Cluster decision-making method and system based on evolution transfer learning

Abstract

The invention discloses a cluster decision-making method and system based on evolutionary transfer learning, and belongs to the technical field of local security protection, and the method comprises the following steps: S1, initializing a chasing scene, and generating a current task parameter group; s2, determining a most similar task parameter group; s3, determining a Pareto frontier solution with a fixed size according to the most similar task parameter group; s4, obtaining an optimal solution of the current task parameter group, and determining the fitness of the optimal solution; and S5, updating the positions of the pursuit agent and the escape agent. According to the method, the historical optimal solution set is dynamically screened, global optimization experience of similar tasks is inherited, the local evolution advantage of adjacent tasks is reserved, inter-generation and inter-time-sequence knowledge migration is achieved, and therefore the solving process of the current task is remarkably accelerated.

Description

Technical Field

[0001] This invention belongs to the field of local security technology, specifically relating to a cluster decision-making method and system based on evolutionary transfer learning. Background Technology

[0002] As unmanned systems become increasingly complex, single-agent systems face a dual bottleneck: high manufacturing costs and reusability. High-precision sensors and dedicated algorithm designs significantly increase the cost of a single agent, yet struggle to meet the generalization requirements of dynamic task scenarios. In contrast, swarm systems based on low-cost, simple agents exhibit enormous potential. Through distributed collaboration mechanisms, swarms can efficiently complete complex tasks (such as wide-area monitoring and collaborative encirclement) without increasing the complexity of individual agents, trading space for precision. This characteristic makes it a key technological direction for achieving a cost-performance balance in the field of near-field security. Against this backdrop, the design of the "brain" of swarm systems—the decision-making architecture—has become a strategic high ground in the competition among intelligent unmanned systems.

[0003] Current mainstream cluster decision-making architectures are mostly based on reinforcement learning paradigms, but their designs have significant limitations: centralized architectures require building a global state-action space model, leading to an exponential increase in computational complexity as the number of agents increases; while distributed architectures reduce local complexity through independent learning by individuals, they are prone to policy conflicts due to neglecting group coupling relationships. Both are limited by the strong coupling between computing resources and agent scale, severely restricting the scale of agents the system can support and making it difficult to leverage the collaborative advantages of large-scale clusters. Therefore, there is an urgent need to build a new decision-making architecture that decouples computing resources from cluster scale—through algorithm-hardware co-design, to break through the scale bottleneck of traditional methods under the same resource conditions and achieve efficient collaboration of clusters of thousands or even larger scales. In addition, existing architectures are not adaptable enough to multi-task scenarios. Since different tasks (such as target tracking and region coverage) have significant differences in environmental dynamics and target coupling, customized solution algorithms are often required. However, existing solutions lack a unified architectural framework that can accommodate diverse algorithms, leading to the need to repeatedly develop underlying modules when system tasks are expanded. By introducing a modular architecture design (such as decoupling task analysis, strategy migration, and optimization solution modules), the algorithm library can be flexibly configured and dynamically invoked, thereby significantly improving the cluster system's tolerance for complex multi-task scenarios.

[0004] It is noteworthy that even with a modular design, the efficiency of policy transfer in dynamic environments still faces significant challenges. Taking warehouse security as an example, the optimal policy trained on the same pursuit algorithm in warehouse A may experience a performance degradation of over 60% when directly deployed to warehouse B with a heterogeneous layout. However, the two scenarios share potential similarities in spatial topology (e.g., shelf distribution) and motion constraints (e.g., turning radius). If the similarity between tasks can be properly utilized to transfer decisions, the time cost of cluster learning can be significantly reduced and the decision-making process accelerated. In conclusion, designing a modular evolutionary learning architecture that measures the similarity between tasks and transfers similar decision-making strategies to achieve efficient and rapid decision-making for large-scale clusters in complex tasks is crucial. Summary of the Invention

[0005] To address the problem that existing cluster decision-making methods and systems cannot quickly and accurately solve different tasks, this invention proposes a cluster decision-making method and system based on evolutionary transfer learning.

[0006] The technical solution of this invention is: a cluster decision-making method based on evolutionary transfer learning, comprising the following steps:

[0007] S1. Initialize the pursuit scenario and generate the current task parameter group;

[0008] S2. Compare the current task parameter set with the task parameter sets in the self-evolutionary decision knowledge base to determine the most similar task parameter set;

[0009] S3. Determine the Pareto front solution of a fixed size based on the most similar task parameter set;

[0010] S4. Input the Pareto front solution of a fixed size into the solver to obtain the optimal solution for the current task parameter set, and determine the fitness of the optimal solution;

[0011] S5. Update the positions of the pursuing and fleeing agents based on the optimal solution.

[0012] Furthermore, in S1, the pursuit scenario includes the locations of the pursuing agent, the escaping agent, and obstacles;

[0013] Hunting intelligent agents The expression is:

[0014] ;

[0015] in, For the pursuit agent labeled 1 in the set of all pursuit agents, For the pursuit agent numbered 2 in the set of all pursuit agents, For all pursuing agents, the set labeled as The pursuing intelligent agent;

[0016] The expression for the fleeing agent is:

[0017] ;

[0018] in, For all escape agents, the escape agent with the index 1 is included. Let the escape agent numbered 2 be the one that escapes from the set of all escape agents. The set of all escaped agents is labeled as The escape agent;

[0019] Location of the obstacle The expression is:

[0020] ;

[0021] in, For the obstacle numbered 1 in the set of all obstacles, For the obstacle numbered 2 in the set of all obstacles, For all obstacles in the set labeled Obstacles;

[0022] In S1, Current task parameter group at any given time The expression is:

[0023] ;

[0024] ;

[0025] ;

[0026] ;

[0027] ;

[0028] in, for The x-coordinate of all pursuing agents in the global coordinate system at any given time. for The ordinate of all pursuing agents in the global coordinate system at any given time. for The x-coordinate of all fleeing agents in the global coordinate system at any given moment. for The x-coordinate of all fleeing agents in the global coordinate system at any given moment. For the number The pursuing intelligent agent in The x-coordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The x-coordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The x-coordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The ordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The ordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t.

[0029] Furthermore, S2 includes the following sub-steps:

[0030] S21. Extract the number of tasks stored in the evolutionary decision knowledge base. If the number of tasks is 0, generate an initial solution randomly and proceed to S25. Otherwise, set the current task parameter group and proceed to S22.

