Variational auto-encoder trajectory planning method for communication and sensing integrated unmanned aerial vehicle network

Abstract

The invention discloses a variational auto-encoder trajectory planning method and device for a communication and inductance integrated unmanned aerial vehicle network, and relates to the technical field of wireless communication. The method comprises the following steps: constructing a double-layer unmanned aerial vehicle cooperative network architecture; configuring orthogonal time-frequency air conditioning parameters and antenna array parameters based on the network architecture; based on the modulation parameters and the antenna array parameters, adopting an orthogonal time-frequency modulation technology to extract delay-Doppler domain channel parameters and calculating a communication rate; defining a comprehensive cost objective function; according to the objective function and the communication rate, constructing an optimal problem of trajectory planning under multiple constraint conditions; constructing a model based on a variational auto-encoder; based on an optimal problem, constructing a composite loss function containing a reconstruction error and a constraint penalty term, and training the model to obtain a trained variational auto-encoder-based model; and based on the trained variational auto-encoder-based model, generating an optimal flight path. According to the invention, the flight energy consumption and safety risk of the unmanned aerial vehicle can be reduced.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a variational autoencoder trajectory planning method and apparatus for integrated sensing and communication unmanned aerial vehicle (UAV) networks. Background Technology

[0002] With the rapid development and widespread adoption of Unmanned Aerial Vehicle (UAV) technology, its role in modern wireless communication systems is becoming increasingly prominent. UAVs, with their high mobility, flexible deployment, and low cost, are widely used in scenarios such as supporting ground communications, emergency disaster relief, aerial reconnaissance and surveillance, and temporary hotspot coverage. Especially in 6G network scenarios, the construction of integrated air-ground-space networks relies heavily on the deep involvement of UAV nodes. However, in practical applications, ensuring reliable wireless communication in dynamic environments faces numerous severe challenges. The high-speed movement of UAVs leads to severe Doppler shift in the wireless channel, complex urban or mountainous environments can trigger multipath fading effects, and the time-varying mobility of users further increases the difficulty of network topology management. These factors combined significantly reduce the stability and reliability of communication links.

[0003] To address the issues of scarce spectrum resources and redundant hardware, Integrated Sensing and Communication (ISAC) technology has emerged. ISAC aims to integrate radar sensing and wireless communication functions onto a single hardware platform and spectrum resource, achieving mutual benefit between communication and sensing through shared RF front-ends and signal processing modules. Introducing ISAC technology into UAV networks has significant strategic importance. On one hand, sensing functions help UAVs acquire real-time information about their surroundings, such as obstacle locations, user distribution, and the status of other UAVs, thereby assisting in safer navigation and more accurate beamforming. On the other hand, communication functions are the foundation for UAVs to perform tasks, transmit data, and coordinate control. Related research shows that utilizing ISAC technology in UAV networks can not only improve spectrum utilization and reduce the need for additional radar spectrum, but also enhance beam alignment accuracy through sensing-assisted communication, thus achieving coexistence of communication and sensing without significantly reducing performance.

[0004] Despite the promising future of ISAC technology, traditional Orthogonal Frequency Division Multiplexing (ISC) modulation techniques face significant challenges in high-mobility UAV scenarios. ISC is highly sensitive to Doppler shift; high-speed movement disrupts the orthogonality between subcarriers, leading to inter-carrier interference and severely degrading communication quality. To address this issue, Orthogonal Time Frequency Space (OTFS) modulation technology has been proposed. OTFS modulates signals in the delay-Doppler domain, transforming time-varying multipath channels into quasi-static and sparse channel interactions in the delay-Doppler domain. This characteristic gives OTFS inherent Doppler immunity, making it suitable for high-mobility communication scenarios in high-speed rail and UAVs.

[0005] However, advanced physical layer technologies alone are insufficient to solve all the problems of UAV networks; trajectory planning is equally crucial. Reasonable trajectory planning can not only shorten communication distances and improve channel quality, but also ensure the continuity of perception and flight safety. Traditional UAV trajectory planning methods mainly rely on convex optimization, dynamic programming, greedy algorithms, or heuristic methods based on graph search. These traditional methods perform reasonably well in simple static environments, but they often fall short when facing complex systems like ISAC UAV networks, which are high-dimensional, nonlinear, and have multiple constraints coupled together. Specifically, traditional optimization algorithms typically require accurate environmental modeling, and their computational complexity increases exponentially with the problem size, making it difficult to meet the real-time online planning requirements of UAVs. Furthermore, these methods often treat communication, perception, and trajectory control separately, failing to fully explore the potential coupling relationships between them, easily getting trapped in local optima, resulting in low overall system energy efficiency. In summary, while the current UAV network field has accumulated some experience with ISAC and OTFS technologies, there is still a gap in the deep integration of these technologies with efficient trajectory planning algorithms. Existing research mostly focuses on single technical aspects, lacking a holistic framework that can uniformly consider the anti-Doppler characteristics of OTFS, the sensing coordination requirements of ISAC, and the kinematic energy consumption constraints of UAVs. Therefore, there is an urgent need to develop a novel trajectory planning scheme that can minimize UAV flight costs and achieve real-time adaptive path adjustment while ensuring communication reliability and sensing accuracy. Summary of the Invention

[0006] To address the technical problems of existing technologies, such as severe Doppler frequency shift, complex multipath fading, and difficulty in coordinating communication and sensing performance optimization in high-dynamic environments, leading to poor communication reliability and high energy consumption in trajectory planning, this invention provides a variational autoencoder trajectory planning method and apparatus for integrated sensing and communication unmanned aerial vehicle (UAV) networks. The technical solution is as follows:

