Efficiency Optimization Method for Wireless Network Communication Based on Federated Learning

Abstract

This invention provides a method for optimizing wireless network communication efficiency in federated learning. It models the wireless network communication efficiency optimization problem in federated learning using the amount of training data per unit of wireless resources. The problem is simplified to a wireless resource allocation subproblem using alternating iterations. The optimal solution is then obtained through transformation and the Lagrange multiplier method. The spectral bandwidth of nodes used to ensure the optimal solution satisfies the constraints of the wireless network communication efficiency optimization problem is calculated, resulting in a surrogate selection subproblem. Parameters for determining whether a node participates in training are obtained. Based on the optimization results, the transmit power, spectral bandwidth, and nodes participating in federated learning are set. This method achieves an optimal trade-off between training accuracy and communication resource consumption, effectively improving network communication efficiency. It has low complexity, significantly increases solution speed, and has strong engineering applicability.

Description

Technical Field

[0001] This invention relates to a method for optimizing wireless network communication efficiency for federated learning, belonging to the field of wireless network communication technology. Background Technology

[0002] Wireless networks can significantly improve network performance by deploying artificial intelligence technologies such as machine learning. However, traditional centralized machine learning can lead to problems such as data security and excessive data transmission. Federated learning technology can effectively solve these problems. As a distributed learning method, federated learning only transmits local training parameters during training, resulting in a smaller data transmission volume. Furthermore, since federated learning does not transmit local data, data security is also protected. Based on these advantages, federated learning has become a common machine learning solution in wireless networks.

[0003] However, deploying federated learning on wireless networks requires addressing communication efficiency issues. On one hand, due to the fading effect of wireless channels, federated learning may experience interruptions during the transmission of training coefficients, affecting training accuracy. On the other hand, due to the limited nature of wireless communication resources, if the federated learning process consumes a large amount of communication resources, its practicality will be significantly reduced.

[0004] Therefore, effectively addressing the communication efficiency issue is a prerequisite for deploying federated learning on wireless networks. Currently, research in this field has made progress:

[0005] For example, the literature [Resource-efficient hierarchical collaborative federated learning method under heterogeneous Internet of Things, Journal of Electronics and Information Technology, 45(8), 2023] studied the impact of the heterogeneity of Internet of Things devices on federated learning and proposed an algorithm to eliminate the impact. However, this method does not involve mathematical modeling of communication efficiency, cannot solve the trade-off between communication resource consumption and training accuracy, and the solution depends on computing power, resulting in high complexity.

[0006] For example, the paper [Federated Learning Over Wireless IoT Networks With Optimized Communication and Resources, IEEE Internet of Things Journal, 9(17), 2022] studies the wireless resource optimization problem of federated learning and proposes a joint resource allocation algorithm. However, this method takes the number of nodes participating in federated learning as the maximization objective and does not consider the power and spectrum resource consumption of the nodes. Therefore, it cannot effectively solve the communication resource consumption problem of federated learning in wireless networks.

[0007] The above-mentioned issues should be considered and addressed in the process of optimizing wireless network communication efficiency for federated learning. Summary of the Invention

[0008] The purpose of this invention is to provide a wireless network communication efficiency optimization method for federated learning to solve the problems of excessive communication resource consumption, high complexity, and the need to improve the communication efficiency of federated learning in wireless networks in the existing technology.

[0009] The technical solution of this invention is:

[0010] A method for optimizing wireless network communication efficiency for federated learning includes the following steps:

[0011] S1. Model the wireless network communication efficiency optimization problem in federated learning using the amount of training data per unit of wireless resources.

[0012] S2. By using alternating iteration, the wireless network communication efficiency optimization problem is simplified into a wireless resource allocation subproblem. Then, the optimal solution is obtained through transformation and the Lagrange multiplier method. And calculate the solution used to ensure the optimal solution. The spectral bandwidth of a node that satisfies the constraints of the wireless network communication efficiency optimization problem. The optimal solution and By simplifying the problem into the wireless network communication efficiency optimization problem, we obtain the agent selection subproblem.

