A full-process cooperative rescue control method based on a heterogeneous cluster

Abstract

The application discloses a kind of full-process collaborative rescue control methods based on heterogeneous cluster, it is related to post-disaster rescue technical field, realize full-process closed loop by the following steps: first type unmanned plane executes wide area reconnaissance, and generates task package to abnormality and end side identification;According to preset relay rule, task package is directly issued to the second type unmanned plane and executes verification;When communication obstruction or space limitation is encountered, activate the third type unmanned plane to enter the limited area and execute fine survey and communication relay;After confirming the target to be rescued, dispatch the fourth type unmanned plane to execute rescue operation;Target state is updated in real time during operation and synchronized to other nodes of cluster;When any node fails, its task is split and redistributed to other nodes.The application realizes autonomous cooperation and dynamic reconstruction from reconnaissance to rescue through constructing multiple types of unmanned plane heterogeneous cluster, and improves the automation level of rescue response, operation efficiency and system invulnerability.

Description

Technical Field

[0001] This invention relates to the field of disaster relief technology, specifically to a full-process collaborative rescue control method based on heterogeneous clusters. Background Technology

[0002] In existing technologies, unmanned aerial vehicle (UAV) systems for emergency rescue have made some progress. Early rescue models mainly relied on human search and single-type UAV operations. Human search is inefficient and poses high safety risks in complex terrain, while a single UAV is limited by its endurance and perception range, making it difficult to achieve rapid coverage in wide-area disaster scenarios. To improve rescue efficiency, the industry has gradually explored multi-UAV collaborative operation solutions, such as using fixed-wing UAVs to conduct large-area disaster surveys and combining them with multi-rotor UAVs for key area verification, forming a two-level collaborative architecture. However, such solutions are mostly limited to simple combinations at the functional division level, lacking deep autonomous collaborative decision-making mechanisms between different models. Each model still operates as an independent operating unit, and the task connection between them relies on manual intervention or preset fixed procedures, making it impossible to dynamically adjust collaborative strategies according to real-time environmental changes. In addition, existing systems generally adopt a centralized control architecture, where all perception data must be transmitted back to the ground command center for operators to analyze before issuing instructions. This model is prone to data congestion and response delays when the cluster size increases or communication conditions are limited, making it difficult to meet the urgent need for rapid response within the "golden rescue time." Regarding adaptability to complex environments, most existing solutions are designed for conventional scenarios. They lack specific adaptation mechanisms for extreme scenarios such as urban building ruins, mountain valleys, and water bodies where signal obstruction, turbulent airflow, or adverse takeoff and landing conditions exist. Significant shortcomings remain in areas such as GPS-denied positioning accuracy, attitude stability in strong winds, and reliability of takeoff and landing on water. In summary, while existing technologies recognize the potential of multi-drone collaborative rescue, there are still technological gaps in areas such as autonomous collaborative logic for heterogeneous drone models, dedicated adaptation mechanisms for extreme scenarios, and the degree of automation in response processes. A closed-loop collaborative rescue system covering the entire process of "reconnaissance—identification—survey—rescue—reconstruction" has not yet been formed. Summary of the Invention

[0003] To establish a closed-loop collaborative rescue system encompassing the entire process of "reconnaissance—identification—survey—rescue—reconstruction," this invention proposes a full-process collaborative rescue control method based on heterogeneous clusters. This method comprises a node cluster consisting of multiple types of unmanned aerial vehicles (UAVs). The collaborative rescue control method includes the following steps: S1: The first type of UAV performs wide-area reconnaissance, processes the collected perception data in real time on the airborne terminal to identify abnormal events, and generates a standardized mission package containing information on abnormal events; S2: In response to the generation of standardized task packages, the second type of drone is determined based on the preset model function relay rules, and the first task package is directly sent to the second type of drone, controlling it to fly to the location of the abnormal event to perform verification. S3: During the verification process of the second type of drone, if it is determined that there is communication obstruction or space access restriction, the third type of drone will be activated to enter the restricted area to perform detailed survey and communication relay, and transmit environmental data back. S4: Once the existence of a target to be rescued is confirmed through verification by a second-type or third-type drone, a fourth-type drone is dispatched to carry out rescue operations for that target. S5: During the fourth type of drone operation, update the status information of the target to be rescued and synchronize the updated information to other nodes in the cluster; S6: When a failure is detected in any node in the cluster, the tasks of the failed node are split and redistributed to other nodes in the cluster.

