CN116582871B - Unmanned aerial vehicle cluster federal learning model optimization method based on topology optimization - Google Patents
Unmanned aerial vehicle cluster federal learning model optimization method based on topology optimization Download PDFInfo
- Publication number
- CN116582871B CN116582871B CN202310828232.5A CN202310828232A CN116582871B CN 116582871 B CN116582871 B CN 116582871B CN 202310828232 A CN202310828232 A CN 202310828232A CN 116582871 B CN116582871 B CN 116582871B
- Authority
- CN
- China
- Prior art keywords
- unmanned aerial
- aerial vehicle
- learning model
- following
- federal learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000005457 optimization Methods 0.000 title claims abstract description 40
- 230000002776 aggregation Effects 0.000 claims abstract description 43
- 238000004220 aggregation Methods 0.000 claims abstract description 43
- 238000004891 communication Methods 0.000 claims abstract description 26
- 230000005540 biological transmission Effects 0.000 claims abstract description 25
- 238000005265 energy consumption Methods 0.000 claims abstract description 24
- 238000004364 calculation method Methods 0.000 claims abstract description 15
- 230000008569 process Effects 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims description 33
- 230000006870 function Effects 0.000 claims description 22
- 238000007667 floating Methods 0.000 claims description 3
- 238000013461 design Methods 0.000 abstract description 7
- 230000002349 favourable effect Effects 0.000 description 5
- 238000005562 fading Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 230000004931 aggregating effect Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000006116 polymerization reaction Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/02—Arrangements for optimising operational condition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W28/00—Network traffic management; Network resource management
- H04W28/02—Traffic management, e.g. flow control or congestion control
- H04W28/08—Load balancing or load distribution
- H04W28/09—Management thereof
- H04W28/0917—Management thereof based on the energy state of entities
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W28/00—Network traffic management; Network resource management
- H04W28/02—Traffic management, e.g. flow control or congestion control
- H04W28/08—Load balancing or load distribution
- H04W28/09—Management thereof
- H04W28/0958—Management thereof based on metrics or performance parameters
- H04W28/0967—Quality of Service [QoS] parameters
- H04W28/0983—Quality of Service [QoS] parameters for optimizing bandwidth or throughput
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W84/00—Network topologies
- H04W84/02—Hierarchically pre-organised networks, e.g. paging networks, cellular networks, WLAN [Wireless Local Area Network] or WLL [Wireless Local Loop]
- H04W84/04—Large scale networks; Deep hierarchical networks
- H04W84/08—Trunked mobile radio systems
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention discloses an unmanned aerial vehicle cluster federal learning model optimization method based on topological optimization, which is oriented to unmanned aerial vehicle clusters and minimizes the weighted sum of energy consumption and time delay of each round of unmanned aerial vehicle cluster federal learning according to bandwidth allocation, calculation time delay, communication time delay and constraint conditions of topological design; further, a topological structure corresponding to each round of unmanned aerial vehicle cluster and related to model parameter transmission is obtained, and based on the topological structure, local federal learning model parameters of all following unmanned aerial vehicles are transmitted to the leading unmanned aerial vehicle; and the leading unmanned aerial vehicle carries out parameter aggregation and updating to obtain global federal learning model parameters in the next iteration, and when the federal learning model precision corresponding to the global federal learning model parameters meets the requirement, the federal learning model corresponding to the unmanned aerial vehicle cluster is obtained. The invention obviously accelerates the federal learning process, greatly improves the robustness of the federal learning of the unmanned aerial vehicle cluster, and ensures that the federal learning can be better suitable for the unmanned aerial vehicle cluster with high dynamic network topology.
Description
Technical Field
The invention belongs to the technical field of unmanned aerial vehicle cluster edge intelligence, and particularly relates to an unmanned aerial vehicle cluster federal learning model optimization method based on topology optimization.
Background
Because unmanned aerial vehicles have unique advantages of flexibility, maneuver, high self-adaptation degree, low cost and the like, the unmanned aerial vehicle is considered to be a promising technology in the next generation network in the aspects of sensing, target tracking, data acquisition, communication service providing and the like. Future wireless networks require unmanned clusters to perform tasks such as remote monitoring and mobile edge computing intelligently and autonomously.
While the ability to enable the intelligence of a drone swarm can be given by the way in which all data needs to be transmitted to the cloud data center and Machine Learning (ML) model training and reasoning is performed in it, this approach is not applicable to a drone swarm, with the following drawbacks: (1) The transmission of huge raw data such as images and videos consumes a lot of bandwidth and energy; (2) The original perceived data may contain a large amount of sensitive information, possibly resulting in serious privacy disclosure and data abuse; (3) Traditional cloud computing is difficult to meet the low-latency requirement of unmanned aerial vehicle clusters for real-time tasks.
Distributed federal learning can provide a viable solution to the above-described problems; however, most of the existing federal learning examples are still centralized, i.e. a central entity is responsible for aggregation and fusion of machine learning models on the whole network, while in the unmanned aerial vehicle cluster, nodes and links are unstable, and meanwhile, dynamic connection and departure of unmanned aerial vehicles cause network topology change, so that the whole federal learning is influenced. Although there are methods to address the relevant problems that suggest unmanned aerial vehicle distributed federal learning that do not require a central server, they are not fully applicable and suffer from the following drawbacks: (1) The lack of a central controller carries out convergence judgment and regulation on the whole learning process; (2) Each unmanned plane can bear additional aggregation and transmission tasks, and the limited energy of the unmanned plane is wasted; (3) Unmanned aerial vehicle clusters in practical application often exist leading unmanned aerial vehicles, and heterogeneous unmanned aerial vehicle clusters can slow the convergence speed of federal learning.
Disclosure of Invention
The invention provides an unmanned aerial vehicle cluster federal learning model optimization method based on topological optimization, which is different from the conventional centralized federal learning method in that each node is required to be directly connected with a central server to transmit and aggregate model parameters.
