A forest fire extinguishing cooperative scheduling system based on multi-unmanned aerial vehicle ad hoc network

Abstract

The present application belongs to the field of unmanned aerial vehicle cluster scheduling and wireless self-organizing network, and relates to a multi-unmanned aerial vehicle forest fire extinguishing cooperative scheduling system based on a Mesh network, comprising: a ground control station responsible for strategy generation, task allocation coordination, remote scheduling control; a plurality of unmanned aerial vehicles equipped with communication and sensing modules, which constitute a cellular Mesh network, realize detection, relay and fire extinguishing; an air Mesh communication network, which is dynamically constructed and contains a backbone layer and an access layer, supports distributed control and cooperative operation; a data sensing and sensing system, which fuses image, temperature, infrared and other information, and identifies fire; a task scheduling and energy management module, which supports task division, path planning and endurance guarantee. The system realizes an integrated closed-loop process of "sensing-communication-scheduling-fire extinguishing", and improves the response speed, resource scheduling efficiency and fire extinguishing accuracy of forest fire extinguishing.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of unmanned aerial vehicle cluster scheduling and wireless self-organizing network, and particularly relates to a forest fire extinguishing cooperative scheduling system based on multi-unmanned aerial vehicle self-organizing network. BACKGROUND

[0002] Unmanned aerial vehicles are widely used in forest fire extinguishing due to their high maneuverability and real-time monitoring capability. At present, the forest fire extinguishing technology based on multi-unmanned aerial vehicles is based on a simple star-shaped topology network, that is, a ground control station establishes a communication connection with each unmanned aerial vehicle, and according to the generated scheduling strategy, real-time tasks are sent to each unmanned aerial vehicle and real-time feedback of each unmanned aerial vehicle is received. However, the multi-unmanned aerial vehicle forest fire extinguishing technology based on the star-shaped topology network has the following defects: on the one hand, forest fires usually occur in complex terrain such as valleys and canyons where conventional signals are difficult to cover, which easily leads to poor communication between unmanned aerial vehicles and the ground control station, task allocation conflicts and untimely task adjustment, and it is difficult to realize multi-unmanned aerial vehicle cooperative operation; on the other hand, the ground control station needs to process the real-time feedback of each unmanned aerial vehicle and generate real-time tasks for each unmanned aerial vehicle, which puts higher requirements on the scientificity of the scheduling strategy, the complexity of the scheduling strategy and the computing capability of the ground control station, and high operation load will affect the response capability of the ground control station, thereby affecting the fire extinguishing efficiency. SUMMARY

[0003] To solve the above technical problems, the present application realizes the following technical scheme:

[0004] Proposed a kind of based on multi-unmanned aerial vehicle ad hoc network's forest fire extinguishing cooperative scheduling system, including: by the Mesh network of multiple unmanned aerial vehicles;Each unmanned aerial vehicle includes: real-time positioning module, for obtaining the real-time position of this unmanned aerial vehicle, the real-time position is broadcast to the rest of each unmanned aerial vehicle;Network division module, for dividing the Mesh network into multiple cellular subnets according to each real-time position, identify the cellular subnet where this unmanned aerial vehicle is located;State monitoring module, for monitoring the real-time state of this unmanned aerial vehicle, the real-time state is sent to the rest of each unmanned aerial vehicle in the cellular subnet;Network topology module, for establishing the hierarchical network topology of the cellular subnet according to each real-time state;Hierarchical network topology includes: backbone layer node and multiple access layer nodes;Selection trigger module, for triggering fire perception module when this unmanned aerial vehicle is access layer node, triggering fire perception module, task allocation module, task analysis module, scheduling request module and resource scheduling module when this unmanned aerial vehicle is backbone layer node;Fire perception module, for real-time perception of fire, the real-time fire is sent to the rest of each node in the cellular subnet;Task allocation module, for generating fire extinguishing tasks of each node in the cellular subnet according to each real-time fire and each real-time state, each fire extinguishing task is correspondingly sent to each node in the cellular subnet;Scheduling request module, for predicting whether the cellular subnet can complete the fire extinguishing task according to each real-time fire and each real-time state in the process of fire extinguishing, if yes, predicting resource surplus, if not, predicting resource demand, the resource demand is sent to each backbone layer node;Resource scheduling module, for matching the resource surplus of the cellular subnet with each resource demand, generating corresponding resource scheduling strategy according to the matching result.

[0005] Compared with the prior art, the present application has the following advantages and beneficial effects: based on the Mesh ad hoc network technology, a Mesh network with multiple unmanned aerial vehicles as nodes is established, realizing mutual communication between multiple unmanned aerial vehicles and solving the problem of poor communication between unmanned aerial vehicles and ground control station;On this basis, a double-layer network structure of "backbone layer + access layer" for each cellular subnet is established, task allocation for each node in the cellular subnet is performed by the backbone layer node, and resource cooperative scheduling between cellular subnets is performed, thereby avoiding ground centralized control, reducing the computing load of ground control station, and realizing efficient forest fire extinguishing. BRIEF DESCRIPTION OF DRAWINGS

[0006] The drawings described herein are used to provide further understanding of the embodiments of the present application, constitute a part of the present application, and do not constitute a limitation on the embodiments of the present application. In the drawings:

[0007] Figure 1 The Mesh network architecture diagram provided for embodiment 1 of the present application is provided. DETAILED DESCRIPTION

[0008] In order to make the objects, technical solutions and advantages of the present application clearer, further detailed description will be made to the present application in combination with embodiments, the illustrative embodiments and the description thereof are only used to explain the present application, and do not limit the present application. The embodiments described below are part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0009] In the following description, a large number of specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without these specific details. In other embodiments, well-known structures, materials or methods have not been specifically described in order to avoid obscuring the present application. The materials, instruments and reagents used in the following embodiments, etc. can be obtained from commercial channels unless otherwise specified. The technical means used in the embodiments are conventional means known to those skilled in the art unless otherwise specified.

[0010] In addition, the terms "first", "second" are only used for description purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0011] Embodiment 1: A multi-UAV ad hoc network-based forest fire extinguishing cooperative scheduling system is provided, which is composed of multiple UAVs Figure 1 The Mesh network is shown. Each UAV is configured with a wireless Mesh communication module, a real-time positioning module, a network division module, a state monitoring module, a network topology module, a selection trigger module, a fire perception module, a fire extinguishing device, a task allocation module, a task analysis module, a scheduling request module, a resource scheduling module, a support request module, a support return module, a cellular subnetwork adjustment module, a blind area monitoring module, a relay node screening module, a relay link construction module, a relay node adjustment module, a three-dimensional model creation module, a flight trajectory adjustment module, a directional antenna control module and an omnidirectional antenna control module. The purposes, internal operation modes and communication relationships with other modules of each of the above modules will be described in detail below.

