Unmanned aerial vehicle distributed task allocation method for reducing calculation and communication overhead

Abstract

The invention discloses an unmanned aerial vehicle distributed task allocation method capable of reducing calculation and communication overhead, and the method reduces the calculation overhead through a local task reordering strategy on the basis of a performance influence algorithm, thereby achieving the addition of an unallocated task or a new task under the condition of not depending on task exchange. Meanwhile, aiming at the problem that an intelligent agent occasionally falls into an infinite loop problem that after one or more tasks are deleted, the intelligent agent is added to the same position of a sequence, a deletion record table and a new task are introduced to add a sequence mark to isolate deadlock. The method further comprises the steps of solving task conflicts in advance and reducing communication overhead through an estimation strategy; according to the novel network simplification algorithm, grouping is carried out by using the preference structure of the unmanned aerial vehicles, the communication network is simplified by reducing the number of closed-loop triangles based on a bidding distributed fragmentation pattern, and the two points can reduce the internal circulation of the algorithm, namely, reduce the number of times of communication between the unmanned aerial vehicles in one iteration of the algorithm.

Description

Technical Field

[0001] The present invention belongs to the field of distributed multi-UAV task allocation, and in particular, relates to a UAV distributed task allocation method that reduces calculation and communication overhead. Background Art

[0002] Judging from the results of many practical applications, drones have demonstrated extremely high combat and execution efficiency in the military and civilian fields. Not only do they effectively undertake reconnaissance and surveillance tasks in battlefield environments, they are also widely involved in precision strikes, material transportation, and communication support tasks, fully demonstrating their versatility and high adaptability. The outstanding performance and wide application of drones have attracted widespread attention from around the world. Countries have increased their investment in the research and development of drone technology, viewing it as an important means to enhance military combat capabilities and cope with complex combat environments. The development of drones has not only promoted the innovation of combat modes, but also profoundly affected the transformation of tactical theories and combat thinking, further accelerating the evolution of future wars towards intelligence and automation.

[0003] However, there are also some challenges in practical applications. Due to the complex and changeable task execution environment, the diversity and complexity of tasks continue to increase, and the limitations of a single drone in terms of endurance, load capacity and computing power make it difficult to efficiently cope with complex and changeable tasks. Therefore, the importance of multiple drones coordinating tasks has become increasingly significant. In addition, compared with the centralized architecture, the distributed architecture has stronger self-organization, weaker dependence on the computing and communication capabilities of the central server, and most importantly, there is no single point of failure. Therefore, distributed multi-drone task allocation has gradually become an important research direction.

[0004] Although domestic and foreign scholars have achieved certain theoretical and practical results in the distributed task allocation of multiple UAVs, with the complexity of the task environment, the increase in task scale and difficulty, the existing task allocation technology has been difficult to adapt to new application requirements or cannot be directly applied. Therefore, the distributed task allocation of multiple UAVs urgently needs further research. At present, most studies focus on a single optimization goal, such as maximizing the number of task assignments, improving task completion benefits, or minimizing task completion time, and the efficiency of task allocation still needs to be improved. In addition, most studies assume that the communication topology is fully connected, the bandwidth is unlimited, the channel is ideal, and reliable communication conditions are available. However, in actual applications, wireless communication between UAVs is affected by factors such as signal attenuation, path loss, and interference, and often presents unreliability, performance degradation, or limitation. Therefore, the present invention takes the rescue scenario as the background and conducts research on distributed multi-UAV task allocation under limited interaction conditions to adapt to multi-target tasks under actual conditions and reduce communication requirements. Summary of the invention

[0005] Purpose of the invention: The present invention aims to provide a method for allocating distributed tasks to drones that reduces computational and communication overheads. In terms of computational overhead, a local task reordering strategy is proposed, and deadlocks are isolated by introducing a task deletion table and adding task flags. In terms of communication overhead, for the outer loop of the algorithm, an estimation strategy is designed to resolve task conflicts in advance, reducing the overall number of iterations of the algorithm and thus greatly reducing the number of communications between drones; for the inner loop of the algorithm, the preference structure of drones is used for grouping, and a new network simplification algorithm based on bidding distributed broken patterns (BDBM) is proposed, which reduces the number of communications between drones in one iteration of the algorithm.

[0006] In order to achieve the above-mentioned object of the invention, the technical solution provided by the present invention is as follows.