[0031] S22. Perform a normalized polar coordinate transformation on the current task parameter set to obtain the normalized polar coordinate distance. ,in, The polar coordinate radius of the center of the pursuing agent is set to a value. , For the angle, take The polar coordinates of the pursuing agent are denoted as ;

[0032] S23. Perform normalized polar coordinate transformation on all task parameter groups stored in the self-evolving decision knowledge base to obtain the normalized self-evolving decision knowledge base; the polar coordinates of the task with the number tempt in the normalized self-evolving decision knowledge base are... Where tempt is any task number in the self-evolutionary decision knowledge base, taken as... The corresponding polar coordinates of the pursuing agent are: ;

[0033] S24. Based on the normalized polar coordinate distance of the current task parameter group and the normalized self-evolutionary decision knowledge base, through... Calculated ,at this time and Overlap, align the task in the self-evolutionary decision knowledge base with the current task, where, These are the polar coordinate angles in the normalized current task parameter set. Let $t$ be the maximum polar coordinate angle of the $t$-th task in the knowledge base. This represents the maximum polar coordinate angle value of the current task parameter set. Let tempt be the maximum polar coordinate radius of the tempt-th task in the knowledge base. The maximum value of the polar coordinate radius of the current task parameter set. This represents the maximum polar coordinate angle in the normalized current task parameter set;

[0034] S25. Based on the alignment results, calculate the similarity using the DTW dynamic curve similarity calculation algorithm. and The curve similarity between them is used to obtain the current task. With the knowledge base Similarity score between two groups of tasks , Let be the polar coordinate radius of the pursuing agent with the number tempt, centered at the escape agent. For the tempt-th numbered angle;

[0035] S26. Determine the number (A number is used to check if the task in the self-evolving knowledge base has finished its cycle. For example, if there are 4 tasks in the knowledge base, tempt=1 and increments. Check if tempt=4. If it equals 4, the cycle is complete, and the most similar task can be determined based on the score. The symbol can be changed, but tempt was chosen to prevent duplication with other symbols.) Is the number of tasks equal to the number of tasks stored in the self-evolving decision knowledge base? If so, find... If the corresponding task parameter set is selected as the most similar task parameter set, proceed to S3; otherwise, proceed to... Add 1, return to step S21; where, The set of all similarity scores. The similarity score between the current task and the first task stored in the knowledge base. The similarity score between the current task and the second task stored in the knowledge base. For the current task and the knowledge base stored in the knowledge base Similarity score for each task.

[0036] Furthermore, S22 includes the following sub-steps:

[0037] S221. Take the average point of all escape agents in the current task parameter group as the origin of the polar coordinate system, and take the counterclockwise direction as the direction of angle growth of the polar coordinate system.

[0038] S222. Perform polar coordinate transformation on the global coordinates of the pursuing agent based on the origin to obtain new polar coordinates;

[0039] S223. Normalize the new polar coordinates, find the pursuing agent with the largest distance, and obtain the normalized polar coordinate distance.

[0040] In S221, the average point of all escaped agents in the current task parameter group The expression is:

[0041] ;

[0042] ;

[0043] in, The x-coordinate of the fleeing agent is the average x-coordinate. The average ordinate of the fleeing agent. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at that moment;

[0044] In S222, the new polar coordinates The expression is:

[0045] ;

[0046] ;

[0047] in, The distance of each pursuing agent in the current task parameter group from the origin. The angle of each pursuing agent in the current polar coordinate system. The polar coordinate radius of the first pursuing agent, with the average point of the escapees as the origin. Let the polar coordinate radius of the second pursuing agent be the average point of the escapees as the origin. For the first The polar coordinate radius of each pursuing agent is taken as the origin from the average point of the escapee. The polar coordinate angle of the first pursuing agent, with the average point of the escapee as the origin. The polar coordinate angle of the second pursuing agent is taken as the origin from the average point of the escapee. For the first The polar coordinate angle of the pursuing intelligent agent with the average point of the escapee as the origin;

[0048] In S223, find the pursuing agent with the greatest distance. The expression is:

[0049] ;

[0050] in, To obtain the maximum value;

[0051] In S223, the normalized polar coordinate distance of the current task parameter set The expression is:

[0052] ;

[0053] in, The polar coordinate radius of the pursuing agent at time t is taken as the origin from the average center of the fleeing agent. for The largest value of the radius in. Let be the polar radius of the first pursuing agent at time t, with the average center of the fleeing agents as the origin. Let be the polar coordinate radius of the second pursuing agent at time t, with the average center of the fleeing agent as the origin. For the first The polar coordinate radius of a pursuing agent at time t is taken as the origin from the average center of the fleeing agent.

[0054] Furthermore, S23 includes the following sub-steps:

[0055] S231. The average point of all escape agents in the task parameter group of the self-evolving decision knowledge base is taken as the origin of the polar coordinate system, and the counterclockwise direction is taken as the direction of angle growth of the polar coordinate system.

[0056] S232. Perform polar coordinate transformation on the global coordinates of the pursuing agent based on the origin to obtain new polar coordinates;

[0057] S233. Normalize the new polar coordinates, find the pursuing agent with the largest distance, and perform normalization operation to obtain the normalized polar coordinate distance.

[0058] In S231, the average point of all fleeing agents in the task parameter set of the self-evolving decision knowledge base. The expression is:

[0059] ;

[0060] ;

[0061] in, Let x be the x-coordinate of the average center point of the escaped agents in the tempt-th task of the self-evolutionary decision knowledge base. Let x be the x-coordinate of the average center point of the escaped agents in the tempt-th task of the self-evolutionary decision knowledge base. Let x be the x-coordinate of the first fleeing agent in the tempt-th task in the self-evolutionary decision knowledge base, in the global coordinate system. Let x be the x-coordinate of the second fleeing agent in the tempt-th task of the self-evolutionary decision knowledge base, in the global coordinate system. For the tempt-th task in the self-evolutionary decision knowledge base, the th... The x-coordinate of the fleeing agent in the global coordinate system Let be the ordinate of the first fleeing agent in the tempt-th task of the self-evolutionary decision knowledge base, in the global coordinate system. Let be the ordinate of the second fleeing agent in the tempt-th task of the self-evolutionary decision knowledge base, in the global coordinate system. For the tempt-th task in the self-evolutionary decision knowledge base, the th... The vertical coordinate of a fleeing agent in the global coordinate system;

[0062] In S232, the new polar coordinates The expression is:

[0063] ;

[0064] ;