[0012] 1. Wireless Mesh communication module

[0013] The wireless Mesh communication module is a communication device supporting a dynamic routing protocol (such as AODV, OLSR), and can realize a self-organizing network between devices. It connects multiple nodes through a wireless manner to form a mesh structure, supports multi-hop transmission and dynamic routing selection. In the embodiment, each unmanned aerial vehicle is configured with the wireless Mesh communication module, and a Mesh network is autonomously established by multiple unmanned aerial vehicles. In the Mesh network, data is relayed and transmitted through multiple intermediate node unmanned aerial vehicles, the coverage is expanded, and the reliability and invulnerability of the network are improved. In addition, the Mesh network can automatically discover and configure nodes, and when a certain node fails, other nodes can automatically reorganize the network structure to ensure the continuity of data transmission. Therefore, by configuring the wireless Mesh communication module on each unmanned aerial vehicle, a Mesh ad hoc network with multiple unmanned aerial vehicles as nodes is established, multi-path data transmission and redundant design are used to reduce the impact of single-point failure, and compared with a star-shaped topology network composed of multiple unmanned aerial vehicles, efficient data transmission and task collaboration can be realized, and the problem of poor communication in mountain valleys and gorges can be solved.

[0014] 2. Real-time positioning module

[0015] The real-time positioning module is used to obtain the real-time position of the unmanned aerial vehicle. The real-time positioning module suitable for unmanned aerial vehicles can be selected from: CUAV C-RTK 9Ps (supporting RTK and PPK, providing centimeter-level accuracy, suitable for applications requiring real-time high-precision navigation), and Xikangtong UM220-III (supporting multiple systems such as Beidou and GPS, and capable of achieving centimeter-level positioning accuracy, suitable for high-precision tasks), and E108-GN03BD (single Beidou positioning module, supporting Beidou II and Beidou III signals, with high integration, low power consumption, and small size). It should be understood that in addition to the above-mentioned real-time positioning modules listed in the embodiment, other devices capable of achieving high-precision real-time positioning can also be selected.

[0016] The real-time position of the unmanned aerial vehicle can be collected by configuring the real-time positioning module on the unmanned aerial vehicle. The real-time positioning module is connected with the Mesh communication module, and the real-time position of the unmanned aerial vehicle is broadcast through the Mesh communication module, so that each node unmanned aerial vehicle in the Mesh network can obtain the real-time position of each other unmanned aerial vehicle, and each unmanned aerial vehicle can know the real-time positional relationship between itself and each other unmanned aerial vehicle.

[0017] 3. Network division module

[0018] Based on the real-time positioning module and the Mesh communication module, the unmanned aerial vehicle can collect the real-time positions of all unmanned aerial vehicles in the Mesh network. The network division module divides the Mesh network into multiple cellular subnets through distributed clustering based on the collected real-time positions, and identifies the cellular subnet where the unmanned aerial vehicle is located. The forest fire extinguishing cooperative scheduling system provided in the embodiment is configured with the network division module on each unmanned aerial vehicle to build a hierarchical and distributed cooperative architecture, and solve the problems of high communication complexity, low task allocation efficiency and extensive resource management in the large-scale unmanned aerial vehicle fire extinguishing scene.

[0019] The network division module comprises a distributed clustering unit and a cellular identification unit.

[0020] (1) Distributed clustering unit

[0021] The distributed clustering unit is configured to perform distributed clustering on the real-time positions to form multiple clusters. Each cluster corresponds to a cellular subnet. Through distributed clustering, each unmanned aerial vehicle only needs to use local rules to gradually form a cluster based on the real-time position of the unmanned aerial vehicle and the real-time positions of other unmanned aerial vehicles, without global topology information. The clustering process is driven by a distributed algorithm between nodes, which avoids overloading or failure of a single node. In addition, distributed clustering can realize that when the unmanned aerial vehicles move, join or exit the cluster, each unmanned aerial vehicle recalculates the adjacency relationship through real-time interaction, triggers the update of the clustering structure, and when individual nodes are out of connection, the remaining nodes can automatically reorganize.

[0022] (2) Cellular identification unit

[0023] The cellular identification unit is configured to match the real-time position of the unmanned aerial vehicle with the real-time positions of all unmanned aerial vehicles in each cellular subnet, and determine the cellular subnet corresponding to the matched real-time position as the cellular subnet where the unmanned aerial vehicle is located. For example, the real-time position of the unmanned aerial vehicle is (x1, y1, z1), and it is found through matching that the unmanned aerial vehicle corresponding to the real-time position (x1, y1, z1) is located in the first cellular subnet, so the first cellular subnet is determined as the cellular subnet where the unmanned aerial vehicle is located.

[0024] 4. State monitoring module

[0025] The state monitoring module is configured to monitor the real-time state of the unmanned aerial vehicle. The real-time state includes the current remaining power, the current extinguishing agent amount, the communication radius, the current received signal strength indicator and the current computing power of the unmanned aerial vehicle. The state monitoring module is connected with the Mesh communication module, and sends the real-time state of the unmanned aerial vehicle to the remaining unmanned aerial vehicles in the cellular subnet through the Mesh communication module, so that each unmanned aerial vehicle in the cellular subnet can know the real-time state of each remaining unmanned aerial vehicle.

[0026] In the state monitoring module: (1) the battery management system (BMS) can be used to monitor the current remaining power of the unmanned aerial vehicle in real time. The BMS is integrated in the unmanned aerial vehicle battery, and real-time collection of battery voltage, current, and temperature data is performed, and the remaining power and health status are calculated. (2) The weight sensor can be used to monitor the current amount of fire extinguishing agent of the unmanned aerial vehicle in real time. The weight sensor directly weighs the container containing the fire extinguishing agent, and the remaining amount of fire extinguishing agent is obtained by deducting the weight of the empty container. (3) The NRF24L01 wireless module can be used to monitor the communication radius and received signal strength indicator in real time. (4) The performance counter (such as the PMU unit of ARM Cortex) can be used to monitor the computing power of the unmanned aerial vehicle in real time. The performance counter is built-in in the control chip of the unmanned aerial vehicle, and the top command or nodelet component is used to monitor the node computing power consumption.

[0027] 5. Network topology module

[0028] Based on the network division module and the state monitoring module, the network topology module establishes a hierarchical network topology for the cellular subnetwork according to the real-time state of the unmanned aerial vehicle and the real-time state of other unmanned aerial vehicles in the cellular subnetwork. The hierarchical network topology is composed of backbone layer nodes and multiple access layer nodes.

[0029] On the basis of the hierarchical and distributed collaborative architecture constructed by the network division module, the network topology module further establishes a two-layer network topology of "backbone layer nodes + access layer nodes" for each cellular subnetwork. The purpose is to:

[0030] Reduce communication complexity and improve collaboration efficiency. On the one hand, local communication is achieved - the access layer nodes in each cellular subnetwork only communicate directly with the backbone layer nodes in the cellular subnetwork. The backbone layer nodes act as "managers" of the cellular subnetwork to aggregate information and issue commands, avoiding the exponential complexity of "two-by-two communication" in star-shaped topology networks. On the other hand, the global communication overhead is reduced - cross-subnetwork collaboration is completed through communication between backbone layer nodes, and unnecessary information is not transmitted across subnetworks, improving communication bandwidth utilization.

[0031] Implement task domain management and adapt to local characteristics of the fire scene. On the one hand, regional task allocation is achieved - each cellular subnetwork corresponds to a local area of the fire scene, and the backbone layer nodes can dynamically adjust the task division of the unmanned aerial vehicles in the cellular subnetwork according to the local fire situation, avoiding the response delay of local requirements based on the unified scheduling form of the ground control station in star-shaped topology networks. On the other hand, strategy flexibility is achieved - different subnetworks can adopt different control strategies to adapt to the multi-dimensional fire extinguishing requirements of the fire scene.

