Multi-unmanned aerial vehicle subsystem real-time integrated control method and system based on edge computing

Abstract

The application relates to the technical field of real-time control of unmanned aerial vehicle systems, and discloses a multi-unmanned aerial vehicle subsystem real-time integrated control method and system based on edge computing, which comprises the following steps: according to collected real-time state data of unmanned aerial vehicles, a scheduling node is screened out, the unmanned aerial vehicles are dynamically distributed into corresponding scheduling areas of the scheduling node according to communication connection states among the unmanned aerial vehicles, the unmanned aerial vehicles are taken as task execution nodes, and an unmanned aerial vehicle scheduling network is constructed; in response to a task instruction, the task is parsed into different types of subtasks, corresponding node units are matched, and a first priority and a second priority are respectively calculated; the unmanned aerial vehicle nodes in the node units are dynamically scheduled, a first control strategy is obtained; priorities among conflict nodes are analyzed, the first control strategy is optimized, and a second control strategy is obtained; and the application can avoid single-point failure and improve the safety and efficiency of a control process of an unmanned aerial vehicle system through task matching and dynamic unmanned aerial vehicle scheduling.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of real-time control of unmanned aerial vehicle systems, and more particularly to a multi-unmanned aerial vehicle subsystem real-time integrated control method and system based on edge computing. BACKGROUND

[0002] Multi-unmanned aerial vehicle cooperative operation can improve task execution efficiency and expand coverage. Traditional multi-unmanned aerial vehicle control methods mostly use centralized control architecture, which relies on ground control centers or cloud servers for task allocation and scheduling. This centralized method has the problems of single point failure risk, high communication delay, poor scalability, etc. In a dynamically changing environment, it is difficult to achieve real-time response and efficient cooperation, which can easily lead to task conflicts, resource waste and insufficient priority management.

[0003] The existing technology has the following problems: relying on a central server or cloud for control, which can easily lead to single point failure, and once the central node fails, the overall function of the control system will be damaged; based on static task allocation and fixed scheduling strategy, it cannot be dynamically adjusted according to the unmanned aerial vehicle state, environmental changes or task requirements, resulting in low resource utilization efficiency and inability to respond to unexpected tasks; in multiple task scenarios, it is difficult to effectively evaluate and compare task priorities, which can lead to low-priority tasks blocking high-priority tasks, affecting the execution process of the task; to solve at least one of the above problems, the present application proposes a multi-unmanned aerial vehicle subsystem real-time integrated control method and system based on edge computing. SUMMARY

[0004] In view of the deficiencies of the prior art, the purpose of the present application is to provide a multi-unmanned aerial vehicle subsystem real-time integrated control method and system based on edge computing, which can effectively solve the problems in the background art. The specific technical solution of the present application is as follows:

[0005] The multi-unmanned aerial vehicle subsystem real-time integrated control method based on edge computing comprises:

[0006] According to the collected real-time state data of the unmanned aerial vehicle, the real-time capability of the unmanned aerial vehicle node is analyzed, the scheduling node is selected, and the unmanned aerial vehicle is dynamically allocated to the corresponding scheduling node corresponding to the scheduling area according to the communication connection state between the unmanned aerial vehicles, as a task execution node, to construct an unmanned aerial vehicle scheduling network;

[0007] In response to the task instruction, the task is parsed into different types of subtasks, and through edge computing, each subtask is matched with a corresponding node unit in the unmanned aerial vehicle scheduling network, and each node unit calculates the first priority of the subtask and the second priority of the currently executing task;

[0008] The first priority and the second priority are compared, and the unmanned aerial vehicle nodes in the node unit are dynamically scheduled to obtain a first control strategy;

[0009] The node conflict between different UAV nodes in the same node unit and the node conflict between different node units are analyzed respectively, the priority between the conflict nodes is analyzed, the first control strategy is optimized, and a second control strategy is obtained.

[0010] Specifically, the real-time state data of the UAVs is collected, the real-time capability of the UAV nodes is analyzed, the scheduling nodes are screened out, the UAVs are dynamically allocated to the scheduling area corresponding to the scheduling nodes according to the communication connection state between the UAVs, as task execution nodes, and a UAV scheduling network is constructed, including:

[0011] The real-time capability of the UAV nodes is analyzed according to the collected real-time state data of the UAVs, and a real-time capability value is obtained;

[0012] The UAV nodes with a real-time capability value greater than a preset capability threshold are screened out, and a scheduling node set is constructed;

[0013] For each scheduling node, the communication connection state between the UAVs is analyzed, and the UAVs are dynamically allocated to the scheduling area corresponding to the scheduling node, as task execution nodes;

[0014] The UAV scheduling network is constructed in combination with the scheduling nodes and the corresponding task execution nodes.

[0015] Specifically, for each scheduling node, the communication connection state between the UAVs is analyzed, and the UAVs are dynamically allocated to the scheduling area corresponding to the scheduling node, as task execution nodes, including:

[0016] The communication connection state between the UAVs is analyzed, and a communication state value is calculated;

[0017] Each scheduling node is taken as a starting point, and the UAV nodes with a communication state value greater than a preset communication threshold are gradually screened out as a corresponding first node set;

[0018] For the coincident nodes in the candidate node set of different scheduling nodes, the scheduling node with the highest communication state value with the coincident node is selected and allocated to the corresponding scheduling area, the first node set is updated, and a second node set is obtained;

[0019] The area formed by the second node set corresponding to each scheduling node is taken as a scheduling area, and each second node is taken as a task execution node.

[0020] Specifically, the task is parsed into different types of sub-tasks in response to the task instruction, and each sub-task is matched with a corresponding node unit in the UAV scheduling network through edge computing, each node unit calculates a first priority of the sub-task and a second priority of the currently executing task, including:

[0021] In response to the task instruction, the task is parsed into different types of sub-tasks by analyzing the task requirements;

[0022] Each sub-task is matched with a corresponding node unit in the UAV scheduling network through edge computing;

[0023] Each node unit calculates a first priority of the sub-task and a second priority of the currently executing task through a preset priority analysis model.

[0024] Specifically, the each sub-task is matched with a corresponding node unit in the UAV scheduling network through edge computing, including:

[0025] According to the real-time state data of each UAV node in the UAV scheduling network, a state feature is extracted through a preset feature extraction model to construct a node state vector;

[0026] Each sub-task is parsed for requirements through a preset task parsing model to construct a requirement vector;

[0027] The similarity between the node state vector and the requirement vector is calculated, and nodes with a similarity greater than a preset similarity threshold are selected as a first matching node set;

[0028] According to the first matching node set, the resource consumption and UAV scheduling of the sub-task execution process are simulated, and the first matching node set is optimized to obtain a corresponding node unit.

[0029] Specifically, the first priority and the second priority are compared, and the UAV nodes in the node unit are dynamically scheduled to obtain a first control strategy, including:

[0030] The difference between the first priority and the second priority is calculated, and a trigger threshold is set according to the load of the UAV in the node unit;

[0031] For the case where the difference is greater than the trigger threshold, a sub-task node set is selected from the node unit to execute the sub-task, and the remaining nodes continue to execute the current task to obtain the first control strategy.

[0032] Specifically, the node conflict between different UAV nodes in the same node unit and the node conflict between different node units are analyzed, the priority between the conflict nodes is analyzed, and the first control strategy is optimized to obtain a second control strategy, including:

[0033] analyze the node conflict situation between different UAV nodes in the same node unit and the node conflict situation between different node units, determine the conflict influence range, analyze the priority between the conflict nodes, and determine the conflict coordination strategy;

[0034] According to the conflict coordination strategy, the first control strategy is optimized to obtain a second control strategy.