[0007] On the one hand, a variational autoencoder trajectory planning method for sensor-integrated UAV networks is provided. This method is implemented by a variational autoencoder trajectory planning device for sensor-integrated UAV networks, and includes: S1. Construct a two-layer UAV collaborative network architecture; configure orthogonal time-frequency control parameters and antenna array parameters based on the network architecture; S2. Based on the orthogonal time-frequency control parameters and antenna array parameters, the orthogonal time-frequency control technology is used to extract the delay-Doppler domain channel parameters and calculate the communication rate. S3. Define a comprehensive cost objective function that includes weights for distance, safety, turning, and altitude; based on the comprehensive cost objective function and communication rate, construct an optimal trajectory planning problem with multiple constraints. S4. Construct a variational autoencoder-based model; based on the optimal problem, construct a composite loss function that includes reconstruction error and constraint penalty term; train the variational autoencoder-based model according to the composite loss function to obtain a trained variational autoencoder-based model. S5. Obtain the current state data of the UAV, input the current state data of the UAV into the trained variational autoencoder-based model, and generate the optimal flight trajectory that satisfies multiple constraints.

[0008] On the other hand, a variational autoencoder trajectory planning device for sensor-integrated UAV networks is provided. This device is applied to a variational autoencoder trajectory planning method for sensor-integrated UAV networks. The device includes: The first building unit is used to construct a two-layer UAV collaborative network architecture; and to configure orthogonal time-frequency control parameters and antenna array parameters based on the network architecture. The calculation unit is used to extract the delay-Doppler domain channel parameters and calculate the communication rate based on the orthogonal time-frequency control parameters and antenna array parameters, using orthogonal time-frequency control technology. The second building unit is used to define a comprehensive cost objective function that includes weights for distance, safety, turning, and altitude; and to construct an optimal trajectory planning problem with multiple constraints based on the comprehensive cost objective function and the communication rate. A training unit is used to construct a variational autoencoder-based model; based on the optimal problem, a composite loss function containing reconstruction error and constraint penalty term is constructed; the variational autoencoder-based model is trained according to the composite loss function to obtain a trained variational autoencoder-based model; The generation unit is used to acquire the current state data of the UAV, input the current state data of the UAV into the trained variational autoencoder-based model, and generate the optimal flight trajectory that satisfies multiple constraints.

[0009] On the other hand, a variational autoencoder trajectory planning device for a sensor-integrated unmanned aerial vehicle (UAV) network is provided. The variational autoencoder trajectory planning device for a sensor-integrated UAV network includes: a processor; a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, any one of the methods described above for variational autoencoder trajectory planning for a sensor-integrated UAV network is implemented.

[0010] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, the at least one instruction being loaded and executed by a processor to implement any of the above-described variational autoencoder trajectory planning methods for sensor-integrated unmanned aerial vehicle networks.

[0011] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: The embodiments of the present invention employ orthogonal time-frequency control technology to solve the Doppler frequency shift problem caused by high mobility, and combine it with the generative capability based on the variational autoencoder model to solve the problem of UAV trajectory planning under multiple constraints, thereby achieving synergistic optimization of communication reliability, perception continuity and flight energy consumption.

[0012] This invention achieves a synergistic improvement in communication reliability and perception continuity, significantly reducing the flight energy consumption and safety risks of UAVs, and enhancing the network's adaptability to dynamic environments. Using this invention can effectively reduce flight costs and improve communication stability. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of a variational autoencoder trajectory planning method for a sensor-integrated unmanned aerial vehicle (UAV) network provided by an embodiment of the present invention. Figure 2 This is a schematic diagram of an ISAC-UAV network system model provided in an embodiment of the present invention; Figure 3 This is a diagram of an OTFS modulation and demodulation signal system architecture provided by an embodiment of the present invention; Figure 4 This is a block diagram of a variational autoencoder trajectory planning device for a sensor-integrated unmanned aerial vehicle (UAV) network provided in an embodiment of the present invention. Figure 5This is a schematic diagram of the structure of a variational autoencoder trajectory planning device for a sensor-integrated unmanned aerial vehicle (UAV) network provided in an embodiment of the present invention. Detailed Implementation

[0015] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0016] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0017] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0018] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0019] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0020] This invention provides a variational autoencoder trajectory planning method for sensor-integrated unmanned aerial vehicle (UAV) networks. This method can be implemented by a variational autoencoder trajectory planning device for sensor-integrated UAV networks, which can be a terminal or a server. Figure 1 The flowchart shown is a variational autoencoder trajectory planning method for sensor-integrated UAV networks. The processing flow of this method may include the following steps:

[0021] S1. Construct a two-layer UAV collaborative network architecture; configure orthogonal time-frequency control parameters and antenna array parameters based on the network architecture.

[0022] Among them, the two-layer UAV collaborative network architecture divides the UAV group into an upper-layer master UAV that performs centralized perception and planning tasks and a lower-layer communication UAV that performs specific communication service tasks, which can achieve task decoupling and resource collaboration in complex dynamic environments.

[0023] In one feasible implementation, the main drone at the upper layer of the network is equipped with Each transmitting antenna transmits signals integrating sensing and communication functions to acquire location information of communication drones, ground users, and obstacles in the environment. The communication drones in the lower layer of the network are equipped with... One transmitting antenna and The main function of the receiving antenna is to dynamically adjust its position to cover ground users and establish a reliable communication link according to the scheduling instructions of the main UAV. Based on the two-layer UAV cooperative network architecture, the main UAV at the upper layer of the network uses the trigonometric theorem to calculate the target distance by measuring the time delay of the echo signal; whereby the target distance is expressed by the following formula (1):

[0024] (1) in, Indicates the distance to the target; Indicates a time delay; It represents the speed of light.

[0025] In this process, by combining measurement data from multiple main UAVs, the precise three-dimensional coordinates of the target are calculated by solving a system of nonlinear equations, thereby constructing a real-time network topology map and providing environmental state input for subsequent trajectory optimization.

