A method, system and device for unmanned aerial vehicle communication based on federated learning
By using a federated learning model to select and allocate communication frequency bands for UAVs, the problem of data packet loss in unreliable communication environments is solved, communication quality and model training accuracy are improved, and efficient communication is achieved in extreme situations.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NAT UNIV OF DEFENSE TECH
- Filing Date
- 2025-08-01
- Publication Date
- 2026-05-19
AI Technical Summary
With limited onboard computing resources, storage capacity, and communication capabilities, drones struggle to effectively transmit model parameters in unreliable communication environments, resulting in high data loss rates and low communication quality. Traditional methods fail to fully consider the dynamic and unreliable nature of wireless channels, impacting the performance and adaptability of federated learning.
A federated learning-based approach is adopted to allocate UAV communication resources through a pre-set federated learning model. UAVs with a high probability of successful transmission are selected as target UAVs, and their corresponding communication frequency bands are allocated for data transmission. The signal-to-noise ratio and channel state information are combined for dynamic updates to optimize communication quality.
It effectively reduced the data packet loss rate, improved the quality of UAV communication and the accuracy of model training, and enhanced system efficiency and robustness in unreliable communication environments.
Smart Images
Figure CN121887257B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of artificial intelligence and communication technology, specifically to a drone communication method, system, and device based on federated learning. Background Technology
[0002] Drones, with their superior maneuverability, three-dimensional maneuverability, and rapid deployment capabilities, have become a key platform for edge intelligence. By integrating advanced machine learning technologies such as neural networks, drones can not only effectively collect data but also perform real-time intelligent analysis onboard, enabling critical applications in agricultural monitoring, logistics delivery, and emergency response. For example, in precision agriculture, AI-equipped drones can detect crop pests and diseases in real time, allowing for the application of targeted treatments. In disaster relief operations, drones can quickly survey affected areas and use onboard AI to locate survivors, significantly improving response efficiency. This "aerial edge computing" approach extends intelligent services from the ground to low-altitude airspace, promoting the deep integration of the Internet of Things (IoT) and artificial intelligence.
[0003] While drone technology has broad application prospects, it faces significant challenges due to limited onboard computing resources, storage capacity, and communication capabilities. Applications such as autonomous navigation and target tracking generate highly distributed data, making traditional centralized data processing inefficient and even infeasible in extreme cases. Due to limitations in drone energy and communication bandwidth, it is difficult to ensure reliable transmission of model parameters from all participating drones to the edge server, resulting in high packet loss rates and ultimately poor communication quality. Summary of the Invention
[0004] This application aims to provide a method, system, and device for drone communication based on federated learning, which can reduce data packet loss rate and improve communication quality.
[0005] The technical solution of this invention is implemented as follows:
[0006] In a first aspect, embodiments of this application provide a federated learning-based drone communication method, the method comprising:
[0007] Determine multiple drones used for communication and the corresponding mission areas for each of the multiple drones;
[0008] A pre-defined federated learning model is used to allocate communication resources among the multiple drones and determine the target drone and its corresponding communication frequency band; wherein, the pre-defined federated learning model is used for joint client selection and resource allocation under unreliable communication conditions;
[0009] Communication between the target drone and the server is conducted based on the communication frequency band corresponding to each target drone.
[0010] In the above scheme, the step of allocating communication resources to the multiple drones and determining the target drone and its corresponding communication frequency band by using a preset federated learning model includes:
[0011] The target drone is determined by selecting clients from the multiple drones using the preset federated learning model; wherein the target drone includes at least two drones.
[0012] The target UAV is allocated bandwidth using the preset federated learning model to determine the corresponding communication frequency band.
[0013] In the above scheme, the communication between the target drone and the server based on the communication frequency band corresponding to each target drone includes:
[0014] Data is collected from the target drone's corresponding mission area to obtain data information related to the target drone.
[0015] Model training is performed based on the data information corresponding to the target UAV to determine the model parameters corresponding to the target UAV.
[0016] For each of the target drones, the model parameters corresponding to each drone are sent to the server through the communication frequency band corresponding to each drone.
[0017] In the above scheme, before determining the target drone and its corresponding communication frequency band by selecting clients and allocating bandwidth to the multiple drones through a preset federated learning model, the method further includes:
[0018] Acquire multiple sample drones and their respective local datasets;
[0019] Client selection is performed on the multiple sample drones to determine the target sample drone;
[0020] The initial federated learning model is trained using the target sample drone and the local dataset to determine the preset federated learning model.
[0021] In the above scheme, the step of selecting a target sample drone from the multiple sample drones by client selection includes:
[0022] Determine the success probability of each of the multiple sample drones;
[0023] Based on the successful transmission probability, the successful transmission probabilities are sorted in descending order to obtain the sorting result;
[0024] Based on the ranking results, the sample drone with a ranking greater than a preset threshold is selected from the plurality of sample drones and determined as the target sample drone.