The invention adopts the following technical scheme:
aiming at an unmanned aerial vehicle cluster formed by a leading unmanned aerial vehicle and a preset number of following unmanned aerial vehicles communicated with the leading unmanned aerial vehicle, the unmanned aerial vehicle cluster federal learning model optimization method based on topology optimization obtains a federal learning model corresponding to the unmanned aerial vehicle cluster based on the model training capability and a local data set of each following unmanned aerial vehicle by the following steps:
step 1: broadcasting global federal learning model parameters to all following unmanned aerial vehicles by the leading unmanned aerial vehicle, judging whether federal learning model accuracy corresponding to the global federal learning model parameters meets a preset requirement, and if so, taking the federal learning model corresponding to the current global federal learning model parameters as a federal learning model corresponding to an unmanned aerial vehicle cluster; if the preset requirement is not met, executing the step 2;
step 2: aiming at each following unmanned aerial vehicle, the following unmanned aerial vehicle performs federal learning model training based on a local data set and combined with global federal learning model parameters broadcasted by the leading unmanned aerial vehicle, and after training is completed, local federal learning model parameters corresponding to each following unmanned aerial vehicle are obtained; executing the step 3;
step 3: the leading unmanned aerial vehicle builds a topological structure corresponding to the unmanned aerial vehicle cluster and related to model parameter transmission based on energy consumption and time delay of the current unmanned aerial vehicle cluster, and each following unmanned aerial vehicle transmits local federal learning model parameters to the leading unmanned aerial vehicle based on the topological structure; executing the step 4;
step 4: and (3) aiming at each received local federal learning model parameter, the leading unmanned aerial vehicle carries out parameter aggregation and updating to obtain the global federal learning model parameter in the next iteration, and returns to the execution step (1).
In the step 1, it is determined whether the precision of the federal learning model corresponding to the federal learning model parameter meets the preset requirement, the global loss function and the preset iteration number are used for determining, and if the global loss function converges or reaches the preset iteration number, the federal learning model parameter is used for determining that the precision of the federal learning model corresponding to the federal learning model parameter meets the preset requirementThe federal learning model precision corresponding to the number reaches the preset requirement; otherwise, the federal learning model precision corresponding to the federal learning model parameters does not reach the preset requirement; global loss functionAs follows;
，
in the formula ,representing following unmanned aerial vehicle +.>Is a local data set sample number; />Representing the total number of following unmanned aerial vehicles; />Representing global federal learning model parameters; />Representing the global federal learning model parameter as +.>Lower->Sample loss function for individual data samples.
In the step 2, when the following unmanned aerial vehicle performs federal learning model training based on a local data set and combined with federal learning model parameters broadcasted by the leading unmanned aerial vehicle, the federal learning model parameters are updated through the following formula to obtain local federal learning model parameters;
，
in the formula ,representing following unmanned aerial vehicle +.>In->Local federal learning model parameters obtained by training a local data set in round iteration; />Indicate->The global federal learning model parameters broadcasted by the unmanned aerial vehicle are followed in the round of iteration; />Representing a preset learning rate; />Representing the global federal learning model parameter as +.>Following unmanned aerial vehicle down +.>Is a local federal learning model loss function.
As a preferred embodiment of the present invention, the step 3 includes the following steps:
step 3.1: the leading unmanned aerial vehicle builds an objective function based on the energy consumption and time delay of the current unmanned aerial vehicle cluster;
step 3.2: the leading unmanned aerial vehicle is based on an objective function, combined with constraint conditions about bandwidth, calculation and communication time and topological connection, and takes the objective function minimization as a target, a tree topology structure which corresponds to the unmanned aerial vehicle cluster and is used for model parameter transmission is constructed, and the tree topology structure takes the leading unmanned aerial vehicle as a root node;
step 3.3: based on a tree topology structure corresponding to the unmanned aerial vehicle cluster, transmitting local federal learning model parameters corresponding to the following unmanned aerial vehicle to the root node direction based on the tree topology structure by taking the following unmanned aerial vehicle corresponding to the leaf node as a starting point until the parameters are transmitted to the leading unmanned aerial vehicle; and the following unmanned aerial vehicle corresponding to each node in the tree topology structure carries out parameter aggregation on the received local federal learning model parameters and the local federal learning model parameters thereof, and the model parameters obtained after the parameter aggregation are updated into the local federal learning model parameters of the node following unmanned aerial vehicle.
As a preferred embodiment of the present invention, in the step 3.1, the objective functionThe formula is as follows:
，
in the formula ,，/>，，/>，
wherein ,representing following unmanned aerial vehicle +.>Is calculated by the energy consumption; />Representing following unmanned aerial vehicle +.>Is calculated by the time delay;representing following unmanned aerial vehicle +.>Is a communication energy consumption of the system; />Representing following unmanned aerial vehicle +.>Is a communication delay of (1); />Representing the total number of following unmanned aerial vehicles; />Representing a preset weighting factor; />Representing coefficients that depend on the chip architecture; />Indicate->Following unmanned aerial vehicle in round iteration +.>A local training round is performed; />Representing following unmanned aerial vehicle +.>Is a local data set sample number; />Representing following unmanned aerial vehicle +.>A floating point operand required to process each training data sample; />Representing following unmanned aerial vehicle +.>Is calculated according to the calculation speed of (2); />Representing following unmanned aerial vehicle +.>The size of the uploaded model parameter data volume; />Representing following unmanned aerial vehicle +.>Is used for the transmission power of the (a); />Representing allocation to following drone +.>Is a sub-bandwidth of (b); />Representation except following unmanned plane->A set of external following drones; />Representing an indicator +.>Representing following unmanned aerial vehicle +.>Transmitting its local model parameters to the following drone +.>，/>Representing following unmanned aerial vehicle +.>Without transmitting its local model parameters to the following drone +.>；/>Represents the path loss constant, +.>Representing a reference distance; />Representing following unmanned aerial vehicle +.>And follow unmanned plane +.>A distance therebetween; />Representing the path coefficients; />Representing following unmanned aerial vehicle +.>Channel noise at.
As a preferred technical solution of the present invention, in the step 3.2, constraints on bandwidth, calculation and communication time, and topology connection are specifically as follows:
constraint 1:；
constraint 2:；
constraint 3:；
in the formula ,representing the total bandwidth that the lead unmanned aerial vehicle can allocate; />Representing the maximum communication sub-bandwidth that can be allocated to the following unmanned aerial vehicle; />Representing a lead drone.
