UAV swarm control system training, formation control method, device and system

Abstract

The present invention discloses a method, device and system for training and formation control of a drone cluster control system. The formation control method obtains an action by inputting the initial flight position state into a prediction model. The action includes a control signal corresponding to a follower drone, a ratio of the number of allocated subcarriers to the total number of carriers and a quantized bit length of the control signal; the number of subcarriers corresponding to the follower drone is calculated according to the ratio of the number of allocated subcarriers to the total number of carriers; the length of the total range of each dimension of the control signal is quantized according to the quantized bit length of the control signal; and the control signal is sent to the corresponding follower drone through the subcarrier corresponding to the follower drone. The control signal is quantized and transmitted based on the control signal data rate and the quantized bit length of the control signal block length, so as to realize the optimal carrier and control data rate allocation of multiple drones, so as to ensure the best control performance of all drones under fairness and reduce the risk of drone collision.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle control, and in particular relates to a method, device and system for training and formation control of an unmanned aerial vehicle cluster control system. Background Art

[0002] Drones play a key role in environmental monitoring, power line inspection, mapping and other fields due to their simple structure, high cost-effectiveness, excellent maneuverability and strong concealment. However, faced with the complexity of the environment and the diversity of tasks, the capabilities of a single drone have inherent limitations. By organizing multiple drones into a collaborative group, drone swarms can achieve information integration and resource sharing, thereby significantly improving the comprehensive ability of information collection and processing. Compared with a single drone, drone swarms can perform multiple tasks at the same time, which not only improves the efficiency of task execution but also enhances the flexibility of operation. In a drone swarm, even if a single drone fails or is damaged, it will not affect the completion of the entire task. The swarm can respond to unexpected situations through collaborative work and distributed decision-making mechanisms, thereby significantly improving the fault tolerance and robustness of the system. In addition, the mutual cooperation between drones also gives the swarm excellent coordination, intelligence and autonomy, which makes drone swarms more advantageous in performing complex tasks. In order to adapt to diverse mission requirements and cope with changing environmental challenges, drone swarms must have flexible formation adjustment capabilities. Traditional drone swarm management methods rely on clustering architecture, dividing a large swarm into several subclusters, each of which consists of a leader drone and several follower drones. The leader drones of these subclusters can share information and remotely wirelessly control the follower drones in their respective clusters.

[0003] Integrated Communication and Control (ICAC) integrates the communication system with the control system to achieve more efficient UAV operation and management. Traditionally, the communication and control functions of UAVs are designed and deployed separately, but with the development of technology and changes in demand, integrated systems have gradually become a new development trend. Under the constraints of limited resources, the competition between control and communication needs to be effectively coordinated in UAV clusters. ICAC systems are generally composed of controllers, actuators, devices, and sensors. The sensors are responsible for monitoring the status of the devices. The controllers generate control signals based on these device status information and transmit the control signals wirelessly to the actuators, thereby completing the control of the devices and real-time status updates. In a UAV cluster, the leading UAV is both a controller and a sensor, while the subsequent UAVs serve as actuators and devices. On the one hand, in a highly dynamic environment, cluster control has a huge demand for communication resources. As the control flexibility increases, the transmission cost of control signals also increases. On the other hand, during the task execution process, data transmission also depends on communication resources, so the quality of service (QoS) of wireless communication will directly affect the control effect of the system. For example, a high block error rate (BLER) may cause the control signal to fail, thereby reducing the overall control effect.

[0004] At present, in the system based on the leading UAV and the following UAV, the method of the leading UAV controlling the following UAV is based on bandwidth allocation, computing delay, communication delay and topology design, such as "A method for optimizing the federated learning model of UAV cluster based on topology optimization" with application number CN202310828232.5. However, the control performance of the above method needs to be further enhanced. Summary of the invention

[0005] In order to solve the problem that the control performance of existing drones is poor and needs to be improved, the present invention proposes a drone cluster control system training, formation control method, device and system, which quantizes and transmits the control signal based on the control signal data rate and the quantized bit length of the control signal block length, and realizes the optimal carrier and control data rate allocation of multiple drones to ensure the best control performance of all drones under fairness and reduce the risk of drone collision.