[0007] A method for allocating distributed tasks of unmanned aerial vehicles (UAVs) with reduced computation and communication overheads comprises the following steps:

[0008] S1. Construct a UAV task allocation scenario, including deploying several UAVs for performing different tasks at different initial positions, and define a Boolean matrix H to reflect the compatibility between the UAV type and the target task. When the UAV type is compatible with the target task type, h = 1, h∈H, otherwise h = 0;

[0009] S2. Define the task allocation scenario: a set of n heterogeneous UAV systems U = [u1, u2, ..., u n ] and a set of m target tasks T = [t1, t2, ..., t m ];

[0010] The heterogeneous UAVs are UAVs with the same flight speed and energy consumption, and the communication between the UAVs is determined according to the initial communication topology;

[0011] S3, based on the task allocation scenario, set each drone to maintain a task list represented by p during task allocation i , its cardinality|p i | indicates drone u i The number of tasks to be executed is allocated, and the bid list c corresponding to the task list is maintained i , then u i The corresponding bidding information is represented by c i，k (p i ), winning drone collection z i ;

[0012] The objective function of the mathematical model of task allocation is to maximize the number of tasks N completed by the drone while minimizing the waiting time C for task execution. The expression is as follows:

[0013] J = min C (max N)

[0014] Constraints:

[0015] Among them, N i The meaning is drone u i The maximum number of Indicates drone u i Task Sequence i The start and deadline of the jth task;

[0016] Constraint C1 represents the UAV u i Can accommodate up to N i tasks;

[0017] Constraint C2 ensures that each task is assigned to at most one UAV, and there will be no task conflict;

[0018] Constraint C3 indicates that any combination of UAV’s task sequences is a subset of the task set T;

[0019] Constraint C4 ensures that UAV u i The tasks will not be executed beyond the start deadline of the corresponding tasks;

[0020] Constraints C5 and C6 jointly constrain UAV u i Only tasks of the type that are suitable for it can be completed, where k represents any target task, and constraint C6 represents the relationship between any task k and the current drone u i The adaptability of the tasks that can be performed;

[0021] Based on the above constraints, in addition to the task list p i and the corresponding bid list c i In addition, each drone also maintains a table of RPI records that remove mission performance impacts. i =[w1, w2, ..., w k , ..., w m ], add task performance impact IPI record table And task deletion record table ψ i , task deletion record table ψ i Used to record the complete mission path p of each UAV before any mission deletion occurs during the mission deletion phase i ;

[0022] Among them, removing task performance affects the RPI record table γ i In, w k Represents task t k From its allocated list p i After the removal operation is performed in , the performance impact of reducing the total cost of task allocation is calculated as The removal operation is represented as The calculation method is:

[0023]

[0024] Where b represents t k In the list p i The position in

[0025] Add task performance impact record table i * middle, Represents task t k Assign to drone u i And its task list p i Performing Add Operation After that, the performance impact of the total cost increase of task allocation is expressed as The calculation method is:

[0026]

[0027] Among them, add operation Indicates that the task t k Add to listp i The lth position of ;

[0028] S4, implement UAV task allocation based on PI-MaxAvg algorithm, and use the result after one execution as the preliminary task sequence, and then use the bidding information and task sequence results in the PI-MaxAvg algorithm to simplify the subsequent UAV cluster grouping and network topology;

[0029] S5. According to the initial task sequence in step S4, in the same group, the task ID corresponding to the lowest RPI value of the drone is the same as the task ID corresponding to the lowest RPI value of another drone. Any group is represented by G p :

[0030]

[0031]

[0032] The grouping based on the initial task sequence includes grouping the drones according to their initial bids, and determining an adapted task type according to the group;

[0033] S6, by reducing the number of closed-loop triangles in the communication network to reduce the communication load when solving large networks, specifically, each drone builds a network G n To describe the edges it thinks should be kept, it does this:

[0034] For drones i , Indicates drone u i Capable of receiving drone u j Message sent when each drone is initialized will be set to all 0 matrices. j Able to work with drones i When communicating, and the mission lists of the two drones conflict, the matrix Will be set to 1;

[0035] The calculation method is:

[0036]

[0037] Next, each drone receives messages from its neighboring drones about the edges they need to maintain, and then adds these edges to its G n In , the operation is defined as:

[0038]

[0039] Final expansion To ensure that you i Directly or indirectly receive the original communication topology with the drone u i A drone that can communicate normally j ;

[0040] S7, each UAV sends a task sequence list p to other UAVs within the simplified communication topology constraints k 、RPI Record Table γ k and drone information update timestamps k ; When the drone i Received from drone u k After the message is received, it will be based on γ k and k Two vector judgments are updated;

[0041] S8. Each drone compares the information in the consensus list with the information in its current list and deletes it based on the consensus result. At the same time, it updates the RPI list γ i To prevent deadlock, the previous task sequence p will be deleted. i Record to task deletion record tableψ i middle;

[0042] S9, for the task-related drones, which are drones that currently have task conflicts, obtain the conflict information from the winning messages of other drones;

[0043] UAV iFind out other drones u in this iteration from the received messages j The winner vector z j , then find out which drone i The scheduling of p i All drones with mission conflicts j , which is calculated as follows:

[0044] L = {j | z j [p i ]≠z i [p i ], j∈N, j≠i}

[0045] The variable L is used to store the information related to the drone u. i There are conflicting drone IDs;

[0046] In each iteration, the drone u i Computational related drones j task fractions of the UAVs and estimate their task selection and the resulting schedule, thus avoiding task conflicts in advance and reducing communication between UAVs;

[0047] For the UAV scheduling process, the estimated h is based on the task-related UAV u k The current schedule p k , bid information k 、UAV k Location k , which is expressed as follows:

[0048]

[0049] S10, each UAV traverses all executable tasks and task lists according to the task status and task adaptability conditions i All addable locations in Calculate the IPI value and determine the feasible location l of the task.