[0035] Specifically, the analysis of the node conflict situation between different UAV nodes in the same node unit and the node conflict situation between different node units, the determination of the conflict influence range, the analysis of the priority between the conflict nodes, and the determination of the conflict coordination strategy include:

[0036] The node conflict situation between different UAV nodes in the same node unit and the node conflict situation between different node units are analyzed, and the conflict nodes with spatial overlap and time overlap are screened out, the influence range of the conflict nodes is analyzed, and the conflict area is obtained;

[0037] In the conflict area, the task failure probability caused by the failure of each conflict node is analyzed, and the conflict priority of the conflict node is determined;

[0038] For the internal conflict of the node unit, the running order of the conflict nodes is coordinated according to the conflict priority from high to low, and for the conflict between different node units, the influence of the simulation coordination process of the scheduling nodes corresponding to different node units on the task execution performance is analyzed, the coordination mode with the optimal performance is screened out, and the conflict coordination strategy is obtained.

[0039] Specifically, the first control strategy is optimized according to the conflict coordination strategy to obtain a second control strategy, which includes:

[0040] According to the conflict coordination strategy, resource allocation is performed on each UAV to obtain a resource allocation mode;

[0041] The first control strategy is optimized according to the resource allocation mode to obtain a second control strategy.

[0042] The multi-UAV subsystem real-time integrated control system based on edge computing is used to realize the multi-UAV subsystem real-time integrated control method based on edge computing, which includes:

[0043] The UAV scheduling network construction module analyzes the real-time capability of the UAV nodes according to the collected real-time state data of the UAVs, screens out scheduling nodes, dynamically allocates the UAVs to the scheduling area corresponding to the scheduling nodes according to the communication connection state between the UAVs, and constructs the UAV scheduling network as a task execution node.

[0044] The task matching module parses the task into different types of subtasks in response to the task instruction, matches a corresponding node unit for each subtask in the UAV scheduling network through edge computing, and calculates a first priority of the subtask and a second priority of a currently executing task by each node unit respectively;

[0045] The control strategy making module dynamically schedules the UAV nodes in the node unit by comparing the first priority and the second priority, and obtains a first control strategy;

[0046] The control strategy optimization module analyzes the node conflict situation between different UAV nodes in the same node unit and the node conflict situation between different node units, analyzes the priority between the conflict nodes, optimizes the first control strategy, and obtains a second control strategy.

[0047] The application has the following beneficial effects: by collecting real-time state data of the UAV, analyzing node capability and screening scheduling nodes, combining with the communication connection state to dynamically allocate task execution nodes, an efficient UAV scheduling network is constructed; in response to a task instruction, a task is parsed into subtasks, and an optimal node unit is matched for each subtask in the UAV scheduling network through edge computing; a first priority of the subtask and a second priority of a current task are calculated by each node unit respectively, the UAV nodes are dynamically scheduled by comparing the priorities and setting a trigger threshold, and high-priority tasks are ensured to be executed in time; the node conflict situation inside the same node unit and between different node units is analyzed respectively, the conflict influence range and the priority are determined, a conflict coordination strategy is made, the control strategy is optimized, and the conflict risk is reduced. Through task matching and dynamic UAV scheduling, single-point failure can be avoided, the continuous execution of tasks can be ensured when part of the nodes fail, the resource utilization rate of the UAV is optimized, the task conflict and collision risk is reduced, and the safety and efficiency of the UAV system control process are improved. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 A work flow diagram of the edge-computing-based real-time integrated control method of the multi-UAV subsystem in Embodiment 1 of the application;

[0049] Figure 2 A schematic diagram of the UAV scheduling network in Embodiment 1 of the application;

[0050] Figure 3 A work flow diagram of matching a corresponding node unit for each subtask in Embodiment 1 of the application;

[0051] Figure 4 A structural schematic diagram of the edge-computing-based real-time integrated control system of the multi-UAV subsystem in Embodiment 1 of the application. DETAILED DESCRIPTION

[0052] The application will be further described below in detail with reference to the drawings and embodiments.

[0053] In the embodiments of the present application, the words "exemplary" or "for example" are used to mean serving as an example, instance, or illustration. Any embodiment or design presented as "exemplary" or "for example" in the embodiments of the present application is not necessarily to be construed as preferred or advantageous over other embodiments or designs. Rather, use of the words "exemplary" or "for example" is intended to present concepts in a concrete manner.

[0054] Hereinafter, the terms "first", "second", and the like are used in a generic sense only and are not to be construed as indicating or implying relative importance or a quantity of the indicated technical features. Thus, a feature defined with "first", "second" can include one or more of the features explicitly or implicitly. In the description of the embodiments of the present application, the meaning of "a plurality of" is two or more, unless otherwise specified.

[0055] Embodiment 1:

[0056] Reference Figure 1 As shown in the embodiments of the present application, the specific embodiments of the multi-unmanned aerial vehicle subsystem real-time integrated control method based on edge computing include:

[0057] S101, according to the collected real-time state data of the unmanned aerial vehicle, analyzing the real-time capability of the unmanned aerial vehicle node, screening out the scheduling node, combining the communication connection state between the unmanned aerial vehicles to dynamically distribute the unmanned aerial vehicles into the corresponding scheduling node corresponding to the scheduling area, as a task execution node, constructing an unmanned aerial vehicle scheduling network;

[0058] S102, in response to the task instruction, parsing the task into different types of subtasks, matching the corresponding node unit for each subtask in the unmanned aerial vehicle scheduling network through edge computing, and each node unit respectively calculating the first priority of the subtask and the second priority of the currently executing task;

[0059] S103, comparing the first priority and the second priority, dynamically scheduling the unmanned aerial vehicle nodes in the node unit to obtain a first control strategy;

[0060] S104, respectively analyzing the node conflict between different unmanned aerial vehicle nodes in the same node unit and the node conflict between different node units, analyzing the priority between the conflict nodes, and optimizing the first control strategy to obtain a second control strategy.

[0061] Traditional multi-UAV control systems often employ a centralized architecture, relying on a ground control center or cloud server for global decision-making. This architecture carries the risk of single points of failure, and the high latency caused by long-distance data transmission cannot meet real-time control requirements. It uses a static task allocation strategy, lacking the ability to dynamically adjust based on UAV real-time status, network conditions, and task priorities, resulting in poor system flexibility and low resource utilization. In complex mission environments, spatiotemporal conflicts can easily occur between multiple UAVs, and traditional conflict resolution methods are often slow to react or computationally complex, making it difficult to guarantee system safety and efficiency.

[0062] In this embodiment, based on the collected real-time status data of the UAVs, including but not limited to remaining battery power, computing load, geographical location, and flight stability, the real-time capabilities of the UAV nodes are analyzed and real-time capability values ​​are calculated. Scheduling nodes are then selected, and UAVs are dynamically assigned to the corresponding scheduling areas of these nodes, serving as task execution nodes, based on the communication connection status between UAVs, thus constructing a UAV scheduling network. By selecting scheduling nodes, the control architecture is transformed from centralized to distributed, effectively eliminating the single point of failure risk associated with a single control center. Dynamically electing scheduling nodes through real-time capability assessment ensures the reliability of the control core. Dynamic area division based on communication connection status guarantees the stability of communication links within each scheduling area, improving the stability and effectiveness of the control process.