In step 3.3, the following unmanned aerial vehicle corresponding to each node in the tree topology structure performs parameter aggregation on the received local federal learning model parameters and the local federal learning model parameters thereof, and the model parameters obtained after the parameter aggregation are updated into the local federal learning model parameters of the node following unmanned aerial vehicle, and an aggregation formula is as follows;
，
wherein ,representing following unmanned aerial vehicle +.>Model parameters obtained after parameter aggregation are carried out; />Representing following unmanned aerial vehicle +.>Model parameters obtained after parameter aggregation are carried out; />Representing following unmanned aerial vehicle +.>The number of aggregation model parameters; />Representing following unmanned aerial vehicle +.>Is a local data set sample number; />Representing following unmanned aerial vehicle +.>Is a model parameter for the original federal learning.
In the step 4, the leading unmanned aerial vehicle performs parameter aggregation and updating for each received local federal learning model parameter according to the following formula to obtain the global federal learning model parameter in the next iteration:, in the formula ,/>Indicate->Global federal learning model parameters in the secondary iterations; />Representing the total number of following unmanned aerial vehicles connected with the leading unmanned aerial vehicle in a topological structure corresponding to the unmanned aerial vehicle cluster and related to model parameter transmission; />Representing a following unmanned aerial vehicle connected with a leading unmanned aerial vehicle in a topological structure corresponding to the unmanned aerial vehicle cluster and related to model parameter transmission; />Representing following unmanned aerial vehicle +.>In->Local federal learning model parameters in round iterations; />Representing following unmanned aerial vehicle +.>Is a local data set sample number; />Representation->The number follows the total number of local data set samples of the drone.
The beneficial effects of the invention are as follows: the invention provides an unmanned aerial vehicle cluster federal learning model optimization method based on topological optimization, which is oriented to unmanned aerial vehicle clusters, and can flexibly select equipment with favorable channels nearby to aggregate and forward local model parameters of a leading unmanned aerial vehicle according to the characteristics of high mobility, availability, simultaneity, expandability and the like of the unmanned aerial vehicle clusters, wherein the following unmanned aerial vehicle with harsh channel conditions can flexibly select the equipment with favorable channels nearby to aggregate and forward the local model parameters according to the bandwidth allocation, calculation time delay, communication time delay and constraint conditions of topological design, so that the weighted sum of the energy consumption and time delay of each round of unmanned aerial vehicle cluster federal learning is minimized. And the dynamic adjustment of the aggregation route of each follow unmanned aerial vehicle local model parameter and the topology structure of unmanned aerial vehicle cluster federal study avoids that the connection between some follow unmanned aerial vehicle and leading unmanned aerial vehicle can suffer high energy consumption and high delay transmission of deep fading in some iterative rounds due to channel fading, saves limited energy of unmanned aerial vehicle and realizes higher energy efficiency. In addition, the intelligent task of following the unmanned aerial vehicle to follow the leading unmanned aerial vehicle is realized by the method, the federal learning process can be remarkably accelerated, the robustness of federal learning of the unmanned aerial vehicle cluster is greatly improved, and meanwhile, the flexibility and the agility of federal learning in the unmanned aerial vehicle cluster can be improved, so that the federal learning can be better adapted to the unmanned aerial vehicle cluster with high dynamic network topology.
Drawings
Fig. 1 is a schematic diagram of a cluster federation learning scenario of an unmanned aerial vehicle based on topology optimization;
fig. 2 is a schematic diagram of a unmanned aerial vehicle module according to the present invention;
FIG. 3 is a flowchart of an unmanned aerial vehicle cluster federal learning optimization method based on topology optimization;
fig. 4 is a flowchart of the following individual unmanned aerial vehicle aggregation topology optimization execution in the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings. The following examples will provide those skilled in the art with a more complete understanding of the invention, but are not intended to limit the invention in any way.
A topology optimization-based unmanned aerial vehicle cluster federal learning model optimization method aims at unmanned aerial vehicle clusters formed by leading unmanned aerial vehicles and preset number of following unmanned aerial vehicles communicated with the leading unmanned aerial vehicles, and based on the fact that each following unmanned aerial vehicle has model training capability and a local data set, a federal learning model corresponding to the unmanned aerial vehicle clusters is obtained through the following steps.
In this embodiment, specific data of model parameters in federal learning are determined by models, and the trained models are different, so that the model parameters are different, and in this embodiment, the model parameters include weight parameters, bias parameters, and regularization parameters, and the weight parameters are parameters in a neural network and are used for calculating the output of neurons. In federal learning, the weight parameters are the main parameters of the model, and their updating is achieved by following training on the drone (local device). After each round of training is finished, the local device sends the updated weight parameters to the leading unmanned aerial vehicle (central server), and the leading unmanned aerial vehicle averages the parameters and sends the averaged parameters back to each following unmanned aerial vehicle. The bias parameters are parameters in the neural network for adjusting the output of the neurons. In federal learning, bias parameters are typically trained and updated along with weight parameters. Regularization of regularization parameters is a technique to avoid model overfitting. In federal learning, regularization parameters are typically trained and updated along with weighting parameters.
Various neural network model structures can be adopted in the federal learning model corresponding to the unmanned aerial vehicle cluster in federal learning, and specific selection depends on factors such as task requirements, data characteristics, computing resources and the like. If the neural network for image recognition is adopted, the image recognition model is obtained through the federal learning training process and is used as the federal learning model corresponding to the unmanned aerial vehicle cluster. And each follow-up unmanned aerial vehicle collects images as a local data set to carry out local federal learning model training corresponding to each follow-up unmanned aerial vehicle. In addition to this, there are many other neural network models, such as deep belief networks (Deep Belief Networks, DBN), automatic encoder networks (Autoencoder Networks), etc., which can be used in federal learning.