[0065] in, Let be the distance of each pursuing agent from the origin in the task parameter set of the self-evolving decision knowledge base. Let be the angle of each pursuing agent in the current polar coordinate system within the task parameter set of the self-evolving decision knowledge base. Let be the radius in polar coordinates of the first pursuing agent in the 'tempt'th task of the self-evolving decision knowledge base. Let be the radius in polar coordinates of the second pursuing agent in the 'tempt'th task within the self-evolving decision knowledge base. For the tempt-th task in the self-evolutionary decision knowledge base, the th... The radius of the pursuing intelligent agent in polar coordinates Let be the polar coordinate angle of the first pursuing agent in the 'tempt'th task within the self-evolving decision knowledge base. Let be the polar coordinate angle of the second pursuing agent in the 'tempt'th task within the self-evolving decision knowledge base. For the tempt-th task in the self-evolutionary decision knowledge base, the th... Angle of a tracking agent in polar coordinates;

[0066] In S233, find the pursuing agent with the greatest distance. The expression is:

[0067] ;

[0068] in, To obtain the maximum value;

[0069] In S23, the normalized polar coordinate distance of the task parameter set in the self-evolutionary decision knowledge base. The expression is:

[0070] ;

[0071] in, Let be the value that maximizes the polar coordinate radius of all pursuing agents in the tempt-th task of the self-evolutionary decision knowledge base. Let be the polar coordinate radius of the first pursuing agent after normalization of the 'tempt'th task in the self-evolutionary decision knowledge base. Let be the polar coordinate radius of the second pursuing agent after normalization of the 'tempt'th task in the self-evolutionary decision knowledge base. For the normalized tempt-th task in the self-evolutionary decision knowledge base, the i-th task is... The polar coordinate radius of the pursuing agent.

[0072] Furthermore, S3 includes the following sub-steps:

[0073] S31. Extract the global optimal solution and the population optimal solution of the most similar task parameter set from the self-evolutionary decision knowledge base;

[0074] S32. Extract the global optimal solution and the population optimal solution of the task parameter set at the previous time step from the self-evolutionary decision knowledge base;

[0075] S33. Generate candidate solutions based on the global optimal solution and population optimal solution of the most similar task parameter set, as well as the global optimal solution and population optimal solution of the task parameter set at the previous time step.

[0076] S34. Generate a loss function for the candidate solutions;

[0077] S35. Based on the loss function of the candidate solutions, determine the Pareto front solution of a fixed size.

[0078] Furthermore, in S33, candidate solutions The expression is:

[0079] ;

[0080] in, The global optimal solution is the parameter set of the most similar task. The population-optimal solution for the most similar set of task parameters. This is the globally optimal solution for the task parameter set at the previous time step. This is the population-optimal solution for the task parameter set at the previous time step;

[0081] Furthermore, in S4, the fitness of the optimal solution The expression is:

[0082] ;

[0083] ;

[0084] ;

[0085] in, This can be a constant that can be modified according to the task requirements. This represents the average distance from all pursuing agents to the target's location within the encirclement. To capture the uniformity of the distribution of the target's position within the encirclement, For the first A pursuing intelligent agent at the current moment The x-coordinate in the global coordinate system For the first A pursuing intelligent agent at the current moment The x-coordinate of the endpoint position on the bounding circle in the global coordinate system. For the first A pursuing intelligent agent at the current moment The ordinate in the global coordinate system For the first A pursuing intelligent agent at the current moment The ordinate of the endpoint position on the bounding circle in the global coordinate system. The number of all pursuing agents. The angles of all endpoint positions in the encirclement, in a polar coordinate system with the escaping agent as the origin, are sorted clockwise. indivual, The angles of all endpoint positions in the encirclement, in a polar coordinate system with the escaping agent as the origin, are sorted clockwise. indivual.

[0086] Furthermore, S5 includes the following sub-steps:

[0087] S51. Move the current pursuing agent toward the optimal solution until the optimal solution is input at the next moment, and obtain the updated position of the pursuing agent after moving forward.

[0088] S52. Move the fleeing agent in a random direction and at a random speed to obtain the updated position of the fleeing agent;

[0089] S53. Determine whether all pursuing agents have reached the encirclement with the fleeing agent as the origin and a radius of dis. If so, all have reached it; otherwise, update the time and return to S2.

[0090] The encirclement is a circle with radius dis, originating from the fleeing agent. If all pursuing agents reach the encirclement, the encirclement is considered successful; otherwise, the encirclement fails, and the decision-making process is repeated until the encirclement is successful.

[0091] Based on the above methods, this invention also proposes a cluster decision-making system based on evolutionary transfer learning, including a task similarity measurement module, a multi-source decision transfer module, and a multi-objective evolutionary optimization module.

[0092] The task similarity measurement module is used to initialize the pursuit scenario, generate the current task parameter set, and compare the current task parameter set with the task parameter set in the self-evolutionary decision knowledge base to determine the most similar task parameter set;

[0093] The multi-source decision transfer module is used to determine a fixed-size Pareto front solution based on the most similar task parameter set, input the fixed-size Pareto front solution into the solver, obtain the optimal solution for the current task parameter set, and determine the fitness of the optimal solution.

[0094] The multi-objective evolution optimization module is used to update the positions of the pursuing and fleeing agents based on the optimal solution.

[0095] The beneficial effects of this invention are:

[0096] (1) In view of the limitation that traditional cluster decision-making methods are only applicable to a single scenario, this invention proposes a modular cluster decision-making method and system based on evolutionary transfer learning that can be applied to multiple tasks. It includes a task similarity measurement module, a multi-source decision transfer module, and a multi-objective evolutionary optimization module. This architecture breaks through the strong coupling characteristics between traditional algorithms and scenarios, supports users to choose evolutionary optimization algorithms according to task characteristics, and significantly improves the generalization and adaptability of the cluster system to diverse tasks.

[0097] (2) To address the bias problem of traditional methods in cross-scene migration, this invention proposes an innovative normalized polar coordinate system alignment method in the task similarity measurement module: a dynamic polar coordinate system is constructed with the centroid of the target population as the origin, and the scene offset effect is eliminated through rotation invariance processing; similarity calculation is performed using DTW dynamic curves to effectively overcome the inherent bias of traditional Euclidean space alignment methods in rotating and scaling scenes; the similarity calculation is performed between the current task and the solved tasks stored in the self-evolutionary decision knowledge base to find the predecessor task with the highest similarity in the self-evolutionary decision knowledge base.

