Multi-intelligent system appointed time task allocation method and system based on game optimization

Abstract

The invention discloses a game optimization-based multi-intelligent system specified time task allocation method and system, and the method comprises the following steps: introducing a non-cooperative game theory, and converting a conventional task allocation model into a cluster unmanned system game allocation model, therefore, a task allocation problem is converted into a function optimization problem under a non-cooperative game framework; a distributed Nash equilibrium search protocol at specified time is designed, and it is ensured that the state of the unmanned system is consistent within preset time; and calculating to obtain a final decision variable, and generating a task allocation result. Compared with the prior art, the method has the advantages that task allocation rapidity is guaranteed, dynamic allocation can be completed among clusters through competition and cooperation, an optimal task allocation scheme is selected, allocation efficiency of the clusters under complex tasks and time constraints can be improved, and the method can be widely applied to scenes such as collaborative operation of various aircrafts and aircrafts.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of cluster command and control, and particularly relates to a multi-intelligent system designated time task allocation method and system based on game optimization. BACKGROUND

[0002] With the rapid development of unmanned platforms, communication networks and artificial intelligence technologies, cluster unmanned systems, as a new type of intelligent combat and operation system, are profoundly changing future action modes. The system is composed of a large number of heterogeneous or homogeneous unmanned agents through a self-organizing network, and through the cooperation between individuals, it can perform complex tasks far beyond the capabilities of a single platform. Its typical applications cover cooperative reconnaissance, saturation attack, electronic countermeasures in the military field, and wide-area search and rescue, intelligent logistics, precision agriculture in the civil field. In these applications, how to efficiently and reasonably allocate tasks to each member in the cluster, i.e., the task allocation problem, is the core key to determine the effectiveness of the entire system and is directly related to the success or failure of the task.

[0003] Compared with centralized allocation, distributed allocation can effectively avoid single-point failure risk, relieve communication bandwidth pressure, and enhance system scalability. Under the distributed allocation framework, the mainstream task allocation methods include auction allocation, behavior incentive allocation and game allocation, etc. Among them, the auction allocation method realizes task allocation by simulating the market bidding mechanism. The agent acts as a bidder and bids according to its own ability and task value. The higher the price, the better. This method can better reflect individual differences and task priorities, but its main defect is that the conflict resolution mechanism is inefficient. When multiple agents compete for the same high-value task, multiple rounds of negotiation are needed to reach an agreement, resulting in a significant increase in communication overhead and a serious impact on convergence speed. The behavior incentive allocation method guides the agent to spontaneously select the task expected by the system by designing a reasonable reward and punishment mechanism. Its advantage is that it can stimulate individual initiative, but continuous dependence on the reward mechanism may lead to a dependence phenomenon of "no incentive to work" for the node, and even some nodes may cheat rewards by means of faking task completion results, affecting the quality of task allocation, and it is difficult for a unified incentive rule to effectively adapt to the heterogeneous characteristics of nodes.

[0004] In contrast, the distributed allocation method based on game theory has unique advantages. This method models each agent as a rational participant in the game, considers task selection as a strategy space, and converts the global optimization problem into a process of solving the Nash equilibrium in a distributed manner by designing a reasonable utility function. Its advantages mainly lie in: 1) the utility function can effectively handle task conflicts and resource competition problems by describing the competition and cooperation relationship between agents; 2) by designing a solution strategy, it can flexibly handle various complex constraint conditions. However, the existing game allocation method still has significant limitations:

[0005] Firstly, the convergence time is uncontrollable, and it is difficult to realize fast real-time allocation. Most algorithms can only guarantee asymptotic convergence, that is, as time tends to infinity, the system state is infinitely close to the Nash equilibrium. This means that in practical applications, the allocation decision needs a long time to reach an acceptable accuracy, which cannot meet the task requirements with strict upper limit on decision time.

[0006] Secondly, the convergence time depends on the initial state of the system. Although some improved algorithms achieve finite time convergence, the upper bound of the convergence time depends heavily on the initial value of the agent state. Different task scenarios and initial conditions will result in different convergence times, making it difficult to predict and guarantee system performance. The uncertainty of task completion time poses challenges for subsequent task planning and cooperative control. SUMMARY

[0007] To solve the above-mentioned limitations of the cluster task allocation method, the present application provides a multi-intelligent system specified time task allocation method and system based on game optimization, which introduces non-cooperative game theory to improve the overall task allocation efficiency through competition and cooperation between clusters. It can be widely applied to various aircraft, vehicle cooperative operation and other scenarios.

