A Dynamic Federated Learning Method for Vehicle Networks Based on Node Segmentation

Abstract

This invention provides a dynamic federated learning method for vehicular networks based on node segmentation. A node distribution map is formed based on the geographical distribution of servers and vehicle terminals. Vehicle terminal nodes select the nearest server node in the node distribution map to form a client-server architecture for federated learning. For vehicle terminal nodes not connected to a server, a minimum spanning tree algorithm is used for grouping. Clients connected to the server perform client-server architecture federated learning, while the remaining nodes are grouped for peer-to-peer architecture federated learning. This invention groups geographically unevenly distributed nodes into nearby groups for federated learning, increasing the participation of vehicle terminals in federated learning based on geographical location, and reducing the impact of excessively long response times caused by too many nodes through grouping.

Description

Technical Field

[0001] This invention relates to the field of federated learning technology for vehicle networks, and in particular to a dynamic federated learning method for vehicle networks based on node segmentation. Background Technology

[0002] With the widespread adoption of internet services, the Internet of Vehicles (IoV), as a direction of the Internet of Things (IoT), has seen a significant increase in data volume, leading to the introduction of machine learning into IoV applications. Traditional machine learning employs a centralized processing model, where servers collect all local data from clients for training. However, this model exposes user data directly to the server, which clearly contradicts current user privacy demands for internet services. Federated learning is a novel distributed machine learning model that transforms the centralized model of machine learning data into one where only the model is centralized, and data is stored locally.

[0003] Federated learning is a distributed machine learning technique. Its core idea is to train models in a distributed manner across multiple data sources that have local data. It does not require exchanging local data, but only exchanges model parameters or intermediate results to build a global model based on virtual fused data, thereby achieving a balance between data privacy protection and data sharing computation.

[0004] The typical steps of federated learning are:

[0005] 1) Initialization: The server distributes the initialization model to the terminal;

[0006] 2) Local training: The terminal uses local data to train the model, obtains a local model, and then transmits it to the server;

[0007] 3) Model aggregation: The server aggregates multiple local models uploaded by the terminal to obtain a global model;

[0008] 4) Model update: The terminal downloads the global model and updates the model.

[0009] Besides the client-server architecture mentioned above, federated learning also has a peer-to-peer architecture, where there is no central server and all interactions are conducted directly between clients. After a client trains the original model, it needs to transmit the local model parameters to the other clients participating in the federated learning process, and then each client aggregates the received models.

[0010] Currently, most federated learning in existing vehicle-to-everything (V2X) environments adopts a client-server architecture. In this architecture, the server aggregates models from clients and undertakes some computational tasks, but this architecture requires server participation. Due to the mobility of vehicles, the locations of vehicle terminals and servers are not fixed. When a vehicle terminal travels far from the server, it may be unable to effectively participate in the federated learning process due to connection instability. There is also a peer-to-peer architecture in federated learning, which allows for local aggregation and updates by exchanging models with surrounding terminals without server involvement. Summary of the Invention

[0011] Embodiments of the present invention provide a dynamic federated learning method for vehicle networks based on node segmentation.

[0012] To achieve the above objectives, the present invention adopts the following technical solution.

[0013] A dynamic federated learning method for vehicular networks based on node segmentation includes:

[0014] A node distribution map is formed based on the geographical distribution of servers and vehicle terminals. The nodes include two categories: server nodes and vehicle terminal nodes.

[0015] The vehicle terminal node selects the server node closest to it from the node distribution map to form a federated learning client-server architecture.

[0016] For vehicle terminal nodes that are not connected to the server, the minimum spanning tree algorithm is used to group them and generate multiple terminal subgraphs.

[0017] In the client-server architecture, the server and vehicle terminal nodes perform federated learning; in the terminal subgraph, each vehicle terminal performs federated learning under a peer-to-peer architecture.

[0018] Preferably, the node distribution map is formed based on the geographical location distribution of servers and vehicle terminals. The nodes include two types: server nodes and vehicle terminal nodes, including:

[0019] Based on the geographical distribution of static servers and dynamic vehicle terminals at a certain moment, a node distribution map is formed. The nodes in the node distribution map include two types: server nodes and vehicle terminal nodes. The distance between nodes is mapped to the weight of the connecting edges between nodes in the node distribution map, and connecting edges with weights greater than a set threshold are removed from the node distribution map.

