A method for allocating search and rescue tasks for UAV swarm based on discrete particle swarm

Abstract

The present invention relates to the technical field of unmanned aerial vehicles, and in particular to a method for allocating unmanned aerial vehicle cluster search and rescue tasks based on discrete particle swarms. The method comprises: determining the legal clusters and legal tasks at the current time t and generating multiple task pairs according to the life cycle of the task and the task execution capability of the cluster. The task value of the task pair is calculated according to a preset task value function, and a task value matrix is ​​generated; random particles are generated based on a random particle model, and the random particles are genetically mutated according to the local optimal particles and / or the global optimal particles to obtain the optimal random particles; the optimal random particles are optimized based on the task value matrix to determine the local optimal particles; multiple local optimal particles are optimized to determine the global optimal particles with the maximum sum of task values ​​and the minimum variance between task values; after a preset number of cycles, the global optimal particles are decoded to obtain multiple optimal task pairs, and a task allocation scheme is generated, so as to achieve high adaptability and high completion rate of task allocation.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, and in particular to a method for allocating search and rescue tasks of unmanned aerial vehicle clusters based on discrete particle swarms. Background Art

[0002] Drone swarm refers to a collaborative working system composed of multiple drones or aircraft that complete complex tasks by sharing information and coordinating actions. These drones can autonomously or semi-autonomously perform various operations such as search, monitoring, and assessment according to preset task allocation methods. In the modern disaster search and rescue process, drone swarms have significantly improved the efficiency and effectiveness of disaster search and rescue work, and won precious time for saving lives.

[0003] Drone swarms can be quickly deployed to disaster-stricken areas to provide an immediate aerial perspective. Using high-resolution cameras, thermal imagers, and other sensors, drones can quickly map disaster areas, identify the location of survivors, and assess the damage to buildings. This efficient coverage capability enables search and rescue teams to obtain key information in the shortest possible time, thereby formulating a more scientific and reasonable search and rescue plan. For example, after the Wenchuan earthquake on May 12, 2018, the post-disaster images taken by the "Hua Ying" and "Clairvoyance" drones provided important information support for subsequent search and rescue work. In addition, drone swarms also have the functions of material transportation and communication relay. In areas where traffic is interrupted or in remote areas, drones can deliver emergency medical supplies or other necessities, and at the same time serve as temporary communication nodes to restore damaged networks and ensure smooth communication between all parties.

[0004] In modern disaster search and rescue, in order to improve the efficiency of drone clusters in disaster search and rescue, the task distribution mechanism of search and rescue is key. For example, the Chinese invention patent application with application publication number CN118819188A discloses a dynamic collaborative optimization method for unmanned clusters facing multiple task requirements, including step 1: obtaining environmental data of drones performing tasks and establishing a dynamic collaborative optimization model for drone clusters; step 2: specifying the urgency and importance of tasks and establishing a hierarchical impact model for task priorities; step 3: intelligently adjusting drone options based on real-time environmental data; step 4: developing an innovative selection model that focuses on selecting drone combinations that carry complementary collaborative resources and have interoperability; step 5: integrating the objective functions of each sub-model, establishing the optimization function of the dynamic collaborative optimization model for drone clusters, and using the particle swarm algorithm to solve the optimal solution of the model. For example, the Chinese invention patent application with application publication number CN119106898A discloses a multi-UAV task allocation method, apparatus, equipment and storage medium, which specifically discloses initializing task target data and UAV cluster data, and calculating the cost information of each UAV in the UAV cluster to execute each task target; based on the cost information of each task target, a dynamic programming strategy is used to select the optimal task target set to be executed; a greedy strategy is used to select the UAV with the lowest cost to execute each optimal task target from the UAV cluster in turn for task allocation, until the task allocation of the optimal task target set is completed.

[0005] However, actual search and rescue scenarios often involve multi-dimensional interference factors, such as the macro-planning requirements of disaster search and rescue, the imbalance between drone cluster resources and search and rescue missions, and the existing search and rescue task issuance mechanism is unable to cope with the resource allocation of complex and diverse search and rescue scenarios, which in turn reduces the overall search and rescue efficiency and affects the success rate and effectiveness of search and rescue operations. Summary of the invention

[0006] The main purpose of the present invention is to provide a method for allocating search and rescue tasks of drone swarms based on discrete particle swarms. In order to solve the above-mentioned technical problems, the present invention specifically adopts the following technical solutions:

[0007] A first aspect of the present invention is to provide a method for allocating search and rescue tasks of drone clusters based on discrete particle swarms, the method comprising:

[0008] S101, obtain N available drone clusters and K search and rescue tasks; filter the clusters and tasks according to the life cycle of each task and the task execution capability of each cluster, and obtain the current time t The set of legal clusters , The set of legal tasks , and generate multiple task pairs ,in, To command the cluster Execute the task The combination of tasks, , , is a positive integer and , is a positive integer and , , is an integer greater than or equal to zero;

[0009] S102, according to the preset task value function , calculate the task value of each task pair, and generate A legitimate cluster, The task value matrix corresponding to the legal tasks ;

[0010] S103: Obtain a preset random particle model, and generate a A first random particle is used as the optimal random particle, wherein the random particles generated by the preset random particle model are particle matrices, and the order of the particle matrix is and The larger value of ;

[0011] S104: Based on the task value matrix right The optimization random particles are optimized to determine the local optimal particle of this iteration, and the local optimal particle of each iteration is recorded; wherein the local optimal particle is the optimization random particle with the largest total task value and the smallest variance between task values; execute step S105;

[0012] S105: Based on the task value matrix Optimize multiple local optimal particles to determine the global optimal particle of this cycle, wherein the global optimal particle is the local optimal particle with the largest total task value and the smallest variance between task values; update the number of cycles , when the number of cycles When the preset number of cycles is reached, step S107 is executed; otherwise, step S106 is executed;

[0013] S106: Generate based on the preset random particle model a second random particle; based on a preset genetic algorithm, according to the local optimal particle or the global optimal particle The second random particles undergo genetic mutation and obtain The optimized random particles are executed; step S104 is executed;

[0014] S107, decoding the global optimal particle based on a preset decoding matrix to obtain multiple optimal task pairs, and generating a task allocation scheme at the current time t according to the multiple optimal task pairs.

[0015] In some embodiments, the set of N available drone clusters is , the set of K search and rescue missions is ; Step S101, comprising: at the current moment , the set of legal clusters is defined as: ; The set of legal tasks is defined as: ;in, Is a cluster The number of drones, is an integer greater than or equal to zero; cluster Available The set of task execution capabilities is For traversing tasks from 1 to 2 Task judgment cluster Whether it has the ability to execute tasks, if there is at least one task type Mission , cluster For Ability to perform tasks of different types ,but ,otherwise ; It is the time criterion used to judge the cluster Can you Arrive at the mission location and perform the mission within the life cycle ; For traversing tasks from 1 to 2 Task judgment cluster Can you The operation of reaching any search and rescue mission location within the life cycle of , cluster Execute the task Time criterion ,but ,otherwise ; For the task The end execution time of For traversing from the first cluster to the Cluster judgment task Whether the operation can be performed by any cluster, if there is at least one cluster , cluster The task type is Mission ability, then ,otherwise ; For traversing from the first cluster to the Cluster judgment task Whether the operation can be completed by any cluster during the life cycle, if there is at least one cluster , cluster Execute the task Time criterion ,but ,otherwise .

[0016] In some embodiments, the task pairs include invalid task pairs and valid task pairs; the preset task value function It includes a task execution capability parameter and a task value evaluation parameter. The task execution capability parameter is used to evaluate the task value of the invalid task pair as 0, and the task value evaluation parameter is used to perform value evaluation on the valid task pair.

[0017] In some embodiments, the preset task value function The expression is: ; Among them, cluster Available The set of task execution capabilities is , cluster Not capable of executing The ability to perform tasks , cluster Possess the ability to execute The ability to perform tasks ; It is the task completion flag. When the task is not completed , when the task has been completed ; is the task execution priority, , is a set of positive integers; It's a task The remaining effective time, Is a cluster Execute the task The cost of time spent.

[0018] In some embodiments, step S102 further includes: , As the matrix side lengths, generate The effective task value matrix of each task pair is filled into the corresponding matrix position as a matrix element; according to the , The difference between them generates a supplementary matrix with zero value, and the supplementary matrix is ​​merged with the effective task value matrix to generate the task value matrix with the same matrix side length.

[0019] In some embodiments, steps S104 to S105 further include: , the particle matrix, determining multiple task values ​​of the current particle matrix; adding the multiple task values ​​to obtain the sum of the task values ​​of the particle matrix, and calculating the variance between the multiple task values ​​to obtain the variance between the task values ​​of the particle matrix.

