Cluster fixed-wing unmanned aerial vehicle operation method, system, equipment and medium

Abstract

The invention relates to the technical field of unmanned aerial vehicles, and discloses a cluster fixed-wing unmanned aerial vehicle operation method, system and device and a medium, and the method comprises the steps: task analysis: receiving a natural language instruction or structured task description input by a user, and analyzing the natural language instruction or structured task description into a standardized task semantic structure through a vertical domain large model; environment information fusion: constructing a three-dimensional situation map in combination with the unmanned aerial vehicle cluster state and the multi-source environment information; operation strategy generation: based on the standardized task semantic structure and the three-dimensional situation map, tasks are distributed to corresponding agents through a vertical domain large model, and operation strategies are generated through the agents; instruction output and execution: converting the operation strategy into an executable control instruction, and issuing the executable control instruction to each unmanned aerial vehicle node; and feedback optimization: recording a task execution result and an abnormal event for subsequent iterative optimization of the vertical domain large model. According to the method, accurate semantic analysis of a natural language task instruction and deep understanding of a low-altitude scene can be realized by combining an aviation vertical domain large model.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) technology, and in particular to a method, system, device, and medium for operating swarms of fixed-wing UAVs. Background Technology

[0002] With the gradual opening of low-altitude airspace and the rapid development of drone technology, fixed-wing drones, with their advantages of long range, high speed, and long endurance, have demonstrated enormous application potential in long-distance, large-scale tasks such as emergency rescue, logistics delivery, power line inspection, and agricultural plant protection. Especially against the backdrop of the low-altitude economy becoming a key emerging industry, drone swarm collaborative operations are considered a crucial means to improve the efficiency of low-altitude resource utilization.

[0003] However, most current UAV swarm control systems are based on traditional centralized or distributed control architectures, relying on pre-set task scripts and structured instruction inputs, lacking the ability to understand complex, ambiguous, or unstructured task descriptions. Operators need professional training to write instructions that meet system requirements, which severely restricts the system's usability and rapid deployment capabilities, making it difficult to adapt to sudden tasks or dynamically changing combat environments.

[0004] A more prominent problem is that existing systems are severely lacking in dynamic perception and intelligent decision-making capabilities for complex low-altitude environments. Low-altitude airspace is highly complex, with not only static restrictions such as fixed no-fly zones and airport clearance zones, but also frequent dynamic risks such as temporary airspace control, sudden weather changes (e.g., sudden strong winds, thunderstorms, low visibility), electromagnetic interference, and intrusions by other aircraft. Traditional control methods typically employ rule-based static obstacle avoidance algorithms or simple path replanning mechanisms, which cannot fuse and analyze multi-source heterogeneous information, and are even less capable of predicting potential conflicts or assessing policy compliance (e.g., whether it violates urban privacy flight regulations or exceeds the permitted beyond visual line of sight flight range). When emergencies occur, the system often reacts slowly, relying on manual intervention, leading to mission interruptions or even safety incidents, making it difficult to meet the autonomous operation requirements of high-safety-level scenarios.

[0005] In recent years, large language models have demonstrated powerful capabilities in natural language understanding, knowledge reasoning, and generative decision-making, providing new technological pathways for solving these problems. However, the direct application of general-purpose large models in the aviation field faces serious challenges: their training data generally lacks aviation expertise, leading to misunderstandings of flight terminology, operating procedures, and airworthiness standards, and even generating erroneous instructions that violate safety logic. For example, the model might suggest flying over no-fly zones or ignore the operational limitations of specific aircraft types in adverse weather conditions, posing serious safety hazards. Furthermore, existing research largely focuses on single-aircraft control or simple formation flight, lacking a systematic architecture that deeply integrates large models with multi-UAV collaborative control, task allocation, and resource scheduling. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes a method, system, device, and medium for operating swarm fixed-wing unmanned aerial vehicles (UAVs). It enhances knowledge and fine-tunes safety constraints in the aviation field, enabling it to understand professional contexts, integrate real-time situational awareness, and generate compliant and robust operating strategies. This truly achieves intelligent, autonomous, and highly reliable operation of UAV swarms in complex low-altitude environments.

