A UAV cluster control method, device, equipment and storage medium

Abstract

The present invention discloses a method, device, equipment and storage medium for controlling a drone cluster. The method comprises: obtaining drone cluster information and attribute information of high-altitude detection balloons that need to be inspected; based on the drone cluster information and the attribute information of the high-altitude detection balloons that need to be inspected, using a drone path planning algorithm to generate an initial drone inspection plan; with the number of iterations reaching a preset iteration number threshold as the goal, executing the following iterative steps to obtain multiple drone inspection plans, wherein the iterative steps comprise: randomly selecting a high-altitude detection balloon from the path of each drone in the drone inspection plan to form a new high-altitude detection balloon set; for the new high-altitude detection balloon set, using the drone path planning algorithm to generate a new drone inspection plan; calculating an evaluation value of each drone inspection plan; and controlling the drone cluster according to the drone inspection plan with the smallest evaluation value.

Description

Technical Field

[0001] The present invention belongs to the technical field of drone cluster control, and relates to a drone cluster control method, device, equipment and storage medium, and specifically to a drone cluster control method based on optimal path planning of a convolutional neural network. Background Art

[0002] A drone is an unmanned aircraft controlled by a radio remote control device and a self-contained program control device, or a device that is completely or intermittently autonomously operated by an onboard computer. With the continuous development of airborne electronic components, the increasingly complex and changeable mission environment is no longer sufficient for a single drone operation. Therefore, multi-agent technology represented by drone swarms has become a current research hotspot. Faced with increasingly diverse mission requirements and highly complex execution environments, the application of agents is gradually tending towards cluster collaborative operations.

[0003] Drone swarms are easily affected by the environment during flight, which makes it easy for drone swarms to deviate from their flight routes and difficult to maintain a stable flight attitude when flying in strong winds and air turbulence. This makes it easy for drone swarms to get out of control, resulting in the inability of drone swarms to complete designated flight missions, reducing the effectiveness of drone control. Summary of the invention

[0004] Purpose: In order to overcome the deficiencies in the prior art, the present invention provides a drone cluster control method, device, equipment and storage medium.

[0005] Technical solution: To solve the above technical problems, the technical solution adopted by the present invention is:

[0006] In a first aspect, the present invention provides a method for controlling a drone cluster, comprising:

[0007] Obtaining the drone cluster information and the attribute information of the high-altitude detection balloon that needs to be inspected; wherein the drone cluster information includes the location of the drone cluster; the attribute information includes the location of the high-altitude detection balloon, the inspection time, the image shooting resolution, and the stability information attribute of the drone cluster during the flight;

[0008] Based on the information of the drone cluster and the attribute information of the high-altitude detection balloons that need to be inspected, the drone path planning algorithm is used to generate the initial drone inspection plan;

[0009] With the goal of reaching a preset iteration number threshold, the following iterative steps are performed to obtain multiple drone inspection plans, wherein the iterative steps include: randomly selecting a high-altitude detection balloon from the path of each drone in the drone inspection plan to form a new high-altitude detection balloon set; for the new high-altitude detection balloon set, using the drone path planning algorithm to generate a new drone inspection plan;

[0010] Calculate the evaluation value of each drone inspection plan;

[0011] The drone cluster is controlled according to the drone inspection plan with the smallest evaluation value.

[0012] In some embodiments, the drone path planning algorithm includes:

[0013] S1, randomly select an unselected drone cluster from the drone set;

[0014] S2. Randomly select a non-selected high-altitude detection balloon from the high-altitude detection balloon set that is closest to the location of the selected drone; if the selected drone cluster cannot complete the inspection task of the selected high-altitude detection balloon, reselect non-selected drones from the drone cluster set until the selected drone cluster can complete the inspection task of the selected high-altitude detection balloon, and add the location of the selected high-altitude detection balloon as an inspection point to the preset path of the selected drone cluster;

[0015] S3. If there are still unselected high-altitude detection balloons, loop through step S2 until there are no unselected high-altitude detection balloons, and output the drone inspection plan.

[0016] Further, in some embodiments, determining whether the selected drone cluster can complete the inspection task of the selected high-altitude detection balloon includes:

[0017] Whether the selected drone cluster can complete the inspection task of the selected high-altitude detection balloon requires that the following conditions are met at the same time:

[0018] (1) The location information of the positioning radar carried by the drone cluster is higher than the longitude and latitude information of the selected high-altitude detection balloon;

[0019] (2) The drone cluster can reach the selected high-altitude detection balloon from the current high-altitude detection balloon and return to the docking point.