[0032] To enhance system robustness and withstand localized failures, a decentralized architecture is established—a two-layer network topology without a global central controller. Each cellular subnet operates autonomously through backbone nodes. When communication in a cellular subnet is interrupted or a backbone node fails, other subnets can still independently execute tasks, avoiding system crashes due to "single points of failure." Furthermore, fault isolation is achieved—faults within a cellular subnet only affect tasks within that region, and backbone nodes can quickly redistribute tasks, minimizing the impact of the fault.

[0033] Enhance system scalability and adapt to dynamic changes in scale for flexible networking capabilities—newly added drones can be quickly added to neighboring subnets through distributed clustering algorithms without reconstructing the entire network topology, supporting the dynamic expansion of drone swarms in firefighting missions.

[0034] Furthermore, the backbone layer nodes and access layer nodes in each cellular subnet are selected through a weight calculation unit, a first filtering unit, and a second filtering unit:

[0035] (1) Weight calculation unit.

[0036] This is used to obtain the overall weight of this UAV in this cellular subnet based on the real-time status, and then send the overall weight to the other UAVs in this cellular subnet.

[0037] Based on the explanation of the status monitoring module, the real-time status of the UAV includes: current remaining battery power, current remaining fire extinguishing agent, communication radius, current received signal strength, and current computing power. Therefore, the overall weight of the UAV in this cellular subnet can be obtained through a weight calculation model. The weight calculation model is expressed as: Overall Weight = a 1× Remaining battery power + a 2×Extinguishing Agent Residual + a 3×communication radius+ a 4× Received signal strength index+ a 5× computing power. Among them, a 1 represents the weighting factor for the remaining battery power. a 2 is the weighting coefficient for the residual extinguishing agent. a 3 is the weighting coefficient for the communication radius. a 4 represents the weighting coefficient for the received signal strength index. a 5 represents the weighting coefficient for computing power.

[0038] It should be noted that: (1) before calculating the comprehensive weight, the remaining, fire extinguishing agent, communication radius, received signal strength index and computing power need to be normalized. The purpose is to convert the indicators of different dimensions and value ranges into unified dimensionless values, which are usually mapped to the interval [0, 1] or [-1, 1]. (2) The weight coefficients are set according to the importance of the real-time state of each type. If the real-time state of each type of task is in the same important position, the weight coefficients can be set to be the same, for example a 1 to a 5 are all 0.2.

[0039] (2) The first screening unit.

[0040] Used to screen out the unmanned aerial vehicles corresponding to the comprehensive weight greater than the threshold.

[0041] (3) The second screening unit.

[0042] Used to screen out the unmanned aerial vehicle closest to the geometric center or geometric center of the current cellular subnetwork as the backbone layer unmanned aerial vehicle according to the real-time position of the screened unmanned aerial vehicle, and screen out the remaining unmanned aerial vehicles of the current cellular subnetwork as access layer nodes.

[0043] It should be noted that: in combination with the role of the above backbone layer node, the unmanned aerial vehicle closest to the geometric center or geometric center of the current cellular subnetwork is taken as the backbone layer unmanned aerial vehicle, which is based on the requirements of the core targets of communication efficiency optimization, coverage balance, network stability and resource utilization maximization of the backbone layer node.

[0044] 6, selection trigger module

[0045] Used to trigger the fire situation awareness module to work when the current unmanned aerial vehicle is an access layer node, and trigger the fire situation awareness module, task allocation module, task analysis module, scheduling request module and resource scheduling module to work when the current unmanned aerial vehicle is a backbone layer node.

[0046] 7, fire situation awareness module.

[0047] Used to realize real-time awareness of fire situation.

[0048] The awareness module is connected with the mesh communication module, and the real-time fire situation is sent to the remaining nodes of the current cellular subnetwork through the mesh communication module. The fire situation awareness module described in the embodiment comprises a region division unit, a visible light camera, a thermal imaging infrared sensor, a temperature sensor, a feature extraction unit and an information processing unit.

[0049] (1) Region division unit

[0050] The fire area is divided into multiple management units according to the hierarchy of each cellular subnetwork, so that one cellular subnetwork corresponds to one management unit. Based on the above introduction of the function of the backbone layer node, the backbone layer node as the "manager" of the cellular subnetwork is responsible for aggregating the information of the cellular subnetwork and issuing task instructions to each node of the cellular subnetwork, as well as realizing cross-subnetwork communication with other cellular subnetworks. Among them, the task instructions issued by the backbone layer node include real-time fire sensing, so each UAV needs to clearly define its own sensing area. Through the area division unit, the entire fire area is divided into cellular subnetworks, so that the cellular subnetworks correspond one by one to the management units, realizing task definition.

[0051] (2) Visible light camera

[0052] The visible light camera is used to collect real-time visible light images of the management unit. The visible light camera can perform conventional flame and smoke recognition, and combined with a convolutional neural network (such as YOLOv5), it can quickly detect fire targets.

[0053] (3) Thermal infrared sensor

[0054] The thermal infrared sensor is used to collect real-time thermal images of the management unit. The thermal infrared sensor can capture hot spots, fire points, and smoke areas in real time, and can even discover fire sources at night or under thick smoke.

[0055] (4) Temperature sensor

[0056] The temperature sensor is used to collect real-time temperature distribution maps of the management unit. The temperature sensor can monitor the local environmental temperature rise, and through image data fusion, it can assist in judging the severity of the fire.

[0057] (5) Feature extraction unit

[0058] The feature extraction unit is used to fuse the visible light image, thermal image, and temperature distribution map using an image fusion algorithm, and use a convolutional neural network to extract features from the fused image to obtain feature information. Specifically, before image fusion, the visible light image, thermal image, and temperature distribution map are preprocessed to unify the image size, eliminate noise, and correct color and grayscale differences. The SIFT algorithm or SURF algorithm is used to extract feature points in the visible light image, thermal image, and temperature distribution map, respectively, and perform feature matching. The matched features are fused, and according to the fused feature information, the fused image is reconstructed by interpolation, transformation, and other methods.

[0059] (6) Information processing unit

[0060] The information processing unit is used to analyze the feature information using a deep learning model (such as the YOLOv5 model) to obtain real-time fire information. Real-time fire information includes fire area affected and flame spread speed.

