Convolutional neural network-based integrated navigation and communication emergency unmanned aerial vehicle deployment method

Abstract

The application provides a convolutional neural network-based integrated communication and positioning emergency unmanned aerial vehicle deployment method, which comprises the following steps: acquiring user distribution data in a current emergency rescue scene, and inputting the user distribution data after feature enhancement based on Gaussian enhancement into a trained convolutional neural network to obtain an optimal deployment position of an unmanned aerial vehicle group; the trained convolutional neural network is obtained through offline training of a predetermined target data set; the target data set is composed of optimal deployment positions of the unmanned aerial vehicle group under historical user distribution data; the optimal deployment positions of the unmanned aerial vehicle group under the historical user distribution data are obtained by solving an unmanned aerial vehicle deployment optimization model through a dynamic particle swarm algorithm; the unmanned aerial vehicle deployment optimization model takes the maximization of a user communication and positioning joint performance function as an objective, and takes signal quality, flight safety performance and load capacity of the unmanned aerial vehicle in the communication and positioning service process as constraints; the communication and positioning joint performance function represents communication performance and positioning performance of the user.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) technology, and in particular to a method for deploying emergency UAVs based on convolutional neural networks that integrates communication and navigation. Background Technology

[0002] During emergency rescue operations, it is essential to restore two critical services: communication and positioning at the disaster site. Unmanned Aerial Vehicles (UAVs) can quickly determine network locations using their onboard base stations, providing wireless coverage to users without relying on existing infrastructure. Due to their inherent advantages in mobility and flexibility, they can provide stable and reliable integrated emergency communication and positioning services in scenarios where ground infrastructure has been damaged after natural disasters.

[0003] Although drone base stations have overcome geographical limitations to some extent, drone-assisted communication and navigation emergency rescue still faces many severe challenges. How to quickly deploy drones in mountainous and forested areas with limited drone resources to meet the communication and positioning needs of ground users remains an urgent problem to be solved. Summary of the Invention

[0004] To address the aforementioned problems in the existing technology, this application provides a method for deploying an integrated communication and navigation emergency drone based on a convolutional neural network.

[0005] In a first aspect, embodiments of this application provide a method for deploying an integrated communication and navigation emergency drone based on a convolutional neural network, including:

[0006] Obtain user distribution data in the current emergency rescue scenario, and perform feature enhancement on the user distribution data based on Gaussian enhancement;

[0007] The user distribution data after feature enhancement is input into a trained convolutional neural network to obtain the optimal deployment location of the drone group.

[0008] The trained convolutional neural network is obtained through offline training on a pre-determined target dataset; the target dataset consists of the optimal deployment locations of drone groups under historical user distribution data; the optimal deployment locations of drone groups under historical user distribution data are obtained by solving a pre-determined drone deployment optimization model using a dynamic particle swarm optimization algorithm.

[0009] The UAV deployment optimization model aims to maximize the user's communication and navigation joint performance function, while taking into account the UAV's signal quality, flight safety performance, and payload capacity during communication and positioning services. The communication and navigation joint performance function is used to characterize the user's communication and positioning performance.

[0010] Secondly, embodiments of this application provide a communication and navigation integrated emergency drone deployment device based on a convolutional neural network, comprising:

[0011] The first acquisition module is used to acquire user distribution data in the current emergency rescue scenario, and to perform feature enhancement on the user distribution data based on Gaussian enhancement;

[0012] The second acquisition module is used to input the user distribution data after feature enhancement into the trained convolutional neural network to obtain the optimal deployment location of the drone group.

[0013] The trained convolutional neural network is obtained through offline training on a pre-determined target dataset; the target dataset consists of the optimal deployment locations of drone groups under historical user distribution data; the optimal deployment locations of drone groups under historical user distribution data are obtained by solving a pre-determined drone deployment optimization model using a dynamic particle swarm optimization algorithm.

[0014] The UAV deployment optimization model aims to maximize the user's communication and navigation joint performance function, while taking into account the UAV's signal quality, flight safety performance, and payload capacity during communication and positioning services. The communication and navigation joint performance function is used to characterize the user's communication and positioning performance.

[0015] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements any of the above-described methods for deploying emergency unmanned aerial vehicles based on convolutional neural networks that integrate communication and navigation.

[0016] Fourthly, embodiments of this application also provide a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described methods for deploying emergency unmanned aerial vehicles based on convolutional neural networks.

[0017] Fifthly, embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements any of the above-described methods for deploying emergency unmanned aerial vehicles based on convolutional neural networks.

[0018] The communication and navigation integrated emergency drone deployment method provided in this application uses a target dataset consisting of the optimal deployment locations of drone groups under historical user distribution data to train the convolutional neural network offline. The convolutional neural network learns the mapping relationship between user distribution and the optimal deployment location of drone groups in emergency rescue scenarios, enabling the direct use of the trained convolutional neural network and current user distribution data to quickly obtain the optimal deployment location of drone groups, meeting the high real-time requirements of emergency rescue scenarios. Simultaneously, Gaussian enhancement is used to enhance the features of the original user distribution data, solving the gradient vanishing problem caused by sparse user distribution. Furthermore, considering the communication and positioning needs in emergency rescue scenarios, a drone deployment optimization model is designed with the communication and navigation joint performance function as the objective and the signal quality, flight safety performance, and load capacity of drones in communication and positioning services as constraints. The solution of this optimization model serves as the optimal deployment location of drone groups under historical user distribution data, making the training dataset of the CNN model more closely aligned with the specific scenarios and the needs of multi-service communication and positioning. Additionally, a dynamic particle swarm optimization algorithm is used to solve the drone deployment optimization model, accelerating the algorithm's convergence speed while effectively preventing the algorithm from getting trapped in local optima, thus improving the quality of the CNN model's training dataset. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of a scenario for the deployment method of an emergency drone based on a convolutional neural network, provided in an embodiment of this application.