[0025] In the above scheme, the step of training the initial federated learning model using the target sample drone and the local dataset to determine the preset federated learning model includes:
[0026] Using the local dataset, the target sample drone is trained locally to determine the local model corresponding to the target sample drone;
[0027] Based on the local model corresponding to the target sample UAV, the initial federated learning model is trained to determine the preset federated learning model.
[0028] In the above scheme, the step of training the initial federated learning model based on the local model corresponding to the target sample UAV to determine the preset federated learning model includes:
[0029] Upload the model parameters of the local model corresponding to the target sample UAV to the server;
[0030] The server processes the model parameters of the local model corresponding to the target sample UAV to determine the global model parameters.
[0031] Based on the global model parameters, the initial federated learning model is trained to determine the preset federated learning model.
[0032] In the above scheme, the step of training the initial federated learning model based on the global model parameters to determine the preset federated learning model includes:
[0033] Based on the global model parameters, determine the parameter information of the initial federated learning model;
[0034] The initial federated learning model is used to allocate bandwidth to the target sample UAV and determine the optimal bandwidth for the target sample UAV.
[0035] The signal-to-noise ratio is calculated using the optimal bandwidth, and channel state information is collected.
[0036] The parameter information of the initial federated learning model is updated using the signal-to-noise ratio and the channel state information to determine the preset federated learning model.
[0037] Secondly, embodiments of this application provide a federated learning-based unmanned aerial vehicle (UAV) communication system, comprising: a determination unit, an allocation unit, and a communication unit, wherein...
[0038] The determining unit is used to determine multiple drones used for communication and the task areas corresponding to each of the multiple drones.
[0039] The allocation unit is used to allocate communication resources to the multiple UAVs through a preset federated learning model, and to determine the target UAV and its corresponding communication frequency band; wherein, the preset federated learning model is used for joint client selection and resource allocation under unreliable communication.
[0040] The communication unit is used to conduct communication between the target drone and the server based on the communication frequency band corresponding to each target drone.
[0041] Thirdly, embodiments of this application provide a drone communication device based on federated learning, the drone communication device based on federated learning comprising: a processor and a memory; wherein,
[0042] The memory is used to store computer programs;
[0043] The processor is configured to call and run the computer program from the memory to perform the method as described in the first aspect.
[0044] This application provides a method, system, and device for drone communication based on federated learning. The method includes: determining multiple drones for communication and their respective task areas; allocating communication resources to the multiple drones using a preset federated learning model to determine a target drone and its corresponding communication frequency band; wherein the preset federated learning model is used for joint client selection and resource allocation under unreliable communication conditions; and conducting communication between the target drone and a server based on the communication frequency band corresponding to each target drone. In unreliable communication scenarios, this approach, by allocating communication resources to multiple drones using a preset federated learning model, determining the target drone and its corresponding communication frequency band, and then conducting communication between the target drone and the server using the communication frequency band, can reduce data packet loss and thus improve communication quality. Attached Figure Description
[0045] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the technical solutions of this application. Obviously, the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.
[0046] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.
[0047] Figure 1 A schematic diagram of a drone communication scenario provided for an embodiment of this application, illustrating a drone communication method based on federated learning.
[0048] Figure 2 An optional flowchart illustrating a federated learning-based unmanned aerial vehicle (UAV) communication method provided for an embodiment of this application;
[0049] Figure 3 A schematic diagram illustrating experimental results of a federated learning-based drone communication method provided in this application embodiment;
[0050] Figure 4 A schematic diagram of the structure of a drone communication system based on federated learning provided for an embodiment of this application;
[0051] Figure 5 This is a schematic diagram of the structure of a drone communication device based on federated learning, provided as an embodiment of this application. Detailed Implementation
[0052] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the specific technical solutions of this application will be further described in detail below with reference to the accompanying drawings of the embodiments of this application. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application.
[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.
[0054] In the following description, references to "some embodiments," "this embodiment," "this application embodiment," and examples, etc., describe a subset of all possible embodiments. However, it is understood that "some embodiments" may be the same subset or different subset of all possible embodiments and may be combined with each other without conflict.
[0055] If the application documents contain similar descriptions such as "first / second", the following explanation shall be added: In the following description, the terms "first / second / third" are used only to distinguish similar objects and do not represent a specific order of objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.