[0050] Furthermore, in step S4, in the PI-MaxAvg algorithm, the inclusion of performance impact and the removal of performance impact are represented as IPI-MaxAss and RPI-MaxAss respectively. If the task can be included in the scheduler and meets the time limit, its IPI-MaxAss is set to no cost, and the calculation method is as follows:

[0051]

[0052] in, These tasks q Compatible with drone features and not yet included in p i and RPI-MaxAss is greater than 0;

[0053] RPI-MaxAss is defined as:

[0054]

[0055] Among them, unassigned tasks are initially set to have a fixed maximum RPI-MaxAss, defined as a constant of U, and assigned tasks are initially set to 0.

[0056] Furthermore, step S6 is to eliminate closed-loop triangles based on Broken-motifs to improve communication efficiency, and the calculation method is as follows:

[0057] If there is a drone u k , k≠i, G BM (i, k)==1 and G BM (k, j)==1, then eliminate u i with u j Direct communication mode, namely G BM (i, j)==0 and G BM (j, i)==0;

[0058] The final step in simplifying the communication topology is to expand Specific process:

[0059] if And it satisfies the Broken-motifs theory conditions, then

[0060] Further, in step S7, when the drone u i Receive data from drone u k After receiving the three types of information, you can choose the following operations:

[0061] (1) Update:

[0062] w i，k =w j，k , z i，k =z j，k

[0063] The above formula represents the UAV u i Full reception of drones k news, and believes that drone u k Able to obtain task j;

[0064] (2) Reset:

[0065] w i，j =0

[0066]

[0067] The above formula represents the UAV u i Think u k The information is unreliable, and the existing knowledge of task j is also cleared;

[0068] (3) Leave:

[0069] w i，j =w i，j , z i，j =z i，j

[0070] The above formula represents the UAV u i Receive all k 's message, but still thought that he had obtained task j.

[0071] Furthermore, the step S8 includes performing the following operations:

[0072] Determine the drone u i List of pending deletions that conflict with the consensus list d i , the calculation method is as follows:

[0073] d i =p i [z i [p i ]≠i]

[0074] To remove the tasks in the deletion list one by one, the calculation method is as follows:

[0075]

[0076] in, It indicates the updated value of RPI after the removal operation. Each time the removal operation is performed, the one with the greatest impact is removed first, until the list of items to be deleted d i Is empty or the above formula no longer satisfies the judgment condition of being greater than 0.

[0077] Further, step S9 takes into account the drone u i Unknown mission related drone j In order to maintain the low traffic advantage of the market-based method, the location information of the drone will not be transmitted additionally. A method based on historical bidding is adopted to predict the location of the drone, in which the performance impact function is only related to the location of the drone, the actual location of the task, and the location of the task in the drone task sequence. After consensus, u i Is a known drone j The bidding information and the task information it obtains;

[0078] The inverse function of the performance impact function is used to obtain the drone u j Distance pj The mathematical calculation method for the distance of the task is as follows:

[0079]

[0080] in, Represents p j The position information of the task in the sequence, v j Indicates drone u j speed;

[0081] The drone u is calculated from the above formula i The distance d to task j ij After that, the drone i The position is located at a distance d from the position of task j. ij On a circle with radius

[0082] By determining the intersection of three circles, we can see that only three different tasks j are needed to obtain drone u. i location.

[0083] Furthermore, the step S10 further includes:

[0084] Calculate the IPI value and compare it with the RPI value to determine the task with the greatest impact:

[0085]

[0086] When q>0, if task t k Not the drone u in step S9 i Prediction p k(t+1) If you want to perform the task in Add the selected tasks to the list i l in the middle;

[0087] Repeat the above calculation until p i If the list cannot accommodate more tasks or q is less than 0 after IPI update, the task addition ends;

[0088] After the task addition phase is over, update and remove the RPI that affects performance.

[0089] Beneficial effects: Compared with the existing distributed task allocation technology based on performance impact, the substantial characteristics and significant effects of the method of the present invention include the following two aspects:

[0090] (1) In problem modeling, a variety of complex real-world requirements are taken into consideration, with a multi-UAV rescue scenario as the background. This includes considering the dual-objective problem, namely maximizing the number of completed tasks while minimizing the waiting time for tasks. These are undoubtedly the two most important goals in a rescue scenario. Strict time window constraints are also considered, so the waiting time for rescue personnel is guaranteed, which is in line with the actual situation, and the established mathematical model is more reasonable.