[0006] The objective of the present invention is achieved through the following technical solutions:

[0007] The first aspect of the present invention discloses a method for training a drone cluster control system, comprising the following steps:

[0008] Constructing a drone cluster, wherein the drone cluster includes a leading drone and at least one follower drone corresponding to the leading drone;

[0009] Constructing a communication system for the drone cluster, in which the leading drone transmits the generated control signal to the following drones through orthogonal frequency division multiplexing technology, and each subcarrier is only allocated to one following drone;

[0010] Determine a state space set, wherein the state space set includes an initial flight position state of at least one control cycle, wherein the initial flight position state is a difference between an actual flight position of the drone and a target flight position in the current control cycle;

[0011] The state space set sequentially inputs the initial flight position state into the prediction model in the order of the control period to obtain an action, wherein the action includes a control signal corresponding to the follower drone, a ratio of the number of allocated subcarriers to the total number of carriers, and a quantized bit length of the control signal;

[0012] According to the quantization bit length, the control signal block length , Coding Rate Determine the control signal data rate corresponding to the follower drone;

[0013] Determine the block error rate of the corresponding follower drone according to the control signal data rate and determine the initial flight position state of the follower drone in the next control cycle;

[0014] The action performed by the following UAV is evaluated according to the initial flight position state of a control cycle, the corresponding initial flight position state of the next control cycle, the action and the environmental reward of the communication system to update the prediction model.

[0015] A second aspect of the present invention discloses a method for controlling a formation of a drone cluster, which is applicable to a leading drone in a drone cluster, wherein the drone cluster includes a leading drone and at least one following drone that communicates with the leading drone through corresponding subcarriers. The method includes the following steps:

[0016] Get the actual flight position of the following drones in the drone cluster;

[0017] Calculate the initial flight position state of the following UAV corresponding to the control period according to the actual flight position and the corresponding target flight position;

[0018] Inputting the initial flight position state into a prediction model trained by a UAV cluster control system training method in the first aspect to obtain an action, wherein the action includes a control signal corresponding to the following UAV, a ratio of the number of allocated subcarriers to the total number of carriers, and a quantized bit length of the control signal;

[0019] The number of subcarriers corresponding to the following drone is calculated based on the ratio of the number of allocated subcarriers to the total number of carriers;

[0020] quantizing the length of the total range of each dimension of the control signal according to the quantization bit length of the control signal;

[0021] The control signals are sent to the corresponding follow-up drones through the subcarriers corresponding to the follow-up drones.

[0022] The third aspect of the present invention discloses a drone cluster formation control device, comprising a memory and a controller that are communicatively connected in sequence, wherein the memory stores a computer program, and the controller is used to read the computer program and execute a drone cluster formation control method described in the second aspect.

[0023] The third aspect of the present invention discloses a drone cluster formation control system, comprising: a leading drone, the leading drone comprising a drone cluster formation control device as described in the second aspect, which is used to generate a control signal; and at least one follower drone, the follower drone being used to receive and execute the control signal sent by the leading drone.

[0024] The beneficial effects of the present invention are:

[0025] The present invention quantizes the length of the total range of each dimension of the control signal based on the quantization bit length of the control signal data rate and the control signal block length, and then transmits it. That is, it comprehensively considers the influence of the control signal data rate and the control signal block length on the control signal quantization noise. In a drone cluster with limited subcarrier resources, the optimal carrier and control data rate allocation of multiple drones is achieved to ensure the best control performance of all drones under fairness and reduce the risk of drone collision. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are now clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0027] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0028] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. indicate positions or positional relationships based on the positions or positional relationships, or the positions or positional relationships in which the product of the invention is usually placed when in use, or the positions or positional relationships commonly understood by those skilled in the art, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific position, be constructed and operated in a specific position, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0029] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0030] The first aspect of the present invention discloses a method for training a drone cluster control system, which includes steps S11 to S17. It should be noted that the step identifiers in this solution are only for the convenience of explaining the method and do not constitute a sequence limitation. The sequence of each step is based on its language description and the sequence of each signal.

[0031] Step S11: construct a drone cluster, wherein the drone cluster includes a leading drone and at least one follower drone corresponding to the leading drone.

[0032] For example, assume that M+1 drones are constructed, one of which is designated as the leading drone, denoted as , and the remaining M drones are used as follow-up drones and are denoted as , M is a natural number greater than or equal to 1.

[0033] Among them, the leading drone plays a key role in adjusting the flight status of the entire drone cluster, and remotely sends control signals to the following drones through the ICAC system. The leading drone has the ability to sense the actual flight position of the following drones, and can generate corresponding control signals based on the actual flight position and the corresponding target flight position. These control signals are then wirelessly transmitted to the following drones through the ICAC system. After receiving the control signal, the following drone will decode and execute it to adjust its own flight position.