[0098] (3) This invention integrates the population optimal solution and global optimal solution of similar tasks with the population optimal solution and global optimal solution of adjacent tasks at the previous time step, breaking through the limitation of single solution migration; by using the environment selection strategy, Pareto front solutions are extracted from the multi-source solution space as initial values, which avoids the waste of random initialization resources and significantly shortens the early exploration stage; by dynamically screening the historical optimal solution set, this mechanism inherits the global optimization experience of similar tasks and retains the local evolution advantages of adjacent tasks, realizing cross-generational and cross-temporal knowledge transfer, thereby significantly accelerating the solution process of the current task. Attached Figure Description

[0099] Figure 1 This is a flowchart of a cluster decision-making method based on evolutionary transfer learning.

[0100] Figure 2 This is an initialization scene diagram for an example.

[0101] Figure 3 The state diagram for the enclosure is provided in the embodiment.

[0102] Figure 4 An example diagram showing the transformation from a global coordinate system to a normalized polar coordinate system;

[0103] Figure 5 This is a schematic diagram illustrating the effect of angle removal order in polar coordinates.

[0104] Figure 6 A schematic diagram of the decision-making framework for the comparative methods;

[0105] Figure 7 This is a schematic diagram of the decision-making architecture of the present invention under the same input conditions;

[0106] Figure 8 This is a structural diagram of a cluster decision-making system based on evolutionary transfer learning. Detailed Implementation

[0107] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0108] like Figure 1 As shown, this invention provides a cluster decision-making method based on evolutionary transfer learning, comprising the following steps:

[0109] S1. Initialize the pursuit scenario and generate the current task parameter group;

[0110] S2. Compare the current task parameter set with the task parameter sets in the self-evolutionary decision knowledge base to determine the most similar task parameter set;

[0111] S3. Determine the Pareto front solution of a fixed size based on the most similar task parameter set;

[0112] S4. Input the Pareto front solution of a fixed size into the solver to obtain the optimal solution for the current task parameter set, and determine the fitness of the optimal solution;

[0113] S5. Update the positions of the pursuing and fleeing agents based on the optimal solution.

[0114] In this embodiment of the invention, in S1, the pursuit scenario includes the positions of the pursuing agent, the escaping agent, and the obstacles;

[0115] Hunting intelligent agents The expression is:

[0116] ;

[0117] in, For the pursuit agent labeled 1 in the set of all pursuit agents, For the pursuit agent numbered 2 in the set of all pursuit agents, For all pursuing agents, the set labeled as The pursuing intelligent agent;

[0118] The expression for the fleeing agent is:

[0119] ;

[0120] in, For all escape agents, the escape agent with the index 1 is included. Let the escape agent numbered 2 be the one that escapes from the set of all escape agents. The set of all escaped agents is labeled as The escape agent;

[0121] Location of the obstacle The expression is:

[0122] ;

[0123] in, For the obstacle numbered 1 in the set of all obstacles, For the obstacle numbered 2 in the set of all obstacles, For all obstacles in the set labeled Obstacles;

[0124] In S1, Current task parameter group at any given time The expression is:

[0125] ;

[0126] ;

[0127] ;

[0128] ;

[0129] ;

[0130] in, for The x-coordinate of all pursuing agents in the global coordinate system at any given time. for The ordinate of all pursuing agents in the global coordinate system at any given time. for The x-coordinate of all fleeing agents in the global coordinate system at any given moment. for The x-coordinate of all fleeing agents in the global coordinate system at any given moment. For the number The pursuing intelligent agent in The x-coordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The x-coordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The x-coordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The ordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The ordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t.

[0131] like Figure 2 As shown, four pursuit agents were randomly generated. One escaped intelligent agent With 2 obstacles The task is for four pursuing agents to form an encirclement around the fleeing agent, thus completing the encirclement. Figure 3 As shown, all agents must not collide with obstacles during their movement, and the paths of escaping agents are randomly generated.

[0132] Figure 2 In The task parameter set at each moment is ,in , , , .

[0133] In this embodiment of the invention, S2 includes the following sub-steps:

[0134] S21. Extract the number of tasks stored in the evolutionary decision knowledge base. If the number of tasks is 0, generate an initial solution randomly and proceed to S25. Otherwise, set the current task parameter group and proceed to S22.

[0135] S22. Perform a normalized polar coordinate transformation on the current task parameter set to obtain the normalized polar coordinate distance. ,in, The polar coordinate radius of the center of the pursuing agent is set to a value. , For the angle, take The polar coordinates of the pursuing agent are denoted as ;

[0136] S23. Perform normalized polar coordinate transformation on all task parameter groups stored in the self-evolving decision knowledge base to obtain the normalized self-evolving decision knowledge base; the polar coordinates of the task with the number tempt in the normalized self-evolving decision knowledge base are... Where tempt is any task number in the self-evolutionary decision knowledge base, taken as... The corresponding polar coordinates of the pursuing agent are: ;

[0137] S24. Based on the normalized polar coordinate distance of the current task parameter group and the normalized self-evolutionary decision knowledge base, through... Calculated ,at this time and Overlap, align the task in the self-evolutionary decision knowledge base with the current task, where, These are the polar coordinate angles in the normalized current task parameter set. Let $t$ be the maximum polar coordinate angle of the $t$-th task in the knowledge base. This represents the maximum polar coordinate angle value of the current task parameter set. Let tempt be the maximum polar coordinate radius of the tempt-th task in the knowledge base. The maximum value of the polar coordinate radius of the current task parameter set. This represents the maximum polar coordinate angle in the normalized current task parameter set;

[0138] S25. Based on the alignment results, calculate the similarity using the DTW dynamic curve similarity calculation algorithm. and The curve similarity between them is used to obtain the current task. With the knowledge base Similarity score between two groups of tasks , Let be the polar coordinate radius of the pursuing agent with the number tempt, centered at the escape agent. For the tempt-th numbered angle;

[0139] S26. Determine the number (A number is used to check if the task in the self-evolving knowledge base has finished its cycle. For example, if there are 4 tasks in the knowledge base, tempt=1 and increments. Check if tempt=4. If it equals 4, the cycle is complete, and the most similar task can be determined based on the score. The symbol can be changed, but tempt was chosen to prevent duplication with other symbols.) Is the number of tasks equal to the number of tasks stored in the self-evolving decision knowledge base? If so, find... If the corresponding task parameter set is selected as the most similar task parameter set, proceed to S3; otherwise, proceed to... Add 1, return to step S21; where, The set of all similarity scores. The similarity score between the current task and the first task stored in the knowledge base. The similarity score between the current task and the second task stored in the knowledge base. For the current task and the knowledge base stored in the knowledge base Similarity score for each task.

[0140] Extracted from the Evolutionary Decision Knowledge Base The number of tasks stored in the middle is They are numbered as follows ;like This indicates that the self-evolutionary decision knowledge base is empty and is randomly generated. Then, skip steps S23 and S24 and proceed to step S25; otherwise, set the index number. This represents the task parameter group. ,in Starting from 1, the maximum is Proceed to step S22.