[0091] (2) In terms of the solution algorithm, since the problem is a combinatorial optimization problem of ST-SR-TA with strong NP difficulty, the algorithm adopted is a distributed task allocation algorithm based on performance impact PI. After actual testing, the algorithm has large computational and communication overheads. In order to improve the performance of the algorithm, different methods are designed in terms of computation and communication. Specifically: in terms of computational overhead, a local task reordering strategy is proposed, and deadlocks are isolated by introducing a task deletion table and adding task flags. In terms of communication overhead, for the outer loop of the algorithm, an estimation strategy is designed to resolve task conflicts in advance, reducing the number of iterations of the algorithm as a whole and thus greatly reducing the number of communications between drones; for the inner loop of the algorithm, the preference structure of drones is used for grouping, and a new network simplification algorithm based on bidding distributed broken patterns (BDBM) is proposed, which reduces the number of communications between drones in one iteration of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 is an overall flow chart of the method of the present invention;

[0093] Figure 2 It is the initial communication topology scheme diagram in the algorithm of the present invention;

[0094] Figure 3 A result diagram of the number of tasks assigned to the present invention;

[0095] Figure 4 A task waiting time result diagram for the task allocation of the present invention;

[0096] Figure 5 The result diagram of the communication times for the task allocation of the present invention. Specific implementation plan

[0097] In order to explain the technical solution disclosed by the present invention in detail, the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0098] The present invention provides a method for allocating distributed tasks of unmanned aerial vehicles (UAVs) with reduced computation and communication overhead. For ease of understanding, the present invention provides the following additional explanations:

[0099] First of all, before applying this method, it is necessary to consider the compatibility of the drone with the tasks that can be performed. For example, the rescue site requires survey drones, communication drones, lighting drones, etc. The present invention can also be applied to scenarios such as the allocation of edge computing tasks. Furthermore, for several drones, grouping issues and deployment scheduling are considered, including communication collaboration between drones. It is also necessary to consider the target task adaptation and current task conflict issues of the drones in the group, and dynamic optimization, adjustment and task update issues in the work of drones. Based on the processing process of the above problems, the present invention is based on the distributed task allocation algorithm PI-MinAvg and PI-MaxAvg for optimization design, so that it can solve multi-objective problems, improve computing efficiency, and isolate deadlocks. In addition, the number of communications between drones is greatly reduced by introducing technologies such as grouping, estimation and prediction, and communication topology simplification.

[0100] Combination Figure 1 The specific implementation steps and processes of the method of the present invention are as follows:

[0101] (1) Deployment of drone scenarios.

[0102] This embodiment uses a multi-UAV rescue mission as the background to set a task allocation problem; the deployment of UAVs includes initially arranging several UAVs at different locations, or deploying UAVs for performing different tasks at the same location.

[0103] Among them, the UAV system: each UAV is in a different initial position and has different mission execution capabilities; classified by heterogeneous UAVs, each type of UAV has the same flight speed and fuel consumption rate. Different types of UAVs have different flight speeds and fuel consumption rates. The communication between UAVs is determined by the initial communication topology, and the example tests Figure 2 The four network topologies described above. The subsequent algorithm will simplify the initial communication topology to achieve the purpose of reducing communication overhead.

[0104] (2) Construct target task scenarios.

[0105] The entire mission scenario is defined as a set of n heterogeneous UAV systems U = [u1, u2, ..., u n ] and a set of m persons to be rescued (target tasks) T = [t1, t2, ..., t m ];

[0106] Personnel to be rescued: Each person to be rescued is in a different position and has a different mission start and end time. Different rescuers require corresponding types of drones. Define a Boolean matrix H to reflect the compatibility of drone types and rescuer types. When the drone type is compatible with the type of the person to be rescued, h = 1, h∈H, otherwise h = 0.

[0107] (3) Establish a mathematical model corresponding to the problem scenario.

[0108] Before building the model, some variables are defined to facilitate understanding and expression: each drone maintains a task list p when assigning tasks i , its cardinality|p i | indicates drone u i The number of tasks to be executed is allocated, and the bid list c corresponding to the task list is maintained i , then u i The corresponding bidding information can be expressed as c i，k (p i ), winning drone collection z i ;

[0109] Objective function: maximize the number of tasks completed by drones while minimizing the waiting time of survivors:

[0110] J = min C (max N)

[0111] Among them, C represents the waiting time of the survivors, and N represents the number of tasks completed by the drones.

[0112] Constraints:

[0113]

[0114] Among them, N i The meaning is drone u i Maximum number of items that can be accommodated; Indicates drone u i Task Sequence i Specifically, C1 represents the start deadline of the jth task of UAV u. i Can accommodate up to N i tasks; C2 ensures that each task is assigned to at most one UAV, and there will be no task conflict; C3 indicates that the combination of any UAV's task sequence is a subset of the task set T; C4 ensures that UAV u i The executed tasks will not exceed the start deadline of the corresponding tasks; C5 and C6 ensure that the drone can only complete tasks of the type it is adapted to.

[0115] In addition to the task list i and the corresponding bid list c i In addition, each drone also maintains a table of RPI records that remove mission performance impacts. i =[w1, w2, ..., w k , ..., w m ], add task performance impact IPI record table And task deletion record table ψ i, which is used to record the complete mission path p of each UAV before any mission deletion occurs during the mission deletion phase i .