[0021] Figure 2 This is one of the flowcharts illustrating the integrated communication and navigation emergency drone deployment method based on convolutional neural networks provided in this application embodiment;

[0022] Figure 3 This is the second flowchart illustrating the integrated communication and navigation emergency drone deployment method based on convolutional neural networks provided in this application embodiment;

[0023] Figure 4 This is a schematic diagram of the structure of the emergency drone deployment device based on a convolutional neural network provided in the embodiments of this application;

[0024] Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Figure 1 This is a schematic diagram of a scenario illustrating the drone deployment method provided in this application embodiment, such as... Figure 1 As shown, this example illustrates an integrated communication and navigation system for emergency rescue in a mountainous forest environment. The system includes at least one emergency communication vehicle, M drones, and N users (essentially referring to N user devices). The drone set and user set are defined as follows: and

[0027] Each user holds a handheld communication and positioning terminal device, and the drone is equipped with a base station (BS) to provide communication and positioning services to users within its coverage area. The emergency communication vehicle is a centralized controller that runs the proposed deployment algorithm to achieve unified scheduling of drones.

[0028] Figure 2 This is one of the flowcharts illustrating the integrated communication and navigation emergency drone deployment method based on convolutional neural networks provided in this application embodiment, such as... Figure 2 As shown, the method includes at least the following steps:

[0029] Step 201: Obtain user distribution data in the current emergency rescue scenario, and perform feature enhancement on the user distribution data based on Gaussian enhancement.

[0030] Step 202: Input the user distribution data after feature enhancement into the trained convolutional neural network to obtain the optimal deployment location of the drone group;

[0031] The trained convolutional neural network is obtained through offline training on a pre-determined target dataset; the target dataset consists of the optimal deployment locations of drone groups under historical user distribution data; the optimal deployment locations of drone groups under historical user distribution data are obtained by solving a pre-determined drone deployment optimization model using a dynamic particle swarm optimization algorithm.

[0032] The UAV deployment optimization model aims to maximize the user's communication and navigation joint performance function, while taking into account the UAV's signal quality, flight safety performance, and payload capacity during communication and positioning services. The communication and navigation joint performance function is used to characterize the user's communication and positioning performance.

[0033] Specifically, for emergency rescue scenarios, quickly obtaining a drone deployment plan is crucial. A well-trained deep learning model can rapidly perform online simulations, meeting the high time requirements of emergency rescue scenarios. Furthermore, considering the spatial relationships among users distributed on the ground, and the significant advantages of Convolutional Neural Networks (CNNs) in extracting spatial features of data, this application utilizes CNNs to achieve emergency drone deployment.

[0034] Considering issues such as gradient vanishing caused by sparse user distribution, after obtaining the original user distribution data in the current emergency rescue scenario, the original user distribution data is enhanced using a Gaussian distribution-based feature enhancement method, namely Gaussian enhancement, and then input into a trained CNN to obtain the optimal deployment location of the drone group.

[0035] By using a target dataset comprised of the optimal deployment locations of drone teams based on historical user distribution data, a CNN is trained offline. This enables the CNN to learn the mapping relationship between user distribution and the optimal deployment locations of drone teams in emergency rescue scenarios. Therefore, during actual rescue operations, the trained CNN model can be directly used, with the current user distribution data in the emergency rescue scenario as input, to quickly determine the optimal deployment locations of drone teams.

[0036] In one possible implementation, after obtaining the target dataset D consisting of the optimal deployment locations of the drone group based on historical user distribution data, a CNN model is trained to obtain the mapping function from user distribution to drone location information. This process is described in the following three parts:

[0037] Input: Without loss of generality, divide the ground area into L×L grids, and define the user distribution matrix O of the k-th data. k In the input layer, feature enhancement is performed on the original user distribution data to address issues such as gradient vanishing caused by sparse user distribution. Optionally, the grid position of user n is defined as (i... n ,j n The value O in the i-th row and j-th column of the matrix is ​​calculated using Gaussian enhancement. k,(i,j) It can be represented as:

[0038]

[0039] Here, ξ represents the Gaussian enhancement standard deviation. This method enriches the image data without interfering with the original user distribution data, making it more conducive to CNN learning.

[0040] Network Structure: The CNN model consists of a feature extraction layer and a decision mapping layer. The feature extraction layer comprises two convolutional layers: the first convolutional layer consists of a 64-pass 7×7 convolutional kernel and a max-pooling layer; the second convolutional layer consists of a 128-pass 3×3 convolutional kernel and a max-pooling layer. The decision mapping layer uses a multilayer perceptron (MLP), with its input layer, hidden layer, and output layer consisting of 1024, 2048, and 3M neurons, respectively; all neurons in the network use ReLU as the activation function.

[0041] Output: The output of the CNN corresponds to the position decision of the drone group, which can be expressed as:

[0042] U=[x1,y1,z1;x2,y2,z2;…,x M ,y M ,z M ]

[0043] Where U represents the deployment location of the drone group, x i ,y i ,z i Let represent the three-dimensional position coordinates of the i-th UAV.

[0044] Optionally, since the output is coordinates, the mean squared error (MSE) is used as the loss function for the network, similar to traditional prediction tasks. Furthermore, given that deep learning is prone to overfitting, a pruning probability can be set for each neuron during training to prevent overfitting. The neural network randomly removes neurons each time, with each neuron having the same pruning probability p. The parameters of the remaining network structure are then trained, and the product of probability p and the trained weights is used as the trained weights.

[0045] Furthermore, in emergency rescue scenarios, rescue personnel not only need network access to maintain communication connections and interact with command centers and other personnel for data transmission, but also need to frequently perform self-location checks to ensure safe rescue and evacuation. Therefore, in order to carry out rescue operations efficiently, it is necessary to restore two key services at the disaster site: communication and positioning.

