Unmanned aerial vehicle path planning and task allocation optimization system and method for disaster rescue detection

Abstract

The invention discloses an unmanned aerial vehicle path planning and task allocation optimization system and method for disaster rescue detection, and the system comprises an edge node layer, a mobile server layer and a cloud server layer, and is constructed to form a three-layer disaster monitoring system architecture of terminal perception-air-edge cooperation-cloud fusion. The edge node layer carries out real-time environment monitoring based on the intelligent monitoring terminal and triggers a task packet to the cloud server; the cloud server layer forms a corresponding action decision; and the mobile server layer allocates task unmanned aerial vehicles according to action decisions, and executes path planning and task allocation strategies of multiple unmanned aerial vehicles. According to the system and the method, efficient management of computing resources and energy resources of an edge node server, a mobile server and a cloud server can be realized, utilization of bandwidth resources among three layers of architectures is optimized, computation-intensive tasks can be processed more efficiently, and a higher resource utilization rate is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) mission execution for disaster rescue detection, and particularly to a UAV path planning and task allocation optimization system and method for disaster rescue detection. Background Art

[0002] Since the 21st century, the acceleration of global climate change and urbanization has led to frequent occurrence of various natural disasters (such as earthquakes, floods, forest fires, etc.), making disaster rescue and emergency response more urgent. After a disaster occurs, quickly and accurately obtaining on-site information, assessing the disaster situation, and real-time transmitting rescue data are the keys to effective decision-making and resource scheduling. Especially in extreme situations such as building collapses and flood encirclements after earthquakes, traditional rescue forces are often difficult to effectively implement life detection and material delivery within the golden 72 hours due to factors such as terrain barriers and infrastructure damage.

[0003] With the breakthrough of edge computing technology, intelligent UAVs equipped with edge computing nodes have shown significant advantages. By migrating computationally intensive tasks such as AI inference and image stitching from the cloud to the UAV side, the system response latency has been reduced from the second level to the millisecond level, which is crucial for real-time video analysis and feedback in the disaster area environment. In recent years, UAVs have been widely used in disaster rescue, environmental monitoring, search and rescue positioning, etc. due to their strong mobility, low cost, and ability to operate efficiently in complex environments. However, a single UAV system has certain limitations in terms of endurance, payload, and field of view coverage, and it is difficult to meet the rescue needs of large-scale disaster areas.

[0004] Therefore, a multi-UAV cooperative operation system has gradually become a trend in technological development. Through the cooperative flight of multiple UAVs, not only can full coverage monitoring of a large area be achieved, but also the real-time and accuracy of data collection can be improved. At the same time, the rapid development of edge computing has promoted the introduction of edge computing nodes into the UAV system to build a distributed data processing platform close to the disaster area. Edge computing nodes can instantaneously process a large amount of data collected by UAVs under low-latency conditions, providing real-time data support for on-site rescue, and significantly improving the rescue response speed and decision-making efficiency.

[0005] However, in a system composed of multiple UAVs and edge computing nodes, how to reasonably allocate tasks and plan flight paths to ensure the efficient coordination of various rescue tasks is still a technical problem to be solved urgently. Existing task offloading and path planning methods usually have difficulty taking into account the energy consumption, task timeliness, communication stability, and flight safety of UAVs. Especially in a complex and changeable environment such as a disaster site, it is easy to cause problems such as task duplication, resource waste, or path conflicts.

[0006] Currently, the research on UAV servers mainly focuses on the following two aspects: On the one hand, regarding the optimization of communication resources, existing research mainly focuses on providing data transmission and forwarding services for edge devices in areas lacking communication infrastructure or with limited communication conditions; using the powerful mobility of UAVs to collect effective data in environments that are difficult for humans to reach, as well as the interconnection between UAV swarms. Although these studies are helpful for the computing tasks of edge devices in certain specific fields, they still face problems such as insufficient computing resource scheduling and unstable UAV wireless communication connections. On the other hand, task offloading and path optimization based on UAV edge computing nodes. UAVs can be dispatched to task-intensive execution areas to directly process tasks on board computing resources, thereby reducing the communication requirements between edge nodes and cloud servers and effectively improving the response time of computing tasks. By optimizing the UAV flight path, offloading task ratio, and user scheduling variables in each time period, the goal is to minimize the target cost among all users within each time slot. However, existing research still lacks sufficient study on the selection of computing offloading destinations and the signal coverage ranges between UAVs and multiple ground terminals.