[0061] For fire influence area calculation: the YOLOv5 model calculates the area of the bounding box according to the bounding box coordinates (top-left corner coordinates, top-right corner coordinates), optimizes the area within the bounding box through morphological operations (such as erosion, dilation), and then converts the pixel area into actual physical area in combination with the resolution of the image. Specifically, first, the YOLOv5 model performs inference on the input image, outputting the bounding box coordinates of the flame target in the form of top-left corner coordinates ( x 1, y 1) and bottom-right corner coordinates ( x 2, y 2), corresponding to the minimum enclosing region of the rectangular box. For the pixel area of a single flame target S px The calculation formula is: S px ( x 2- x 1)×( y 2- y 1), if there are multiple flame targets, the area of each target needs to be calculated and added up to obtain the total pixel area. Then, through morphological operations, noise (such as small areas of false detection) in the detection results is removed, holes in the flame area are filled, and the boundary is smoothed to improve the accuracy of area calculation, including: erosion operation - using a specified size of structural element (such as a 3x3 or 5x5 rectangular or elliptical kernel) to traverse the binary flame mask within the bounding box, deleting edge pixels and eliminating small particle noise; dilation operation - the shape and size of the structural element need to be adjusted according to the resolution and noise characteristics of the flame image, and the optimal value (such as a kernel size of 5x5 and an iteration number of 1-2 times) is usually determined through experiments. Finally, the pixel area is converted to physical area. The mapping relationship between the resolution of the image and the actual physical size needs to be calibrated in advance, i.e., the actual area (such as meters 2 / pixel) corresponding to a unit pixel. There are two cases: if the camera position is fixed, the pixel density , , where d 1 represents the actual physical width (in meters); d 2 represents the image width (in pixels), h 1 represents the actual physical height (in meters), h 2 represents the image height (in pixels). If perspective transformation (such as oblique shooting) is involved, the image needs to be corrected through a homography matrix (Homography Matrix) and then the physical area is calculated after converting to an aerial view. Finally, the total physical area .

[0062] For flame spread speed calculation: after the YOLOv5 model obtains continuous multiple frames of images, it respectively detects the flame in each frame of image to obtain the bounding box coordinates of multiple flame targets, determines the correspondence relationship of the flame targets between different frames, and indirectly calculates the flame spread speed by calculating the displacement distance of the flame centroid between different frames. Specifically, first, continuous frame flame detection and target tracking are performed - N consecutive frames of images (such as 30 frames per second, taking the last 5 frames) are sequentially detected by YOLOv5 to obtain the bounding box coordinates of the flame targets in each frame; a target tracking algorithm (such as IOU-based greedy tracking or DeepSORT) is used to associate the same flame target in different frames to ensure the accuracy of the cross-frame correspondence; the Hungarian Algorithm is used to match the intersection over union (IOU) of the current frame detection box and the previous frame tracking box to avoid target ID jumping. Then, the centroid is calculated - for the bounding box of a single flame target, the centroid ( c x , c y ) is taken as the center of the bounding box: c x =( x 1+ x 2) / 2, c y =( y 1+ y 2) / 2. Next, the frame displacement is calculated - assuming that the centroid of the target in the t th frame is ( c x t , c y t ) and the centroid of the target in the t +1th frame is ( c x t+1 , c y t+1 ), the pixel displacement distance is d px , , and the pixel displacement distance is converted into actual physical displacement d phys , , is the ratio factor between the pixel displacement distance and the actual physical displacement, which can be obtained through experiments or calibration process. Finally, the flame spread speed is calculated - assuming that the video frame rate is f (such as 30 FPS), then the time interval △ t =1 / f second between adjacent two frames, and the average speed v of the flame in the v= d phys / △ t (m / s).

[0063] 8、task allocation module

[0064] for generating fire extinguishing tasks of each node in the current cellular subnetwork according to each real-time fire situation and each real-time state, and sending each fire extinguishing task to each node in the current cellular subnetwork.

[0065] Under the unmanned aerial vehicle cellular subnetwork architecture, after the backbone layer node receives the real-time fire situation information and node state information of the current cellular subnetwork, through multi-dimensional data processing and intelligent algorithm, the dynamic and accurate allocation of the fire extinguishing tasks of the access layer node is realized, so as to improve the overall fire extinguishing efficiency and resource utilization. Specifically, the task allocation module comprises:

[0066] (1) fire situation analysis unit

[0067] for dividing the current management unit into multiple regions according to the real-time fire situation. The multiple regions include: a flame front region, a flame spread region and a cooling region. Among them, the flame front region is the region where the fire is currently most intense, the flame directly burns and is in a state of advancement; the flame spread region is located behind the flame front region, which is a region that may be affected by the flame; the cooling region is a region that has been extinguished or the fire has been effectively controlled, and the temperature is relatively low. For the flame front region: edge detection (such as Canny operator) can be performed on the fused image to extract the flame edge contour, and the boundary region of the front end of the flame is determined as the flame front region. For the flame spread region: machine learning algorithms (such as LSTM, GRU) can be used to learn the historical movement data of the flame front region to predict the future spread direction and range of the flame. For the cooling region: in the temperature distribution map, the region with a temperature lower than the set threshold and no obvious flame is marked as the cooling region.

[0068] (2) priority quantization unit

[0069] for quantifying the priority of each flame front region, the priority of each flame spread region and the priority of each cooling region by a fire situation priority quantization model. The quantification results include: the priority calculation result of each flame front region, the priority calculation result of each flame spread region and the priority calculation result of each cooling region;

[0070] The purpose of priority quantization is:

[0071] 1) to realize the reasonable allocation of resources: in the fire extinguishing task, the number of unmanned aerial vehicles, the carrying capacity of fire extinguishing agent, the flight time and other resources are limited. By quantifying the priority of different regions, the emergency degree and resource demand degree of each region can be clearly judged, so as to preferentially invest limited resources in regions with high priority.

[0072] 2) Promote multi-drone collaborative operation: After the priority of different areas is quantified, each drone can clearly understand the importance and objectives of its own mission, which helps to achieve better collaboration.

[0073] The expression for the fire priority quantification model is: (1); In equation (1), w 1 indicates the weight of the flame spread rate. w 2 indicates the weight of the area affected by the fire. V Indicates the speed of flame spread. V mas Indicates the maximum spread rate of the flame. S Indicates the area affected by the fire. S max Indicates the maximum fire area; w 1 and w 2. Calculated using the Analytic Hierarchy Process (AHP). For example, using a fire priority quantification model, fire priorities are divided into three levels: First Priority (… P ≥0.7), second priority (0.3 < P <0.7), third priority ( P (≤0.3). It should be noted that before calculating the overall weight, the flame spread rate, the area affected by the fire, and the maximum flame spread rate need to be normalized.

[0074] In addition, the quantification results are sent to the remaining backbone nodes through the mesh communication module to provide data support for subsequent priority matching.

[0075] (3) Priority matching unit

[0076] This is used to match the combined weights from each weight calculation unit with the priority levels from each priority quantization unit using a support vector machine, and to establish a priority matching table.

[0077] (4) Task allocation unit

[0078] This is used to assign firefighting tasks to each node in the current cellular subnet based on a priority matching table and a multi-objective optimization algorithm. Specifically:

[0079] First, define the core objective function for multi-objective optimization, including:

[0080] Fire extinguishing efficiency function: ,in, F eff This measures the area of ​​fire extinguished per unit of time. j This refers to the node numbering within the cellular subnet. k This indicates the number of nodes in the subnet. Sshed,j Indicates the first j The area of ​​the fire extinguished by each node t total This indicates the total firefighting time.

[0081] Resource utilization function: ,in, F res Indicates the overall resource utilization rate. m use,j Indicates the first j The amount of extinguishing agent already used at each node, m total,j Indicates the first j The total amount of extinguishing agent at each node, E left,j Indicates the first j The remaining power of each node E total,j Indicates the first j Total power of each node.

[0082] Task completion time function: F time =max{ t finish,1 , t finish,2 ,…, t finish,k},in, F time This represents the maximum value of all tasks completed. t finish,j Indicates the first j Task completion time for each node. j=1 , 2 , … , k .