[0063] Specifically, in response to task instructions, the task is parsed into different types of subtasks. Edge computing is used to analyze the matching degree between drone nodes and corresponding subtasks. In the drone scheduling network, a corresponding node unit is matched for each subtask. Each node unit calculates the first priority of the subtask and the second priority of the currently executing task. Through task parsing and node matching, real-time configuration between task requirements and drone resource capabilities can be achieved, improving the professionalism and efficiency of task execution. Distributed matching calculations through edge computing reduce the computational pressure on individual nodes and improve node matching speed. By analyzing task priorities, accurate task value evaluation criteria are provided for the drone scheduling and control process. Dynamic scheduling of drone resources is possible when resources are limited, improving drone resource utilization and task execution efficiency.

[0064] Specifically, by comparing the first and second priorities, the difference between the first priority of the new subtask and the second priority of the current task is calculated to reflect the relative urgency of the new task. The load of the UAVs in the node unit is analyzed, and the trigger threshold is dynamically adjusted according to the load. The UAV nodes in the node unit are dynamically scheduled by combining the difference and the trigger threshold to obtain the first control strategy. By setting a dynamic trigger threshold, it can be ensured that high-priority tasks can interrupt low-priority tasks in time to obtain resources, while avoiding frequent and unnecessary task switching caused by small differences in priority or excessive system load. This improves the stability of the system and enhances the task throughput and ability to respond to emergencies of the UAV control system.

[0065] Specifically, the system analyzes node conflicts between different UAV nodes within the same node unit and between different node units. It identifies conflicting nodes with overlapping trajectories in space and time, determines the conflict regions corresponding to their impact areas, analyzes the mission failure probability caused by the failure of each conflicting node within these regions, determines the priority of conflicting nodes, and optimizes the first control strategy to obtain a second control strategy. Through intra- and inter-unit conflict analysis, collision risks and resource competition can be proactively identified and eliminated, improving the safety of the UAV swarm operation. By analyzing mission failure probabilities to determine conflict priorities, the system prioritizes ensuring that the most critical tasks are not affected during coordination. The resulting second control strategy, while meeting task priority requirements, improves both system safety and execution efficiency.

[0066] This application constructs an efficient UAV scheduling network by collecting real-time UAV status data, analyzing node capabilities, and selecting scheduling nodes. It dynamically allocates task execution nodes based on communication connection status. Responding to task commands, the network parses tasks into subtasks and matches optimal node units for each subtask within the UAV scheduling network using edge computing. Each node unit calculates the first priority of the subtask and the second priority of the current task. By comparing priorities and setting trigger thresholds, UAV nodes are dynamically scheduled to ensure timely execution of high-priority tasks. The application analyzes node conflicts within the same node unit and between different node units to determine the scope and priority of conflict impact, formulate conflict coordination strategies, optimize control strategies, and reduce conflict risks. Through task matching and dynamic UAV scheduling, single points of failure can be avoided, ensuring continuous task execution even when some nodes fail, optimizing UAV resource utilization, reducing task conflict and collision risks, and improving the safety and efficiency of the UAV system control process.

[0067] Furthermore, based on the collected real-time status data of the drones, the real-time capabilities of the drone nodes are analyzed, scheduling nodes are selected, and drones are dynamically allocated to the corresponding scheduling areas of the scheduling nodes as task execution nodes, based on the communication connection status between drones, thus constructing a drone scheduling network, including:

[0068] S201. Based on the collected real-time status data of the UAV, analyze the real-time capability of the UAV node through a preset status analysis model to obtain the real-time capability value.

[0069] S202. Select drone nodes whose real-time capability values ​​are greater than the preset capability threshold and construct a scheduling node set;

[0070] S203. For each scheduling node, analyze the communication connection status between UAVs and dynamically allocate the UAVs to the scheduling area corresponding to the scheduling node as task execution nodes.

[0071] S204. Combine scheduling nodes and corresponding task execution nodes to construct a UAV scheduling network.

[0072] In this embodiment, based on the collected real-time status data of the UAV, including but not limited to remaining battery power, CPU and memory utilization, GPS signal strength and attitude angle data, and the current load status of the communication module, the real-time capability of the UAV node is analyzed through a preset status analysis model. The status analysis model includes, but is not limited to, a fully connected feedforward neural network model pre-trained using a large amount of historical UAV status data. The number of nodes in the model's input layer is equal to the dimension of the preprocessed feature vector. The output layer consists of a single node, using the Sigmoid activation function, outputting a scalar value between 0 and 1, representing the real-time capability value. One or two hidden layers may be included, using the ReLU activation function. The model's training dataset comes from historical flight logs and manually labeled data.

[0073] Specifically, a large number of multi-dimensional time-series state data fragments of drone nodes during historical flights are collected as samples. Performance analysis is performed by domain experts or based on post-flight tasks. For each data fragment, a comprehensive capability score is manually assigned to the drone node at the corresponding time point, exemplarily ranging from 0 to 1, with 1 representing the most suitable scheduling node. The score comprehensively considers the node's reliability, remaining mission endurance potential, and computational margin at that time. Supervised learning is used for training, with historical state data as input and the score as the target label. The training objective is to minimize the gap between the model's predicted values ​​and the labeled scores. The loss function is mean squared error, and backpropagation and the Adam optimizer are used for parameter updates. Training hyperparameters include: a set learning rate, initially set to 0.001, which can be decayed using a learning rate decay strategy, performing a sufficient number of training epochs until the loss on the validation set no longer decreases significantly. After training, the model's weight parameters are fixed as a preset portion and deployed in the system. The real-time acquired and preprocessed state feature vectors are input into a pre-trained fully connected feedforward neural network model. The output value obtained by forward propagation is the real-time capability value of the node. Nodes with strong computing power, sufficient battery power, and stable flight will obtain higher real-time capability values. By calculating the real-time capability values ​​of UAV nodes, the UAV state information is quantified, providing an accurate data foundation for selecting and scheduling nodes, improving the reliability of the selected scheduling nodes, and enhancing the stability and efficiency of the distributed control network.

[0074] Furthermore, drone nodes with real-time capability values ​​greater than preset capability thresholds are selected. Capability thresholds are set based on the accuracy requirements of drone control and task execution processes, reflecting the system's expectation of the minimum capability required by the scheduling nodes. The selected drone nodes are then used as a set of scheduling nodes. By setting capability thresholds, scheduling nodes can be quickly selected, improving the reliability of node scheduling functions while ensuring that the scheduling nodes have a basic capability baseline. Dynamically setting thresholds can enhance the system's flexibility in adapting to different task requirements and drone cluster states.

[0075] Specifically, for each scheduling node, the communication connection status between UAVs is analyzed, and the UAVs are dynamically assigned to the scheduling area corresponding to the scheduling node as task execution nodes. Through communication quality analysis and dynamic node allocation, a stable and low-latency communication link can be ensured between each scheduling node and the corresponding task execution node. This provides reliable communication guarantees for real-time command issuance, status monitoring, and collaborative control within the local area, avoiding the risk of edge nodes losing control due to poor connection quality and improving the reliability of the control system in complex communication environments.

[0076] like Figure 2As shown, a UAV scheduling network is constructed by combining scheduling nodes and corresponding task execution nodes and establishing communication connections between each scheduling node and its corresponding task execution node. By building this UAV scheduling network, a well-structured and clearly defined distributed control architecture is created, eliminating the single point of failure risk of centralized control. Dynamically allocating task execution nodes reduces system communication latency and improves the efficiency and reliability of the UAV system's operation.