[0004] This invention significantly improves the automation level, operational efficiency, and system resilience of rescue response by constructing a heterogeneous cluster of multiple types of UAVs and realizing autonomous collaboration and dynamic reconfiguration throughout the entire process from reconnaissance to rescue.

[0005] Furthermore, in step S1, the real-time processing of the collected sensing data on the airborne end to identify abnormal events specifically involves: Various types of drones deploy heterogeneous lightweight inference models on the airborne end to process the raw data collected by their respective sensors in real time and generate preliminary identification results including anomaly type and confidence level. When the confidence level of the initial identification result is lower than the preset confidence threshold, or when the anomaly type belongs to the preset high-risk event type, it is determined that multi-source data fusion confirmation is required, and a collaborative reasoning mechanism based on federated learning is triggered. In the collaborative reasoning mechanism, the drones involved in the abnormal event only encrypt and upload the key parameters of the local model to the consensus node in the cluster. The consensus node performs secure aggregation to form a global judgment and synchronizes the global judgment result to the relevant nodes.

[0006] Furthermore, in step S2, the preset model function relay rule is a relay logic based on priority-driven and scenario-triggered logic, specifically as follows: When the abnormal event identified by the first type of drone is a suspected life sign, the specific model of the second type of drone that is suitable is automatically determined based on the environmental parameters of the environment in which the abnormal event occurs. The environmental parameters include at least one of the following: GPS signal strength, communication signal strength, obstacle density, and wind speed; If the environmental parameters monitored in real time reach the preset boundary parameters during the verification operation of the second type of drone, the next level drone will be triggered to perform the relay operation. The boundary parameters include at least one of the following: operating height threshold, space width threshold, and communication signal strength threshold for different models.

[0007] Furthermore, in step S3, activating the third type of UAV to enter the restricted area to perform refined reconnaissance and communication relay specifically includes: Construct a three-tiered communication architecture: "High-altitude hub layer - Low-altitude relay layer - Blind spot compensation layer". The high-altitude hub layer is constructed by a first type of UAV carrying a long-distance communication module at a set altitude to establish the main communication link; The low-altitude relay layer is constructed by a third type of UAV in the low-altitude airspace using a Mesh self-organizing network. Each third type of UAV acts as a mobile relay node, covering signal blind spots through multi-hop relay. The blind spot compensation layer includes a ground mobile node. When the communication signal attenuates to below a preset signal threshold, the cooperative communication link between the ground mobile node and the third type of UAV is activated. When any communication node goes offline, the system automatically identifies the communication quality of neighboring nodes within a set time and replans the relay path based on an improved routing algorithm.

[0008] Furthermore, in step S4, scheduling the fourth type of UAV to perform rescue operations against the target to be rescued also includes a step of dynamically adjusting the task priority based on the strength of vital signs and / or the degree of environmental hazard: Once the existence of a target to be rescued is confirmed, the intensity of the target's vital signs signal and the environmental hazard parameters of the area are obtained; When the intensity of vital signs signals is lower than the preset intensity of vital signs signals or the environmental hazard parameters are higher than the preset environmental hazard intensity, the rescue priority of the target to be rescued will be raised from the current level to a higher level, and drones with stronger carrying capacity and / or faster speed will be dispatched first to carry out rescue operations.

[0009] Furthermore, in step S4, dispatching the fourth type of drone to perform rescue operations targeting the object to be rescued specifically involves: If there are take-off and landing points in the target area that meet the size requirements and the ambient wind speed is lower than the take-off and landing wind speed, then the fourth type of drone will be dispatched to carry out rescue operations in the air delivery mode. If there are no open take-off and landing points in the target area or the ambient wind speed exceeds the take-off and landing wind speed, the fourth type of UAV will be dispatched to perform rescue operations in an air-ground collaborative mode. The air-ground collaborative mode includes: the fourth type of UAV carrying a ground mobile robot, airdropping the ground mobile robot over the target area, the ground mobile robot autonomously moving to the location of the target to be rescued to perform rescue operations, and transmitting ground perception data back to the UAV cluster through an ad hoc network.