[0083] The multi-objective optimization algorithm described in this embodiment is the NSGA-II algorithm. Based on the NSGA-II algorithm, fire suppression tasks are assigned to each node in this cellular subnet. Specific operations include: population initialization, fitness calculation, non-dominated sorting, crowding calculation, genetic operations, population update iteration, and scheme selection.

[0084] Initialize the population – randomly generate a certain number (e.g., 100) of task allocation schemes as the initial population, ensuring that each task is covered by at least one node (avoiding invalid solutions) and satisfying node energy and communication constraints.

[0085] Fitness calculation – Based on the three objective functions mentioned above, calculate the fitness value for each individual.

[0086] Non-dominated sorting - sort the individuals in the population according to non-dominated relationship, individuals in the same layer are non-dominated to each other, the smaller the layer number, the better the individual. For example, for any two individuals A and B in the population, if A is not inferior to B in all objectives, and at least one objective is better than B, then A dominates B. By layer-by-layer sorting, the population is divided into different non-dominated levels (level 1 is the best).

[0087] Crowding degree calculation - reflects the distribution density of individuals in the solution space, used to maintain population diversity. The crowding degree calculation is the sum of the distances between adjacent individuals in each objective dimension within the same non-dominated layer. The larger the crowding degree, the sparser the distribution of individuals around the individual, the better the diversity.

[0088] Genetic operation - including: (1) selection operation: adopt tournament selection method, randomly select k individuals from the population, select individuals with high non-dominated level and large crowding degree into the mating pool; (2) crossover operation: adopt two-point crossover or uniform crossover, randomly exchange task allocation fragments of selected chromosomes, while ensuring the legality of the solution after crossover (such as task coverage constraint); (3) mutation operation: flip the gene in the chromosome with a low probability (0→1 or 1→0), and repair the solution that violates the constraint (such as reassigning nodes to unallocated tasks).

[0089] Population update iteration - merge the parent and child populations, retain the top 100 individuals (based on non-dominated sorting and crowding degree), form a new generation of population; repeat the selection, crossover, and mutation operations until the maximum number of iterations is reached or the Pareto frontier converges.

[0090] Scheme selection - select the most suitable task allocation scheme for the current fire situation and node state from the Pareto frontier solution set. Specifically, from the Pareto frontier solution set, select the most suitable task allocation scheme for the current fire situation and node state, combine dynamic evaluation indicators, real-time data feedback, and decision-making strategies to convert multi-objective optimization results into executable specific schemes:

[0091] (1) Dynamic evaluation indicators

[0092] Including: fire emergency degree (flame spread speed, threat area population / asset value and burning material type), task timeliness (remaining time from the best fire extinguishing window period), spatial distribution (fire source location, spread direction), unmanned aerial vehicle resources (remaining power, range, fire extinguishing agent capacity, current position and distance from the fire source), historical task load (number of executed tasks, cumulative flight time), etc.

[0093] (2) Real-time data feedback

[0094] According to real-time data, weights are assigned to each target, and the multi-objective problem is converted into a single-objective optimization problem. For example, the weights assigned to each target are adjusted in real time according to preset expert field knowledge. For example, when the fire level is greater than or equal to 3, the weight of “fire extinguishing speed” is greater than 0.6, and the weight of “drone safety” is greater than 0.3. When there are people trapped in a certain area, the weight of “priority coverage of the area” is increased to 0.5. Real-time adjustment of the weight assigned to each target can dynamically correct the weight through sensor data (such as thermal imaging and wind speed). For example, when the wind speed increases, the weight of “quickly blocking the direction of fire spread” increases.

[0095] (3) Decision strategy

[0096] A fuzzy reasoning system is constructed, the input variables are the fire emergency level (low / medium / high) and the node health level (good / fair / poor), and the output is the weight coefficient of each target. For example, when the “fire emergency level” is high and the “node health level” is good, the weight of “task timeliness” is 0.7, and the weight of “energy consumption” is 0.3. When the “fire emergency level” is medium and the “node health level” is poor, the weight of “balanced load” is 0.6, and the weight of “completion rate” is 0.4.

[0097] (4) Scheme screening

[0098] Each scheme in the Pareto solution set is quantitatively scored using the evaluation index, and the scheme most suitable for the current state is selected. The method of sorting the ideal solution can be used for screening. The steps are as follows: first, standardize the index data of each scheme (such as normalized power and distance); then, determine the “positive ideal solution” (the optimal value of each index) and the “negative ideal solution” (the worst value of each index); finally, calculate the Euclidean distance between each scheme and the positive / negative ideal solution to obtain the comprehensive score: C i = d i - / ( d i - + d i + ), wherein C i represents the comprehensive score, C i the closer to 1, the better the scheme, d i - is the negative ideal solution, d i + is the positive ideal solution.

[0099] 9. Scheduling request module

[0100] For in the process of extinguishing fire, according to each real-time fire and each real-time state, predict whether the current sub-net can complete the task of extinguishing fire, if yes, predict the resource surplus, if not, predict the resource demand, and send the resource demand to each backbone layer node.

[0101] Wherein, the resource surplus refers to the resource remaining after the fire in the management unit is completely extinguished, including: surplus power and extinguishing agent surplus; the resource demand refers to the resource required to completely extinguish the fire in the management unit, including: power demand and extinguishing agent demand.

[0102] The scheduling request module comprises:

[0103] (1) Fire condition and state classification unit

[0104] For inputting the real-time fire condition obtained by each node in the current sub-net and the real-time state of each node into the random forest model, and outputting the classification result. The classification result includes: can complete the task of extinguishing fire and cannot complete the task of extinguishing fire.

[0105] (2) Surplus power prediction unit

[0106] For inputting the endurance time, power consumption rate, current flight speed, current flight height and current extinguishing agent amount of each node in the current sub-net into the time series prediction model (such as LSTM / GRU neural network), and outputting the surplus power of each node. Specifically, the essence of surplus power prediction is to capture the dynamic correlation between multiple features and power consumption through time series modeling, and to provide endurance capability prediction for task scheduling. Before prediction, first, the endurance time, power consumption rate, current flight speed, current flight height and current extinguishing agent amount are processed by data cleaning, normalization and feature extraction, and the processed data is input into the trained time series prediction model (taking long short-term memory network as an example). In the LSTM, the input layer receives sequence data in the time window, and the input dimension is Kx5 (K is the length of the time window, and 5 is the number of features); in the LSTM hidden layer, the number of layers is selected according to the complexity of the data (such as 1-3 layers), and each layer contains N hidden units (such as N=64); in the output layer, the output of the LSTM is mapped to a single value prediction through a fully connected layer (Dense), and the surplus power is output.

[0107] (3) Extinguishing agent surplus prediction unit

[0108] This method inputs the area of ​​the extinguishing zone already executed by each node in the cellular subnet, the average temperature of the management unit, the average extinguishing time of each node, and the flame spread rate into a regression model (such as Gradient Boosting Tree XGBoost / LightGBM), and outputs the remaining extinguishing agent for each node. Specifically, the Gradient Boosting Tree algorithm can capture complex nonlinear relationships between features (such as the synergistic effect of temperature and spread rate), is insensitive to outliers, is suitable for handling noise that may exist in sensor data, and can output the contribution of each feature to the prediction result, helping to understand the key driving factors of extinguishing agent consumption. In practical tasks, an appropriate Gradient Boosting Tree is selected based on the amount of data, accuracy requirements, scenario complexity, and real-time requirements. This embodiment selects the LightGBM model, which is suitable for scenarios with large amounts of data and high real-time requirements (such as dynamic updates of the cellular subnet). Similarly, before prediction, the area of ​​the extinguishing zone, the average temperature of the management unit, the average extinguishing time of each node, and the flame spread rate are cleaned, normalized, and feature extracted. The processed data is then input into the trained LightGBM model. In the LightGBM model, the model outputs the excess fire extinguishing agent by weighted summation of the prediction results from multiple trees.