[0020] In some embodiments, the method for calculating the iteration number threshold Δn includes:

[0021]

[0022] Among them, ρ is the preset proportional operator value, N is the number of paths of the drone cluster, and d n The path of the nth drone includes the data of the high-altitude detection balloon.

[0023] In some embodiments, calculating the evaluation value of each drone inspection plan includes:

[0024]

[0025] Among them, S is the evaluation value of the drone inspection plan, N is the number of paths of the drone cluster in the drone inspection plan, and P i n is the inspection energy consumption of the nth UAV completing the i-th inspection task in the UAV inspection scheme, ε1 is the inspection energy consumption weight, T i n is the inspection time for the nth UAV to complete the i-th inspection task in the UAV inspection scheme, ε2 is the weight of the inspection time, and I is the number of high-altitude detection balloons.

[0026] In some embodiments, the drone path planning algorithm uses a bidirectional long short-term memory layer network unit and a CNN network unit;

[0027] The network unit of the bidirectional long short-term memory layer is used to extract features with preset path information of t-sequence drones according to the location of the drone cluster and the attribute information of the high-altitude detection balloons that need to be inspected;

[0028] The CNN network unit is used to determine the optimal path according to the characteristics of the preset path information of the t-sequence drone.

[0029] Furthermore, in some embodiments, the network unit of the bidirectional long short-term memory layer is used to extract features having preset path information of t-sequence drones according to the location of the drone cluster and the attribute information of the high-altitude detection balloons that need to be inspected, including:

[0030] Given the assigned drone sequence U = {U1,U2,..,U NU} and the task sequence T = {T1, T2, .., T NT} Position information of the task sequence to be executed, assuming that the initial position information of the i-th UAV is The location of the jth task point is The path distance between each UAV and the mission point is It is expressed as:

[0031]

[0032] Where: ||·|| represents the bi-norm: i=1,...,N U,j=1,...,N T ; N U is the total number of drones, N T is the total number of mission points;

[0033] The path distance Γ of all UAVs to each mission point m Indicated as N U ×N T The matrix For its elements, m represents the location point of the current path.

[0034] Further, in some embodiments, the CNN network unit is used to determine the optimal path according to the characteristics of the preset path information of the t-sequence drone, including: D is the loss function of the preset path. The smaller the loss value, the better the preset path. It is defined as:

[0035]

[0036] Where: β is the preset path loss coefficient, which is used to determine the training and updating coefficient of the evaluation model parameters; β = [μ1, μ2] ∈ R 1×2 , μ1 represents the probability that the sequence is a generated sequence, μ2 represents the probability that the sequence is a real reserved sequence, and R represents all real numbers. Indicates drone U i The path distance between the jth task point; Indicates drone U i Whether to implement the preset path distance

[0037] In a second aspect, the present invention provides a drone cluster control device, including a processor and a storage medium;

[0038] The storage medium is used to store instructions;

[0039] The processor is configured to operate according to the instructions to execute the method according to the first aspect.

[0040] In a third aspect, the present invention provides a device, comprising:

[0041] Memory;

[0042] processor;

[0043] as well as

[0044] Computer programs;

[0045] The computer program is stored in the memory and is configured to be executed by the processor to implement the method described in the first aspect above.

[0046] In a fourth aspect, the present invention provides a storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described in the first aspect.

[0047] Beneficial effects: The drone cluster control method, device, equipment and storage medium provided by the present invention have the following advantages: by using the drone cluster information determined by the high-precision positioning radar and the attribute information of the high-altitude detection balloon that needs to be inspected, the drone path planning algorithm is used to generate a drone inspection plan, and according to the evaluation value of each drone inspection plan, the drone inspection plan with the smallest evaluation value is found as the optimal solution, so as to achieve stable flight of the drone cluster and optimize the flight path in real time;

[0048] Furthermore, it can effectively solve the problem that when a drone cluster encounters strong winds and air turbulence, the drone cluster deviates from the flight route and has difficulty maintaining a stable flight attitude. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a schematic diagram of a method flow according to an embodiment of the present invention;