[0010] Furthermore, in step S5, the updated information is synchronized to other nodes in the cluster using an event-triggered synchronization mechanism: Establish a unique digital profile for each target to be rescued, recording data from discovery to the completion of the rescue. When the state of a digital archive changes, a publish / subscribe pattern is used, pushing the update information only to nodes in the cluster that have subscribed to that type of state.

[0011] Furthermore, in step S6, splitting and redistributing the tasks of the faulty node to other nodes in the cluster specifically involves: Each node in the cluster monitors its own operating status in real time, and when a fault is detected, it diagnoses the type and severity of the fault. The current task of the faulty node is divided into several sub-tasks according to the task type and the capability weight of the remaining nodes; The remaining nodes bid for subtasks based on a distributed auction algorithm, using the "reward-cost" ratio as the evaluation metric. The reward is the priority of the subtask, and the cost is the flight distance and power consumption.

[0012] Furthermore, the various types of drones include: Type 1 drones are used for wide-area reconnaissance, Type 2 drones are used for vital sign identification, Type 3 drones are used for detailed reconnaissance in confined spaces, and Type 4 drones are used for rescue operations. The fourth type of drone further includes a first subtype for delivering supplies and a second subtype for air-ground coordinated rescue and casualty evacuation.

[0013] Compared with the prior art, the present invention has at least the following beneficial effects: (1) The present invention proposes a full-process collaborative rescue control method based on heterogeneous clusters, which upgrades the rescue response from the traditional manual serial process to a parallel closed-loop system driven by machine intelligence by constructing an autonomous collaborative mechanism for heterogeneous UAV clusters. (2) At the perception level, the UAV processes data and generates task packages in real time on the edge. Combined with federated learning, it realizes collaborative verification of multi-source information, which not only greatly shortens the response time from anomaly detection to task issuance, but also effectively reduces the communication bandwidth pressure in high-obstruction scenarios. (3) At the decision-making and execution level, based on the relay rules of the model function and the three-level communication architecture, the system can dynamically call the adapted model to enter narrow space or communication blind spot to perform reconnaissance and relay according to environmental changes, and extend the rescue operation to areas that are difficult for vehicles and personnel to reach through the air-ground coordination mode. (4) At the level of system resilience and information synchronization, an event triggering mechanism and a distributed task redistribution strategy are adopted. Even if some nodes fail, the cluster can still quickly reassemble the task chain and maintain information consistency among the nodes. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the steps of a full-process collaborative rescue control method based on heterogeneous clusters. Detailed Implementation

[0015] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings to further illustrate the technical solutions of the present invention. However, the present invention is not limited to these embodiments.

[0016] The core of the heterogeneous cluster-based full-process collaborative rescue control method proposed in this invention lies in constructing an intelligent cluster system composed of multiple types of UAVs, and realizing a closed-loop process from disaster reconnaissance, target identification, blind spot exploration to rescue operations and system self-healing through a complete set of collaborative control logic. In a typical embodiment of this invention, the heterogeneous cluster includes five functionally specialized UAV types: vertical takeoff and landing fixed-wing UAVs, infrared search UAVs, racing drones, multi-functional quadcopter UAVs, and air-to-ground integrated rescue composite-wing UAVs. Each plays the role of a "wide-area reconnaissance node," "precise identification node," "blind spot breakthrough node," "precise delivery node," and "comprehensive rescue node," forming an organic whole. Figure 1 As shown, the method mainly includes the following steps: S1: The first type of UAV performs wide-area reconnaissance, processes the collected perception data in real time on the airborne terminal to identify abnormal events, and generates a standardized mission package containing information on abnormal events; S2: In response to the generation of standardized task packages, the second type of drone is determined based on the preset model function relay rules, and the first task package is directly sent to the second type of drone, controlling it to fly to the location of the abnormal event to perform verification. S3: During the verification process of the second type of drone, if it is determined that there is communication obstruction or space access restriction, the third type of drone will be activated to enter the restricted area to perform detailed survey and communication relay, and transmit environmental data back. S4: Once the existence of a target to be rescued is confirmed through verification by a second-type or third-type drone, a fourth-type drone is dispatched to carry out rescue operations for that target. S5: During the fourth type of drone operation, update the status information of the target to be rescued and synchronize the updated information to other nodes in the cluster; S6: When a failure is detected in any node in the cluster, the tasks of the failed node are split and redistributed to other nodes in the cluster.