[0013] S3. For the agent selection subproblem, it is approximated as a knapsack problem, and an approximate solution is obtained using cost-effectiveness search. This yields the parameters that determine whether a node participates in training.

[0014] S4. Based on the optimization results, including the optimal solution to the wireless resource allocation subproblem obtained in step S2. With regard to the value of the spectrum bandwidth And the parameters obtained in step S3 for determining whether a node participates in training. Configure the node's transmit power, spectrum bandwidth, and the nodes participating in federated learning.

[0015] Furthermore, in step S1, the efficiency optimization problem of wireless network communication in federated learning is modeled using the amount of training data per unit of wireless resources. Specifically,

[0016] S11, The amount of training data U per unit of wireless resources is:

[0017]

[0018] Where i represents the wireless node index, M represents the set of wireless nodes, and t represents the training index. S represents the parameter used to determine whether node i participates in the training process during the t-th training iteration. i This represents the amount of training sample data for node i. Let represent the spectral bandwidth of node i in the t-th training iteration. Let represent the transmit power of node i in the t-th training iteration;

[0019] S12. Modeling the efficiency optimization problem of wireless network communication in federated learning:

[0020] objective function

[0021] Constraints

[0022]

[0023]

[0024]

[0025] in, This represents the floor function, E represents the maximum tolerable uplink delay, and B represents the floor function. T P represents the total bandwidth of the network spectrum. max Represents the maximum transmit power of a node; the training parameter backhaul time of node i in the t-th training iteration. Where D represents the amount of training parameter data; and the transmission rate of node i in the t-th training iteration is... in, Let σ represent the wireless channel gain from node i to the base station during the t-th training iteration. 2 Indicates background noise power;

[0026] Furthermore, in step S2, the wireless network communication efficiency optimization problem is simplified into a wireless resource allocation subproblem using alternating iterations, and then the optimal solution is obtained through transformation and the Lagrange multiplier method. The optimal solution and By simplifying the problem into the wireless network communication efficiency optimization problem, we obtain the agent selection subproblem, specifically:

[0027] S21, Fixed Combining equations (1-B) and (2), we obtain

[0028]

[0029] S22, combine formula (3) and Substituting into formula (1), the wireless network communication efficiency optimization problem is simplified into a wireless resource allocation subproblem as follows:

[0030] objective function

[0031] Constraints

[0032]

[0033] S23. Solve the wireless resource allocation subproblem using the Lagrange multiplier method, i.e., formula (4), to obtain the optimal solution to the wireless resource allocation subproblem. Implicit expressions;

[0034] S24. Based on the optimal solution of the wireless resource allocation subproblem The implicit expression is used to iterate and obtain the optimal solution to the wireless resource allocation subproblem.

[0035] S25, Order Substituting into formula (3) yields the result. Indicates to ensure Satisfying constraint (1-B) Values;

[0036] S26, will and Substituting the original problem (1) into the simplified problem, we obtain the following proxy selection subproblem:

[0037] objective function

[0038] Constraints

[0039]

[0040] Where M represents the set of wireless nodes, B T This indicates the total bandwidth of the network spectrum.

[0041] Furthermore, in step S23, the Lagrange multiplier method is used to solve the wireless resource allocation subproblem, i.e., formula (4), to obtain the optimal solution to the wireless resource allocation subproblem. The implicit expression is, specifically,

[0042] S231. Expand the Lagrangian function L of the wireless resource allocation subproblem, i.e., formula (4), as follows:

[0043]

[0044] Where η is the Lagrange multiplier corresponding to constraint (4-A), u i and λ iThese are the Lagrange multipliers corresponding to the constraint (1-D);

[0045] S232. Based on the Kuntak conditions, establish the following system of equations:

[0046]

[0047] η≥0,u i ≥0,λ i ≥0

[0048]

[0049]

[0050]

[0051] S233. Solve the system of equations from step S232 to obtain the optimal solution to the wireless resource allocation subproblem. The implicit expression is as follows:

[0052]

[0053] Furthermore, in step S24, based on the optimal solution of the wireless resource allocation subproblem... The implicit expression is used to iterate and obtain the optimal solution to the wireless resource allocation subproblem. Specifically,

[0054] S241, Describe the optimal solution to the wireless resource allocation subproblem. The value of the kth iteration is initialized with the iteration number k = 1, and the maximum number of iterations K is set. max ,

[0055] S242, Test for k > K max Does it satisfy the condition? If so, output the result. The iteration stops; if the condition is not met, then let... Proceed to the next step, S243;

[0056] S243. Let k = k + 1, then return to step S242.