[0007] The technical solution adopted in this invention is as follows: A method for operating a cluster of fixed-wing unmanned aerial vehicles (UAVs) includes: Task parsing: Receives natural language instructions or structured task descriptions input by the user and parses them into a standardized task semantic structure using a vertical domain large model; the standardized task semantic structure includes task objectives, geographical scope, time constraints, and compliance requirements; Environmental information fusion: Combining the status of the UAV swarm and multi-source environmental information to construct a three-dimensional situation map; the multi-source environmental information includes airspace information, meteorological data, and policy dynamics; Operation strategy generation: Based on the standardized task semantic structure and the three-dimensional situation map, the task is distributed to the corresponding intelligent agent through the vertical domain large model, and the intelligent agent generates the operation strategy; the operation strategy includes sub-task allocation scheme, optimal route, potential risk identification results and emergency response plan. Command output and execution: The operation strategy is converted into executable control commands and sent to each UAV node; Feedback optimization: Record task execution results and abnormal events for subsequent iterative optimization of the large-scale model in the vertical domain.

[0008] Furthermore, the task parsing specifically includes: User voiceprint features are extracted using a voiceprint recognition method for identity verification. The voiceprint recognition method is based on speaker identification or speaker confirmation patterns, uses a Gaussian mixture model to extract voiceprint feature vectors, and uses cosine similarity to match a pre-registered user voiceprint database. The user's audio signal is converted into natural language text using a speech recognition method, which employs an end-to-end model. By combining an aviation knowledge base, the natural language text is converted into a standardized task semantic structure in a preset format through a large vertical model.

[0009] Furthermore, the environmental information fusion specifically includes: By combining the drone swarm status and multi-source environmental information, asynchronous sensor data is spatiotemporally aligned using Kalman filtering. A three-dimensional airspace graph structure is constructed using a graph neural network. In the three-dimensional airspace graph structure, nodes represent drones, obstacles, and no-fly zones, and the weights of edges represent the probability of conflict. Output a standardized three-dimensional situational structure in a preset format.

[0010] Furthermore, the generation of the operation strategy specifically includes: Subtask allocation: Based on the standardized task semantic structure and the three-dimensional situation map, the Hungarian optimization method is used to achieve optimal matching between UAVs and tasks. The allocation strategy is dynamically adjusted by combining the near-end strategy optimization method to generate a subtask allocation scheme. Route planning: Based on the sub-task allocation scheme and the three-dimensional situation map, global path planning is performed using the A-satellite path search method, combined with the random response method and random sampling extended tree structure for dynamic obstacle avoidance, generating the optimal route.

[0011] Furthermore, the operation strategy generation also includes: Risk identification: Based on the optimal flight path and real-time environmental data, a graph neural network is used to predict airspace conflicts, a fuzzy logic system is used to assess the risk level, and a risk assessment report is generated. Emergency Response: Based on the aforementioned risk assessment report, suitable solutions are selected from the preset emergency rule base using a vertical domain big data model to generate a tiered emergency response plan.

[0012] Furthermore, the instruction output and execution specifically include: combining the UAV communication protocol, decomposing the operation strategy into executable control instructions through a large vertical domain model, and executing instruction distribution through the data distribution service middleware of the robot operating system.

[0013] Furthermore, the feedback optimization specifically includes: recording task execution logs, holographic data of abnormal events, and user confirmation feedback; optimizing the large vertical domain model through periodic incremental training combined with labeled abnormal events; compressing the model parameter scale using knowledge distillation technology; and synchronously updating the aviation knowledge base and compliance check logic.