[0109] (4) Electricity demand forecasting unit

[0110] This is used to establish a power demand prediction model. The mission flight distance, average flight speed, total firefighting operation time, average power, and remaining flight time of each node are input into the power demand prediction model. The model outputs the power demand of this cellular subnet and sends the power demand to the other backbone nodes in the form of a request signal.

[0111] The expression for the electricity demand forecasting model is: (3); In equation (3), j This refers to the node numbering within the cellular subnet. k This indicates the number of nodes in the subnet. T req This indicates the amount of electricity required. P j Indicates the first j Average power of each node L j Indicates the first j The mission flight distance of each node V j Indicates the first j The average flight speed of each node T total,j Indicates the first j Total firefighting operation time for each node, T already,j Indicates the first j The remaining battery life of each node.

[0112] a fire extinguishing agent demand prediction unit for establishing a fire extinguishing agent demand prediction model, inputting the area of each region in the management unit, the fire extinguishing agent dosage per unit area in each region, the priority calculation result of each region in the management unit and the current fire extinguishing agent residual amount of each node into the fire extinguishing agent demand prediction model, and outputting the fire extinguishing agent demand of the current cellular subnetwork, and sending the fire extinguishing agent demand to the remaining backbone layer nodes in the form of a request signal.

[0113] The expression of the fire extinguishing agent demand prediction model is: (2); in formula (2), Q represents the fire extinguishing agent demand, i is the region number in the management unit, m represents the number of regions divided in the management unit, S i represents the area of the i th region, q i represents the fire extinguishing agent dosage per unit area in the i th region, represents the priority calculation result of the i th region, C j represents the current fire extinguishing agent residual amount of the j th node.

[0114] 10. A resource scheduling module

[0115] for matching the resource surplus of the current cellular subnetwork with each resource demand, and generating a corresponding resource scheduling strategy according to the matching result.

[0116] Specifically, the resource scheduling module comprises:

[0117] (1) a demand analysis unit

[0118] for determining whether the resource surplus of the current cellular subnetwork simultaneously satisfies the conditions of surplus power > power demand and fire extinguishing agent surplus > fire extinguishing agent demand, if yes, triggering the demand response unit and the timing analysis unit to work, otherwise, no demand response is performed.

[0119] (2) a demand response unit

[0120] for generating a demand response signal, and sending the demand response signal and the generation time of the demand response signal to the remaining backbone layer nodes.

[0121] (3) a timing analysis unit

[0122] The generation time of each demand response signal is sorted in a chronological order to form a generation time queue, and it is determined whether the generation time of the demand response signal of the node is at the head of the queue, if yes, the resource scheduling unit is triggered to work, otherwise, the next request signal is received.

[0123] (4) Resource scheduling unit

[0124] The resource scheduling strategy is generated according to the resource demand amount, and the resource scheduling strategy includes: the number of unmanned aerial vehicles.

[0125] 11, Support request module

[0126] When condition A is met, a support request signal is sent to the ground control station according to the resource demand amount. Condition A: the classification result is that the fire extinguishing task cannot be completed, and no demand response signal is received.

[0127] 12, Support return module

[0128] When condition B or condition C is met, after the task is completed, a control signal is sent to each node in the cellular subnetwork to return to the ground control station. Condition B: the classification result is that the fire extinguishing task can be completed, and no request signal is received during the task execution. Condition C: the classification result is that the fire extinguishing task can be completed, and no demand response is made during the task execution.

[0129] 13, Cellular subnetwork adjustment module

[0130] The newly added unmanned aerial vehicle is used as an access layer node of the cellular subnetwork, and the hierarchical structure of the cellular subnetwork is updated. When an unmanned aerial vehicle exits the cellular subnetwork, the hierarchical structure of the cellular subnetwork is updated.

[0131] Further, the selection trigger module is further used to trigger the cellular structure adjustment module to work when the unmanned aerial vehicle is a backbone layer node and an unmanned aerial vehicle joins / leaves the cellular subnetwork.

[0132] 14, Blind area monitoring module

[0133] The received signal strength indicator of each communication link in the current cellular subnetwork is monitored in real time, and the communication links with received signal strength indicators less than a threshold value are screened out. The area within the management unit where each screened-out communication link is projected is marked as a communication blind area edge, and the communication blind area edge is sent to the remaining backbone layers. The received signal strength indicator (RSSI) is a parameter for measuring wireless signal strength. The receiver of the wireless Mesh communication module determines the RSSI value by measuring the power of the signal. Further, the communication blind area edge refers to the boundary of an area in a communication system where effective signals cannot be received or the signal strength is extremely low due to limitations of signal propagation, resulting in communication failure. Natural and man-made obstacles such as mountains, buildings, and tunnels can block signals, forming shadow areas or blind areas.

[0134] 15, relay node screening module

[0135] The relay nodes in the current cellular subnetwork are screened out according to the hierarchical structure of the current cellular subnetwork and the remaining power of each node. Specifically:

[0136] (1) Construct a node model and weight graph of the cellular subnetwork

[0137] Construct a weighted undirected graph G ( V , E ). Among them, V represents the node set of the current cellular subnetwork, E represents the communication link between nodes, and the weight on the communication link is , wherein, represents the comprehensive weight of the communication link, represents the weight coefficient of the distance, and distance represents the distance between the two nodes of the communication link, represents the weight coefficient of the remaining circuit, energy represents the remaining power of the node. The closer the distance, the lower the communication energy consumption, and the higher the remaining power of the node, which is more suitable as a relay. In this embodiment, the weight of the remaining power is set to the inverse of the remaining power.

[0138] (2) Determine the candidate relay node set

[0139] Preliminarily screen out nodes that meet the basic conditions from the cellular subnetwork as the candidate set C ∈ V The conditions include: the remaining power is higher than a threshold value (such as the remaining power ≥ 50% of the initial power); the communication coverage range can connect at least two other nodes (to avoid isolated nodes); and the geographical location is uniformly distributed to cover potential communication blind areas.

[0140] (3) Application of minimum spanning tree algorithm to generate relay link

[0141] The minimum spanning tree is generated by Kruskal algorithm on the weighted graph G First, the edge set E is sorted by weight from small to large; then, the Union-Find structure is created, and each node is initialized as an independent set. Finally, each edge is traversed in the order of sorting, and if the two nodes corresponding to the current query edge belong to different sets, the sets are merged, and the single signature query edge is added to the spanning tree until all nodes are connected.

[0142] (4) Extracting relay nodes from relay link

[0143] After generating the minimum spanning tree, mark the non-leaf nodes in the minimum spanning tree as relay nodes.

[0144] 16, Relay link construction module

[0145] Used to construct relay links containing relay nodes through topology-aware algorithms.

