A method for adaptive cluster-based intelligent streaming task allocation in complex adversarial environments

Abstract

This invention discloses an adaptive intelligent streaming task allocation method for unmanned swarms in complex adversarial environments. The method includes: selecting UAV B as the target of the current point-to-point communication; determining whether to generate a new task to be allocated based on environmental information obtained by UAV A, and adding it to UAV A's task list; UAV B generating a streaming task message body; updating UAV A's task list, processing task list, and completed task list based on the received streaming task message body; UAV A submitting a bid for the highest-profit task, forming a temporary allocation result; determining whether a set iteration cycle has been reached; if so, moving the temporary allocation result from the processing task list to UAV A's completed task list; otherwise, returning to continue iteration. This invention improves the adaptability of distributed decision-making for dynamic tasks in heterogeneous unmanned swarms to communication conditions between individual UAVs within the swarm.

Description

Technical Field

[0001] This invention relates to an adaptive cluster intelligent streaming task allocation method for communication in complex adversarial environments, belonging to the field of artificial intelligence technology group collaboration. Background Technology

[0002] The target allocation problem is widely used in unmanned aerial vehicle (UAV) swarms to handle multiple objectives, and distributed decision-making capability is considered one of the manifestations of UAV swarm intelligence, attracting widespread attention. However, the results of distributed UAV swarm decision-making algorithms are easily affected by communication interference, making it highly valuable to reduce or even eliminate the impact of communication on distributed decision-making results.

[0003] Compared to other distributed task methods, market-like mechanisms offer rigorous mathematical derivations, demonstrable computational complexity, clear theoretical explanations, and fewer restrictions on network size and topology. In practical applications, individual drones in the cluster only need to calculate bidding or tender information, resulting in minimal computational load, making them suitable for the limited computing power of distributed unmanned swarm platforms. Particularly in the future military field, where unmanned swarms are large-scale, network topologies are complex, and battlefield environments change rapidly, the aforementioned characteristics of market-like mechanisms can better address unmanned swarm task planning problems in such scenarios. However, current methods require all drones to process a fixed number of tasks simultaneously, which does not meet the requirements for streaming task processing. In real-world environments, tasks are often not generated simultaneously but initiated by each drone in the cluster; therefore, modifying distributed algorithms to support streaming data input has significant application value. Summary of the Invention

[0004] The technical problem solved by this invention is to overcome the shortcomings of the prior art and propose an adaptive cluster intelligent streaming task allocation method for complex adversarial environment communication, which reduces the impact of communication obstruction in adversarial environment on the consistency of cluster task allocation results.

[0005] The technical solution of this invention is: a method for adaptive cluster intelligent streaming task allocation in complex adversarial environments, comprising:

[0006] 1) Based on the current communication history between UAV A and other UAVs in the entire UAV cluster, select UAV B as the target for this point-to-point communication;

[0007] 2) Based on the environmental information obtained by UAV A, determine whether to generate a new task to be assigned, and add it to the task list of UAV A.

[0008] 3) The target drone B generates a streaming task message body based on its stored list of tasks to be assigned, list of tasks in progress, and list of completed tasks, and sends it to the current drone A;

[0009] 4) Update the list of tasks to be assigned, the list of tasks in process, and the list of completed tasks of the current drone A based on the streaming task message body sent by the target drone B;

[0010] 5) Based on the updated task list of drone A, drone A submits a bid for the highest-profit target, forming a temporary allocation result, which is then stored in the task list.

[0011] 6) Determine whether the tasks in the current task list of UAV A have reached the set iteration round. If yes, temporarily assign the results from the task list to the completed task list of UAV A. If not, return to step 1) to continue iterating.

[0012] In step 1), the target drone B for this point-to-point communication is selected based on the communication history between the current drone A and other drones in the entire drone cluster. This includes: using the negative power function method of communication frequency to give the probability that other drones in the entire drone cluster other than the current drone A will be selected as the communication target of the current drone A, and extracting the target drone B according to this probability distribution.