[0014] An operating system for swarm fixed-wing unmanned aerial vehicles (UAVs) includes: The task parsing module is configured to receive natural language instructions or structured task descriptions input by the user, and parse them into a standardized task semantic structure through a vertical domain big model; the standardized task semantic structure includes task objectives, geographical scope, time constraints, and compliance requirements. The environmental information fusion module is configured to combine the status of the UAV cluster and multi-source environmental information to construct a three-dimensional situation map; the multi-source environmental information includes airspace information, meteorological data, and policy dynamics. The operation strategy generation module is configured to distribute tasks to corresponding agents through a vertical domain large model based on the standardized task semantic structure and the three-dimensional situation map, and generate operation strategies through the agents; the operation strategies include sub-task allocation schemes, optimal routes, potential risk identification results, and emergency response plans. The instruction output and execution module is configured to convert the operation strategy into executable control instructions and send them to each UAV node; The feedback optimization module is configured to record task execution results and abnormal events for subsequent iterative optimization of the large vertical model.

[0015] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method for operating a cluster of fixed-wing unmanned aerial vehicles.

[0016] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for operating a cluster of fixed-wing unmanned aerial vehicles.

[0017] The beneficial effects of this invention are as follows: 1. This invention integrates professional knowledge in the aviation field with the deep semantic understanding capabilities of large models to construct an intelligent operation framework for the entire process from task parsing, environmental perception, strategy generation, command execution to feedback optimization. It effectively solves key technical problems in traditional UAV swarm control, such as fuzzy task understanding, lagging environmental response, inefficient strategy generation, poor command adaptability, and long system iteration cycle.

[0018] 2. This invention, by combining a large-scale aviation vertical model, achieves accurate semantic parsing of natural language task commands and a deep understanding of low-altitude scenarios. It integrates the real-time status of UAV swarms, dynamic airspace information, meteorological conditions, and policy constraints, generating a comprehensive operational plan from the large model that includes sub-task allocation, optimal route planning, potential risk identification, and emergency response strategies, and transforming it into executable flight control commands. This invention significantly improves the autonomous decision-making capability and mission execution safety of UAV swarms in complex low-altitude environments, effectively solving problems such as task comprehension bias, delayed risk response, and rigid strategy generation in traditional control methods. It is applicable to typical low-altitude economic application scenarios such as emergency rescue, low-altitude logistics, and urban inspection, and has high intelligent and practical application value. Attached Figure Description

[0019] Figure 1 This is a flowchart of a cluster fixed-wing UAV operation method according to Embodiment 1 of the present invention.

[0020] Figure 2 This is a flowchart of environmental perception and situational awareness fusion in Embodiment 2 of the present invention. Detailed Implementation

[0021] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments are now described. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention; that is, the described embodiments are only a part of the embodiments of the invention, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0022] Example 1 like Figure 1 As shown, this embodiment provides a method for operating a cluster of fixed-wing UAVs, including: Task parsing: Receives natural language instructions or structured task descriptions from users and parses them into a standardized task semantic structure using a large-scale vertical model; the standardized task semantic structure includes task objectives, geographical scope, time constraints, and compliance requirements; Environmental information fusion: Combining the status of UAV swarms and multi-source environmental information to construct a three-dimensional situation map; multi-source environmental information includes airspace information, meteorological data, and policy dynamics; Operational strategy generation: Based on the standardized task semantic structure and 3D situation map, tasks are distributed to corresponding agents through a large vertical domain model, and operational strategies are generated by the agents. The operational strategies include sub-task allocation schemes, optimal routes, potential risk identification results, and emergency response plans. Command output and execution: Transform the operation strategy into executable control commands and send them to each UAV node; Feedback optimization: Record task execution results and abnormal events for subsequent iterative optimization of the large-scale model in the vertical domain.

[0023] Preferably, task parsing specifically includes: The voiceprint recognition method extracts user voiceprint features for identity verification, avoiding interference from non-user voices. The voiceprint recognition method is based on speaker identification or speaker confirmation patterns, and uses a Gaussian mixture model to extract voiceprint feature vectors (such as Mel frequency cepstral coefficients). It uses cosine similarity to match the pre-registered user voiceprint database to verify identity. This process can effectively reduce the impact of environmental noise, microphone differences and user emotional fluctuations on recognition performance. The user's audio signal is converted into natural language text by combining speech recognition methods. The speech recognition method adopts an end-to-end model and outputs natural language text. By combining an aviation knowledge base, a large vertical model is used to convert natural language text into a standardized task semantic structure in a preset format (such as JSON format), which specifically includes fields such as task objectives, geographical scope, time constraints, and compliance requirements.