[0034] Specifically, the leading drone and the following drone The interaction process between the two is divided into multiple discrete cycles. Each cycle consists of two main parts: a transmission time slot and a control time slot. The duration of the transmission time slot is In this stage, the leading UAV senses the actual flight position of the following UAV, generates a control signal and transmits it to the following UAV. . Follow the drone The received control signal block length is The duration of the control time slot is , follow the drone during the control time slot Execute the received control signal to update its own flight position.

[0035] Step S12: construct a communication system for the drone cluster, in which the leading drone transmits the generated control signal to the following drones through orthogonal frequency division multiplexing technology OFDM, and each subcarrier is only allocated to one following drone.

[0036] Specifically, the control signal generated by the head unit is transmitted to the drone via OFDM ,Each follow-up drone is assigned a different subcarrier. To avoid interference, each subcarrier is assigned to at most one drone. The number of subcarriers is recorded as ,have ,in is the total number of subcarriers, the control signal block length It is expressed as:

[0037]

[0038] in, is the bandwidth of a single subcarrier, is the control signal transmission duration.

[0039] Follow the drone's wireless channel Modeled as:

[0040]

[0041] in, represents the path loss, Represents small-scale fading:

[0042]

[0043] in, , To follow the actual flight position of the drone, To follow the target flight position of the drone, . is the speed of light, is the signal transmission gain, is the signal receiving gain, is the signal carrier frequency.

[0044] Small scale fading Introducing the block fading channel model:

[0045]

[0046] in, Generated by Rayleigh, is the correlation coefficient of the channel.

[0047] Signal-to-Noise Ratio for:

[0048]

[0049] in, is the signal transmission power, is the average noise power.

[0050] In the tth cycle, follow the drone The channel capacity is expressed as:

[0051]

[0052] Follow the drone The control signal data rate is expressed as:

[0053]

[0054] Among them, channel dispersion for:

[0055]

[0056] Is a follow-up drone Control the block error rate BLER of signal transmission, is a function The inverse function of The block error rate of the control signal can be expressed as:

[0057]

[0058] Step S13: determine a state space set, wherein the state space set includes an initial flight position state of at least one control cycle, and the initial flight position state is a difference between an actual flight position and a target flight position of the drone in the current control cycle.

[0059] State Space Set .

[0060] Since the initial flight position state in the tth cycle is the flight position state of the M following drones after being controlled in the t-1th cycle, the state is continuous.

[0061]

[0062] Step S14: input the initial flight position state of the state space set into the prediction model in sequence according to the control cycle to obtain an action, wherein the action includes a control signal corresponding to the follow-up drone, a ratio of the number of allocated subcarriers to the total number of carriers, and a quantization bit length of the control signal.

[0063] In order to improve control efficiency, the prediction model gives priority to the first actor network model, that is, the first Actor network model, to improve the efficiency of concurrent transaction processing. The Actor network model here is implemented based on the actor network theory.

[0064] That is the first Actor network model The flight position state is used as input and the action is the output, namely:

[0065]

[0066] in, is random noise with mean 0 and variance .

[0067]

[0068] in, Represents an action set, .

[0069] During the training process, by continuously optimizing the prediction model, the ratio of the number of allocated subcarriers to the total number of carriers and the quantization bit length of the control signal can be optimized. for , is an integer. For a given coding rate, To control the signal block length, To control the signal data rate, is the ratio of the number of allocated subcarriers to the total number of carriers, For control signal.

[0070] in,

[0071] ,

[0072] .

[0073] Specifically, the prediction model optimizes the optimization problem based on constraints, that is, the number of collisions between drone clusters in each control cycle. The constraints include conditions C1, C2, C3, and C4.

[0074] Among them, condition C1 is ,

[0075] In the formula, is the quantization bit length, For a given coding rate, To control the signal block length, To follow the drone The channel capacity of

[0076] Condition C2 is ,

[0077] In the formula, is the total number of subcarriers between the leading UAV and the following UAV, To follow the drone The number of allocated subcarriers, M is the number of following drones in the drone cluster;

[0078] Condition C3 is ,

[0079] In the formula, is the mean square error of the control following the UAV in the tth control cycle, , that is, the actual state of the following drone in the tth control cycle With target status The smaller the control mean square error, the better the drone is controlled and the closer it is to the target state.