[0020] Preferably, the vehicle terminal node selects the nearest server node from the node distribution map to form a federated learning client-server architecture, including:

[0021] When examining the neighboring nodes of a vehicle terminal node on the node distribution map, if a vehicle terminal node has a connection with at least one server node, the vehicle terminal node selects the server node closest to it to form a federated learning client-server architecture, retains the connection edge with the server node with the lowest weight, deletes the remaining connection edges of the vehicle terminal node, and adds the vehicle terminal node to the connection user list of the selected server.

[0022] Each server can connect to a maximum of m vehicle terminals. After a server has connected to m vehicle terminals, the server node deletes the edges between itself and nodes not on the list of connected users on the server. After performing the above operations on all vehicle terminal nodes and server nodes in the node distribution map, the node distribution map is divided into multiple subgraphs.

[0023] Preferably, for vehicle terminal nodes that are not connected to the server, the minimum spanning tree algorithm is used to group them and generate multiple terminal subgraphs, including:

[0024] For vehicle terminal nodes that are not connected to the server, the minimum spanning tree algorithm is used to group them. A vehicle terminal node that is not connected to the server is randomly selected and divided into a subgraph. The minimum weight edge between the node and other nodes in the subgraph is repeatedly searched, and the adjacent nodes of the minimum weight edge are added to the subgraph in turn until the number in the subgraph reaches the set value n. After completing the subgraph cutting, a node is randomly selected from the remaining nodes and the above steps are repeated until only isolated vehicle terminal nodes remain in the node distribution map.

[0025] Preferably, the method further includes:

[0026] For the last remaining isolated vehicle terminal node in the node distribution map, check if the isolated vehicle terminal node has an adjacent edge with the already cut terminal subgraph. If it does, assign the isolated vehicle terminal node to the terminal subgraph where the adjacent edge with the smallest weight is located; if it does not, divide the isolated vehicle terminal node into its own terminal subgraph.

[0027] Preferably, the various vehicle terminals in the terminal subgraph perform federated learning under a peer-to-peer architecture, including:

[0028] After training its local model, the vehicle terminal node in the terminal subgraph exchanges its local model with its neighboring nodes, and then aggregates the received model with the local model to obtain the updated local model for that node.

[0029] As can be seen from the technical solutions provided by the embodiments of the present invention above, the embodiments of the present invention dynamically allocate all terminal nodes on the current map according to the server distribution in geographical location. When the terminal is close to the server, the client-server architecture is selected first. When it is far away, the terminal is divided into federated learning under a peer architecture with other nearby terminals through node segmentation, thereby deploying the two architectures on a map.

[0030] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and will become apparent from the description or may be learned by practice of the invention. Attached Figure Description

[0031] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 A flowchart illustrating a dynamic vehicular network federated learning method based on node segmentation, provided in an embodiment of the present invention.

[0033] Figure 2 This is a schematic diagram illustrating the segmentation of a node distribution map provided in an embodiment of the present invention. Detailed Implementation

[0034] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0035] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or couplings. The term “and / or” as used herein includes any and all combinations of one or more of the associated listed items.

[0036] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0037] To facilitate understanding of the embodiments of the present invention, the following will provide further explanation and description with reference to the accompanying drawings and several specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0038] This invention addresses the geographical distribution of servers by dynamically allocating all vehicle terminal nodes on the current map. When a terminal is close to the server, a client-server architecture is prioritized. When a terminal is far from the server, node segmentation is used to form a federated learning architecture with other nearby terminals, thereby deploying both architectures on a single map.

[0039] To ensure that all terminals participate in the federated learning process, this invention applies two federated learning architectures to the same vehicle network node map, proposing a node-segmentation-based dynamic federated learning method for vehicle networks. This method maps geographically located server nodes and vehicle terminal nodes into a node distribution graph, recording the distances between nodes. For vehicle terminal nodes connected to server nodes, priority is given to connecting to the server, with each server connecting to a maximum of m vehicle terminal nodes. Vehicle terminal nodes connected to the server form a client-server federated learning architecture, performing client-server federated learning while severing connections with other vehicle terminal nodes. For the remaining vehicle terminal nodes not connected to the server, the minimum spanning tree principle is used to segment the remaining nodes into multiple terminal subgraphs, each with a maximum of n nodes, implementing peer-to-peer federated learning.

[0040] The processing flow of a dynamic vehicular network federated learning method based on node segmentation provided in this embodiment of the invention is as follows: Figure 1 As shown, the processing steps include the following:

[0041] Step S10: Based on the geographical distribution of servers and vehicle terminals, a node distribution map is formed, which includes two types of nodes: server nodes and vehicle terminal nodes.