[0007] From this, the following problems can be seen. First, the signal coverage range of a single UAV is limited, which requires the UAV to move continuously to ensure that all edge devices can obtain offloading services. At the same time, the computing resources of a single UAV are limited and cannot meet the offloading requests of a large number of edge devices. In addition, existing methods usually adopt weighted optimization of time and energy consumption and cannot consider time, energy consumption, and other key optimization indicators simultaneously. Therefore, the present invention proposes a multi-UAV cooperation-based video analysis system aimed at solving the above problems. Specifically, the present invention first uses a UAV cluster to construct a distributed edge computing node cluster, which can not only provide a wider service range but also more sufficient computing resources, thereby simplifying the movement of UAVs and shortening the response time of edge devices. Secondly, a multi-objective optimization algorithm is adopted to consider multiple factors such as time, energy consumption, and task collection volume during the task processing process. This method can make more full use of system resources to achieve performance optimization, resource conservation, and efficient task processing. Based on the currently widely used multi-agent deep reinforcement learning (MADRL) technology, the present invention proposes a multi-objective optimization algorithm for minimizing energy consumption, minimizing task processing time, and maximizing task completion volume in a multi-UAV cooperation-based video analysis system for disaster scenarios. This algorithm can effectively generate task offloading and path planning schemes that meet the requirements of real-time, accuracy, and efficiency. Through this method, the system can significantly reduce the energy consumption of task processing in mobile edge video analysis, reduce the task processing time, and increase the task collection volume.

[0008] For example, the invention application with the application number 202110911316.6 discloses an edge swarm intelligence method based on UAV collaborative networking. Although this application mentions the application of UAV collaborative networking in forest fire smoke detection, its solution does not form a disaster monitoring system of edge nodes - mobile services - cloud services, and the rescue detection efficiency and response speed cannot be guaranteed.

[0009] Therefore, in reality, there is an urgent need for a path planning and task allocation optimization method for disaster rescue detection. Through intelligent algorithms and real-time scheduling strategies, this method gives full play to the advantages of the multi-UAV mobile server system to improve the overall rescue detection efficiency and response speed, providing more reliable and efficient technical support for disaster emergency rescue. Summary of the Invention

[0010] Aiming at the above existing problems, the purpose of the present invention is to provide a UAV path planning and task allocation optimization system and method for disaster rescue detection. By forming intelligent algorithms and real-time scheduling strategies in layers, it gives full play to the advantages of the multi-UAV mobile server system, improves the overall rescue detection efficiency and response speed, and provides more reliable and efficient technical support for disaster emergency rescue.

[0011] The embodiments of the present invention provide a UAV path planning and task allocation optimization system and method for disaster rescue detection.

[0012] First aspect: A UAV path planning and task allocation optimization system for disaster rescue detection, comprising:

[0013] The edge node layer, relying on intelligent monitoring terminals deployed in the disaster monitoring area, conducts real-time environmental monitoring. When an abnormal event is detected, it triggers a task package to the cloud server;

[0014] The cloud server layer, relying on the cloud server, forms corresponding action decisions based on the triggered task package and the geographical and status information of the intelligent monitoring terminals and multi-UAVs;

[0015] The mobile server layer, relying on multi-UAVs deployed at the edge of the disaster monitoring area, receives the action decisions, performs dynamic channel modeling to execute the path planning of the multi-UAVs, and hierarchically executes the task allocation strategy;

[0016] Among them, edge computing nodes are deployed on the intelligent monitoring terminals, mobile servers are deployed on the multi-UAVs, and multi-agent reinforcement learning models are deployed on the cloud server. The edge node layer, mobile server layer, and cloud server layer construct a three-layer disaster monitoring system architecture of "terminal perception - air-edge collaboration - cloud integration".

[0017] Optionally, the:

[0018] Dynamic channel modeling includes: determining the UAV signal coverage range and the ground intelligent monitoring terminals within this range based on the Euclidean distance, path loss, and Shannon's theorem in the three-dimensional space of the UAV, calculating the task bandwidth, and obtaining the corresponding task transmission data time and task calculation time based on the task bandwidth.

[0019] The hierarchical task allocation strategy includes: performing real-time mobile server inference on delay-sensitive tasks, and at the same time forwarding data-intensive tasks to the cloud server through the 5G wireless link.

[0020] Optionally, the cloud server layer includes:

[0021] A receiving query module: used to receive device status and task information and perform centralized processing;

[0022] A system device information collection module: generating action decisions based on the deployed multi-agent reinforcement learning model;

[0023] A task offloading and path planning module: formulating the best path planning and task allocation scheme based on the action decision and transmitting the generated scheme to the corresponding device;

[0024] A data volume receiving module: used to collect task execution data of the edge intelligent monitoring terminal and the UAV computing cluster;

[0025] An execution record module: used to generate a log file based on the status information of each device in the task and the task execution data.

[0026] Optionally, the mobile server layer includes:

[0027] A status sending module: used to transmit information such as the geographical location, battery power, and flight speed of the UAV itself to the cloud server;

[0028] A task processing module: used to process tasks on the mobile server carried by the UAV;

[0029] A task collection module: used for the UAV to collect task data at the task location;

[0030] A task sending module: used to establish a network connection with the cloud server and send the task data to the cloud server side according to the allocation decision;

[0031] A flight command processing module: used to parse the UAV path planning strategy issued by the cloud server to make the UAV move to the best hovering position.