[0080] To control the mean square error MSE threshold;

[0081] Condition C4 is ;

[0082] The optimization problem is:

[0083]

[0084] In the formula, is the number of collisions of the UAV cluster in the tth control cycle, T is the total number of control cycles, For the control signal.

[0085] The optimized variables are quantization bit length and following drone The number of allocated subcarriers, where condition C1 represents that the control signal data rate of each follower drone does not exceed the channel capacity; condition C2 represents that the total number of subcarriers must not exceed the limit; condition C3 constrains the control performance of the follower drone; condition C4 represents that the number of quantization bits and the number of subcarriers are positive integers.

[0086] Step S15: according to the quantization bit length and the control signal block length , Coding Rate Determine the control signal data rate corresponding to the follower drone.

[0087] According to the formula , the control signal data rate can be calculated ,Right now

[0088] =( / ( )).

[0089] Step S16: determining the block error rate of the corresponding follower drone according to the control signal data rate and determining the initial flight position state of the follower drone in the next control cycle.

[0090] The block error rate of this step is calculated according to equations (1) to (9) in step S12. Specifically, the control signal block length, channel dispersion and channel capacity are calculated respectively, and then the block error rate of the following drone is calculated according to the control signal block length, channel dispersion and channel capacity and the control signal data rate, as shown in equation (9).

[0091] Specifically, the control signal block length is calculated based on the number of subcarriers allocated to the following drone, the bandwidth of a single subcarrier, and the duration of the transmission time slot, as shown in formula (1).

[0092] The calculation of channel dispersion and channel capacity first calculates the path loss based on the actual flight position of the following UAV and the target flight position, see formula (3); then calculates the wireless channel based on the path loss, see formulas (2) and (4); then calculates the signal-to-noise ratio based on the wireless channel, see formula (5); finally, calculates the channel dispersion and channel capacity based on the signal-to-noise ratio, see formulas (8) and (6).

[0093] ‌‌Block error rate is an important indicator to measure the quality of data transmission. It indicates the percentage of blocks with errors during transmission among all the blocks sent. Follow the drone decoding status Following the 0-1 distribution, the following drone will make an error in decoding the control signal, and its execution control signal will fail. , The probability of is the block error rate; the following drone successfully decodes and executes the control signal, at this time, Therefore, the position state of the following UAV after executing the control signal at the end of the current control cycle, that is, the initial flight position state of the next control cycle, can be updated according to the block error rate.

[0094] Then the initial flight position state of the next control cycle is Updated to:

[0095] ,

[0096] in, and is the parameter matrix of the control system, is the Gaussian controlled noise, The mean is 0 and the variance is .

[0097] Step S17: evaluating the action performed by the follower UAV according to the initial flight position state of a control cycle, the corresponding initial flight position state of the next control cycle, the action and the environmental reward of the communication system to update the prediction model.

[0098] This step optimizes the prediction performance of the prediction model by evaluating and updating the following UAV's execution of the action.

[0099] Taking the above Actor network model as an example, the evaluation optimization can be based on the second Actor network model , the first value network model , the second value network model .

[0100] Specifically, the environmental reward received by the following drone in the tth control cycle is for:

[0101] ,

[0102] Before training, initialize the minimum batch size , Experience Replay Pool , the learning rate of the first value network model and the second value network model And the learning rate of the first Actor network model and the second Actor network model , learning interval , attenuation coefficient and the discount factor . Set the standard deviation of the exploration random noise and exploration threshold , randomly initialize the first Actor network model Second Actor Network Model , the first value network model , the second value network model , initialization round number .

[0103] During the evaluation, steps S171 to S176 are used:

[0104] Step S171: Initialization ; Get the initial flight position status ;if is an integer, for the learning rate and to update.

[0105] in:

[0106] ,

[0107] ,

[0108] if , then let .

[0109] Step S172: The action generated by the leading drone is

[0110] in, is random noise with mean 0 and variance .

[0111] The signal received by the following drone is , according to the probability that the following drone fails to successfully decode and execute the control signal, that is, the block error rate, we can get .

[0112] Step S173: Storage To the experience replay pool .

[0113] Step S174: If is an integer, then the random experience replay pool Medium Sampling Multidimensional array of groups , calculate the state-action pair Target value ,

[0114]

[0115] Update the first critic network by minimizing the loss function :

[0116]

[0117] Gradient update first Actor network model :

[0118]

[0119] Update according to the following formula and Network parameters and :

[0120]

[0121]

[0122] in is the soft learning parameter;

[0123] if If it is not an integer, skip this step;

[0124] Step S175: Update time ;like , then go to step S176, otherwise return to step S172;

[0125] Step S176: Update the number of rounds ;like , then the process ends, otherwise, it returns to step S171. is the maximum number of rounds of the algorithm.