[0050] Figure 2 FIG. 2 is a schematic diagram of a practical application according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The present invention will be further described below in conjunction with the accompanying drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0052] In the description of the present invention, "several" means more than one, "many" means more than two, "greater than", "less than", "exceed", etc. are understood to exclude the number itself, and "above", "below", "within", etc. are understood to include the number itself. If there is a description of "first" or "second", it is only used for the purpose of distinguishing the technical features, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0053] In the description of the present invention, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0054] Example 1

[0055] In a first aspect, this embodiment provides a method for controlling a drone cluster, including:

[0056] Obtaining the drone cluster information and the attribute information of the high-altitude detection balloon that needs to be inspected; wherein the drone cluster information includes the location of the drone cluster; the attribute information includes the location of the high-altitude detection balloon, the inspection time, the image shooting resolution, and the stability information attribute of the drone cluster during the flight;

[0057] Based on the information of the drone cluster and the attribute information of the high-altitude detection balloons that need to be inspected, the drone path planning algorithm is used to generate the initial drone inspection plan;

[0058] With the goal of reaching a preset iteration number threshold, the following iterative steps are performed to obtain multiple drone inspection plans, wherein the iterative steps include: randomly selecting a high-altitude detection balloon from the path of each drone in the drone inspection plan to form a new high-altitude detection balloon set; for the new high-altitude detection balloon set, using the drone path planning algorithm to generate a new drone inspection plan;

[0059] Calculate the evaluation value of each drone inspection plan;

[0060] The drone cluster is controlled according to the drone inspection plan with the smallest evaluation value.

[0061] In some embodiments, the drone path planning algorithm includes:

[0062] S1, randomly select an unselected drone cluster from the drone set;

[0063] S2. Randomly select a non-selected high-altitude detection balloon from the high-altitude detection balloon set that is closest to the location of the selected drone; if the selected drone cluster cannot complete the inspection task of the selected high-altitude detection balloon, reselect non-selected drones from the drone cluster set until the selected drone cluster can complete the inspection task of the selected high-altitude detection balloon, and add the location of the selected high-altitude detection balloon as an inspection point to the preset path of the selected drone cluster;

[0064] S3. If there are still unselected high-altitude detection balloons, loop through step S2 until there are no unselected high-altitude detection balloons, and output the drone inspection plan.

[0065] Further, in some embodiments, determining whether the selected drone cluster can complete the inspection task of the selected high-altitude detection balloon includes:

[0066] Whether the selected drone cluster can complete the inspection task of the selected high-altitude detection balloon requires that the following conditions are met at the same time:

[0067] (1) The location information of the positioning radar carried by the drone cluster is higher than the longitude and latitude information of the selected high-altitude detection balloon;

[0068] (2) The drone cluster can reach the selected high-altitude detection balloon from the current high-altitude detection balloon and return to the docking point.

[0069] In some embodiments, the method for calculating the iteration number threshold Δn includes:

[0070]

[0071] Among them, ρ is the preset proportional operator value, N is the number of paths of the drone cluster, and d n The path of the nth drone includes the data of the high-altitude detection balloon.

[0072] In some embodiments, calculating the evaluation value of each drone inspection plan includes:

[0073]

[0074] Among them, S is the evaluation value of the drone inspection plan, N is the number of paths of the drone cluster in the drone inspection plan, is the inspection energy consumption of the nth UAV completing the i-th inspection task in the UAV inspection scheme, ε1 is the weight of the inspection energy consumption, is the inspection time for the nth UAV to complete the i-th inspection task in the UAV inspection scheme, ε2 is the weight of the inspection time, and I is the number of high-altitude detection balloons.

[0075] The quality of the learning rate can directly affect the final accuracy of the model. The multi-classification cross entropy function is used to represent the loss value of the model. By setting the training rounds, the learning rate value is halved every 20 rounds, and the learning rate is reduced. By continuously optimizing the learning rate, the loss value of the model is minimized and recently enters a steady state. Adam's update rule is as follows: Assuming ε = 0, the effective descent step length in time step t and parameter space is There are two suprema of the effective descent step length: In this case, the supremum of the effective step length satisfies and in other cases satisfying |Δ t |≤α. The magnitude of the effective step size of each time step in the parameter space is approximately limited by the step size factor α, that is,

[0076] An epoch process has only one iteration and one update data. Loss Function formula:

[0077]

[0078] The formula for updating weight parameters using the Loss function is:

[0079]

[0080] The prediction result error evaluation standard adopts the mean absolute percentage error (MAPE), that is, the formula:

[0081]

[0082] In the formula is the predicted value; i is the true value; n is the data sample size.