[0056] Implementing federated learning in real-world wireless networks presents several challenges: First, limited wireless resources restrict the number of clients that can participate in each training round. Second, due to the limitations of UAV energy and communication bandwidth, it is difficult to ensure reliable transmission of model parameters from all participating UAVs to the edge server, ultimately leading to a decrease in overall model accuracy. This renders traditional data processing methods unsuitable for various extreme situations. Therefore, researching a federated learning method that can adapt to unreliable communication environments is of paramount practical significance for improving the intelligence level and application efficiency of UAV systems.
[0057] Existing technologies mainly focus on client selection and resource allocation strategies to optimize federated learning performance. For example, some studies have proposed optimizing federated learning performance through client selection and resource allocation, but these methods fail to fully consider the dynamics and unreliability of wireless channels; some studies use retransmission mechanisms to improve transmission success rates, but frequent retransmissions incur significant communication overhead; other studies attempt to mitigate the impact of packet loss by filtering important samples and clients, but these studies typically rely on fixed packet loss probability models and fail to fully explore the dynamic relationship between wireless resource allocation and packet loss probability.
[0058] Existing methods fail to adequately consider the dynamic and unreliable nature of wireless channels, which may lead to performance instability in practical applications. Methods relying on fixed packet loss probability models fail to effectively explore the dynamic relationship between wireless resource allocation and packet loss probability, limiting their adaptability and performance optimization capabilities in complex and ever-changing wireless environments. While retransmission mechanisms can improve transmission success rates, frequent retransmissions result in significant communication overhead, thus impacting the overall system efficiency.
[0059] The specific research scenario of this application is shown in Figure 1, where there are N drones in the air (i.e., Figure 1 Client drone 1, client drone 2, ..., and client drone N are located in different mission areas (i.e., Figure 1Data is collected in task regions 1, 2, ..., and N. An edge server is located on the ground. The drone swarm and the server collaborate on training through federated learning. However, due to the limited onboard resources of the drones (battery capacity, communication bandwidth, and transmission power, etc.), the number of drones that can participate in training is limited. In addition, the communication environment in wireless networks is usually unreliable, and data packets may be lost when the drones communicate with the server. These factors will affect the performance of the federated learning model.
[0060] Based on this, embodiments of this application provide a drone communication method based on federated learning. Figure 2 This is an optional flowchart illustrating a federated learning-based drone communication method provided in an embodiment of this application, which will combine... Figure 2 The steps shown are explained.
[0061] S101. Determine the multiple drones used for communication and the corresponding mission areas for each drone.
[0062] In some embodiments of this application, an unmanned aerial vehicle (UAV) refers to an unmanned aircraft controlled by a radio remote control device or a preset program, which has the ability to fly autonomously.
[0063] In some embodiments of this application, the drone is mainly a civilian drone.
[0064] Civilian drones: covering consumer-grade (such as aerial photography and agricultural monitoring) and industrial-grade (logistics and surveying) applications, must comply with the real-name registration and flight permit system.
[0065] It should be noted that this application uses civilian drones as an example for illustration.
[0066] In some embodiments of this application, the task area corresponding to the UAV refers to the area where the UAV collects data. For multiple UAVs, each UAV has its own corresponding task area, and the task areas between any two UAVs do not overlap.
[0067] In some embodiments of this application, the federated learning-based UAV communication method is applicable to scenarios involving federated client selection and resource allocation under unreliable communication conditions.
[0068] In some embodiments of this application, the drone communication method based on federated learning is implemented by a drone and a server.
[0069] In some embodiments of this application, the area to be monitored and the swarm of drones responsible for monitoring are obtained. The drone swarm includes multiple drones, which can communicate with a server. Based on the number of drones, the area to be monitored is divided into multiple task areas, and a mapping relationship is established between the task areas and the drones, i.e., one drone corresponds to one task area.
[0070] It should be noted that each drone has its corresponding mission area, and a reasonable flight route is formulated for the drone to monitor the mission area.
[0071] S102. Through a preset federated learning model, communication resources are allocated to multiple UAVs to determine the target UAV and its corresponding communication frequency band; wherein, the preset federated learning model is used for joint client selection and resource allocation under unreliable communication.
[0072] In some embodiments of this application, a pre-defined federated learning model is used for federated client selection and resource allocation in unreliable communication. The pre-defined federated learning model is trained using sample data.
[0073] In some embodiments of this application, Federated Learning (FL) is a distributed machine learning framework that allows multiple participants to collaboratively train a global model without sharing the original data. Its core objective is to solve the data silo problem and ensure privacy and security.
[0074] In some embodiments of this application, the basic principles of the federated learning model include: local training, parameter aggregation, and global iteration.
[0075] Local training: Each participant uses local data to update model parameters or gradients; Parameter aggregation: The central server integrates the updated parameters from each node using a secure aggregation algorithm (such as FedAvg); Global iteration: The updated model is redistributed locally for the next round of training until convergence.