[0013] In step 3), the streaming task message body generated by the target drone B includes the current list of tasks to be assigned, the list of tasks in process, and the list of completed tasks. The list of tasks to be assigned stores the task tag and task start time of each task; the list of tasks in process stores the task tag, task start time, current quote, and quote drone ID of each task.

[0014] The current quote refers to the fact that each task in the task list being processed will be temporarily given a cost for the current task execution by a certain drone.

[0015] The list of completed tasks includes task tags, task start time, current quote, and quote drone ID; the task tags contain information used to calculate the revenue of drone A performing this task.

[0016] In step 4), the unassigned task list, the task-in-process list, and the completed task list of the current drone A are updated, including:

[0017] Update the task list to be assigned: Remove duplicate tasks from the task list to be assigned, set the quoted drone for each task in the task list to the current drone A and set the current quote to 0, and then move the task to the task list to be processed.

[0018] Update the list of tasks in process: Remove duplicate tasks from the list of tasks in process, and merge the quotes in the list of tasks in process for the current drone A and the quotes in the list of tasks in process for the target drone B.

[0019] Update the list of completed tasks: Remove duplicate tasks from the list of completed tasks, and merge the quotes and quote drone IDs in the list of completed tasks.

[0020] When removing duplicate tasks from the task list to be assigned, the task list to be processed, and the task list to be completed, the criterion for determining duplicate tasks is: based on the description in the task tag, determine whether the tasks from the current drone A and the target drone B are the same task.

[0021] The merging of quotes in the current processing task list of drone A and the quotes in the processing task list of target drone B includes: for each task in the processing task list of target drone B, if it also exists in the current drone A, the current quote in the processing task list of the current drone A is updated to the higher quote, and the quote drone ID in the processing task list of the current drone A is updated; if it is not in the processing task list of the current drone A, the task is added to the processing task list of the current drone A, and the quote and quote drone are the same as the data in target drone B.

[0022] The merging of quotes and quote drone IDs in the completed task list includes: for each task in the completed task list of target drone B, if it also exists in the current drone A, the current quote in the completed task list of the current drone A is updated to the higher quote, and the quote drone ID in the completed task list of the current drone A is updated; if it is not in the completed task list of the current drone A, the task is added to the completed task list of the current drone A, and the quote and quote drone are the same as the data in target drone B.

[0023] In step 5), based on the updated task list of drone A, drone A makes a bid for the highest-profit task, forming a temporary allocation result. This includes: drone A calculates the revenue of each task through the task tags in the task list, calculates the net revenue of each task based on the difference between the revenue of each task and the current bid, selects the task with the highest net revenue, bids for it, and modifies the current bid and the bid drone ID for this task.

[0024] The beneficial effects of the present invention are:

[0025] 1. Existing distributed task allocation methods typically only consider allocating a certain number of tasks to a certain number of drones, while this invention adopts a streaming task message body, which accommodates more task information and can support distributed task generation and streaming input.

[0026] 2. Existing distributed task allocation methods adopt a fixed communication assumption. However, based on real-world application scenarios, this invention proposes a communication object selection method that uses probability to select objects with fewer communication cycles, which can greatly increase communication efficiency and avoid many application problems caused by the original method's fixed communication order.

[0027] 3. Existing distributed task allocation methods can be proven to converge, both theoretically and experimentally, on the premise of unmanned swarm communication network connectivity. This requires a stable communication network structure, but in real-world applications, communication networks cannot remain stable. The probabilistic communication method used in this invention relaxes the strict requirements on the communication network structure, making it better suited to real-world application scenarios. Attached Figure Description

[0028] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0029] like Figure 1 As shown, this invention proposes an adaptive intelligent streaming task allocation method for unmanned swarms in complex adversarial environments, the steps of which include: (distributed deployment and operation on each UAV in the unmanned swarm)

[0030] The first step is to use probability P j Select drone j, which can communicate with the current drone A, as the target of this communication:

[0031]

[0032] Where n is the total number of drones that have communicated with the current drone, μj is the number of times the current drone communicates with drone j, μk is the number of times the current drone communicates with drone k, and the extracted drone is called the target drone B.