[0126] Based on the prediction model obtained by training using a drone cluster control system training method according to the first aspect, the second aspect of the present invention discloses a drone cluster formation control method, which is suitable for use on the leading drone side in a drone cluster. The drone cluster includes a leading drone and at least one follower drone that communicates with the leading drone through corresponding subcarriers, and the method includes steps S21 to S26.

[0127] Step S21: Obtain the actual flight position of the following drones in the drone cluster.

[0128] Step S22: Calculate the initial flight position state of the following UAV corresponding to the control cycle according to the actual flight position and the corresponding target flight position.

[0129] Step S23: input the initial flight position state into the prediction model trained by the method in the first aspect to obtain an action, wherein the action includes a control signal corresponding to the follow-up drone, a ratio of the number of allocated subcarriers to the total number of carriers, and a quantized bit length of the control signal. .

[0130] Step S24: Calculate the number of subcarriers corresponding to the following drone based on the ratio of the number of allocated subcarriers to the total number of carriers.

[0131] Step S25: quantize the length of the total range of each dimension of the control signal according to the quantization bit length of the control signal.

[0132] The length of the total range of each dimension of the control signal is quantized. for:

[0133] ,

[0134] in, It is the control signal The total range of each dimension of the control signal is .

[0135] The control signal quantization noise is ,in Evenly distributed in .

[0136] Step S26: Send the control signal to the corresponding follow-up drones through the subcarriers corresponding to the follow-up drones.

[0137] In the existing UAV swarm formation control technology, the influence of control signal block length and control signal data rate on the quantization noise during control signal transmission is ignored. In theory, the increase of control signal block length and control signal data rate enables more control information to be transmitted, thus reducing the quantization error. However, a higher control signal data rate will also lead to a higher block error rate, which in turn will reduce the control performance. Although a longer control signal block length can reduce the block error rate and reduce the quantization noise, this advantage is determined by the number of subcarriers allocated by the ICAC system based on orthogonal frequency division multiple access. Since the carrier resources in the UAV swarm are limited, in order to ensure fairness among the UAVs and achieve the target formation within the specified time, it is necessary to perform optimal resource allocation. This requires us to not only consider the efficiency of signal transmission, but also balance the use of carrier resources to achieve efficient collaborative operation of the entire UAV swarm. In addition, if the current flight state of the UAV is ignored and the control signal is generated using a fixed control gain, the control MSE will continue to decrease with fluctuations, and the collision problem caused by the UAVs being too close cannot be avoided. Therefore, real-time optimization of the control signal is particularly important.

[0138] In this regard, the scheme method is adopted to quantize the length of the total range of each dimension of the control signal based on the quantization bit length of the control signal data rate and the control signal block length before transmission, that is, the influence of the control signal data rate and the control signal block length on the control signal quantization noise is comprehensively considered, and the optimal carrier and control data rate allocation of multiple drones is realized in the drone cluster with limited subcarrier resources to ensure the best control performance of all drones under fairness. At the same time, considering the collision risk of drones in actual flight, optimizing the control signal can effectively reduce the collision risk of drones.

[0139] The third aspect of the present invention discloses a drone cluster formation control device, including a memory and a controller that are sequentially connected in communication, the memory stores a computer program, and the controller is used to read the computer program and execute a drone cluster formation control method described in the second aspect. For example, the memory may include, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a flash memory (Flash Memory), a first-in-first-out memory (FIFO) and / or a first-in-last-out memory (FILO), etc.; the controller may not be limited to a microcontroller of the STM32F105 series. In addition, the computer device may also include, but is not limited to, a power supply unit, a display screen and other necessary components.

[0140] The fourth aspect of the present invention discloses a drone cluster formation control system, comprising: a leading drone, the leading drone comprising a drone cluster formation control device as described in the second aspect, which is used to generate a control signal; and at least one follower drone, the follower drone being used to receive and execute the control signal sent by the leading drone.

[0141] The operating principles of the devices and systems disclosed in the third and fourth aspects of the present invention are detailed in the first and second aspects and will not be elaborated here.