[0116] Among them, removing task performance affects the RPI record table γ i In, w k Represents task t k From its allocated list p i Perform the removal operation in After that, the performance impact of reducing the total cost of task allocation is expressed as The calculation method is:

[0117]

[0118] Where b represents t k In the list p i The position in

[0119] Add task performance impact record table i * middle, Represents task t k Assign to drone u i And its task list p i Performing Add Operation After that, the performance impact of the total cost increase of task allocation is expressed as The calculation method is:

[0120]

[0121] Among them, add operation Indicates that the task t k Add to list a j The lth position of ;

[0122] (4) Preliminary iteration considering performance impact

[0123] The result obtained by executing the PI-MaxAvg algorithm once is used as the preliminary result. The bidding information and task sequence results in this algorithm are used for subsequent drone cluster grouping and network topology simplification.

[0124] The PI-MaxAvg algorithm follows the same algorithmic flow as PI-MinAvg, iterating the task inclusion phase, communication, and conflict resolution phase. However, it uses different definitions for including performance impact and removing performance impact, denoted as IPI-MaxAss and RPI-MaxAss, respectively. If a task can be included in the scheduler and meets the time constraint, its IPI-MaxAss is set to no cost, which is calculated as follows:

[0125]

[0126] in, These missions are compatible with the capabilities of the drone and are not yet included in the p i and RPI-MaxAss is greater than 0.

[0127] RPI-MaxAss is defined as:

[0128]

[0129] The formula shows that unassigned tasks are initially set to have a fixed maximum RPI - MaxAss (a constant defined as U), and assigned tasks are initially set to 0.

[0130] (5) Drone grouping and mission matching

[0131] In the distributed task allocation problem, conflicts will occur between drones that prefer the same task, which means that agents with similar task preferences can be grouped in the same group. In this way, drones in the same group only need to communicate within the group. After the same group of cluster drones reaches a consensus, they only need to arbitrarily select a drone that can communicate with the surrounding cluster groups to achieve the purpose of information consensus of the entire multi-drone system. According to the initial task sequence in step S4, if the task ID corresponding to the lowest RPI value of one drone is the same as the task ID corresponding to the lowest RPI value of another drone, then the two drones are in the same group, and the calculation method is:

[0132]

[0133]

[0134] After the drones are grouped according to their initial bids, the missions also need to be grouped. is the collection of the task lists that were originally assigned to the same group of drones, as well as the tasks that were not assigned to any drone.

[0135] (6) Reduce the number of closed-loop triangles in the communication network to reduce the communication load when solving large networks.

[0136] Each drone will build a necessary network G n to describe the edges (links) it thinks should be kept. i , Indicates a i Can receive drone a j Message sent. Each drone is initialized will be set to all 0 matrices. jCan be with u i When communicating and the mission lists of the two drones conflict, the matrix The specific calculation method is:

[0137]

[0138] Next, each drone receives messages from its neighboring drones about the edges they need to maintain, and then adds these edges to its G n Define this operation as:

[0139]

[0140] Broken-motifs theory eliminates closed-loop triangles to improve communication efficiency. The calculation method is as follows:

[0141] If there is a drone u k , k≠i, G BM (i, k)==1 and G BM (k, j) == 1, then u can be eliminated i with u j Direct communication mode, namely G BM (i, j)==0 and G BM (j,i)==0.

[0142] Final expansion To ensure that you i It can directly or indirectly receive the original communication topology and u i A drone that can communicate normally j The specific process is as follows: If And it satisfies the Broken-motifs theory conditions, then

[0143] (7) Task Update

[0144] Each time a task is added, the UAVs need to update the latest task information. Specifically, each UAV sends a task sequence list p to other UAVs within the simplified communication topology. k 、RPI Record Table γ k and drone information update timestamps k When the drone i Received from drone u k After the message is received, it will be based on γ k and k Two vector judgments are updated.

[0145] When the drone i Receive data from drone u kAfter receiving the three types of information, the following three actions may be taken:

[0146] (1) Update:

[0147] w i，k =w j，k , z i，k =z j，k

[0148] Indicates drone u i Receive all k The news that u k Able to obtain task j.

[0149] (2) Reset:

[0150]

[0151] Indicates drone u i Think u k The information is not reliable, and the existing knowledge about task j will also be cleared.

[0152] (3) Leave:

[0153] w i，j =w i，j , z i，j =z i，j

[0154] Indicates drone u i Receive all k 's message, but still thought that he had obtained task j.

[0155] UAV i With drone u k The consensus rules for task j during information exchange are as follows:

[0156]

[0157]

[0158] (8) Optimize UAV mission sequence.

[0159] Each drone will follow the consensus result, that is, the winning drone set list z of each drone i and task performance impact RPI record table γ i will be consistent, and then the information in the consensus list will be compared with the information in the current list and deleted, and the RPI list will be updated at the same time. i To prevent deadlock, the previous task sequence p will be deleted. i Recorded to i middle.

[0160] After the communication and negotiation operation, each drone will reach a consensus, specifically:

[0161] (8.1) Save the task sequence before deleting it i to i middle.