[0077] Furthermore, for each scheduling node, the communication connection status between UAVs is analyzed, and the UAVs are dynamically assigned to the scheduling area corresponding to the scheduling node as task execution nodes, including:

[0078] S301. Analyze the communication connection status between drones and calculate the communication status value;

[0079] S302. Starting from each scheduling node, gradually filter out the drone nodes whose communication status value is greater than the preset communication threshold, and use them as the corresponding first node set.

[0080] S303. For overlapping nodes in the candidate node sets of different scheduling nodes, select the scheduling node with the highest communication status value with the overlapping node, assign it to the corresponding scheduling area, update the first node set, and obtain the second node set.

[0081] S304. The region consisting of the set of second nodes corresponding to each scheduling node shall be the scheduling region, and each second node shall be the task execution node.

[0082] In this embodiment, communication link parameters between relevant UAV nodes in the network are continuously measured, including but not limited to received signal strength indication, signal-to-noise ratio, and data packet round-trip time. The impact of each communication link parameter on the communication state is analyzed, and corresponding weights are assigned. The normalized communication link parameters are then weighted and summed according to their weights to obtain the communication state value. By quantifying the communication state and calculating the corresponding communication state value, accurate data references are provided for the region division and node allocation process, avoiding the bias and inaccuracy of judgments based on a single indicator, and improving the communication connectivity of the UAV scheduling network.

[0083] Specifically, starting with each scheduling node, a communication threshold is set based on the system's communication and control accuracy requirements. This threshold reflects the minimum communication quality required to ensure reliable and timely transmission of control commands. Starting with all adjacent nodes directly communicating with the scheduling node, nodes with communication status values ​​greater than the preset threshold are added to the first node set. Using these newly added nodes as the starting point, the analysis continues with their connected adjacent nodes, adding nodes with communication status values ​​greater than the preset threshold to the first node set. This process is iterated until no more nodes meeting the criteria can be selected, resulting in the first node set for each scheduling node. By analyzing communication quality, it is ensured that each scheduling node maintains an effective control link with all UAVs within its first node set, improving communication quality and reliability.

[0084] Specifically, nodes that exist in the first node set of multiple scheduling nodes are identified as overlapping nodes. Overlapping nodes maintain qualified communication links with multiple scheduling nodes. By comparing the communication status values ​​between the overlapping node and all scheduling nodes corresponding to its first node set, the scheduling node with the highest communication status value with the overlapping node is selected, and the overlapping node is assigned to this scheduling node. This node is then removed from the first node set of other scheduling nodes. The first node set of each affected scheduling node is updated to eliminate overlapping areas, resulting in the second node set.

[0085] It should be noted that by coordinating the competition among overlapping nodes, the problems of multiple command and resource conflicts among drone nodes can be avoided. By selecting the strongest connection node, each task execution node can be assigned to the scheduling node with the best communication link quality, which optimizes the network communication efficiency, reduces the transmission delay and packet loss risk of control commands, ensures the relative balance of network load among various scheduling nodes, and improves the stability and performance of the drone collaborative control system.

[0086] Specifically, the region consisting of the set of second nodes corresponding to each scheduling node is designated as the scheduling region. The scheduling node is responsible for the management, computation, and coordination of UAV nodes. Each UAV node in the set of second nodes acts as a task execution node, receiving instructions from its assigned scheduling node and efficiently and accurately executing the specific sub-tasks assigned, such as flight, reconnaissance, or operations. By defining the task roles and scheduling regions of UAV nodes, a distributed UAV scheduling network can be established. Dividing the scheduling regions allows for the decomposition of global control tasks, which can then be processed in parallel by each local region, improving the system's scalability and robustness. The master-slave and collaborative relationship between scheduling nodes and task execution nodes provides structural support for task decomposition, dynamic scheduling, and conflict coordination.

[0087] Furthermore, in response to task instructions, the task is parsed into different types of subtasks. Each subtask is matched with a corresponding node unit in the UAV scheduling network via edge computing. Each node unit calculates the first priority of the subtask and the second priority of the currently executing task, including:

[0088] S401. In response to task instructions, analyze task requirements and parse the task into different types of subtasks;

[0089] S402. Match corresponding node units for each subtask in the UAV scheduling network through edge computing;

[0090] S403. Using a preset priority analysis model, each node unit calculates the first priority of the subtask and the second priority of the currently executing task.

[0091] In this embodiment, in response to task instructions, the task requirements are analyzed and parsed into different types of subtasks. The task instructions are then parsed using a requirement parsing model, which includes, but is not limited to, a neural network model based on a sequence-to-sequence architecture. The model's input is the vectorized representation of the task instruction text. The output is a sequence where each element represents a standardized description code for a subtask. These codes correspond to a predefined library of subtask types, such as high-definition imaging reconnaissance, infrared heat source monitoring, fixed-point material hovering delivery, wide-area video stream monitoring, and multispectral scanning. The training dataset is constructed using historical task records and manual annotation. Specifically, a large number of historically successfully executed complex task cases are collected, each containing the original task instruction text. Domain experts annotate these cases as a series of standard subtask types based on the actual decomposition scheme of the task execution. The model is trained using supervised learning. The cross-entropy loss function is used to calculate the difference between the predicted subtask sequence and the manually annotated subsequence. Training uses a teacher-mandated strategy and the Adam optimizer.

[0092] Key training parameters include: setting an initial learning rate, using gradient clipping to prevent gradient explosion, and using early stopping on the validation set to determine the number of training epochs until the model can stably output reasonable subtask sequences on unseen task instruction text. The standardized subtask type sequence output by the model will be instantiated into specific subtask objects. Each subtask object inherits and binds specific parameters from the original instruction. For example, the high-definition imaging reconnaissance subtask will bind the eastern warehouse as the target area and inherit constraints such as the high shooting frequency required by the fire monitoring task. The model outputs a complete list of subtasks of different types. By parsing task instructions into multiple subtasks, task instructions can be transformed into clearly defined basic operational units that can be directly understood and executed by the UAV system. This provides accurate data support for UAV resource matching and scheduling processes tailored to the characteristics of each subtask, avoiding resource misallocation and execution deviations caused by unclear task definitions, and improving the effectiveness and efficiency of task execution.

[0093] Specifically, edge computing is used to match corresponding node units for each subtask in the UAV scheduling network; this enables accurate and efficient matching between subtasks and task execution nodes, which can improve the effectiveness of task allocation and task execution quality. Parallel processing by edge nodes can reduce the latency of task matching, ensuring that the matched node units can complete the subtasks efficiently and collaboratively, thereby improving the reliability and effectiveness of the task execution process.

[0094] Specifically, after a subtask is matched to a corresponding node unit, the scheduling node in that unit calculates the first priority of the subtask and the second priority of the currently executing task using a pre-defined priority analysis model. This priority analysis model includes, but is not limited to, a gradient boosting decision tree model pre-trained using a large amount of historical task data. The model input is a multi-dimensional feature vector, and the output is a real value, i.e., a priority score. A higher score indicates that the task should be executed with higher priority in the current context. Training data comes from historical task execution records and human evaluations, collecting historical cases. Each case contains a feature vector of a task at a specific time and in a specific context. Humans assign an ideal priority score to the task based on its actual urgency and impact on the global objective. The model is trained using supervised learning. Since predicting priority scores is a regression problem, the loss function uses mean squared error, and training is performed using stochastic gradient descent or the Adam optimizer. Training parameters include the number and depth of decision trees, and the learning rate. The training objective is to minimize the difference between the priority score predicted by the model and the manually labeled score. The node unit inputs the feature vectors of the new subtask and the current task into the pre-trained model to obtain two priority scores, which are used as the first and second priorities, respectively.

