Heterogeneous perception federated learning method through unmanned aerial vehicle sampling and D2D communication

Abstract

According to the heterogeneous sensing federated learning method based on unmanned aerial vehicle sampling and D2D communication, all problems caused by unmanned aerial vehicle heterogeneity, limited communication resources and significant overlapping of local data distribution of unmanned aerial vehicles are considered, and the heterogeneous sensing federated learning method based on unmanned aerial vehicle sampling and D2D communication is provided and utilizes a submodule maximization algorithm and a greedy algorithm to jointly optimize unmanned aerial vehicle sampling and end-to-end communication. The invention provides a fast convergence unmanned aerial vehicle sampling and end-to-end resource allocation method, which not only can improve the efficiency of federal learning training, but also can ensure the participation degree of the unmanned aerial vehicle so as to improve the performance of a global model while protecting the data privacy of the unmanned aerial vehicle. The method provided by the invention is fast in convergence and high in efficiency when all models are updated, especially for the problem of low utilization rate of unmanned aerial vehicle sampling and computing resources in a dynamic heterogeneous environment, fast convergence can be realized under the condition of data heterogeneous, the time delay requirement can be met, and meanwhile, the convergence speed of the algorithm and the accuracy of the models are improved.

Description

Technical Field

[0001] The present invention belongs to the field of communication technologies, and particularly relates to a heterogeneous-aware federated learning method through drone sampling and D2D (Device-to-Device) communication. Background Art

[0002] The centralized training method of traditional artificial intelligence models requires aggregating all raw data onto a central server, which is impractical for wireless networks due to bandwidth consumption and privacy issues. Federated learning provides a solution by supporting iterative model training across different participants, coordinated by a central server, thus balancing data privacy and model performance and proving to be more effective in a wireless network environment.

[0003] Federated learning has been applied to many applications, such as optimizing resource allocation in vehicle-to-vehicle communication and content recommendation on smartphones. However, existing federated learning methods cannot effectively solve the problems caused by client heterogeneity, limited communication resources, and significant overlap in the local data distribution of drones, resulting in low utilization of computing resources and slow convergence speed. Therefore, there is an urgent need for a method to effectively solve the above problems. Summary of the Invention

[0004] To solve the above problems existing in the prior art, the present invention provides a heterogeneous-aware federated learning method through drone sampling and D2D communication. The technical problems to be solved by the present invention are realized through the following technical solutions: A heterogeneous-aware federated learning method through drone sampling and D2D communication is applied to an edge federated learning system composed of an edge server and multiple drones. The heterogeneous-aware federated learning method through drone sampling and D2D communication includes: S100, in the current training round, the edge server transmits the globally updated model of the previous training round to all drones; S200, in the current training round, the edge server selects sampling drones from the drones that meet the latency constraint and divides them into first sampling drones for direct communication and second sampling drones for indirect communication using the total latency; S300, in the current training round, both the first sampling drones and the second sampling drones locally train the globally updated model transmitted by the edge server using local datasets; the second sampling drones transmit the trained model parameters to the matching first sampling drones, and the first sampling drones aggregate their own model parameters and the model parameters sent by the second sampling drones and then upload them to the edge server; S400, in the current training round, the edge server performs weighted average aggregation on the model parameters uploaded by each drone to obtain aggregated model parameters, and uses the aggregated model parameters to update the global model of the previous training round to obtain the global model of the current training round; S500, repeating the process from S100 to S400 until the global model converges.

[0005] Beneficial effects: The present invention takes into account all the problems caused by the heterogeneity of drones, limited communication resources, and significant overlap of local data distribution of drones, and provides a heterogeneous perception federated learning method through drone sampling and D2D communication. The method uses a submodule maximization algorithm and a greedy algorithm to jointly optimize drone sampling and end-to-end communication, and realizes the effective use of computing resources while protecting the privacy of drone data. In addition, the present invention proposes a fast-converging drone sampling and end-to-end resource allocation method, which can not only improve the efficiency of federated learning training, but also ensure the participation of drones to improve the performance of the global model. The method proposed by the present invention converges quickly and efficiently when updating all models, especially for the problem of low utilization of drone sampling and computing resources in dynamic heterogeneous environments. It can converge quickly and meet the delay requirements in the case of data heterogeneity, while improving the convergence speed of the algorithm and the accuracy of the model.

[0006] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 It is a flow chart of the heterogeneous perception federated learning method through drone sampling and D2D communication provided by the present invention. DETAILED DESCRIPTION

[0008] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0009] The present invention provides a heterogeneous perception federated learning method through drone sampling and D2D communication, which is applied to an edge federated learning system composed of an edge server and multiple drones. The multiple drones in the edge federated learning system constitute a drone set, which is expressed as , each drone has a size of Local dataset ;in, is the total number of drones, The serial number of the drone.

[0010] The goal of this invention is to train a global model that performs well on data from all drones. , each drone uses a size of Local dataset , and uses gradient descent to train this global model, but in federated learning, each drone computes its local gradients, then uploads these gradients to an edge server (or aggregates them securely), which then updates the global model.

[0011] like Figure 1 As shown, the present invention provides a heterogeneous perception federated learning method through drone sampling and D2D communication, including: S100, in the current training round, the edge server transmits the global model updated in the previous training round to all drones; S200, in the current training round, the edge server selects sampling drones from the drones that meet the delay constraint, and divides them into first sampling drones for direct communication and second sampling drones for indirect communication using the total delay; It is worth noting that the allocation to drones The more bandwidth a drone has, the The smaller the delay is. Since the delay of each round of federated learning is determined by the slowest UAV, the present invention can reallocate bandwidth from faster UAVs to slower UAVs to reduce the delay of federated learning. Therefore, the optimal solution can only be achieved when all bandwidths are allocated and all selected UAVs have the same completion time. The binary search method can be used to find the optimal delay value and bandwidth allocation, that is, determine the sampling UAVs, and divide the sampling UAVs into indirect and direct UAVs, and then perform the process of updating the global model through federated learning.

[0012] The present invention can sample and obtain a set of sampled drones, which then need to be divided into two categories: one is a set of drones that communicate directly with the edge server , one is a collection of drones that communicate indirectly with edge servers The purpose of dividing the sampling drones into two categories is to reduce the overall system latency and alleviate the impact of stragglers. The specific implementation details are described in detail below.

[0013] In one embodiment of the present invention, S200 includes: S210, in the current training round, the edge server determines a drone that satisfies a delay constraint condition among all drones; S220, in the current training round, the edge server selects a sampling drone from the drones that meet the delay constraint condition; In this step, in the current training round, the edge server uses the submodule maximization algorithm to select a predetermined number of drones from the drones that meet the delay constraint conditions to form a sampling drone set. ; Among them, the drones that meet the delay constraint conditions are the drones within the predetermined delay range.

[0014] After S220, the heterogeneous perception federated learning method through drone sampling and D2D communication further includes: In the current training round, the edge server targets the sampled drone set For each sampling drone in The distance between other sampling drones in the sampling process, if the distance is not greater than the maximum coverage range of the sampling drone , then add the number of the other sampling drone to the end-to-end communication drone set corresponding to the sampling drone middle.

[0015] Specifically, for sampling drones Perform the following steps: Computational Sampling Drones The distance between other drones, if the distance between them is less than or equal to the maximum coverage range of the drone , then add each other to their respective end-to-end communication drone sets, that is, , Indicates sampling drone A collection of end-to-end communication drones, Indicates the numbers of other sampling drones; represents the set of end-to-end communicating drones of other sampling drones, It is a sampling drone Number.

[0016] S230: In the current training round, the edge server divides the sampling drones into first sampling drones for direct communication and second sampling drones for indirect communication using the total delay.

[0017] In an optional implementation manner of the present invention, S230 includes: S231, in the current training round, the edge server initializes the set of drones for direct communication and indirectly communicate with drone collections is empty; S232, in the current training round, the edge server uses a greedy algorithm to select a sampling drone with the smallest total delay and add it to the set of drones for direct communication In and from the sampling drone collection Remove the sampling drone number; S233, in the current training round, the edge server performs a , if the sampling drone If the membership condition and delay condition are met, the sampling drone will be Added the number of the drone to the collection of indirect communication drones In and from the sampling drone collection Remove the sampling drone The number of; wherein the membership condition is the sampling drone The number of peer-to-peer communication drones in other sampling drones collection The time delay condition is the sampling drone Local training latency and sampling drones To the sampling drone The sum of the transmission delays of the sampling drone is no greater than The total delay of

[0018] It is worth noting that for sampling drones If the sampling drone and If so, add a sampling drone arrive , and from Removed sampling drone , Indicates sampling drone A collection of end-to-end communication drones, Indicates that in the current training round , sampling drone To the sampling drone The transmission delay of It is a sampling drone The local training delay is In the current training round , sampling drone The total delay of

[0019] The total delay of federated learning mainly includes the transmission delay of the model parameters transmitted by the drone and the local training delay. All drones share the bandwidth resources of the edge server, so the sampling drone The total delay is expressed as: ; ; ; In the formula, In the current training round , sampling drone The total delay of In the current training round , sampling drone transmission delay; For sampling drones Local training latency; is the size of the model parameters; In the current training round , sampling drone The transmission rate; In the current training round , sampling drone a bandwidth allocated by the edge server; , is the total upstream bandwidth, In the current training round , sampling drone The transmission power; In the current training round , sampling drone Channel gain to edge servers; is the power spectral density of Gaussian noise.

[0020] S234, the edge server repeats S232 to S233 until the sampling drone set is an empty set, and the set of drones that communicate directly is obtained and indirectly communicate with drone collections ; wherein the drone set The first sampling drone is included in the At least one second sampling drone is included.

[0021] S300, in the current training round, the first sampling drone and the second sampling drone both train the global model transmitted by the edge server locally using a local data set; the second sampling drone transmits the trained model parameters to the matching first sampling drone, and the first sampling drone aggregates its own model parameters and the model parameters sent by the second sampling drone and then uploads them to the edge server; The model parameters include the gradient obtained by training the global model using the gradient descent method.

[0022] S400, in the current training round, the edge server performs weighted average aggregation on the model parameters uploaded by each drone to obtain aggregated model parameters, and uses the aggregated model parameters to update the global model of the previous training round to obtain the global model of the current training round; It is worth noting that the edge server does not participate in the training of the global model. Its main function is to perform weighted average aggregation of the model parameters trained by each drone to obtain the aggregated parameters of the global model of the current training round, and provide an updated global model for the distributed (drone) training of the next training round. The following is a detailed description of the aggregation and update process: After receiving updates from multiple drones, the edge server calculates a weighted average based on the amount of data from each drone. The weighted formula is , It is The amount of data from drones, is the total number of drones participating in the update, yes One of the drones with updated parameters; The server uses the aggregated weights to update the global model parameters. , the model parameters of the updated global model It can be calculated by the following formula: ; in It is The updated model parameters of the UAV itself.

[0023] For example, suppose there are three drones, A, B, and C, with 100, 200, and 300 data records, respectively. After a round of training of the global model, each drone uploads its own updated model parameters. The model parameters of drone A are , the data volume is 100. The model parameters of drone B are , the data volume is 200. The model parameters of UAV C are , the data volume is 300. First, the model parameters of each drone are calculated, expressed as: ; ; ; The edge server then performs weighted aggregation on these model parameters to generate aggregated parameters, which are used to generate model parameters of the new global model. : ; In this way, the edge server obtains the aggregated model parameters, which integrate the contributions of all participating drones and more accurately reflect the overall data distribution. Knowing the aggregated model parameters, the aggregated model parameters can be used to update the global model of the previous training round to obtain the global model of the current training round, that is, the updated global model.

[0024] S500, repeating the process from S100 to S400 until the global model converges.

[0025] Because each round of federated learning is synchronous, the latency depends on the slowest drone, so In the current training round The total delay used to update all models is expressed as ,in, In the current training round , sampling drone transmission delay; The value of is 1; For sampling drones Local training latency; In the current training round , sampling drone The total delay of

[0026] The present invention uses a submodule maximization algorithm and a greedy algorithm to jointly optimize drone sampling and end-to-end communication. This method is aimed at the existing federated learning method and cannot effectively solve all the problems caused by the heterogeneity of drones, limited communication resources and significant overlap of local data distribution of drones, resulting in low computing resource utilization, slow convergence speed and other problems. The technical solution of the present invention takes into account all the problems caused by the heterogeneity of drones, limited communication resources and significant overlap of local data distribution of drones. Through adaptive drone sampling and end-to-end transmission, it realizes the effective utilization of computing resources while protecting the privacy of drone data, and proposes a fast-converging drone sampling and end-to-end allocation method, which can not only improve the efficiency of federated learning training, but also ensure the participation of drones (i.e. drones) to improve model performance. The method proposed by the present invention is novel, converges quickly and efficiently when updating the global model, especially for the problem of low utilization of drone sampling and computing resources in dynamic heterogeneous environments, and can converge quickly and meet the delay requirements in the case of data heterogeneity, while improving the convergence speed of the algorithm and the accuracy of the model.

[0027] It is worth noting that the terms "first" and "second" in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0028] Although the present application is described herein in conjunction with various embodiments, in the process of implementing the claimed application, those skilled in the art may understand and implement other variations of the disclosed embodiments by reviewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality of components or steps.

[0029] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A heterogeneous sensing federated learning method through drone sampling and D2D communication, characterized in that: Applied to an edge federated learning system consisting of an edge server and multiple drones, the heterogeneous perception federated learning method through drone sampling and D2D communication includes: S100, in the current training round, the edge server transmits the global model updated in the previous training round to all drones; S200, in the current training round, the edge server selects sampling drones from the drones that meet the delay constraint, and divides them into first sampling drones for direct communication and second sampling drones for indirect communication using the total delay; S300, in the current training round, the first sampling drone and the second sampling drone both train the global model transmitted by the edge server locally using a local data set; the second sampling drone transmits the trained model parameters to the matching first sampling drone, and the first sampling drone aggregates its own model parameters and the model parameters sent by the second sampling drone and then uploads them to the edge server; S400, in the current training round, the edge server performs weighted average aggregation on the model parameters uploaded by each drone to obtain aggregated model parameters, and uses the aggregated model parameters to update the global model of the previous training round to obtain the global model of the current training round; S500, repeating the process from S100 to S400 until the global model converges.

2. The heterogeneous sensing federated learning method through drone sampling and D2D communication according to claim 1 is characterized in that: Multiple drones in the edge federated learning system constitute a drone set, which is represented by , each drone has a size of Local dataset ;in, is the total number of drones, is the serial number of the drone; all drones share the bandwidth resources of the edge server.

3. The heterogeneous sensing federated learning method through drone sampling and D2D communication according to claim 2 is characterized in that: S200 includes: S210, in the current training round, the edge server determines a drone that satisfies the delay constraint condition among all drones; S220, in the current training round, the edge server selects a sampling drone from drones that meet the delay constraint condition; S230: In the current training round, the edge server divides the sampling drones into first sampling drones for direct communication and second sampling drones for indirect communication using the total delay.

4. The heterogeneous sensing federated learning method through drone sampling and D2D communication according to claim 3 is characterized in that: S220 includes: In the current training round, the edge server uses the submodule maximization algorithm to select a predetermined number of drones from the drones that meet the delay constraint conditions to form a sampling drone set. ; Among them, the drones that meet the delay constraint conditions are the drones within the predetermined delay range.

5. The heterogeneous sensing federated learning method through drone sampling and D2D communication according to claim 4 is characterized in that: After S220, the heterogeneous perception federated learning method through drone sampling and D2D communication further includes: In the current training round, the edge server targets the sampled drone set For each sampling drone in The distance between other sampling drones in the sampling process, if the distance is not greater than the maximum coverage range of the sampling drone , then add the number of the other sampling drone to the end-to-end communication drone set corresponding to the sampling drone middle.

6. The heterogeneous sensing federated learning method through drone sampling and D2D communication according to claim 4 is characterized in that: S230 includes: S231, in the current training round, the edge server initializes the set of drones for direct communication and indirectly communicate with drone collections is empty; S232, in the current training round, the edge server uses a greedy algorithm to select a sampling drone with the smallest total delay and add it to the set of drones for direct communication In and from the sampling drone collection Remove the sampling drone number; S233, in the current training round, the edge server performs a , if the sampling drone If the membership condition and delay condition are met, the sampling drone will be Added the number of the drone to the collection of indirect communication drones In and from the sampling drone collection Remove the sampling drone Serial number; S234, the edge server repeats S232 to S233 until the sampling drone set is an empty set, and the set of drones that communicate directly is obtained and indirectly communicate with drone collections ; wherein the drone set The first sampling drone is included in the Includes at least one second sampling drone.

7. The heterogeneous sensing federated learning method through drone sampling and D2D communication according to claim 6 is characterized in that: The membership condition is the sampling drone The number of peer-to-peer communication drones in other sampling drones collection The time delay condition is the sampling drone Local training latency and sampling drones To the sampling drone The sum of the transmission delays of the sampling drone is no greater than The total delay.

8. The heterogeneous sensing federated learning method through drone sampling and D2D communication according to claim 7 is characterized in that: The sampling drone The total delay is expressed as: ； ； ； In the formula, In the current training round , sampling drone The total delay of In the current training round , sampling drone transmission delay; For sampling drones Local training latency; is the size of the model parameters; In the current training round , sampling drone The transmission rate; In the current training round , sampling drone a bandwidth allocated by the edge server; , is the total upstream bandwidth, In the current training round , sampling drone The transmission power; In the current training round , sampling drone Channel gain to edge servers; is the power spectral density of Gaussian noise.

9. The heterogeneous sensing federated learning method through drone sampling and D2D communication according to claim 1, characterized in that: The total delay used to update all models in the current training round is expressed as ,in, In the current training round , sampling drone transmission delay; The value of is 1; For sampling drones Local training latency; In the current training round , sampling drone The total delay.

10. The heterogeneous sensing federated learning method through drone sampling and D2D communication according to claim 1, characterized in that: The model parameters include gradients obtained by training the global model using a gradient descent method.

Images