; the remaining task graphs are assigned to the corresponding agents and enter the task execution process.

Images