[0057] Furthermore, in step S3, the agent selection subproblem is approximated as a knapsack problem, and an approximate solution is obtained using cost-effectiveness search. This yields the parameters that determine whether a node participates in training. Specifically,

[0058] S31. Initialize the parameters for determining whether node i participates in the training during the t-th training iteration. Weight parameter W = 0, set Φ = M, iteration counter k = 1;

[0059] S32, Test W≤B T If both k ≤ |M| and k ≤ |M| are true, then search for index j in set Φ such that... Proceed to the next step S33; if not both conditions are met, the iteration terminates, and the parameters determining whether node i participates in training during the t-th training iteration are output.

[0060] S33, Order k = k + 1, Φ = Φ - {j}, return to step S32.

[0061] Further, in step S4, the transmit power, spectrum bandwidth, and participating nodes of the nodes are set, specifically as follows:

[0062] S41. Let set Φ be empty. For each node i∈M, determine whether node i participates in training based on the parameters obtained in step S3 during the t-th training iteration. test If the condition is true, let Φ = Φ + {i}, establish a link between node i and the base station, and include node i in the federated learning process; if the condition is false, exclude node i from the federated learning process.

[0063] S42. For each element i in set Φ, based on the optimal solution of the wireless resource allocation subproblem obtained in step S24... Set the transmit power of node i to According to step S25 Set the spectrum bandwidth of node i to

[0064] The beneficial effects of this invention are as follows: This method for optimizing wireless network communication efficiency in federated learning uses the amount of training data per unit of wireless resources to mathematically model the efficiency of wireless network communication in federated learning, thereby achieving an optimal trade-off between training accuracy and communication resource consumption, effectively improving network communication efficiency; this invention uses interruption constraints in the model to transform and simplify the model, simplifying the model while ensuring that the solutions to subproblems satisfy the interruption constraints; this invention uses the Lagrange multiplier method to solve subproblems, resulting in a low-complexity iterative process, greatly improving the solution speed, and demonstrating strong engineering applicability. Attached Figure Description

[0065] Figure 1 This is a flowchart illustrating the wireless network communication efficiency optimization method for federated learning according to an embodiment of the present invention.

[0066] Figure 2 This is a schematic diagram of the federated learning parameter training and transmission model in the wireless network of this embodiment.

[0067] Figure 3 This is a schematic diagram comparing the amount of training data per unit of wireless resources in the simulation experiment of the wireless network communication efficiency optimization method for federated learning in this embodiment and the benchmark method. Detailed Implementation

[0068] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0069] Example

[0070] A method for optimizing wireless network communication efficiency for federated learning, such as Figure 1 This includes the following steps:

[0071] S1. Model the wireless network communication efficiency optimization problem in federated learning using the amount of training data per unit of wireless resources.

[0072] Figure 2 This is a schematic diagram of the federated learning parameter training and transmission model in the wireless network of this embodiment. Figure 2 In this example, the wireless network consists of one base station and M wireless nodes. Each wireless node has the capability to collect local samples and train a federated learning model, denoted by a set M. A single node in the set is denoted by i, and the amount of training sample data for node i is denoted by S. i The training of federated learning models is periodic, with each training iteration involving multiple rounds of parameter exchange, denoted by t. The base station and parameter server are directly connected. In the t-th round of parameter exchange, the parameter server first sends the global model parameters w to the wireless node via the base station's downlink. t Node i will w t The collected samples are then used to train the local model, yielding the local model parameters. Then via the node uplink The parameter server sends the parameters to the base station. The parameter server aggregates all local model parameters to obtain the global model parameters w for round t+1. t+1 .use This indicates whether node i participates in the training process during the t-th training iteration. This indicates that node i participates in the training. This indicates that node i does not participate in training. Let represent the spectral bandwidth of node i in the t-th training iteration. Let represent the transmit power of node i in the t-th training iteration.