[0046] In existing technologies, research on UAV deployment often focuses on either communication or positioning alone, rarely considering both communication and positioning requirements simultaneously. Therefore, this application considers the integration of multiple services, including communication and positioning. Furthermore, simply using machine learning or deep learning to solve the UAV deployment problem often yields limited results. Therefore, this application further considers combining the integration of multiple services, including communication and positioning, with deep learning to address the UAV deployment issue.

[0047] Optionally, the optimal deployment location of the UAV group under the historical user distribution data is obtained by solving a pre-determined UAV deployment optimization model using a dynamic particle swarm optimization algorithm; wherein, the UAV deployment optimization model aims to maximize the user's communication and navigation joint performance function, and is constrained by the UAV's signal quality, flight safety performance, and load capacity during communication and positioning services; the communication and navigation joint performance function is used to characterize the user's communication performance and positioning performance.

[0048] Specifically, the optimal deployment locations of drone groups under historical user distribution data are determined and used as the training dataset for the CNN. These optimal deployment locations are determined based on a pre-defined drone deployment optimization model. In other words, the optimal solution of the pre-defined drone deployment optimization model is used as the optimal deployment location for the drone groups under the historical user distribution data.

[0049] The UAV deployment optimization model aims to maximize the user's joint communication and navigation performance function, while constraining the UAV's signal quality, flight safety performance, and payload capacity during communication and positioning services. The joint communication and navigation performance function characterizes the user's communication and positioning performance.

[0050] Specifically, emergency rescue often occurs in forest and mountainous areas. Dense forests have complex environments with many trees blocking the signal, resulting in more severe and irregular signal attenuation. This makes existing urban and rural channel models unusable directly. Therefore, the characteristics of mountainous and forest areas should be fully considered, and UAV deployment design should be optimized based on the channel model specific to forest areas.

[0051] Combination Figure 1Considering the downlink transmission process of UAV base station signals, a signal service cycle is divided into two phases chronologically: the communication phase and the positioning phase. During the communication phase, each user can establish a communication connection with at most one UAV. During the positioning phase, each user establishes positioning connections with three UAVs before location calculation. Within each service cycle, all UAVs share the same frequency band, and each UAV base station uses Frequency Division Multiple Access (FDMA) technology to provide communication and positioning services to users. Therefore, the communication and positioning phases are orthogonal in time, and all users connected to each UAV do not interfere with each other.

[0052] Establish a three-dimensional Cartesian coordinate system, where the coordinates of the UAV m and the user n can be represented as u. m =[x m ,y m ,z m ] and o n =[x n ,y n If ,0], then the spatial distance d between drone m and user n is... m,n and horizontal distance r m,n They can be represented as:

[0053]

[0054]

[0055] To ensure the safety of drones, in order to avoid collisions, the spatial distance between any two drones must be no less than the minimum spatial distance d. min ,Right now On the other hand, to avoid collisions between drones and mountains or trees, the drone's flight altitude must not be less than the minimum flight altitude h. min ,Right now

[0056] Since the drone and the user are in a forest environment, a forest channel model is adopted. Considering the large-scale fading of wireless signal propagation, i.e., free space path loss (FSPL), and the shadowing effect, the total path loss (in dB) between the drone m and the user n can be expressed as:

[0057]

[0058] in, X represents the free space path loss when the reference distance between the drone m and the user n is d0, α represents the path loss factor, and X represents the path loss factor. σThis represents shadow fading, where the shadow fading is a zero-mean Gaussian random variable with a standard deviation of σ. This represents the additional losses incurred between drone m and user n in the forest area.

[0059] Free space path loss This can be further expressed as:

[0060]

[0061] Where f represents the carrier frequency, d0 represents the reference distance between the drone m and the user n, and c represents the speed of light.

[0062] Shadow Decay X σ The probability density function can be further expressed as:

[0063]

[0064] Additional losses This can be further expressed as:

[0065]

[0066] Where A, C, E, G, and H represent environmental parameters, This represents the radian value from user n to drone m.

[0067] The channel power gain from UAV m to user n can be expressed as:

[0068]

[0069] During the communication phase, matrix A is defined. C To represent the communication relationship between the drone and the user, a binary variable is defined. (A C The row (m-n) indicates whether drone m provides communication services to user n. That is, when drone m provides communication services to user n... otherwise To ensure that the served users can achieve a reasonable communication rate, it is assumed that each drone can serve a maximum of K users during a single communication phase. C One user.

[0070] Because each drone reuses the same frequency, when Consider the signal-to-interference-plus-noise ratio (SINR) γ received by user n from drone m. m,n for:

[0071]

[0072] Where, pm,n g represents the transmit power allocated by drone m to user n. m,n This represents the channel power gain from drone m to user n. Let ψ represent the interference caused by other drones, and ψ be the ambient noise power. Assume that each drone distributes its power equally among all the users it serves, i.e.:

[0073]

[0074] Where, p max This indicates the total transmission power of the drone. This indicates whether drone m provides communication services to user j. To ensure the quality of the communication signal for the served user, the SINR value of the served user needs to be greater than the minimum signal-to-noise ratio of the communication signal.

[0075] When drones provide communication services to users, throughput is used to measure communication performance. According to Shannon's formula, the achievable communication rate R for user n is... n for:

[0076] R n =b m.n log(1+γ m,n )

[0077] Among them, b m.n Let m represent the bandwidth allocated by drone m to user n. Assuming each drone distributes its power equally among all users it serves, and the total available bandwidth of the drones is B, then the following holds true:

[0078]

[0079] During the positioning phase, ground user equipment calculates the location of the UAV by estimating the Time Difference of Arrival (TDOA) of the signals transmitted by the UAV. The UAV is equipped with a Global Positioning System (GPS) module, which can obtain its own location via satellite.