[0017] In a specific application scenario, such as an earthquake occurring in the core area of ​​a city, causing building collapses and creating a complex urban ruin environment, with severe damage to communication infrastructure, the intelligent cluster system using the method of this invention will be deployed to perform emergency rescue missions. First, during the system initialization phase, a drone nest pre-deployed near the emergency command center or deployed with the rescue team will automatically open. After the operator sets the mission area via the ground station, the drones in the nest will take off sequentially according to preset priority and mission requirements. In this embodiment, the first type of drone specifically refers to one or more vertical take-off and landing fixed-wing drones, characterized by long endurance and long range, responsible for conducting a wide-area disaster survey of the entire affected area. This type of drone is equipped with a hyperspectral camera and synthetic aperture radar. After climbing to a predetermined altitude of approximately 500 meters, it begins scanning according to the optimal route planned by the improved A* algorithm. During flight, the drone's onboard computing unit processes raw data from sensors in real time, and a lightweight YOLOv5s-Tiny target detection model is deployed internally. When the model identifies abnormal features of suspected collapsed buildings from hyperspectral imagery or detects surface deformation from synthetic aperture radar data, it immediately generates a preliminary abnormal event identification result at the edge and packages it into a standardized mission package. This task package includes not only the type of anomalous event (such as "suspected collapsed building" or "suspected vital signs") and its geographic coordinates (latitude, longitude, and elevation), but also the confidence level of the event and the priority of the preliminary assessment. This process is completed entirely on the UAV side, without the need to transmit raw video or radar data back, greatly saving communication bandwidth.

[0018] However, in complex disaster environments, relying solely on data from a single sensor of a single drone model is often insufficient for making a final judgment. For example, while a VTOL fixed-wing UAV can detect large-scale building collapses from an altitude of 500 meters, it cannot penetrate the rubble to confirm whether there are signs of life inside. To address this, this invention introduces a collaborative reasoning mechanism based on federated learning to solve the challenge of high-confidence judgments. When the confidence level of an abnormal event generated by a first-type UAV is lower than a preset threshold, or if the event type is predefined as a high-risk event (such as "vital signs"), it is determined that multi-source data fusion confirmation is required. At this time, the system triggers collaborative reasoning. The VTOL fixed-wing UAV does not send out the original remote sensing images. Instead, it encrypts the key model parameters (such as gradients or weight increments) generated by its local model (i.e., YOLOv5s-Tiny) after training, and then uploads them to a consensus node within the cluster via a self-organizing network communication link. This consensus node is an edge server at the ground station. The consensus node securely aggregates the model parameters received from multiple related UAVs to form a global judgment model that integrates features from multiple data sources. Subsequently, this global judgment result is synchronized back to all relevant nodes. This process not only ensures that the original data does not leave the endpoint, meeting data security requirements, but more importantly, it transforms complex data fusion tasks that cannot be completed due to communication interruptions in traditional centralized processing into a low-bandwidth, high-efficiency distributed computing process.