[0076] In some embodiments of this application, the federated learning model is divided into: horizontal federation, vertical federation, and federated transfer learning. Horizontal federation: suitable for scenarios with significant feature overlap and large sample differences (such as cross-bank user behavior prediction); Vertical federation: for scenarios with significant sample overlap and large feature differences (such as cross-institutional joint risk control modeling); Federated transfer learning: solves the problem of significant differences in data distribution.
[0077] In some embodiments of this application, a target drone is determined by selecting clients from multiple drones using a preset federated learning model; wherein, the target drone includes at least two drones; and the target drone is allocated bandwidth using a preset federated learning model to determine the communication frequency band corresponding to the target drone.
[0078] In some embodiments of this application, a preset federated learning model is used to select clients from multiple drones, and the drone with the highest communication success rate is selected as the target drone; the preset federated learning model is used to allocate bandwidth to the target drones and determine the communication frequency band of each drone in the target drones.
[0079] For example, a federated learning model can be represented as follows:
[0080] For each sample In local dataset In this context, the features and labels are represented as follows: and The FL global model consists of model parameters. Representation. Using the loss function. To evaluate model parameters For the sample The prediction error. Then, regarding UAV... The local loss function can be defined as:
[0081]
[0082] Therefore, the global loss function for all distributed datasets is expressed as:
[0083]
[0084] in, Indicates the client The weight, By expecting model parameters To minimize the global loss function N represents the set of all clients.
[0085] It should be noted that, These are model parameters; These are the desired model parameters.
[0086] S103. Based on the communication frequency bands corresponding to the target drones, conduct communication between the target drones and the server.
[0087] In some embodiments of this application, data is collected from the task area corresponding to the target drone to obtain data information corresponding to the target drone; model training is performed based on the data information corresponding to the target drone to determine the model parameters corresponding to the target drone; for each drone in the target drone, the model parameters corresponding to each drone are sent to the server through the communication frequency band corresponding to each drone.
[0088] In some embodiments of this application, the target drone collects data from the mission area according to a planned route, obtaining data information corresponding to the target drone. Model training is performed based on this data information to determine the model parameters corresponding to the target drone. Any one of the target drones can send the model parameters to the server via its corresponding communication frequency band.
[0089] It should be noted that the server is a edge server.
[0090] Understandably, the process involves identifying multiple drones used for communication and their respective task areas; allocating communication resources among these drones using a pre-defined federated learning model to determine the target drone and its corresponding communication frequency band; and then conducting communication between the target drone and the server based on its respective communication frequency band. In unreliable communication scenarios, the pre-defined federated learning model allocates communication resources among the drones, determines the target drone and its corresponding communication frequency band, and facilitates communication between the target drone and the server through these frequency bands, thereby reducing data packet loss and improving communication quality.
[0091] In some embodiments of this application, before S102, the drone communication method based on federated learning further includes S104-S106, as follows:
[0092] S104. Obtain multiple sample drones and their respective local datasets.
[0093] In some embodiments of this application, multiple sample drones and their respective local datasets are obtained; the local datasets are data information collected by the multiple sample drones in their respective task areas.
[0094] S105. Select the target sample drone from multiple sample drones.
[0095] In some embodiments of this application, the drone with the highest successful transmission probability among multiple drones is selected as the target sample drone.
[0096] In some embodiments of this application, the success transmission probability of each of the multiple sample drones is determined; based on the success transmission probability, the success transmission probabilities are sorted in descending order to obtain a sorting result; based on the sorting result, the sample drone with a sorting value greater than a preset threshold is selected from the multiple sample drones and determined as the target sample drone.
[0097] S106. Train the initial federated learning model using the target sample drone and the local dataset to determine the preset federated learning model.
[0098] In some embodiments of this application, the target sample drone is trained using a local dataset. After determining the local model of the target sample drone, the initial federated learning model is trained to determine the preset federated learning model.
[0099] In some embodiments of this application, the target sample drone is trained locally using a local dataset to determine the local model corresponding to the target sample drone; based on the local model corresponding to the target sample drone, the initial federated learning model is trained to determine the preset federated learning model.
[0100] In some embodiments of this application, the model parameters of the local model corresponding to the target sample UAV are uploaded to the server; the server processes the model parameters of the local model corresponding to the target sample UAV to determine the global model parameters; based on the global model parameters, the initial federated learning model is trained to determine the preset federated learning model.
[0101] In some embodiments of this application, the parameter information of the initial federated learning model is determined based on the global model parameters; the bandwidth of the target sample UAV is allocated through the initial federated learning model to determine the optimal bandwidth corresponding to the target sample UAV; the signal-to-noise ratio is calculated through the optimal bandwidth; and channel state information is collected; the parameter information of the initial federated learning model is updated through the signal-to-noise ratio and channel state information to determine the preset federated learning model.
[0102] For example, determining the preset federated learning model can be achieved through the following steps:
[0103] Step 1: System modeling and dynamic channel analysis.
[0104] The model constructs a drone federated learning network, comprising N = {1, 2, ..., N} drones and an edge server, where the drones act as clients utilizing local datasets. During model training, the server is responsible for aggregating global model parameters. Based on the Rayleigh fading channel model, the uplink channel gain and packet loss probability are derived. Among these, the signal-to-noise ratio (SNR) is... With bandwidth allocation Directly related. By analyzing dynamic channel characteristics, a quantitative relationship between communication quality and resource allocation is established. Channel state information is collected in real time, and the distance and channel gain between the UAV and the edge server are dynamically updated to ensure that the model adapts to UAV movement and channel fluctuations.
[0105] Step 2: Derivation of convergence theory and transformation of optimization problem.
[0106] Based on the convergence analysis of federated learning, the global loss function is derived. The convergence rate closed-form expression is derived, revealing the mechanism by which the packet loss rate affects training performance. Specifically, by assuming the L-smoothness of the loss function and the boundedness of the gradient variance, it is proved that the upper bound of the global loss is:
[0107]
[0108] in, and To be related to learning rate Local iteration count Relevant constants; Let these be the desired model parameters. Based on this theory, the original problem (minimizing the global loss F( This problem is transformed into a mixed-integer nonlinear programming (MINLP) problem, with the objective function being to minimize the average packet loss probability. The constraints include the maximum energy consumption of a single wheel. End-to-end delay and total bandwidth limit This achieves a balance between problem complexity and theoretical guidance.
[0109] Step 3: Implementation and dynamic adaptation of the alternating iterative optimization algorithm.
[0110] We propose the FedCB algorithm, which achieves joint resource management by iteratively optimizing client selection and bandwidth allocation. First, given a fixed set of client selections, we solve for the optimal bandwidth allocation.
[0111] For a fixed client selection, each drone will be selected in round k. The bandwidth allocation optimization problem is decomposed as follows:
[0112]
[0113]
[0114] in, , The signal-to-noise ratio threshold indicating when data is successfully decoded. This represents Gaussian white noise. It is a drone The transmission power, Indicates drone Distance to the server; Let P2 represent the given set of clients. This is a convex optimization problem. That is, the objective function P2(b) is the bandwidth. The monotonically decreasing function ( Therefore, to maximize problem P2(b), it is only necessary to minimize the bandwidth of each drone. However, as the bandwidth of the drone decreases, the communication latency and energy consumption of the drone will increase. ( The minimum bandwidth that can be allowed by the time delay and energy consumption constraints can be derived from constraints (12a) and (12b). and Therefore, under the constraints, the optimal solution can only be the maximum of the two. The following is given:
[0115]
[0116] Based on the above-mentioned client selection strategy, the calculation for each drone can be performed. The optimal bandwidth allocation is determined and substituted into the problem, thus simplifying it into the following customer choice problem:
[0117]
[0118]
[0119] It should be noted that, The corresponding constraints are explained in detail below.
[0120] In some embodiments of this application, the UAV communication method based on federated learning involves latency and energy consumption models, packet loss probability, optimization problem statements, and simulation parameters. Specifically:
[0121] 1. Delay and Energy Consumption Model
[0122] Since edge servers typically possess powerful computing capabilities and a stable power supply, their latency and energy consumption were not considered in the optimization model. In the federated learning network under consideration, although flight-related costs are the primary overhead for the drone, the focus is on optimizing latency and energy consumption during local training and model parameter uploading. This ensures that the drone will not be interrupted due to excessive energy consumption during mission execution.
[0123] Local training: Indicates drone CPU frequency, Indicates in drones The number of CPU cycles required to process one sample. (Drone) The computation time for training the local model is:
[0124]
[0125] The corresponding energy cost can be expressed as:
[0126]
[0127] in, It is the effective capacitance coefficient.
[0128] Wireless Transmission: After local training, each selected drone uploads its local model parameters to the edge server. Consider a specific wireless multiple access scheme, Orthogonal Frequency Division Multiple Access (OFDMA), where each drone is assigned a unique frequency band for communication. Considering Rayleigh fading, the achievable uplink data rate for drone n is:
[0129]
[0130] in, This represents the transmit power of drone n. This represents the channel bandwidth of the drone n. This represents the power spectral density of Gaussian noise; Let S represent the channel gain between the drone n and the edge server. Let S represent the data size of the local model. The upload time for model updates is:
[0131]
[0132] Therefore, the energy consumed by drone n in uploading its parameters to the edge server is:
[0133]
[0134] 2. Data packet loss probability
[0135] In this work, a cyclic redundancy check (CRC)-based error detection mechanism is employed to identify potential data corruption in the local model received by the edge server. Packet loss probability is used as a metric to characterize the unreliability of UAV uplink transmission. Simultaneously, downlink transmission is assumed to remain reliable, ensuring successful reception of the global model transmitted from the edge server to the UAV. This assumption is reasonable because edge servers typically possess significantly higher transmission power capabilities than UAVs, thus guaranteeing robustness in downlink communication.
[0136] set up Let represent the signal-to-noise ratio (SNR) threshold required for successful data decoding. Therefore, the probability that drone n successfully transmits through the channel in the kth round is given by the following formula:
[0137]
[0138] drones The corresponding packet loss probability is:
[0139]
[0140] 3. Optimize the problem statement
[0141] In federated learning, the global loss function after K rounds of training is minimized by optimizing client selection and bandwidth allocation. Indicates whether drone n is selected (when selected) =1, otherwise =0) participates in the kth round of training. Let represent the bandwidth allocated to drone n. = { } express The set, B(k) = { } express The set of . The optimization problem can be formulated as:
[0142]
[0143]
[0144]
[0145]
[0146]
[0147]
[0148]
[0149] Federated learning requires multiple training rounds to complete the entire training process, and the longest time for a single training round of federated learning is [duration missing]. Constraint (12b) states that the total energy consumption of each drone in a single round is... Each round of spending should not exceed the budget. Constraint (12c) stipulates that the number of drones participating in each round of training shall not exceed M. Constraint (12d) limits the total bandwidth allocated to the drones. Constraints (12e) and (12f) describe the range of optimization variables.
[0150] 4. Simulation parameters
[0151] In the simulation, an edge server is located at the center of a 500m × 500m square network area, and N client drones are randomly deployed within the coverage area. Since each drone has a specific target, they need to fly to a fixed mission scenario at different times. Therefore, the flight trajectory and landing position for each time slot are pre-planned. On the server side, to simulate packet loss in an unreliable communication environment, the neural network layer parameters of the models uploaded by different client drones are randomly zeroed. The zeroing probability is determined by the corresponding PDP. The complete simulation parameters are recorded in Table 1, and the specific parameter settings are shown in Table 1.
[0152] Table 1
[0153]
[0154] The experimental study used two different datasets: MNIST and CIFAR-10. MNIST is a handwritten digit dataset where each grayscale image is 28x28 pixels in size. CIFAR-10 is a color image dataset containing 32x32 pixel images of various objects. For the MNIST dataset, a multilayer perceptron (MLP) model with two hidden layers (containing 159,010 parameters) was used, while for the CIFAR-10 dataset, a more complex VGG16 architecture (containing 33,645,666 parameters) was used.
[0155] The performance of the proposed method is evaluated by comparison with three benchmark methods: Only-CS (average bandwidth allocation and optimized client selection only), Only-BA (random client selection and optimized bandwidth), and FedAvg (random client selection and uniform bandwidth allocation). Experiments were conducted on different datasets using different models, and the model performance of the different methods was compared. The experimental results are shown below. Figure 3 As shown, Figure 3 (a) in the figure is experimental data on the MNIST dataset; Figure 3 (b) in the figure is experimental data on the CIFAR-10 dataset; Figure 3 In (a) and (b), the horizontal axis represents the number of rounds, in units of rounds; the vertical axis represents the test accuracy and loss value, respectively. The test accuracy and loss value are scalar data, with values between 0 and 1.
[0156] Understandably, by selecting drones with superior wireless channel conditions for training and allocating bandwidth resources appropriately, the negative impact of unreliable communication links can be effectively mitigated, significantly improving model accuracy. Experimental results show that the FedCB framework achieves convergence in only 8 iterations, demonstrating a faster convergence speed. Furthermore, the FedCB framework exhibits stronger robustness in unreliable communication environments, effectively mitigating the impact of packet loss on model performance.
[0157] In some embodiments of this application, the FedCB (Federated Learning Framework) is a federated learning framework suitable for unmanned aerial vehicle (UAV) networks. It optimizes system performance through a joint client selection and bandwidth allocation strategy. An algorithm based on alternating iterative optimization is used to solve the joint optimization problem, effectively coordinating client selection and bandwidth allocation, thus improving the efficiency and robustness of the federated learning system. Dynamic resource allocation and client selection: Clients are dynamically selected and bandwidth resources are allocated based on wireless channel conditions to adapt to unreliable communication environments, ensuring the efficiency and accuracy of model training.
[0158] Based on the above embodiments of a federated learning-based UAV communication method, this application also provides a federated learning-based UAV communication system, such as... Figure 4 As shown, Figure 4 This application provides a schematic diagram of the structure of a federated learning-based unmanned aerial vehicle (UAV) communication system 4. The federated learning-based UAV communication system 4 includes: a determination unit 401, an allocation unit 402, and a communication unit 403.
[0159] The determining unit 401 is used to determine multiple drones used for communication and the task areas corresponding to each of the multiple drones.
[0160] The allocation unit 402 is used to allocate communication resources to the multiple UAVs through a preset federated learning model, and to determine the target UAV and its corresponding communication frequency band; wherein, the preset federated learning model is used for joint client selection and resource allocation under unreliable communication.
[0161] The communication unit 403 is used to conduct communication between the target drone and the server based on the communication frequency band corresponding to each target drone.
[0162] In some embodiments of this application, the determining unit 401 is further configured to select clients for the plurality of drones through the preset federated learning model, and determine the target drone; wherein the target drone includes at least two drones; and to allocate bandwidth to the target drone through the preset federated learning model, and determine the communication frequency band corresponding to the target drone.
[0163] In some embodiments of this application, the federated learning-based unmanned aerial vehicle (UAV) communication system 4 further includes an acquisition unit 404 and a transmission unit 405; wherein,
[0164] The acquisition unit 404 is used to collect data from the task area corresponding to the target drone through the target drone and obtain the data information corresponding to the target drone.
[0165] The determining unit 401 is further configured to perform model training based on the data information corresponding to the target UAV, and determine the model parameters corresponding to the target UAV; for each UAV in the target UAV, the model parameters corresponding to each UAV are sent to the server through the communication frequency band corresponding to each UAV.
[0166] The sending unit 405 is used to send the model parameters corresponding to each UAV in the target UAV to the server via the communication frequency band.
[0167] In some embodiments of this application, the acquisition unit 404 is further used to acquire multiple sample drones and their respective local datasets before the client selection and bandwidth allocation of the multiple drones through a preset federated learning model to determine the target drone and the communication frequency band corresponding to the target drone.
[0168] The determining unit 401 is further configured to perform client selection on the plurality of sample drones to determine the target sample drone; and to train the initial federated learning model using the target sample drone and the local dataset to determine the preset federated learning model.
[0169] In some embodiments of this application, the determining unit 401 is further configured to determine the success transmission probability of each of the plurality of sample drones;
[0170] The acquisition unit 404 is further configured to sort the successful transmission probabilities in descending order based on the successful transmission probabilities, and obtain a sorting result.
[0171] The determining unit 401 is further configured to select, based on the sorting result, a sample drone whose sorting value is greater than a preset threshold from the plurality of sample drones and determine it as the target sample drone.
[0172] In some embodiments of this application, the determining unit 401 is further configured to perform local training on the target sample drone using the local dataset to determine the local model corresponding to the target sample drone; and to train the initial federated learning model based on the local model corresponding to the target sample drone to determine the preset federated learning model.
[0173] In some embodiments of this application, the sending unit 405 is further configured to upload the model parameters of the local model corresponding to the target sample UAV to the server;
[0174] The determining unit 401 is further configured to process the model parameters of the local model corresponding to the target sample UAV through the server to determine the global model parameters; and to train the initial federated learning model based on the global model parameters to determine the preset federated learning model.
[0175] In some embodiments of this application, the determining unit 401 is further configured to: determine the parameter information of the initial federated learning model based on the global model parameters; allocate bandwidth to the target sample UAV using the initial federated learning model to determine the optimal bandwidth corresponding to the target sample UAV; calculate the signal-to-noise ratio using the optimal bandwidth; and collect channel state information; update the parameter information of the initial federated learning model using the signal-to-noise ratio and the channel state information to determine the preset federated learning model.
[0176] Based on the above embodiments of a federated learning-based UAV communication method, this application also provides a federated learning-based UAV communication device, such as... Figure 5 As shown, Figure 5 This is a schematic diagram of a federated learning-based drone communication device provided in an embodiment of this application. The federated learning-based drone communication device 5 includes a processor 501 and a memory 502. The memory 502 is used to store computer programs; the processor 501 is used to call and run the computer programs from the memory to execute a federated learning-based drone communication method as described in the above embodiment.
[0177] In the embodiments of this application, the processor 501 described above can be at least one of the following: Application-Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that for different devices, the electronic device used to implement the above processor function can also be other types, and the embodiments of this application do not specifically limit it.
[0178] Furthermore, in the embodiments of this application, the functional modules 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. The integrated unit can be implemented in hardware or as a software functional module.
[0179] If the integrated unit is implemented as a software functional module and is not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment, in essence, or the part that contributes to the prior art, or all or 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.) or processor to execute all or part of the steps of the method of this embodiment. 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.
[0180] It should be understood that the phrases "one embodiment," "an embodiment," or "some embodiments" mentioned throughout the specification mean that a specific feature, structure, or characteristic related to an embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment," "in one embodiment," or "in some embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential 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 this application. The sequence numbers of the above-described embodiments are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. The descriptions of the various embodiments above tend to emphasize the differences between the various embodiments; their similarities or commonalities can be referred to mutually, and for the sake of brevity, they will not be repeated here.
[0181] The modules described above as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules. They may be located in one place or distributed across multiple network units. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.
[0182] In addition, each functional module in the various embodiments of this application can be integrated into one processing unit, or each module can be a separate unit, or two or more modules can be integrated into one unit; the integrated modules can be implemented in hardware or in the form of hardware plus software functional units.
[0183] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.
[0184] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.
[0185] The features disclosed in the several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.
[0186] The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method or device embodiments.
[0187] The above description is merely an embodiment of this application, but the protection scope of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.
Claims
1. A drone communication method based on federated learning, characterized in that, The method includes: Determine multiple drones used for communication and the corresponding mission areas for each of the multiple drones; A pre-defined federated learning model is used to allocate communication resources among the multiple drones and determine the target drone and its corresponding communication frequency band; wherein, the pre-defined federated learning model is used for joint client selection and resource allocation under unreliable communication conditions; Based on the communication frequency bands corresponding to each of the target drones, communication between the target drones and the server is carried out. Before determining the target drone and its corresponding communication frequency band by selecting clients and allocating bandwidth to the multiple drones through a preset federated learning model, the method further includes: Acquire multiple sample drones and their respective local datasets; Client selection is performed on the multiple sample drones to determine the target sample drone; The initial federated learning model is trained using the target sample drone and the local dataset to determine the preset federated learning model; The step of selecting a target sample drone from the plurality of sample drones by client selection includes: Determine the success probability of each of the multiple sample drones; Based on the successful transmission probability, the successful transmission probabilities are sorted in descending order to obtain the sorting result; Based on the sorting results, the sample drone with a sorting value greater than a preset threshold is selected from the plurality of sample drones and determined as the target sample drone. The step of training an initial federated learning model using the target sample drone and the local dataset to determine the preset federated learning model includes: Using the local dataset, the target sample drone is trained locally to determine the local model corresponding to the target sample drone; Based on the local model corresponding to the target sample UAV, the initial federated learning model is trained to determine the preset federated learning model; The step of training the initial federated learning model based on the local model corresponding to the target sample UAV to determine the preset federated learning model includes: Upload the model parameters of the local model corresponding to the target sample UAV to the server; The server processes the model parameters of the local model corresponding to the target sample UAV to determine the global model parameters. Based on the global model parameters, determine the parameter information of the initial federated learning model; The initial federated learning model is used to allocate bandwidth to the target sample UAV and determine the optimal bandwidth for the target sample UAV. The signal-to-noise ratio is calculated using the optimal bandwidth, and channel state information is collected. The parameter information of the initial federated learning model is updated using the signal-to-noise ratio and the channel state information to determine the preset federated learning model.
2. The method according to claim 1, characterized in that, The step of allocating communication resources among the multiple drones using a pre-set federated learning model to determine the target drone and its corresponding communication frequency band includes: The target drone is determined by selecting clients from the multiple drones using the preset federated learning model; wherein the target drone includes at least two drones. The target UAV is allocated bandwidth using the preset federated learning model to determine the corresponding communication frequency band.
3. The method according to claim 1, characterized in that, The communication between the target drone and the server based on the communication frequency band corresponding to each target drone includes: Data is collected from the target drone's corresponding mission area to obtain data information related to the target drone. Model training is performed based on the data information corresponding to the target UAV to determine the model parameters corresponding to the target UAV. For each of the target drones, the model parameters corresponding to each drone are sent to the server through the communication frequency band corresponding to each drone.
4. A drone communication system based on federated learning, characterized in that, The method described in any one of claims 1-3 includes: a determining unit, an allocation unit, and a communication unit, wherein, The determining unit is used to determine multiple drones used for communication and the task areas corresponding to each of the multiple drones. The allocation unit is used to allocate communication resources to the multiple UAVs through a preset federated learning model, and to determine the target UAV and its corresponding communication frequency band; wherein, the preset federated learning model is used for joint client selection and resource allocation under unreliable communication. The communication unit is used to conduct communication between the target drone and the server based on the communication frequency band corresponding to each target drone.
5. A drone communication device based on federated learning, characterized in that, include: Processor and memory, of which, The memory is used to store computer programs; The processor is configured to call and run the computer program from the memory to perform the method as described in any one of claims 1 to 3.