[0142] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for training a drone cluster control system, characterized in that: The following steps are involved: Constructing a drone cluster, wherein the drone cluster includes a leading drone and at least one follower drone corresponding to the leading drone; Constructing a communication system for the drone cluster, in which the leading drone transmits the generated control signal to the following drones through orthogonal frequency division multiplexing technology, and each subcarrier is only allocated to one following drone; Determine a state space set, wherein the state space set includes an initial flight position state of at least one control cycle, wherein the initial flight position state is a difference between an actual flight position of the drone and a target flight position in the current control cycle; The state space set sequentially inputs the initial flight position state into the prediction model in the order of the control period to obtain an action, wherein the action includes a control signal corresponding to the follower drone, a ratio of the number of allocated subcarriers to the total number of carriers, and a quantized bit length of the control signal; Determine the control signal data rate corresponding to the following drone according to the quantization bit length, the control signal block length, and the coding rate; Determine the block error rate of the corresponding follower drone according to the control signal data rate and determine the initial flight position state of the follower drone in the next control cycle; According to the initial flight position state of a control cycle, the initial flight position state of the corresponding next control cycle, the action and the environmental reward of the communication system, the following UAV is evaluated to update the prediction model; The prediction model optimizes the optimization problem based on constraint conditions, wherein the constraint conditions include condition C1, condition C2, condition C3 and condition C4. Among them, condition C1 is , In the formula, is the quantization bit length, For a given coding rate, To control the signal block length, To follow the drone The channel capacity of Condition C2 is , In the formula, is the total number of subcarriers between the leading UAV and the following UAV, To follow the drone The number of allocated subcarriers, M is the number of following drones in the drone cluster; Condition C3 is , In the formula, is the control mean square error of the following UAV in the t-th control cycle, To control the mean square error threshold; Condition C4 is ; The optimization problem is: , In the formula, is the number of collisions of the UAV cluster in the tth control cycle, T is the total number of control cycles, For the control signal.

2. A UAV swarm control system training method according to claim 1, characterized in that: The control signal data rate corresponding to the following drone is determined according to the quantization bit length, the control signal block length, and the coding rate: The control signal data rate = (quantization bit length / (control signal block length*coding rate)).

3. A UAV swarm control system training method according to claim 1, characterized in that: Determining the block error rate of the corresponding follower drone according to the control signal data rate includes: Calculate the path loss based on the actual flight position of the following drone and the target flight position; Calculating a wireless channel according to the path loss; Calculate a signal-to-noise ratio according to the wireless channel; Calculate the channel dispersion and channel capacity according to the signal-to-noise ratio; The control signal block length is calculated based on the number of subcarriers allocated to the following UAV, the bandwidth of a single subcarrier, and the duration of the transmission time slot; The block error rate of the following drone is calculated based on the control signal block length, channel dispersion, channel capacity and control signal data rate.

4. A method for controlling a formation of a drone cluster, which is applicable to a leading drone in a drone cluster, wherein the drone cluster includes a leading drone and at least one following drone that communicates with the leading drone through corresponding subcarriers, and is characterized in that: The method comprises the following steps: Get the actual flight position of the following drones in the drone cluster; Calculate the initial flight position state of the following UAV corresponding to the control period according to the actual flight position and the corresponding target flight position; Inputting the initial flight position state into a prediction model trained by any one of the UAV cluster control system training methods of claims 1 to 3 to obtain an action, wherein the action includes a control signal corresponding to the follower UAV, a ratio of the number of allocated subcarriers to the total number of carriers, and a quantized bit length of the control signal; The number of subcarriers corresponding to the following drone is calculated based on the ratio of the number of allocated subcarriers to the total number of carriers; quantizing the length of the total range of each dimension of the control signal according to the quantization bit length of the control signal; The control signals are sent to the corresponding follow-up drones through the subcarriers corresponding to the follow-up drones.

5. A method for controlling the formation of a drone swarm according to claim 4, characterized in that: The length of the total range of each dimension of the control signal is quantized according to the quantization bit length of the control signal as follows: The length of the total range of each dimension of the control signal is quantized for: , in, It is the control signal The total range of each dimension of the control signal is .

6. A drone cluster formation control device, comprising a memory and a controller that are sequentially connected in communication, wherein a computer program is stored in the memory, characterized in that: The controller is used to read the computer program and execute the drone cluster formation control method described in claim 4 or 5.

7. A UAV cluster formation control system, characterized in that: include: A leading UAV, the leading UAV comprising the UAV cluster formation control device according to claim 6; At least one follower drone, the follower drone is used to receive and execute the control signal sent by the lead drone.

Images