[0019] Once an abnormal event (e.g., a target identified as having a "high probability of harboring vital signs") is confirmed, the system enters the task execution phase. At this point, the system automatically determines and schedules the most suitable second-type drone to perform the verification task based on preset priority-driven and scenario-triggered drone function relay rules. In this embodiment, the second-type drone specifically refers to one or more infrared search drones. This relay rule is not a simple function list, but a complex set of decision-making logic. It comprehensively evaluates the environmental parameters of the current abnormal event. For example, based on data transmitted by the first-type drone, the system knows that the target area is located in the core of a high-rise building ruin, with a GPS signal strength below -130dBm and a high density of surrounding obstacles. Based on these parameters, the system automatically determines that a standard quadcopter drone would have difficulty entering the area, therefore prioritizing the scheduling of infrared search drones with strong anti-interference capabilities and infrared penetration capabilities. This task package is directly sent to the flight control system of the infrared search drone via the MAVLink protocol, eliminating the need for manual relay from the ground command center. After arriving at the target area, the infrared search drone hovers at a low altitude of 100-200 meters and activates its onboard high-resolution infrared thermal imager and life detection radar. The infrared thermal imager can penetrate some gaps in the rubble and smoke, while the life detection radar emits low-frequency electromagnetic waves that can penetrate concrete walls to detect subtle human movements and breathing. By fusing data from these two sensors, the drone can accurately locate the trapped individuals.

[0020] During the verification operation, the UAV monitors environmental parameters in real time. When the infrared search UAV detects a target trapped in an extremely narrow gap in the ruins (e.g., an entrance only 0.8 meters wide), its own size preventing it from entering, and the communication signal in the area attenuates to below -85dBm due to building obstruction, the system's environmental monitoring module determines that the current parameters have reached preset boundary parameters. At this point, the third type of UAV is automatically activated. In this embodiment, the third type of UAV specifically refers to one or more drones. These drones are extremely small, only 0.6 meters wide, allowing them to flexibly navigate through gaps as narrow as 0.8 meters. Upon receiving a "blind spot replacement request" from the infrared search UAV, it immediately takes off from its standby point and penetrates into the ruins. Simultaneously, it activates its onboard visual SLAM system, constructing a three-dimensional environmental map in real time using cameras and an IMU in the absence of GPS signals.

[0021] More importantly, the racing drones also act as mobile communication relays. The entire drone swarm constructs a three-tiered communication architecture: a high-altitude hub layer, a low-altitude relay layer, and a blind spot coverage layer. The high-altitude hub layer consists of vertical take-off and landing fixed-wing drones maintaining the main link at an altitude of 500 meters. The low-altitude relay layer consists of multiple racing drones forming a mesh self-organizing network at an altitude of about 100 meters, with each drone acting as a node in the network. Through multi-hop relays, signals are transmitted from one node to another, effectively covering blind spots for GPS and 4G signals. The blind spot coverage layer includes ground mobile nodes, such as robotic dogs. When a racing drone penetrates deep into the ruins and the communication signal attenuates to a preset threshold, the system automatically activates the collaborative communication link with the ground robotic dogs. The SDR-A3 Pro module carried by the robotic dogs penetrates the concrete walls and transmits data back to the racing drones at low altitude, which then relay the data back to the swarm, thus forming a complete "air-ground-air" communication link. When any communication node in the cluster goes offline for any reason, the system can automatically identify the communication quality of neighboring nodes through the improved AODV routing algorithm and quickly plan a new relay path to ensure uninterrupted communication.

[0022] After relay verification by infrared search drones and drones, the precise location and vital signs of the trapped personnel were confirmed. At this point, the system entered the rescue operation scheduling phase, that is, scheduling the fourth type of drone to perform the rescue operation. In this embodiment, the fourth type of drone is further subdivided into two subtypes: the first subtype is a multi-functional quadcopter drone, responsible for material delivery; the second subtype is an air-to-ground integrated rescue composite wing drone, responsible for air-to-ground coordinated rescue and casualty transfer. Which subtype to schedule depends on further assessment of the target area. The system will obtain environmental parameters of the target area, such as whether there are open take-off and landing points, current wind speed, etc. If there is a relatively flat area of ​​more than 2 square meters on the top of the target ruins, and the real-time wind speed is below level 6 (about 13.8 m / s), the system will prioritize scheduling the multi-functional quadcopter drone. This drone carries emergency supplies such as first aid kits, drinking water, and energy replenishment. It hovers 0.5 meters directly above the target point and uses visual guidance to accurately drop supplies, with the delivery error controlled within 0.3 meters, ensuring that the supplies are accurately delivered to the trapped personnel.