[0083] A network with a bidirectional long short-term memory (BiLSTM) layer is optionally designed for fast estimation of the best path found. This is to generate realistic paths for training learning-based networks. BiLSTM provides extremely fast path planning for real-time drone applications. It is valuable for real-time autonomous drone applications in complex and large environments because it provides a new method with fast path planning capabilities.

[0084] The output feature vector of the information feature extraction network is H As the hidden state input of the sequence generation network, H It is possible to provide constraints on each generated sequence. Then set the initialization input x1 = N U +1, the network calculation is as follows (6).

[0085]

[0086]

[0087]

[0088]

[0089] x t+1 =O t

[0090] Where: W z , W r is the network parameter, σ(·) is sigmoid. The entire calculation process of formula (6) can be expressed as a BiLSTM unit. Set the input data at time t to x t , the hidden state input is where x tis the result calculated by the BiLSTM unit at the previous moment, is the hidden state output calculated by the BiLSTM unit at the previous moment. t , Substitute into formula (6) to calculate the hidden state output at this moment contains all the preset path information of the drones before time t. Then the output O at time t is obtained. t .

[0091] Optionally, the features with t-sequence drone preset path information extracted by multiple BiLSTM units are input into the CNN network for optimal analysis of the preset path.

[0092] The CNN model is optionally trained to adapt the behavior strategy of the UAV. In addition, since the dataset contains global environmental information and optimal paths, the trained CNN model can estimate the distribution of environmental information. The estimated information can be regarded as implementing local environmental information and target positions to infer more appropriate behaviors, thereby effectively mitigating the first conflict. And use the optimal path planning algorithm to calculate their optimal UAV paths. Then extract the state and position of the environmental information in the detection range at different times during the path planning process, and train the CNN model. When the state of the new scene is input into the training model, the predicted behavior of the UAV is regarded as a guide to the heading direction in the case of real-time path planning. First, various scenes are collected and their optimal UAV paths are calculated using the optimal path planning algorithm. Then extract the state and position of the environmental information in the detection range at different times during the path planning process, and train the CNN model. When the state of the new scene is input into the training model, the predicted behavior of the UAV is regarded as a guide to the heading direction in the case of real-time path planning. First, various scenes are collected and their optimal UAV paths are calculated using the optimal path planning algorithm. Then extract the state and position of the environmental information in the detection range at different times during the path planning process, and train the CNN model.

[0093] In some embodiments, the drone path planning algorithm uses a bidirectional long short-term memory layer network unit and a CNN network unit;

[0094] The network unit of the bidirectional long short-term memory layer is used to extract features with preset path information of t-sequence drones according to the location of the drone cluster and the attribute information of the high-altitude detection balloons that need to be inspected;

[0095] The CNN network unit is used to determine the optimal path according to the characteristics of the preset path information of the t-sequence drone.

[0096] Furthermore, in some embodiments, the network unit of the bidirectional long short-term memory layer is used to extract features having preset path information of t-sequence drones according to the location of the drone cluster and the attribute information of the high-altitude detection balloons that need to be inspected, including:

[0097] Given the assigned drone sequence U = {U1,U2,..,U NU} and the task sequence T = {T1, T2, .., T NT} Position information of the task sequence to be executed, assuming that the initial position information of the i-th UAV is The location of the jth task point is The path distance between each UAV and the mission point is It is expressed as:

[0098]

[0099] Where: ||·|| represents the bi-norm: i=1,...,N U ,j=1,...,N T ; N U is the total number of drones, N T is the total number of mission points;

[0100] The path distance Γ of all UAVs to each mission point m Indicated as N U ×N T The matrix For its elements, m represents the location point of the current path.