[0141] Alignment and Firstly, because the distance in polar coordinates is normalized, there must exist a point with a value of 1. and Calculate the difference between the two angles. = ; then on Each element in the array is rotated and aligned. .

[0142] In this embodiment of the invention, the task parameter group Perform normalized polar coordinate transformation, as shown in the diagram below. Figure 4 As shown. S22 includes the following sub-steps:

[0143] S221. Take the average point of all escape agents in the current task parameter group as the origin of the polar coordinate system, and take the counterclockwise direction as the direction of angle growth of the polar coordinate system.

[0144] S222. Perform polar coordinate transformation on the global coordinates of the pursuing agent based on the origin to obtain new polar coordinates;

[0145] S223. Normalize the new polar coordinates, find the pursuing agent with the largest distance, and obtain the normalized polar coordinate distance.

[0146] In S221, the average point of all escaped agents in the current task parameter group The expression is:

[0147] ;

[0148] ;

[0149] in, The x-coordinate of the fleeing agent is the average x-coordinate. The average ordinate of the fleeing agent. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at that moment;

[0150] In S222, the new polar coordinates The expression is:

[0151] ;

[0152] ;

[0153] in, The distance of each pursuing agent in the current task parameter group from the origin. The angle of each pursuing agent in the current polar coordinate system. The polar coordinate radius of the first pursuing agent, with the average point of the escapees as the origin. Let the polar coordinate radius of the second pursuing agent be the average point of the escapees as the origin. For the first The polar coordinate radius of each pursuing agent is taken as the origin from the average point of the escapee. The polar coordinate angle of the first pursuing agent, with the average point of the escapee as the origin. The polar coordinate angle of the second pursuing agent is taken as the origin from the average point of the escapee. For the first The polar coordinate angle of the pursuing intelligent agent with the average point of the escapee as the origin;

[0154] In S223, find the pursuing agent with the greatest distance. The expression is:

[0155] ;

[0156] in, To obtain the maximum value;

[0157] In S223, the normalized polar coordinate distance of the current task parameter set The expression is:

[0158] ;

[0159] in, The polar coordinate radius of the pursuing agent at time t is taken as the origin from the average center of the fleeing agent. for The largest value of the radius in. Let be the polar radius of the first pursuing agent at time t, with the average center of the fleeing agents as the origin. Let be the polar coordinate radius of the second pursuing agent at time t, with the average center of the fleeing agent as the origin. For the first The polar coordinate radius of a pursuing agent at time t is taken as the origin from the average center of the fleeing agent.

[0160] In this embodiment of the invention, S23 includes the following sub-steps:

[0161] S231. The average point of all escape agents in the task parameter group of the self-evolving decision knowledge base is taken as the origin of the polar coordinate system, and the counterclockwise direction is taken as the direction of angle growth of the polar coordinate system.

[0162] S232. Perform polar coordinate transformation on the global coordinates of the pursuing agent based on the origin to obtain new polar coordinates;

[0163] S233. Normalize the new polar coordinates, find the pursuing agent with the largest distance, and perform normalization operation to obtain the normalized polar coordinate distance.

[0164] In S231, the average point of all fleeing agents in the task parameter set of the self-evolving decision knowledge base. The expression is:

[0165] ;

[0166] ;

[0167] in, Let x be the x-coordinate of the average center point of the escaped agents in the tempt-th task of the self-evolutionary decision knowledge base. Let x be the x-coordinate of the average center point of the escaped agents in the tempt-th task of the self-evolutionary decision knowledge base. Let x be the x-coordinate of the first fleeing agent in the tempt-th task in the self-evolutionary decision knowledge base, in the global coordinate system. Let x be the x-coordinate of the second fleeing agent in the tempt-th task of the self-evolutionary decision knowledge base, in the global coordinate system. For the tempt-th task in the self-evolutionary decision knowledge base, the th... The x-coordinate of the fleeing agent in the global coordinate system Let be the ordinate of the first fleeing agent in the tempt-th task of the self-evolutionary decision knowledge base, in the global coordinate system. Let be the ordinate of the second fleeing agent in the tempt-th task of the self-evolutionary decision knowledge base, in the global coordinate system. For the tempt-th task in the self-evolutionary decision knowledge base, the th... The vertical coordinate of a fleeing agent in the global coordinate system;

[0168] In S232, the new polar coordinates The expression is:

[0169] ;

[0170] ;

[0171] in, Let be the distance of each pursuing agent from the origin in the task parameter set of the self-evolving decision knowledge base. Let be the angle of each pursuing agent in the current polar coordinate system within the task parameter set of the self-evolving decision knowledge base. Let be the radius in polar coordinates of the first pursuing agent in the 'tempt'th task of the self-evolving decision knowledge base. Let be the radius in polar coordinates of the second pursuing agent in the 'tempt'th task within the self-evolving decision knowledge base. For the tempt-th task in the self-evolutionary decision knowledge base, the th... The radius of the pursuing intelligent agent in polar coordinates Let be the polar coordinate angle of the first pursuing agent in the 'tempt'th task within the self-evolving decision knowledge base. Let be the polar coordinate angle of the second pursuing agent in the 'tempt'th task within the self-evolving decision knowledge base. For the tempt-th task in the self-evolutionary decision knowledge base, the th... Angle of a tracking agent in polar coordinates;

[0172] In S233, find the pursuing agent with the greatest distance. The expression is:

[0173] ;

[0174] in, To obtain the maximum value;

[0175] In S23, the normalized polar coordinate distance of the task parameter set in the self-evolutionary decision knowledge base. The expression is:

[0176] ;

[0177] in, Let be the value that maximizes the polar coordinate radius of all pursuing agents in the tempt-th task of the self-evolutionary decision knowledge base. Let be the polar coordinate radius of the first pursuing agent after normalization of the 'tempt'th task in the self-evolutionary decision knowledge base. Let be the polar coordinate radius of the second pursuing agent after normalization of the 'tempt'th task in the self-evolutionary decision knowledge base. For the normalized tempt-th task in the self-evolutionary decision knowledge base, the i-th task is... The polar coordinate radius of the pursuing agent.

[0178] In this embodiment of the invention, S3 includes the following sub-steps:

[0179] S31. Extract the global optimal solution and the population optimal solution of the most similar task parameter set from the self-evolutionary decision knowledge base;

[0180] S32. Extract the global optimal solution and the population optimal solution of the task parameter set at the previous time step from the self-evolutionary decision knowledge base;

[0181] S33. Generate candidate solutions based on the global optimal solution and population optimal solution of the most similar task parameter set, as well as the global optimal solution and population optimal solution of the task parameter set at the previous time step.

[0182] S34. Generate a loss function for the candidate solutions;

[0183] S35. Based on the loss function of the candidate solutions, determine the Pareto front solution of a fixed size.

[0184] Extract the global optimal solution and population optimal solution of the most similar previous generation task from the self-evolutionary decision knowledge base; then extract the global optimal solution and population optimal solution of the previous time step task, merge the four solutions and calculate the fitness function, extract the Pareto front solution of a fixed size to construct a new solution.

[0185] Calculating candidate solutions based on loss function ,in It can be modified according to different mission requirements. Taking the pursuit-encirclement mission as an example, it can be... Set to:

[0186] ;

[0187] ;

[0188] in, Represents all the intelligent agents involved in the pursuit. To the target location The global average distance is the smallest. The range of values ​​is the distance from the origin of the polar coordinates. On the circle, among which This is a set constant that can be changed according to task requirements; since the candidate solutions are stored as the polar coordinate angles of the corresponding pursuing agents. ,therefore It can be calculated using the following formula:

[0189] ;

[0190] ;

[0191] This represents the uniformity of the distribution of the pursuing agents within the encirclement, where for Variables that are reordered according to size, such as Figure 5 As shown. Simultaneously, for solutions that violate the constraints, the corresponding loss is set to... This ensures that the solution is dominated by other solutions. The constraints can also be adjusted according to different tasks; for example, in a pursuit-encirclement task, they can be set as follows:

[0192]

[0193]

[0194]

[0195] The first constraint means that the target positions of the pursuing agents cannot overlap, i.e., they cannot collide; the second constraint means that the positions of the pursuing agents at the next moment after taking one step toward the current target position cannot overlap, i.e., they cannot collide; the third constraint means that the position of the pursuing agents at the next moment after taking one step toward the current target position cannot overlap with the position of the fleeing agents, i.e., they cannot collide.

[0196] Based on the calculated candidate solution loss function Find a Pareto front solution of fixed size, denoted as... The size can be set according to the scale of the problem.

[0197] In this embodiment of the invention, in S33, candidate solutions The expression is:

[0198] ;

[0199] in, The global optimal solution is the parameter set of the most similar task. The population-optimal solution for the most similar set of task parameters. This is the globally optimal solution for the task parameter set at the previous time step. This is the population-optimal solution for the task parameter set at the previous time step.

[0200] get Then, this is used as the initial solution for the task and input into the selected solver. The solver can choose different evolutionary algorithms based on the desired results, such as the common SMPSO and NSGA-II. Taking SMPSO as an example, the input task parameter set... Initialization solution Loss function Running the algorithm can yield the globally optimal solution for the current task. Population optimal solution At the same time , , Stored in the self-evolutionary decision knowledge base.

[0201] In this embodiment of the invention, in S4, since the global optimal solution is a set of many Pareto front solutions, the fitness of the optimal solution is selected using the following formula, and the corresponding [solution / finding] is found. Fitness of the optimal solution The expression is:

[0202] ;

[0203] ;

[0204] ;

[0205] in, This can be a constant that can be modified according to the task requirements. This represents the average distance from all pursuing agents to the target's location within the encirclement. To capture the uniformity of the distribution of the target's position within the encirclement, For the first A pursuing intelligent agent at the current moment The x-coordinate in the global coordinate system For the first A pursuing intelligent agent at the current moment The x-coordinate of the endpoint position on the bounding circle in the global coordinate system. For the first A pursuing intelligent agent at the current moment The ordinate in the global coordinate system For the first A pursuing intelligent agent at the current moment The ordinate of the endpoint position on the bounding circle in the global coordinate system. The number of all pursuing agents. The angles of all endpoint positions in the encirclement, in a polar coordinate system with the escaping agent as the origin, are sorted clockwise. indivual, The angles of all endpoint positions in the encirclement, in a polar coordinate system with the escaping agent as the origin, are sorted clockwise. indivual.

[0206] In this embodiment of the invention, S5 includes the following sub-steps:

[0207] S51. Move the current pursuing agent toward the optimal solution until the optimal solution is input at the next moment, and obtain the updated position of the pursuing agent after moving forward.

[0208] S52. Move the fleeing agent in a random direction and at a random speed to obtain the updated position of the fleeing agent;

[0209] S53. Determine whether all pursuing agents have reached the encirclement with the fleeing agent as the origin and a radius of dis. If so, all have reached it; otherwise, update the time and return to S2.

[0210] The encirclement is a circle with radius dis, originating from the fleeing agent. If all pursuing agents reach the encirclement, the encirclement is considered successful; otherwise, the encirclement fails, and the decision-making process is repeated until the encirclement is successful.

[0211] Traditional intelligent algorithms suffer from parameter dimensionality bottlenecks in cluster decision-making: each agent within the cluster requires separate modeling, such as... Figure 6 As shown. Assume a solver has the following parameters: A 3D matrix, then Each intelligent agent needs maintenance. The parameter space is 3D. In contrast, an evolutionary solver is used for... When multiple agents perform collaborative optimization, the parameter dimension is reduced to Wei, such as Figure 7 As shown, this not only greatly simplifies computational complexity, but also breaks through the cluster size limitation of traditional methods, that is, the upper limit of the number of agents is significantly increased.

[0212] Based on the above methods, such as Figure 8 As shown, this invention also proposes a cluster decision-making system based on evolutionary transfer learning, including a task similarity measurement module, a multi-source decision transfer module, and a multi-objective evolutionary optimization module;

[0213] The task similarity measurement module is used to initialize the pursuit scenario, generate the current task parameter set, and compare the current task parameter set with the task parameter set in the self-evolutionary decision knowledge base to determine the most similar task parameter set;

[0214] The multi-source decision transfer module is used to determine a fixed-size Pareto front solution based on the most similar task parameter set, input the fixed-size Pareto front solution into the solver, obtain the optimal solution for the current task parameter set, and determine the fitness of the optimal solution.

[0215] The multi-objective evolution optimization module is used to update the positions of the pursuing and fleeing agents based on the optimal solution.

[0216] First, the task is input into the task similarity measurement module, which searches the self-evolutionary decision knowledge base for the task most similar to the current task. Then, the solutions of similar tasks are input into the multi-source decision transfer module, which integrates the optimal solutions of similar tasks and the optimal solutions of adjacent tasks from the previous time step as the initial solution for the current task. Finally, the optimizer module uses an evolutionary learning algorithm to solve the problem, achieving online policy fine-tuning and constraint satisfaction verification, ultimately outputting the optimal decision.