[0023] However, if the target area lacks open take-off and landing sites, or if the ambient wind speed exceeds the safe take-off and landing threshold for ordinary rotorcraft, the system will activate a more complex air-to-ground collaborative rescue mode. At this point, the second sub-type, a compound-wing UAV, is activated. It takes off from its pod and quickly reaches the target area. This UAV not only possesses vertical take-off and landing capabilities and high-speed cruising, but also carries a small tracked ground robot on its underside. Once the UAV reaches the target area, it uses its onboard LiDAR to scan the terrain below, searching for a suitable drop point. After confirming the drop point, the UAV releases the robot at an altitude of approximately 100 meters, with a landing error of no more than 0.5 meters. After landing, the robot's visual SLAM system immediately activates, constructing a local environmental map from a ground perspective and transmitting it back in real time via a self-organizing network. Simultaneously, the UAV hovers in the air, providing communication relay and lighting support for the robot. The robot then autonomously moves to the trapped personnel to perform more complex tasks, such as delivering gas masks, assisting the trapped personnel in donning basic protective equipment, or guiding them to a safe area. In this process, air and ground data achieve bidirectional driving: the drone's aerial perspective guides the robot's global path planning, while the robot's ground perspective provides the drone with more detailed local environmental information. For example, if an unstable wall is detected, the drone swarm will immediately adjust its flight path to avoid the risky area.

[0024] During rescue operations, this invention also introduces a dynamic priority adjustment mechanism to handle emergencies. For example, when an infrared search drone or a drone detects a trapped person's heart rate below 60 beats per minute using life detection radar, and the earthquake monitoring sensors in the area show an increasing aftershock frequency, the system automatically upgrades the rescue priority of that target from "high" to "extremely high." At this point, even if a simple supply delivery was originally planned to be performed by a quadcopter drone, the system will forcibly interrupt the plan and prioritize dispatching a compound-wing drone with a stronger payload capacity and higher speed to perform the casualty evacuation mission, striving to transfer the trapped person from the danger zone in the shortest possible time.

[0025] The entire rescue process operates on a closed-loop and transparent information flow. This invention establishes a unique digital profile for each discovered target awaiting rescue. This profile is created from the moment the target is first identified, recording the discovery time, precise coordinates, changes in vital signs, material delivery records, and transfer details—a complete record of the entire process. When any status in the profile changes, such as "materials delivered" or "personnel rescued," the system does not use periodic polling for synchronization but instead employs an efficient event-triggered synchronization mechanism. Specifically, communication within the cluster uses a "publish / subscribe" model. Each drone node can "subscribe" to specific types of information based on its mission needs. For example, the drone responsible for material delivery only cares about updates to "target location confirmed" and "material requirements," while the command drone responsible for overall situational awareness subscribes to all information. When an event occurs (such as "rescue completed"), the system immediately pushes the event and the updated profile data to all nodes that have subscribed to that type of information. This mechanism keeps data synchronization latency below 10 milliseconds, significantly reducing network load and ensuring a high degree of consistency in the understanding of the situation on-site among all relevant nodes. Meanwhile, all digital archive data is encrypted and stored using blockchain technology, ensuring the immutability and traceability of the information, and providing a basis for subsequent rescue review and decision optimization.

[0026] Finally, to address the unpredictability of rescue scenarios—for example, a drone performing a communication relay mission might be damaged by strong winds and collide with an obstacle, leading to a power system failure—the system immediately activates a damage adaptive reconstruction mechanism. First, the drone's own sensor network monitors the abnormal drop in motor speed in real time and broadcasts the fault information to the cluster. Then, the task management terminal in the cluster splits the faulty drone's task (e.g., "providing communication relay for area A"). The splitting rules consider the task type and the capability weights of the remaining nodes. For example, a communication relay task can be split into two sub-tasks: "high-altitude coverage" and "low-altitude coverage." Finally, the remaining healthy nodes autonomously bid based on a distributed auction algorithm. Each drone, as an independent agent, assesses its own capabilities (e.g., battery power, location, payload) and calculates the "benefit-cost" ratio for executing a sub-task (benefit being the priority of the sub-task, and cost being the flight distance and estimated battery consumption). All nodes complete the bidding within a short time, and the drone with the "closest location, sufficient battery power, and matching capabilities" ultimately wins the bid and takes over the faulty task.