[0033] The second step is to initialize the allocation results stored in all drones. The actual task constraints determine whether a new task to be assigned has been generated. If so, the newly generated task to be assigned is allocated to the drone that proposed the task, and the drone bids 0 for it.

[0034] The third step is for the selected drone B to generate a streaming task message body from its stored list of tasks to be assigned, list of tasks in progress, and list of completed tasks, and send it to the current drone A.

[0035] The fourth step is to perform a price merging operation on the current drone A and the target drone B. That is, for tasks that exist in both drones' price lists, the merged price is the larger of the two original prices. For tasks that exist only in one of them, they are added to the other drone's price list and the price remains unchanged.

[0036] Fifth, based on the merged quotes, the current drone i makes a bid;

[0037] For the task with the highest profit in the current task list of drone A, increase the price by Δp:

[0038] Δp=p max -p smax +ε

[0039] Where, p max To process the revenue of the task with the highest net profit in the task list, p max To handle the revenue of the second-largest net-revenue task in the task list, ε is a hyperparameter that needs to be adjusted based on its actual value. A larger ε results in faster convergence but may lead to consistency issues. If the number of tasks currently being handled by a drone exceeds its limit, tasks without a bid will be assigned to other drones.

[0040] Step 6: Jump back to step 1 until the method iteration limit is reached.

[0041] This invention discloses an adaptive cluster-based intelligent streaming task allocation method for complex adversarial environments, which involves two processes requiring detailed description. One is an improvement to the communication network. Due to the instability of the communication network structure and the inconsistent availability of task information within each local area network, a priority weight for communication is designed for each UAV. The communication targets and frequency percentages of each UAV are recorded, with higher weights given to communications with lower frequency percentages. This helps to obtain more overall data and make better judgments.

[0042] Suppose that within a certain communication time interval, the communication with node p... i The connected n nodes communicated μ respectively. j If this happens once, then the probability of each communication node being selected the next time a communication target is chosen is:

[0043]

[0044] Furthermore, a detailed explanation is needed regarding the conflicts that arise when merging the two existing communication networks. Observing the auction method from a holistic perspective, it's not difficult to see that at any given moment, it can be determined that any two connected nodes p... i ∈N(i),p j Each ∈N(j) can represent a uniquely restored communication node between two communication networks N(i) and N(j) that were once broken into two parts. Let R(i) and R(j) represent the original allocation schemes that the two networks reached a consensus. Then, in p i ,p jIf the unified new allocation result R(i)′=R(j)′ after communication can be generated using the original two allocation schemes, the distributed operation of the method can be completed. The method aims to construct a fluid distributed task allocation method, in which new tasks are continuously generated and added to the allocation list, and allocation results are continuously output to the execution unit for execution, rather than the naive distributed allocation method that distributes tasks based on all known weight information and then ensures global consistency through a consistency method, and obtains the processing result of that time segment after a specific deadline.

[0045] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. A method for adaptive cluster intelligent streaming task allocation in complex adversarial environments, characterized in that... include: 1) Based on the current communication history between UAV A and other UAVs in the entire UAV cluster, select UAV B as the target for this point-to-point communication; 2) Based on the environmental information obtained by UAV A, determine whether to generate a new task to be assigned, and add it to the task list of UAV A. 3) The target drone B generates a streaming task message body based on its stored list of tasks to be assigned, list of tasks in progress, and list of completed tasks, and sends it to the current drone A; 4) Update the list of tasks to be assigned, the list of tasks in process, and the list of completed tasks of the current drone A based on the streaming task message body sent by the target drone B; 5) Based on the updated task list of drone A, drone A submits a bid for the highest-profit target, forming a temporary allocation result, which is then stored in the task list. 6) Determine whether the tasks in the current task list of drone A have reached the set iteration round. If so, temporarily assign the results from the task list to the completed task list of drone A. If not, return to step 1) and continue iterating; In step 1), the target drone B for this point-to-point communication is selected based on the communication history between the current drone A and other drones in the entire drone cluster. This includes: using the negative power function method of communication frequency to give the probability that other drones in the entire drone cluster other than the current drone A will be selected as the communication target of the current drone A, and extracting the target drone B according to this probability distribution. With probability P j Select drone j, which can communicate with the current drone A, as the target of this communication: Where n is the total number of drones that have communicated with the current drone, and μ j μ is the number of times the drone communicates with drone j. k This represents the number of communications between the current drone and drone k. The extracted drone is called the target drone B.