[0026] In one feasible implementation, Figure 2 This is a schematic diagram of an ISAC-UAV network system model provided in an embodiment of the present invention; a two-layer ISAC UAV network environment for 6G communication scenarios is constructed. The simulation area is set as follows. The network architecture is logically divided into two layers: the upper layer deploys three main UAVs, with their flight altitude set between 100 and 150 meters to ensure they are within line-of-sight paths of most ground targets; the lower layer deploys 20 communication UAVs, with their flight altitude set between 30 and 80 meters for flexible maneuverability and proximity to ground users. The number of ground users is set to 50, randomly distributed within the area, and simulated to move irregularly at speeds up to 120 km / h. The system communication carrier frequency is set in the millimeter-wave band (e.g., 28 GHz), and the system bandwidth is set to 10 MHz. To implement orthogonal time-frequency control, the time-frequency resource grid is divided into... Subcarriers and One time slot symbol, subcarrier spacing The frequency is set to 15kHz. The main UAV is equipped with a high-gain radar sensing antenna array, which is responsible for periodically transmitting ISAC signals to scan the environment; the communication UAV is equipped with an omnidirectional communication antenna, which is responsible for receiving control commands from the main UAV and providing downlink data transmission services to ground users.

[0027] Among them, such as Figure 3 The figure shown is an architecture diagram of an OTFS modulation and demodulation signal system provided by an embodiment of the present invention.

[0028] S2. Based on the orthogonal time-frequency control parameters and antenna array parameters, the orthogonal time-frequency control technology is used to extract the delay-Doppler domain channel parameters and calculate the communication rate.

[0029] Optionally, the specific implementation process of S2 includes S21-S25: S21. Using orthogonal time-frequency modulation technology, the transmitting end maps the delay-Doppler domain data symbols to the time-frequency domain through inverse symplectic finite Fourier transform to obtain the time-frequency domain signal; In one feasible implementation, the process of obtaining the time-frequency domain signal is represented by the following formula (2): (2) in, Indicates the delay-Doppler domain data symbol; n represents the time slot index; m represents the subcarrier index; k represents the Doppler tap index; Indicates a delayed tap index; This represents a time-frequency domain signal.

[0030] S22. Use Heisenberg transform to convert the time-frequency domain signal into a continuous-time transmitted waveform; The continuous-time transmission waveform is represented by the following formula (3): (3) in, This represents the continuous-time transmitted waveform; T represents the symbol period, satisfying... ; t represents a continuous-time variable; Indicates the transmission pulse; This indicates the subcarrier spacing.

[0031] In one feasible implementation, after the signal is transmitted through a time-varying multipath channel, the time-domain signal is received at the receiving end. The receiver uses the received pulse pairs to perform a Wigner transform, mapping the signal from the time domain back to the time-frequency domain, to obtain the time-frequency domain discrete sampled signal after the Wigner transform at the receiving end, which is expressed by the following formula (4):

[0032] (4) in, Represents a time-domain signal; Receive pulses; This represents the time-frequency domain discrete sampled signal after Wigner transformation.

[0033] In one feasible implementation, based on the Wigner transform-derived discrete sampled signal in the time-frequency domain, the symplectic finite Fourier transform is used to restore the signal from the time-frequency domain to the delay-Doppler domain, so as to perform channel estimation and data detection in this domain; the process of restoring the signal from the time-frequency domain to the delay-Doppler domain is expressed by the following formula (5): (5) in, This indicates the delay-Doppler domain.

[0034] Through the above transformation, the Doppler frequency shift generated by rapid movement manifests as a cyclic shift of data in the delay-Doppler domain, rather than inter-carrier interference in the OFDM system, thus ensuring the reliability of the link.

[0035] In one feasible implementation, regarding the sensing function, the main UAV utilizes a matched filter to process the received echo signal. Let the transmitted signal be... Let the time-domain signal be... For the echo signal, the impulse response of the matched filter is: The output of the matched filter It can be expressed by the following formula (6):

[0036] (6) Among them, through detection The peak position of the target is used to estimate the round-trip delay and Doppler shift. Based on these estimates, the main UAV calculates the target position using triangulation. Assume the positions of the three main UAVs are as follows: The measured distances are respectively The target coordinates can then be calculated by solving the following system of equations. It can be expressed by the following formula (7):

[0037] (7) S23. Establish a downlink sensing channel model and obtain the channel impulse response of the main UAV's transmitted signal during propagation. In one feasible implementation, the channel impulse response is represented by the following formula (8): (8) in, Indicates the first Complex reflection coefficients of the path; Indicates the first Doppler frequency shift along the path; Indicates the first Path delay for each path; Indicates the launch steering vector; This represents the channel impulse response; t represents the continuous-time variable. Represents the time delay variable; Indicates the azimuth angle variable; Represents the Dirac impulse function; This represents the total number of scattering paths.

[0038] In one feasible implementation, for the uplink echo reception channel, considering the bistatic radar geometry, the channel response is expressed by the following formula (9): (9) in, Indicates the channel response; This includes the effects of path loss and the target's radar cross section; This represents the bistatic Doppler frequency generated by the relative motion between the UAV and the target.

[0039] S24. Establish a multi-input multi-output orthogonal time-frequency space-space channel model for inter-UAV communication, map the physical channel to a discrete delay-Doppler grid, and obtain the equivalent channel response; Among them, for the communication channel between the communication UAV and the user, a multi-input multi-output orthogonal time-frequency space-space channel model is adopted, and the time-varying channel matrix is ​​constructed by the following formula (10): (10) in, Indicates complex gain; Represents the time-varying channel matrix; Indicates the receiving array steering vector; Indicates the transmission array steering vector; This indicates the total number of paths that exist in the communication link.