In this embodiment we considerAnd the set of the following unmanned aerial vehicles executes the federal learning task under the management of the leading unmanned aerial vehicle. The federal learning process includes iterative local model computation, uploading, and global model updating. Data communication between the drones is via a wireless link. Assuming that the lead unmanned aerial vehicle has perfect knowledge of the wireless channel gain and the computational characteristics of the following unmanned aerial vehicle, the result can be obtained through feedback.
The method is as shown in fig. 3, and a federal learning model corresponding to the unmanned aerial vehicle cluster is obtained through steps 1-4.
Step 1: broadcasting global federal learning model parameters to all following unmanned aerial vehicles by the leading unmanned aerial vehicle, judging whether federal learning model accuracy corresponding to the global federal learning model parameters meets a preset requirement, and if so, taking the federal learning model corresponding to the current global federal learning model parameters as a federal learning model corresponding to an unmanned aerial vehicle cluster; if the preset requirement is not met, executing the step 2. In this embodiment, the initial value of the global federal learning model parameter is obtained by initializing the leading unmanned aerial vehicle.
In the step 1, judging whether the precision of the federal learning model corresponding to the federal learning model parameter meets a preset requirement or not, judging through a global loss function and a preset iteration number, and if the global loss function converges or reaches the preset iteration number, the precision of the federal learning model corresponding to the federal learning model parameter meets the preset requirement; otherwise, the federal learning model precision corresponding to the federal learning model parameters does not reach the preset requirement; global loss functionAs follows;, in the formula ,/>Representing following unmanned aerial vehicle +.>Is a local data set sample number; />Representing the total number of following unmanned aerial vehicles; />Representing global federal learning model parameters; />Representing the global federal learning model parameter as +.>Lower->Sample loss function for individual data samples.
Step 2: aiming at each following unmanned aerial vehicle, the following unmanned aerial vehicle performs federal learning model training based on a local data set and combined with global federal learning model parameters broadcasted by the leading unmanned aerial vehicle, and after training is completed, local federal learning model parameters corresponding to each following unmanned aerial vehicle are obtained; step 3 is performed.
In the step 2, when the following unmanned aerial vehicle performs federal learning model training based on a local data set and combined with federal learning model parameters broadcasted by the leading unmanned aerial vehicle, the federal learning model parameters are updated through the following formula to obtain local federal learning model parameters:, in the formula ,/>Representing following unmanned aerial vehicle +.>In->Local federal learning model parameters obtained by training a local data set in round iteration; />Indicate->The global federal learning model parameters broadcasted by the unmanned aerial vehicle are followed in the round of iteration; />Representing a preset learning rate; />Representing the global federal learning model parameter as +.>Following unmanned aerial vehicle down +.>A local federal learning model loss function; />Representing vector differentiationOperators, in high numbers represent gradient operators. Following the model parameters broadcast by the unmanned aerial vehicle on reception of the leading unmanned aerial vehicle in the t-th training>The local data set is then used locally to independently complete the updating of the partial model parameters.
Step 3: the leading unmanned aerial vehicle builds a topological structure corresponding to the unmanned aerial vehicle cluster and related to model parameter transmission based on energy consumption and time delay of the current unmanned aerial vehicle cluster, and each following unmanned aerial vehicle transmits local federal learning model parameters to the leading unmanned aerial vehicle based on the topological structure; step 4 is performed.
The step 3 comprises the following steps:
step 3.1: the leading unmanned aerial vehicle builds an objective function based on the energy consumption and time delay of the current unmanned aerial vehicle cluster;
in the step 3.1, the objective functionThe formula is as follows:
，
in the formula ,，/>，，/>，
wherein ,representing following unmanned aerial vehicle +.>Is of the meter(s)Calculating energy consumption; />Representing following unmanned aerial vehicle +.>Is calculated by the time delay;representing following unmanned aerial vehicle +.>Is a communication energy consumption of the system; />Representing following unmanned aerial vehicle +.>Is a communication delay of (1); />Representing the total number of following unmanned aerial vehicles; />Representing a preset weighting factor; />Representing coefficients that depend on the chip architecture; />Indicate->Following unmanned aerial vehicle in round iteration +.>A local training round is performed; />Representing following unmanned aerial vehicle +.>Is a local data set sample number; />Representing following unmanned aerial vehicle +.>A floating point operand required to process each training data sample; />Representing following unmanned aerial vehicle +.>Is calculated according to the calculation speed of (2); />Representing following unmanned aerial vehicle +.>The size of the uploaded model parameter data volume; />Representing following unmanned aerial vehicle +.>Is used for the transmission power of the (a); />Representing allocation to following drone +.>Is a sub-bandwidth of (b); />Representation except following unmanned plane->A set of external following drones; />Representing an indicator +.>Representing following unmanned aerial vehicle +.>Transmitting its local model parameters to the following drone +.>，/>Representing following unmanned aerial vehicle +.>Without transmitting its local model parameters to the following drone +.>；/>Represents the path loss constant, +.>Representing a reference distance; />Representing following unmanned aerial vehicle +.>And follow unmanned plane +.>A distance therebetween; />Representing the path coefficients; />Representing following unmanned aerial vehicle +.>Channel noise at.
Topology structure corresponding to unmanned aerial vehicle cluster and related to model parameter transmission, in particular following unmanned aerial vehicleJudging whether other following unmanned aerial vehicles are required to be assisted to aggregate and transmit local model parameters (local federal learning model parameters) and selecting whether nearby following unmanned aerial vehicles are required to aggregate and forward the local model parameters to the leading unmanned aerial vehicle; the flow is shown in fig. 4, and includes the following steps:
(1) Following unmanned aerial vehicleJudging whether other following unmanned aerial vehicles need to be assisted to aggregate and forward local model parameters of the other following unmanned aerial vehicles to the leading unmanned aerial vehicle by receiving topology selection signals from the other following unmanned aerial vehicles;
(2) If the other following unmanned aerial vehicles need to be assisted in aggregation and forwarding the local model parameters to the leading unmanned aerial vehicle, the local model parameters of the other following unmanned aerial vehicles and the local model parameters of the other following unmanned aerial vehicles are received for aggregation; otherwise, directly carrying out the next step;
(3) Following unmanned aerial vehicleTopology selection is performed to judge whether nearby following unmanned aerial vehicle is selected>To aggregate and forward its local model parameters to the lead drone.