[0032] Optionally, the functional modules of the edge node layer include:

[0033] A video stream monitoring module: automatically collecting target environment video stream data relying on the intelligent monitoring terminals deployed in the disaster monitoring area;

[0034] Task packaging module: used to periodically compress the collected data into a custom data format;

[0035] Signal broadcasting module: used to establish a link with nearby drones to prepare for task data transmission;

[0036] Disaster data task transmission module: used to transmit task data to drones.

[0037] Second aspect: A method for optimizing the path planning and task allocation of drones for disaster rescue detection, including:

[0038] S1. Relying on the intelligent monitoring terminals deployed within the disaster monitoring area, perform real-time environmental monitoring. When an abnormal event is detected, trigger a task package to the cloud server;

[0039] S2. Based on the task package and device information, rely on the cloud server to execute the path planning and task allocation algorithms for multiple drones, and form corresponding action decisions;

[0040] S3. According to the action decisions, allocate task drones, perform dynamic channel modeling to execute the path planning of multiple drones, and execute a hierarchical task allocation strategy.

[0041] Optionally, when relying on the cloud server to execute the path planning and task allocation algorithms for multiple drones and form corresponding action decisions in S2; based on the multi-agent reinforcement learning model deployed on the cloud server, according to the task time, bandwidth, and energy consumption, generate the path planning and task allocation of multiple drones, and form corresponding action decisions.

[0042] Optionally, the data processing flow of the cloud server layer includes:

[0043] S11. Receive the task package and the device status information of the edge node layer and the mobile server layer;

[0044] S12. Determine whether the multi-agent reinforcement learning model is started. If not, perform exception handling and then execute S12. If it has been started, execute S13;

[0045] S13. Rely on the multi-agent reinforcement learning model to form corresponding action decisions for the path flight and task allocation of multiple drones;

[0046] S14. Determine whether the intelligent monitoring terminal is connected to the drone. If not, perform exception handling and then execute S14. If it has been connected, execute S15;

[0047] S15. Execute the hierarchical task allocation strategy of the edge computing node, mobile server, and cloud server, receive task data, and generate a log file;

[0048] S16. Determine whether all tasks are completed. If all tasks are completed, end the task; if not, continue to execute S11.

[0049] Optionally, the data processing flow of the mobile server layer includes:

[0050] S21. Send device status information to the cloud server;

[0051] S22. Determine whether the multi-agent reinforcement learning model of the cloud server is started. If not, perform exception handling and then execute S21. If it has been started, execute S23;

[0052] S23. Receive the action decision sent by the cloud server and perform path flight and task allocation for multiple UAVs;

[0053] S24. Determine whether the task is offloaded to the mobile server. If the task is offloaded to the mobile server, execute S25; if the task is not offloaded to the mobile server, execute S26;

[0054] S25. Establish a process for the disaster video data analysis task, receive the data stream from the edge intelligent monitoring terminal, and send the result to the cloud after processing

[0055] S26. Receive the data stream from the edge intelligent monitoring terminal, establish a communication link with the cloud server, and directly transmit the data stream to the cloud server

[0056] S27. After the task is completed, send the task execution time and energy consumption data to the cloud server and record them in the log file;

[0057] S28. Determine whether all tasks are completed. If all tasks are completed, end the task; if not, continue to execute S21.

[0058] Optionally, the data processing flow of the edge node layer includes:

[0059] S31. Perform real-time environmental monitoring. When an abnormal event is detected, trigger a task package to the cloud server; periodically collect environmental information;

[0060] S32. Determine whether there are available UAVs nearby. If not, perform exception handling and then execute S32. If there are, execute S33;

[0061] S33. Establish a transmission process with the UAV and transmit the data to the UAV. After the task is transmitted, send an end signal to the UAV.

[0062] Third aspect: An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the program, the steps of the method provided in the second aspect are implemented.

[0063] Fourth aspect: A non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method provided in the second aspect are implemented.

[0064] Advantages of the present invention:

[0065] 1. The present invention proposes a flexible "cloud-edge-end" collaborative mobile edge video analysis system composed of edge computing nodes, mobile servers, and cloud servers. Based on the characteristics of drones, a task computing offloading strategy is studied to achieve efficient management of the computing resources and energy resources of edge nodes, mobile servers, and cloud servers, while optimizing the utilization of bandwidth resources between the three-layer architectures. This mobile server scheduling scheme can process compute-intensive tasks more efficiently and achieve higher resource utilization.

[0066] 2. The system of the present invention adopts a three-layer computing resource architecture of cloud, edge, and end. Among them, the cloud server platform is responsible for big data analysis, storage, and global scheduling; the edge computing nodes are close to the disaster area, enabling low-latency data processing and real-time response; the drones carrying mobile servers are responsible for on-site data collection and preliminary processing. Through the collaborative cooperation of resources at each layer, while maintaining the data processing ability, the system significantly reduces the response time, ensuring that accurate rescue information can be quickly obtained at the disaster site; at the same time, edge computing reduces network load and data transmission latency, improving the overall stability and robustness of the system, especially outstanding in emergency situations with complex network conditions.