[0024] Preferably, environmental information fusion specifically includes: By combining the status of UAV swarms and multi-source environmental information (airspace information, meteorological data, policy dynamics, etc.), Kalman filtering is used to perform spatiotemporal alignment of asynchronous sensor data such as GNSS (Global Navigation Satellite System) and ADS-B (Automatic Dependent Surveillance-Broadcast System). Among them, Kalman filtering reduces noise interference and ensures data synchronization by fusing sensor data with prediction models. Based on this, a three-dimensional airspace graph structure is constructed using graph neural networks. In the three-dimensional airspace graph structure, nodes represent drones, obstacles, and no-fly zones, and the weights of the edges represent the probability of conflict. The output is a standardized three-dimensional situation structure in a preset format (e.g., JSON format).

[0025] Preferably, the operation strategy generation specifically includes: Subtask allocation: Based on the standardized task semantic structure and 3D situation map, the Hungarian optimization method is used to achieve optimal matching between UAVs and tasks. Combined with the near-end strategy optimization method, the allocation strategy is dynamically adjusted to generate a subtask allocation scheme. Among them, the Hungarian optimization method allocates UAVs and tasks by minimizing the cost matrix, while the near-end strategy optimization method dynamically adjusts the strategy through the reward function. Route planning: Based on the sub-task allocation scheme and 3D situation map, the A-satellite path search method is used for global path planning. Combined with the random response method and random sampling extended tree structure for dynamic obstacle avoidance, the optimal route is generated. Among them, the A-satellite path search method searches for the shortest path based on the digital map, and then adapts to real-time environmental changes through random sampling extended tree structure. Risk identification: Based on optimal flight paths and real-time environmental data, graph neural networks are used to predict airspace conflicts, and a fuzzy logic system is used to assess the risk level and generate a risk assessment report; Emergency Response: Based on the risk assessment report, suitable solutions are selected from the preset emergency rule base through the vertical domain big data model to generate a tiered emergency response plan.

[0026] Preferably, the instruction output and execution specifically includes: combining the UAV communication protocol, decomposing the operation strategy into executable control instructions through a large-scale vertical model, and distributing the instructions through the robot operating system's data distribution service middleware to ensure real-time performance. Users can monitor the task progress through a visual interface and take over in case of emergency. The output of this step is user confirmation feedback and correction instructions for manual intervention.

[0027] Preferably, the feedback optimization specifically includes: recording task execution logs, holographic data of abnormal events, and user confirmation feedback for subsequent model iteration and optimization; optimizing the large-scale model in the vertical domain through periodic incremental training combined with labeled abnormal events; and compressing the model parameter scale using knowledge distillation technology while simultaneously updating the aviation knowledge base and compliance check logic. This closed-loop optimization mechanism ensures that the model continuously adapts to new scenarios and policy changes.

[0028] In summary, this method integrates aviation expertise with the deep semantic understanding capabilities of large models to construct a fully intelligent operation framework that spans from task parsing, environmental perception, policy generation, command execution to feedback optimization. This effectively solves key technical challenges in traditional UAV swarm control, such as ambiguous task understanding, delayed environmental response, inefficient policy generation, poor command adaptability, and long system iteration cycles.

[0029] Example 2 This embodiment is based on embodiment 1: This embodiment provides a method for operating a cluster of fixed-wing UAVs, including the following steps: During the mission initiation phase, the rescue commander issued a voice command via a handheld terminal: "Deliver rescue packages to P1, P2, and P3 within 2 hours, avoiding no-fly zones." Step 1: Task Analysis Phase Step 1.1: The system first verifies the user's identity through a voiceprint recognition algorithm. If the threshold is greater than 0.95, it is confirmed that the user is the rescue commander.