[0040] In one feasible implementation, in order to overcome the Doppler effect, the physical channel is mapped to the delay-Doppler domain using orthogonal time-frequency modulation techniques, and the equivalent channel response is expressed by the following formula (11): (11) in, Represents the equivalent channel response; P represents the total number of propagation paths; Indicates the first The complex channel gain of each path; Indicates the corresponding number Discrete delay tap index for each path; Indicates the corresponding number Discrete Doppler tap index of the path; Indicates the first Doppler frequency shift along the path; Indicates the first Path delay for each path.

[0041] Among them, using sparse channel characteristics in the delay-Doppler domain for channel estimation and equalization can eliminate the impact of Doppler frequency shift on communication quality.

[0042] S25. Calculate the communication rate based on the channel impulse response and the channel impulse response.

[0043] In one feasible implementation, based on the input-output relationship And based on mutual information theory, the communication rate is calculated and expressed by the following formula (12): (12) in, Represents the equivalent channel matrix; Represents the covariance matrix of the transmitted signal; Indicates noise power; Indicates communication rate; The dimension is The identity matrix; This represents the transpose of the equivalent channel matrix.

[0044] S3. Define a comprehensive cost objective function that includes weights for distance, safety, turning, and altitude; based on the comprehensive cost objective function and communication rate, construct an optimal trajectory planning problem with multiple constraints.

[0045] Optionally, the comprehensive cost objective function of S3, which includes weights for distance, safety, steering, and altitude, is expressed by the following formula (13): (13) in, Represent the overall cost objective function; The distance cost is the sum of the Euclidean distances between adjacent waypoints; This represents the safety cost, which is a penalty based on the safety distance threshold and the distance to the obstacle. The altitude cost is the sum of the absolute values ​​of altitude changes between adjacent waypoints. Indicates the cost of switching; This represents the first weighting coefficient, set to 1.0; This represents the second weighting coefficient, set to 10.0; This represents the third weighting coefficient, set to 2.0; This represents the fourth weighting coefficient, set to 1.5.

[0046] In one feasible implementation, the trajectory planned by the UAV is... A discrete waypoint sequence Composition, then distance from cost This can be expressed by the following formula (14): (14) in, Represents the i-th discrete waypoint; Let represent the (i+1)th discrete waypoint; N represents the total number of discrete waypoints. Distance cost directly reflects the basic energy consumption of the UAV flight.

[0047] In one feasible implementation, a safe distance threshold is preset. For each point on the trajectory, the distance to the nearest obstacle or no-fly zone is calculated. If this distance is less than the safe distance threshold, a penalty is incurred; wherein, the safety cost... This can be expressed by the following formula (15): (15) in, Indicates the safe distance threshold; This indicates the distance to obstacles. By calculating safety costs, it is ensured that the generated trajectory can automatically avoid known obstacles, reducing the risk of collision.

[0048] In one feasible implementation, high cost The energy consumption resulting from vertical maneuvering is expressed by the following formula (16): (16) in, Indicates the first The altitude of each waypoint; Indicates the first The altitude of each waypoint.

[0049] In one feasible implementation, to ensure smooth flight and reduce energy consumption and control instability caused by sharp turns, it is necessary to calculate the heading change at each waypoint; among these, the turning cost... Used to penalize abrupt changes in heading in order to ensure flight smoothness, it is expressed by the following formula (17): (17) in, It is a vector with vector The angle between the two points. When the trajectory is a straight line, the angle is 0 and the cost is 0; as the turning angle increases, the cost increases significantly.

[0050] The optimal problem of trajectory planning under multiple constraints is expressed by the following formula (18): (18) Where C1 represents the communication rate constraint, t represents the time slot index, and u represents the user index; C2 represents the transmit power constraint. C3 indicates the maximum transmit power; C3 indicates the UAV's position constraints. C4 indicates a safe zone; C4 indicates a flight speed constraint. C5 indicates the maximum flight speed; C5 indicates a combined constraint on velocity and acceleration. Indicates speed, C6 represents acceleration; C6 represents acceleration constraint. C7 represents the acceleration amplitude; C7 represents the single-step maximum flight distance constraint. Indicates the time step. C8 represents the maximum allowable flight distance per time slot; C8 represents the energy budget constraint. C9 represents the total energy budget for the drone mission; C9 represents the bandwidth resource constraint. C10 represents the maximum available bandwidth of the system; C10 represents the total power constraint at the user level. This represents the total power allocated to user u at time t. Indicates the maximum permissible transmission power; where, denoted as the comprehensive cost objective function; z represents the latent variable in the latent space of the variational autoencoder, used to parameterize the UAV trajectory; C1 ensures that the communication rate meets the threshold; C2 and C10 limit the transmission power; C3 to C7 limit the kinematic parameters of the UAV, such as position, speed, acceleration, and flight distance; C8 and C9 limit the total energy budget and bandwidth resources, respectively.

[0051] Optionally, the multiple constraints include: communication rate constraints, transmit power constraints, UAV position constraints, flight speed constraints, speed and acceleration coordination constraints, acceleration constraints, single-step flight distance constraints, energy budget constraints, bandwidth resource constraints, and total user power constraints.

[0052] S4. Construct a variational autoencoder-based model; based on the optimality problem, construct a composite loss function that includes reconstruction error and constraint penalty terms; train the variational autoencoder-based model according to the composite loss function to obtain a trained variational autoencoder-based model.

[0053] Optionally, the variational autoencoder-based model of S4 includes: an encoder network and a decoder network; The encoder network consists of a multi-layer one-dimensional convolutional neural network, including an input layer, three convolutional layers, and a ReLU activation function; the encoder network is used to extract the temporal features of the trajectory. Among them, the input layer receiving dimension is The trajectory matrix, The time step is 9, which represents the three-dimensional position, three-dimensional velocity, and three-dimensional acceleration.

[0054] The decoder network adopts a structure symmetrical to the encoder network, which is used to map latent variables back to a high-dimensional trajectory space to generate a reconstructed trajectory.