[0095] It should be noted that by calculating task priority, scheduling decisions that are beneficial to the overall goal can be selected when resources are limited or when task conflicts occur. This ensures that system resources can prioritize serving the most critical and urgent tasks, thereby improving the UAV system's responsiveness and task completion efficiency in complex scenarios such as sudden tasks and multi-task parallelism.

[0096] like Figure 3 As shown, edge computing is used to match corresponding node units for each subtask in the UAV scheduling network, including:

[0097] S501. Based on the real-time status data of each UAV node in the UAV scheduling network, extract the status features through a preset feature extraction model and construct the node status vector.

[0098] S502. Analyze the requirements of each subtask using a preset task parsing model and construct a requirement vector.

[0099] S503. Calculate the similarity between the node state vector and the demand vector, and select nodes with a similarity greater than a preset similarity threshold as the first set of matching nodes;

[0100] S504. Optimize the first matching node set based on the resource consumption and UAV scheduling of the simulated subtask execution process, and obtain the corresponding node units.

[0101] In this embodiment, based on the real-time status data of each UAV node in the UAV scheduling network, including but not limited to geographic coordinates, remaining battery percentage, sensor type and status, and computing load rate, state features are extracted using a pre-defined feature extraction model. This model includes, but is not limited to, a stacked autoencoder, which consists of an encoder and a decoder. The input is a pre-processed high-dimensional vector of the node state. The encoder typically consists of multiple fully connected layers, with the number of neurons decreasing layer by layer. The ReLU activation function is used for transformation, ultimately outputting a significantly reduced-dimensional vector at the bottleneck layer—the node state vector. The decoder takes this bottleneck layer vector as input and reconstructs the original high-dimensional input using a symmetrical fully connected layer structure. The output is the reconstructed vector. The goal of model training is to make the reconstructed vector as close as possible to the original input. The training dataset does not require manual annotation; it directly uses snapshots of a large number of UAV node state data collected during historical flights as samples. The model employs a self-supervised learning approach, with the learning objective being the reconstruction of the input data itself. The loss function uses mean squared error, calculating the difference between the reconstructed vector and the original input vector. The Adam optimizer is used for training, and the learning rate can be set to a small value. Training continues until the reconstruction error on the validation set no longer decreases significantly. After training, the decoder is discarded, and only the encoder is retained for online feature extraction.

[0102] The real-time state data of the UAV nodes, after being collected and preprocessed, is input into the trained encoder. The output of its bottleneck layer is the node state vector, which includes the essential features and internal structure of the original state data. By constructing the node state vector and quantizing the state data to obtain the corresponding state vector data, the computational load of the matching process can be reduced, thus improving computational efficiency.

[0103] Specifically, for each subtask parsed, a pre-defined task parsing model is used to analyze its requirements. This model includes, but is not limited to, rule-based and lookup-based mappers. Based on the subtask's type label, the model retrieves the corresponding basic requirement vector template from a predefined task type-capability requirement template library. This template is a vector, where each dimension represents the general requirement intensity for that dimension's capability. For example, for high-definition fixed-point reconnaissance, the requirement value for optical sensor resolution is high, while the requirement value for maximum payload is low. The model then instantiates and adjusts the template based on the specific parameters of the subtask. For example, if the area parameter is large, the requirement value for endurance time will be increased proportionally; if real-time image transmission is required, the requirement value for communication bandwidth will be increased. The adjustment rules are pre-defined based on domain knowledge. The vector obtained after instantiation and adjustment is the requirement vector for that subtask. Its dimensions are exactly the same as the node state vector, and the value of each dimension represents the quantitative requirement of the UAV's corresponding capabilities for executing the subtask.

[0104] It should be noted that by constructing a demand vector that is isomorphic to the node state vector, a comparative analysis between tasks and resources can be performed, improving the accuracy and effectiveness of task matching, avoiding resource mismatch caused by differences in demand understanding, and improving the accuracy and efficiency of the task analysis and matching process.

[0105] Furthermore, the cosine similarity between the node state vector and the demand vector is calculated. A similarity threshold is set according to the accuracy requirements of the task matching analysis, and nodes with similarity greater than the preset similarity threshold are selected as the first set of matching nodes. By filtering through the similarity threshold, nodes that meet the task requirements can be quickly selected, thereby improving the task response capability of the system.

[0106] Specifically, the simulation process simulates the resource consumption and UAV scheduling of subtasks based on the first set of matching nodes. It simulates the complete process of each node in the first set executing the subtask and calculates key simulation indicators, including but not limited to calculating the total resource consumption for the entire task cycle, assessing whether a single node is overloaded, and predicting the total task completion time. Based on the simulation results, the first set of matching nodes is optimized, including but not limited to replacing nodes with lower loads within the set with nodes that are overloaded. The first set of matching nodes is iteratively optimized to determine the optimal node combination in terms of resource utilization, execution efficiency, and collaborative security, thus obtaining the node unit corresponding to the subtask. By dynamically simulating and analyzing resource bottlenecks and path conflicts, proactive optimization and adjustments can be made before task execution, improving the quality and effectiveness of task allocation schemes, enhancing the efficiency and collaborative effect of node units in completing tasks, and ultimately optimizing UAV resource utilization and task execution success rate.

[0107] Furthermore, by comparing the first priority and the second priority, the UAV nodes in the node unit are dynamically scheduled to obtain the first control strategy, including:

[0108] S601. Calculate the difference between the first priority and the second priority, and set the trigger threshold according to the load of the UAV in the node unit;

[0109] S602. When the difference is greater than the trigger threshold, select a set of subtask nodes from the node unit to execute the subtask, and the remaining nodes continue to execute the current task to obtain the first control strategy.

[0110] In this embodiment, the difference between the first priority and the second priority is calculated to quantify the relative importance and urgency of new tasks. Based on the load of the UAVs in the node unit, the global load of the node unit is evaluated in real time. The load is analyzed by monitoring the computing resource utilization, communication channel utilization, current task execution progress, and criticality of each UAV within the node unit. The indicators are normalized and weighted to calculate the load value. A trigger threshold is then set based on this weighted load value. When the node unit is lightly loaded, it can flexibly respond to new tasks with similar priorities, setting a lower trigger threshold. When the node unit is under high load, a higher trigger threshold is set to avoid system oscillations caused by frequent task switching. By dynamically setting the trigger threshold, a balance between response timeliness and system stability can be effectively achieved. This ensures rapid response to various beneficial task requests when system resources are sufficient, improving task response speed. It also prevents frequent and unnecessary task rescheduling due to minor priority differences under high load conditions, improving system robustness and resource utilization efficiency.

[0111] Specifically, when the difference exceeds the trigger threshold, the scheduling mechanism is activated. A set of subtask nodes is selected from the node unit to execute the subtask. The matching degree between the task node and the new subtask, the node's remaining energy, and the proximity of the node's current location to the subtask area are analyzed to select the minimum number of nodes that meet the subtask requirements as the subtask node set for execution. UAV nodes not selected for the subtask node set continue executing their original tasks, resulting in the first control strategy. By selecting some UAV nodes to execute new tasks while retaining others to continue their original tasks, the system can ensure immediate response to high-priority tasks while minimizing or eliminating disruption to ongoing tasks. This improves system task continuity and overall efficiency, reduces resource waste and task interruption losses, and achieves efficient parallel processing of multiple tasks within the node unit, enhancing the system's throughput and overall performance in complex, multi-task concurrent environments.