2. The method for adaptive cluster intelligent streaming task allocation in complex adversarial environments according to claim 1, characterized in that, In step 3), the streaming task message body generated by the target drone B includes the current list of tasks to be assigned, the list of tasks in process, and the list of completed tasks. The list of tasks to be assigned stores the task tag and task start time of each task; the list of tasks in process stores the task tag, task start time, current quote, and quote drone ID of each task.

3. The method for adaptive cluster intelligent streaming task allocation in complex adversarial environments according to claim 2, characterized in that, The current quote refers to the fact that each task in the task list being processed will be temporarily given a cost for the current task execution by a certain drone.

4. The method for adaptive cluster intelligent streaming task allocation in complex adversarial environments according to claim 2, characterized in that, The list of completed tasks includes task tags, task start time, current quote, and quote drone ID; the task tags contain information used to calculate the revenue of drone A performing this task.

5. The method for adaptive cluster intelligent streaming task allocation in complex adversarial environments according to claim 4, characterized in that, In step 4), the unassigned task list, the task-in-process list, and the completed task list of the current drone A are updated, including: Update the task list to be assigned: Remove duplicate tasks from the task list to be assigned, set the quoted drone for each task in the task list to the current drone A and set the current quote to 0, and then move the task to the task list to be processed. Update the list of tasks in process: Remove duplicate tasks from the list of tasks in process, and merge the quotes in the list of tasks in process for the current drone A and the quotes in the list of tasks in process for the target drone B. Update the list of completed tasks: Remove duplicate tasks from the list of completed tasks, and merge the quotes and quote drone IDs in the list of completed tasks.

6. The method for adaptive cluster intelligent streaming task allocation in complex adversarial environments according to claim 5, characterized in that, When removing duplicate tasks from the task list to be assigned, the task list to be processed, and the task list to be completed, the criterion for determining duplicate tasks is: based on the description in the task tag, determine whether the tasks from the current drone A and the target drone B are the same task.

7. The method for adaptive cluster intelligent streaming task allocation in complex adversarial environments according to claim 5, characterized in that, The merging of quotes in the current processing task list of drone A and the quotes in the processing task list of target drone B includes: for each task in the processing task list of target drone B, if it also exists in the current drone A, the current quote in the processing task list of the current drone A is updated to the higher quote, and the quote drone ID in the processing task list of the current drone A is updated; if it is not in the processing task list of the current drone A, the task is added to the processing task list of the current drone A, and the quote and quote drone are the same as the data in target drone B.

8. The method for adaptive cluster intelligent streaming task allocation in complex adversarial environments according to claim 5, characterized in that, The merging of quotes and quote drone IDs in the completed task list includes: for each task in the completed task list of target drone B, if it also exists in the current drone A, the current quote in the completed task list of the current drone A is updated to the higher quote, and the quote drone ID in the completed task list of the current drone A is updated; if it is not in the completed task list of the current drone A, the task is added to the completed task list of the current drone A, and the quote and quote drone are the same as the data in target drone B.

9. The method for adaptive cluster intelligent streaming task allocation in complex adversarial environments according to claim 1, characterized in that, In step 5), based on the updated task list of drone A, drone A submits a bid for the highest-profit task, forming a temporary allocation result, including: Drone A calculates the revenue of each task by processing the task tags in the task list. It then calculates the net revenue of each task based on the difference between the revenue of each task and the current bid. Drone A selects the task with the highest net revenue, bids for it, and modifies the current bid and the bidder's ID for that task.

Images