[0067] 3. The collaborative work of multiple drones in the present invention can achieve comprehensive and rapid information collection and task execution within the area, avoiding blind spots and task bottlenecks brought by the limitations of single drones. The collaborative work of multiple drones can not only cover a wider range but also significantly improve the timeliness and accuracy of rescue tasks. In addition, collaborative operations can also reduce the load and energy consumption risk of individual drones, improving the task carrying capacity and fault tolerance of the overall system.

[0068] 4. The system of the present invention introduces a multi-agent reinforcement learning model for task offloading and path planning algorithms. During the task offloading and path planning process, multiple objectives such as energy consumption, time cost, and task data collection are considered simultaneously. This algorithm can autonomously learn and optimize in a complex and dynamic disaster site environment, find the best balance point among various objectives, thereby achieving efficient utilization of resources and optimization of task execution. While ensuring the timeliness of tasks, it can effectively control the energy consumption of drones, extend the flight time, and improve the task success rate and safety.

[0069] 5. The hierarchical task offloading adopted by the present invention utilizes edge computing nodes or mobile servers to deploy distributed computing nodes nearby, constructs an elastic resource scheduling system, and can achieve two technical breakthroughs: 1) The utilization rate of computing resources is increased to more than 82%; 2) The core network bandwidth requirement is reduced by about 60%, effectively alleviating the network congestion problem. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a schematic structural diagram of an unmanned aerial vehicle path planning and task allocation optimization system for disaster rescue detection according to the present invention;

[0071] Figure 2 It is a schematic flowchart of an unmanned aerial vehicle path planning and task allocation optimization method for disaster rescue detection according to the present invention;

[0072] Figure 3 It is a schematic structural diagram of the cloud server layer according to the present invention;

[0073] Figure 4 It is a schematic structural diagram of the mobile server layer according to the present invention;

[0074] Figure 5 It is a schematic structural diagram of the edge node layer according to the present invention;

[0075] Figure 6 It is a schematic flowchart of the operation of the cloud server layer according to the present invention;

[0076] Figure 7 It is a schematic flowchart of the operation of the mobile server layer according to the present invention;

[0077] Figure 8 It is a schematic flowchart of the operation of the edge node layer according to the present invention;

[0078] Figure 9 It is a schematic structural diagram of an electronic device according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0079] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar symbols represent the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation of the present invention.

[0080] When existing unmanned aerial vehicles are applied to disaster rescue detection tasks, there are: (1) The signal coverage range of a single unmanned aerial vehicle is limited, and the computing resources of a single unmanned aerial vehicle are limited, unable to meet the offloading task requests of a large number of edge devices; (2) Existing methods usually adopt time and energy consumption weighted optimization and cannot comprehensively consider time, energy consumption, and other key optimization indicators at the same time.

[0081] To address the above problems, the present invention provides an optimized system for UAV path planning and task allocation for disaster rescue detection.

[0082] Figure 1 FIG. 4 is a schematic structural diagram of the optimized system for UAV path planning and task allocation for disaster rescue detection provided by an embodiment of the present invention. The system includes: an edge node layer, a mobile server layer, a cloud server layer, etc.

[0083] The edge node layer relies on intelligent monitoring terminals deployed in the disaster monitoring area to conduct real-time environmental monitoring. When an abnormal event is detected, a task package is triggered to the cloud server.

[0084] The intelligent monitoring terminal can use intelligent cameras deployed in the disaster monitoring area as distributed sensing nodes. The intelligent monitoring terminal has the following core functions: 1) Real-time environmental monitoring: Through the embedded YOLOv5 target detection model, it continuously identifies disaster features such as fires and structural collapses, with an accuracy rate of up to 92.3%; 2) Task triggering mechanism: When an abnormal event is detected, it can automatically generate a calculation task package containing a timestamp, geographical coordinates, and video segments; 3) Near-field communication triggering: Adopting a mixed protocol of 5G and Wi-Fi 6, when the UAV enters a 50-meter communication radius, task uploading is started to avoid energy consumption waste caused by continuous broadcasting.

[0085] The functional modules of the edge node layer, as Figure 5 shown, include a video stream monitoring module, a task packaging module, a signal broadcasting module, a disaster data task transmission module, etc. Among them:

[0086] The video stream monitoring module automatically collects environmental video stream task data containing targets based on intelligent monitoring terminals deployed in the disaster monitoring area, and provides it for other servers to collect and calculate.

[0087] The task packaging module periodically compresses task data into a custom data format, which is convenient for network link transmission, reduces the risk of data loss, and is convenient for detecting and processing data loss.

[0088] The signal broadcasting module is used to continuously broadcast task processing signals to the surrounding area by the edge computing node after the video stream monitoring module and the task packaging module complete task processing. After nearby UAVs receive the signal, a connection is established to prepare for task transmission.