[0030] Step 1.2: Use the speech recognition module to convert the speech into text. The converted text result is: "Deliver rescue packages to P1, P2, and P3 within 2 hours, avoiding no-fly zones."

[0031] Step 1.3: Input the text results from Step 1.2 into the aviation vertical domain large model. The model automatically parses out the structured task semantics, such as task ID, task type, material delivery coordinates and type, task deadline, no-fly zone, maximum flight time, whether weather avoidance is enabled, number of drones in this mission, and takeoff point, etc.

[0032] Step 2, the environmental perception and situational awareness fusion stage, such as Figure 2 As shown.

[0033] Step 2.1: Data from Unit 2 (UAV_02) and Unit 3 (UAV_03) is transmitted to Unit 1 (UAV_01).

[0034] Step 2.2: Use Kalman filtering to perform spatiotemporal alignment of heterogeneous data from GNSS (Global Navigation Satellite System) and IMU (Inertial Measurement Unit) of Units 1, 2 and 3, as well as external ADS-B (Automatic Dependent Surveillance-Broadcast System), meteorological, air traffic control and other sources.

[0035] Step 2.3: Construct a three-dimensional airspace graph structure using GNN (Graph Neural Network). Nodes represent drones, obstacles, and no-fly zones, while edge weights reflect the probability of conflict.

[0036] Step 2.4: Use a graph neural network to dynamically assess risks such as airspace conflicts and sudden weather changes. Output results include: current situation map generation time, shared situational awareness data, delivery targets, target coordinates and status, no-fly zones, temporary weather zones, obstacle types, obstacle heights and locations, current status of the swarm of drones, etc.

[0037] Step 3: Strategy Generation Stage Step 3.1: The subtask allocation agent uses the output of Step 1.3, the output of Step 2.4, the drone cluster status information, and the preset task priority rules as inputs, where the priority of this delivery task is load balancing.

[0038] Step 3.2: Based on the Hungarian algorithm, allocate drones and tasks by minimizing the cost matrix according to the input information.

[0039] Step 3.3: Further optimize the task allocation strategy using the PPO (Proximity Policy Optimization) algorithm to maximize the task success rate. Finally, target points P1, P2, and P3 are assigned to the three drones respectively.

[0040] Step 3.4: Using the output of Step 3.3, namely the current coordinates of each drone and the coordinates of the delivery point, a global path is generated for each drone in combination with the A-satellite path search algorithm.

[0041] Step 3.5: During flight, use the RRT (Random Tree Faster) algorithm for local dynamic obstacle avoidance.

[0042] Step 3.6 During the flight, the GNN (Graph Neural Network) detected that there was a risk of intersection between UAV 2 and UAV 3 at the position [30.115, 103.412]. The predicted minimum approach distance was less than 600 meters, which triggered the fuzzy logic system to evaluate the risk level as "medium" and generate a red warning.

[0043] Step 3.7: Based on the risk level, select the "Altitude Difference Avoidance + Communication Negotiation" strategy from the preset rule base. Instruct UAV 2 to maintain its altitude, while UAV 3 climbs to 1100 meters, and broadcast the coordination intent via ADS-B. Output results include: task allocation decision generation time, UAV ID, delivery target point, 3D waypoints, conflict type, involved UAV numbers, conflict location, predicted minimum distance, risk level, solution (altitude difference avoidance: UAV 2 climbs to 1000m, UAV 3 climbs to 1100m), whether a no-fly zone was violated, whether the deadline was met, and whether the battery power is sufficient, etc.

[0044] Step 4: Command Execution and Cluster Control Phase Step 4.1: Input the output of Step 3.7 into the large-scale aviation vertical model. The large-scale model, combined with the UAV communication protocol, decomposes it into executable standard MAVLink (UAV communication protocol) control commands. Output results include: unique identifier of the mission, UAV execution commands, navigation to the target point, item delivery, altitude increase, communication quality, communication middleware, transmission reliability, maximum communication latency, etc.