[0217] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A cluster decision-making method based on evolutionary transfer learning, characterized in that, Includes the following steps: S1. Initialize the pursuit scenario and generate the current task parameter group; S2. Compare the current task parameter set with the task parameter sets in the self-evolutionary decision knowledge base to determine the most similar task parameter set; S3. Determine the Pareto front solution of a fixed size based on the most similar task parameter set; S4. Input the Pareto front solution of a fixed size into the solver to obtain the optimal solution for the current task parameter set, and determine the fitness of the optimal solution; S5. Update the positions of the pursuing and fleeing agents based on the optimal solution.

2. The cluster decision-making method based on evolutionary transfer learning according to claim 1, characterized in that, In S1, the pursuit scenario includes the locations of the pursuing agent, the escaping agent, and the obstacles; The pursuit agent The expression is: ； in, For the pursuit agent labeled 1 in the set of all pursuit agents, For the pursuit agent numbered 2 in the set of all pursuit agents, For all pursuing agents, the set labeled as The pursuing intelligent agent; The expression for the escape agent is: ； in, For all escape agents, the escape agent with the index 1 is included. Let the escape agent numbered 2 be the one that escapes from the set of all escape agents. The set of all escaped agents is labeled as The escape agent; The location of the obstacle The expression is: ； in, For the obstacle numbered 1 in the set of all obstacles, For the obstacle numbered 2 in the set of all obstacles, For all obstacles in the set labeled Obstacles; In S1, Current task parameter group at any given time The expression is: ； ； ； ； ； in, for The x-coordinate of all pursuing agents in the global coordinate system at any given time. for The ordinate of all pursuing agents in the global coordinate system at any given time. for The x-coordinate of all fleeing agents in the global coordinate system at any given moment. for The x-coordinate of all fleeing agents in the global coordinate system at any given moment. For the number The pursuing intelligent agent in The x-coordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The x-coordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The x-coordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The ordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The ordinate in the global coordinate system at time t. For the number The pursuing intelligent agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t.

3. The cluster decision-making method based on evolutionary transfer learning according to claim 1, characterized in that, S2 includes the following sub-steps: S21. Extract the number of tasks stored in the evolutionary decision knowledge base. If the number of tasks is 0, generate an initial solution randomly and proceed to S25. Otherwise, set the current task parameter group and proceed to S22. S22. Perform a normalized polar coordinate transformation on the current task parameter set to obtain the normalized polar coordinate distance. ,in, To track down an agent that is trying to escape, the polar coordinate radius of the agent's center is... For angle; S23. Perform normalized polar coordinate transformation on all task parameter groups stored in the self-evolutionary decision knowledge base to obtain the normalized self-evolutionary decision knowledge base. S24. Based on the normalized polar coordinate distance of the current task parameter group and the normalized self-evolutionary decision knowledge base, through... Calculated ,at this time and Overlap, align the task in the self-evolutionary decision knowledge base with the current task, where, These are the polar coordinate angles in the normalized current task parameter set. Let $t$ be the maximum polar coordinate angle of the $t$-th task in the knowledge base. This represents the maximum polar coordinate angle value of the current task parameter set. Let tempt be the maximum polar coordinate radius of the tempt-th task in the knowledge base. The maximum value of the polar coordinate radius of the current task parameter set. This represents the maximum polar coordinate angle in the normalized current task parameter set; S25. Based on the alignment results, calculate the similarity using the DTW dynamic curve similarity calculation algorithm. and The curve similarity between them is used to obtain the current task. With the knowledge base Similarity score between two groups of tasks , Let be the polar coordinate radius of the pursuing agent with the number tempt, centered at the escape agent. For the tempt-th numbered angle; S26. Determine the number Does it equal the number of tasks stored in the self-evolutionary decision knowledge base? If so, then find... If the corresponding task parameter set is selected as the most similar task parameter set, proceed to S3; otherwise, proceed to... Add 1, return to step S21; where, The set of all similarity scores. The similarity score between the current task and the first task stored in the knowledge base. The similarity score between the current task and the second task stored in the knowledge base. For the current task and the knowledge base stored in the knowledge base Similarity score for each task.

4. The cluster decision-making method based on evolutionary transfer learning according to claim 1, characterized in that, S22 includes the following sub-steps: S221. Take the average point of all escape agents in the current task parameter group as the origin of the polar coordinate system, and take the counterclockwise direction as the direction of angle growth of the polar coordinate system. S222. Perform polar coordinate transformation on the global coordinates of the pursuing agent based on the origin to obtain new polar coordinates; S223. Normalize the new polar coordinates, find the pursuing agent with the largest distance, and obtain the normalized polar coordinate distance. In S221, the average point of all escaped agents in the current task parameter group The expression is: ； ； in, The x-coordinate of the fleeing agent is the average x-coordinate. The average ordinate of the fleeing agent. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The x-coordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at time t. For the number The escape agent in The ordinate in the global coordinate system at that moment; In S222, the new polar coordinates The expression is: ； ； in, The distance of each pursuing agent in the current task parameter group from the origin. The angle of each pursuing agent in the current polar coordinate system. The polar coordinate radius of the first pursuing agent, with the average point of the escapees as the origin. Let the polar coordinate radius of the second pursuing agent be the average point of the escapees as the origin. For the first The polar coordinate radius of each pursuing agent is taken as the origin from the average point of the escapee. The polar coordinate angle of the first pursuing agent, with the average point of the escapee as the origin. The polar coordinate angle of the second pursuing agent is taken as the origin from the average point of the escapee. For the first The polar coordinate angle of the pursuing intelligent agent with the average point of the escapee as the origin; In step S223, the pursuing agent with the greatest distance is found. The expression is: ； in, To obtain the maximum value; In S223, the normalized polar coordinate distance of the current task parameter group The expression is: ； in, The polar coordinate radius of the pursuing agent at time t is taken as the origin from the average center of the fleeing agent. for The largest value of the radius in. Let be the polar radius of the first pursuing agent at time t, with the average center of the fleeing agents as the origin. Let be the polar coordinate radius of the second pursuing agent at time t, with the average center of the fleeing agent as the origin. For the first The polar coordinate radius of a pursuing agent at time t is taken as the origin from the average center of the fleeing agent.