[0112] Furthermore, the node conflict situations between different UAV nodes within the same node unit and between different node units are analyzed separately. The priority of conflicting nodes is analyzed, and the first control strategy is optimized to obtain a second control strategy, including:

[0113] S701. Analyze the node conflict between different UAV nodes in the same node unit and the node conflict between different node units respectively, determine the scope of conflict impact, analyze the priority between conflicting nodes, and determine the conflict coordination strategy.

[0114] S702. Based on the conflict coordination strategy, the first control strategy is optimized to obtain the second control strategy.

[0115] In this embodiment, based on the preset trajectories and task timelines of all UAV nodes in the first control strategy, conflict detection analysis is performed. The conflict situations between different UAV nodes in the same node unit and between different node units are analyzed to determine the scope of conflict impact. The probability of task failure that may result from the failure of each conflict node due to conflict is analyzed to calculate the priority between conflict nodes. Conflict coordination is performed on each conflict node to determine the conflict coordination strategy. By identifying collision risks and task interference in advance through conflict analysis, and determining the priority of conflict coordination based on the probability of task failure, it can be ensured that nodes that are crucial to the success of the overall task are prioritized when coordinating resources. The final conflict coordination strategy is the comprehensive optimal solution for the overall system performance.

[0116] Specifically, based on the conflict coordination strategy as the optimization constraint and behavioral instruction, the first control strategy is optimized to obtain the second control strategy. By optimizing the first control strategy, the safety and global coordination of the UAV collaborative control process can be improved, and the reliability and overall efficiency of large-scale UAV swarm missions in complex environments can be enhanced.

[0117] Furthermore, the node conflict situations between different UAV nodes within the same node unit and between different node units are analyzed separately to determine the scope of conflict impact, analyze the priorities between conflicting nodes, and determine conflict coordination strategies, including:

[0118] S801. Analyze the node conflict between different UAV nodes in the same node unit and the node conflict between different node units respectively, screen out the conflict nodes that overlap in space and time, analyze the influence range of the conflict nodes, and obtain the conflict area.

[0119] S802. In the conflict zone, analyze the probability of task failure caused by the failure of each conflict node, and determine the conflict priority of the conflict nodes.

[0120] S803. For conflicts within a node unit, coordinate the running order of conflicting nodes according to the conflict priority from high to low. For conflicts between different node units, simulate the impact of the coordination process on task execution performance based on the scheduling nodes corresponding to different node units, select the optimal coordination method, and obtain the conflict coordination strategy.

[0121] In this embodiment, based on the predetermined flight trajectory generated by each UAV node according to the first control strategy, the spatial position of the UAV changes over time. The node conflict between different UAV nodes in the same node unit and the node conflict between different node units are analyzed respectively. It is calculated whether the spatial positions of any two UAV nodes have an intersection in the same time segment. When it is detected that the spatial positions of two or more UAV nodes have spatial overlap and the overlap lasts for one or more overlapping time periods, the corresponding node is identified as a conflict node. The spatial union of the spatial positions of the conflict node in the conflict period is calculated, the influence range of the conflict node is analyzed, and the conflict area is obtained.

[0122] It should be noted that by detecting spatiotemporal conflicts, potential collision risk points can be identified in advance, providing a sufficient time window for proactive coordination of drones. By quantifying the conflict area, accurate spatial boundaries are provided for drone coordination, and accurate spatial references are provided for the drone coordination process of avoidance path planning, thereby improving space utilization efficiency and system safety.

[0123] Within the conflict zone, for each conflict node, the analysis considers the irreplaceability of the subtask undertaken by the node within the overall task, the global priority of the subtask itself, the completion status of the node's current task stage, and whether the node's task interruption or delay due to avoidance would affect subsequent critical task stages. Quantitative analysis indicators are then used to calculate the probability of overall system task failure if the node fails due to conflict coordination needs. Based on this failure probability, a weighted average is applied to determine the conflict priority of each node. A higher probability indicates a greater destructive impact of the node's failure on the overall task, and thus a higher conflict priority. By quantitatively analyzing the impact of each conflict node on the overall task, priority can be given to nodes crucial to task success during coordination, improving the reliability of system decisions in complex conflict scenarios.

[0124] Specifically, for conflicts within a node unit, the scheduling node of that node unit makes coordination decisions based on conflict priority. Following a descending order of priority, the right-of-way for high-priority nodes is guaranteed, while the execution order of low-priority nodes is coordinated by adjusting routes, temporary hovering, changing altitude layers, or delaying departure times. For conflicts between different node units, the scheduling nodes corresponding to each node unit communicate and interact, simulating various coordination processes based on global task information. They evaluate the impact of each coordination method on their own and the other party's task execution performance, including but not limited to changes in task completion time, total energy consumption, and task completion percentage. By comparing simulation results, the scheduling nodes jointly select the optimal coordination method that minimizes damage to global performance. All internal coordination decisions and cross-unit negotiation results are integrated to obtain the conflict coordination strategy.

[0125] It is important to emphasize that by adopting a hybrid coordination mechanism within and between different node units, conflicts within a unit can be quickly and centrally processed, ensuring the timeliness of coordination. Distributed simulation and negotiation of cross-unit conflicts can filter out the overall optimal solution through performance simulation, avoiding decisions that are locally optimal but globally suboptimal. Through a hierarchical and domain-based coordination strategy generation method, various conflicts can be resolved efficiently, improving the feasibility and efficiency of collaborative operations of UAV swarms in dense and complex airspaces.

[0126] Furthermore, based on the conflict coordination strategy, the first control strategy is optimized to obtain a second control strategy, including:

[0127] S901. Allocate resources to each UAV according to the conflict coordination strategy to obtain the resource allocation method;

[0128] S902. Optimize the first control strategy according to the resource allocation method to obtain the second control strategy.

[0129] In this embodiment, resources are allocated to each UAV according to the conflict coordination strategy. For each UAV node in the conflict coordination strategy, the specific parameters of various key resources required to execute the task are calculated. Resources include, but are not limited to, airspace resources, time resources, and communication resources. The allocation process must ensure that within the conflict area, there is no longer any competitive overlap of similar resources between any two nodes in the spatiotemporal dimension. A corresponding resource list is generated for each node to obtain the resource allocation method. By setting resource occupation instructions for each UAV node, accurate resource occupation can be provided for each UAV, avoiding resource conflicts and improving the security and reliability of the final control strategy.

[0130] Specifically, the first control strategy is optimized according to the resource allocation method. The new waypoint sequence planned for each node in the resource allocation method replaces the original waypoints in the first control strategy; the adjusted time parameters are updated in the task timing table; and the allocated altitude layer, communication channel, and other parameters are written into the corresponding control command fields. After completing the parameter update and command adjustment, the second control strategy is obtained. Optimizing the first control strategy through resource optimization improves the flexibility of the second control strategy in responding to task dynamics, and enhances the success rate and overall efficiency of multi-UAV systems performing tasks in complex environments.

[0131] like Figure 4 As shown, a real-time integrated control system for multiple unmanned aerial vehicle (UAV) subsystems based on edge computing is used to implement a real-time integrated control method for multiple UAV subsystems based on edge computing, including:

[0132] The UAV scheduling network construction module analyzes the real-time capabilities of UAV nodes based on the collected real-time status data of UAVs, selects scheduling nodes, and dynamically allocates UAVs to the scheduling areas corresponding to the scheduling nodes based on the communication connection status between UAVs, so as to serve as task execution nodes and build a UAV scheduling network.