[0020] In some embodiments, steps S104 to S106 further include: The local optimal particle of the iteration Defined as: ;in, is a set of positive integers; is the random particle number of this iteration, It is The iteration generated The first of the random particles Random particles; is the optimal random particle, when =1, is the first random particle, when hour, Including part of the second random particles , and some The local optimal particle of the iteration and / or part of the global optimal particles ; It is Random particles The random permutation matrix, matrix is the optimal random particle The permutation matrix of ; It means calculating the cumulative sum of all elements of the matrix in brackets; For this iteration traverse the optimal random particles and obtain the particle matrix with the largest total task value; Used to calculate the standard deviation of all non-zero elements in the matrix; It is used to obtain the local optimal particle with the smallest variance between task values ​​from the particle matrix with the largest total task value;

[0021] The global optimal particle of this cycle Defined as: ;in, is the number of cycles of this algorithm, is the iteration number, It is The local optimal particle in the iteration A random permutation matrix of ; It means calculating the cumulative sum of all elements of the matrix in brackets; Used to traverse the generated local optimal particles and obtain the particle matrix with the largest total task value; Used to calculate the standard deviation of all non-zero elements in the matrix; It is used to obtain the global optimal particle with the smallest variance between task values ​​from the particle matrix with the largest total task value;

[0022] The preset genetic algorithm is defined as: ;in, It is a genetic operation that copies part of the column data of one matrix to the corresponding column data of another matrix; is the global optimal particle The permutation matrix of .

[0023] In some embodiments, the preset decoding matrix includes a preset mapping matrix and a permutation matrix of the global optimal particle; S107 includes: multiplying the global optimal particle with the inverse of the permutation matrix of the global optimal particle, and multiplying the permutation matrix with the preset mapping matrix to obtain a task allocation scheme matrix , the calculation formula is: ;in, is the global optimal particle, is the permutation matrix of the global optimal particle, is the preset mapping matrix, Representation Matrix The task allocation scheme matrix The task pair in is taken as the optimal task pair, and the set of multiple optimal task pairs is: ;in, Represents the task allocation matrix No. Line The optimal task pair for the column.

[0024] In some embodiments, when Greater than When the method further comprises: determining the task with the greatest task value for each of the clusters according to the task value, and generating a set of spare task pairs ; Get the set of optimal task pairs The used clusters in the task list; and filtering out the set of backup task pairs according to the used clusters. The task pairs corresponding to the unused clusters in the cluster are obtained to obtain the remaining cluster task pair set ; The set of the optimal task pairs , the remaining cluster task pair set The task pairs in are used as target task pairs, and the task allocation scheme at the current time t is updated according to the multiple target task pairs.

[0025] In some embodiments, when Less than When at least one of the clusters completes the corresponding task, the method further includes: updating the current time , the life cycle of the task, the task execution capability of the cluster; re-execute steps S101 to 107 to generate an updated current time task allocation plan.

[0026] Beneficial technical effects:

[0027] The present invention provides a method for allocating search and rescue tasks for drone swarms based on discrete particle swarms with high adaptability and high completion rate. The method realizes balanced screening of maximizing value and minimizing variance through a multi-level progressive optimization strategy from local to global based on full matrix operation with redundant design, thus ensuring the adaptability, balance and high quality of task allocation and execution, and meeting the strict requirements of search and rescue operations for rapid response and accurate decision-making. The specific performance is as follows:

[0028] 1. Local to global multi-level progressive optimization strategy. In the process of optimizing random particles to determine the global optimal particle, multi-level continuous optimization from the local level to the global level is achieved, which improves the quality and accuracy of task allocation. Specifically, at the beginning of each iteration, new random particles are generated to achieve mutation. These particles randomly cross-transfer genes with the local optimal particles determined in the historical iteration or the global optimal particles determined in the historical cycle to simulate the genetic process, ensuring that the particles are adjusted based on the latest information, ensuring the diversity of the particle group, and promoting the transmission of excellent features. Subsequently, all local optimal particles are optimized again to finally determine the global optimal particle.

[0029] 2. Balanced screening for maximizing value and minimizing variance. The task allocation scheme represented by the particle with the largest total task value can provide higher overall benefits. In order to improve the balance of task completion, multiple high-task value particles are evaluated twice, and the particle with the smallest variance in task value is selected as the final result to ensure the balance of task allocation, so that the task completion quality is higher and the time consumed is shorter. Especially in actual scenarios where the number of drone clusters is less than the number of search and rescue tasks, it is necessary to allocate and complete the search and rescue tasks in batches. At this time, a more balanced task allocation can be selected based on the variance between the task values, so as to better optimize the priority of multiple search and rescue tasks according to the characteristics of clusters and tasks in actual search and rescue scenarios. As a result, the first batch of search and rescue tasks can be completed more quickly, while reducing the possibility of delays in the golden search and rescue period of the second batch of tasks, thereby improving the overall completion rate of search and rescue tasks.

[0030] 3. Lightweight matrix operations accelerate task allocation. In order to speed up the calculation speed and improve the timeliness of the allocation plan, a full matrix method is used to complete the value calculation of multiple possible task allocation schemes at one time, and quickly determine the best decision from a variety of feasible task allocation schemes in a short time, which greatly shortens the calculation time and is highly adapted to the changing situations and limited task allocation time in search and rescue missions, ensuring that search and rescue operations are carried out in a timely and effective manner. At the same time, efficient matrix calculation methods reduce the computing power required for task allocation, and can operate efficiently even in resource-constrained environments, which is highly adapted to the limited computing performance of onboard computers. In addition, lightweight matrix operations help extend the flight time of drones, because less computing load means lower energy consumption, thereby better supporting long-term, large-scale search and rescue missions.

[0031] Furthermore, by simply matching legal clusters and legal tasks one by one to generate task pairs, various types of task allocation possibilities can be included quickly and comprehensively. For the redundant invalid task pairs (the clusters in the task pair cannot execute the tasks in the task pair), in the particle optimization process, balanced task allocation is achieved by minimizing the variance between task values, and the influence of invalid task pairs without task values ​​in task allocation is further eliminated. In addition, the matrix cells of the task value matrix generated based on the task pairs are also redundant. These redundant cells cleverly form a matching relationship with the particle matrix of random particles, and the redundant matrix cells are reasonably used, which simplifies the subsequent algorithm process of optimizing random particles based on the task value matrix, and also speeds up the calculation speed and improves the timeliness of the allocation plan through the full matrix.

[0032] Furthermore, a systematic task allocation mechanism has been built, which can adaptively and quickly allocate tasks regardless of whether the drone clusters are equal or not to the tasks. First, in actual scenarios where the number of drone clusters is greater than the number of search and rescue tasks, the drone clusters that have not been assigned tasks are reassigned tasks so that they can perform the corresponding high-value tasks. Some search and rescue tasks are completed collaboratively by at least two drone clusters, which accelerates the execution speed of search and rescue tasks and further improves the practicality and adaptability of task allocation in actual search and rescue scenarios. Second, in actual scenarios where the number of drone clusters is less than the number of search and rescue tasks, the task execution status of each cluster is continuously monitored, and when it is detected that a cluster has completed its assigned task, multiple key parameters are updated, and tasks are reassigned and executed. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, each element or part is not necessarily drawn according to the actual scale. Obviously, the drawings described below are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without paying creative labor.

[0034] Figure 1 is a schematic flow chart of a method for allocating search and rescue tasks of drone clusters based on discrete particle swarms provided by an embodiment of the present invention;

[0035] Figure 2 It is a schematic diagram of a drone cluster search and rescue task allocation scenario based on discrete particle swarms provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly and completely described in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention. Herein, the orientation or position relationship indicated by the terms "upper", "lower", "inner", "outer", "front", "back", "one end", "other end", etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood in specific circumstances. Herein, "and / or" includes any and all combinations of one or more listed related items. Herein, "multiple" means two or more, that is, it includes two, three, four, five, etc. It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0037] The existing task allocation methods for drone swarms use a variety of algorithms, mainly including optimization methods, heuristic methods, and market auction methods based on contract networks. Among them, optimization methods include exhaustive methods, integer programming methods, constraint programming, and graph theory methods. These methods require the allocation problem to have an optimal solution based on certain ideal assumptions, but in actual application, due to the complex derivation formulas, they cannot adapt to the short-term and rapid requirements of large-scale task allocation scenarios, and there are often multiple local optimal solutions in practice. Heuristic algorithms include list planning, evolutionary algorithms, swarm intelligence algorithms, etc., which have large computational complexity and high requirements for initial data, and cannot ensure the optimal solution. Existing algorithms are difficult to cope with multi-dimensional interference factors in actual search and rescue scenarios, and the effectiveness of the corresponding task allocation is difficult to guarantee, which affects the success rate and effect of search and rescue operations.

[0038] The task allocation of drone clusters is a task allocation scenario under multi-dimensional discrete conditions. First, tasks are discrete, and each task is usually a specific and independent work unit, such as searching a specific area or reconstructing communications in a specific area. Secondly, resources are discrete. As an independent execution unit, the capabilities and states (such as remaining power, load capacity, etc.) of each drone cluster are discrete quantities. Finally, the task allocation scheme is discrete, and the assignment relationship of each drone cluster to which task is assigned is discrete. Particle Swarm Optimization (PSO) is an optimization algorithm based on swarm intelligence. By simulating the behavior patterns of groups in nature, it can efficiently explore the solution space and find the optimal solution. Discrete Particle Swarm Optimization (DPSO) is a variant of PSO, which is suitable for dealing with problems with discrete variables.