[0042] For the geographical distribution of static servers and dynamic vehicle terminals at a given moment, a node distribution map is created. The nodes in the node distribution map include both server nodes and vehicle terminal nodes. The distance between nodes is mapped to the weight of the connecting edges in the node distribution map. Because when the distance between nodes is too large, even if there is a connection, it will be unstable. Therefore, after mapping the actual geographical locations to the node distribution map, connecting edges with weights greater than a set threshold for stable connections are removed. For example, the threshold can be set to 50m, that is, edges with weights greater than 50 are removed from the node distribution map to ensure the stability of connections between nodes.

[0043] Step S20: The vehicle terminal node selects the server node closest to it in the node distribution map to form a federated learning client-server architecture.

[0044] When examining the neighboring nodes of a vehicle terminal node, if a vehicle terminal node has a connection with at least one server node, the vehicle terminal node selects the server node closest to it to form a federated learning client-server architecture. That is, it only retains the edge with the smallest weight between itself and the server node, deletes the rest of the connection edges between the vehicle terminal node, and adds the vehicle terminal node to the server's list of connected users.

[0045] Each server can connect to a maximum of m vehicle terminals, with m being between 10 and 20. Once a server has connected to m vehicle terminals, it removes edges between itself and nodes not on its connected user list. Servers and vehicle terminals are mapped in a node distribution graph. Server nodes only participate in client-server architecture federated learning, with a maximum terminal connection limit of m. For vehicle terminal nodes, when connected to a server, they prioritize client-server architecture federated learning and therefore disconnect from other nodes.

[0046] After performing the above operations on all vehicle terminal nodes and server nodes in the node distribution map, the node distribution map is divided into multiple subgraphs. Each subgraph contains at most m vehicle terminal nodes and 1 server node to form a client-server architecture for federated learning operations.

[0047] Step S30: For vehicle terminal nodes that are not connected to the server, the minimum spanning tree algorithm is used to group them and generate multiple terminal subgraphs containing at most n vehicle terminal nodes. The number of n can fluctuate between 15 and 30. The vehicle terminal nodes in each terminal subgraph learn the peer architecture.

[0048] Randomly select a vehicle terminal node and partition it into a subgraph. Then, repeatedly search for the minimum weight edge between the node and other nodes in the subgraph, and add the adjacent nodes of the minimum weight edge to the subgraph in turn until the number of nodes in the subgraph reaches n. That is, after randomly selecting a vehicle terminal node v and adding it to the subgraph node set U (i.e., U = {v}), repeatedly find the minimum weight edge among all edges (u, w) of u ∈ U, w ∈ VU, and add the node w to set U until |U| = n. After completing one subgraph partition, randomly select another node from the remaining nodes and repeat the above steps until only isolated vehicle terminal nodes remain in the node distribution map.

[0049] Minimum Spanning Tree Algorithm: Given an undirected graph, if any two vertices in its subgraph are connected and form a tree structure, then this tree is called a spanning tree. When the graph connecting the vertices has weights, the tree structure with the smallest sum of weights is the minimum spanning tree.

[0050] Prim's algorithm:

[0051] 1) The minimum spanning tree can be found starting from any node;

[0052] 2) After a point is added to the selected points, all new edges originating from that point are unlocked;

[0053] 3) Select the smallest edge from all the unlocked edges, and then see if this edge forms a cycle;

[0054] 4) If you know how, don't use the current edge, continue to examine the smallest edge among the remaining unlocked edges, and repeat step 3);

[0055] 5) If you don't know how, to select the current edge, add the point that the edge points to to the selected points, and repeat step 2);

[0056] 6) Once all points have been selected, the minimum spanning tree is obtained.

[0057] Step S40: For the last remaining isolated vehicle terminal node in the node distribution map, check if the isolated vehicle terminal node has an adjacent edge with the already segmented terminal subgraph. If it does, assign the isolated vehicle terminal node to the terminal subgraph where the adjacent edge with the smallest weight is located; if it does not, divide the isolated vehicle terminal node into its own terminal subgraph.

[0058] At this point, the node distribution map has been completely divided into multiple subgraphs. Subgraphs with server nodes implement client-server architecture federated learning, while subgraphs without server nodes implement peer-to-peer architecture federated learning. Figure 2 This is a schematic diagram illustrating the segmentation of a node distribution map provided in an embodiment of the present invention.

[0059] Step S50: For client-server architecture federated learning, the classic federated learning mode is used; while for peer-to-peer architecture federated learning, after the vehicle terminal node in the subgraph trains its local model, it exchanges its local model with its neighboring nodes, and aggregates the received model with the local model to obtain the updated local model of the node. This ensures that all nodes can participate in the federated learning process as much as possible.

[0060] In summary, the embodiments of the present invention address the problem of terminals being unable to participate in federated learning due to their distance from the server by simultaneously implementing client-server and peer-to-peer federated learning architectures.

[0061] After several steps, the server node takes on the task of collecting and aggregating local models from nearby terminal nodes. Terminal nodes that are geographically distant and unable to participate in client-server federated learning form a peer-to-peer federated learning architecture with nearby nodes through the minimum spanning tree algorithm.

[0062] Through this invention, nodes that are geographically unevenly distributed are grouped together for federated learning. Terminal nodes will not be unable to participate in federated learning due to their distance from server nodes. This ensures that the datasets collected by each vehicle terminal node can be used to train the global model as much as possible, thereby increasing the participation of vehicle terminals in federated learning under different geographical locations. Furthermore, grouping reduces the impact of excessive response time caused by too many nodes.

[0063] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0064] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0065] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for apparatus or system embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The apparatus and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0066] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A dynamic federated learning method for vehicular networks based on node segmentation, characterized in that, include: A node distribution map is formed based on the geographical distribution of servers and vehicle terminals. The nodes include two categories: server nodes and vehicle terminal nodes. The vehicle terminal node selects the server node closest to it from the node distribution map to form a federated learning client-server architecture. For vehicle terminal nodes that are not connected to the server, the minimum spanning tree algorithm is used to group them and generate multiple terminal subgraphs. Federated learning is performed between the server and vehicle terminal nodes in the client-server architecture. After the vehicle terminal node in the terminal subgraph has trained its local model, it exchanges its local model with its neighboring nodes and aggregates the received model with the local model to obtain the updated local model for that node. The aforementioned node distribution map, based on the geographical location distribution of servers and vehicle terminals, includes two types of nodes: server nodes and vehicle terminal nodes. Based on the geographical distribution of static servers and dynamic vehicle terminals at a certain moment, a node distribution map is formed. The nodes in the node distribution map include two types: server nodes and vehicle terminal nodes. The distance between nodes is mapped to the weight of the connection edges between nodes in the node distribution map, and connection edges with weights greater than a set threshold are removed from the node distribution map. The vehicle terminal node selects the nearest server node from the node distribution map to form a federated learning client-server architecture, including: Step S1: View the neighboring nodes of the vehicle terminal node in the node distribution map. When there is a connection between the vehicle terminal node and at least one server node, the vehicle terminal node selects the server node closest to it to form a federated learning client-server architecture, retains the connection edge with the smallest weight between the vehicle terminal node and the server node, deletes the remaining connection edges of the vehicle terminal node, and adds the vehicle terminal node to the connection user list of the selected server. Step S2: Each server can connect to a maximum of m vehicle terminals. After a server connects to m vehicle terminals, the server node deletes the edges between itself and nodes not on the connected user list on the server. After performing the operation in step S1 on all vehicle terminal nodes and server nodes in the node distribution map, the node distribution map is cut into multiple subgraphs.

2. The method according to claim 1, characterized in that, For vehicle terminal nodes that are not connected to the server, the minimum spanning tree algorithm is used to group them and generate multiple terminal subgraphs, including: For vehicle terminal nodes that are not connected to the server, the minimum spanning tree algorithm is used for grouping. Randomly select a vehicle terminal node that is not connected to the server, and perform a subgraph cutting operation on the vehicle terminal node. The subgraph cutting operation includes: dividing it into subgraphs, repeatedly finding the minimum weight edge between the node in the subgraph and other nodes, and sequentially adding the adjacent nodes of the minimum weight edge to the subgraph, until the number in the subgraph reaches a set value n. After completing one subgraph cutting... Randomly select another vehicle terminal node from the remaining nodes, and perform the subgraph cutting operation on the vehicle terminal node. Repeat the above steps until only isolated vehicle terminal nodes remain in the node distribution map.

3. The method according to claim 2, characterized in that, The method further includes: For the last remaining isolated vehicle terminal node in the node distribution map, check if the isolated vehicle terminal node has an adjacent edge with the already cut terminal subgraph. If it does, assign the isolated vehicle terminal node to the terminal subgraph where the adjacent edge with the smallest weight is located; if it does not, divide the isolated vehicle terminal node into its own terminal subgraph.

Images