[0089] The disaster data task transmission module, after establishing a wireless network connection with nearby UAVs through the signal broadcasting module, transmits the task data to the mobile server of the UAV. To address the problem of data loss during transmission, this module will confirm data integrity after transmission.

[0090] The mobile server layer relies on multiple drones deployed at the edge of the disaster monitoring area, receives action decisions, performs dynamic channel modeling to execute the path planning of multiple drones, and hierarchically executes the task allocation strategy.

[0091] The multiple drones can consist of multiple drones equipped with NVIDIA Jetson NX computing modules and can complete the following: 1) Dynamic channel modeling: When the location of the intelligent monitoring terminal is determined, the bandwidth is calculated through the Euclidean distance, path loss, and Shannon's theorem in three-dimensional space. The signal coverage range of the drones and the edge computing nodes and cloud servers within this range are determined based on the calculated bandwidth, and the corresponding transmission data time and task execution time are obtained based on the bandwidth. 2) Hierarchical task processing: It can perform real-time local (mobile server) inference on latency-sensitive tasks (such as secondary collapse warnings), and at the same time, it can forward data-intensive tasks (such as full-region video stitching) to the cloud server through the 5G wireless link. The task allocation decision is completed based on the multi-agent reinforcement learning model of the cloud server.

[0092] The functional modules of the mobile server layer, such as Figure 4 shown in the figure, include a status sending module, a task processing module, a task collection module, a task sending module, and a flight command processing module, etc. Among them:

[0093] The status sending module is used to transmit information such as the geographical location, battery power, and flight speed of the drone itself to the cloud server for subsequent inference, and 5G wireless link communication is used between devices.

[0094] The task processing module relies on the mobile server carried by the drone to process the task data stream obtained by the task collection module and provides computing power support of approximately 117 TOPS.

[0095] The task collection module, according to the action strategy given by the cloud server, after the drone moves to the best task collection position assigned, activates the communication with multiple ground intelligent monitoring terminals through the wireless link.

[0096] The task sending module parses the task allocation decision of the drone, and after the task collection module completes the task data acquisition, it establishes a network connection with the cloud server and sends the task data to the cloud server side after processing according to the task allocation decision.

[0097] The flight command processing module parses the drone path planning strategy issued by the cloud server, and transmits instructions such as flight direction and distance to the flight controller through the drone (such as DJI OSDK-ROS) development platform to control the drone to execute the path planning flight, so that the drone moves to the best hovering position to perform task offloading and forwarding services.

[0098] The cloud server layer relies on cloud servers and, based on triggered tasks and a multi-agent reinforcement learning model, intelligently monitors the geographical distribution of terminals and the status information of multiple drones to form corresponding action decisions.

[0099] As a computing power reserve, the cloud server mainly undertakes two types of tasks: 1) Execute the path planning and task allocation algorithms for multiple drones on the cloud server. Based on the multi-agent reinforcement learning model, according to the geographical location distribution of intelligent monitoring terminals and the flight status information of multiple drones, corresponding action decisions are given; 2) Provide algorithmic services. The cloud server layer is configured with sufficient computing power devices and is powered by a wired power supply. Therefore, the system does not need to consider the power conservation of the cloud server, but only needs to consider the transmission time cost of the cloud server.

[0100] The functional modules of the cloud server layer, such as Figure 3 shown, include a receiving and querying module, a system device information collection module, a task offloading and path planning module, a data volume receiving module, an execution record module, etc. Among them:

[0101] The receiving and querying module is used to receive the status and task information of all devices in the system (including multiple drones and intelligent monitoring terminals, etc.) and centrally process them.

[0102] The system device information collection module deploys a multi-agent reinforcement learning model on the cloud server. The model depends on the status information of each device as input to generate action decisions. For this reason, the system device information collection module is used to sense surrounding devices and communicate with them through a wireless network to collect the required device information. In addition, this module is also responsible for preprocessing the collected information on the cloud server side and using the processing result as the model input.

[0103] The task offloading and path planning module, after the cloud server receives the offloading task request, the system takes the relevant device status information and task information as the basis, preprocesses them and inputs them into the deployed multi-agent reinforcement learning algorithm model, and can quickly formulate the best path planning and task allocation scheme for multiple drones. Finally, the generated scheme is transmitted to the corresponding device to execute task offloading.

[0104] The data volume receiving module, on the cloud server side, starts a task offloading process or a task execution result collection process for each device in units of devices. 1) Task offloading process: When the task of a certain device is assigned to the cloud server, the system allocates computing resources for this task offloading to the task calculation module and allocates communication resources to the data stream receiving module to receive and process the data stream uploaded by the device. 2) Task collection process: When the task of a certain device is executed on an edge computing node or a mobile server, the system only needs to collect the time and energy consumption data generated during the execution after the task is completed.

[0105] The execution record module displays the status information of each device before task offloading and shows the device status and task completion status during task execution. Meanwhile, the system receives the execution time and energy consumption data from the edge computing node and the mobile server, and records them together with the execution data of the cloud server in the log file.