[0039] The embodiment of the present invention introduces discrete particle swarm optimization, and based on this, optimizes the diversified scenarios in actual search and rescue, and provides a drone cluster search and rescue task allocation method based on discrete particle swarms with high adaptability and high completion rate. Based on full matrix operation with redundant design, a balanced screening of value maximization and variance minimization is achieved through a multi-level progressive optimization strategy from local to global.

[0040] In some embodiments, in order to accurately match the drone swarm with the search and rescue mission and determine the best decision to improve the success rate and effect of the search and rescue operation, mathematical modeling is performed on the search and rescue mission and the drone swarm, and the mission requirements and the characteristics of the drone swarm are fully considered in combination with the actual search and rescue scenario. , a set of K search and rescue missions , each drone cluster can be used to perform search and rescue missions, and each mission can be performed by at least one cluster. Different missions may require drone clusters to perform search and rescue in different flight formations, which is not limited here. Pre-write or select cluster behavior algorithms suitable for specific mission scenarios, such as formation flying, distributed search, etc. Compared with the precise allocation and activation of a single drone, cluster-based overall activation does not require re-formation, which can save more time and complete the search and rescue mission as efficiently as possible within the golden search and rescue period. In addition, some drones will be lost during the search and rescue process. Cluster-based overall activation is more in line with the requirements of redundant design in search and rescue missions. Therefore, modeling based on drone clusters is more advantageous.

[0041] In some embodiments, a single search and rescue mission is generated through information such as task index, search and rescue priority, search and rescue location, and search and rescue required arrival time, and a single drone cluster is generated by combining information such as drone cluster index, average movement speed of drone cluster, ability of drone cluster to perform tasks, and number of surviving members of drone cluster.

[0042] Define the state attributes of the drone cluster, including but not limited to: drone cluster index, drone cluster performance parameters (such as the average movement speed of the drone cluster, the ability of the drone cluster to perform tasks), the number of surviving members of the drone cluster, etc. Cluster , , which has the following properties: 1. Cluster The number of surviving drones in is , is an integer greater than or equal to zero; 2. Moment, Cluster The geometric center coordinates are , is the matrix transpose symbol, , and They are Moment Cluster The north direction coordinate, east direction coordinate and the height coordinate perpendicular to the horizontal plane of the center coordinate; 3. Moment, Cluster The average flight speed is , is the matrix transpose symbol, , and Clusters The north velocity component, east velocity component and vertical upward velocity component of the average flight speed; 4. Cluster Available The set of task execution capabilities is ,in is a signed number equal to 0 or 1, Representative cluster Not capable of executing The ability to perform tasks, Representative cluster Possess the ability to execute Ability to perform tasks.

[0043] The characteristic attributes of the search and rescue mission are defined, including but not limited to: mission index, which is used to uniquely identify the number or code of each mission; mission type, which is used to define the specific nature of the mission; search and rescue priority, which is used to characterize the urgency or importance of the mission; search and rescue location, which specifies the geographical coordinates or area where the mission occurs; geographical environment, which is used to describe the terrain, climate and other environmental conditions in the mission area; search and rescue time, that is, the life cycle of the mission, which is used to define the time window for the start or completion of the mission, etc. For example, for the mission numbered Search and rescue mission , , which has the following attributes: 1. Task type , , is a set of positive integers; 2. The task completion mark is , is a signed number equal to 0 or 1, Indicates that the task has not been completed. Indicates that the task has been completed; 3. The task execution priority is , is a set of positive integers, The larger the absolute value of represents, the higher the priority of the search and rescue mission; 4. Time, Mission The geometric center coordinates of the area are , is the matrix transpose symbol, , and They are Time Mission The north direction coordinate, east direction coordinate and the height coordinate perpendicular to the horizontal plane of the center coordinate; 4. Task The start execution time and end time are defined as and , , , .

[0044] Among them, the task type is determined according to the different needs of different types of search and rescue scenarios (such as earthquakes, floods, forest fires), such as communication reconstruction tasks, object search tasks, reconnaissance modeling tasks, and item transportation tasks. Correspondingly, clusters can be divided into communication reconstruction clusters, object search clusters, reconnaissance modeling clusters, and item transportation clusters according to their task execution capabilities; further, the functions of clusters can also be complex, and a cluster can have multiple task execution capabilities, which are not limited here.

[0045] Some embodiments of the present application are described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the following embodiments and features of the embodiments can be combined with each other. Figure 1 , Figure 1 is a schematic flow chart of a method for allocating search and rescue tasks of drone clusters based on discrete particle swarms provided by an embodiment of the present invention, such as Figure 1 As shown, an embodiment of the present invention provides a method for allocating search and rescue tasks of drone clusters based on discrete particle swarms, and the method includes S101 to S107.

[0046] S101, obtain N available drone clusters and K search and rescue tasks; filter the clusters and tasks according to the life cycle of each task and the task execution capability of each cluster, and obtain the current time t The set of legal clusters , The set of legal tasks , and generate multiple task pairs ,in, To command the cluster Execute the task The combination of tasks, , , is a positive integer and , is a positive integer and , , is an integer greater than or equal to zero;

[0047] S102, according to the preset task value function , calculate the task value of each task pair, and generate A legitimate cluster, The task value matrix corresponding to the legal tasks ;

[0048] S103: Obtain a preset random particle model, and generate a A first random particle is used as the optimal random particle, wherein the random particles generated by the preset random particle model are particle matrices, and the order of the particle matrix is and The larger value of ;

[0049] S104: Based on the task value matrix right The optimization random particles are optimized to determine the local optimal particle of this iteration, and the local optimal particle of each iteration is recorded; wherein the local optimal particle is the optimization random particle with the largest total task value and the smallest variance between task values; execute step S105;

[0050] S105: Based on the task value matrix Optimize multiple local optimal particles to determine the global optimal particle of this cycle, wherein the global optimal particle is the local optimal particle with the largest total task value and the smallest variance between task values; update the number of cycles , when the number of cycles When the preset number of cycles is reached, step S107 is executed; otherwise, step S106 is executed;

[0051] S106: Generate based on the preset random particle model a second random particle; based on a preset genetic algorithm, according to the local optimal particle or the global optimal particle The second random particles undergo genetic mutation and obtain The optimized random particles are executed; step S104 is executed;

[0052] S107, decoding the global optimal particle based on a preset decoding matrix to obtain multiple optimal task pairs, and generating a task allocation scheme at the current time t according to the multiple optimal task pairs.

[0053] Among them, the mission execution capability refers to the ability of the cluster to perform search and rescue missions, which is determined based on a comprehensive evaluation of multiple factors such as the cluster's flight performance (such as maximum flight distance, average flight speed, endurance, etc.), load capacity, and the type and number of sensors equipped. Specifically, it can be expressed as whether a certain cluster can successfully complete a certain type of search and rescue mission. If it can, it has the ability to perform that type of mission, otherwise it does not. A drone cluster can have multiple mission execution capabilities or only one mission execution capability.

[0054] Among them, the task pair is a possible task combination generated by matching legal clusters with legal tasks one by one, that is, for each legal drone cluster, try to match it with each legal search and rescue task to form multiple task pairs. The task pair in the best solution that can maximize the predetermined goal (that is, the solution with the largest total task value and the smallest variance between task values) is the optimal task pair. Exemplarily, the task pair includes invalid task pairs and valid task pairs. Since the legal cluster and legal task screening requires that the legal cluster can at least perform any task and the legal task can at least be performed by any cluster, the one-to-one matching task pair generation method may generate redundant invalid task pairs, that is, the cluster in the task pair cannot perform the task in the task pair. It should be understood that the legal cluster and the legal task are time-sensitive and only effective at the current time t. As time goes on, the legal cluster and the legal task may change accordingly, which improves the accuracy of search and rescue task allocation and the dynamic adaptability to complex search and rescue scenarios.

[0055] Among them, the preset task value function is used to evaluate the current time t UAV cluster Execute the task The benefit is the task value. For example, the preset task value function It includes task execution capability parameters, task value evaluation parameters, and task completion status parameters. The task execution capability parameters are used to evaluate the task value of the invalid task pair as 0. The task completion status parameters are used to evaluate the task value of the invalid task pair corresponding to the completed task as 0. The task value evaluation parameters are used to evaluate the value of the valid task pair. According to the correlation and change relationship between a series of task value evaluation parameters (such as task priority weighting, cluster task execution capability, task remaining time and execution time cost, execution distance cost, energy consumption cost, etc.) and the task value, a preset task value function is set to make the most effective resource allocation within a limited time and maximize the search and rescue benefits and success rate. For example, the time cost required to complete the task is inversely proportional to the task value. A longer execution time will reduce the task value of the task pair and increase the task cost of the task pair. It should be noted that the greater the task value, the smaller the corresponding task cost. Based on this correlation and change relationship, the preset task value function can also be designed from the perspective of task cost to indirectly reflect the task value.

[0056] The task value matrix is ​​generated based on the task value of each task pair, and is used to complete the screening of task allocation schemes through matrix operations. For example, the matrix structure of the task value matrix can be clustered For line, task is a column, and the cells are task pairs The task value or task cost of the cluster For columns, tasks For the line.

