Federal learning training method and device based on location awareness

Abstract

The invention relates to the technical field of intelligent network connection automobiles and edge computing, in particular to a federated learning training method and device based on position awareness, and can solve the problem that an existing federated learning method cannot well adapt to a VEFL application scene to a certain extent. The method comprises the following steps: acquiring a historical track and a resource condition of a client, wherein the resource condition comprises the computing capability and the communication capability of the client; in each round of training, track prediction is carried out according to the historical tracks, a track prediction result of each client is obtained, and response delay of each client is obtained through estimation according to the resource conditions and the track prediction results; selecting a corresponding sampling client set and a test client set for each cluster based on the response delay of each client and the trajectory prediction result; and allocating corresponding cluster models to the sampling client set and the test client set to carry out local training and test tasks, and circulating the above process until a training ending condition is met.

Description

Technical Field

[0001] The present application relates to the field of intelligent connected vehicles and edge computing technology, and more specifically, to a location-aware federated learning training method and device. Background Art

[0002] Federated Learning (FL) is an emerging machine learning paradigm that has great potential in data privacy protection and resource utilization. Through client-side edge federated learning (VEFL), smart clients can train models locally and only share the updated results of the model without transmitting sensitive original data. This not only protects user privacy, but also reduces the latency and overhead of data transmission, and improves the accuracy and real-time performance of the model.

[0003] In the application scenario of VEFL, the data collected by the client from different geographical areas have different statistical characteristics, and the high mobility of the client further aggravates the statistical heterogeneity, that is, the distribution of data collected by the client may change with the change of its location. Therefore, the existing scheme proposes to bind the FL model to the geographical area and train the same FL model in the area where the data follows a highly similar distribution to alleviate the adverse effects of statistical heterogeneity.

[0004] However, the client is both a data collector and a model user in VEFL. Due to the change of the client's location, if the cluster model is assigned to the client only according to the current location, the data collected by the client in the past may not match the data distribution suitable for the model, resulting in deviations in model training; if the corresponding cluster model is assigned to the client only according to the data that the client has already collected, the model may not be suitable for the data currently and in the future collected by the client, and reasonable test results and ideal application accuracy cannot be obtained, which affects the real-time performance of the model; and different clients have different computing and communication capabilities, that is, system heterogeneity, and resource-poor clients can easily become "laggards", greatly reducing training efficiency; making the existing federated learning methods not well adapted to VEFL application scenarios. Summary of the invention

[0005] In order to solve the problem that the existing federated learning methods cannot adapt well to VEFL application scenarios, the present application provides a location-aware federated learning training method and device to adapt to mobile scenarios and improve training efficiency.

[0006] The embodiment of the present application is implemented as follows:

[0007] In a first aspect, the present application provides a location-aware federated learning training method, comprising:

[0008] Acquire the historical track and resource status of the client, wherein the resource status includes the computing power and communication power of the client;

[0009] In each round of training, trajectory prediction is performed based on the historical trajectory to obtain a trajectory prediction result of each client, and a response delay of each client is estimated based on the resource situation and the trajectory prediction result;

[0010] Based on the response delay of each client and the trajectory prediction result, select a corresponding sampling client set and a test client set for each cluster;

[0011] Corresponding cluster models are allocated to the sampling client set and the test client set to perform local training and testing tasks, and the above process is repeated until the end condition of the training is met.

[0012] In a possible implementation, in each round of training, trajectory prediction is performed according to the historical trajectory to obtain a trajectory prediction result of each client, and a response delay of each client is estimated according to the resource situation and the trajectory prediction result, further comprising:

[0013] Perform trajectory prediction based on the historical trajectory to obtain a trajectory prediction result;

[0014] According to the historical trajectory and trajectory prediction results, estimate the size of the training data set of the client;

[0015] Based on the training data set size, computing power, model size and communication capacity, the response delay of each client is estimated.

[0016] In a possible implementation, the selecting a corresponding sampling client set and a test client set for each cluster based on the response delay and the trajectory prediction result of each client further includes:

[0017] The entire training area is divided into multiple non-overlapping sub-areas, each of which is bound to a dedicated cluster model;

[0018] Determine a training client set for each cluster according to the historical trajectory of the client and the trajectory prediction result;

[0019] According to the response delay and sampling ratio of the client, a sampling client set of each cluster is determined in the training client set.