5. The cluster decision-making method based on evolutionary transfer learning according to claim 1, characterized in that, S23 includes the following sub-steps: S231. The average point of all escape agents in the task parameter group of the self-evolving decision knowledge base is taken as the origin of the polar coordinate system, and the counterclockwise direction is taken as the direction of angle growth of the polar coordinate system. S232. Perform polar coordinate transformation on the global coordinates of the pursuing agent based on the origin to obtain new polar coordinates; S233. Normalize the new polar coordinates, find the pursuing agent with the largest distance, and perform normalization operation to obtain the normalized polar coordinate distance. In S231, the average point of all escaped agents in the task parameter set of the self-evolving decision knowledge base is... The expression is: ； ； in, Let x be the x-coordinate of the average center point of the escaped agents in the tempt-th task of the self-evolutionary decision knowledge base. Let x be the x-coordinate of the average center point of the escaped agents in the tempt-th task of the self-evolutionary decision knowledge base. Let x be the x-coordinate of the first fleeing agent in the tempt-th task in the self-evolutionary decision knowledge base, in the global coordinate system. Let x be the x-coordinate of the second fleeing agent in the tempt-th task of the self-evolutionary decision knowledge base, in the global coordinate system. For the tempt-th task in the self-evolutionary decision knowledge base, the th... The x-coordinate of the fleeing agent in the global coordinate system Let be the ordinate of the first fleeing agent in the tempt-th task of the self-evolutionary decision knowledge base, in the global coordinate system. Let be the ordinate of the second fleeing agent in the tempt-th task of the self-evolutionary decision knowledge base, in the global coordinate system. For the tempt-th task in the self-evolutionary decision knowledge base, the th... The vertical coordinate of a fleeing agent in the global coordinate system; In S232, the new polar coordinates The expression is: ； ； in, Let be the distance of each pursuing agent from the origin in the task parameter set of the self-evolving decision knowledge base. Let be the angle of each pursuing agent in the current polar coordinate system within the task parameter set of the self-evolving decision knowledge base. Let be the radius in polar coordinates of the first pursuing agent in the 'tempt'th task of the self-evolving decision knowledge base. Let be the radius in polar coordinates of the second pursuing agent in the 'tempt'th task within the self-evolving decision knowledge base. For the tempt-th task in the self-evolutionary decision knowledge base, the th... The radius of the pursuing intelligent agent in polar coordinates Let be the polar coordinate angle of the first pursuing agent in the 'tempt'th task within the self-evolving decision knowledge base. Let be the polar coordinate angle of the second pursuing agent in the 'tempt'th task within the self-evolving decision knowledge base. For the tempt-th task in the self-evolutionary decision knowledge base, the th... Angle of a tracking agent in polar coordinates; In step S233, the pursuing agent with the greatest distance is found. The expression is: ； in, To obtain the maximum value; In S23, the polar coordinate distance of the task parameter set of the self-evolutionary decision knowledge base after normalization is... The expression is: ； in, Let be the value that maximizes the polar coordinate radius of all pursuing agents in the tempt-th task of the self-evolutionary decision knowledge base. Let be the polar coordinate radius of the first pursuing agent after normalization of the 'tempt'th task in the self-evolutionary decision knowledge base. Let be the polar coordinate radius of the second pursuing agent after normalization of the 'tempt'th task in the self-evolutionary decision knowledge base. For the normalized tempt-th task in the self-evolutionary decision knowledge base, the i-th task is... The polar coordinate radius of the pursuing agent.

6. The cluster decision-making method based on evolutionary transfer learning according to claim 1, characterized in that, S3 includes the following sub-steps: S31. Extract the global optimal solution and the population optimal solution of the most similar task parameter set from the self-evolutionary decision knowledge base; S32. Extract the global optimal solution and the population optimal solution of the task parameter set at the previous time step from the self-evolutionary decision knowledge base; S33. Generate candidate solutions based on the global optimal solution and population optimal solution of the most similar task parameter set, as well as the global optimal solution and population optimal solution of the task parameter set at the previous time step. S34. Generate a loss function for the candidate solutions; S35. Based on the loss function of the candidate solutions, determine the Pareto front solution of a fixed size.

7. The cluster decision-making method based on evolutionary transfer learning according to claim 6, characterized in that, In S33, the candidate solution The expression is: ； in, The global optimal solution is the parameter set of the most similar task. The population-optimal solution for the most similar set of task parameters. This is the globally optimal solution for the task parameter set at the previous time step. This is the population-optimal solution for the task parameter set at the previous time step.

8. The cluster decision-making method based on evolutionary transfer learning according to claim 1, characterized in that, In S4, the fitness of the optimal solution The expression is: ； ； ； in, It is a constant. This represents the average distance from all pursuing agents to the target's location within the encirclement. To capture the uniformity of the distribution of the target's position within the encirclement, For the first A pursuing intelligent agent at the current moment The x-coordinate in the global coordinate system For the first A pursuing intelligent agent at the current moment The x-coordinate of the endpoint position on the bounding circle in the global coordinate system. For the first A pursuing intelligent agent at the current moment The ordinate in the global coordinate system For the first A pursuing intelligent agent at the current moment The ordinate of the endpoint position on the bounding circle in the global coordinate system. The number of all pursuing agents. The angles of all endpoint positions in the encirclement, in a polar coordinate system with the escaping agent as the origin, are sorted clockwise. indivual, The angles of all endpoint positions in the encirclement, in a polar coordinate system with the escaping agent as the origin, are sorted clockwise. indivual.

9. The cluster decision-making method based on evolutionary transfer learning according to claim 1, characterized in that, S5 includes the following sub-steps: S51. Move the current pursuing agent toward the optimal solution until the optimal solution is input at the next moment, and obtain the updated position of the pursuing agent after moving forward. S52. Move the fleeing agent in a random direction and at a random speed to obtain the updated position of the fleeing agent; S53. Determine whether all pursuing agents have reached the encirclement with the fleeing agent as the origin and a radius of dis. If so, all have reached it; otherwise, update the time and return to S2.

10. A cluster decision-making system based on evolutionary transfer learning, characterized in that, It includes a task similarity measurement module, a multi-source decision transfer module, and a multi-objective evolution optimization module; The task similarity measurement module is used to initialize the pursuit scenario, generate the current task parameter set, and compare the current task parameter set with the task parameter set in the self-evolutionary decision knowledge base to determine the most similar task parameter set. The multi-source decision migration module is used to determine a Pareto front solution of a fixed size based on the most similar task parameter set, input the Pareto front solution of the fixed size into the solver, obtain the optimal solution of the current task parameter set, and determine the fitness of the optimal solution. The multi-objective evolution optimization module is used to update the positions of the pursuing agent and the fleeing agent based on the optimal solution.

Images