[0057] Among them, the preset random particle model is used to generate random particles used in each iteration, such as the first random particle and the second random particle. The first random particle is the random particle used in the first round of iteration. Since it is the first round of iteration, the first random particle is directly used as the optimization random particle; each round of iteration after the first round of iteration will generate different second random particles, and the second random particles will be used as the optimization random particle after genetic mutation. Different random particles are selected from all possible task pairs to construct different task allocation schemes, and then the overall performance of different task allocation schemes is compared by task value, and then the best scheme that can maximize the satisfaction of the predetermined goal is selected.

[0058] Among them, the number of random particles generated in each iteration It is flexibly set according to actual needs and is not limited here. In order to promote group diversity and avoid premature convergence to the local optimal solution, It is both the total number of particles in this round of iteration and the total number of calculations in this round of iteration. By calculating each particle, the solution space can be explored more comprehensively, enhancing the robustness and reliability of the algorithm.

[0059] It should be noted that in order to avoid repeated assignment of the same cluster or task, the preset random particle model is set with unique constraints on clusters and tasks to ensure that each cluster is only assigned to one task at a time point, and each task is only executed by one cluster at the same time point.

[0060] Among them, the order of the particle matrix is and The larger value in is to correspond to the task value matrix to complete the relevant calculation of the task value and adapt to the special situation where the number of clusters and the number of tasks in different search and rescue scenarios are not equal. Exemplarily, the random particles generated by the preset random particle model are particle matrices, and each row and column in the particle matrix has only one cell value of 1, and the values ​​of the other cells are all 0, thereby ensuring the unique allocation of each cluster and task.

[0061] Among them, the local optimal particle is the optimization random particle with the largest total task value and the smallest variance between task values ​​determined in a round of iteration, that is, each round of iteration will select a local optimal particle from multiple optimization random particles. The global optimal particle is the local optimal particle with the largest total task value and the smallest variance between task values ​​in this round of cycle, that is, after each round of iteration, multiple local optimal particles (including the latest local optimal particle selected by the current round of iteration) will be cyclically compared to re-determine the global optimal particle of this cycle or this algorithm. The preset number of cycles for particle optimization is also flexibly set according to actual needs, such as 25 times, which is not limited here.

[0062] Among them, the preset genetic algorithm is used to randomly select a part from the local optimal particles and / or the global optimal particles to replace a part of the second random particles. The population size, selection strategy, crossover probability, mutation probability and other related parameters of the genetic algorithm are predefined. The specific values ​​can be flexibly set according to actual needs and are not limited here. In the particle swarm optimization, the idea of ​​genetic algorithm can be combined to enhance the ability of PSO to jump out of the local optimum by introducing genetic operations, and select part from the local optimal particles and / or the global optimal particles to replace part of the second random particles, thereby improving the optimization efficiency and diversity.

[0063] It should be understood that, in essence, the first random particle, the second random particle, the local optimal particle, and the global optimal particle are all particle matrices, and the number of rows and columns of these particle matrices are the same, which are all larger values ​​of the number of legal clusters and the number of legal tasks. A portion of rows or columns are randomly selected from the second random particle as the first object to be replaced, and then the same number of rows or columns in the local optimal particle and / or the global optimal particle are selected as the second object, and the second object is used to replace the first object to achieve random genetic variation. It is also possible to select the local optimal particle or the global optimal particle for crossover to generate a new replacement particle. The specific crossover operation can be performed by single-point crossover, multi-point crossover or uniform crossover. Correspondingly, in order to avoid repeated allocation of the same cluster or task, the preset genetic algorithm is also provided with uniqueness constraints of clusters and tasks.

[0064] Among them, the preset decoding matrix is ​​used to decode the particle matrix and restore it to the matrix of task pairs. It should be understood that the global optimal particle is essentially a random particle after genetic mutation. Random mutation and inheritance will randomly disrupt the order of rows and columns of the particle matrix. Therefore, it is necessary to use the preset decoding matrix to uniformly organize the randomly disrupted column order during the algorithm operation, and then complete the decoding through the original arrangement order of the task pairs and matrix cells.

[0065] Specifically, N available drone clusters and K search and rescue missions are preliminarily screened according to the life cycle of the mission and the mission execution capability of the cluster. A legal mission can be completed by at least one cluster within its life cycle, and the mission is within the executable validity period; a legal cluster can complete at least one search and rescue mission, and the mission start time or mission completion time is within the life cycle of the search and rescue mission. This ensures that the drone clusters and search and rescue missions participating in the search and rescue mission allocation are valid, improves the efficiency and accuracy of task allocation, and thus increases the success rate of the search and rescue operation. Match the legal clusters and legal tasks one by one to obtain multiple task pairs. , and then according to the preset task value function , calculate the task value of each task pair, and then generate A legitimate cluster, The task value matrix corresponding to the legal tasks .

[0066] A preset random particle model is obtained, and a group of first random particles are generated based on the model. These first random particles represent the initial task allocation scheme, and each particle is a potential solution (i.e., a possible task allocation scheme). These first random particles are used as optimization random particles to execute step S104.

[0067] Step S104 is to determine the task value of the task allocation scheme corresponding to each optimization random particle based on the task value matrix, reflecting the contribution of each task to the overall goal. By evaluating all optimization random particles, the local optimal particle of this round of iteration is determined, and all local optimal particles are further evaluated to determine the global optimal particle of this cycle, and the number of cycles is updated. If the number of cycles reaches the preset number of cycles, the next step is entered; otherwise, the particle optimization is continued. It should be noted that since there is only one local optimal particle in the first round of iteration, the local optimal particle is the global optimal particle of this cycle or algorithm. Except for this special case, after each subsequent round of iteration confirms the local optimal particle of this round, the local optimal particles of multiple rounds of iterations are compared based on the task value matrix to determine the global optimal particle of this cycle.

[0068] When the preset number of cycles is not reached, a group of second random particles is generated based on the preset random particle model, and a preset genetic algorithm is applied to perform random genetic mutation operations on these second random particles according to the previously determined local optimal particles or global optimal particles. The genetic algorithm introduces new diversity through operations such as selection, crossover and mutation, thereby avoiding falling into the local optimal trap. After genetic mutation, the new particles obtained become the optimal random particles again, and step S104 is executed to perform a new round of optimization.

[0069] When the number of cycles reaches the preset number of cycles, the final determined global optimal particle is decoded based on the preset decoding matrix. The decoding process converts the global optimal particle into a specific task allocation scheme, that is, multiple optimal task pairs. Based on these optimal task pairs, the task allocation scheme at the current time t is generated to guide the resource allocation in the actual search and rescue operation.

[0070] In the embodiment provided by the present invention, at the beginning of each iteration, new random particles are generated to achieve mutation. These particles are randomly cross-transferred with the local optimal particles determined in the historical iteration or the global optimal particles in the historical cycle to simulate the genetic process, ensure that the particles are adjusted based on the latest information, ensure the diversity of the particle group, and promote the transmission of excellent features. Subsequently, by optimizing all local optimal particles again, a secondary evaluation of multiple high-value solutions is completed, and the solution with the smallest variance of task value is selected as the global optimal particle, realizing multi-level continuous optimization from the local level to the global level, and improving the quality and accuracy of task allocation. It should be understood that the local optimal particle refers to the particle with the largest sum of task values ​​and the smallest variance between task values. The maximum sum of task values ​​means that the task allocation scheme represented by the particle can provide higher overall benefits; and the minimum variance between task values ​​ensures the balance of task allocation and avoids the value of some tasks being too high or too low. After each round of iteration, these local optimal particles are recorded for further optimization. The global optimal particle also requires that the total value of tasks is maximized and the variance between task values ​​is minimized, but it is the best solution selected from multiple local optimal particles, further improving the overall efficiency and balance of task allocation.

[0071] The total value of different task allocation schemes may be the same. For example, in the first task allocation scheme, the task value of task pair A is 1, and the task value of task pair B is 8; in the second task allocation scheme, the task value of task pair C is 4, and the task value of task pair D is 5. In the first task allocation scheme, the task value of task pair A is too low, and the matching degree between the cluster and the task in task pair A is low. The drone cluster may consume too much time cost, distance cost, etc. when executing, and the quality of task completion may be affected. In the second task allocation scheme, the variance between task values ​​is smaller, the task allocation is more balanced, and the completion quality of each task is higher and the time consumed is shorter. In other words, extreme high-value tasks and low-value tasks may be combined in task allocation. The task load is kept relatively uniform through the variance between task values ​​to ensure high-quality execution of tasks. Especially in the actual scenario where the number of drone clusters is less than the number of search and rescue tasks, it is necessary to allocate and complete the search and rescue tasks in batches. At this time, according to the variance between task values, a more balanced task allocation can be selected while the resource allocation is highly efficient, so that the task allocation can adapt to the actual scenario of scarce execution resources, improve the quality of task completion, and reduce the time cost.

[0072] When the number of drone clusters is less than the number of search and rescue missions, it is necessary to selectively complete the missions in batches. The missions executed in the first mission allocation scheme may be different from those executed in the second mission allocation scheme. Taking the remaining effective time of the mission as an example, since the ratio between the remaining effective time and the cost time is proportional to the mission value, the longer the remaining effective time and the shorter the cost time, the higher the mission value. The corresponding mission pair B may have more time for the mission. If the search and rescue mission is executed first, and other tasks are postponed to the second batch, they may be difficult to complete due to time delays. Therefore, choosing the second mission allocation scheme can effectively improve the overall mission completion rate.