Definition of the definitionAs topology selection signal indicator, model parameter transmission and aggregation topology in a cluster federal learning system of unmanned aerial vehicles is uniquely specified if unmanned aerial vehicles are followed +.>Transmitting its local model parameters to the following drone +.>Equal to 1, otherwise 0. Topology selection signal indicator constructs unmanned aerial vehicle by leading unmanned aerial vehicle based on energy consumption and time delay of current unmanned aerial vehicle clusterThe topology of the cluster corresponding to the model parameter transmission is provided. Furthermore, during the design of the topology, i.e. following the unmanned plane + ->Follow-up unmanned aerial vehicle in judging whether to select the vicinity +.>When the local model parameters are aggregated and forwarded to the leading unmanned aerial vehicle, the weighted sum of the energy consumption and the time delay of each round of unmanned aerial vehicle cluster federal learning is minimized according to the constraint conditions of bandwidth allocation, calculation time delay, communication time delay and topological design, as shown in step 3.2. The energy consumption of each wheel comprises: />The total calculated energy consumption sum of the frame following unmanned aerial vehicle +.>The frame follows the total communication energy consumption of the unmanned aerial vehicle; the total time delay of each round comprises: />The frame follows the total calculated delay sum of the unmanned aerial vehicle +.>The rack follows the total communication delay of the drone.
Step 3.2: the leading unmanned aerial vehicle is based on an objective function, constraint conditions about bandwidth, calculation and communication time and topological connection are combined, the objective function is minimized, a tree topology structure corresponding to the unmanned aerial vehicle cluster and about model parameter transmission is constructed, and the tree topology structure takes the leading unmanned aerial vehicle as a root node.
In the step 3.2, constraint conditions about bandwidth, calculation and communication time and topology connection are as follows:
constraint 1: the bandwidth constraint for all unmanned aerial vehicles is specifically:；
constraint 2: the calculation and communication time constraints between all connected unmanned aerial vehicles are specifically:；
wherein a following drone (e.g. a following drone) as an aggregate forwarding) The local model parameters should be received and aggregated from its neighboring drones, so that it can only be received from the slave with +.>Is->The transmission is started after all model parameters are received. If follow unmanned plane->Is shorter than the following unmanned aerial vehicle +.>If it has to have an extra latency, obviously the extra latency has to be 0 in the optimal case, in which case we can reduce the following drone +.>To reduce energy consumption;
constraint 3: the topology constraint of the unmanned aerial vehicle is specifically as follows:；
wherein each following unmanned aerial vehicle is assumedOnly its local model parameters can be transmitted to one following drone and at least one following drone is directly connected to the lead drone.The topology of the unmanned aerial vehicle cluster federal learning system is defined as a tree topology, and the tree topology ensures that all local model parameters are aggregated once and only once;
in the formula ,representing the total bandwidth that the lead unmanned aerial vehicle can allocate; />Representing the maximum communication sub-bandwidth that can be allocated to the following unmanned aerial vehicle; />Representing a lead drone.
Step 3.3: based on a tree topology structure corresponding to the unmanned aerial vehicle cluster, transmitting local federal learning model parameters corresponding to the following unmanned aerial vehicle to the root node direction based on the tree topology structure by taking the following unmanned aerial vehicle corresponding to the leaf node as a starting point until the parameters are transmitted to the leading unmanned aerial vehicle; and the following unmanned aerial vehicle corresponding to each node in the tree topology structure carries out parameter aggregation on the received local federal learning model parameters and the local federal learning model parameters thereof, and the model parameters obtained after the parameter aggregation are updated into the local federal learning model parameters of the node following unmanned aerial vehicle.
The topological structure is shown in fig. 1, and fig. 1 is a schematic diagram of a scene of an unmanned aerial vehicle cluster federal learning model optimization method based on topological optimization. In the schematic scene, the leading unmanned aerial vehicle is used as a central server to receive local models of other unmanned aerial vehicles for model aggregation; for unmanned aerial vehicle No. 1, unmanned aerial vehicle No. 4 and unmanned aerial vehicle No. 6, they establish wireless links with leading unmanned aerial vehicle directly in order to transmit the local model respectively; for the unmanned aerial vehicle No. 3 and the unmanned aerial vehicle No. 5, the unmanned aerial vehicle No. 2 and the unmanned aerial vehicle No. 4 which are nearby and have favorable channels are flexibly selected to aggregate and forward the gradient to the leading unmanned aerial vehicle; for unmanned aerial vehicle No. 2, it not only needs to receive the local model that comes from unmanned aerial vehicle No. 3 transmission to carry out the model on this basis and update, later select to transmit the local model after the polymerization update to nearby unmanned aerial vehicle No. 1 to aggregate and forward its gradient to leading unmanned aerial vehicle.
In the step 3.3, the following unmanned aerial vehicle corresponding to each node in the tree topology structure performs parameter aggregation on the received local federal learning model parameters and the local federal learning model parameters thereof, and the model parameters obtained after the parameter aggregation are updated into the local federal learning model parameters of the node following unmanned aerial vehicle, wherein an aggregation formula is as follows;
，
wherein ,representing following unmanned aerial vehicle +.>Model parameters obtained after parameter aggregation are carried out; />Representing following unmanned aerial vehicle +.>Model parameters obtained after parameter aggregation are carried out; />Representing following unmanned aerial vehicle +.>The number of aggregation model parameters; />Representing following unmanned aerial vehicle +.>Is a local data set sample number; />Representing following unmanned aerial vehicle +.>Original federal learning model of (1)Type parameters.
Step 4: and (3) aiming at each received local federal learning model parameter, the leading unmanned aerial vehicle carries out parameter aggregation and updating to obtain the global federal learning model parameter in the next iteration, and returns to the execution step (1).