[0055] Optionally, S4 trains the variational autoencoder-based model based on the loss function to obtain a trained variational autoencoder-based model, including: S41. Initial flight trajectory sample data is generated using a random walk algorithm; the initial flight trajectory sample data is normalized and input into the input layer of the encoder network based on the variational autoencoder model; through three convolutional layers and the ReLU activation function, the mean and log-variance of the latent variables are output; based on the mean of the latent variables, the reparameterization technique is used to sample random noise from the standard normal distribution, and a latent vector is generated through linear transformation. Each flight trajectory sample data contains a series of state vectors for the UAV, namely position, velocity, and acceleration. The obtained flight trajectory sample data is normalized to the interval [0,1].

[0056] In one feasible implementation, the process of sampling random noise from a standard normal distribution and generating a latent vector through linear transformation using a reparameterization technique is represented by the following formula (19): (19) in, Represents the latent vector; Indicates random noise. ; This represents the mean of the latent variables; The Hadamard product represents the element-wise product of a matrix or vector.

[0057] S42. Input the latent vector into the decoder network, and use the deconvolution layer to restore the latent vector to the trajectory matrix of the original dimension; S43. Based on the trajectory matrix, the hard constraints are transformed into differentiable soft constraints by constructing a composite loss function that includes reconstruction error and constraint penalty terms. The composite loss function is minimized by using the stochastic gradient descent algorithm, and the parameters of the encoder network and the decoder network are updated until the variational autoencoder model converges and the training is completed. The trained encoder network and the trained decoder network are then output.

[0058] In one feasible implementation, this embodiment of the invention employs a variational autoencoder-based model to transform trajectory planning into a probabilistic generation problem. Specifically, the variational autoencoder-based model is trained by maximizing a variational lower bound.

[0059] Optionally, the composite loss function that includes reconstruction error and constraint penalty term includes KL divergence loss, reconstruction loss and constraint penalty term; In one feasible implementation, the KL divergence loss is used to constrain the latent spatial distribution to approximate a standard normal distribution, as expressed by the following formula (20): (20) in, Represents the log-variance of the latent variable; This represents the square of the mean; Indicates variance.

[0060] In one feasible implementation, the reconstruction loss uses mean squared error to ensure that the generated trajectory is geometrically consistent with the distribution of the training data, as expressed by the following formula (21): (twenty one) Where K represents the number of samples; Indicates the currently generated trajectory; This represents the original input trajectory data.

[0061] In one feasible implementation, the constraint penalty term is the core of constraint optimization. This embodiment of the invention transforms hard constraints into differentiable soft constraints. For example, for communication rate constraints... The penalty term is constructed and expressed by the following formula (22):

[0062] (twenty two) in, This represents the penalty term corresponding to the communication rate constraint. Represents the first Lagrange multiplier; Indicates the rate threshold; This indicates the communication rate corresponding to the currently generated trajectory; Based on the calculated channel matrix and communication rate corresponding to the current generated trajectory, if the rate is lower than the set rate threshold, a gradient is generated to drive the network parameter updates. Similarly, penalty terms are constructed for constraints such as maximum flight distance, maximum speed, and safe zone.

[0063] In one feasible implementation, a constraint penalty loss function is constructed based on multiple constraints, and is expressed by the following formula (23): (twenty three) Specifically, formula (23) can be expanded into the following formula (24): (twenty four) in, Indicates communication quality; Indicates the maximum flight range; Indicates maximum speed; Indicates the magnitude of acceleration; Represents from the latent vector The generated trajectory is decoded; Let k denote a Lagrange multiplier, where k takes values ​​from 1 to 5.

[0064] In one feasible implementation, the multi-constraint penalty loss function, by introducing Lagrange multipliers and a squared penalty term, forces the generated trajectory to strictly satisfy physical constraints such as communication quality, maximum flight range, maximum speed, speed-acceleration coordination, and acceleration amplitude while minimizing flight costs. By training a variational autoencoder-based model, the trajectory planning system can directly sample from the latent space and generate the optimal trajectory.

[0065] In one feasible implementation, based on the above-mentioned reconstruction loss, constraint penalty term loss, and KL divergence loss, a composite overall loss function including reconstruction error and constraint penalty term is constructed, which is expressed by the following formula (25): (25) in, This represents the composite overall loss, which includes reconstruction error and constraint penalty terms. Indicates the reconstruction loss; The weighting coefficients represent the KL divergence loss. Indicates divergence loss; Let k represent a Lagrange multiplier, where k takes values ​​in the range [1,5]. Indicates a constraint or penalty item; This represents the mathematical expectation of the total flight cost of the current trajectory.

[0066] In one feasible implementation, based on the above-mentioned composite global loss function, the hard constraint is transformed into a differentiable soft constraint, that is, the composite loss function containing reconstruction error and constraint penalty term is expressed by the following formula (26): (26) in, This represents the composite loss, which includes reconstruction error and constraint penalty terms. Represents the parameters of the decoder neural network; These represent the parameters of the encoder neural network; This represents the input drone flight trajectory data; This represents the KL divergence, used to constrain the potential spatial distribution; Indicates the encoder network; This represents the prior probability distribution of the latent variable z; Let k represent a Lagrange multiplier, where k takes values ​​in the range [1,5]. Indicates the first The squared penalty term for the constraint; This represents the mathematical expectation of the total flight cost of the current trajectory.

[0067] in, The specific form is expressed by the following formulas (27)-(29): (27) (28) (29) in, This indicates a communication quality constraint penalty item; This indicates the penalty for the maximum flight distance constraint; This indicates the penalty for the maximum flight speed constraint; This indicates the actual communication quality of the currently generated trajectory; Indicates the communication threshold quality; Indicates the maximum flight speed; Indicates the maximum flight range.

[0068] S5. Obtain the current state data of the UAV, input the current state data of the UAV into the trained variational autoencoder-based model, and generate the optimal flight trajectory that satisfies multiple constraints.