[0146] Based on the minimum spanning tree generated in the relay node screening module, a path from the root node to the node is a relay link containing relay nodes.

[0147] 17, Three-dimensional model creation module

[0148] Used to create a three-dimensional model of the communication blind area of the current cell subnet according to the edge of the communication blind area and the fused image.

[0149] (1) Establish a spatial coordinate system

[0150] UTM projection or local coordinate system is adopted to ensure that the model accurately corresponds to the actual geographical position, and the Z-axis is defined based on the mean sea level (MSL) or ground elevation.

[0151] (2) Three-dimensionalization of blind area boundary

[0152] First, combine the two-dimensional blind area edge points with the terrain elevation data to assign corresponding Z coordinates to each boundary point; then, use B-spline curve or polynomial fitting algorithm to smooth the discrete boundary points and generate continuous three-dimensional boundary curves.

[0153] (3) Generate blind area region

[0154] First, apply Delaunay triangulation to the region enclosed by the three-dimensional boundary curve to generate a triangular mesh covering the blind area; then, according to the terrain undulation and obstacle distribution, assign height values to each triangular patch.

[0155] (4) Establishing a three-dimensional model

[0156] Convert the triangular mesh model into a voxel representation, each voxel is marked as "blind area" or "non-blind area", and a three-dimensional reconstruction algorithm (point cloud-based reconstruction algorithm) is used to establish a three-dimensional model of the communication blind area.

[0157] 18. Flight trajectory adjustment module

[0158] For dynamically adjusting the flight trajectory of each unmanned aerial vehicle in the current cellular subnetwork according to the terrain elevation and the three-dimensional model of the communication blind area, the NSGA-II algorithm is used. Refer to the method principle and process of assigning fire extinguishing tasks to each node in the current cellular subnetwork based on the NSGA-II algorithm.

[0159] The selection trigger module is also used to trigger the three-dimensional model creation module and the flight trajectory adjustment module when the current unmanned aerial vehicle is a backbone layer node and receives the edge of the communication blind area.

[0160] 19. Directional antenna control module

[0161] For starting the directional antenna to work and adjusting the pointing direction of the directional antenna.

[0162] 20. Omnidirectional antenna control module

[0163] For starting the omnidirectional antenna to work and adjusting the pointing direction of the directional antenna.

[0164] The selection trigger module is also used to trigger the directional antenna control module when the current unmanned aerial vehicle is a backbone layer node, and trigger the omnidirectional antenna control module when the current unmanned aerial vehicle is an access layer node.

[0165] The above specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

[0166] It should be noted that the structure, proportion, size, etc. shown in the drawings attached to the present specification are only used to cooperate with the content disclosed in the specification, so that those skilled in the art can understand and read, and are not used to limit the conditions that the present application can be implemented, so they do not have technical significance. Any modification of the structure, change of the proportion relationship or adjustment of the size, without affecting the effect and purpose that the present application can produce, should still fall within the scope of the technical content disclosed by the present application. At the same time, the terms such as "upper", "lower", "left", "right", "middle" and the like in the present specification are only for the convenience of clear description, and are not used to limit the scope of the present application, and the change or adjustment of the relative relationship is also considered as the implementation of the present application without substantial changes in technical content.

Claims

1. A forest fire fighting collaborative dispatch system based on a multi-UAV self-organizing network, characterized in that, include: A mesh network consisting of multiple drones; Each drone includes: The real-time positioning module is used to obtain the real-time location of this drone and broadcast the real-time location to other drones. The network segmentation module is used to divide the Mesh network into multiple cellular subnets based on each real-time location and identify the cellular subnet where the drone is located; The status monitoring module is used to monitor the real-time status of this UAV and send the real-time status to the other UAVs in this cellular subnet; the real-time status includes: remaining battery power, fire extinguishing agent balance, communication radius, received signal strength, and computing power; The network topology module is used to establish a hierarchical network topology for this cellular subnet based on various real-time states; the hierarchical network topology includes: backbone layer nodes and multiple access layer nodes; The network topology module includes: The weight calculation unit is used to obtain the comprehensive weight of this UAV in this cellular subnet based on the real-time status, and send the comprehensive weight to the other UAVs in this cellular subnet; The first filtering unit is used to filter out drones corresponding to multiple comprehensive weights that are greater than the threshold. The second filtering unit is used to select the drone closest to the geometric center or geometric centroid of the cellular subnet as the backbone drone based on the real-time location of the selected drones, and to use the remaining drones of the cellular subnet as access layer nodes. The selected trigger module is used to trigger the fire perception module when the UAV is an access layer node, and to trigger the fire perception module, task allocation module, task parsing module, scheduling request module and resource scheduling module when the UAV is a backbone layer node. The fire sensing module is used to sense fires in real time and send the real-time fire information to the other nodes in this cellular subnet. The task allocation module is used to generate fire-fighting tasks for each node in the cellular subnet based on the real-time fire situation and status, and send each fire-fighting task to the corresponding node in the cellular subnet. The scheduling request module is used to predict whether the current cellular subnet can complete the firefighting task based on the real-time fire situation and status during the firefighting process. If so, it predicts the resource surplus; if not, it predicts the resource demand and sends the resource demand to each backbone layer node. The resource scheduling module is used to match the resource surplus of this cellular subnet with the resource demand of each resource, and generate corresponding resource scheduling strategies based on the matching results.

2. A forest fire fighting collaborative dispatch system based on a multi-UAV self-organizing network as described in claim 1, characterized in that, The network partitioning module includes: Distributed clustering units are used to perform distributed clustering of each real-time location to obtain multiple clusters; one cluster corresponds to one cellular subnet. The cellular identification unit is used to match the real-time location of this UAV with the real-time location of each UAV in each cellular subnet, and to determine the cellular subnet corresponding to the successfully matched real-time location as the cellular subnet where this UAV is located.

3. A forest fire fighting collaborative dispatch system based on a multi-UAV self-organizing network as described in claim 1 or 2, characterized in that, The fire perception module includes: The area division unit is used to divide the fire area into multiple management units according to the hierarchical structure of each cellular subnet; one cellular subnet corresponds to one management unit. A visible light camera is used to acquire visible light images of this management unit in real time. A thermal imaging infrared sensor is used to acquire thermal images of this management unit in real time. Temperature sensor, used to collect temperature distribution map of this management unit in real time; The feature extraction unit is used to fuse visible light images, thermal images, and temperature distribution maps using an image fusion algorithm, and to extract features from the fused image using a convolutional neural network to obtain feature information. The information processing unit is used to analyze feature information using a deep learning model to obtain real-time fire information, including the fire's affected area and the rate of flame spread.

4. A forest fire fighting collaborative dispatch system based on a multi-UAV self-organizing network as described in claim 3, characterized in that, The task allocation module includes: The fire analysis unit is used to divide this management unit into multiple zones based on the real-time fire situation; these zones include: the flame front zone, the flame spread zone, and the cooling zone. The priority quantization unit is used to quantify the priority of each flame front zone, each flame spread zone, and each cooling zone using the fire priority quantization model, and send the quantization results to the other backbone layer nodes. The quantization results include: the priority calculation results of each flame front zone, each flame spread zone, and each cooling zone. The priority matching unit is used to match the comprehensive weight of each node with each priority level through the support vector machine to establish a priority matching table. The priority levels include: first priority, second priority and third priority. The priority calculation result corresponding to the first priority is ≥0.7, the priority calculation result corresponding to the second priority is in the range of (0.3, 0.7), and the priority calculation result corresponding to the third priority is ≤0.