In the step 4, the leading unmanned aerial vehicle performs parameter aggregation and updating according to the received local federal learning model parameters through the following formula to obtain global federal learning model parameters in the next iteration:, in the formula ,/>Indicate->Global federal learning model parameters in the secondary iterations; />Representing the total number of following unmanned aerial vehicles connected with the leading unmanned aerial vehicle in a topological structure corresponding to the unmanned aerial vehicle cluster and related to model parameter transmission; />Representing a following unmanned aerial vehicle connected with a leading unmanned aerial vehicle in a topological structure corresponding to the unmanned aerial vehicle cluster and related to model parameter transmission; />Representing following unmanned aerial vehicle +.>In->Local federal learning model parameters in round iterations; />Representing following unmanned aerial vehicle +.>Is a local data set sample number; />Representation->The number follows the total number of local data set samples of the drone.
According to the unmanned aerial vehicle cluster federation learning model optimization method based on topology optimization, unlike a conventional centralized federation learning method, each node is required to be directly connected with a central server to transmit and aggregate model parameters, the unmanned aerial vehicle cluster federation learning model optimization method based on topology optimization optimizes the aggregation topology, minimizes the weighted sum of energy consumption and time delay of each round of unmanned aerial vehicle cluster federation learning under the constraint conditions of bandwidth allocation, calculation time delay, communication time delay and topology design, enables federation learning to be better suitable for the high dynamic property of unmanned aerial vehicle cluster network topology, improves the communication efficiency and federation learning process of unmanned aerial vehicle clusters, enhances the robustness of unmanned aerial vehicle ad hoc network federation learning, and optimizes the overall performance of unmanned aerial vehicle cluster federation learning.
Based on the above method, as shown in fig. 2, in this embodiment, each following unmanned aerial vehicle is installed with a learning system for participating in federal learning model training, and the learning system includes: the model receiving module is used for receiving model parameters sent from other unmanned aerial vehicles in the training process of each wheel; the model updating module is used for training a model by using the unmanned aerial vehicle local data and updating local model parameters; the model aggregation module is used for aggregating the local model received from other unmanned aerial vehicles with the local model of the unmanned aerial vehicle; and the model sending module is used for sending the updated local model to other unmanned aerial vehicles. Wherein a learning system for federal learning model training management is installed on the lead unmanned aerial vehicle, the learning system includes: the model receiving module is used for receiving local model parameters sent from other unmanned aerial vehicles in the training process of each wheel; the aggregation module is used for aggregating local model parameters received from other unmanned aerial vehicles and updating a global model; the broadcasting module is used for broadcasting the updated new round of model parameters to each adjacent following unmanned aerial vehicle; and the learning management module is used for carrying out convergence judgment and regulation on the whole federal learning model training process.
The invention designs an unmanned aerial vehicle cluster federal learning model optimization method based on topology optimization, which is oriented to unmanned aerial vehicle clusters, and can flexibly select equipment with favorable channels nearby to aggregate and forward local model parameters of a leading unmanned aerial vehicle according to the characteristics of high mobility, availability, simultaneity, expandability and the like of the unmanned aerial vehicle clusters, wherein the following unmanned aerial vehicle with harsh channel conditions can flexibly select equipment with favorable channels nearby to aggregate and forward the local model parameters according to bandwidth allocation, calculation time delay, communication time delay and constraint conditions of topology design, and the energy consumption and the weighted sum of time delay of each round of unmanned aerial vehicle cluster federal learning are minimized. And the dynamic adjustment of the aggregation route of each follow unmanned aerial vehicle local model parameter and the topology structure of unmanned aerial vehicle cluster federal study avoids that the connection between some follow unmanned aerial vehicle and leading unmanned aerial vehicle can suffer high energy consumption and high delay transmission of deep fading in some iterative rounds due to channel fading, saves limited energy of unmanned aerial vehicle and realizes higher energy efficiency. In addition, the intelligent task of following the unmanned aerial vehicle to follow the leading unmanned aerial vehicle is realized by the method, the federal learning process can be remarkably accelerated, the robustness of federal learning of the unmanned aerial vehicle cluster is greatly improved, and meanwhile, the flexibility and the agility of federal learning in the unmanned aerial vehicle cluster can be improved, so that the federal learning can be better adapted to the unmanned aerial vehicle cluster with high dynamic network topology.
Although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that the foregoing embodiments may be modified or equivalents substituted for some of the features thereof. All equivalent structures made by the content of the specification and the drawings of the invention are directly or indirectly applied to other related technical fields, and are also within the scope of the invention.