[0069] In one feasible implementation, the current state data of the UAV includes: position, velocity, and acceleration; the current state data of the UAV is input into a trained variational autoencoder-based model, and the trained encoder network is used to extract features from the data to construct latent variables; the latent variables are input into a trained decoder for decoding, and the optimal flight trajectory that satisfies communication and perception constraints is output.

[0070] Among them, a trained decoder network is used to sample and generate the optimal flight trajectory that satisfies the communication and perception constraints from the potential space, and the communication UAV is driven to perform the task by issuing commands from the master UAV.

[0071] In one feasible implementation, during the inference phase, the main UAV only needs to randomly sample a latent vector from a standard normal distribution and input it into a trained decoder to generate an optimized flight trajectory within milliseconds. Based on the optimized flight trajectory, the main UAV sends the waypoint coordinate sequence of the trajectory to the communication UAV via control signaling. The communication UAV uses its onboard flight controller to track the trajectory, thereby achieving an optimal balance between energy consumption and safety while ensuring communication quality.

[0072] The embodiments of the present invention employ orthogonal time-frequency control technology to solve the Doppler frequency shift problem caused by high mobility, and combine it with the generative capability based on the variational autoencoder model to solve the problem of UAV trajectory planning under multiple constraints, thereby achieving synergistic optimization of communication reliability, perception continuity and flight energy consumption.

[0073] This invention achieves a synergistic improvement in communication reliability and perception continuity, significantly reducing the flight energy consumption and safety risks of UAVs, and enhancing the network's adaptability to dynamic environments. Using this invention can effectively reduce flight costs and improve communication stability.

[0074] Figure 4 This is a block diagram of a variational autoencoder trajectory planning device for a sensor-integrated unmanned aerial vehicle (UAV) network, provided by an embodiment of the present invention. This device is used in a variational autoencoder trajectory planning method for sensor-integrated UAV networks. (Refer to...) Figure 4 The device includes a first construction unit 410, a computing unit 420, a second construction unit 430, a training unit 440, and a generation unit 450. Wherein:

[0075] The first construction unit 410 is used to construct a two-layer UAV cooperative network architecture; and to configure orthogonal time-frequency control parameters and antenna array parameters based on the network architecture. The calculation unit 420 is used to extract the delay-Doppler domain channel parameters and calculate the communication rate based on the orthogonal time-frequency control parameters and antenna array parameters, using orthogonal time-frequency control technology. The second construction unit 430 is used to define a comprehensive cost objective function that includes weights for distance, safety, turning, and altitude; and to construct an optimal trajectory planning problem with multiple constraints based on the comprehensive cost objective function and the communication rate. A training unit is used to construct a variational autoencoder-based model; based on the optimal problem, a composite loss function containing reconstruction error and constraint penalty term is constructed; the variational autoencoder-based model is trained according to the composite loss function to obtain a trained variational autoencoder-based model; The generation unit 450 is used to acquire the current state data of the UAV, input the current state data of the UAV into the trained variational autoencoder-based model, and generate the optimal flight trajectory that satisfies multiple constraints.

[0076] Optionally, the computing unit 420 is used for: Using orthogonal time-frequency modulation technology, the transmitter maps the delay-Doppler domain data symbols to the time-frequency domain through inverse symplectic finite Fourier transform to obtain the time-frequency domain signal; The Heisenberg transform is used to convert the time-frequency domain signal into a continuous-time transmitted waveform; Establish a downlink sensing channel model and obtain the channel impulse response of the main UAV's transmitted signal during propagation; A multi-input multi-output orthogonal time-frequency space-space channel model for inter-UAV communication is established, and the physical channel is mapped to a discrete delay-Doppler grid to obtain the equivalent channel response; The communication rate is calculated based on the channel impulse response and the equivalent channel response.

[0077] Optionally, the comprehensive cost objective function, which includes weights for distance, safety, steering, and altitude, is expressed by the following formula (1): (1) in, Represent the overall cost objective function; The distance cost is the sum of the Euclidean distances between adjacent waypoints; This represents the safety cost, which is a penalty based on the safety distance threshold and the distance to the obstacle. The altitude cost is the sum of the absolute values ​​of altitude changes between adjacent waypoints. Indicates the cost of switching; Indicates the first weighting coefficient; This represents the second weighting coefficient; This represents the third weighting coefficient; This represents the fourth weighting coefficient.

[0078] Optionally, the multiple constraints include: communication rate constraints, transmit power constraints, UAV position constraints, flight speed constraints, speed and acceleration coordination constraints, acceleration constraints, single-step flight distance constraints, energy budget constraints, bandwidth resource constraints, and total user power constraints.

[0079] Optionally, the variational autoencoder-based model includes: an encoder network and a decoder network; The encoder network consists of a multi-layer one-dimensional convolutional neural network, including an input layer, three convolutional layers, and a ReLU activation function; the encoder network is used to extract the temporal features of the trajectory. The decoder network adopts a structure symmetrical to the encoder network, which is used to map latent variables back to a high-dimensional trajectory space to generate a reconstructed trajectory.

[0080] Optionally, the training unit 440 is used for: The initial flight trajectory sample data is generated using a random walk algorithm. The initial flight trajectory sample data is normalized and input into the input layer of the encoder network based on the variational autoencoder model. Through three convolutional layers and the ReLU activation function, the mean and log-variance of the latent variables are output. Based on the mean of the latent variables, the reparameterization technique is used to sample random noise from the standard normal distribution and generate latent vectors through linear transformation. The latent vectors are input into the decoder network, and the latent vectors are restored to the trajectory matrix of the original dimension through the deconvolution layer; Based on the trajectory matrix, the hard constraints are transformed into differentiable soft constraints by constructing a composite loss function that includes reconstruction error and constraint penalty terms. The composite loss function is minimized by the stochastic gradient descent algorithm, and the parameters of the encoder network and the decoder network are updated until the variational autoencoder model converges and the training is completed, outputting the trained encoder network and the trained decoder network.