[0073] Thus, the amount of training data U per unit of wireless resources is:

[0074]

[0075] Where i represents the wireless node index, M represents the set of wireless nodes, and t represents the training index. S represents the parameter used to determine whether node i participates in the training process during the t-th training iteration. i This represents the amount of training sample data for node i. Let represent the spectral bandwidth of node i in the t-th training iteration. Let represent the transmit power of node i in the t-th training iteration;

[0076] In the above formula, the numerator represents the total amount of training data incorporated into federated learning, and the denominator represents the total amount of wireless communication resources consumed. The advantage of using this metric is that it reflects the amount of training data that can be obtained per unit of wireless resources consumed, and can intuitively characterize communication efficiency.

[0077] Modeling the efficiency optimization problem of wireless network communication in federated learning:

[0078] objective function

[0079] Constraints

[0080]

[0081]

[0082]

[0083] in, This represents the floor function, E represents the maximum tolerable uplink delay, and B represents the floor function. T P represents the total bandwidth of the network spectrum. max Represents the maximum transmit power of a node; the training parameter backhaul time of node i in the t-th training iteration. Where D represents the amount of training parameter data; and the transmission rate of node i in the t-th training iteration is... in, Let σ represent the wireless channel gain from node i to the base station during the t-th training iteration. 2 Indicates background noise power;

[0084] The objective function formula (1) represents the amount of training data that can be obtained per unit of wireless resources consumed. It is an indicator reflecting communication efficiency and is expressed as the ratio of the total amount of training data included in federated learning to the total amount of wireless communication resources consumed. The larger this indicator is, the more training data can be obtained per unit of wireless resources consumed, i.e., the higher the communication efficiency. Constraint (1-A) indicates that node i can only have two states: participating in training or not participating in training. Constraint (1-B) indicates that node i can only participate in training if its uplink is not interrupted. Constraint (1-C) indicates that the sum of the spectrum bandwidth consumed by all participating nodes cannot exceed the total network spectrum bandwidth. Constraint (1-D) indicates that the transmit power of node i is non-negative and cannot exceed the maximum transmit power.

[0085] S2. By using alternating iteration, the wireless network communication efficiency optimization problem is simplified into a wireless resource allocation subproblem. Then, the optimal solution is obtained through transformation and the Lagrange multiplier method. And calculate the solution used to ensure the optimal solution. The spectral bandwidth of a node that satisfies the constraints of the wireless network communication efficiency optimization problem. The optimal solution and By simplifying the problem into the wireless network communication efficiency optimization problem, we obtain the agent selection subproblem.

[0086] S21, Fixed Combining equations (1-B) and (2), we obtain

[0087]

[0088] The advantage of the above simultaneous equations is that the constraint condition (1-B) is embedded in formula (3) by derivation, so that the solution that satisfies formula (3) naturally satisfies the constraint condition (1-B).

[0089] S22, combine formula (3) and Substituting into formula (1), the wireless network communication efficiency optimization problem is simplified into a wireless resource allocation subproblem as follows:

[0090] objective function

[0091] Constraints

[0092]

[0093] The advantage of the above simplification is that it eliminates the optimization variables using formula (3). This greatly reduces the complexity of the model.

[0094] S23. Solve the wireless resource allocation subproblem using the Lagrange multiplier method, i.e., formula (4), to obtain the optimal solution to the wireless resource allocation subproblem. Implicit expressions;

[0095] S231. Expand the Lagrangian function L of the wireless resource allocation subproblem, i.e., formula (4), as follows:

[0096]

[0097] Where η is the Lagrange multiplier corresponding to constraint (4-A), u i and λ i These are the Lagrange multipliers corresponding to the constraint (1-D);

[0098] S232. Based on the Kuntak conditions, establish the following system of equations:

[0099]

[0100] η≥0,u i ≥0,λ i ≥0

[0101]

[0102]

[0103]

[0104] S233. Solve the system of equations from step S232 to obtain the optimal solution to the wireless resource allocation subproblem. The implicit expression is as follows:

[0105]

[0106] The advantage of the above expression is that it is derived from the Kuhn-Tak condition, which guarantees the optimality of the solution.

[0107] S24. Based on the optimal solution of the wireless resource allocation subproblem The implicit expression is used to iterate and obtain the optimal solution to the wireless resource allocation subproblem.

[0108] S241, Describe the optimal solution to the wireless resource allocation subproblem. The value of the kth iteration is initialized with the iteration number k = 1, and the maximum number of iterations K is set. max =1000,

[0109] S242, Test for k > K max Does it satisfy the condition? If so, output the result. The iteration stops; if the condition is not met, then let... Proceed to the next step, S243;

[0110] S243. Let k = k + 1, then return to step S242.

[0111] The advantage of using the above iteration is that it can obtain high-quality solutions with low complexity.

[0112] S25, Order Substituting into formula (3) yields the result. Indicates to ensure Satisfying constraint (1-B) Values;

[0113] S26, will and Substituting the original problem (1) into the simplified problem, we obtain the following proxy selection subproblem:

[0114] objective function

[0115] Constraints

[0116]

[0117] Where M represents the set of wireless nodes, B T This indicates the total bandwidth of the network spectrum.

[0118] S3. For the agent selection subproblem, it is approximated as a knapsack problem, and an approximate solution is obtained using cost-effectiveness search. This yields the parameters that determine whether a node participates in training.

[0119] S31. Initialize the parameters for determining whether node i participates in the training during the t-th training iteration. Weight parameter W = 0, set Φ = M, iteration counter k = 1;

[0120] S32, Test W≤B T If both k ≤ |M| and k ≤ |M| are true, then search for index j in set Φ such that... Proceed to the next step S33; if not both conditions are met, the iteration terminates, and the parameters determining whether node i participates in training during the t-th training iteration are output.

[0121] S33, Order k = k + 1, Φ = Φ - {j}, return to step S32.

[0122] In step S3, the advantage of using cost-effectiveness search is that it can obtain an approximate optimal solution to the surrogate selection subproblem formula (6) with lower complexity.

[0123] S4. Based on the optimization results, including the optimal solution to the wireless resource allocation subproblem obtained in step S2. With regard to the value of the spectrum bandwidth And the parameters obtained in step S3 for determining whether a node participates in training. Configure the node's transmit power, spectrum bandwidth, and the nodes participating in federated learning.

[0124] S41. Let set Φ be empty. For each node i∈M, determine whether node i participates in training based on the parameters obtained in step S3 during the t-th training iteration. test If the condition is true, let Φ = Φ + {i}, establish a link between node i and the base station, and include node i in the federated learning process; if the condition is false, exclude node i from the federated learning process.

[0125] S42. For each element i in set Φ, based on the optimal solution of the wireless resource allocation subproblem obtained in step S24... Set the transmit power of node i to According to step S25 Set the spectrum bandwidth of node i to

[0126] This method for optimizing wireless network communication efficiency in federated learning uses the amount of training data per unit of wireless resources to mathematically model the efficiency of wireless network communication in federated learning, thereby achieving an optimal trade-off between training accuracy and communication resource consumption, effectively improving network communication efficiency. The invention transforms and simplifies the model using interruption constraints, simplifying the model while ensuring that the solutions to subproblems satisfy the interruption constraints. Furthermore, the invention employs the Lagrange multiplier method to solve subproblems, resulting in a low-complexity iterative process that significantly improves solution speed and demonstrates strong engineering applicability.

[0127] This method for optimizing wireless network communication efficiency in federated learning addresses issues such as mathematical modeling of communication efficiency, failure to consider communication resource consumption in the optimization objective, excessive communication resource consumption, and high complexity. It models wireless network communication efficiency in federated learning using the amount of training data per unit of wireless resources, and then achieves efficient solution of the model through alternating iteration, problem transformation, Lagrange multiplier method, and cost-effectiveness search. This method can effectively improve the communication efficiency of federated learning in wireless networks and is suitable for wireless edge intelligent networks.

[0128] The effects of the present invention will be further explained below with reference to simulation data.

[0129] 1. Experimental conditions

[0130] The effectiveness of the method of the present invention is further illustrated below by comparing it with the training data volume under unit wireless resources using existing methods. The existing method is the scheme in the literature [Federated Learning Over Wireless IoT NetworksWith Optimized Communication and Resources, IEEE Internet of Things Journal, 9(17), 2022], whose optimization objective is to maximize the number of nodes participating in training.

[0131] The simulation experiment models a wireless network with a coverage radius of 1 kilometer, deploying one base station and 20 wireless nodes. The base station is directly connected to a parameter server and located at the center of the coverage area. The wireless nodes are evenly distributed within the coverage area. The channel propagation model is 10³ + 27 log₂(t / t). 10 (d), where d represents the distance between the two points. Total network spectrum bandwidth B T =10MHz, background noise power σ 2 = -109dBm, maximum node transmit power P max =15dBm. The global model used in the simulation is a convolutional neural network, consisting of two convolutional layers, one fully connected layer, and one output layer. The training dataset is from MNIST, with S samples per node. i The training consists of 2000-3000 images. The number of training sessions is 1000.

[0132] 2. Implementation process

[0133] Based on the experimental parameters above, the set of wireless nodes in model (1) is instantiated as M = {1, 2, ..., 20}. In each simulation, the distance d between the base station and the wireless nodes is calculated based on their coordinates, and then 103 + 27log 10 (d) Channel gain Instantiate the data. For example, if the coordinates of the base station are (500, 500) and the coordinates of wireless node 3 are (20, 70), then the distance between them is... Substitute it into 103+27log 10 (d) Calculated Instantiated and σ 2 Substituting -109dBm into (5), we can get The instantiation iteration formula is as follows. Similarly, the distance from each wireless node to the base station is calculated based on the coordinates, and then all... The instantiation iteration formula is then used. Then, using the iteration in step S24, the instantiation is obtained. and Then instantiated and Substituting the cost-effectiveness search from step S3, calculate the instantiation. Find the instantiated Finally, based on the calculated instantiation Set the connection relationship between node i and the base station in the simulation, based on the instantiated... and Configure the transmit power and spectrum bandwidth of node i.

[0134] 3. Analysis of Experimental Results

[0135] Figure 3 A comparison chart showing the amount of training data consumed per unit of wireless resources between the method of this invention and existing methods under the parameter settings of the experimental examples; from Figure 3 As can be seen, the method of the present invention can effectively increase the amount of training data and thus improve communication efficiency while consuming the same amount of wireless resources, thereby proving the effectiveness of the modeling and solving steps of the present invention in improving communication efficiency.

[0136] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing wireless network communication efficiency for federated learning, characterized in that: Includes the following steps, S1. Model the wireless network communication efficiency optimization problem in federated learning using the amount of training data per unit of wireless resources. S11, Training data volume per unit of wireless resources for: ， in, Indicates the wireless node number. Represents a set of wireless nodes. Indicates the training sequence number. Indicates the first Nodes are determined during this training. Parameters that are included in training Represents a node The amount of training sample data, Indicates the first Nodes in this training session Spectrum bandwidth, Indicates the first Nodes in this training session The transmission power; S12. Modeling the efficiency optimization problem of wireless network communication in federated learning: objective function (1) Constraints (1-A) （1-B） （1-C） （1-D） in, This represents the floor function. This indicates the maximum tolerable uplink latency. This represents the total bandwidth of the network spectrum. Indicates the node's maximum transmit power; the first Nodes in this training session Training parameter backhaul time : (2), where, Indicates the amount of training parameter data; the first Nodes in this training session transmission rate : ,in, Indicates the first Nodes in this training session The wireless channel gain to the base station Indicates background noise power; S2. By using alternating iteration, the wireless network communication efficiency optimization problem is simplified into a wireless resource allocation subproblem. Then, the optimal solution is obtained through transformation and the Lagrange multiplier method. And calculate the solution to ensure optimality. The spectral bandwidth of a node that satisfies the constraints of the wireless network communication efficiency optimization problem. , the optimal solution and By simplifying the problem into the wireless network communication efficiency optimization problem, we obtain the agent selection subproblem. S21, Fixed Combining this with formulas (1-B) and (2), we obtain （3） S22, combine formula (3) and Substituting into formula (1), the wireless network communication efficiency optimization problem is simplified into a wireless resource allocation subproblem as follows: objective function (4) Constraints (4-A) (1-D) S23. Solve the wireless resource allocation subproblem using the Lagrange multiplier method, i.e., formula (4), to obtain the optimal solution to the wireless resource allocation subproblem. Implicit expressions; S24. Based on the optimal solution of the wireless resource allocation subproblem The implicit expression is used to iterate and obtain the optimal solution to the wireless resource allocation subproblem. ; S241, Describe the optimal solution to the wireless resource allocation subproblem. The value of the kth iteration, initialized with the iteration number. Set the maximum number of iterations. , ; S242, Inspection Does it satisfy the condition? If so, output the result. If the condition is not met, then let the iteration stop; otherwise, let Proceed to the next step, S243; S243, Order Then return to step S242; S25, Order Substituting into formula (3) yields the result. , Indicates to ensure Satisfying constraint (1-B) Values; S26, will and Substituting the original problem (1) into the simplified problem, we obtain the following proxy selection subproblem: objective function (6) Constraints in, Represents a set of wireless nodes. Indicates the total bandwidth of the network spectrum; S3. For the agent selection subproblem, it is approximated as a knapsack problem, and an approximate solution is obtained using cost-effectiveness search. This yields the parameters that determine whether a node participates in training. ; S4. Based on the optimization results, including the optimal solution to the wireless resource allocation subproblem obtained in step S2. With regard to the value of the spectrum bandwidth And the parameters obtained in step S3 for determining whether a node participates in training. Configure the node's transmit power, spectrum bandwidth, and the nodes participating in federated learning.

2. The wireless network communication efficiency optimization method for federated learning as described in claim 1, characterized in that: In step S23, the Lagrange multiplier method is used to solve the wireless resource allocation subproblem, i.e., formula (4), to obtain the optimal solution to the wireless resource allocation subproblem. The implicit expression is, specifically, S231, the Lagrangian function of the wireless resource allocation subproblem, i.e., formula (4). Expand as ， in, These are the Lagrange multipliers corresponding to constraint (4-A). and These are the Lagrange multipliers corresponding to the constraint (1-D); S232. Based on the Kuntak conditions, establish the following system of equations: ， S233. Solve the system of equations from step S232 to obtain the optimal solution to the wireless resource allocation subproblem. The implicit expression is as follows: （5）。 3. The wireless network communication efficiency optimization method for federated learning as described in claim 1, characterized in that: In step S3, the agent selection subproblem is approximated as a knapsack problem, and an approximate solution is obtained using cost-effectiveness search. This yields the parameters that determine whether a node participates in training. Specifically, S31, Initialize the first Nodes are determined during this training. Parameters whether to participate in training : Weight parameters ,gather Iteration counter ; S32, Inspection and Do they both hold true at the same time? If they both hold true, then in the set... Search sequence number , making Proceed to the next step S33; if not both conditions are met, the iteration terminates and the output is the first step. Nodes are determined during this training. Parameters whether to participate in training ; S33, Order , , , Return to step S32.

4. The wireless network communication efficiency optimization method for federated learning as described in claim 1, characterized in that: In step S4, the transmit power, spectrum bandwidth, and participating nodes of the nodes are set, specifically as follows: S41, Let set For an empty set, for each node According to the first step S3, Nodes are determined during this training. Parameters whether to participate in training ,test Is it true? If it is true, then... Establish nodes The link with the base station will connect the nodes. Incorporate into the federal learning process; If not true, then the node Excluded from the federal learning process; S42, For sets Each element in Based on the optimal solution of the wireless resource allocation subproblem obtained in step S24 Set up nodes The transmission power is According to step S25 Set up nodes The spectral bandwidth is .

Images