Claims (6)
1. The unmanned aerial vehicle cluster federal learning model optimization method based on topological optimization is characterized in that aiming at an unmanned aerial vehicle cluster formed by a leading unmanned aerial vehicle and a preset number of following unmanned aerial vehicles communicated with the leading unmanned aerial vehicle, the federal learning model corresponding to the unmanned aerial vehicle cluster is obtained based on the model training capability and a local data set of each following unmanned aerial vehicle by the following steps:
step 1: broadcasting global federal learning model parameters to all following unmanned aerial vehicles by the leading unmanned aerial vehicle, judging whether federal learning model accuracy corresponding to the global federal learning model parameters meets a preset requirement, and if so, taking the federal learning model corresponding to the current global federal learning model parameters as a federal learning model corresponding to an unmanned aerial vehicle cluster; if the preset requirement is not met, executing the step 2;
step 2: aiming at each following unmanned aerial vehicle, the following unmanned aerial vehicle performs federal learning model training based on a local data set and combined with global federal learning model parameters broadcasted by the leading unmanned aerial vehicle, and after training is completed, local federal learning model parameters corresponding to each following unmanned aerial vehicle are obtained; executing the step 3;
step 3: the leading unmanned aerial vehicle builds a topological structure corresponding to the unmanned aerial vehicle cluster and related to model parameter transmission based on energy consumption and time delay of the current unmanned aerial vehicle cluster, and each following unmanned aerial vehicle transmits local federal learning model parameters to the leading unmanned aerial vehicle based on the topological structure; executing the step 4;
step 4: the leading unmanned aerial vehicle carries out parameter aggregation and updating aiming at each received local federal learning model parameter to obtain a global federal learning model parameter in the next iteration, and returns to the execution step 1;
the step 3 comprises the following steps 3.1 to 3.3;
step 3.1: the leading unmanned aerial vehicle constructs an objective function O formula based on the energy consumption and time delay of the current unmanned aerial vehicle cluster as follows:
in the formula ,
wherein ,representing the calculated energy consumption of the following unmanned aerial vehicle i; />Representing a calculation time delay following the unmanned aerial vehicle i; />Representing the communication energy consumption of the following unmanned aerial vehicle i; />Representing the communication time delay following the unmanned plane i; k represents the total number of following unmanned aerial vehicles; μ represents a preset weighting factor; gamma represents a coefficient depending on the chip architecture; n (N) i (t) represents the local training round followed by drone i in the t-th round of iteration; d (D) i Representing the number of samples of the local data set following the drone i; θ i Representing floating point operands required to process each training data sample following unmanned plane i; f (f) i Representing a calculated speed of the following unmanned aerial vehicle i; s is S i The size of the data volume of the model parameters uploaded by the following unmanned aerial vehicle i is represented; p (P) i Representing the transmission power of the following unmanned aerial vehicle i; b i Representing the sub-bandwidth allocated to the following drone i; k (K) 0 Representing a set of following unmanned aerial vehicles other than following unmanned aerial vehicle i; i i,j Representing indicators, I i,j =1 means that the following unmanned aerial vehicle I transmits its local model parameters to the following unmanned aerial vehicle j, I i,j =0 means that following unmanned plane i does notTransmitting its local model parameters to the following unmanned aerial vehicle j; g 0 Represents the path loss constant, d 0 Representing a reference distance; d, d i,j Representing a distance between the following unmanned aerial vehicle i and the following unmanned aerial vehicle j; alpha represents a path coefficient; />Representing channel noise at following drone i;
step 3.2: the leading unmanned aerial vehicle is based on an objective function, combined with constraint conditions about bandwidth, calculation and communication time and topological connection, and takes the objective function minimization as a target, a tree topology structure which corresponds to the unmanned aerial vehicle cluster and is used for model parameter transmission is constructed, and the tree topology structure takes the leading unmanned aerial vehicle as a root node;
step 3.3: based on a tree topology structure corresponding to the unmanned aerial vehicle cluster, transmitting local federal learning model parameters corresponding to the following unmanned aerial vehicle to the root node direction based on the tree topology structure by taking the following unmanned aerial vehicle corresponding to the leaf node as a starting point until the parameters are transmitted to the leading unmanned aerial vehicle; and the following unmanned aerial vehicle corresponding to each node in the tree topology structure carries out parameter aggregation on the received local federal learning model parameters and the local federal learning model parameters thereof, and the model parameters obtained after the parameter aggregation are updated into the local federal learning model parameters of the node following unmanned aerial vehicle.
2. The optimization method of the unmanned aerial vehicle cluster federation learning model based on the topological optimization according to claim 1 is characterized in that in the step 1, whether the federation learning model precision corresponding to federation learning model parameters meets a preset requirement is judged, the global loss function and the preset iteration times are used for judging, and if the global loss function converges or reaches the preset iteration times, the federation learning model precision corresponding to federation learning model parameters meets the preset requirement; otherwise, the federal learning model precision corresponding to the federal learning model parameters does not reach the preset requirement; the global loss function L (ω) is as follows;
in the formula ,Di Representing the number of samples of the local data set following the drone i; k represents the total number of following unmanned aerial vehicles; omega represents a global federal learning model parameter; l (ω, δ) represents the sample loss function of the δ -th data sample at ω as a global federal learning model parameter.
3. The optimization method of the federal learning model of the unmanned aerial vehicle cluster based on the topological optimization according to claim 1, wherein in the step 2, when the federal learning model training is performed by following the unmanned aerial vehicle based on a local data set and combining federal learning model parameters broadcasted by the leading unmanned aerial vehicle, the federal learning model parameters are updated through the following formula to obtain local federal learning model parameters;
in the formula ,representing local federal learning model parameters obtained by training the following unmanned aerial vehicle i by utilizing a local data set in a t-th round of iteration; omega t Representing global federal learning model parameters following unmanned aerial vehicle broadcasting in the t-th iteration; ρ represents a preset learning rate; l (L) i (ω t ) Representing global federal learning model parameters as ω t The local federal learning model loss function of unmanned aerial vehicle i follows.
4. The optimization method of the unmanned aerial vehicle cluster federal learning model based on the topological optimization according to claim 1, wherein in the step 3.2, constraint conditions about bandwidth, calculation and communication time and topological connection are as follows:
constraint 1:
constraint 2:
constraint 3:
wherein B represents the total bandwidth which can be allocated by the leading unmanned aerial vehicle; b max Representing the maximum communication sub-bandwidth that can be allocated to the following unmanned aerial vehicle; k+1 represents the leading unmanned aerial vehicle.
5. The optimization method of the unmanned aerial vehicle cluster federation learning model based on the topological optimization according to claim 1, wherein in the step 3.3, the following unmanned aerial vehicle corresponding to each node in the tree topology structure performs parameter aggregation on the received local federation learning model parameters and the local federation learning model parameters thereof, and the model parameters obtained after the parameter aggregation are updated into the local federation learning model parameters of the node following unmanned aerial vehicle, and an aggregation formula is as follows;
wherein ,representing model parameters obtained after parameter aggregation of the following unmanned aerial vehicle j; />Representing model parameters obtained after parameter aggregation following the unmanned plane i; />Representing the number of aggregation model parameters at the following unmanned plane i; d (D) j Representing local number following unmanned aerial vehicle jThe number of data set samples; u (u) j And representing the parameters of the original federal learning model following the unmanned aerial vehicle j.