[0080] Similar to the communication phase, in the positioning phase, matrix A is defined. P To represent the location association between the drone and the user, a binary variable is defined. (A P The m-th row and n-th column () represents whether drone m provides location services to user n, i.e., when drone m provides location services to user n. otherwise To ensure the quality of the positioning signals for the served users, it is assumed that each drone can serve a maximum of K users during the positioning phase. PFor each user, the SINR value of the location signal must be greater than the minimum signal-to-noise ratio of the location signal.

[0081] When using drone swarms to locate users, positioning performance is measured by positioning accuracy. Positioning accuracy is measured by factors such as the coverage area of ​​the drone base station and the complexity of the drone swarm deployment.

[0082] In some embodiments, the positioning performance is determined based on the user's location accuracy factor and a pre-set location accuracy factor threshold.

[0083] Specifically, considering the needs of engineering practice, this application uses the Position Dilution of Precision (PDOP) to measure the geometric relationship of the positioning anchor node, that is, to evaluate the positioning results of the UAV group, and thus as an indicator to measure the positioning accuracy.

[0084] When the location of user n can be calculated, this application defines the UAV positioning service group for user n as follows: Assuming that the ranging information error of each drone is independent and identically distributed, the PDOP of user n can be expressed as:

[0085]

[0086] Among them, H n The measurement matrix is ​​for the navigation system. Assume the drone k... n,1 For the TDOA reference station of the location service group, For drones k n,1 The three-dimensional distance to user n is H. n It can be represented as:

[0087]

[0088] To ensure compatibility with both location-enabled and non-location-enabled users in terms of location performance, this application defines a piecewise function G. n Let's represent the location performance of user n:

[0089]

[0090] Where, ρ max The position accuracy factor threshold is a pre-set value. If the position accuracy factor of a drone group is greater than this threshold, the positioning error caused by the drone group is too large and it can be regarded as invalid positioning.

[0091] The emergency drone deployment method based on convolutional neural networks provided in this application determines the user's positioning performance through PDOP and a pre-set PDOP threshold value, which is more suitable for the needs of actual engineering.

[0092] In some embodiments, the method further includes:

[0093] The user's communication performance and positioning performance are determined based on a pre-determined forest area channel model;

[0094] The communication performance and the positioning performance are weighted to determine the joint communication and navigation performance function.

[0095] Specifically, after determining the user's communication and positioning performance using the forest area channel model, the communication and positioning performance are weighted to obtain the user's joint communication and navigation performance function.

[0096] Optionally, the joint performance function of the conduction and passivity of user n satisfies:

[0097]

[0098] Where λ and μ (0<λ<1, 0<μ<1) represent the communication and positioning performance weights, respectively.

[0099] Optionally, since communication rate and positioning accuracy have different dimensions and orders of magnitude, in order to eliminate the influence caused by the different dimensions between optimization indicators, this application adopts the minimax normalization method for R. n and The two optimization metrics are standardized. The normalization function is as follows:

[0100]

[0101] Optionally, after obtaining the user's joint communication and navigation performance function, the constraints of the UAV deployment optimization model are further considered. Specifically, the constraints are the signal quality, flight safety performance, and payload capacity of the UAV during communication and positioning services.

[0102] The emergency drone deployment method based on convolutional neural networks provided in this application takes into account the complexity of the forest and mountainous environment and pre-designs a forest channel model. On this basis, it also considers the communication and positioning needs in emergency rescue scenarios and designs a drone deployment optimization model with the joint communication and navigation performance function as the objective and the signal quality, flight safety performance and load capacity of the drone in the communication and positioning service process as constraints. The solution of this optimization model is used as the optimal deployment location of the drone group under historical user distribution data, so that the training dataset of the CNN model is more in line with the needs of special scenarios and multiple communication and positioning services.

[0103] In some embodiments, the optimal deployment location of the drone group based on the historical user distribution data is determined through the following steps:

[0104] The joint communication and navigation performance function that maximizes the user is decomposed into a sub-problem of UAV-user association and a sub-problem of UAV deployment location optimization.

[0105] Based on the user's communication link benefits and positioning link benefits, determine the association strategy between the drone and the user;

[0106] Given a defined association strategy between drones and users, the optimal solution to the drone deployment location optimization subproblem is determined using a dynamic particle swarm optimization algorithm. This optimal solution is then used as the optimal deployment location for the drone group based on the historical user distribution data.

[0107] Specifically, the problem of maximizing the joint performance function of user communication and navigation can be decomposed into two sub-problems: the UAV-user relationship sub-problem and the UAV deployment location optimization sub-problem. The relationship between the UAV and the user includes communication relationship and positioning relationship.

[0108] Based on the analysis above, let This indicates the deployment strategy of the drone team; all drones are deployed directly above the disaster area. This application describes the association strategy between drones and users, and its objective is to jointly optimize the deployment strategy of drone groups. And the strategy of associating drones with users This maximizes the communication and navigation fusion performance for all ground users. The optimization objective is:

[0109]

[0110] In one possible implementation, the design constraints of this application include:

[0111] C1:

[0112] C2:

[0113] C3:

[0114] C4:

[0115] C5:

[0116] C6:

[0117] C7:

[0118] C8:

[0119] Wherein, C1 indicates that each user can be provided with communication services by a maximum of one drone, C2 indicates that a maximum of three drones can coordinate the positioning of the user, C3 and C4 indicate the upper limit of the number of communication users and the upper limit of the number of positioning users of the drone, respectively, C5 and C6 indicate the minimum signal-to-noise ratio of the user's communication signal and the minimum signal-to-noise ratio limit of the positioning signal, respectively, C7 indicates the minimum flight altitude, and C8 indicates the minimum safe distance between drones.

[0120] Based on this, the modeling of the UAV deployment optimization model in this application was realized.

[0121] Optionally, when solving one of the two sub-problems, it is necessary to ensure that the other sub-problem is determined. Specifically, regarding the sub-problem of drone-user association, the association strategy between drones and users is determined based on the user's communication link benefits and positioning link benefits. Regarding the sub-problem of optimizing drone deployment locations, given the determined association strategy, the optimal solution to the sub-problem is determined using a dynamic particle swarm optimization algorithm. Therefore, the optimal solution to the sub-problem of optimizing drone deployment locations is used as the optimal deployment location for the drone group based on historical user distribution data.

[0122] In some embodiments, determining the association strategy between the drone and the user based on the user's communication link benefits and location link benefits includes:

[0123] For all users within the communication coverage area, identify the user with the highest communication link benefit and the corresponding drone group;

[0124] For all users within the location coverage area, identify the user with the highest location link benefit and the corresponding drone group;

[0125] Associate the user with the highest benefit from the communication link with the corresponding drone group, and associate the user with the highest benefit from the positioning link with the corresponding drone group.

[0126] Specifically, regarding the sub-problem of the association between drones and users, this application comprehensively considers user service quality and drone service capacity constraints, and proposes a benefit-based greedy association strategy (BGAS).

[0127] This application defines the communication link revenue for user n. and location link benefits when Once determined, for all users within the communication or positioning coverage area, this application calculates both types of benefits separately, associating the user with the highest benefit from the communication link with the corresponding UAV group, and associating the user with the highest benefit from the positioning link with the corresponding UAV group.

[0128] Optionally, determining the user and corresponding drone group with the highest communication link revenue or location link revenue can be done by calculating the revenue of all users, sorting all users in descending order, and taking the user and corresponding drone group with the highest ranking.

[0129] Optionally, during the communication phase, user n's communication benefits... It can be represented as:

[0130]

[0131]

[0132] Where K C (i) represents the remaining number of communication connections for drone i. This represents the set of drones that user n observes and that provide communication services.

[0133] Optionally, during the location phase, user n's location benefits It can be represented as:

[0134]

[0135]

[0136] Where K P (m) represents the remaining number of positioning connections for drone m, W n This refers to a collection of drones that can provide location services.

[0137] The emergency drone deployment method based on convolutional neural networks provided in this application comprehensively considers user service quality and drone service capacity constraints, and proposes a BGAS to determine the association strategy between drones and users.

[0138] In some embodiments, determining the optimal solution to the UAV deployment location optimization subproblem based on the dynamic particle swarm optimization algorithm includes:

[0139] The position and velocity of each particle are initialized based on the K-means algorithm, and the optimal position of the individual and the optimal position of the group are initialized; whereby the position of each particle includes the positions of all drones in the drone group;

[0140] Based on the association strategy between the UAV and the user and the fitness function, the fitness value of each particle is determined;

[0141] The individual optimal position and the population optimal position of each particle are updated based on the fitness value of each particle;

[0142] The inertial weight and learning factor for each particle are determined based on the number of updates.

[0143] The update position and update speed of each particle are determined based on the inertia weight and the learning factor;

[0144] When the maximum number of iterations is reached, the current optimal position of the group is determined as the optimal deployment position of the drone group;

[0145] The fitness function is associated with the communication and navigation joint performance function and the flight safety performance constraints of the UAV. The learning factor includes a group learning factor and an individual learning factor. During the iteration process, the inertia weight is dynamically updated in a non-linear decreasing manner, the group learning factor is dynamically updated in a non-linear decreasing manner, and the individual learning factor is dynamically updated in a non-linear increasing manner.

[0146] Specifically, the subproblem of optimizing UAV deployment locations can be solved using traversal methods, heuristic algorithms, and so on. However, traversal methods consume a significant amount of time and computational resources. Therefore, this application prioritizes heuristic algorithms—Particle Swarm Optimization (PSO).

[0147] However, the traditional PSO algorithm suffers from problems such as local optima and slow convergence speed. Considering the passivity of CNN model deployment decisions, this application further improves the PSO algorithm by proposing the Dynamic Particle Swarm Optimization (DPSO) algorithm to improve the quality of the dataset.

[0148] First, generate K. D Each piece of ground user information (user location information) is defined as the set of user locations for the kth piece. Each piece of user information is randomly generated based on the scenario. After generation, K needs to be obtained. D The corresponding drone deployment locations.

[0149] Each particle position l contains the positions of M drones:

[0150]

[0151] The main process of DPSO in solving the UAV position optimization sub-problem is as follows:

[0152] Initialization phase: The initial population contains k particles, i.e., l = {l1, l2, ..., ln}.k Optionally, to reduce the iteration time of the algorithm, the position l of each particle i is determined using the KMeans algorithm. i and velocity v i Initialize the data, and simultaneously initialize the individual's optimal position p. i And the optimal position g of the group.

[0153] Evaluation Phase: Based on the association between the UAV and the user and the fitness function, the fitness value of each particle is determined. Based on the aforementioned BGAS, the UAV-user association strategy for each particle l can be derived, and then the fitness value of each particle is calculated using the fitness function. Optionally, this application defines a fitness function F(l) to evaluate each particle. F(l) is related to the optimization objective and its constraints, and can be expressed as:

[0154]

[0155] Where f(l) is the optimization objective function—the joint performance function of conduction and bandwidth, f min It is the minimum fitness value in the current group. C7 & C8 indicate that the position of the drone corresponding to the current particle satisfies the C7 and C8 constraints, which are the flight safety performance constraints of the drone.

[0156] Update Phase: The individual optimal position and the population optimal position of each particle are updated based on its fitness value. For each particle i, if its current position fitness value F(l)... i ) is better than its individual optimal fitness F(p) i If the individual's optimal position p is updated, then the optimal position p is updated. i If the current position fitness value F(l) i ) is better than its population optimal fitness F(g) i If the optimal position of the group is updated, then the optimal position g is updated. i .

[0157] Then, the inertia weight and learning factor of each particle are determined based on the number of updates; the update position and update rate of each particle are determined based on the inertia weight and learning factor.

[0158] Optionally, for each particle i, the velocity v is updated according to the following formula. i and position l i :

[0159]

[0160]

[0161] Where t represents the number of iterations, w t It is an inertia weight, used to balance the ability of local search and global search. and These are the individual learning factor and the group learning factor, representing the learning ability of individuals and groups, respectively. r1 and r2 are randomly generated numbers between [0,1] to introduce randomness.

[0162] Generally speaking, w t A larger value indicates stronger global optimization ability but weaker local optimization ability. Unlike traditional PSO, DPSO makes w t Non-linear decrease, w in the early stage of the search t The size is relatively large, giving particles ample space for global search, in the later stages w t Smaller size is advantageous for finding local optimizations.

[0163] Optionally, w t It can be represented as:

[0164]

[0165] Among them, w ini ,w end Let t represent the initial and final values ​​of the inertia weight during the iteration process, respectively. max It represents the maximum number of iterations.

[0166] for when The motion of particles tends to favor the individual optimal direction; conversely, it tends to favor the group optimal direction. Therefore, DPSO makes... Non-linear decrease, The algorithm exhibits non-linear incrementality, emphasizing global search capabilities in the early stages and local search capabilities in later stages. Optionally, They share a single formula, expressed as:

[0167]

[0168] Among them, c ini ,c end These represent the initial and final values ​​of the learning factor during the iteration process, respectively.

[0169] Finally, when the algorithm reaches the maximum number of iterations T iter When the optimal position g of the current population is reached, output it as the solution to the problem.

[0170] Alternatively, the BGAS+DPSO method can be used for each piece of user information. Able to calculate near-optimal drone deployment sets at a relatively low cost Repeat BGAS+DPSO method K D Next, we can get K. D The optimal policy dataset D for n samples is denoted as:

[0171]

[0172] The emergency drone deployment method based on convolutional neural networks provided in this application introduces a nonlinear dynamic weight update strategy and a nonlinear dynamic learning factor update strategy on the basis of the traditional PSO algorithm. On the one hand, it accelerates the convergence speed of the algorithm, and on the other hand, it can effectively prevent the algorithm from getting stuck in local optima. The quality optimization of the training dataset of CNN is improved through DPSO.

[0173] The following example illustrates the deployment method of the integrated communication and navigation emergency drone based on convolutional neural networks provided in this application.

[0174] Figure 3 This is the second flowchart illustrating the integrated communication and navigation emergency drone deployment method based on convolutional neural networks provided in this application embodiment. Figure 3 As shown, the method includes at least two stages:

[0175] Offline Phase: Offline dataset construction and CNN training. Historical user distribution data is solved using DPSO. The optimal deployment location for the unmanned aerial vehicle (UAV) group is determined, including steps such as initializing UAV positions using the K-means algorithm, using BGAS to determine the association between UAVs and users, updating UAV positions, and finally outputting the UAV deployment locations. This allows us to construct a training dataset for the CNN, train the CNN, and learn the mapping relationship between user distribution and the optimal deployment location of the drone group.

[0176] Online Phase: CNN Online Decision Making. Using a pre-trained CNN, the deployment results of drone groups in communication and navigation drone deployment scenarios are quickly obtained.

[0177] The following describes the communication and navigation integrated emergency drone deployment device based on convolutional neural networks provided in this application. The communication and navigation integrated emergency drone deployment device described below and the communication and navigation integrated emergency drone deployment method described above can be referred to in correspondence.

[0178] Figure 4 This is a schematic diagram of the structure of the integrated communication and navigation emergency drone deployment device based on a convolutional neural network provided in the embodiments of this application, as shown below. Figure 4 As shown, the device includes at least:

[0179] The first acquisition module 401 is used to acquire user distribution data in the current emergency rescue scenario, and to perform feature enhancement on the user distribution data based on Gaussian enhancement;

[0180] The second acquisition module 402 is used to input the user distribution data after feature enhancement into the trained convolutional neural network to obtain the optimal deployment location of the drone group.

[0181] The trained convolutional neural network is obtained through offline training on a pre-determined target dataset; the target dataset consists of the optimal deployment locations of drone groups under historical user distribution data; the optimal deployment locations of drone groups under historical user distribution data are obtained by solving a pre-determined drone deployment optimization model using a dynamic particle swarm optimization algorithm.

[0182] The UAV deployment optimization model aims to maximize the user's communication and navigation joint performance function, while taking into account the UAV's signal quality, flight safety performance, and payload capacity during communication and positioning services. The communication and navigation joint performance function is used to characterize the user's communication and positioning performance.

[0183] In some embodiments, the optimal deployment location of the drone group based on the historical user distribution data is determined through the following steps:

[0184] The joint communication and navigation performance function that maximizes the user is decomposed into a sub-problem of UAV-user association and a sub-problem of UAV deployment location optimization.

[0185] Based on the user's communication link benefits and positioning link benefits, determine the association strategy between the drone and the user;

[0186] Given a defined association strategy between drones and users, the optimal solution to the drone deployment location optimization subproblem is determined using a dynamic particle swarm optimization algorithm. This optimal solution is then used as the optimal deployment location for the drone group based on the historical user distribution data.

[0187] In some embodiments, determining the association strategy between the drone and the user based on the user's communication link benefits and location link benefits includes:

[0188] For all users within the communication coverage area, identify the user with the highest communication link benefit and the corresponding drone group;

[0189] For all users within the location coverage area, identify the user with the highest location link benefit and the corresponding drone group;

[0190] Associate the user with the highest benefit from the communication link with the corresponding drone group, and associate the user with the highest benefit from the positioning link with the corresponding drone group.

[0191] In some embodiments, determining the optimal solution to the UAV deployment location optimization subproblem based on the dynamic particle swarm optimization algorithm includes:

[0192] The position and velocity of each particle are initialized based on the K-means algorithm, and the optimal position of the individual and the optimal position of the group are initialized; whereby the position of each particle includes the positions of all drones in the drone group;

[0193] Based on the association strategy between the UAV and the user and the fitness function, the fitness value of each particle is determined;

[0194] The individual optimal position and the population optimal position of each particle are updated based on the fitness value of each particle;

[0195] The inertial weight and learning factor for each particle are determined based on the number of updates.

[0196] The update position and update speed of each particle are determined based on the inertia weight and the learning factor;

[0197] When the maximum number of iterations is reached, the current optimal position of the group is determined as the optimal deployment position of the drone group;

[0198] The fitness function is associated with the communication and navigation joint performance function and the flight safety performance constraints of the UAV. The learning factor includes a group learning factor and an individual learning factor. During the iteration process, the inertia weight is dynamically updated in a non-linear decreasing manner, the group learning factor is dynamically updated in a non-linear decreasing manner, and the individual learning factor is dynamically updated in a non-linear increasing manner.

[0199] In some embodiments, the apparatus further includes:

[0200] The second determining module is used to determine the user's communication performance and positioning performance based on a pre-determined forest area channel model;

[0201] The third determining module is used to perform weighted processing on the communication performance and the positioning performance to determine the communication and navigation joint performance function.

[0202] In some embodiments, the positioning performance is determined based on the user's location accuracy factor and a pre-set location accuracy factor threshold.

[0203] Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application, such as... Figure 5As shown, the electronic device may include: a processor 501, a communication interface 502, a memory 503, and a communication bus 504, wherein the processor 501, the communication interface 502, and the memory 503 communicate with each other via the communication bus 504. The processor 501 can call logical instructions in the memory 503 to execute any of the convolutional neural network-based emergency drone deployment methods provided in the above-described method embodiments, the method including:

[0204] Obtain user distribution data in the current emergency rescue scenario, and perform feature enhancement on the user distribution data based on Gaussian enhancement;

[0205] The user distribution data after feature enhancement is input into a trained convolutional neural network to obtain the optimal deployment location of the drone group.

[0206] The trained convolutional neural network is obtained through offline training on a pre-determined target dataset; the target dataset consists of the optimal deployment locations of drone groups under historical user distribution data; the optimal deployment locations of drone groups under historical user distribution data are obtained by solving a pre-determined drone deployment optimization model using a dynamic particle swarm optimization algorithm.

[0207] The UAV deployment optimization model aims to maximize the user's communication and navigation joint performance function, while taking into account the UAV's signal quality, flight safety performance, and payload capacity during communication and positioning services. The communication and navigation joint performance function is used to characterize the user's communication and positioning performance.

[0208] Furthermore, the logical instructions in the aforementioned memory 503 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, 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 application. 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.

[0209] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the emergency drone deployment method based on a convolutional neural network provided in the above-described method embodiments. The method includes:

[0210] Obtain user distribution data in the current emergency rescue scenario, and perform feature enhancement on the user distribution data based on Gaussian enhancement;

[0211] The user distribution data after feature enhancement is input into a trained convolutional neural network to obtain the optimal deployment location of the drone group.

[0212] The trained convolutional neural network is obtained through offline training on a pre-determined target dataset; the target dataset consists of the optimal deployment locations of drone groups under historical user distribution data; the optimal deployment locations of drone groups under historical user distribution data are obtained by solving a pre-determined drone deployment optimization model using a dynamic particle swarm optimization algorithm.

[0213] The UAV deployment optimization model aims to maximize the user's communication and navigation joint performance function, while taking into account the UAV's signal quality, flight safety performance, and payload capacity during communication and positioning services. The communication and navigation joint performance function is used to characterize the user's communication and positioning performance.

[0214] Furthermore, this application also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the convolutional neural network-based emergency drone deployment method provided in the above-described method embodiments. The method includes:

[0215] Obtain user distribution data in the current emergency rescue scenario, and perform feature enhancement on the user distribution data based on Gaussian enhancement;

[0216] The user distribution data after feature enhancement is input into a trained convolutional neural network to obtain the optimal deployment location of the drone group.

[0217] The trained convolutional neural network is obtained through offline training on a pre-determined target dataset; the target dataset consists of the optimal deployment locations of drone groups under historical user distribution data; the optimal deployment locations of drone groups under historical user distribution data are obtained by solving a pre-determined drone deployment optimization model using a dynamic particle swarm optimization algorithm.

[0218] The UAV deployment optimization model aims to maximize the user's communication and navigation joint performance function, while taking into account the UAV's signal quality, flight safety performance, and payload capacity during communication and positioning services. The communication and navigation joint performance function is used to characterize the user's communication and positioning performance.

[0219] The device embodiments described above are merely illustrative. 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0220] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0221] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for deploying emergency unmanned aerial vehicles (UAVs) based on convolutional neural networks, characterized in that, include: Obtain user distribution data in the current emergency rescue scenario, and perform feature enhancement on the user distribution data based on Gaussian enhancement; The user distribution data after feature enhancement is input into a trained convolutional neural network to obtain the optimal deployment location of the drone group. The trained convolutional neural network is obtained through offline training on a pre-determined target dataset; the target dataset consists of the optimal deployment locations of drone groups under historical user distribution data; the optimal deployment locations of drone groups under historical user distribution data are obtained by solving a pre-determined drone deployment optimization model using a dynamic particle swarm optimization algorithm. The UAV deployment optimization model aims to maximize the user's joint communication and navigation performance function, while taking the UAV's signal quality, flight safety performance, and payload capacity during communication and positioning services as constraints; the joint communication and navigation performance function is used to characterize the user's communication performance and positioning performance. The optimal deployment location of the drone group based on the historical user distribution data is determined through the following steps: The joint communication and navigation performance function that maximizes the user is decomposed into a sub-problem of UAV-user association and a sub-problem of UAV deployment location optimization. Based on the user's communication link benefits and positioning link benefits, determine the association strategy between the drone and the user; Given the association strategy between drones and users, the optimal solution to the drone deployment location optimization subproblem is determined based on the dynamic particle swarm optimization algorithm, and the optimal solution to the drone deployment location optimization subproblem is used as the optimal deployment location of the drone group under the historical user distribution data. The step of determining the optimal solution to the UAV deployment location optimization subproblem based on the dynamic particle swarm optimization algorithm includes: The position and velocity of each particle are initialized based on the K-means algorithm, and the optimal position of the individual and the optimal position of the group are initialized; whereby the position of each particle includes the positions of all drones in the drone group; Based on the association strategy between the UAV and the user and the fitness function, the fitness value of each particle is determined; The individual optimal position and the population optimal position of each particle are updated based on the fitness value of each particle; The inertial weight and learning factor for each particle are determined based on the number of updates. The update position and update speed of each particle are determined based on the inertia weight and the learning factor; When the maximum number of iterations is reached, the current optimal position of the group is determined as the optimal deployment position of the drone group; The fitness function is associated with the communication and navigation joint performance function and the flight safety performance constraints of the UAV. The learning factor includes a group learning factor and an individual learning factor. During the iteration process, the inertia weight is dynamically updated in a non-linear decreasing manner, the group learning factor is dynamically updated in a non-linear decreasing manner, and the individual learning factor is dynamically updated in a non-linear increasing manner.

2. The deployment method for integrated communication and navigation emergency drones based on convolutional neural networks according to claim 1, characterized in that, The method for determining the association strategy between the drone and the user based on the user's communication link revenue and positioning link revenue includes: For all users within the communication coverage area, identify the user with the highest communication link benefit and the corresponding drone group; For all users within the location coverage area, identify the user with the highest location link benefit and the corresponding drone group; Associate the user with the highest benefit from the communication link with the corresponding drone group, and associate the user with the highest benefit from the positioning link with the corresponding drone group.

3. The deployment method for integrated communication and navigation emergency drones based on convolutional neural networks according to claim 1, characterized in that, The method further includes: The user's communication performance and positioning performance are determined based on a pre-determined forest area channel model; The communication performance and the positioning performance are weighted to determine the joint communication and navigation performance function.

4. The deployment method for integrated communication and navigation emergency drones based on convolutional neural networks according to claim 3, characterized in that, The positioning performance is determined based on the user's location accuracy factor and a pre-set location accuracy factor threshold.

5. A communication and navigation integrated emergency drone deployment device based on convolutional neural networks, characterized in that, include: The first acquisition module is used to acquire user distribution data in the current emergency rescue scenario, and to perform feature enhancement on the user distribution data based on Gaussian enhancement; The second acquisition module is used to input the user distribution data after feature enhancement into the trained convolutional neural network to obtain the optimal deployment location of the drone group. The trained convolutional neural network is obtained through offline training on a pre-determined target dataset; the target dataset consists of the optimal deployment locations of drone groups under historical user distribution data; the optimal deployment locations of drone groups under historical user distribution data are obtained by solving a pre-determined drone deployment optimization model using a dynamic particle swarm optimization algorithm. The UAV deployment optimization model aims to maximize the user's joint communication and navigation performance function, while taking the UAV's signal quality, flight safety performance, and payload capacity during communication and positioning services as constraints; the joint communication and navigation performance function is used to characterize the user's communication performance and positioning performance. The optimal deployment location of the drone group based on the historical user distribution data is determined through the following steps: The joint communication and navigation performance function that maximizes the user is decomposed into a sub-problem of UAV-user association and a sub-problem of UAV deployment location optimization. Based on the user's communication link benefits and positioning link benefits, determine the association strategy between the drone and the user; Given the association strategy between drones and users, the optimal solution to the drone deployment location optimization subproblem is determined based on the dynamic particle swarm optimization algorithm, and the optimal solution to the drone deployment location optimization subproblem is used as the optimal deployment location of the drone group under the historical user distribution data. The step of determining the optimal solution to the UAV deployment location optimization subproblem based on the dynamic particle swarm optimization algorithm includes: The position and velocity of each particle are initialized based on the K-means algorithm, and the optimal position of the individual and the optimal position of the group are initialized; whereby the position of each particle includes the positions of all drones in the drone group; Based on the association strategy between the UAV and the user and the fitness function, the fitness value of each particle is determined; The individual optimal position and the population optimal position of each particle are updated based on the fitness value of each particle; The inertial weight and learning factor for each particle are determined based on the number of updates. The update position and update speed of each particle are determined based on the inertia weight and the learning factor; When the maximum number of iterations is reached, the current optimal position of the group is determined as the optimal deployment position of the drone group; The fitness function is associated with the communication and navigation joint performance function and the flight safety performance constraints of the UAV. The learning factor includes a group learning factor and an individual learning factor. During the iteration process, the inertia weight is dynamically updated in a non-linear decreasing manner, the group learning factor is dynamically updated in a non-linear decreasing manner, and the individual learning factor is dynamically updated in a non-linear increasing manner.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the emergency drone deployment method based on convolutional neural networks as described in any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the emergency drone deployment method based on convolutional neural networks as described in any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the emergency drone deployment method based on convolutional neural networks as described in any one of claims 1 to 4.

Images