[0020] In a possible implementation, the training method further includes a cluster model allocation strategy, and the cluster model allocation strategy includes:

[0021] Determining a sub-area where the client will stay in the future according to the trajectory prediction result of the client;

[0022] The client is assigned to a test client set of a cluster in the sub-region where the client will stay in the future.

[0023] In a possible implementation, the cluster model allocation strategy includes:

[0024] According to the historical track of the client, it is determined whether the client has just entered the training area. If the client has just entered the training area, it will not immediately participate in the training task.

[0025] In a possible implementation, the cluster model allocation strategy includes:

[0026] According to the trajectory prediction result of the client, it is determined whether the client is about to leave the training area. If the client is about to leave the training area, it will not participate in the test task.

[0027] In a possible implementation, the cluster model allocation strategy includes:

[0028] When each client in the sampled client set receives a corresponding cluster model training task, it will filter the data collected within the coverage of the cluster as a training data set;

[0029] When each client in the test client set receives a corresponding cluster model test task, the data recently collected by the client is screened and used as a test data set.

[0030] In a possible implementation, the assigning of corresponding cluster models to the sampling client set and the test client set to perform local training and testing tasks, and looping the above process until the end condition of the training is met, further includes:

[0031] In each round of training, the training client set is divided into multiple layers according to the response delay of each client by using a clustering algorithm, and the corresponding delays between clients in each layer are close;

[0032] After selecting a layer of the training client set, a group of clients is randomly selected according to the sampling ratio as a sampling client set to perform the training task.

[0033] In a possible implementation, determining the cluster currently participating in training of the client according to the historical trajectory of the client further includes:

[0034] Determine the time the client stays in each cluster according to the historical trajectory;

[0035] The cluster where the client stays the longest is determined as the cluster currently participating in the training of the client.

[0036] In a second aspect, the present application provides a location-aware federated learning training device, comprising:

[0037] Data collection module: used to obtain the historical track and resource status of the client, the resource status includes the computing power and communication capacity of the client;

[0038] Data processing module: used to perform trajectory prediction according to the historical trajectory in each round of training, obtain the trajectory prediction result of each client, and estimate the response delay of each client according to the resource situation and the trajectory prediction result;

[0039] Data partitioning module: used to select a corresponding sampling client set and a test client set for each cluster based on the response delay and the trajectory prediction result of each client;

[0040] Training and testing module: used to assign corresponding cluster models to the sampling client set and the test client set to perform local training and testing tasks, and loop the above process until the end condition of the training is met.

[0041] The technical solution provided by this application can at least achieve the following beneficial effects:

[0042] The location-aware federated learning training method and device provided in the present application utilize the predictability of client movement to predict the client's future location trajectory based on the client's historical trajectory, thereby determining the cluster model that best suits the distribution of each client data set and estimates the response delay of each client, alleviating the adverse effects of device heterogeneity and further improving the training efficiency. The dual identity of the client in VEFL is taken into account, and considering that the client is both a data collector and a model user, appropriate cluster model testing and application are assigned to the client based on the historical trajectory and trajectory prediction results to ensure the effectiveness of the test and the accuracy of the application. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0044] Figure 1 is a flowchart of a location-aware federated learning training method shown in an exemplary embodiment of the present application;

[0045] Figure 2 It is a schematic diagram of an implementation process of a location-aware federated learning training algorithm shown in an exemplary embodiment of the present application;

[0046] Figure 3 is a schematic diagram of federated learning training area and cluster division shown in an exemplary embodiment of the present application;

[0047] Figure 4 is a schematic diagram of the client position and sampling status of the first three rounds shown in an exemplary embodiment of the present application;

[0048] Figure 5 It is a structural diagram of a location-aware federated learning training device shown in an exemplary embodiment of the present application.

[0049] Reference numerals:

[0050] 1. Data acquisition module; 2. Data processing module; 3. Data division module; 4. Training and testing module. DETAILED DESCRIPTION

[0051] In order to make the purpose, implementation mode and advantages of the present application clearer, the exemplary implementation mode of the present application will be clearly and completely described below in conjunction with the drawings in the exemplary embodiments of the present application. Obviously, the described exemplary embodiments are only part of the embodiments of the present application, not all of the embodiments. It should be understood that the specific embodiments described here are only used to explain the present application and are not used to limit the present application.

[0052] It should be noted that the brief description of terms in this application is only for the convenience of understanding the embodiments described below, and is not intended to limit the embodiments of this application. Unless otherwise specified, these terms should be understood according to their common and usual meanings.