3. The task allocation unit is used to allocate firefighting tasks to each node of the cellular subnet according to the priority matching table and a multi-objective optimization algorithm. Resource surplus includes: surplus electricity and surplus fire extinguishing agent; resource demand includes: electricity demand and fire extinguishing agent demand. The scheduling request module includes: The fire situation and status classification unit is used to input the real-time fire situation and real-time status of each node in this cellular subnet into the random forest model and output the classification results; the classification results include: able to complete the fire extinguishing task and unable to complete the fire extinguishing task. The surplus power prediction unit is used to input the flight time, power consumption rate, current flight speed, current flight altitude and current fire extinguishing agent balance of each node in this cellular subnet into the time series prediction model, and output the surplus power of each node. The fire extinguishing agent surplus prediction unit is used to input the fire extinguishing area of ​​each node in this cell subnet, the average temperature of this management unit, the average fire extinguishing time of each node, and the flame spread rate into the regression model, and output the fire extinguishing agent surplus of each node. The power demand prediction unit is used to establish a power demand prediction model. It inputs the mission flight distance, average flight speed, total firefighting operation time, average power and endurance time of each node into the power demand prediction model, outputs the power demand of this cellular subnet, and sends the power demand to the other backbone layer nodes in the form of a request signal. The fire extinguishing agent demand prediction unit is used to establish a fire extinguishing agent demand prediction model. It inputs the area of ​​each region in this management unit, the fire extinguishing agent dosage per unit area in each region, the priority calculation results of each region in this management unit, and the current fire extinguishing agent balance of each node into the fire extinguishing agent demand prediction model, outputs the fire extinguishing agent demand of this cellular subnet, and sends the fire extinguishing agent demand to the other backbone layer nodes in the form of a request signal. The expression for the fire priority quantification model is: (1); In equation (1), w 1 indicates the weight of the flame spread rate. w 2 indicates the weight of the area affected by the fire. V Indicates the speed of flame spread. V mas Indicates the maximum spread rate of the flame. S Indicates the area affected by the fire. S max Indicates the maximum fire area; w 1 and w 2. Calculated using the Analytic Hierarchy Process (AHP).

5. A forest fire fighting collaborative dispatch system based on a multi-UAV self-organizing network according to claim 4, characterized in that, The expression for the electricity demand forecasting model is: (3); In equation (3), j This refers to the node numbering within the cellular subnet. k This indicates the number of nodes in the subnet. T req This indicates the amount of electricity required. P j Indicates the first j Average power of each node L j Indicates the first j The mission flight distance of each node V j Indicates the first j The average flight speed of each node T total,j Indicates the first j Total firefighting operation time for each node, T already,j Indicates the first j The remaining battery life of each node; The expression for the fire extinguishing agent demand forecasting model is: (2); In equation (2), Q This indicates the required amount of fire extinguishing agent. i For the area number in the management unit, m This indicates the number of areas divided within this management unit. S i Indicates the first i The area of ​​each region q i Indicates the first i The amount of extinguishing agent per unit area in each region. Indicates the first i Priority calculation results for each region C j Indicates the first j The current remaining amount of extinguishing agent at each node.

6. A forest fire fighting collaborative dispatch system based on a multi-UAV self-organizing network as described in claim 4 or 5, characterized in that, The resource scheduling module includes: The demand analysis unit is used to determine whether the resource surplus of this cellular subnet simultaneously satisfies the conditions that surplus power > power demand and fire extinguishing agent surplus > fire extinguishing agent demand. If so, the demand response unit and timing analysis unit are triggered to work; otherwise, no demand response is performed. The demand response unit is used to generate a demand response signal and send the demand response signal and its generation time to the other backbone layer nodes. The timing analysis unit is used to sort the generation times of each demand response signal in chronological order, generate a time queue, and determine whether the generation time of the demand response signal of this node is at the head of the queue. If so, the resource scheduling unit is triggered to work; otherwise, it waits to receive the next request signal. The resource scheduling unit is used to generate corresponding resource scheduling strategies based on resource demand; the resource scheduling strategy includes: the number of drones.

7. A forest fire fighting collaborative dispatch system based on a multi-UAV self-organizing network as described in claim 6, characterized in that, Also includes: The support request module is used to send a support request signal to the ground control station according to the resource requirements when condition A is met. The support return module is used to send control signals back to the ground control station to each node in this cellular subnet after the task is completed, when either condition B or condition C is met. Condition A: The classification result is that the firefighting task cannot be completed and no demand response signal has been received; Condition B: The classification result is that the firefighting task can be completed and no request signal is received during the execution of the task; Condition C: The classification result is that the firefighting task can be completed but no demand response is made during the execution of the task.

8. A forest fire fighting collaborative dispatch system based on a multi-UAV self-organizing network as described in claim 1 or 2, characterized in that, Also includes: The cellular subnet adjustment module is used to add newly added drones as access layer nodes of the cellular subnet, update the hierarchical structure of the cellular subnet, and update the hierarchical structure of the cellular subnet when a drone leaves the cellular subnet. The selection trigger module is also used to trigger the cellular structure adjustment module when the UAV is a backbone node and a UAV joins / leaves from the cellular subnet.

9. A forest fire fighting collaborative dispatch system based on a multi-UAV self-organizing network according to claim 3, characterized in that, Also includes: The blind spot monitoring module is used to monitor the received signal strength index of each communication link in this cellular subnet in real time, filter out communication links with received signal strength index < threshold, mark the area projected by each filtered communication link in this management unit as the edge of the communication blind spot, and send the edge of the communication blind spot to the other backbone layers. The relay node filtering module is used to filter relay nodes in the cellular subnet based on the hierarchical structure of the cellular subnet and the remaining power of each node. The relay link construction module is used to construct relay links containing relay nodes using a topology-aware algorithm. The trigger selection module is also used to trigger the blind zone detection module, relay node screening module, and relay link construction module to work when the UAV is a backbone layer node and receives a communication blind zone edge.

10. A forest fire fighting collaborative dispatch system based on a multi-UAV self-organizing network according to claim 9, characterized in that, Also includes: The 3D model creation module is used to create a 3D model of the communication blind zone of this cellular subnet based on the edge of the communication blind zone and the fused image. The flight trajectory adjustment module is used to dynamically adjust the flight trajectory of each UAV in this cellular subnet based on the terrain elevation and the three-dimensional model of the communication blind spot using the NSGA-II algorithm. The trigger selection module is also used to trigger the 3D model creation module and flight trajectory adjustment module to work when the UAV is a backbone node and receives a signal at the edge of the communication blind zone.

11. A forest fire fighting collaborative dispatch system based on a multi-UAV self-organizing network as described in claim 1 or 2, characterized in that, Also includes: The directional antenna control module is used to start the directional antenna and adjust its direction. The omnidirectional antenna control module is used to start the omnidirectional antenna and adjust the direction of the directional antenna; The trigger selection module is also used to trigger the directional antenna control module to work when the UAV is a backbone layer node, and to trigger the omnidirectional antenna control module to work when the UAV is an access layer node.

Images