[0162] (8.2) Determine the drone u i List of pending deletions that conflict with the consensus list d i , which is calculated as follows:

[0163] d i =p i [z i [p i ]≠i]

[0164] (8.3) Remove the tasks in the deletion list one by one. The calculation method is as follows:

[0165]

[0166] Among them, the use of It indicates the updated value of RPI after the removal operation. Each time the removal operation is performed, the one with the greatest impact is removed first, until d i The list is empty or the above formula no longer satisfies the judgment condition of being greater than 0.

[0167] (8.4) If the task sequence |p i | decreases, the deletion operation ends; otherwise, the task deletion record is cleared ψ i .

[0168] (9) Task-related optimization mainly involves sampling various tasks and selecting the optimal schedule without conflict. Each sample obtains a conflict-free schedule by estimating the schedules of other drones related to the conflicting task. Calculations for too many other drones will greatly increase the computing time. Therefore, an appropriate increase in the amount of computing should be used on "competitive" drones that may compete for tasks. The core idea of ​​selecting these competitive drones is task relevance.

[0169] Mission-related drones are drones that currently have mission conflicts. Conflict information can be obtained from the winning messages of other drones. Drone i finds the winner vector z of other drones j in this iteration from the received message j , and then find the scheduling p with UAV i i The calculation method for all UAVs j with mission conflicts is as follows:

[0170] L = {j | z j [p i ]≠z i [pi ], j∈N, j≠i}

[0171] The variable L stores the information related to the drone u i There are conflicting drone IDs.

[0172] In each iteration, UAV i calculates the task scores of related UAVs j and estimates their task selection and the resulting schedule, thus avoiding task conflicts in advance and reducing communication between UAVs.

[0173] UAV i Unknown mission related drone j The location information of the drone is not transmitted in addition to the low traffic advantage of the market-based method. Here, a method based on historical bidding is used to predict the location of the drone. The performance impact function is only related to the location of the drone, the actual location of the task, and the location of the task in the drone task sequence. After consensus, u i Is a known drone j The bidding information of the drone u and the task information it obtains can be obtained through the inverse function of the performance impact function. j Distance p j The distance of the task in is calculated as follows:

[0174]

[0175] in, Represents p j The position information of the task in the sequence, v j Indicates drone u j speed.

[0176] Calculate the drone u i The distance d to task j ij After that, the drone i The position is located at a distance d from the position of task j. ij From the intersection of the three circles, we can see that only three different tasks j are needed to obtain the drone u. i location.

[0177] The process of estimating the scheduling of relevant UAVs is to estimate h, which is mainly based on the number of mission-related UAVs u k The current schedule p k , bid information k 、UAV k Location k , which is expressed as follows:

[0178]

[0179] (10) Location Update

[0180] Each drone traverses all executable tasks and task lists according to the task status and task adaptability conditions. i All addable locations in Calculate the IPI value and determine the feasible location l of the task.

[0181] (10.1) Calculate the IPI value and compare it with the RPI value to determine the task with the greatest impact:

[0182]

[0183] When q>0, if task t k Not S9.1 i Prediction p k(t+1) If you want to perform the task in Add the selected tasks to the list i At l in .

[0184] (10.2) Repeat (10.1) until p i If the list cannot accommodate more tasks or q is less than 0 after IPI update, the task addition ends.

[0185] (10.3) After the task adding phase is over, update and remove the RPI that affects performance.

[0186] Based on the implementation process of the present invention described above, the entire algorithm is divided into two parts: internal and external groups. Once the internal consensus is reached, the groups will reach external consensus. When multiple external consensuses are conducted and no task conflicts are found, the entire algorithm converges and the task allocation ends. Otherwise, continue to execute steps (7), (8), (9) and (10) until the entire algorithm converges.

[0187] Furthermore, combined with Figures 3 to 5 The specific implementation results of the method of the present invention are as follows:

[0188] Figure 2The task allocation quantity index of the algorithm of the present invention is compared with that of other methods, indicating that the algorithm of the present invention outperforms the other two methods in any case. When n is very small, the number of tasks completed by the algorithm of the present invention and PI-MaxAss is similar. The occurrence of this phenomenon can be attributed to the limited number of tasks and drones, resulting in the result of PI-MinAvg being close to the best. The black line segment between the algorithm of the present invention and the PI-MaxAss line represents the difference in the number of task allocations between the two algorithms. The number next to the black line segment represents the percentage of the number of tasks completed by the algorithm of the present invention compared with PI-MaxAss. It can be clearly seen that as n increases, the advantage of the algorithm of the present invention in the number of allocations gradually emerges. This is because as the size of the task increases, the drone has more tasks to choose from. The chance of obtaining a feasible time by reordering unassigned tasks and local task sequences with a large number of tasks is definitely greater than that based on the insertion strategy.

[0189] Figure 3 The average waiting time of tasks in the algorithm of the present invention is compared with other methods, which shows that the algorithm of the present invention can achieve a shorter average waiting time than PI-MaxAss. In addition, it is noted that PI-MinAvg is always the smallest waiting time because it completes the least number of tasks, which is difficult to play a role in actual task allocation requirements.