[0081] Optionally, the composite loss function including reconstruction error and constraint penalty term includes KL divergence loss, reconstruction loss and constraint penalty term; The composite loss function, which includes reconstruction error and constraint penalty term, is expressed by the following formula (2): (2) in, This represents the composite loss, which includes reconstruction error and constraint penalty terms. Represents the parameters of the decoder neural network; These represent the parameters of the encoder neural network; This represents the input drone flight trajectory data; This represents the KL divergence, used to constrain the potential spatial distribution; Indicates the encoder network; This represents the prior probability distribution of the latent variable z; Let k represent a Lagrange multiplier, where k takes values ​​in the range [1,5]. Indicates the first The squared penalty term for the constraint; This represents the mathematical expectation of the total flight cost of the current trajectory.

[0082] The embodiments of the present invention employ orthogonal time-frequency control technology to solve the Doppler frequency shift problem caused by high mobility, and combine it with the generative capability based on the variational autoencoder model to solve the problem of UAV trajectory planning under multiple constraints, thereby achieving synergistic optimization of communication reliability, perception continuity and flight energy consumption.

[0083] This invention achieves a synergistic improvement in communication reliability and perception continuity, significantly reducing the flight energy consumption and safety risks of UAVs, and enhancing the network's adaptability to dynamic environments. Using this invention can effectively reduce flight costs and improve communication stability.

[0084] Figure 5 This is a schematic diagram of the structure of a variational autoencoder trajectory planning device for an integrated sensor-unmanned aerial vehicle (UAV) network, as provided in an embodiment of the present invention. Figure 5 As shown, the variational autoencoder trajectory planning device for integrated sensor-unmanned aerial vehicle (UAV) networks may include the above-mentioned... Figure 4 The illustrated variational autoencoder trajectory planning device for a sensor-integrated unmanned aerial vehicle (UAV) network. Optionally, the variational autoencoder trajectory planning device 510 for a sensor-integrated UAV network may include a first processor 2001.

[0085] Optionally, the variational autoencoder trajectory planning device 510 for a sensor-integrated unmanned aerial vehicle (UAV) network may also include a memory 2002 and a transceiver 2003.

[0086] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0087] The following is combined Figure 5 The components of the variational autoencoder trajectory planning device 510 for integrated sensor-unmanned aerial vehicle (UAV) networks are described in detail below: The first processor 2001 is the control center of the variational autoencoder trajectory planning device 510 for the integrated sensor-unmanned aerial vehicle (UAV) network. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0088] Optionally, the first processor 2001 can perform various functions of the variational autoencoder trajectory planning device 510 for the sensor-integrated unmanned aerial vehicle network by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0089] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 5 CPU0 and CPU1 are shown in the diagram.

[0090] In a specific implementation, as one example, the variational autoencoder trajectory planning device 510 for integrated sensor-unmanned aerial vehicle (UAV) networks may also include multiple processors, such as... Figure 5 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0091] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0092] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or exist independently, and may be connected via the interface circuit of the variational autoencoder trajectory planning device 510 for the integrated sensor-unmanned aerial vehicle network. Figure 5 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0093] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0094] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 5(Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0095] Optionally, the transceiver 2003 can be integrated with the first processor 2001 or exist independently, and can be connected via the interface circuit of the variational autoencoder trajectory planning device 510 for a sensor-integrated unmanned aerial vehicle network. Figure 5 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0096] It should be noted that, Figure 5 The structure of the variational autoencoder trajectory planning device 510 for a sensor-integrated unmanned aerial vehicle (UAV) network shown in the figure does not constitute a limitation on the router. Actual variational autoencoder trajectory planning devices for sensor-integrated UAV networks may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0097] Furthermore, the technical effects of the variational autoencoder trajectory planning device 510 for integrated sensory drone networks can be referenced from the technical effects of the variational autoencoder trajectory planning method for integrated sensory drone networks described in the above method embodiments, and will not be repeated here.

[0098] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or it may be any conventional processor, etc.

[0099] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0100] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0101] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0102] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0103] It should be understood that, in various embodiments of the present invention, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0104] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0105] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0106] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0107] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0108] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0109] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0110] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A variational autoencoder trajectory planning method for sensor-integrated unmanned aerial vehicle (UAV) networks, characterized in that, The method includes: S1. Construct a two-layer UAV collaborative network architecture; configure orthogonal time-frequency control parameters and antenna array parameters based on the network architecture; S2. Based on the orthogonal time-frequency control parameters and antenna array parameters, the orthogonal time-frequency control technology is used to extract the delay-Doppler domain channel parameters and calculate the communication rate. S3. Define a comprehensive cost objective function that includes weights for distance, safety, turning, and altitude; based on the comprehensive cost objective function and communication rate, construct an optimal trajectory planning problem with multiple constraints. S4. Construct a variational autoencoder-based model; based on the optimal problem, construct a composite loss function that includes reconstruction error and constraint penalty term; train the variational autoencoder-based model according to the composite loss function to obtain a trained variational autoencoder-based model. S5. Obtain the current state data of the UAV, input the current state data of the UAV into the trained variational autoencoder-based model, and generate the optimal flight trajectory that satisfies multiple constraints.

2. The variational autoencoder trajectory planning method for sensor-integrated unmanned aerial vehicle (UAV) networks according to claim 1, characterized in that, S2, based on the orthogonal time-frequency modulation parameters and antenna array parameters, employs orthogonal time-frequency modulation technology to extract delay-Doppler domain channel parameters and calculate the communication rate, including: S21. Using orthogonal time-frequency modulation technology, the transmitting end maps the delay-Doppler domain data symbols to the time-frequency domain through inverse symplectic finite Fourier transform to obtain the time-frequency domain signal; S22. Use Heisenberg transform to convert the time-frequency domain signal into a continuous-time transmitted waveform; S23. Establish a downlink sensing channel model and obtain the channel impulse response of the main UAV's transmitted signal during the propagation process; S24. Establish a multi-input multi-output orthogonal time-frequency space-space channel model for inter-UAV communication, map the physical channel to a discrete delay-Doppler grid, and obtain the equivalent channel response; S25. Calculate the communication rate based on the channel impulse response and the equivalent channel response.