[0027] In summary, the present invention proposes a full-process collaborative rescue control method based on heterogeneous clusters. By constructing an autonomous collaborative mechanism for heterogeneous UAV clusters, the rescue response is upgraded from a traditional manual serial process to a parallel closed-loop system driven by machine intelligence.

[0028] At the perception level, the drone processes data and generates task packages in real time on the edge. Combined with federated learning, it achieves collaborative verification of multi-source information, which not only significantly shortens the response time from anomaly detection to task issuance, but also effectively reduces the communication bandwidth pressure in high-obstruction scenarios.

[0029] At the decision-making and execution level, based on the relay rules of aircraft functions and the three-level communication architecture, the system can dynamically call on the appropriate aircraft to enter narrow spaces or communication blind spots to perform reconnaissance and relay according to environmental changes, and extend rescue operations to areas that are difficult for vehicles and personnel to reach through air-ground collaborative mode.

[0030] At the system resilience and information synchronization level, an event-triggered mechanism and a distributed task redistribution strategy are adopted. Even if some nodes fail, the cluster can still quickly reassemble the task chain and maintain information consistency among the nodes.

[0031] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0032] Furthermore, in this invention, descriptions involving terms such as "first," "second," and "a" are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0033] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0034] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

Claims

1. A method for end-to-end collaborative rescue control based on heterogeneous clusters, characterized in that, The collaborative rescue control method, comprising a node cluster consisting of multiple types of drones, includes the following steps: S1: The first type of UAV performs wide-area reconnaissance, processes the collected perception data in real time on the airborne terminal to identify abnormal events, and generates a standardized mission package containing information on abnormal events; S2: In response to the generation of standardized task packages, the second type of drone is determined based on the preset model function relay rules, and the first task package is directly sent to the second type of drone, controlling it to fly to the location of the abnormal event to perform verification. S3: During the verification process of the second type of drone, if it is determined that there is communication obstruction or space access restriction, the third type of drone will be activated to enter the restricted area to perform detailed survey and communication relay, and transmit environmental data back. S4: Once the existence of a target to be rescued is confirmed through verification by a second-type or third-type drone, a fourth-type drone is dispatched to carry out rescue operations for that target. S5: During the fourth type of drone operation, update the status information of the target to be rescued and synchronize the updated information to other nodes in the cluster; S6: When a failure is detected in any node in the cluster, the tasks of the failed node are split and redistributed to other nodes in the cluster.

2. The method for end-to-end collaborative rescue control based on heterogeneous clusters as described in claim 1, characterized in that, In step S1, the real-time processing of the collected sensing data on the airborne terminal to identify abnormal events specifically involves: Various types of drones deploy heterogeneous lightweight inference models on the airborne end to process the raw data collected by their respective sensors in real time and generate preliminary identification results including anomaly type and confidence level. When the confidence level of the initial identification result is lower than the preset confidence threshold, or when the anomaly type belongs to the preset high-risk event type, it is determined that multi-source data fusion confirmation is required, and a collaborative reasoning mechanism based on federated learning is triggered. In the collaborative reasoning mechanism, the drones involved in the abnormal event only encrypt and upload the key parameters of the local model to the consensus node in the cluster. The consensus node performs secure aggregation to form a global judgment and synchronizes the global judgment result to the relevant nodes.

3. The method for end-to-end collaborative rescue control based on heterogeneous clusters as described in claim 1, characterized in that, In step S2, the preset model function relay rule is a relay logic based on priority-driven and scenario-triggered logic, specifically as follows: When the abnormal event identified by the first type of drone is a suspected life sign, the specific model of the second type of drone that is suitable is automatically determined based on the environmental parameters of the environment in which the abnormal event occurs. The environmental parameters include at least one of the following: GPS signal strength, communication signal strength, obstacle density, and wind speed; If the environmental parameters monitored in real time reach the preset boundary parameters during the verification operation of the second type of drone, the next level drone will be triggered to perform the relay operation. The boundary parameters include at least one of the following: operating height threshold, space width threshold, and communication signal strength threshold for different models.

4. The method for end-to-end collaborative rescue control based on heterogeneous clusters as described in claim 1, characterized in that, In step S3, activating the third type of UAV to enter the restricted area to perform detailed reconnaissance and communication relay specifically includes: Construct a three-tiered communication architecture: "High-altitude hub layer - Low-altitude relay layer - Blind spot compensation layer". The high-altitude hub layer is constructed by a first type of UAV carrying a long-distance communication module at a set altitude to establish the main communication link; The low-altitude relay layer is constructed by a third type of UAV in the low-altitude airspace using a Mesh self-organizing network. Each third type of UAV acts as a mobile relay node, covering signal blind spots through multi-hop relay. The blind spot compensation layer includes a ground mobile node. When the communication signal attenuates to below a preset signal threshold, the cooperative communication link between the ground mobile node and the third type of UAV is activated. When any communication node goes offline, the system automatically identifies the communication quality of neighboring nodes within a set time and replans the relay path based on an improved routing algorithm.

5. The method for end-to-end collaborative rescue control based on heterogeneous clusters as described in claim 1, characterized in that, In step S4, dispatching the fourth type of UAV to perform rescue operations against the target to be rescued also includes the step of dynamically adjusting the task priority based on the strength of vital signs and / or the degree of environmental hazard: Once the existence of a target to be rescued is confirmed, the intensity of the target's vital signs signal and the environmental hazard parameters of the area are obtained; When the intensity of vital signs signals is lower than the preset intensity of vital signs signals or the environmental hazard parameters are higher than the preset environmental hazard intensity, the rescue priority of the target to be rescued will be raised from the current level to a higher level, and drones with stronger carrying capacity and / or faster speed will be dispatched first to carry out rescue operations.

6. The method for end-to-end collaborative rescue control based on heterogeneous clusters as described in claim 1, characterized in that, In step S4, dispatching the fourth type of drone to perform rescue operations targeting the object to be rescued specifically involves: If there are take-off and landing points in the target area that meet the size requirements and the ambient wind speed is lower than the take-off and landing wind speed, then the fourth type of drone will be dispatched to carry out rescue operations in the air delivery mode. If there are no open take-off and landing points in the target area or the ambient wind speed exceeds the take-off and landing wind speed, the fourth type of UAV will be dispatched to perform rescue operations in an air-ground collaborative mode. The air-ground collaborative mode includes: the fourth type of UAV carrying a ground mobile robot, airdropping the ground mobile robot over the target area, the ground mobile robot autonomously moving to the location of the target to be rescued to perform rescue operations, and transmitting ground perception data back to the UAV cluster through an ad hoc network.

7. The method for full-process collaborative rescue control based on heterogeneous clusters as described in claim 1, characterized in that, In step S5, the updated information is synchronized to other nodes in the cluster using an event-triggered synchronization mechanism: Establish a unique digital profile for each target to be rescued, recording data from discovery to the completion of the rescue. When the state of a digital archive changes, a publish / subscribe pattern is used, pushing the update information only to nodes in the cluster that have subscribed to that type of state.

8. The method for end-to-end collaborative rescue control based on heterogeneous clusters as described in claim 1, characterized in that, In step S6, splitting and redistributing the tasks of the faulty node to other nodes in the cluster specifically involves: Each node in the cluster monitors its own operating status in real time, and when a fault is detected, it diagnoses the type and severity of the fault. The current task of the faulty node is divided into several sub-tasks according to the task type and the capability weight of the remaining nodes; The remaining nodes bid for subtasks based on a distributed auction algorithm, using the "reward-cost" ratio as the evaluation metric. The reward is the priority of the subtask, and the cost is the flight distance and power consumption.

9. The method for full-process collaborative rescue control based on heterogeneous clusters as described in claim 1, characterized in that, The various types of drones include: Type 1 drones are used for wide-area reconnaissance, Type 2 drones are used for vital sign identification, Type 3 drones are used for detailed reconnaissance in confined spaces, and Type 4 drones are used for rescue operations. The fourth type of drone further includes a first subtype for delivering supplies and a second subtype for air-ground coordinated rescue and casualty evacuation.

Images