6. The optimization method of the unmanned aerial vehicle cluster federation learning model based on the topological optimization according to claim 1, wherein in the step 4, the leading unmanned aerial vehicle performs parameter aggregation and updating for each received local federation learning model parameter by the following formula to obtain the global federation learning model parameter in the next iteration:
in the formula ,ωt+1 Representing global federal learning model parameters in the t+1st iteration; n represents the total number of following unmanned aerial vehicles connected with the leading unmanned aerial vehicle in a topological structure corresponding to the unmanned aerial vehicle cluster and related to model parameter transmission; n represents a following unmanned aerial vehicle connected with a leading unmanned aerial vehicle in a topological structure corresponding to the unmanned aerial vehicle cluster and related to model parameter transmission;representing local federal learning model parameters of the following unmanned aerial vehicle n in the t-th round of iteration; d (D) n Representing the number of samples of the local data set following unmanned plane n; d= Σ n∈N D n Representing the total number of samples of the local data set of the N following drones.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310828232.5A CN116582871B (en) | 2023-07-07 | 2023-07-07 | Unmanned aerial vehicle cluster federal learning model optimization method based on topology optimization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310828232.5A CN116582871B (en) | 2023-07-07 | 2023-07-07 | Unmanned aerial vehicle cluster federal learning model optimization method based on topology optimization |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116582871A CN116582871A (en) | 2023-08-11 |
| CN116582871B true CN116582871B (en) | 2023-10-13 |
Family
ID=87545568
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310828232.5A Active CN116582871B (en) | 2023-07-07 | 2023-07-07 | Unmanned aerial vehicle cluster federal learning model optimization method based on topology optimization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116582871B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117135693B (en) * | 2023-10-27 | 2024-01-23 | 四川长虹新网科技有限责任公司 | Real-time service distribution method based on federal learning under multi-AP environment |
| CN117289725B (en) * | 2023-11-27 | 2024-02-27 | 清华大学 | Unmanned plane distributed general calculation integrated resource scheduling method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113971461A (en) * | 2021-10-26 | 2022-01-25 | 南京航空航天大学 | Distributed federal learning method and system for unmanned aerial vehicle ad hoc network |
| CN114169412A (en) * | 2021-11-23 | 2022-03-11 | 北京邮电大学 | Federal learning model training method for large-scale industrial chain privacy calculation |
| CN115376031A (en) * | 2022-10-24 | 2022-11-22 | 江西省科学院能源研究所 | Road unmanned aerial vehicle routing inspection data processing method based on federal adaptive learning |
| WO2023072049A1 (en) * | 2021-10-28 | 2023-05-04 | 华为技术有限公司 | Federated learning method and related apparatus |
-
2023
- 2023-07-07 CN CN202310828232.5A patent/CN116582871B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113971461A (en) * | 2021-10-26 | 2022-01-25 | 南京航空航天大学 | Distributed federal learning method and system for unmanned aerial vehicle ad hoc network |
| WO2023072049A1 (en) * | 2021-10-28 | 2023-05-04 | 华为技术有限公司 | Federated learning method and related apparatus |
| CN114169412A (en) * | 2021-11-23 | 2022-03-11 | 北京邮电大学 | Federal learning model training method for large-scale industrial chain privacy calculation |
| CN115376031A (en) * | 2022-10-24 | 2022-11-22 | 江西省科学院能源研究所 | Road unmanned aerial vehicle routing inspection data processing method based on federal adaptive learning |
Non-Patent Citations (2)
| Title |
|---|
| 基于边端协同的ＣＮＮ 推理加速框架;郭永安，周金粮，王宇翱;《南京邮电大学学报》;全文 * |
| 边缘智能驱动的无人机网络计算任务卸载及隐私保护方案研究;郑镐;《中国优秀硕士论文全文数据库》;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116582871A (en) | 2023-08-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116582871B (en) | Unmanned aerial vehicle cluster federal learning model optimization method based on topology optimization | |
| CN113162679B (en) | DDPG algorithm-based IRS (intelligent resilient software) assisted unmanned aerial vehicle communication joint optimization method | |
| CN108337034B (en) | Full-duplex unmanned aerial vehicle mobile relay system and path optimization method thereof | |
| WO2021017227A1 (en) | Path optimization method and device for unmanned aerial vehicle, and storage medium | |
| CN110968426B (en) | Edge cloud collaborative k-means clustering model optimization method based on online learning | |
| CN113132943B (en) | Task unloading scheduling and resource allocation method for vehicle-side cooperation in Internet of vehicles | |
| CN113406974B (en) | Learning and resource joint optimization method for unmanned aerial vehicle cluster federal learning | |
| CN109327252B (en) | Online learning beam selection method based on contact context | |
| CN114650567B (en) | Unmanned aerial vehicle auxiliary V2I network task unloading method | |
| CN114980169A (en) | Unmanned aerial vehicle auxiliary ground communication method based on combined optimization of track and phase | |
| CN113448703A (en) | Unmanned aerial vehicle swarm dynamic reconnaissance task scheduling system and method based on perception array | |
| CN115278905B (en) | Multi-node communication opportunity determination method for unmanned aerial vehicle network transmission | |
| CN116847293A (en) | Combined buffer decision and track optimization method under unmanned aerial vehicle auxiliary vehicle networking | |
| CN115133977B (en) | Unmanned aerial vehicle communication perception integrated system position optimization method based on information age minimization | |
| CN106792982B (en) | Multi-target direct positioning method based on self-adaptive clustering strategy | |
| CN115665805A (en) | Point cloud analysis task-oriented edge computing resource scheduling method and system | |
| KR20230060505A (en) | Communication method for federated learning and device performing the same | |
| Zhang et al. | Cybertwin-driven multi-intelligent reflecting surfaces aided vehicular edge computing leveraged by deep reinforcement learning | |
| WO2023116787A1 (en) | Intelligent model training method and apparatus | |
| WO2023279967A1 (en) | Intelligent model training method and device | |
| CN114745693B (en) | PSO-GA hybrid algorithm-based UAV auxiliary Internet of vehicles resource allocation method | |
| CN117055621B (en) | Data acquisition-oriented multi-unmanned aerial vehicle path planning method | |
| CN116778363B (en) | Low-traffic reservoir area water environment risk identification method based on federal learning | |
| Liu et al. | Intelligent Beamforming for UAV Assisted IIoT Based on Hypergraph Inspired Explainable Deep Learning | |
| Wu et al. | UAV-Assisted Data Synchronization for Digital-Twin-Enabled Vehicular Networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |