Multi-agent collaborative task allocation method based on improved particle swarm optimization algorithm with non-dominated sorting

Abstract

The present invention discloses a multi-agent collaborative task allocation method based on an improved particle swarm algorithm with non-dominated sorting, establishes a model for multi-agent collaborative task allocation, and realizes the simultaneous optimization of multiple indicators such as attack benefits, resource consumption, and damage probability. In view of the characteristics of the problem of multi-agent collaborative task allocation, the present invention combines the multi-objective particle swarm algorithm with the non-dominated sorting algorithm, incorporates the non-dominated sorting algorithm and the cross-mutation mechanism, designs a nonlinear method for improving the inertia weight value, and proposes a maximum distance method for obtaining the optimal solution of the Pareto solution obtained by the algorithm, which greatly improves the global search and optimization ability of the algorithm and its engineering application value. Compared with traditional optimization methods, the method of the present invention has better convergence and accuracy when solving multi-objective optimization problems in dynamic environments.

Description

Technical Field

[0001] The present invention belongs to the field of multi-agent collaborative control, and in particular relates to a multi-agent collaborative task allocation method based on an improved particle swarm algorithm with non-dominated sorting. Background Art

[0002] With the development of unmanned technology, intelligent agents are gradually being able to replace humans in performing various complex tasks in dull, harsh, and dangerous environments, such as rescue detection, large-scale search, air defense suppression and strike, electronic attack, intelligence reconnaissance and surveillance. Such complex tasks cannot be completed by a single agent alone; they often require the collaborative and coordinated efforts of multiple agents. Therefore, the rationality and effectiveness of task allocation schemes are crucial to the overall operation.

[0003] To effectively allocate tasks among multiple agents, scholars at home and abroad have conducted a number of studies. Current research primarily relies on intelligent optimization algorithms, intelligent search algorithms, and other approaches to solve problems. For example, when using genetic algorithms to solve problems, shortcomings such as non-standard encoding and premature convergence are common. While the most common particle swarm algorithm addresses the issue of non-standard encoding in genetic algorithms, it can be difficult to adjust parameters and cannot effectively balance the global and local search capabilities of particles. Furthermore, for specific multi-objective optimization problems, common intelligent optimization algorithms generally transform multiple objectives into single-objective optimization problems through weighted approaches. Due to the subjectivity of weight selection, this can easily lead to problems where multiple optimization objectives cannot be properly balanced. Summary of the Invention

[0004] The purpose of the present invention is to address the problems existing in the above-mentioned prior art and provide a multi-agent collaborative task allocation method based on an improved particle swarm optimization algorithm based on non-dominated sorting, the method comprising the following steps:

[0005] Step 1: Combine the working environment information to establish the target benefit model of multi-agent task allocation;

[0006] Step 2: Combined with the working environment information, establish a loss cost model for multi-agent task allocation;

[0007] Step 3: Based on the multi-objective functions obtained in steps 1 and 2 and combined with the constraints in the agent's task execution process, a multi-agent collaborative task allocation model is established;

[0008] Step 4, using the improved particle swarm optimization algorithm based on non-dominated sorting to solve the model obtained in step 3 and obtain the Pareto solution set;

[0009] Step 5: Based on the Pareto solution set obtained in step 4, the Pareto optimal solution is obtained by the maximum distance method.

[0010] Preferably, step 1 establishes a target benefit model for multi-agent task execution based on the operating environment information, specifically including:

[0011] The equipment model used by the i-th agent to perform the task on target j is Equipment suitability is The strike rate against target j is Then the damage degree P of the i-th agent when performing the task on target j is ij It is determined by both, namely The value corresponding to target j is V j , then the target value benefit of the agent formation is:

[0012] P ij ×V j

[0013] Therefore, the overall benefit model of the task allocation of the entire multi-agent system is as follows:

[0014]

[0015] Where M is the number of agents, N is the number of targets, V j is the value of target j, V max Represents the maximum value of the target; X ij is the agent's allocation plan, represented by a task allocation decision matrix, which is defined as follows:

[0016]

[0017] Preferably, step 2 establishes a loss cost function for agent task allocation based on the operating environment information, specifically including:

[0018] (1) Shortest flight distance indicator f2

[0019] set up is the flight length of the i-th agent when choosing path p. Since all agents may face multiple target tasks, target k is the first target point that the agent flies to. Represents other target points after k, where T max Indicates the maximum number of goals that the agent can execute in one task; D ik is the flight distance of the i-th agent from the initial position to the k-th target point, is the flight distance of the ith agent from the kth target point to the rth target point; the flight distance of the agent formation is expressed as:

[0020]

[0021] Where, L max is the maximum flight distance of a unit agent when performing a task, M is the number of participating agents, N is the number of targets, and L max M is the normalization factor;

[0022]

[0023] Where D ikmax It represents the maximum flight range of the i-th agent from the initial position to the k-th target point; It represents the maximum flight range of the i-th agent from the k-th target point to the r-th target point; T max -1 refers to the maximum number of times the agent can perform a task;

[0024] Therefore, the shortest flight distance indicator of the agent is:

[0025]

[0026] (2) Minimum index of self-loss cost f3

[0027] The formalization of the loss cost minimum indicator is shown in the following formula:

[0028]

[0029] In the formula, the equipment model used to perform the task on target j is for Unit cost of model equipment,

[0030] (3) Maximum index of sub-goal coverage f4

[0031] The formalization of the maximum index of sub-goal coverage is shown in the following formula:

[0032]

[0033] Preferably, step 3 is based on the model obtained in steps 1 and 2, and incorporates various constraints faced by the agent formation when performing tasks to establish an overall model for multi-agent collaborative task allocation, specifically including:

[0034] Combining the above four indicators, multi-objective optimization is performed to obtain the overall evaluation function, that is, the overall model of multi-agent collaborative task allocation:

[0035] min f=[f1;f2;f3;f4]

[0036] Constraints include:

[0037] (1) Multi-agent collaboration constraint c1: In order to ensure the collaboration of multiple agents in the process of executing tasks and prevent collaboration problems such as invalid tasks, repeated tasks, and mismatch between the number of agents and the number of targets, it is necessary to add certain constraints to the established model. The main ones are the following two points:

[0038] For the agent, any target point can only be executed once by the agent at most, that is:

[0039]

[0040] For the target point, the number of tasks that the agent performs cannot exceed the task load that it can bear, that is:

[0041]

[0042] Among them, Z imax is the task load that the i-th agent can carry;

[0043] For tasks, all task numbers must be executed, namely:

[0044]

[0045] Among them, N type Represents the number of task types executed;

[0046] (2) Multi-agent operation radius constraint c2: In the process of multi-agent collaborative execution of tasks, there are certain constraints on the radius of the task, namely:

[0047]

[0048] Among them, R i ,i=1,2,...,M represents the operating radius of the i-th agent.

[0049] Preferably, step 4 solves the model obtained in step 3, and the specific steps are as follows:

[0050] Step 4.1: Input the constraints of the multi-agent task allocation according to step 3, initialize the particle swarm, randomly set the speed and position of each particle, set t = 0, and randomly generate the initial solution;

[0051] Step 4.2, solve the fitness of each particle according to the overall model of multi-agent collaborative task allocation, and store the position and fitness of the particle in the individual extreme value p of the particle. best In the best The individual position and fitness value of the best fitness value in the global extreme value g best middle;

[0052] Step 4.3, update particle position and velocity

[0053] x i,j (t+1)=x i,j (t)+v i,j (t+1)

[0054] v i,j (t+1)=ω·v i,j (t)+c1r1[p i,j -x i,j (t)]+c2r2[p g,j -x i,j (t)]

[0055] In the formula, x is the position information of the particle, v represents the velocity information of the particle, c1 and c2 are the learning factors of the particle, r1 and r2 are random numbers between (0,1), and p i,j is the local optimal value of the particle, p g,j is the global optimal value of the particle, and w is the inertia weight. The difference between the particle position and the current optimal position is used to guide the value of the inertia weight w. The size of the inertia weight w is adjusted nonlinearly according to its value. The difference between the i-th particle at moment j and the global optimal particle is for

[0056]

[0057]

[0058] Where D is the dimension of the solution space; x max 、x min are the upper and lower bounds of the particle position components respectively; represents the global optimal particle position at time j, represents the position of the i-th particle at time j; is the inertia weight of particle i at time j; w start 、w end are the initial and final values ​​of w respectively;

[0059] Step 4.4, compare the fitness value of each particle with the best position of the particle. If the difference between the two is within the preset range, the fitness value of the current particle is used as the best position of the particle; and compare all the current p best and g best , update g best ;

[0060] Step 4.5 introduces crossover and mutation operations: based on the difference X between the particle position component and the global optimal position, it determines whether to crossover and mutate, so that the particle can quickly jump out of the local optimal position; the specific steps are as follows:

[0061] ①Determine the threshold value X min , crossover rate p c and mutation rate p m ;

[0062] ② Determine the difference X of particle i i Is it less than the threshold? If yes, it will be executed; otherwise, it will be skipped.

[0063] ③ Select a random number r from [0,1] for each dimension of particle i, and the jth dimension is r ij , if r ij <p m , then perform mutation operation:

[0064] x ij =x min +(x max -x min )r

[0065] ④ Then determine the random number r corresponding to the j-th dimension of the particle ij Is it less than the crossover rate p? c , then perform a crossover operation on the j-th dimension, the crossover object is the global optimal solution, and assign the global optimal solution of the j-th dimension to the j-th dimension of the particle;

[0066] Step 4.6: Use the non-dominated sorting method to sort each objective function and select the next generation population based on the sorting results;

[0067] Step 4.7: When the algorithm reaches the preset stopping condition, stop searching and output the results; otherwise, go to step 4.3 to continue searching.

[0068] Preferably, a Pareto optimal solution is obtained by a maximum distance method, specifically including:

[0069] For each non-dominated solution in the Pareto solution set, calculate the maximum value of each objective function and determine the vector of the worst indicator function as follows:

[0070]

[0071] in, represents the value of the nth objective function in the i-th non-dominated solution;

[0072] The distance formula between each non-dominated solution and the worst indicator function is established as follows:

[0073]

[0074] We can get the distance set D={D1,D2,...,D i}, where D iDenotes the distance value obtained by the i-th non-dominated solution. The solution with the largest distance D value will be selected as the final solution. Worst_F represents the worst indicator function. Represents the value of the nth objective function in the i-th non-dominated solution, and n represents the sequence number of the objective function.

[0075] The present invention also discloses a multi-agent collaborative task allocation system based on an improved particle swarm optimization algorithm based on non-dominated sorting, the system comprising:

[0076] The first building module is used to establish a task allocation target benefit model during the agent's task execution process based on the working environment information;

[0077] The second building block is used to establish a loss cost model during the agent's task execution based on the working environment information;

[0078] The third building block is used to build an overall model of multi-agent collaborative task allocation by adding constraints during the agent's task execution based on the model obtained by the first building block and the second building block;

[0079] The solution module is used to solve the model obtained by the third construction module by using an improved particle swarm optimization algorithm based on non-dominated sorting.

[0080] The present invention also discloses a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method of the present invention when executing the computer program.

[0081] The present invention also discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method of the present invention are implemented.

[0082] Beneficial effects:

[0083] The present invention improves on the multi-objective particle swarm algorithm, guides the value of the inertia weight according to the degree of gap between the particle and the contemporary optimal particle, incorporates the non-dominated sorting algorithm and the crossover mutation mechanism, and designs a maximum distance method to select the optimal solution from the Pareto solution set. Compared with traditional algorithms, the present invention has better convergence and accuracy, improves the problem of premature maturity and easy falling into local optimality of the particle swarm algorithm, makes the distribution of Pareto solutions more uniform, and makes the selection more accurate, thus ensuring the diversity of the population and greatly improving the search performance of the algorithm. It has reference significance for the problem of multi-agent task allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 This is a flow chart of the multi-agent collaborative task allocation method based on the improved particle swarm algorithm of non-dominated sorting of the present invention. DETAILED DESCRIPTION

[0085] In order to make the purpose, technical solutions and advantages of this application more clear, this application is further described in detail below with reference to the accompanying drawings and embodiments.

[0086] In one embodiment, combined Figure 1 , provides a multi-agent collaborative task allocation method based on an improved particle swarm optimization algorithm based on non-dominated sorting, including the following steps:

[0087] Step 1: Combine the working environment information to establish the target benefit model of multi-agent task allocation;

[0088] Step 2: Combined with the working environment information, establish a loss cost model for multi-agent task allocation;

[0089] Step 3: Based on the multi-objective functions obtained in steps 1 and 2 and combined with the constraints in the agent's task execution process, a multi-agent collaborative task allocation model is established;

[0090] Step 4: Use the improved particle swarm optimization algorithm based on non-dominated sorting to solve the model obtained in step 3.

[0091] Step 5: Based on the Pareto solution set obtained in step 4, the Pareto optimal solution is obtained by the maximum distance method.

[0092] Furthermore, in one embodiment, step 1 establishes a target benefit model for the multi-agent task execution process based on the operating environment information, specifically including:

[0093] The value gain of a multi-agent attack on a single sub-target is the product of the damage degree of the sub-target and the value of the sub-target. When multi-agents perform tasks in clusters, the damage degree to the sub-target refers to the damage degree of the multi-agent cluster to the sub-target.

[0094] The equipment model used by the i-th agent to perform the task on target j is Equipment suitability is The strike rate against target j is Then the damage degree P of the i-th agent when performing the task on target j is ij It is determined by both, namely The value corresponding to target j is V j , then the target value benefit of the agent formation is:

[0095] P ij ×V j

[0096] Therefore, the overall benefit model of the task allocation of the entire multi-agent system is as follows:

[0097]

[0098] Where M is the number of agents, N is the number of targets, V j is the value of each sub-goal, V max Represents the maximum value of the target; X ij The allocation scheme for the agent can be represented by a task allocation decision matrix, which is defined as follows:

[0099]

[0100] Furthermore, in one embodiment, step 2 establishes a loss cost model for the multi-agent task execution process based on the operating environment information, specifically including:

[0101] (1) Shortest flight distance indicator f2

[0102] set up is the length of flight of the i-th agent formation when selecting path p, and the k-th target node is the first target point on this path. Represents the other nodes after the k target node, where T max Indicates the maximum number of goals that the agent can execute in one task. ik is the flight distance of the i-th agent from its initial position to the k-th target node. is the flight distance of the i-th agent from the k-target node to the r-target node. The agent formation flight distance can be expressed as:

[0103]

[0104] Where, L max is the maximum flight distance of a unit agent when performing a task, M is the number of participating agents, and L max M is the normalization factor.

[0105]

[0106] Where D ikmax It represents the maximum flight range of the i-th agent from the initial position to the k target node. It represents the maximum flight range of the i-th agent from the k-target node to the r-target node. max -1 refers to the maximum number of times the agent can perform the task.

[0107] Therefore, the shortest flight distance indicator of the agent is:

[0108]

[0109] (2) Minimum index of self-loss cost f3

[0110] The formalization of the loss cost minimum indicator is shown in the following formula:

[0111]

[0112] In the formula, the equipment model used to perform the task on target j is for Unit cost of model equipment,

[0113] (3) Maximum index of sub-goal coverage f4

[0114] The formalization of the maximum index of sub-goal coverage is shown in the following formula:

[0115]

[0116] Furthermore, in one embodiment, based on the models obtained in steps 1 and 2, the constraints in the multi-agent task execution process are added to establish an overall model of multi-agent collaborative task allocation, which specifically includes:

[0117] Combining the above four indicators, multi-objective optimization is performed to obtain the overall evaluation function, that is, the overall model of multi-agent collaborative task allocation:

[0118] min f=[f1;f2;f3;f4]

[0119] Constraints include:

[0120] (1) Multi-agent collaboration constraint c1: In order to ensure the collaboration of multiple agents in the process of executing tasks and prevent collaboration problems such as invalid tasks, repeated tasks, and mismatch between the number of agents and the number of targets, it is necessary to add certain constraints to the established model. The main ones are the following two points:

[0121] For the agent, any target point can only be executed once by the agent at most, that is:

[0122]

[0123] For the target point, the number of tasks that the agent performs cannot exceed the task load that it can bear, that is:

[0124]

[0125] Among them, Z imax is the task load that the i-th agent can carry.

[0126] For tasks, all task numbers must be executed, namely:

[0127]

[0128] Among them, N type Represents the number of task types executed.

[0129] (2) Multi-agent operating radius constraint c2: In the process of multi-agent collaborative execution of tasks, the radius of the task execution is subject to certain constraints, taking into account factors such as the fuel that the agent can carry. This also needs to be considered during the modeling process. The flight range of the task execution should meet the agent's own operating radius, that is:

[0130]

[0131] Among them, R i ,i=1,2,...,M represents the operating radius of the i-th agent.

[0132] Furthermore, in one embodiment, the step 4 uses the improved particle swarm optimization algorithm based on non-dominated sorting to solve the model obtained in step 3, specifically including:

[0133] Step 4.1: Input the constraints of the multi-agent task allocation according to step 3, initialize the particle swarm, randomly set the speed and position of each particle, set t = 0, and randomly generate the initial solution;

[0134] Step 4.2, solve the fitness of each particle according to the overall model of multi-agent collaborative task allocation, and store the position and fitness of the particle in the individual extreme value p of the particle. best In the best The individual position and fitness value of the best fitness value in the global extreme value g best middle;

[0135] Step 4.3, update particle position and velocity

[0136] x i,j (t+1)=x i,j (t)+v i,j (t+1)

[0137] v i,j (t+1)=ω·v i,j (t)+c1r1[p i,j -x i,j (t)]+c2r2[p g,j -x i,j (t)]

[0138] In the formula, x is the position information of the particle, v represents the velocity information of the particle, c1 and c2 are the learning factors of the particle, r1 and r2 are random numbers between (0,1), and p i,j is the local optimal value of the particle, p g,j is the global optimal value of the particle, and w is the inertia weight. Since the commonly used inertia weight w is usually linearly decreasing or increasing, there is no guidance from the particle on the w value during the iteration process. Therefore, in this algorithm, the difference between the particle position and the current optimal position is used to guide the value of the inertia weight w. The size of the inertia weight w is nonlinearly adjusted according to its value. The difference between the i-th particle at time j and the global optimal particle is for

[0139]

[0140]

[0141] Where D is the dimension of the solution space; x max 、x min are the upper and lower bounds of the particle position components respectively; represents the global optimal particle position at time j, represents the position of the i-th particle at time j; is the inertia weight of particle i at time j; w start 、w end are the initial and final values ​​of w respectively;

[0142] Step 4.4, compare the fitness value of each particle with the best position of the particle. If the difference between the two is within the preset range, the fitness value of the current particle is used as the best position of the particle; and compare all the current p best and g best , update g best ;

[0143] Step 4.5 introduces crossover and mutation operations: based on the difference X between the particle position component and the global optimal position, it determines whether to crossover and mutate, so that the particle can quickly jump out of the local optimum. The specific steps are as follows:

[0144] ①Determine the threshold value X min , crossover rate p c and mutation rate p m ;

[0145] ② Determine the difference X of particle i i Is it less than the threshold? If yes, it will be executed; otherwise, it will be skipped.

[0146] ③ Select a random number r from [0,1] for each dimension of particle i, and the jth dimension is r ij , if r ij <pm , then perform mutation operation:

[0147] x ij =x min +(x max -x min )r

[0148] ④ Then determine the random number r corresponding to the j-th dimension of the particle ij Is it less than the crossover rate p? c , then a crossover operation is performed on the j-th dimension, the crossover object is the global optimal solution, and the global optimal solution of the j-th dimension is assigned to the j-th dimension of the particle.

[0149] Step 4.6: Use the non-dominated sorting method to sort each objective function and select the next generation population based on the sorting results;

[0150] Step 4.7: When the algorithm reaches the preset stopping condition, stop searching and output the results; otherwise, go to step 4.3 to continue searching.

[0151] Furthermore, in one embodiment, the maximum value of each objective function is calculated based on each non-dominated solution in the obtained Pareto solution set, and the vector of the worst indicator function is determined as follows:

[0152]

[0153] in, Represents the value of the nth objective function in the i-th non-dominated solution.

[0154] The distance formula between each non-dominated solution and the worst indicator function is established as follows:

[0155]

[0156] We can get the distance set D={D1,D2,...,D i}, where D i Denotes the distance value obtained by the i-th non-dominated solution. i The solution with the largest distance D value in the set is taken as the final solution, and Worst_F represents the worst indicator function. Represents the value of the nth objective function in the i-th non-dominated solution, and n represents the sequence number of the objective function.

[0157] The present invention also discloses a multi-agent collaborative task allocation system based on an improved particle swarm optimization algorithm based on non-dominated sorting, the system comprising:

[0158] The first building module is used to establish a task allocation target benefit model during the agent's task execution process based on the working environment information;

[0159] The second building block is used to establish a loss cost model during the agent's task execution based on the working environment information;

[0160] The third building block is used to build an overall model of multi-agent collaborative task allocation by adding constraints during the agent's task execution based on the model obtained by the first building block and the second building block;

[0161] The solution module is used to solve the model obtained by the third construction module by using an improved particle swarm optimization algorithm based on non-dominated sorting.

[0162] The present invention also discloses a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method of the present invention when executing the computer program.

[0163] The present invention also discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method of the present invention are implemented.

[0164] This paper solves the multi-agent task allocation problem based on an improved particle swarm algorithm with non-dominated sorting. A cost function is established to measure the quality of task allocation solutions. Based on this model, an improved particle swarm algorithm is designed. A method for determining the inertia weight is devised, and a non-dominated sorting and crossover mutation mechanism are introduced. Finally, a method for finding the optimal solution in the Pareto set is proposed. Compared to traditional particle swarm algorithms, this algorithm improves convergence accuracy and speed, resulting in better convergence results.

[0165] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-agent collaborative task allocation method based on an improved particle swarm optimization algorithm with non-dominated sorting, the method comprising the following steps: Step 1: Combined with the working environment information, establish a target benefit model for multi-agent task allocation; specifically, it includes: The equipment model used by the i-th agent to perform the task on target j is Equipment suitability is The strike rate against target j is Then the damage degree P of the i-th agent when performing the task on target j is ij It is determined by both, namely The value corresponding to target j is V j , then the target value benefit of the agent formation is: P ij ×V j Therefore, the overall benefit model of the task allocation of the entire multi-agent system is as follows: Where M is the number of agents, N is the number of targets, V j is the value of target j, V max Represents the maximum value of the target; X ij is the agent's allocation plan, represented by a task allocation decision matrix, which is defined as follows: i∈[1,M],j∈[1,N]; Step 2: Combined with the work environment information, establish a loss cost model for multi-agent task allocation; specifically, it includes: (1) Shortest flight distance indicator f2 set up is the flight length of the i-th agent when choosing path p. Since all agents may face multiple target tasks, target k is the first target point that the agent flies to. Represents other target points after k, where T max Indicates the maximum number of goals that the agent can execute in one task; D ik is the flight distance of the i-th agent from the initial position to the k-th target point, is the flight distance of the ith agent from the kth target point to the rth target point; the flight distance of the agent formation is expressed as: Where, L max is the maximum flight distance of a unit agent when performing a task, M is the number of participating agents, N is the number of targets, and L max M is the normalization factor; Where D ikmax It represents the maximum flight range of the i-th agent from the initial position to the k-th target point; It represents the maximum flight range of the i-th agent from the k-th target point to the r-th target point; T max -1 refers to the maximum number of times the agent can perform a task; Therefore, the shortest flight distance indicator of the agent is: (2) Minimum index of self-loss cost f3 The formalization of the loss cost minimum indicator is shown in the following formula: In the formula, the equipment model used to perform the task on target j is for Unit cost of model equipment, (3) Maximum index of sub-goal coverage f4 The formalization of the maximum index of sub-goal coverage is shown in the following formula: Step 3: Based on the multi-objective functions obtained in Steps 1 and 2 and combined with the constraints in the agent's task execution process, a multi-agent collaborative task allocation model is established; specifically, it includes: Combining the above four indicators, multi-objective optimization is performed to obtain the overall evaluation function, that is, the overall model of multi-agent collaborative task allocation: min f=[f1;f2;f3;f4] Constraints include: (1) Multi-agent collaboration constraint c1: In order to ensure the collaboration of multiple agents in the process of executing tasks and prevent collaboration problems such as invalid tasks, repeated tasks, and mismatch between the number of agents and the number of targets, it is necessary to add certain constraints to the established model. The main ones are the following two points: For the agent, any target point can only be executed once by the agent at most, that is: For the target point, the number of tasks that the agent performs cannot exceed the task load that it can bear, that is: Among them, Z imax is the task load that the i-th agent can carry; For tasks, all task numbers must be executed, namely: Among them, N type Represents the number of task types executed; (2) Multi-agent operation radius constraint c2: In the process of multi-agent collaborative execution of tasks, there are certain constraints on the radius of the task, namely: Among them, R i ,i=1,2,...,M represents the operation radius of the i-th agent; Step 4, using the improved particle swarm optimization algorithm based on non-dominated sorting to solve the model obtained in step 3 and obtain the Pareto solution set; Step 5: Based on the Pareto solution set obtained in step 4, the Pareto optimal solution is obtained by the maximum distance method. The specific steps are as follows: Step 4.1: Input the constraints of the multi-agent task allocation according to step 3, initialize the particle swarm, randomly set the speed and position of each particle, set t = 0, and randomly generate the initial solution; Step 4.2, solve the fitness of each particle according to the overall model of multi-agent collaborative task allocation, and store the particle position and fitness value in the particle's individual extreme value p best In the best The individual position and fitness value of the best fitness value in the global extreme value g best middle; Step 4.3, update particle position and velocity x i,j (t+1)=x i,j (t)+v i,j (t+1) v i,j (t+1)=ω·v i,j (t)+c1r1[p i,j -x i,j (t)]+c2r2[p g,j -x i,j (t)] In the formula, x is the position information of the particle, v represents the velocity information of the particle, c1 and c2 are the learning factors of the particle, r1 and r2 are random numbers between (0,1), and p i,j is the local optimal value of the particle, p g,j is the global optimal value of the particle, and w is the inertia weight. The difference between the particle position and the current optimal position is used to guide the value of the inertia weight w. The size of the inertia weight w is adjusted nonlinearly according to its value. The difference between the i-th particle at moment j and the global optimal particle is for Where D is the dimension of the solution space; x max 、x min are the upper and lower bounds of the particle position components respectively; represents the global optimal particle position at time j, represents the position of the i-th particle at time j; is the inertia weight of particle i at time j; w start 、w end are the initial and final values ​​of w respectively; Step 4.4, compare the fitness value of each particle with the best position of the particle. If the difference between the two is within the preset range, the fitness value of the current particle is used as the best position of the particle; and compare all the current p best and g best , update g best ; Step 4.5 introduces crossover and mutation operations: based on the difference X between the particle position component and the global optimal position, it determines whether to crossover and mutate, so that the particle can quickly jump out of the local optimal position; the specific steps are as follows: ①Determine the threshold value X min , crossover rate p c and mutation rate p m ; ② Determine the difference X of particle i i Is it less than the threshold? If yes, it will be executed; otherwise, it will be skipped. ③ Select a random number r from [0,1] for each dimension of particle i, and the jth dimension is r ij , if r ij <p m , then perform mutation operation: x ij =x min +(x max -x min )r ④ Then determine the random number r corresponding to the j-th dimension of the particle ij Is it less than the crossover rate p? c , then perform a crossover operation on the j-th dimension, the crossover object is the global optimal solution, and assign the global optimal solution of the j-th dimension to the j-th dimension of the particle; Step 4.6: Use the non-dominated sorting method to sort each objective function and select the next generation population based on the sorting results; Step 4.7: When the algorithm reaches the preset stopping condition, stop searching and output the results; otherwise, go to step 4.3 to continue searching.

2. The method for obtaining the Pareto solution set obtained in step 4 according to claim 1, characterized in that: A Pareto optimal solution is obtained through a maximum distance method, which includes: For each non-dominated solution in the Pareto solution set, calculate the maximum value of each objective function and determine the vector of the worst indicator function as follows: in, represents the value of the nth objective function in the i-th non-dominated solution; The distance formula between each non-dominated solution and the worst indicator function is established as follows: We can get the distance set D={D1,D2,...,D i }, where D i represents the distance value obtained by the i-th non-dominated solution; the solution with the largest distance D value will be selected as the final solution, Worst_F represents the worst indicator function, Represents the value of the nth objective function in the i-th non-dominated solution, and n represents the sequence number of the objective function.

3. A system for multi-agent collaborative task allocation method based on the improved particle swarm optimization algorithm based on non-dominated sorting according to any one of claims 1 or 2, characterized in that: The system comprises: The first building module is used to establish a task allocation target benefit model during the agent's task execution process based on the working environment information; The second building block is used to establish a loss cost model during the agent's task execution based on the working environment information; The third building block is used to build an overall model of multi-agent collaborative task allocation by adding constraints during the agent's task execution based on the model obtained by the first building block and the second building block; The solution module is used to solve the model obtained by the third construction module by using an improved particle swarm optimization algorithm based on non-dominated sorting.

4. A computer 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 steps of the method according to any one of claims 1 to 3 are implemented.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

Images