[0008] The technical solution to achieve the purpose of the present application is:

[0009] A multi-intelligent system specified time task allocation method based on game optimization, comprising:

[0010] According to the actual task requirements, set the task allocation time, calculate the task income of each agent to each target and initialize the decision variables and cluster communication topology;

[0011] Based on the task income and initialized decision variables, a multi-objective allocation optimization model of cluster agents is established;

[0012] Based on the multi-objective allocation optimization model of cluster agents, a game optimization model of cluster agent target allocation is constructed;

[0013] The constraint of the target allocation game optimization problem is eliminated by the barrier function method, the adjusting dynamics of the target allocation decision variable is introduced, and the game optimization model of cluster agent target allocation is reconstructed;

[0014] Based on the consistency protocol and gradient descent algorithm, a preset time convergence mechanism and a distributed Nash equilibrium search protocol are designed;

[0015] The reconstructed game optimization model of cluster agent target allocation is solved by the preset time convergence mechanism and the distributed Nash equilibrium search protocol, and the final decision variable of cluster task allocation is generated.

[0016] Further, the multi-objective allocation optimization model of cluster agents is:

[0017]

[0018]

[0019] in Let the agent be the decision variable. Determine the target , ,otherwise , For intelligent agents Intercept target Cost estimates Let be the total number of intelligent agents. The total number of targets.

[0020] Furthermore, decision variables are introduced. Representation in the continuous domain and intelligent agents Target other members in the cluster Decision estimation Establish a game-theoretic optimization model for target allocation in clustered intelligent agents.

[0021] Furthermore, the target allocation game optimization model for the clustered intelligent agents is as follows:

[0022]

[0023]

[0024] in and They are respectively and The vector form, for In matrix form, The optimal local estimation matrix for the cost of cluster attacks. The target number corresponding to the largest element in the array is the agent. The task allocation results.

[0025] Furthermore, the reconstructed target allocation game optimization model for swarm intelligence agents is as follows:

[0026]

[0027]

[0028] in For the barrier function, To optimize the game constraints for the original attack allocation method. To constrain the total number, a designed search protocol.

[0029] Further, the constraints of the target assignment game optimization problem are eliminated by barrier function method, including: for inequality constraints, using interior point penalty function method to eliminate constraints, for equality constraints, using exterior point penalty function method to eliminate constraints.

[0030] Further, the barrier function under interior point penalty function method is:

[0031] ;

[0032] The barrier function under exterior point penalty function method is:

[0033] ;

[0034] Wherein, corresponding to the th inequality constraint, is the total number of inequality constraints; corresponding to the th equality constraint, is the total number of equality constraints.

[0035] Further, the preset time convergence mechanism and the distributed Nash equilibrium search protocol are:

[0036]

[0037]

[0038] Wherein is an element in the adjacency matrix, representing the interaction coefficient between agent and agent , which is related to the cluster topology; is the th column vector in ; is a preset time convergence mechanism set by human being, is a designed search protocol, and T is a task assignment time window.

[0039] Further, the preset time convergence mechanism uses a function designed to converge to 0 at a specified time , and the function is , is a control gain.

[0040] A multi-intelligent system specified time task assignment system based on game optimization, comprising:

[0041] An initialization unit sets a task allocation time according to actual task requirements, calculates a task benefit of each agent to each target, and initializes a decision variable and a communication topology matrix between clusters;

[0042] A multi-target allocation optimization model establishment unit establishes a multi-target allocation optimization model of the cluster agent based on the task benefit and the initialized decision variable;

[0043] A target allocation game optimization model establishment unit constructs a target allocation game optimization model of the cluster agent based on the multi-target allocation optimization model of the cluster agent, introduces a regulation dynamics of a target allocation decision variable, and reconstructs the target allocation game optimization model of the cluster agent;

[0044] A convergence mechanism and search protocol design unit designs a preset time convergence mechanism and a distributed Nash equilibrium search protocol based on a consistency protocol and a gradient descent algorithm;

[0045] A decision variable solving unit solves the reconstructed target allocation game optimization model of the cluster agent through the preset time convergence mechanism and the distributed Nash equilibrium search protocol, and generates a final decision variable of the cluster task allocation.

[0046] Compared with the prior art, the present application has the following advantages: 1) the present application introduces a representation quantity of the decision variable in a continuous domain, and converts a traditional discrete domain problem into a game optimization continuous domain problem; 2) the present application designs a convergence mechanism under a preset time, introduces the convergence mechanism under a distributed Nash equilibrium search protocol, and realizes rapid task allocation; 3) the present application solves through a dynamics model, and introduces a cooperative and competitive behavior between clusters into a game process, so that each agent maximizes the overall benefit while optimizing the task completion time; 4) the present application can enable the agents to select an optimal task allocation scheme through competition and cooperation to improve the allocation efficiency of the cluster under complex tasks and time constraints. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 is a target allocation game optimization principle diagram in the present application.

[0048] Figure 2 is a game allocation strategy solving flowchart in the present application.

[0049] Figure 3 is a game allocation decision variable simulation result diagram in the present application. DETAILED DESCRIPTION

[0050] The technical solutions of the present application will be specifically described below with reference to the drawings.

[0051] Combined with the drawings, Figure 1The application discloses a multi-intelligent system designated time task allocation method based on game optimization.

[0052] First, according to actual task requirements, a task allocation time window is set , the task income of each intelligent agent to each target is calculated and the decision variable is initialized and the adjacency matrix A of the topological relationship between clusters , if two intelligent agents have information interaction ; otherwise . The calculation of the task income adopts a known calculation method, and different known calculation methods are adopted based on different scene calculation methods, for example, the target and the intelligent agent state can be calculated.

[0053] Secondly, a cluster game allocation strategy model is established. In order to obtain a cluster intelligent agent target allocation game optimization model, a cluster intelligent agent multi-target allocation optimization model is established, and the mathematical form is as follows:

[0054]

[0055]

[0056] In the above model, the decision variable is , if the intelligent agent determines to attack the target , , otherwise . The objective function aims to minimize the attack cost, wherein is the cost estimate of the intelligent agent intercepting the target . The constraint condition considers the number of targets attacked by each intelligent agent and the allocation of each target to the intelligent agent attack. Since the model is a mixed integer centralized optimization model, and the cluster has a distributed decision feature, the is introduced in the continuous domain and the decision estimate of the intelligent agent to the target attacked by other members in the cluster, so that the distributed game allocation strategy model can be obtained. For the intelligent agent , the mathematical form of the attack allocation game optimization model is

[0057]

[0058]

[0059] in and They are respectively and The vector form, for In matrix form, For the local estimation matrix of the cost of cluster attack, the optimal... The target number corresponding to the largest element in the array is the agent. The task allocation results.

[0060] Next, a search protocol is designed such that each agent completes a search within a preset time. The internal mechanism adjusts its decision-making behavior based on local information to converge to a Nash equilibrium of objective allocation, thereby minimizing its own attack cost. The constraint of the objective allocation game optimization problem is eliminated as much as possible through the interior point method of the obstacle function, and an objective allocation decision quantity is introduced. By understanding the dynamics of regulation, a game-solving model that is easy to design with search protocols is obtained, and its mathematical form is:

[0061]

[0062]

[0063] in For the barrier function, To optimize the game constraints for the original attack allocation method. To constrain the total number, The designed search protocol is as follows. Based on the consensus protocol and gradient descent algorithm, a pre-defined time convergence mechanism and a distributed Nash equilibrium search protocol are designed as follows:

[0064]

[0065]

[0066] in For intelligent agents and intelligent agents The interaction coefficients between clusters are related to the cluster topology. for The first in Column vector; A pre-defined time convergence mechanism is used, employing a specified time. Function design that converges to 0, for example ,here To control the gain, the specific game allocation strategy solution process is as follows:Figure 2 as shown.

[0067] Finally, the cluster task allocation final decision variable is generated, and the final allocation result is obtained according to the size of the decision variable when the agent executes each task. The interception allocation decision variable The error with the Nash equilibrium is analyzed based on the Lyapunov stability analysis method, the reachability of the Nash equilibrium is analyzed, and it is proved that the convergence time meets the preset time requirement.

[0068] The application also provides a multi-intelligent system specified time task allocation system based on game optimization, which comprises:

[0069] The initialization unit sets the task allocation time according to the actual task requirement, calculates the task income of each agent to each target, and initializes the decision variable and the communication topology matrix between clusters;

[0070] The multi-target allocation optimization model establishment unit establishes a multi-target allocation optimization model of cluster agents based on the task income and the initialized decision variable;

[0071] The target allocation game optimization model establishment unit constructs a target allocation game optimization model of cluster agents based on the multi-target allocation optimization model of cluster agents, introduces the adjustment dynamics of the target allocation decision variable, and reconstructs the target allocation game optimization model of cluster agents;

[0072] The convergence mechanism and search protocol design unit designs a preset time convergence mechanism and a distributed Nash equilibrium search protocol based on the consistency protocol and the gradient descent algorithm;

[0073] The decision variable solving unit solves the reconstructed target allocation game optimization model of cluster agents through the preset time convergence mechanism and the distributed Nash equilibrium search protocol, and generates a cluster task allocation final decision variable.

[0074] Embodiment 1

[0075] In this embodiment, the UAV cluster is taken as an example, and it is assumed that 3 air attack targets need to be allocated to 3 UAVs, and all tasks need to be completed within a preset time of 10 seconds. A multi-intelligent system specified time task allocation method based on game optimization comprises the following four specific implementation steps.

[0076] Step 1: Set the task allocation time, calculate the task income, initialize the decision variable and the cluster communication topology.

[0077] Set the task specified completion time s, calculate the task income of each agent to each target and initialize the decision variable The task revenue calculation results of each agent to each target are as follows:

[0078]

[0079] The initial decision variable matrix is set as follows:

[0080]

[0081] The adjacency matrix of the topological relationship between unmanned aerial vehicles is set as follows:

[0082]

[0083] Step 2: Establish a cluster game allocation strategy model.

[0084] Based on the multi-target allocation optimization model of unmanned aerial vehicles, a game optimization model for unmanned aerial vehicle target allocation is established.

[0085] According to the value of the target, the task revenue of the unmanned aerial vehicle to its own target and the task revenue of the unmanned aerial vehicle to the target of other unmanned aerial vehicles Target allocation is performed. The purpose is to minimize the strike cost, that is:

[0086]

[0087]

[0088] Among them, the single unmanned aerial vehicle task quantity constraint is: , the target uniqueness constraint is: .

[0089] The continuous domain representation quantity ( ) and the decision estimation of the agent to the target of other members in the cluster are introduced to this model, that is, the target allocation model is converted into a distributed game optimization model, and the local optimization target of the unmanned aerial vehicle is:

[0090]

[0091]

[0092] Then the adjustment dynamics of the target allocation decision quantity is introduced, and a game solution model easy to search protocol design is obtained, and its mathematical form is

[0093]

[0094] ​

[0095] where is the barrier function, in this simulation, for inequality constraints, the interior point penalty function method is used for unconstrained solving, the initial penalty factor is 1, and the reduction factor is 0.95; for equality constraints, the exterior point penalty function method is used for unconstrained solving, the initial penalty factor is 1, and the reduction factor is 1.155. The barrier function under the interior point penalty function method is as follows:

[0096]

[0097] The barrier function under the exterior point penalty function method is as follows:

[0098]

[0099] where, corresponds to the th inequality constraint, is the total number of inequality constraints; corresponds to the th equality constraint, is the total number of equality constraints.

[0100] In the game solving model, the is the game optimization constraint for the original attack allocation to be absorbed, is the total number of constraints, is the designed search protocol.

[0101] Step 3: Design the Nash equilibrium search protocol under the preset time.

[0102] Based on the target allocation game optimization model, the gradient descent and consistency protocol are combined to solve it, so as to obtain new decision variables:

[0103]

[0104] where, is the preset time convergence mechanism, which can be set artificially according to the actual situation; .

[0105] Here, is , here is the control gain, adjusting its parameters can ensure converges to Nash equilibrium, in this simulation is 1.

[0106] Step 4: Generate the cluster task allocation result.

[0107] The final decision variables are obtained according to the search protocol designed above as follows:

[0108]

[0109] According to the decision variables obtained by solving, the final task allocation result is: the unmanned aerial vehicle 1 attacks the target 1; the unmanned aerial vehicle 2 attacks the target 2; and the unmanned aerial vehicle 3 attacks the target 3. The decision variable simulation result is shown in the following table. Figure 3 Then, under the preset time convergence mechanism and the distributed Nash equilibrium search protocol, there is a constant , so that When , the state of all unmanned aerial vehicles can converge to the Nash equilibrium of the game solving model designed by the search protocol within the preset time

[0110]

[0111] Figure 3 It is shown that the application guarantees the rapidity of task allocation, can make the clusters complete dynamic allocation through competition and cooperation, thereby selecting an optimal task allocation scheme, and is helpful to improve the allocation efficiency of the clusters under complex tasks and time constraints, and can be widely applied to various aircraft, vehicle cooperative operation and the like scenes.