[0073] Furthermore, in order to ensure that resources can be effectively utilized in emergency situations, the priority of the search and rescue mission will be set in advance, but the flexibility of the priority set in advance for a single task is limited, which may lead to uneven resource allocation in actual search and rescue scenarios. By screening the task allocation scheme based on the variance between task values, the priority limit can be dynamically broken through when the execution resources are less than the task, and the priority of multiple search and rescue tasks can be better optimized according to the characteristics of the cluster and the task in the actual search and rescue scenario. As a result, the first batch of search and rescue tasks can be completed more quickly, while reducing the possibility of delays in the golden search and rescue period of the second batch of tasks, thereby improving the overall completion rate of the search and rescue tasks to minimize losses and save lives. At the same time, this breakthrough in the priority limit is also limited by the total task value, taking into account both the task execution efficiency and the task completion rate. In some embodiments, when a high-priority task is not executed in the first batch, a corresponding reminder can be generated to enable human intervention to confirm whether to enable the execution of the backup drone cluster.

[0074] In order to speed up the calculation and improve the timeliness of the allocation plan, a full matrix approach is adopted to complete the value calculation of multiple possible task allocation schemes at one time, and quickly determine the best decision from a variety of feasible task allocation schemes in a short time, which greatly shortens the calculation time. It is highly adaptable to the needs of changeable situations and limited task allocation time in search and rescue missions, ensuring that search and rescue operations are carried out in a timely and effective manner.

[0075] In some embodiments, the set of N available drone clusters is , the set of K search and rescue missions is ; Step S101, comprising:

[0076] At the present moment , the set of legal clusters is defined as:

[0077] ;

[0078] The set of legal tasks is defined as:

[0079] ;

[0080] in, Is a cluster The number of drones, is an integer greater than or equal to zero; cluster Available The set of task execution capabilities is , when the cluster Not capable of executing The ability to perform tasks , when the cluster Possess the ability to execute The ability to perform tasks ;

[0081] For traversing tasks from 1 to 2 Task judgment cluster Whether it has the ability to execute tasks, if there is at least one task type Mission , cluster For Ability to perform tasks of different types ,but ,otherwise ;

[0082] It is the time criterion used to judge the cluster Can you Arrive at the mission location and perform the mission within the life cycle ; For traversing tasks from 1 to 2 Task judgment cluster Can you The operation of reaching any search and rescue mission location within the life cycle of , cluster Execute the task Time criterion ,but ,otherwise ; Among them, the life cycle of a task is used to limit the time window for the start or completion of a task, that is, the executable time period of the task. Correspondingly, To determine the cluster Can you Arrive at the task location within the task start time window, and / or cluster Can you The task is completed within the time window of the task completion ;

[0083] For the task The end execution time of the task During task execution time , indicating the task The executable task time has not yet ended;

[0084] For traversing from the first cluster to the Cluster judgment task Whether the operation can be performed by any cluster, if there is at least one cluster , cluster The task type is Mission ability, then ,otherwise ;

[0085] For traversing from the first cluster to the Cluster judgment task Whether the operation can be completed by any cluster during the life cycle, if there is at least one cluster , cluster Execute the task Time criterion ,but ,otherwise .

[0086] Specifically, for search and rescue tasks, it is determined whether the current time is between the task start time and the task end time, and whether each task can be completed by a certain cluster within its life cycle. A legal task requires that at least one cluster has the task execution capability required by the task, and the task is within the executable validity period, and at least one cluster can arrive at the task location in time to perform the task within the required executable time period under its time and space constraints. For drone clusters, it is confirmed whether the drone cluster has enough energy and surviving members to complete the task, and whether the drone cluster has the ability to complete the search and rescue task within the life cycle of any search and rescue task. A legal cluster requires that the number of cluster members is greater than 0, that it has the task execution capability required by at least one task in the legal task, and that the task location and task start time of at least one task in the legal task can be timely (within the executable time period of the task) to arrive and execute under the position and speed constraints of the cluster at the current time t. It should be understood that by establishing a legality criterion, filtering out legal search and rescue tasks and drone clusters, generating a set of legal tasks that can be assigned and a set of legal clusters, it is ensured that the drone clusters and search and rescue tasks participating in the search and rescue task assignment are valid, and that resources are used most effectively, thereby improving the success rate of search and rescue operations. Exemplarily, the tasks involved in the selection of legal tasks and legal clusters are not completed, i.e. .

[0087] In some embodiments, the time criterion Defined as: ;

[0088] Among them, drone swarm The average flight speed is and the center position is ,Task The center position is , is an operator for calculating the magnitude of a three-dimensional vector, It's a task The start execution time of the task. Thus, the time required for the cluster to reach the task location under the position and speed constraints at the current time t is calculated. Combining the start execution time of the task and the current time t, it can be determined whether the cluster can reach the task location in time and execute the task.

[0089] In some embodiments, according to the characteristics of the search and rescue mission, the positive and negative relationship and enabling relationship of the mission and cluster-related attributes on the mission value are determined, and the mission value function is designed. The preset mission value function The expression is: ; Among them, cluster Available The set of task execution capabilities is , cluster Not capable of executing The ability to perform tasks , cluster Possess the ability to execute The ability to perform tasks ; It is the task completion flag. When the task is not completed , when the task has been completed ; is the task execution priority, , is a set of positive integers; It's a task The remaining effective time, Is a cluster Execute the task The cost time spent. Among them, the task execution capability parameter is , the task completion status parameters are ,task value assessment parameters include but are not limited to: , , .

[0090] First, by presetting the task value function The task execution capability parameters and task completion status parameters are used to conduct basic feasibility screening. To identify the task pair Medium Cluster Is the task execution capability consistent with the task? Corresponding to the task type, the cluster Possess the ability to execute The ability to perform tasks , the task value can be calculated normally, otherwise the task value is always 0, thereby quickly identifying invalid task pairs; Used to identify completed tasks. , corresponding to If the value of is 1, the task value can be calculated normally, otherwise the task value is always 0, so that the completed invalid tasks can be quickly identified. As a result, the calculation speed of the task value is improved, redundant data is quickly eliminated, the waste of computing resources is avoided, and the timeliness of task allocation is improved.

[0091] It should be understood that in the screening of N available drone clusters and K search and rescue tasks, legal clusters and legal tasks are selected. Since the task may be completed by multiple clusters, the amount of calculation required to accurately generate task pairs based on the task type and the task execution capability of the cluster is large, and the subsequent generation of a standardized task value matrix requires secondary sorting, which is a complicated and inefficient process. The embodiment of the present invention generates task pairs through simple one-to-one matching, which can quickly and comprehensively include various types of task allocation possibilities. By presetting the task value function The clusters and tasks are quickly screened for feasibility, the task value of invalid task pairs is evaluated as 0, the task value variance between redundant invalid task pairs and valid task pairs is expanded, and in the particle optimization process, the influence of invalid task pairs in task allocation is further eliminated by minimizing the variance between task values, thereby eliminating the error influence of redundant design and effectively improving the generation efficiency of the task value matrix. Furthermore, there is also redundancy in the matrix cells of the task value matrix. The embodiment of the present invention makes these redundant cells and the particle matrix of random particles cleverly form a matching relationship, and the reasonable use of the redundant part simplifies the subsequent algorithm process of optimizing the random particles based on the task value matrix on the one hand, and on the other hand, the full matrix speeds up the calculation speed and improves the timeliness of the allocation plan.

[0092] Furthermore, by presetting the task value function The task value evaluation parameter evaluates the task value of the task pair. Introducing the task priority setting, task execution priority Positively correlated with task value, ensuring that high-priority tasks are given priority in resource allocation; the remaining effective time of the task and cost time The ratio between is positively correlated with the task value. The cost time is the time cost required for the cluster to complete the task. When the time cost increases, the overall task value will decrease, so that the task value depends not only on the importance of the task itself, but also on the feasibility of actual execution. Among them, the time cost is determined based on the average flight speed of the cluster, the location of the cluster, the location of the task, and the number of drones in the cluster. The time cost is positively correlated with the time it takes for the cluster to reach the task. The longer the time, the higher the cost. The time cost is negatively correlated with the number of drones in the cluster. The more the number, the higher the cost.

[0093] It should be understood that a dynamic adjustment scoring mechanism for task pairs can be realized through the preset task value function, which comprehensively considers multiple factors such as task priority, cluster capability, remaining time and time cost, and comprehensively evaluates the suitability of drone clusters to perform tasks, ensuring that the score reflects both the urgency of the task and the feasibility and efficiency of actual execution, thereby maximizing the search and rescue efficiency and success rate.

[0094] at the same time, All parameters in the dynamic change over time, the task progress and the environment change correspondingly Especially in scenarios where fewer drone clusters are needed for search and rescue missions, when one or more clusters complete the first batch of tasks and then reassign tasks, the calculation is recalculated. The time is based on the latest task status, cluster status and time parameters (such as , ), ensuring that task allocation is a dynamic response to changing task requirements and environmental conditions, improving the adaptability of task allocation to search and rescue scenarios, and thus improving the success rate of search and rescue operations. In other words, from the selection of tasks and clusters, the design of task value functions, the establishment of random particle models, to the final combination of particle swarm optimization and genetic algorithms, taking into account the high adaptability and high completion rate of task allocation from multiple dimensions such as task allocation quality, resource task asymmetry, and task allocation computational efficiency, each step is closely linked, and can make fast and accurate decisions in a complex and changing environment, ensuring the adaptability, balance, and high quality of task allocation and execution.

[0095] In some embodiments, the task Start execution time and end time , the difference between the two is the life cycle of the task; the actual remaining time of the task at the current time t , defined as , then the task The remaining effective time is defined as:

[0096] ;

[0097] In order to adapt to the actual situations that may exist in various search and rescue missions in different search and rescue scenarios, the remaining effective time is flexibly determined in three situations. First, when the actual remaining time is greater than the life cycle, the remaining effective time is taken as the life cycle. Second, when the actual remaining time is less than 1 second, that is, the mission has ended or there is less than 1 second left to end, the remaining effective time is taken as 1 second. Third, when the actual remaining time is less than the life cycle, the remaining effective time is taken as the actual remaining time.

[0098] In some embodiments, the cluster Execute the task The time spent is , defined as:

[0099]

[0100] Among them, drone swarm The average flight speed is and the center position is , the number of drones is ,Task The center position is , It is an operator that calculates the Euclidean distance between two three-dimensional coordinate points. The distance between the cluster position and the task position is proportional to the cost time, and the speed and quantity of the cluster are inversely proportional to the cost time. Therefore, the cost time comprehensively considers the distance factor, speed factor, and quantity factor, and can accurately reflect the cluster Execute the task The time cost.

[0101] In some embodiments, the method further comprises: for OK The identity matrix of the column is the original particle before random transformation, and the expression is: ;in, ; represents the order of the matrix, which is the larger value of the total number of elements in the legal cluster set and the total number of elements in the legal task set. is the total number of legal clusters, is the total number of legal tasks, is an operator for calculating the number of elements in a collection. Is the maximum comparison output operator.

[0102] By random exchange The diagonal matrix of order The random permutation matrix obtained by the column vector position order is expressed as: ;

[0103] is a diagonal matrix and is expressed as: ;in, is used to randomly swap diagonal matrices Random operators for column vectors; The matrix is Order square array.

[0104] Generate a preset random particle model based on the random permutation matrix and the identity matrix to obtain the random particle matrix It should be understood that the preset random particle model is used to compare the legal tasks and legal clusters to obtain the larger value of the two numbers, and this value is used as the order of the random permutation matrix, and the random permutation matrix is ​​generated by randomly arranging the column order of the random permutation matrix. random particles. For example, It is a full-rank sparse matrix. The full-rank sparse matrix can accelerate matrix calculation and further improve the timeliness of task allocation.

[0105] In some embodiments, step S102 further includes: , As the matrix side lengths, generate The effective task value matrix of each task pair is filled into the corresponding matrix position as a matrix element; according to the , The difference between them generates a supplementary matrix with zero value, and the supplementary matrix is ​​merged with the effective task value matrix to generate the task value matrix with the same matrix side length.

[0106] In some embodiments, an order is initialized to The zero matrix of is the number of legal clusters The number of legal tasks The larger value of is taken, with legal tasks as columns and legal clusters as rows, and the preset task value function is used to calculate the task pairs under each column index and row index. The human-machine cluster label and the search and rescue task label are used as the row index and column index respectively, and the task value is filled into the corresponding cell of the zero matrix to obtain the task value matrix.

[0107] In some embodiments, the task value matrix is defined as:

[0108] ;

[0109] in, The matrix is Order square array; and They are the number of legal clusters and the number of legal tasks, respectively; is the number of legal clusters The number of legal tasks The larger value of is the matrix transpose operator; yes A column vector with all zeros in row 1 and column 2; Is OK The effective task value matrix composed of columns, Defined as:

[0110] ;

[0111] in, The matrix is OK Matrix of columns.

[0112] All matrices in the matrix algorithm must be square. In order to adapt to the special situation of unequal numbers of drone clusters and tasks in search and rescue scenarios, the larger value of the number is used as the matrix side length so that each task pair can find the corresponding position in the matrix to calculate and store the task value, realizing the adaptive editing operation of the matrix. For example, 3 legal clusters, such as , , ; 2 legal tasks, such as , ; Legal tasks are columns and legal clusters are rows, which are paired to form multiple task pairs, such as , , , , , . Calculate the task values ​​of the above 6 task pairs respectively, fill the results into the corresponding positions of the particle matrix according to the rows and columns of the task pairs, and get the effective task value matrix. The 3 legal clusters and 2 legal tasks generate a 3-row 3-column zero matrix, one part of which is the effective task value matrix, and there will be vacancies in the remaining part of the matrix. The task value matrix is ​​obtained by filling in zeros.

[0113] In some embodiments, the optimization random particle and the local optimal particle are particle matrices; steps S104 to S105 further include: according to the task value matrix , the particle matrix, determine the multiple task values ​​of the current particle matrix; add the multiple task values ​​to obtain the sum of the task values ​​of the particle matrix, and calculate the variance between the multiple task values ​​to obtain the variance between the task values ​​of the particle matrix. Specifically, the task value matrix records the task values ​​of each task pair, and the numerical distribution in the particle matrix defines a task allocation scheme containing multiple task pairs, and the task values ​​of these task pairs are selected from the task value matrix. Add all the individual task values ​​to obtain the sum of the task values ​​of the task allocation scheme corresponding to the particle matrix. In order to further measure the balance between different task values, calculate the standard deviation or variance of these individual task values.

[0114] In some embodiments, the number of iterations of the particle optimization algorithm is set to a positive integer. , the number of random particles generated in each iteration is a positive integer Steps S104 to S106 also include:

[0115] No. The local optimal particle of the iteration Defined as:

[0116] ;

[0117] in, is a set of positive integers; is the random particle number of this iteration, It is The iteration generated The first of the random particles Random particles; is the optimal random particle, when =1, is the first random particle, when hour, Including part of the second random particles , and some The local optimal particle of the iteration and / or part of the global optimal particles ; It is Random particles The random permutation matrix, matrix is the optimal random particle The permutation matrix of ; Represents the matrix and Perform matrix multiplication, Express and Perform matrix dot multiplication calculation; It means calculating the cumulative sum of all elements of the matrix in brackets; For this iteration traverse the optimal random particles and obtain the particle matrix with the largest total task value; Used to calculate the standard deviation of all non-zero elements in the matrix; It is used to obtain the local optimal particle with the smallest variance between task values ​​from the particle matrix with the largest total task value.

[0118] In some embodiments, the global optimal particle of this cycle Defined as:

[0119] ;

[0120] in, is a set of positive integers; is the number of cycles of this algorithm, is the iteration number, It is The local optimal particle in the iteration A random permutation matrix of ; Represents the matrix and Perform matrix multiplication, Express and Perform matrix dot multiplication calculation; It means calculating the cumulative sum of all elements of the matrix in brackets; Used to traverse the generated local optimal particles and obtain the particle matrix with the largest total task value; Used to calculate the standard deviation of all non-zero elements in the matrix; It is used to obtain the global optimal particle with the smallest variance between task values ​​from the particle matrix with the largest total task value.

[0121] It should be understood that each round of iteration requires The optimization random particles are optimized separately, that is, each round of iteration needs to be Calculate from The local optimal particle is selected from the optimization random particles. The local optimal particle generated by the first round of iteration is There are local optimal particles. After each round of iteration, a loop is entered, and each loop needs to The optimal particle is searched among several local optimal particles, and a global optimal particle is selected from these local optimal particles.

[0122] In some embodiments, the preset genetic algorithm is defined as:

[0123] ;

[0124] in, is the global optimal particle The permutation matrix of ; It is a genetic operation that copies part of the column data of one matrix to the corresponding column data of another matrix, for example It is a random Copy some columns of the matrix to the matrix The genetic operation of the corresponding column data location. The operation result of the preset genetic algorithm can ensure that each column of its matrix has only one non-zero element with a value of 1 to achieve the uniqueness constraint. Among them, the number of iterations ,when Sometimes and Established.

[0125] It should be understood that at the beginning of each optimization iteration, random particles are generated and randomly cross-transferred with local optimal particles and / or global optimal particles to achieve genetic variation. During each optimization iteration, the value and variance of each particle matrix are calculated using the preset task value function, and the local optimal particle with the largest value and the smallest variance in this iteration and the global optimal particle of this algorithm are saved.

[0126] In some embodiments, Including part of the second random particles , and some The local optimal particle of the iteration and / or part of the global optimal particles , according to The random genetic variation content and the permutation matrix of the local optimal particle and / or the permutation matrix of the global optimal particle The matrix columns are exchanged, that is, the permutation matrix is ​​generated according to the random changes of the corresponding particle matrix so that and Maintaining the correct permutation relationship, the permutation matrix can restore the randomly changed particles to the position order that can be decoded by the preset mapping matrix.