3. The variational autoencoder trajectory planning method for sensor-integrated unmanned aerial vehicle (UAV) networks according to claim 1, characterized in that, The comprehensive cost objective function of S3, which includes distance, safety, steering, and altitude weights, is expressed by the following formula (1): （1） in, Represent the overall cost objective function; The distance cost is the sum of the Euclidean distances between adjacent waypoints; This represents the safety cost, which is a penalty based on the safety distance threshold and the distance to the obstacle. The altitude cost is the sum of the absolute values ​​of altitude changes between adjacent waypoints. Indicates the cost of switching; Indicates the first weighting coefficient; This represents the second weighting coefficient; This represents the third weighting coefficient; This represents the fourth weighting coefficient.

4. The variational autoencoder trajectory planning method for sensor-integrated unmanned aerial vehicle (UAV) networks according to claim 1, characterized in that, The multiple constraints include: communication rate constraint, transmission power constraint, UAV position constraint, flight speed constraint, speed and acceleration coordination constraint, acceleration constraint, single-step flight distance constraint, energy budget constraint, bandwidth resource constraint, and total user power constraint.

5. The variational autoencoder trajectory planning method for sensor-integrated unmanned aerial vehicle (UAV) networks according to claim 1, characterized in that, The variational autoencoder-based model of S4 includes: an encoder network and a decoder network; The encoder network consists of a multi-layer one-dimensional convolutional neural network, including an input layer, three convolutional layers, and a ReLU activation function; the encoder network is used to extract the temporal features of the trajectory. The decoder network adopts a structure symmetrical to the encoder network, which is used to map latent variables back to a high-dimensional trajectory space to generate a reconstructed trajectory.

6. The variational autoencoder trajectory planning method for sensor-integrated unmanned aerial vehicle (UAV) networks according to claim 5, characterized in that, The step S4 involves training the variational autoencoder-based model based on the composite loss function to obtain a trained variational autoencoder-based model, including: S41. Initial flight trajectory sample data is generated using a random walk algorithm; the initial flight trajectory sample data is normalized and input into the input layer of the encoder network based on the variational autoencoder model; through three convolutional layers and the ReLU activation function, the mean and log-variance of the latent variables are output; based on the mean of the latent variables, the reparameterization technique is used to sample random noise from the standard normal distribution, and a latent vector is generated through linear transformation. S42. Input the latent vector into the decoder network, and use the deconvolution layer to restore the latent vector to the trajectory matrix of the original dimension; S43. Based on the trajectory matrix, the hard constraints are transformed into differentiable soft constraints by constructing a composite loss function that includes reconstruction error and constraint penalty terms. The composite loss function is minimized by using the stochastic gradient descent algorithm, and the parameters of the encoder network and the decoder network are updated until the variational autoencoder model converges and the training is completed. The trained encoder network and the trained decoder network are then output.

7. The variational autoencoder trajectory planning method for sensor-integrated unmanned aerial vehicle (UAV) networks according to claim 1, characterized in that, The composite loss function, which includes reconstruction error and constraint penalty term, comprises KL divergence loss, reconstruction loss, and constraint penalty term; The composite loss function, which includes reconstruction error and constraint penalty term, is expressed by the following formula (2): （2） in, This represents the composite loss, which includes reconstruction error and constraint penalty terms. Represents the parameters of the decoder neural network; These represent the parameters of the encoder neural network; This represents the input drone flight trajectory data; This represents the KL divergence, used to constrain the potential spatial distribution; Indicates the encoder network; This represents the prior probability distribution of the latent variable z; Let k represent a Lagrange multiplier, where k takes values ​​in the range [1,5]. Indicates the first The squared penalty term for the constraint; This represents the mathematical expectation of the total flight cost of the current trajectory.

8. A variational autoencoder trajectory planning device for a sensor-integrated unmanned aerial vehicle (UAV) network, wherein the variational autoencoder trajectory planning device for a sensor-integrated UAV network is used to implement the variational autoencoder trajectory planning method for a sensor-integrated UAV network as described in any one of claims 1-7, characterized in that, The device includes: The first building unit is used to construct a two-layer UAV collaborative network architecture; and to configure orthogonal time-frequency control parameters and antenna array parameters based on the network architecture. The calculation unit is used to extract the delay-Doppler domain channel parameters and calculate the communication rate based on the orthogonal time-frequency control parameters and antenna array parameters, using orthogonal time-frequency control technology. The second building unit is used to define a comprehensive cost objective function that includes weights for distance, safety, turning, and altitude; and to construct an optimal trajectory planning problem with multiple constraints based on the comprehensive cost objective function and the communication rate. A training unit is used to construct a variational autoencoder-based model; based on the optimal problem, a composite loss function containing reconstruction error and constraint penalty term is constructed; the variational autoencoder-based model is trained according to the composite loss function to obtain a trained variational autoencoder-based model; The generation unit is used to acquire the current state data of the UAV, input the current state data of the UAV into the trained variational autoencoder-based model, and generate the optimal flight trajectory that satisfies multiple constraints.

9. A variational autoencoder trajectory planning device for integrated sensor-unmanned aerial vehicle (UAV) networks, characterized in that, The variational autoencoder trajectory planning device for integrated sensor-unmanned aerial vehicle (UAV) networks includes: processor; A memory storing computer-readable instructions that, when executed by the processor, implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 7.

Images