[0045] Step 4.2: Send the standard MAVLink control commands to each drone through the DDS (Data Distribution Service) middleware of ROS2 (Robot Operating System).

[0046] Step 4.3: Commanders can view real-time information such as the drone's location, flight path, battery level, dynamic obstacle avoidance path, and material delivery progress through a visual interface.

[0047] Step 4.4: When Drone No. 2 approaches the target area, the system proactively reports: "Expected to arrive at point P2 in 2 minutes. Current wind speed is 18 m / s. It is recommended to reduce the drop height to 50 meters." After the commander confirms, the system automatically executes the correction command to ensure accurate delivery of supplies.

[0048] Step 5: Feedback and Learning Phase Step 5.1: After the task is completed, the system automatically archives the following data for model optimization, including: task execution logs, exception event records, commander feedback scores, etc. Output results include: the task's unique identifier, task completion status, task completion time, cluster performance metrics, average path deviation, number of successful conflict avoidances, energy utilization efficiency during task execution, compliance rate, whether any unexpected events occurred during task execution, user feedback, and training sample labels, etc.

[0049] Step 5.2: Periodically trigger incremental training to iteratively optimize the large model for the vertical domain.

[0050] Example 3 This embodiment provides a clustered fixed-wing UAV operating system, including: The task parsing module is configured to receive natural language instructions or structured task descriptions input by the user, and parse them into a standardized task semantic structure through a vertical domain big model; the standardized task semantic structure includes task objectives, geographical scope, time constraints, and compliance requirements. The environmental information fusion module is configured to combine the status of the UAV cluster and multi-source environmental information to construct a three-dimensional situation map; the multi-source environmental information includes airspace information, meteorological data, and policy dynamics. The operation strategy generation module is configured to distribute tasks to corresponding agents through a vertical domain large model based on the standardized task semantic structure and the three-dimensional situation map, and generate operation strategies through the agents; the operation strategies include sub-task allocation schemes, optimal routes, potential risk identification results, and emergency response plans. The instruction output and execution module is configured to convert the operation strategy into executable control instructions and send them to each UAV node; The feedback optimization module is configured to record task execution results and abnormal events for subsequent iterative optimization of the large vertical model.

[0051] Example 4 This embodiment is based on embodiment 1: This embodiment provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement a swarm fixed-wing UAV operation method according to Embodiment 1. The computer program can be in the form of source code, object code, executable file, or some intermediate form.

[0052] Example 5 This embodiment is based on embodiment 1: This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a swarm fixed-wing UAV operation method according to Embodiment 1. The computer program can be in the form of source code, object code, executable file, or some intermediate form. The storage medium includes any entity or device capable of carrying computer program code, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the content contained in the storage medium can be appropriately added to or subtracted according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the storage medium does not include electrical carrier signals and telecommunication signals.

[0053] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

[0054] It should be noted that, for the sake of simplicity, the foregoing method embodiments are described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

Claims

1. A method for operating a cluster of fixed-wing unmanned aerial vehicles (UAVs), characterized in that, include: Task parsing: Receives natural language instructions or structured task descriptions input by the user and parses them into a standardized task semantic structure through a vertical domain big data model; the standardized task semantic structure includes task objectives, geographical scope, time constraints, and compliance requirements; Environmental information fusion: Combining the status of the UAV swarm and multi-source environmental information to construct a three-dimensional situation map; the multi-source environmental information includes airspace information, meteorological data, and policy dynamics; Operation strategy generation: Based on the standardized task semantic structure and the three-dimensional situation map, the task is distributed to the corresponding intelligent agent through the vertical domain large model, and the intelligent agent generates the operation strategy; the operation strategy includes sub-task allocation scheme, optimal route, potential risk identification results and emergency response plan. Command output and execution: The operation strategy is converted into executable control commands and sent to each UAV node; Feedback optimization: Record task execution results and abnormal events for subsequent iterative optimization of the large-scale model in the vertical domain.