[0106] The present invention relies on the multi-agent reinforcement learning model deployed by the cloud server and is implemented based on the multi-agent reinforcement learning model to form corresponding action decisions for the path planning and task allocation algorithms of multiple unmanned aerial vehicles. Specifically:

[0107] Based on the running time, data transmission bandwidth, and device energy consumption of devices (including multiple unmanned aerial vehicles and intelligent monitoring terminals, etc.), an action strategy including the path planning and task allocation of multiple unmanned aerial vehicles is generated.

[0108] This action strategy takes the positions of intelligent monitoring terminals in the disaster area, the amount of computation and data of task requests, and the status of each device in the three-layer architecture as input variables, and realizes the path planning and task allocation of multiple unmanned aerial vehicles with the help of a pre-trained multi-agent reinforcement learning model.

[0109] Specifically, the system transforms the problem into a multi-objective optimization problem with the goal of minimizing time, normalizing energy consumption, and maximizing the number of task executions under energy consumption constraints, and its output is the selection of the offloading destination for each task and the relative positions of the unmanned aerial vehicles.

[0110] To improve the evaluation accuracy, a method based on multi-agent reinforcement learning is adopted, and the Q-value calculation method is improved, replacing the original Q-value evaluation method with a dual-network scheme combining a state network and a value network.

[0111] In addition, by field-testing parameters such as the flight energy consumption, flight speed, task monitoring time, and power consumption of the unmanned aerial vehicles, the multi-agent reinforcement learning model is trained using real-scene data, so that the model can better adapt to practical applications.

[0112] After the cloud server obtains the status of each device and the trigger task package data, it inputs them into the model for reasoning, and finally generates the action strategies of each device. Finally, the intelligent monitoring terminals in the disaster area transmit the data to the unmanned aerial vehicles, and complete data processing and report the disaster target detection results on the cloud server.

[0113] Taking DJI drones as an example, DJI drones provide an OSDK (a toolkit for developing drone applications). Applications developed based on the OSDK can run on the drone's on-board computer (such as Manifold2). Developers can obtain various types of drone data by calling the specified interfaces in the OSDK, and then combine the designed software logic and algorithm framework to process and calculate the data, thereby generating control instructions to achieve automatic control and automated flight of the drone's actions. The automatic flight control program written using the OSDK can guide the drone to fly to the task offloading target area and execute disaster data analysis task offloading and data forwarding on the intelligent monitoring terminal node.

[0114] The present invention also provides a method for optimizing the path planning and task allocation of drones for disaster rescue detection, as Figure 2 shown, the method steps include:

[0115] S1. Relying on the intelligent monitoring terminals deployed within the disaster monitoring area, conduct real-time environmental monitoring. When an abnormal event is detected, trigger a task package to the cloud server;

[0116] S2. Based on the task package and device information, rely on the cloud server to execute the path planning and task allocation algorithms for multiple drones to form corresponding action decisions;

[0117] S3. According to the action decisions, allocate task drones, perform dynamic channel modeling to execute the path planning of multiple drones, and execute a hierarchical task allocation strategy.

[0118] In the present invention, edge computing nodes are deployed on the intelligent monitoring terminals, and mobile servers are deployed on multiple drones. The edge node layer, mobile server layer, and cloud server layer are constructed to form a three-layer disaster monitoring system architecture of "terminal perception - air-edge collaboration - cloud integration". The system uses discrete time slots T as the basic scheduling unit.

[0119] The system defines a fixed computing cycle of T seconds. In each cycle, first, the edge nodes within the wireless communication range of the mobile edge computing nodes send offloading request information and device status information to the cloud to achieve fine-grained scheduling cycles.

[0120] Among them, the cloud server layer performs data processing, as Figure 6 shown, the process includes:

[0121] S11. Receive the task package and the device status information of the edge node layer and the mobile server layer;

[0122] S12. Determine whether the multi-agent reinforcement learning model is started. If not, perform exception handling and then execute S12. If it has been started, execute S13;

[0123] S13. Based on the multi-agent reinforcement learning model, form corresponding action decisions for the multi-UAVs to perform path flight and task allocation;

[0124] S14. Determine whether the intelligent monitoring terminal is connected to the UAV. If not connected, perform exception handling and then execute S14. If connected, execute S15;

[0125] S15. Execute the hierarchical task allocation strategy of the edge computing node, mobile server, and cloud server, receive task data, and generate a log file;

[0126] S16. Determine whether all tasks are completed. If all tasks are completed, end the task. If not, continue to execute S11.