[0053] The terms "first", "second", "third", etc. in the specification and claims of this application and the above drawings are used to distinguish similar or similar objects or entities, and do not necessarily mean to limit a specific order or sequence, unless otherwise noted. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances.

[0054] The terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device comprising a list of components is not necessarily limited to all the components expressly listed but may include other components not expressly listed or inherent to such product or device.

[0055] To facilitate the technical solution of the application, some concepts involved in the application are first explained below.

[0056] Before explaining the location-aware federated learning training method provided in the embodiment of the present application, the application scenario and implementation environment of the embodiment of the present application are first introduced.

[0057] Federated Learning (FL) is an emerging machine learning paradigm that has great potential in data privacy protection and resource utilization. Through client-side edge federated learning (VEFL), smart clients can train models locally and only share the updated results of the model without transmitting sensitive original data. This not only protects user privacy, but also reduces the latency and overhead of data transmission, and improves the accuracy and real-time performance of the model.

[0058] In the application scenario of VEFL, the data collected by the client from different geographical areas have different statistical characteristics, and the high mobility of the client further aggravates the statistical heterogeneity, that is, the distribution of data collected by the client may change with the change of its location. Therefore, the existing scheme proposes to bind the FL model to the geographical area and train the same FL model in the area where the data follows a highly similar distribution to alleviate the adverse effects of statistical heterogeneity.

[0059] However, in VEFL, the client is both a data collector and a model user. Due to changes in the client's location, if the client is assigned a cluster model only based on the current location, the data collected by the client in the past may not match the data distribution suitable for the model, resulting in deviations during model training; if the client is assigned a corresponding cluster model only based on the data it has collected, the model may not be suitable for the data currently and in the future collected by the client, and reasonable test results and ideal application accuracy cannot be obtained, affecting the real-time performance of the model; and different clients have different computing and communication capabilities, that is, system heterogeneity, and resource-poor clients can easily become "laggards", greatly reducing training efficiency; making the existing federated learning methods not well adapted to VEFL application scenarios.

[0060] Based on this, the present application provides a location-aware federated learning training method and device, which guides the training and application process according to the client's resource conditions and mobility prediction results to adapt to actual mobile scenarios and improve training efficiency.

[0061] Next, the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems will be described in detail through embodiments and in combination with the accompanying drawings. The embodiments may be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. Obviously, the described embodiments are part of the embodiments of the present application, not all of them.

[0062] Figure 1 It is a flowchart of a location-aware federated learning training method shown in an exemplary embodiment of the present application.

[0063] In an exemplary embodiment, Figure 1 As shown, a location-aware federated learning training method is provided. In this embodiment, the method may include the following steps:

[0064] Step 100: Obtain the historical track and resource status of the client, wherein the resource status includes the computing capability and communication capability of the client;

[0065] Step 200: In each round of training, trajectory prediction is performed according to the historical trajectory to obtain a trajectory prediction result of each client, and a response delay of each client is estimated according to the resource situation and the trajectory prediction result;

[0066] Step 300: Based on the response delay and the trajectory prediction result of each client, select a corresponding sampling client set and a test client set for each cluster;

[0067] Step 400: assigning corresponding cluster models to the sampling client set and the test client set to perform local training and testing tasks, and looping the above process until the end condition of the training is met.

[0068] It can be seen that some embodiments of the present application take into account the geographical dispersion of data and the mobility of clients, which further aggravate statistical heterogeneity. The predictability introduced by mobility is used to send appropriate cluster models to clients to perform training or testing tasks, thereby alleviating the adverse effects of statistical heterogeneity. In addition, a location-aware stratified sampling strategy is designed for the system heterogeneity between clients in the VEFL application scenario. A set of sampling clients with similar response delays is selected for each cluster according to the client's resource situation and location trajectory, so as to improve training efficiency.

[0069] In one possible implementation, considering that the client is both a data collector and a model user, and due to its high mobility, it can cross areas covered by different clusters in a short time, so the cluster model it participates in training may be different from the cluster model used in testing and application.

[0070] To this end, some embodiments of the present application utilize location prediction methods to fully exploit the predictability of movement.

[0071] Assuming that the entire training area is divided into K clusters, for each federated learning training round r, if the server Θ determines that the client has stayed more frequently in the coverage of the kth cluster based on the client's historical and predicted location trajectory, it will be assigned to the set of clients N participating in the training process of cluster k in the rth round. k (r) medium;

[0072] If it is most likely to move within the coverage of the kth cluster in the future, server Θ will classify it into the rth round of testing or apply the model w of cluster k k (r) Client Collection middle.

[0073] When the server Θ finds that the client i has just entered the entire FL training area, that is, i has not collected much data in the area, it will not let i participate in the training task immediately.

[0074] In addition, when the server Θ predicts that client i is about to leave the entire FL training area, i will not be allowed to participate in subsequent test tasks to avoid introducing errors.

[0075] On this basis, in order to further reduce the impact of statistical heterogeneity caused by mobility, when each client i∈N k (r) Receive the kth cluster model w k (r-1) When preparing for local training, it will filter the data collected within the coverage area of ​​the kth cluster as the training dataset.

[0076] and The client in should execute the cluster model w using the most recently collected data k The test task of (r) is more reasonable and more in line with the actual application scenario.

[0077] Figure 2 It is a schematic diagram of the implementation process of a location-aware federated learning training algorithm shown in an exemplary embodiment of the present application.

[0078] In one possible implementation, Figure 2 As shown, in the training process of each cluster model, some embodiments of the present application adopt a stratified sampling strategy.

[0079] That is, each round in N k (r) Select a group of similar response delays T i (r) Client S k (r) Performing actual training tasks to mitigate the adverse effects of system heterogeneity.

[0080] The specific implementation process of the stratified sampling strategy is as follows: In each round r, the server Θ first obtains the historical trajectory J of each client ii (t) and the current computing power C i (t) and communication capability θ i (t), and estimate the training dataset size D of client i based on the historical and predicted location trajectories i (t);

[0081] Then, according to the dataset size D i (t), computing power C i (t), model size α and communication capacity θ i (t) Estimate the delay T of each client i i (r),

[0082] Then use the clustering algorithm according to T i (r) N k (r) partitioned into K′ layers, where the latency differences between clients in the same layer are relatively small;

[0083] Finally, Θ selects one layer and randomly selects a group of clients S according to the sampling ratio σ k (r) Perform training tasks. The parameter K′ and how to select one of the K′ layers for random sampling can be adjusted according to different levels of device heterogeneity.

[0084] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the instructions, these steps are not necessarily executed in sequence according to the order of the instructions. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the above-mentioned embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the steps or stages in other steps.

[0085] Corresponding to the aforementioned embodiment of the location-awareness-based federated learning training method and adopting the same technical concept, the present application also provides an embodiment of the location-awareness-based federated learning training device.

[0086] Figure 5 It is a structural diagram of a location-aware federated learning training device shown in an exemplary embodiment of the present application.

[0087] In an exemplary embodiment, Figure 5 As shown, the location-aware federated learning training device includes:

[0088] Data collection module 1: used to obtain the historical trajectory and resource status of the client, the resource status includes the computing power and communication capacity of the client;

[0089] Data processing module 2: used to perform trajectory prediction according to the historical trajectory in each round of training to obtain the trajectory prediction result of each client, and estimate the response delay of each client according to the resource situation and the trajectory prediction result;

[0090] Data partitioning module 3: used to select a corresponding sampling client set and a test client set for each cluster based on the response delay and the trajectory prediction result of each client;

[0091] Training and testing module 4: used to assign corresponding cluster models to the sampling client set and the testing client set to perform local training and testing tasks, and to repeat the above process until the end condition of the training is met.

[0092] For the specific limitations of the location-aware federated learning training device, please refer to the limitations of the location-aware federated learning training method above, which will not be repeated here. Each module in the above-mentioned location-aware federated learning training device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0093] Working process:

[0094] In order to make the technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with an implementation method of the present application. Obviously, the described embodiments are embodiments of part of the present application, not all of the embodiments.

[0095] Figure 3 is a schematic diagram of the federated learning training area and cluster division shown in an exemplary embodiment of the present application, Figure 4 It is a schematic diagram of the client position and sampling status in the first three rounds shown in an exemplary embodiment of the present application.

[0096] like Figure 3As shown, it is assumed that the entire training area is divided into two parts, namely, the coverage range M1 of the first cluster and the coverage range M2 of the second cluster. At the beginning, there are 8 cars in the entire training area, numbered 1, 4, 5, 8, 3, 6, 7, and 9. In each round, the client first uploads its resource status and historical trajectory to the server. The server predicts the location of the client and estimates the response delay of each client. Some embodiments of the present application assume that the actual delay time of each car is as shown in Table 1. The car number also represents the sorting order of its delay time. The smaller the number, the lower the delay of the corresponding car, and it is assumed that the delay time of the car is unchanged in each round.

[0097] Table 1 Client delay time example

[0098] Client 1 2 3 4 5 6 7 8 9 Delay time 1 2 3 4 5 16 17 18 19

[0099] Next, some embodiments of the present application will introduce the specific implementation process of each round of cluster model allocation and stratified sampling strategy with a simple example.

[0100] like Figure 3 As shown in the figure, in the first round, the server knows from the historical trajectories that cars 1, 4, 6, and 8 have been staying in the coverage area M1 of the first cluster, while cars 3, 5, 7, and 9 have been staying in the coverage area M2 of the second cluster. Based on the position prediction results, the server knows that car 3 is about to enter M1 and car 4 is about to enter M2, thus determining N 1 (1) = {1,4,6,8}, N 2 (1) = {3,5,7,9}, Next, the server uses the K-Means clustering algorithm to cluster N 1 (1) and N 2 (1) are divided into 2 layers, wherein the layer with lower delay is {1,4} and {3,5}, and the layer with higher delay is {6,8} and {7,9}. In each subsequent round, some embodiments of the present application will alternately select the layer with higher delay or lower delay for sampling. Therefore, in the first round, some embodiments of the present application select the layer with lower delay for random sampling. Due to the small number of clients, the sampling ratio σ is set to 100%, that is, S 1 (1) = {1, 4}, S 2 (1) = {3, 5}. Afterwards, the server will initialize the cluster model w 1 (0) and the range of M1, w 2 The ranges of (0) and M2 are sent to S 1 (1) and S 2 (1), sampling client i∈S1 (1)∪S 2 (1) Filter the data collected within the coverage area of ​​the cluster as the training data set to train the corresponding cluster model. After the training, the new local model w i (1) Sent to the server, the server has a local model for each cluster k Aggregate and get the new cluster model w 1 (1) and w 2 (1) Then send them to and Perform the test and upload the results to the server after the test is completed.

[0101] like Figure 4 As shown in the figure, in the second round, the server knows from the historical trajectory that in the past period of time, cars No. 1, 3, 6, and 8 stayed in M1, cars No. 4, 5, 7, and 9 stayed in M2, and car No. 2 just entered M1 not long ago. According to the position prediction results, the server knows that car No. 1 is about to leave M1, thus determining N 1 (2) = {1, 3, 6, 8}, N 2 (2) = {4, 5, 7, 9}, Next, the server uses the K-Means clustering algorithm to cluster N 1 (2) and N 2 (2) They are divided into two layers, wherein the layer with lower delay is {1, 3} and {4, 5}, and the layer with higher delay is {6, 8} and {7, 9}. In the second round, some embodiments of the present application select the layer with higher delay for random sampling with σ=100%, that is, S 1 (2) = {6, 8}, S 2 (2) = {7, 9}. Afterwards, the server will model w 1 (1) and the range of M1, w 2 (1) and M2 ranges are sent to S 1 (2) and S 2 (2), sampling client i∈S 1 (2)∪S 2 (2) Use the filtered data as the training data set to train the corresponding cluster model. After the training, the new local model w i (2) Sent to the server, the server has a local model for each cluster k Aggregate and get the new cluster model w 1 (2) and w 2 (2) Then send them to and Perform the test and upload the results to the server after the test is completed.

[0102] like Figure 4 As shown in the figure, in the third round, the server knows from the historical trajectory that in the past period of time, cars No. 2, 3, 6, and 8 stayed in M1, cars No. 4, 5, 7, and 9 stayed in M2, and car No. 1 has left M1. According to the position prediction results, the eight cars No. 2 to 9 will still move in the current area in the future, so N is determined. 1 (3) = {2, 3, 6, 8}, N 2 (3) = {4, 5, 7, 9}, Next, the server also uses the clustering algorithm to cluster N 1 (3) and N 2 (3) are divided into 2 layers, wherein the layer with lower delay is {2, 3} and {4, 5}, and the layer with higher delay is {6, 8} and {7, 9}. In the third round, some embodiments of the present application select the layer with lower delay for random sampling with σ=100%, that is, S 1 (3) = {2, 3}, S 2 (3) = {4, 5}.

[0103] The subsequent training and testing steps are similar to the first two rounds. The above steps are repeated in each round until the training deadline is reached, the test results reach the expected accuracy, or the cluster model converges.

[0104] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0105] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A location-aware federated learning training method, characterized in that: include: Acquire the historical track and resource status of the client, wherein the resource status includes the computing power and communication power of the client; In each round of training, trajectory prediction is performed based on the historical trajectory to obtain a trajectory prediction result of each client, and a response delay of each client is estimated based on the resource situation and the trajectory prediction result; Based on the response delay of each client and the trajectory prediction result, select a corresponding sampling client set and a test client set for each cluster; Corresponding cluster models are allocated to the sampling client set and the test client set to perform local training and testing tasks, and the above process is repeated until the end condition of the training is met.

2. The location-aware federated learning training method according to claim 1, characterized in that: In each round of training, trajectory prediction is performed according to the historical trajectory to obtain a trajectory prediction result of each client, and the response delay of each client is estimated according to the resource situation and the trajectory prediction result, and further includes: Perform trajectory prediction based on the historical trajectory to obtain a trajectory prediction result; According to the historical trajectory and trajectory prediction results, estimate the size of the training data set of the client; Based on the training data set size, computing power, model size and communication capacity, the response delay of each client is estimated.

3. The location-aware federated learning training method according to claim 1, characterized in that: The selecting a corresponding sampling client set and a test client set for each cluster based on the response delay and the trajectory prediction result of each client further includes: The entire training area is divided into multiple non-overlapping sub-areas, each of which is bound to a dedicated cluster model; Determine a training client set for each cluster according to the historical trajectory of the client and the trajectory prediction result; According to the response delay and sampling ratio of the client, a sampling client set of each cluster is determined in the training client set.

4. The location-aware federated learning training method according to claim 3, characterized in that: The training method further includes a cluster model allocation strategy, wherein the cluster model allocation strategy includes: Determining a sub-area where the client will stay in the future according to the trajectory prediction result of the client; The client is assigned to a test client set of a cluster in the sub-region where the client will stay in the future.

5. The location-aware federated learning training method according to claim 4, characterized in that: The cluster model allocation strategy includes: According to the historical track of the client, it is determined whether the client has just entered the training area. If the client has just entered the training area, it will not immediately participate in the training task.

6. The location-aware federated learning training method according to claim 4, characterized in that: The cluster model allocation strategy includes: According to the trajectory prediction result of the client, it is determined whether the client is about to leave the training area. If the client is about to leave the training area, it will not participate in the test task.

7. The location-aware federated learning training method according to claim 4, characterized in that: The cluster model allocation strategy includes: When each client in the sampled client set receives a corresponding cluster model training task, it will filter the data collected within the coverage of the cluster as a training data set; When each client in the test client set receives a corresponding cluster model test task, the data recently collected by the client is screened and used as a test data set.

8. The location-aware federated learning training method according to claim 1 or 2, characterized in that: The assigning of corresponding cluster models to the sampling client set and the test client set to perform local training and testing tasks, and looping the above process until the end condition of the training is met, further includes: In each round of training, the training client set is divided into multiple layers according to the response delay of each client by using a clustering algorithm, and the corresponding delays between clients in each layer are close; After selecting a layer of the training client set, a group of clients is randomly selected according to the sampling ratio as a sampling client set to perform the training task.

9. The location-aware federated learning training method according to claim 1, characterized in that: The step of determining the cluster currently participating in the training of the client according to the historical track of the client further includes: Determine the time the client stays in each cluster according to the historical trajectory; The cluster where the client stays the longest is determined as the cluster currently participating in the training of the client.

10. A location-aware federated learning training device, characterized in that: include: Data collection module: used to obtain the historical track and resource status of the client, the resource status includes the computing power and communication capacity of the client; Data processing module: used to perform trajectory prediction according to the historical trajectory in each round of training, obtain the trajectory prediction result of each client, and estimate the response delay of each client according to the resource situation and the trajectory prediction result; Data partitioning module: used to select a corresponding sampling client set and a test client set for each cluster based on the response delay and the trajectory prediction result of each client; Training and testing module: used to assign corresponding cluster models to the sampling client set and the test client set to perform local training and testing tasks, and loop the above process until the end condition of the training is met.

Images