[0127] In some embodiments, the local optimal particle may also be randomly mutated based on the global optimal particle and / or the local optimal particles of other rounds, and the global optimal particle is selected as the local optimal particle with the largest total task value and the smallest variance between task values ​​among all mutated local optimal particles.

[0128] In some embodiments, the preset decoding matrix includes a preset mapping matrix and a permutation matrix of the global optimal particle; S107 includes:

[0129] The task allocation scheme matrix is ​​obtained by multiplying the global optimal particle with the inverse of the permutation matrix of the global optimal particle and multiplying it with the preset mapping matrix. , the calculation formula is:

[0130] ;in, is the global optimal particle, is the permutation matrix of the global optimal particle, is the preset mapping matrix, Representation Matrix Inverse matrix operation of ; Representation Matrix The inverse matrix operation of Represents the matrix With the matrix Perform matrix multiplication, Represents the matrix With the matrix Perform matrix dot product operation.

[0131] The task allocation scheme matrix The task pair in is taken as the optimal task pair, and the set of multiple optimal task pairs is:

[0132] ;

[0133] in, Represents the task allocation matrix No. Line The optimal task pair for the column.

[0134] Specifically, the global optimal particle is inversely multiplied with the permutation matrix of the global optimal particle to achieve unified arrangement of the randomly disrupted column order during the algorithm operation, so that the unified global optimal particle can correspond one-to-one with the original task pair arrangement order of the preset mapping matrix, and the decoding is completed after the two are multiplied.

[0135] In some embodiments, the preset mapping matrix is defined as:

[0136] ;

[0137] in, The matrix is Order square array; and They are the number of legal clusters and the number of legal tasks, respectively; is the number of legal clusters The number of legal tasks The larger value of is the matrix transpose operator; yes A column vector with all zeros in row 1 and column 2; Is OK The effective mapping matrix composed of columns, Defined as:

[0138] ;

[0139] in, The matrix is OK Matrix of columns.

[0140] Specifically, the side length of the preset mapping matrix is ​​equal to the side length of the global optimal particle. According to the row number and column number of the cell of each matrix element in the preset mapping matrix, it can correspond to the task pair. and preset mapping matrix When multiplied, the global optimal particle The only element with a value of 1 in each column will select the matrix element at the corresponding position and translate it to get the optimal task pair. Due to the uniqueness constraint, only one matrix element is selected for each column or row, and the matching relationship between clusters and tasks is one-to-one. It should be understood that the preset mapping matrix has an associated relationship with the task value matrix. The task value matrix is ​​used to matrix the task pairs, and the preset mapping matrix is ​​used to restore the matrix to task pairs.

[0141] See also Figure 2 , Figure 2 FIG. 1 is a schematic diagram of a UAV cluster search and rescue task allocation scenario based on discrete particle swarms provided by an embodiment of the present invention. Figure 2 As shown, it contains four drone clusters , , , and three search and rescue missions , , Among them, drone swarm , , , The position coordinates are , , , ; Drone swarm , , , The velocity vectors are , , , ; Drone swarm , , , The number of cluster members is , , , integer ; Drone swarm It only has the ability to perform the second and third types of tasks, that is, for drone swarms have and ; Drone swarm It only has the ability to perform the third type of mission, that is, for drone swarms have ; Drone swarm It only has the ability to perform the first type of mission, that is, for drone swarms have ; Drone swarm Possess the ability to execute The capability of the type of mission, namely for drone swarms have In addition, search and rescue missions , , The position coordinates are , , ; Search and rescue mission , , The types are respectively Class 2 ( )、Category 3( ), No. kind( ); Search and rescue mission , , The tasks have the same priority and satisfy other legal task constraints.

[0142] like Figure 2 As shown in the figure, based on the above content, the UAV cluster search and rescue task allocation method based on discrete particle swarm finally generates the task allocation scheme including: 1) UAV cluster Assigned to perform search and rescue missions that are close in distance and of matching mission type 2) Drone swarm Assigned to perform search and rescue missions that are close in distance and of matching mission type , drone swarm Although it also has the ability to perform search and rescue missions capability, but compared to drone swarms The farther the distance, the lower the value of the preset task value function; 3) Drone swarm Not assigned to any search and rescue missions due to the drone swarm There is no capability to perform any existing mission type, so drone swarms Will automatically return to the airport; 4) Drone swarm Assigned to perform search and rescue missions that are close in distance and of matching mission type .

[0143] In practical applications, three drone clusters are , , , and three search and rescue missions , , The algorithm is fed with the relevant attributes, including the total number of drone members, positions, speeds, executable task types (i.e., task execution capabilities) of the drone cluster, the task type of the search and rescue task, the completion mark, the start and end time of the task, and the location of the task location. Based on the preset task value function, a collection of behavioral tasks (i.e., a collection of legal tasks) and a task value table (i.e., a valid task value matrix) listed as a collection of drone clusters (i.e., a collection of legal clusters) are generated, and based on this, the global optimal particle of this round of the algorithm is determined. The allocation plan results calculated by this round of the algorithm are generated according to the particle matrix. The calculation time consumed by the operation algorithm is about 0.093 seconds. The value calculation of multiple task allocation plans is completed at one time using the full matrix method, and the best decision is quickly determined from multiple task allocation plans in a short time. It is highly adapted to the needs of changing situations and limited task allocation time in search and rescue tasks, ensuring that the search and rescue operations are carried out in a timely and effective manner.

[0144] In some embodiments, in order to highly adapt to the actual search and rescue scenario, when the number of drone clusters is greater than the number of tasks, an embodiment of the present invention provides a drone cluster search and rescue task allocation method based on discrete particle swarms. Greater than When the method further comprises steps S201 to S203: S201, determining the task with the greatest task value of each cluster according to the task value, and generating a set of spare task pairs ; S202, obtain the set of the optimal task pairs The used clusters in the task list; and filtering out the set of backup task pairs according to the used clusters. The task pairs corresponding to the unused clusters in the cluster are obtained to obtain the remaining cluster task pair set ; S203, the set of the optimal task pairs , the remaining cluster task pair set The task pairs in are used as target task pairs, and the task allocation scheme at the current time t is updated according to the multiple target task pairs.

[0145] Specifically, each task pair is calculated using a preset task value function, and the task pairs of the maximum value tasks that each cluster can perform are recorded to generate a set of spare task pairs. The unassigned drone clusters in the set of optimal task pairs are supplemented with tasks according to the recorded maximum value tasks to obtain the final task allocation plan. It should be understood that in actual scenarios where the number of drone clusters is greater than the number of search and rescue tasks, oversaturation execution is adopted, and secondary task allocation is performed on drone clusters that are not assigned tasks so that they perform corresponding high-value tasks. Some search and rescue tasks are completed collaboratively by at least two drone clusters, which accelerates the execution speed of search and rescue tasks and further improves the practicality and adaptability of task allocation in actual search and rescue scenarios. Close collaboration between multiple drones forms a cluster operation mode. For example, through division of labor and collaboration, multiple tasks such as search, search and rescue, and material delivery are completed, and available resources are fully utilized to improve search and rescue efficiency.

[0146] In some embodiments, subtracting the drone cluster that has been assigned a task in the optimal task pair set from the backup task pair set to obtain the remaining cluster task pair set, and then combining the remaining cluster task pair set and the optimal task pair set to determine the target task pair set, including: determining the task with the greatest task value for each cluster according to the task value, and generating a backup task pair set , the expression is: ;in, Represents the set of legal tasks for the cluster The task with the largest task value is output; the target task pair set is determined based on the optimal task pair set and the backup task pair set ; ;in, is the optimal task pair set; It is a conditional set subtraction operator. Used in Exclude The optimal task in the used drone cluster; is a set addition and merging symbol; updates the task allocation scheme at the current time t according to the plurality of target task pairs.

[0147] In some embodiments, in order to highly adapt to the actual search and rescue scenario, when the number of drone clusters is less than the number of tasks, an embodiment of the present invention provides a drone cluster search and rescue task allocation method based on discrete particle swarms. Less than The method further includes S301 to S302: S301, when at least one of the clusters completes the corresponding task, updating the current time , the life cycle of the task, and the task execution capability of the cluster; S302, re-execute steps S101 to S107 to generate an updated current time task allocation plan.

[0148] Specifically, the task execution status of each cluster is continuously monitored. When a cluster is detected to have completed its assigned task, the task life cycle and the task execution capability of the cluster are updated. Exemplarily, multiple key parameters are updated synchronously, including updating the task execution capability according to the loss of the drone cluster during the task execution, the number of surviving drones, etc.; marking the cluster that has completed the task as available, making the cluster status an available drone cluster with task execution capability; updating the completion mark of the completed task, reducing the search and rescue tasks to be completed, etc. At the same time, the current time is updated. , re-execute the UAV cluster search and rescue task allocation method based on discrete particle swarm in steps S101 to 107, so as to re-execute the task allocation process to generate a new task allocation plan, and realize batch allocation and execution of tasks.