[0190] Figure 4 The algorithm of the present invention is compared with other methods in terms of the number of communications between drones, which shows that the algorithm of the present invention can achieve fewer communications than the PI algorithm. Because the proposed task removal strategy and task-related optimization method eliminate the possible task conflicts with other drones in advance, the speed of reaching a consensus between drones is faster. In addition, the grouping and network topology simplification methods only require a small number of drones to communicate to achieve the effect of reaching a consensus that would otherwise require a large number of drones to communicate.

Claims

1. A method for allocating distributed tasks of unmanned aerial vehicles with reduced computation and communication overhead, characterized in that: The method comprises the following steps: S1. Construct a UAV task allocation scenario, including deploying several UAVs for performing different tasks at different initial positions, and define a Boolean matrix H to reflect the compatibility between the UAV type and the target task. When the UAV type is compatible with the target task type, h = 1, h∈H, otherwise h = 0; S2. Define the task allocation scenario: a set of n heterogeneous UAV systems U = [u1,u2,…,u n ] and a set of m target tasks T = [t1, t2, …, t m ]; The heterogeneous UAVs are UAVs with the same flight speed and energy consumption, and the communication between the UAVs is determined according to the initial communication topology; S3, based on the task allocation scenario, set each drone to maintain a task list represented by p during task allocation. i , its cardinality|p i | indicates drone u i The number of tasks to be executed is allocated, and the bid list c corresponding to the task list is maintained i , then u i The corresponding bidding information is represented by c j,k (p i ), winning drone collection z i ; The objective function of the mathematical model of task allocation is to maximize the number of tasks N completed by the drone while minimizing the waiting time C for task execution. The expression is as follows: J = minC(maxN) Constraints: C1: C2: C3: C4: C5: C6: Among them, N i The meaning is drone u i The maximum number of Indicates drone u i Task Sequence i The start and deadline of the jth task; Constraint C1 represents the UAV u i Can accommodate up to N i tasks; Constraint C2 ensures that each task is assigned to at most one UAV, and there will be no task conflict; Constraint C3 indicates that any combination of UAV’s task sequences is a subset of the task set T; Constraint C4 ensures that UAV u i The tasks will not be executed beyond the start deadline of the corresponding tasks; Constraints C5 and C6 jointly constrain UAV u i Only tasks of the type that are suitable for it can be completed, where k represents any target task, and constraint C6 represents the relationship between any task k and the current drone u i The adaptability of the tasks that can be performed; Based on the above constraints, in addition to the task list p i and the corresponding bid list c i In addition, each drone also maintains a table of RPI records that remove mission performance impacts. i =[w1,w2,...,w k ,...,w m ], add task performance impact IPI record table And task deletion record table ψ i , task deletion record table ψ i Used to record the complete mission path p of each drone before any mission deletion occurs during the mission deletion phase i ; Among them, removing task performance affects the RPI record table γ i In, w k Represents task t k From its allocated list p i After the removal operation is performed in , the performance impact of reducing the total cost of task allocation is calculated as The removal operation is represented as The calculation method is: Where b represents t k In the list p i The position in Add task performance impact record table i * middle, Represents task t k Assigned to drone u i And its task list p i Performing Add Operation After that, the performance impact of the total cost increase of task allocation is expressed as The calculation method is: Among them, add operation Indicates that the task t k Add to listp i The lth position of ; S4, implement UAV task allocation based on PI-MaxAvg algorithm, and use the result after one execution as the preliminary task sequence, and then use the bidding information and task sequence results in the PI-MaxAvg algorithm to simplify the subsequent UAV cluster grouping and network topology; S5. According to the initial task sequence in step S4, in the same group, the task ID corresponding to the lowest RPI value of the drone is the same as the task ID corresponding to the lowest RPI value of another drone. Any group is represented by G p : The grouping based on the initial task sequence includes grouping the drones according to their initial bids, and determining an adapted task type according to the group; S6, by reducing the number of closed-loop triangles in the communication network to reduce the communication load when solving large networks, specifically, each drone builds a network G n To describe the edges it thinks should be kept, it does this: For drones i , Indicates drone u i Capable of receiving drone u j Message sent when each drone is initialized will be set to all 0 matrices. j Able to work with drones i When communicating, and the mission lists of the two drones conflict, the matrix Will be set to 1; The calculation method is: Next, each drone receives messages from its neighboring drones about the edges they need to maintain, and then adds these edges to its G n In , the operation is defined as: Final expansion To ensure that you i Directly or indirectly receive the original communication topology with the drone u i A drone that can communicate normally j ; S7, each UAV sends a task sequence list p to other UAVs within the simplified communication topology constraints k 、RPI Record Table γ k and drone information update timestamps k ; When the drone i Received from drone u k After the message is received, it will be based on γ k and k Two vector judgments are updated; S8. Each drone compares the information in the consensus list with the information in its current list and deletes it based on the consensus result. At the same time, it updates the RPI list γ i To prevent deadlock, the previous task sequence p will be deleted. i Record to task deletion record tableψ i middle; S9, for the task-related drones, which are drones that currently have task conflicts, obtain the conflict information from the winning messages of other drones; UAV i Find out other drones u in this iteration from the received messages j The winner vector z j , then find out which drone i The scheduling of p i All drones with mission conflicts j , which is calculated as follows: L={j|z j [p i ]≠z i [p i ],j∈N,j≠i} The variable L is used to store the information related to the drone u. i There are conflicting drone IDs; In each iteration, the drone u i Computational related drones j task fractions of the UAVs and estimate their task selection and the resulting schedule, thus avoiding task conflicts in advance and reducing communication between UAVs; For the UAV scheduling process, the estimated h is based on the task-related UAV u k The current schedule p k , bid information k 、UAV k Location k , which is expressed as follows: S10, each UAV traverses all executable tasks and task lists according to the task status and task adaptability conditions i All addable positions in Calculate the IPI value and determine the feasible location l of the task.