2. The method for operating a cluster of fixed-wing UAVs according to claim 1, characterized in that, The task parsing specifically includes: User voiceprint features are extracted using a voiceprint recognition method for identity verification. The voiceprint recognition method is based on speaker identification or speaker confirmation patterns, uses a Gaussian mixture model to extract voiceprint feature vectors, and uses cosine similarity to match a pre-registered user voiceprint database. The user's audio signal is converted into natural language text using a speech recognition method, which employs an end-to-end model. By combining an aviation knowledge base, the natural language text is converted into a standardized task semantic structure in a preset format through a large vertical model.

3. The method for operating a cluster of fixed-wing UAVs according to claim 1, characterized in that, The environmental information fusion specifically includes: By combining the drone swarm status and multi-source environmental information, asynchronous sensor data is spatiotemporally aligned using Kalman filtering. A three-dimensional airspace graph structure is constructed using a graph neural network. In the three-dimensional airspace graph structure, nodes represent drones, obstacles, and no-fly zones, and the weights of edges represent the probability of conflict. Output a standardized three-dimensional situational structure in a preset format.

4. The method for operating a cluster of fixed-wing UAVs according to claim 1, characterized in that, The generation of the operation strategy specifically includes: Subtask allocation: Based on the standardized task semantic structure and the three-dimensional situation map, the Hungarian optimization method is used to achieve optimal matching between UAVs and tasks. Combined with the near-end strategy optimization method, the allocation strategy is dynamically adjusted to generate a subtask allocation scheme. Route planning: Based on the sub-task allocation scheme and the three-dimensional situation map, global path planning is performed using the A-satellite path search method, combined with the random response method and random sampling extended tree structure for dynamic obstacle avoidance, generating the optimal route.

5. The method for operating a cluster of fixed-wing UAVs according to claim 4, characterized in that, The operation strategy generation also includes: Risk identification: Based on the optimal flight path and real-time environmental data, a graph neural network is used to predict airspace conflicts, a fuzzy logic system is used to assess the risk level, and a risk assessment report is generated. Emergency Response: Based on the aforementioned risk assessment report, suitable solutions are selected from the preset emergency rule base using a vertical domain big data model to generate a tiered emergency response plan.

6. The method for operating a cluster of fixed-wing unmanned aerial vehicles according to claim 1, characterized in that, The instruction output and execution specifically include: combining the UAV communication protocol, decomposing the operation strategy into executable control instructions through a large vertical domain model, and executing instruction distribution through the data distribution service middleware of the robot operating system.

7. The method for operating a cluster of fixed-wing UAVs according to claim 1, characterized in that, The feedback optimization specifically includes: recording task execution logs, holographic data of abnormal events, and user confirmation feedback; optimizing the large vertical domain model through periodic incremental training combined with labeled abnormal events; compressing the model parameter scale using knowledge distillation technology; and synchronously updating the aviation knowledge base and compliance check logic.

8. An operating system for clustered fixed-wing unmanned aerial vehicles, characterized in that, include: The task parsing module is configured to receive natural language instructions or structured task descriptions input by the user, and parse them into a standardized task semantic structure through a vertical domain big model; the standardized task semantic structure includes task objectives, geographical scope, time constraints, and compliance requirements. The environmental information fusion module is configured to combine the status of the UAV cluster and multi-source environmental information to construct a three-dimensional situation map; the multi-source environmental information includes airspace information, meteorological data, and policy dynamics. The operation strategy generation module is configured to distribute tasks to corresponding agents through a vertical domain large model based on the standardized task semantic structure and the three-dimensional situation map, and generate operation strategies through the agents; the operation strategies include sub-task allocation schemes, optimal routes, potential risk identification results, and emergency response plans. The instruction output and execution module is configured to convert the operation strategy into executable control instructions and send them to each UAV node; The feedback optimization module is configured to record task execution results and abnormal events for subsequent iterative optimization of the large vertical model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements a method for operating a clustered fixed-wing unmanned aerial vehicle as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for operating a clustered fixed-wing unmanned aerial vehicle as described in any one of claims 1-7.

Images