[0149] In addition to updating multiple key parameters, as time goes by, tasks are completed, and clusters are lost during search and rescue, legal clusters and legal tasks will change. Correspondingly, the task value matrix, the preset random particle model, and the random particles generated by it will be updated adaptively, such as updating the order of the particle matrix to highly adapt to the actual situation of the current search and rescue scene. At the same time, All parameters in the search and rescue scene are also adaptively changed with the dynamic changes of the search and rescue scene. When the task is assigned again, Recalculate All are based on the latest task status, cluster status and time parameters (e.g. , ), for example, the cost time will be recalculated based on the current location of the drone (which may be the location of the first mission) to enable the execution of the next batch of tasks nearby. This ensures that task allocation is a dynamic response to changing task requirements and environmental conditions, improves the adaptability of task allocation to search and rescue scenarios, and thus improves the success rate of search and rescue operations.

[0150] The embodiment of the present invention also provides a drone cluster search and rescue task allocation system based on discrete particle swarms, the system comprising: a central control station for centralized management and coordination of path planning and real-time monitoring of multiple drone clusters; each drone cluster includes multiple drone units; the drone cluster includes a first cluster and a second cluster; the first cluster is also used to implement the steps of the drone cluster search and rescue task allocation method based on discrete particle swarms as provided in any embodiment of the present invention, and the specific implementation of the steps can refer to the previous embodiments, which will not be repeated here. It should be understood that the drone cluster search and rescue task allocation method based on discrete particle swarms provided in the embodiment of the present invention adopts an efficient matrix calculation method, occupies a low amount of calculation and memory, can meet distributed deployment, reduce the computing power required for task allocation, can be used in the onboard computer of one or more drones in the first cluster, can reduce the dependence of drone search and rescue on ground stations, and can better meet the emergency use of search and rescue scenes. At the same time, lightweight matrix operations help to extend the endurance of drones, because less computing load means lower energy consumption, thereby better supporting long-term and large-scale search and rescue tasks.

[0151] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for allocating search and rescue tasks of drone clusters based on discrete particle swarms, characterized in that: The method comprises: S101, obtain N available drone clusters and K search and rescue tasks; filter the clusters and tasks according to the life cycle of each task and the task execution capability of each cluster, and obtain the current time t The set of legal clusters , The set of legal tasks , and generate multiple task pairs ,in, To command the cluster Execute the task The combination of tasks, , , is a positive integer and , is a positive integer and , , is an integer greater than or equal to zero; S102, according to the preset task value function , calculate the task value of each task pair, and generate A legitimate cluster, The task value matrix corresponding to the legal tasks ; The preset task value function The expression is: ； Among them, cluster Available The set of task execution capabilities is , cluster Not capable of executing The ability to perform tasks , cluster Possess the ability to execute The ability to perform tasks ; It is the task completion flag. When the task is not completed , when the task has been completed ; is the task execution priority, , is a set of positive integers; It's a task The remaining effective time, Is a cluster Execute the task The cost of time spent; S103: Obtain a preset random particle model, and generate a A first random particle is used as the optimal random particle, wherein the random particles generated by the preset random particle model are particle matrices, and the order of the particle matrix is and The larger value of ; S104: Based on the task value matrix right The optimization random particles are optimized to determine the local optimal particle of this iteration, and the local optimal particle of each iteration is recorded; wherein the local optimal particle is the optimization random particle with the largest total task value and the smallest variance between task values; execute step S105; S105: Based on the task value matrix Optimize multiple local optimal particles to determine the global optimal particle of this cycle, wherein the global optimal particle is the local optimal particle with the largest total task value and the smallest variance between task values; update the number of cycles , when the number of cycles When the preset number of cycles is reached, step S107 is executed; otherwise, step S106 is executed; S106: Generate based on the preset random particle model a second random particle; based on a preset genetic algorithm, according to the local optimal particle or the global optimal particle The second random particles undergo genetic mutation and obtain The optimized random particles are executed; step S104 is executed; S107, decoding the global optimal particle based on a preset decoding matrix to obtain multiple optimal task pairs, and generating a task allocation scheme at the current time t according to the multiple optimal task pairs.

2. The method according to claim 1, characterized in that The set of N available drone clusters is , the set of K search and rescue missions is ; Step S101 includes: At the present moment , the set of legal clusters is defined as: ； The set of legal tasks is defined as: ； in, Is a cluster The number of drones, is an integer greater than or equal to zero; cluster Available The set of task execution capabilities is ; For traversing tasks from 1 to 2 Task judgment cluster Whether it has the ability to execute tasks, if there is at least one task type Mission , cluster For Ability to perform tasks of different types ,but ,otherwise ; It is the time criterion used to judge the cluster Can you do it in the mission Arrive at the mission location and perform the mission within the life cycle ; For traversing tasks from 1 to 2 Task judgment cluster Can you do it in the mission The operation of reaching any search and rescue mission location within the life cycle of , cluster Execute the task Time criterion ,but ,otherwise ; For the task The end execution time of For traversing from the first cluster to the Cluster judgment task Whether the operation can be performed by any cluster, if there is at least one cluster , cluster The type of task to be executed is Mission ability, then ,otherwise ; For traversing from the first cluster to the Cluster judgment task Whether the operation can be completed by any cluster during the life cycle, if there is at least one cluster , cluster Execute the task Time criterion ,but ,otherwise .

3. The method according to claim 1, characterized in that The task pairs include invalid task pairs and valid task pairs; the preset task value function It includes a task execution capability parameter and a task value evaluation parameter. The task execution capability parameter is used to evaluate the task value of the invalid task pair as 0, and the task value evaluation parameter is used to perform value evaluation on the valid task pair.

4. The method according to claim 1, characterized in that: Step S102 also includes: by , As the matrix side lengths, generate An effective task value matrix is ​​obtained, and the task value of each task pair is filled into the corresponding matrix position as a matrix element; According to the , The difference between them generates a supplementary matrix with zero value, and the supplementary matrix is ​​merged with the effective task value matrix to generate the task value matrix with the same matrix side length.

5. The method according to claim 1, characterized in that Steps S104 to S105 also include: According to the task value matrix , the particle matrix, determining a plurality of the task values ​​of the current particle matrix; A plurality of the task values ​​are added together to obtain the sum of the task values ​​of the particle matrix, and the variances between the plurality of the task values ​​are calculated to obtain the variance between the task values ​​of the particle matrix.

6. The method according to claim 1, characterized in that Steps S104 to S106 also include: No. The local optimal particle of the iteration Defined as: ； in, is a set of positive integers; is the random particle number of this iteration, It is The iteration generated The first of the random particles Random particles; is the optimal random particle, when =1, is the first random particle, when hour, Including part of the second random particles , and some The local optimal particle of the iteration and / or part of the global optimal particles ; It is Random particles The random permutation matrix, matrix is the optimal random particle The permutation matrix of ; It means calculating the cumulative sum of all elements of the matrix in brackets; For this iteration traverse the optimal random particles and obtain the particle matrix with the largest total task value; Used to calculate the standard deviation of all non-zero elements in the matrix; It is used to obtain the local optimal particle with the smallest variance between task values ​​from the particle matrix with the largest total task value; The global optimal particle of this cycle Defined as: ； in, is the number of cycles of this algorithm, is the iteration number, It is The local optimal particle in the iteration A random permutation matrix of ; It means calculating the cumulative sum of all elements of the matrix in brackets; Used to traverse the generated local optimal particles and obtain the particle matrix with the largest total task value; Used to calculate the standard deviation of all non-zero elements in the matrix; It is used to obtain the global optimal particle with the smallest variance between task values ​​from the particle matrix with the largest total task value; The preset genetic algorithm is defined as: ； in, It is a genetic operation that copies part of the column data of one matrix to the corresponding column data of another matrix; is the global optimal particle The permutation matrix of .

7. The method according to claim 6, characterized in that The preset decoding matrix includes a preset mapping matrix and a permutation matrix of the global optimal particle; S107 includes: The task allocation scheme matrix is ​​obtained by multiplying the global optimal particle with the inverse of the permutation matrix of the global optimal particle and multiplying it with the preset mapping matrix. , the calculation formula is: ； in, is the global optimal particle, is the permutation matrix of the global optimal particle, is the preset mapping matrix, Representation Matrix Inverse matrix operation of ; The task allocation scheme matrix The task pair in is taken as the optimal task pair, and the set of multiple optimal task pairs is: ； in, Represents the task allocation matrix No. Line The optimal task pair for the column.

8. The method according to claim 1 or 7, characterized in that: when Greater than When, the method further comprises: Determine the task with the greatest task value for each cluster according to the task value, and generate a set of spare task pairs ; Get the set of optimal task pairs in the The clusters used in; Filter the set of standby task pairs according to the used cluster The task pairs corresponding to the unused clusters in the cluster are obtained to obtain the remaining cluster task pair set ; The set of the optimal task pairs , the remaining cluster task pair set The task pairs in are used as target task pairs, and the task allocation scheme at the current time t is updated according to the multiple target task pairs.

9. The method according to claim 1, characterized in that: when Less than When the method further comprises: When at least one of the clusters completes the corresponding task, the current time is updated , the life cycle of the task, and the task execution capability of the cluster; Re-execute steps S101 to S107 to generate an updated current time task allocation plan.

Images