2. The method for allocating distributed tasks of unmanned aerial vehicles with reduced computing and communication overheads as claimed in claim 1, characterized in that: In step S4, in the PI-MaxAvg algorithm, the inclusion of performance impact and the removal of performance impact are represented as IPI-MaxAss and RPI-MaxAss respectively. If the task can be included in the scheduler and meets the time limit, its IPI-MaxAss is set to no cost, and its calculation method is as follows: in, These tasks q Compatible with drone features and not yet included in p i and RPI-MaxAss is greater than 0; RPI-MaxAss is defined as: Among them, unassigned tasks are initially set to have a fixed maximum RPI-MaxAss, defined as a constant of U, and assigned tasks are initially set to 0.

3. The method for allocating distributed tasks of unmanned aerial vehicles with reduced computing and communication overheads as claimed in claim 1, characterized in that: Step S6 is to eliminate closed-loop triangles based on Broken-motifs to improve communication efficiency, and its calculation method is as follows: If there is a drone u k ,k≠i,G BM (i,k)==1 and G BM (k,j)==1, then u is eliminated i with u j Direct communication mode, namely G BM (i,j)==0 and G BM (j,i)==0; The final step in simplifying the communication topology is to expand Specific process: if And it satisfies the Broken-motifs theory conditions, then 4. The method for allocating distributed tasks of unmanned aerial vehicles with reduced computing and communication overheads as claimed in claim 1, characterized in that: In step S7, when the drone u i Receive data from drone u k After receiving the three types of information, you can choose the following operations: (1) Update: In i,k =in j,k ,With i,k =from j,k The above formula represents the UAV u i Full reception of drones k news, and believes that drone u k Able to obtain task j; (2) Reset: w i,j ＝0 The above formula represents the UAV u i Think u k The information is unreliable, and the existing knowledge of task j is also cleared; (3) Leave: In i,j =in i,j ,With i,j =from i,j The above formula represents the UAV u i Receive all k 's message, but still thought that he had obtained task j.

5. The method for allocating distributed tasks of unmanned aerial vehicles with reduced computation and communication overheads as claimed in claim 1, characterized in that: The step S8 includes performing the following operations: Determine the drone u i List of pending deletions that conflict with the consensus list d i , the calculation method is as follows: d i =p i [z i [p i ]≠i] To remove the tasks in the deletion list one by one, the calculation method is as follows: in, It indicates the updated value of RPI after the removal operation. Each time the removal operation is performed, the one with the greatest impact is removed first, until the list of items to be deleted d i Is empty or the above formula no longer satisfies the judgment condition of being greater than 0.

6. The distributed task allocation method for reducing computing and communication overhead according to claim 1, characterized in that: Step S9 takes into account the drone u i Unknown mission related drone j In order to maintain the low traffic advantage of the market-based method, the location information of the drone will not be transmitted additionally. A method based on historical bidding is adopted to predict the location of the drone, in which the performance impact function is only related to the location of the drone, the actual location of the task, and the location of the task in the drone task sequence. After consensus, u i Is a known drone j The bidding information and the task information it obtains; The inverse function of the performance impact function is used to obtain the drone u j Distance p j The mathematical calculation method for the distance of the task is as follows: in, Represents p j The position information of the task in the sequence, v j Indicates drone u j speed; The UAV u is calculated from the above formula i The distance d to task j ij After that, the drone i The position is located at a distance d from the position of task j. ij On a circle with radius By determining the intersection of three circles, we can see that only three different tasks j are needed to obtain drone u. i location.

7. The method for allocating distributed tasks of unmanned aerial vehicles with reduced computation and communication overheads as claimed in claim 1, characterized in that: The step S10 further includes: Calculate the IPI value and compare it with the RPI value to determine the task with the greatest impact: When q>0, if task t k Not the drone u in step S9 i Prediction p k(t+1) , then perform the add operation p i ⊕ l t k , add the selected task to the list p i l in the middle; Repeat the above calculation until p i If the list cannot accommodate more tasks or q is less than 0 after IPI update, the task addition ends; After the task addition phase is over, update and remove the RPI that affects performance.

Images