[0133] The task matching module, in response to the task instruction, parses the task into different types of subtasks, and matches each subtask with a corresponding node unit in the UAV scheduling network through edge computing. Each node unit calculates the first priority of the subtask and the second priority of the currently executing task.

[0134] The control strategy formulation module compares the first priority and the second priority, and dynamically schedules the UAV nodes in the node unit to obtain the first control strategy.

[0135] The control strategy optimization module analyzes the node conflict between different UAV nodes in the same node unit and the node conflict between different node units, analyzes the priority between conflicting nodes, optimizes the first control strategy, and obtains the second control strategy.

[0136] In this embodiment, the UAV scheduling network construction module selects scheduling nodes by evaluating the comprehensive capabilities of each UAV node, and classifies surrounding UAVs into its scheduling area to form a collaborative subnet based on the quality of real-time communication links. This allocation of control responsibilities eliminates the risk of single-point failures caused by a single control node, and the division of communication areas improves the stability of control links within the control network. The task matching module responds to upper-layer task instructions, parses them into multiple subtasks, and utilizes the distributed computing capabilities of edge computing to accurately match the resource requirements of each subtask with the real-time status and functions of each node in the UAV scheduling network. It allocates suitable node units and quantifies the priority of subtasks and the current tasks of node units. By configuring the relationship between tasks and execution resources, the effectiveness and efficiency of task execution can be improved.

[0137] Specifically, the control strategy formulation module, based on the priority information output by the task matching module, generates a preliminary first control strategy by comparing the priority difference between the new subtask and the current task, combined with the real-time load of the node unit. This rapid task response ensures timely execution of high-priority tasks and avoids unnecessary task switching overhead, ensuring system resources are prioritized for the most critical tasks while improving system stability and task throughput. The control strategy optimization module detects and analyzes spatiotemporal conflicts between UAVs within the same node unit and between different node units, assesses the relative importance of conflicting nodes, and formulates corresponding conflict coordination strategies to obtain the final second control strategy. Conflict analysis effectively prevents collisions and task interference between UAVs, while conflict coordination strategies reduce the impact on task performance while ensuring safety.

[0138] Example 2:

[0139] This embodiment uses a specific scenario of combined emergency monitoring and material delivery in a mountainous fire area to illustrate the overall process of this technical solution. This embodiment assumes a cluster of six heterogeneous UAVs initially dispersed in a mountainous airspace.

[0140] The initial cluster status is as follows: Drones U1 and U2, equipped with high-definition visible light cameras and infrared thermal imagers, have battery levels of 85% and 78% respectively, and are located over the eastern side of the fire. Drones U3 and U4, equipped with multispectral scanners and gas sensors, have battery levels of 90% and 82% respectively, and are located over the western side of the fire. Drones U5 and U6, cargo models carrying emergency supply packages, have battery levels of 88% and 80% respectively, and are located at the rear base. All drones maintain connectivity via an ad hoc network communication link, but link quality varies due to terrain obstruction.

[0141] In the first phase, the system first constructs a drone dispatch network. The drone dispatch network construction module begins working, collecting real-time status data from each drone. A preset status analysis model, based on battery power, computational load, sensor completeness, and historical reliability data, calculates the real-time capability value for each drone. Assuming that the calculated capability values ​​for U1, U3, and U5 are 0.87, 0.89, and 0.84 respectively, all exceeding the preset capability threshold of 0.80, they are selected as dispatch nodes, forming the initial dispatch node set. Subsequently, the module analyzes the communication connection status between all drones and calculates the communication status value. Taking dispatch node U1 as an example, its communication status value with U2 is 0.95, the best; with U5, it is 0.85, second best; while its communication status value with U4, which is obstructed by a mountain, is 0.60, poor. Based on the strongest connection principle and the preset communication threshold of 0.70, U2 is dynamically assigned to U1's dispatch area, and U5 is also assigned to U1's dispatch area. Similarly, U3 forms another dispatch area with U4 and U6. Ultimately, two scheduling node regions were constructed: Region 1 includes scheduling node U1 and task execution nodes U2 and U5; Region 2 includes scheduling node U3 and task execution nodes U4 and U6. This process was completed in a distributed manner among candidate nodes through edge computing, forming a decentralized control infrastructure.

[0142] In the second phase, the task matching module responds to the incoming emergency task instruction: to perform precise fire monitoring within a 5-kilometer radius of coordinates X (East longitude) and Y (North latitude) and deliver two boxes of fire extinguishing bombs to coordinate Z. The module parses this instruction and decomposes it into two different types of sub-tasks using a pre-defined task parsing model: Sub-task A involves performing high-resolution visible light and infrared fusion imaging reconnaissance on a designated area, and sub-task B involves delivering two boxes of fire extinguishing bombs to the designated coordinate Z. Edge computing is used to match node units for each sub-task within the pre-built scheduling network. For sub-task A, the module extracts its requirement vector, which emphasizes high-resolution imaging, infrared capabilities, endurance, and location proximity. Simultaneously, each scheduling node extracts the node state vector of the UAVs under its jurisdiction. By calculating the cosine similarity between the demand vector and the state vector of each node, U1, U2, and U3, whose similarity is higher than a preset similarity threshold, are initially selected to form the first set of matching nodes. The module simulates the process of this set performing a reconnaissance mission, considering coverage path, endurance, and data return. After optimization, it is determined that U1 and U2 form a node unit to execute sub-task A. For sub-task B, its demand vector emphasizes load capacity, delivery accuracy, and range. After matching and optimization, it is determined that U5 and U6 form a node unit to execute it, and U5 temporarily assumes the coordination responsibility of this unit. The scheduling node within each node unit calculates the first priority of its newly assigned sub-task. Sub-tasks A and B are both assigned high priority, and the second priority of the currently executing task is evaluated.

[0143] In the third stage, the control strategy formulation module begins its work. Comparing the first priority (0.9) of subtask A with the second priority (0) of the current task, if the difference exceeds the trigger threshold, U1 and U2 are immediately dynamically scheduled to perform fire reconnaissance tasks; similarly, U5 and U6 are scheduled to perform material delivery tasks. Based on this, the first control strategy is generated, defining the scanning paths for U1 and U2 to the fire site, and the delivery paths and timelines for U5 and U6 to coordinate Z.

[0144] In the fourth stage, the control strategy optimization module performs conflict analysis and optimization on the first control strategy. The module analysis identified two potential conflicts: U1 and U2 have spatial and temporal overlap at a turn on their reconnaissance paths; U5's delivery route intersects with U1's reconnaissance route in a certain airspace and their timings are close. For internal conflicts, the analysis considers U1, as a scheduling node carrying critical sensors, to have a higher probability of mission failure due to its failure; therefore, its conflict priority is higher, and the coordination strategy is to instruct U2 to adjust its speed or slightly detour to avoid U1. For cross-unit conflicts, scheduling nodes U1 and U5 communicate and negotiate, simulating different coordination schemes. Simulation results show that allowing U5 to briefly climb and pass over U1 has the least impact on the overall completion time of the two sub-tasks. Based on this, the module determines the conflict coordination strategy. Finally, the first control strategy is optimized based on this strategy: fine-tuning U2's local path and speed, adjusting U5's flight altitude layer, and re-coordinating the time window. The optimized second control strategy is then issued to all UAVs for execution.

[0145] In summary, this embodiment demonstrates a complete closed-loop process from cluster initialization, dynamic networking, task parsing and intelligent matching, priority-based dynamic scheduling, to final conflict prediction and strategy optimization. By introducing an edge computing architecture, computing is distributed across scheduling nodes such as U1, U3, and U5, achieving low-latency local decision-making and coordination. A two-level priority mechanism ensures resource acquisition and global security for critical tasks, guaranteeing the practical feasibility and efficiency of task allocation. The entire system exhibits adaptive, collaborative, secure, and efficient operation capabilities when facing sudden complex tasks, solving the problems of high latency, single point of failure, and the difficulty of handling dynamic conflicts using traditional methods.

[0146] The above description is merely a preferred embodiment of this application. The scope of protection of this application is not limited to the above embodiments. All technical solutions falling within the scope of this application's concept are within the scope of protection of this application. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of this application should also be considered within the scope of protection of this application.

Claims

1. A real-time integrated control method for multiple unmanned aerial vehicle (UAV) subsystems based on edge computing, characterized in that, include: Based on the collected real-time status data of UAVs, the real-time capabilities of UAV nodes are analyzed, scheduling nodes are selected, and UAVs are dynamically allocated to the scheduling areas corresponding to the scheduling nodes based on the communication connection status between UAVs, serving as task execution nodes, thus constructing a UAV scheduling network. In response to task instructions, analyze task requirements and parse the task into different types of subtasks; Based on the real-time status data of each drone node in the drone scheduling network, status features are extracted using a preset feature extraction model to construct a node status vector. Each subtask is analyzed for requirements using a pre-defined task analysis model, and a requirement vector is constructed. Calculate the similarity between the node state vector and the demand vector, and select nodes with a similarity greater than a preset similarity threshold as the first set of matching nodes; Based on the resource consumption and UAV scheduling of the simulated subtask execution process of the first matching node set, the first matching node set is optimized to obtain the corresponding node unit; Using a pre-defined priority analysis model, each node unit calculates the first priority of the subtask and the second priority of the currently executing task. By comparing the first priority and the second priority, the UAV nodes in the node unit are dynamically scheduled to obtain the first control strategy; The node conflict situation between different UAV nodes in the same node unit and the node conflict situation between different node units are analyzed respectively. The priority between conflicting nodes is analyzed, and the first control strategy is optimized to obtain the second control strategy.

2. The real-time integrated control method for multiple unmanned aerial vehicle (UAV) subsystems based on edge computing according to claim 1, characterized in that, The process involves analyzing the real-time capabilities of drone nodes based on collected real-time status data, selecting scheduling nodes, and dynamically allocating drones to the corresponding scheduling areas of the scheduling nodes based on the communication connection status between drones, thus constructing a drone scheduling network as task execution nodes. This includes: Based on the collected real-time status data of the UAV, the real-time capabilities of the UAV nodes are analyzed through a preset status analysis model to obtain real-time capability values. Select drone nodes whose real-time capability values ​​are greater than a preset capability threshold and construct a set of scheduling nodes; For each scheduling node, the communication connection status between drones is analyzed, and the drones are dynamically assigned to the scheduling area corresponding to the scheduling node as task execution nodes. By combining scheduling nodes and corresponding task execution nodes, a drone scheduling network is constructed.

3. The real-time integrated control method for multiple unmanned aerial vehicle (UAV) subsystems based on edge computing according to claim 2, characterized in that, For each scheduling node, the communication connection status between UAVs is analyzed, and the UAVs are dynamically allocated to the scheduling area corresponding to the scheduling node as task execution nodes, including: Analyze the communication connection status between drones and calculate the communication status values; Starting from each scheduling node, drone nodes whose communication status values ​​are greater than a preset communication threshold are selected step by step and used as the corresponding first node set. For overlapping nodes in the candidate node sets of different scheduling nodes, select the scheduling node with the highest communication status value with the overlapping node, assign it to the corresponding scheduling area, update the first node set, and obtain the second node set. The region consisting of the set of second nodes corresponding to each scheduling node is designated as the scheduling region, and each second node is designated as the task execution node.

4. The real-time integrated control method for multiple unmanned aerial vehicle (UAV) subsystems based on edge computing according to claim 1, characterized in that, The comparison of the first priority and the second priority, and the dynamic scheduling of the UAV nodes in the node unit to obtain the first control strategy, includes: Calculate the difference between the first priority and the second priority, and set the trigger threshold according to the load of the drone in the node unit; If the difference is greater than the trigger threshold, a set of subtask nodes is selected from the node units to execute the subtask, while the remaining nodes continue to execute the current task, thus obtaining the first control strategy.

5. The real-time integrated control method for multiple unmanned aerial vehicle (UAV) subsystems based on edge computing according to claim 1, characterized in that, The process involves analyzing node conflicts between different UAV nodes within the same node unit and between different node units, analyzing the priorities of conflicting nodes, optimizing the first control strategy, and obtaining a second control strategy, including: The node conflict situation between different UAV nodes in the same node unit and between different node units are analyzed to determine the scope of the conflict, analyze the priority between conflicting nodes, and determine the conflict coordination strategy. Based on the conflict coordination strategy, the first control strategy is optimized to obtain the second control strategy.

6. The real-time integrated control method for multiple unmanned aerial vehicle (UAV) subsystems based on edge computing according to claim 5, characterized in that, The analysis of node conflicts between different UAV nodes within the same node unit and between different node units determines the scope of conflict impact, analyzes the priorities of conflicting nodes, and determines conflict coordination strategies, including: The node conflict situation between different UAV nodes in the same node unit and the node conflict situation between different node units were analyzed respectively. Conflicting nodes with spatial overlap and temporal overlap were screened out, and the influence range of conflicting nodes was analyzed to obtain the conflict area. In the conflict zone, analyze the probability of task failure caused by the failure of each conflict node, and determine the conflict priority of the conflict nodes; For conflicts within a node unit, the running order of the conflicting nodes is coordinated according to the conflict priority from high to low. For conflicts between different node units, the impact of the coordination process on task execution performance is simulated based on the scheduling nodes corresponding to different node units, and the optimal coordination method is selected to obtain the conflict coordination strategy.

7. The real-time integrated control method for multiple unmanned aerial vehicle (UAV) subsystems based on edge computing according to claim 6, characterized in that, The step of optimizing the first control strategy according to the conflict coordination strategy to obtain the second control strategy includes: Based on the conflict coordination strategy, resources are allocated to each drone to obtain the resource allocation method; The first control strategy is optimized according to the resource allocation method to obtain the second control strategy.

8. A real-time integrated control system for multiple unmanned aerial vehicle (UAV) subsystems based on edge computing, characterized in that, The method for implementing the real-time integrated control of multiple unmanned aerial vehicle (UAV) subsystems based on edge computing as described in any one of claims 1 to 7 includes: The UAV scheduling network construction module analyzes the real-time capabilities of UAV nodes based on the collected real-time status data of UAVs, selects scheduling nodes, and dynamically allocates UAVs to the scheduling areas corresponding to the scheduling nodes based on the communication connection status between UAVs, so as to serve as task execution nodes and build a UAV scheduling network. The task matching module, in response to the task instruction, parses the task into different types of subtasks, and matches each subtask with a corresponding node unit in the UAV scheduling network through edge computing. Each node unit calculates the first priority of the subtask and the second priority of the currently executing task. The control strategy formulation module compares the first priority and the second priority, and dynamically schedules the UAV nodes in the node unit to obtain the first control strategy. The control strategy optimization module analyzes the node conflict between different UAV nodes in the same node unit and the node conflict between different node units, analyzes the priority between conflicting nodes, optimizes the first control strategy, and obtains the second control strategy.

Images