[0101] Further, in some embodiments, the CNN network unit is used to determine the optimal path according to the characteristics of the preset path information of the t-sequence drone, including: D is the loss function of the preset path. The smaller the loss value, the better the preset path. It is defined as:

[0102]

[0103] Where: β is the preset path loss coefficient, which is used to determine the training and updating coefficient of the evaluation model parameters; β = [μ1, μ2] ∈ R 1×2 , μ1 represents the probability that the sequence is a generated sequence, μ2 represents the probability that the sequence is a real reserved sequence, and R represents all real numbers. Indicates drone U i The path distance between the jth task point; Indicates drone U i Whether to implement the preset path distance

[0104] Example 2

[0105] In a second aspect, based on Embodiment 1, this embodiment provides a drone cluster control device, including a processor and a storage medium;

[0106] The storage medium is used to store instructions;

[0107] The processor is configured to operate according to the instructions to execute the method according to embodiment 1.

[0108] In some embodiments, a drone cluster control system includes the above-mentioned drone cluster control device.

[0109] Furthermore, the drone cluster control system also includes: a high-altitude meteorological detection balloon, a ground positioning radar, and an inertial measurement unit;

[0110] High-altitude meteorological balloons can monitor the wind speed and direction around the drone cluster. Ground positioning radars can observe the flight position of the drone cluster in real time through the ground control center. When the drone cluster deviates from the position, it can be discovered in time through the positioning radar. The inertial measurement unit (IMU) measures the horizontal state, speed, and position information of the drone during flight. When the drone cluster deviates from the path, it can be controlled and corrected in time through the ground control center.

[0111] The ground control console presets the drone cluster to perform flight missions through the control device. When the drone cluster encounters strong winds during flight, the high-altitude meteorological detection balloon will transmit weather information such as wind speed, wind direction, temperature and humidity to the control device to provide a reference for the control device. At the same time, when the transmitted wind force is greater than the maximum tolerance range of the drone cluster, the control device will generate an alarm message to remind the staff to recall the drone cluster and stop the flight mission, effectively preventing accidental losses caused by the forced flight of the drone cluster. When the wind force sensed by the drone cluster has no effect on the flight of the drone, the staff can adjust the attitude, speed and heading of the drone cluster through the ground control panel so that it can fly steadily in line with the wind force.

[0112] At the same time, when the drone cluster is out of control of the control device during flight, the drone automatically starts sending positioning information, and the warning light is automatically triggered, emitting an audible and visual alarm to facilitate staff to find it; when the drone cluster falls abnormally, the airbag can be triggered to reduce damage to the drone, thereby improving the flight safety of the drone cluster.

[0113] Example 3

[0114] In a third aspect, based on embodiment 1, this embodiment provides a device, including:

[0115] Memory;

[0116] processor;

[0117] as well as

[0118] Computer programs;

[0119] The computer program is stored in the memory and is configured to be executed by the processor to implement the method described in Example 1.

[0120] Example 4

[0121] In a fourth aspect, based on Embodiment 1, this embodiment provides a storage medium on which a computer program is stored, and when the computer program is executed by a processor, the method described in Embodiment 1 is implemented.