[0112] Any person skilled in the art within the technical range disclosed by the application can easily think of changes or replacements, which should be covered in the protection range of the application. Therefore, the protection range of the application should be subject to the protection range described in the claims.

Claims

1. A multi-intelligent system designated time task allocation method based on game optimization, characterized in that, The application relates to a cluster intelligent agent target allocation game optimization method. According to actual task requirements, task allocation time is set, the task benefits of each intelligent agent to each target are calculated, and decision variables and cluster communication topologies are initialized; Based on the task benefits and the initialized decision variables, a cluster intelligent agent multi-target allocation optimization model is established; Based on the cluster intelligent agent multi-target allocation optimization model, a cluster intelligent agent target allocation game optimization model is constructed; Through an obstacle function method, the constraints of the target allocation game optimization problem are eliminated, and adjusting dynamics of target allocation decision variables are introduced to reconstruct the cluster intelligent agent target allocation game optimization model; Based on a consensus protocol and a gradient descent algorithm, a preset time convergence mechanism and a distributed Nash equilibrium search protocol are designed; The reconstructed cluster intelligent agent target allocation game optimization model is solved through the preset time convergence mechanism and the distributed Nash equilibrium search protocol, and final decision variables of cluster task allocation are generated.

2. The method of claim 1, wherein, The cluster intelligent agent multi-target allocation optimization model is: ； ； wherein is a decision variable, if the agent decides to attack the target , , otherwise , is a cost estimate for the agent intercepting the target , is a total number set of agents, is a total number set of targets.

3. The method of claim 2, wherein, Introducing decision variables Characteristics in continuous domain And intelligent agent Strike targets on other members in the cluster Decision estimates Establish a cluster of intelligent agent target allocation game optimization model.

4. The method of claim 3, wherein, The cluster intelligent agent target allocation game optimization model is: ； ； wherein and respectively are and in vector form, is in matrix form, is a local estimation matrix of the swarm strike cost, and the optimal corresponding to the maximum element in the matrix is the task assignment result of the intelligent agent .

5. The method of claim 4, wherein, The reconstructed cluster intelligent agent target allocation game optimization model is: ； ； wherein is a barrier function, is a constraint for the optimization of the original strike allocation game, is a constraint total number, is a designed search protocol.

6. The method of claim 5, wherein, Through the obstacle function method, the constraints of the target allocation game optimization problem are eliminated, including: for inequality constraints, an interior point penalty function method is used to eliminate constraints, and for equality constraints, an exterior point penalty function method is used to eliminate constraints.

7. The method of claim 6, wherein, The obstacle function under the interior point penalty function method is: ； The obstacle function under the exterior point penalty function method is: ； wherein, corresponding to the inequality constraint, is the total number of inequality constraints; corresponding to the equality constraint, is the total number of equality constraints.

8. The method of claim 5, wherein, The preset time convergence mechanism and the distributed Nash equilibrium search protocol are: ； ； wherein is an element in the adjacency matrix, representing the interaction coefficient between agent and agent , which is related to the cluster topology; is the th column vector in ; is a preset time convergence mechanism set by human, is the designed search protocol, and T is the task allocation time window.

9. The method of claim 8, wherein, The preset time convergence mechanism Adopting a specified time Converge to 0 function design, the function is , Is the control gain.

10. A multi-intelligent system designated time task allocation system based on game optimization, which implements the method of any one of claims 1-9, characterized in that, The application relates to a cluster intelligent agent target allocation game optimization method. An initialization unit sets task allocation time according to actual task requirements, calculates the task benefits of each intelligent agent to each target, and initializes decision variables and a communication topology matrix among clusters; A multi-target allocation optimization model establishment unit establishes a cluster intelligent agent multi-target allocation optimization model based on task benefits and initialized decision variables; A target allocation game optimization model establishment unit constructs a cluster intelligent agent target allocation game optimization model based on the cluster intelligent agent multi-target allocation optimization model, introduces adjusting dynamics of target allocation decision variables, and reconstructs the cluster intelligent agent target allocation game optimization model; A convergence mechanism and search protocol design unit designs a preset time convergence mechanism and a distributed Nash equilibrium search protocol based on a consensus protocol and a gradient descent algorithm; A decision variable solving unit solves the reconstructed cluster intelligent agent target allocation game optimization model through the preset time convergence mechanism and the distributed Nash equilibrium search protocol, and generates final decision variables of cluster task allocation.

Images