[0127] The mobile server layer performs data processing. As Figure 7 shown, the process includes:

[0128] S21. Send device status information to the cloud server;

[0129] S22. Determine whether the multi-agent reinforcement learning model of the cloud server is started. If not started, perform exception handling and then execute S21. If started, execute S23;

[0130] S23. Receive the action decisions sent by the cloud server for the multi-UAVs to perform path flight and task allocation;

[0131] S24. Determine whether the task is offloaded to the mobile server. If the task is offloaded to the mobile server, execute S25. If the task is not offloaded to the mobile server, execute S26;

[0132] S25. Establish a process for the disaster video data analysis task, receive the data stream from the edge intelligent monitoring terminal, and send the result to the cloud after processing

[0133] S26. Receive the data stream from the edge intelligent monitoring terminal, establish a communication link with the cloud server, and directly transmit the data stream to the cloud server

[0134] S27. After the task is completed, send the task execution time and energy consumption data to the cloud server and record them in the log file;

[0135] S28. Determine whether all tasks are completed. If all tasks are completed, end the task. If not, continue to execute S21.

[0136] The edge node layer performs data processing. As Figure 8 shown, the process includes:

[0137] S31. Conduct real-time environmental monitoring. When an abnormal event is detected, trigger a task package to the cloud server; periodically collect environmental information;

[0138] S32. Determine whether there is an available drone nearby. If not, perform exception handling and then execute S32. If so, execute S33;

[0139] S33. Establish a transmission process with the drone and transmit the data to the drone. After the task transmission is completed, send an end signal to the drone.

[0140] The method of the present invention performs long-term monitoring by deploying fixed intelligent monitoring terminals (such as cameras, infrared monitors, etc.) inside the disaster monitoring area to obtain disaster information. To ensure that the computing tasks can be fed back in a timely manner, the computing power and storage capacity of the fixed intelligent monitoring terminals are the primary considerations; on the other hand, since there is usually a lack of wired power connections in the monitoring area, the bandwidth limit of energy use is also an issue that needs to be considered.

[0141] The fixed intelligent monitoring terminal transmits the collected data to a cloud server with powerful computing power through manual or existing networks, and then performs centralized computing and processing. By offloading the computing tasks to the cloud server side or the mobile server side, the computing power and processing efficiency can be effectively improved. A monitoring process with offloaded task computing is formed.

[0142] The present invention also provides an electronic device, Figure 9 which is a schematic structural diagram of the electronic device provided by the embodiment of the present invention. As Figure 9 shown, the electronic device may include: a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete mutual communication through the communication bus. The processor can call the logical instructions in the memory, for example, to execute the following method:

[0143] S1. Rely on the intelligent monitoring terminals deployed inside the disaster monitoring area to perform real-time environmental monitoring. When an abnormal event is detected, trigger a task packet to the cloud server;

[0144] S2. Based on the task packet and device information, rely on the cloud server to execute the path planning and task allocation algorithms for multiple drones to form corresponding action decisions;

[0145] S3. According to the action decisions, allocate task drones, perform dynamic channel modeling to execute the path planning of multiple drones, and execute the hierarchical task allocation strategy.

[0146] In addition, when the logical instructions in the above-mentioned memory are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0147] Embodiments of the present invention also provide a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is used to execute the methods provided in the above-mentioned various embodiments, for example, including:

[0148] S1. Relying on the intelligent monitoring terminals deployed within the disaster monitoring area, conduct real-time environmental monitoring. When an abnormal event is detected, trigger a task package to the cloud server;

[0149] S2. Based on the task package and device information, rely on the cloud server to execute the path planning and task allocation algorithms for multiple unmanned aerial vehicles, and form corresponding action decisions;

[0150] S3. According to the action decisions, allocate task unmanned aerial vehicles, perform dynamic channel modeling to execute the path planning of multiple unmanned aerial vehicles, and execute a hierarchical task allocation strategy.

[0151] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative efforts.

[0152] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An optimized system for UAV path planning and task allocation in disaster rescue detection, characterized in that, Including: Edge node layer, relying on intelligent monitoring terminals deployed in the disaster monitoring area, conducts real-time environmental monitoring. When an abnormal event is detected, it triggers a task package to the cloud server; Cloud server layer, relying on the cloud server, forms corresponding action decisions based on the triggered task package and the geographical and status information of intelligent monitoring terminals and multiple unmanned aerial vehicles (UAVs); Mobile server layer, relying on multiple UAVs deployed at the edge of the disaster monitoring area, receives the action decisions, performs dynamic channel modeling to execute the path planning of multiple UAVs, and hierarchically executes the task allocation strategy; Among them, the intelligent monitoring terminals are deployed with edge computing nodes, the multiple UAVs are deployed with mobile servers, and the cloud server is deployed with a multi-agent reinforcement learning model. The edge node layer, mobile server layer, and cloud server layer construct a three-layer disaster monitoring system architecture of "terminal perception - air-edge collaboration - cloud integration".

2. The optimization system according to claim 1, wherein The following: Dynamic channel modeling includes: determining the UAV signal coverage range and the ground intelligent monitoring terminals within this range based on the Euclidean distance, path loss, and Shannon's theorem in the three-dimensional space of the UAV, calculating the task bandwidth, and obtaining the corresponding task transmission data time and task calculation time according to the task bandwidth; Hierarchical task allocation strategy includes: performing real-time mobile server inference on delay-sensitive tasks, and at the same time forwarding data-intensive tasks to the cloud server through the 5G wireless link.