[0122] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0123] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0124] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0125] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0126] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for controlling a drone cluster, characterized in that: The method comprises: Obtaining the drone cluster information and the attribute information of the high-altitude detection balloon that needs to be inspected; wherein the drone cluster information includes the location of the drone cluster; the attribute information includes the location of the high-altitude detection balloon, the inspection time, the image shooting resolution, and the stability information attribute of the drone cluster during the flight; Based on the information of the drone cluster and the attribute information of the high-altitude detection balloons that need to be inspected, the drone path planning algorithm is used to generate the initial drone inspection plan; With the goal of reaching a preset iteration number threshold, the following iterative steps are performed to obtain multiple drone inspection plans, wherein the iterative steps include: randomly selecting a high-altitude detection balloon from the path of each drone in the drone inspection plan to form a new high-altitude detection balloon set; for the new high-altitude detection balloon set, using the drone path planning algorithm to generate a new drone inspection plan; Calculate the evaluation value of each drone inspection plan; The drone cluster is controlled according to the drone inspection plan with the smallest evaluation value; The UAV path planning algorithm adopts a network unit of a bidirectional long short-term memory layer and a CNN network unit; the network unit of the bidirectional long short-term memory layer is used to extract features with t-sequence UAV preset path information according to the location of the UAV cluster and the attribute information of the high-altitude detection balloon that needs to be inspected; the CNN network unit is used to determine the optimal path according to the features with t-sequence UAV preset path information; The network unit of the bidirectional long short-term memory layer is used to extract features with preset path information of t-sequence drones according to the location of the drone cluster and the attribute information of the high-altitude detection balloons that need to be inspected, including: Given the assigned drone sequence and pending task sequence The position information of the task sequence to be executed, assuming that the initial position information of the i-th UAV is The location of the jth task point is The path distance between each UAV and the mission point is It is expressed as: Where: ||·|| represents the bi-norm: i=1,...,N U ,j=1,...,N T ; N U is the total number of drones, N T is the total number of mission points; The path distance Γ of all UAVs to each mission point m Indicated as N U ×N T The matrix is the element therein, m represents the position point of the current path; The output feature vector of the information feature extraction network is H As the hidden state input of the sequence generation network Make Ι H It is possible to provide constraints on each generated sequence; then set the initialization input x1 = N U +1, the network calculation is as follows: xt+1=Ot Where: W z , W r is the network parameter, σ(·) is sigmoid; the entire calculation process is represented as a BiLSTM unit, and the input data at time t is set to x t , the hidden state input is where x t is the result calculated by the BiLSTM unit at the previous moment, is the hidden state output calculated by the BiLSTM unit at the previous moment; x t , Substitute into the above formula to calculate the hidden state output at this moment contains all the preset path information of the UAVs retained before time t; then the output O at time t is obtained t , that is, x t+1 ; The CNN network unit is used to determine the optimal path according to the characteristics of the preset path information of the t-sequence UAV, including: D is the loss function of the preset path. The smaller the loss value, the better the preset path. It is defined as: Where: β is the preset path loss coefficient, which is used to determine the training update coefficient of the evaluation model parameters; β = [μ1, μ2] ∈ R 1 ×2 , μ1 represents the probability that the sequence is a generated sequence, μ2 represents the probability that the sequence is a real reserved sequence, and R represents all real numbers. Indicates drone U i The path distance between the jth task point; Indicates drone U i Whether to implement the preset path distance The CNN network unit is trained by first collecting various scenes and using the optimal path planning algorithm to calculate the optimal drone path for various scenes, and then extracting the state and position of the environmental information within the detection range at different times during the path planning process to train the CNN network unit; the data set contains global environmental information and the optimal path, and the trained CNN network unit can estimate the distribution of environmental information.

2. The drone cluster control method according to claim 1, characterized in that: The UAV path planning algorithm includes: S1, randomly select an unselected drone cluster from the drone set; S2. Randomly select a non-selected high-altitude detection balloon from the high-altitude detection balloon set that is closest to the location of the selected drone; if the selected drone cluster cannot complete the inspection task of the selected high-altitude detection balloon, reselect non-selected drones from the drone cluster set until the selected drone cluster can complete the inspection task of the selected high-altitude detection balloon, and add the location of the selected high-altitude detection balloon as an inspection point to the preset path of the selected drone cluster; S3. If there are still unselected high-altitude detection balloons, loop through step S2 until there are no unselected high-altitude detection balloons, and output the drone inspection plan.

3. The drone cluster control method according to claim 2, characterized in that: Determine whether the selected drone cluster can complete the inspection task of the selected high-altitude detection balloon, including: Whether the selected drone cluster can complete the inspection task of the selected high-altitude detection balloon requires that the following conditions are met at the same time: (1) The location information of the positioning radar carried by the drone cluster is higher than the longitude and latitude information of the selected high-altitude detection balloon; (2) The drone cluster can reach the selected high-altitude detection balloon from the current high-altitude detection balloon and return to the docking point.

4. The drone cluster control method according to claim 1, characterized in that: The calculation method of the iteration number threshold Δn includes: Among them, ρ is the preset proportional operator value, N is the number of paths of the drone cluster, and d n The path of the nth drone includes the data of the high-altitude detection balloon.

5. The drone cluster control method according to claim 1, characterized in that: Calculate the evaluation value of each drone inspection plan, including: Among them, S is the evaluation value of the drone inspection plan, N is the number of paths of the drone cluster in the drone inspection plan, is the inspection energy consumption of the nth UAV completing the i-th inspection task in the UAV inspection scheme, ε1 is the inspection energy consumption weight, T i n is the inspection time for the nth UAV to complete the i-th inspection task in the UAV inspection scheme, ε2 is the weight of the inspection time, and I is the number of high-altitude detection balloons.

6. A drone cluster control device, characterized in that: including processor and storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the method according to any one of claims 1 to 5.

7. A storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images