3. The optimization system according to claim 1, characterized in that, The cloud server layer includes: Receiving query module: used to receive device status and task information and centrally process them; System device information collection module: generates action decisions based on the deployed multi-agent reinforcement learning model; Task offloading and path planning module: formulates the best path planning and task allocation scheme based on the action decisions and transmits the generated scheme to the corresponding devices; Data volume receiving module: used to collect the task execution data of edge intelligent monitoring terminals and UAV computing clusters; Execution record module: used to generate a log file based on the status information of each device in the task and the task execution data.

4. The optimization system according to claim 1, wherein The mobile server layer includes: Status sending module: used to transmit information such as the geographical location, battery power, and flight speed of the UAV itself to the cloud server; Task processing module: used to process tasks on the mobile server carried by the UAV; Task collection module: used for the UAV to collect task data at the task location; Task sending module: used to establish a network connection with the cloud server and send the task data to the cloud server side according to the allocation decision; Flight command processing module: used to parse the UAV path planning strategy issued by the cloud server to make the UAV move to the best hover position.

5. The optimization system according to claim 1, characterized in that The functional modules of the edge node layer include: Video stream monitoring module: automatically collects target environment video stream data relying on intelligent monitoring terminals deployed in the disaster monitoring area; Task packaging module: used to periodically compress the collected data into a custom data format; Signal broadcasting module: used to establish a link with nearby UAVs to prepare for task data transmission; Disaster data task transmission module: used to transmit task data to the UAV.

6. An optimization method for UAV path planning and task allocation for disaster relief detection based on the optimization system according to any one of claims 1 to 5, characterized in that, Including steps: S1. Rely on the intelligent monitoring terminals deployed within the disaster monitoring area to conduct real-time environmental monitoring. When an abnormal event is detected, trigger a task package to the cloud server; S2. Based on the task package and device information, rely on the cloud server to execute the path planning and task allocation algorithms for multiple drones, and form corresponding action decisions; S3. According to the action decisions, allocate task drones, perform dynamic channel modeling for the path planning of multiple drones, and execute the hierarchical task allocation strategy.

7. The optimization method according to claim 6, characterized in that When in S2, relying on the cloud server to execute the path planning and task allocation algorithms for multiple drones and form corresponding action decisions; Based on the multi-agent reinforcement learning model deployed on the cloud server, generate the path planning and task allocation for multiple drones according to the task time, bandwidth, and energy consumption, and form corresponding action decisions.

8. The optimization method according to claim 7, wherein The data processing flow of the cloud server layer includes: S11. Receive the task package and the device status information of the edge node layer and the mobile server layer; S12. Judge whether the multi-agent reinforcement learning model is started. If it is not started, perform exception handling and then execute S12. If it is started, execute S13; S13. Rely on the multi-agent reinforcement learning model to form corresponding action decisions for the path flight and task allocation of multiple drones; S14. Judge whether a connection is established between the intelligent monitoring terminal and the drone. If not, perform exception handling and then execute S14. If it is connected, execute S15; S15. Execute the hierarchical task allocation strategy of the edge computing node, mobile server, and cloud server, receive task data, and generate a log file; S16. Judge whether all tasks are completed. If all tasks are completed, end the task. If not, continue to execute S11.

9. The optimization method according to claim 7, characterized in that The data processing flow of the mobile server layer includes: S21. Send device status information to the cloud server; S22. Judge whether the multi-agent reinforcement learning model of the cloud server is started. If it is not started, perform exception handling and then execute S21. If it is started, execute S23; S23. Receive the action decisions sent by the cloud server and perform path flight and task allocation for multiple drones; S24. Judge whether the task is unloaded on the mobile server. If the task is unloaded on the mobile server, execute S25. If the task is not unloaded on the mobile server, execute S26; S25. Establish a process for the disaster video data analysis task, receive the data stream from the edge intelligent monitoring terminal, and send the processed result to the cloud; S26. Receive the data stream from the edge intelligent monitoring terminal, establish a communication link with the cloud server, and directly transmit the data stream to the cloud server; S27. After the task is completed, send the task execution time and energy consumption data to the cloud server and record them in the log file; S28. Judge whether all tasks are completed. If all tasks are completed, end the task. If not, continue to execute S21.

10. The optimization method according to claim 7, wherein The data processing flow of the edge node layer includes: S31. Conduct real-time environmental monitoring. When an abnormal event is detected, trigger a task package to the cloud server; periodically collect environmental information; S32. Judge whether there are available drones nearby. If not, perform exception handling and then execute S32. If there are, execute S33; S33. Establish a transmission process with the drone and transmit the data to the drone. After the task transmission is completed, send an end signal to the drone.

Images