Method, device, computer equipment and storage medium for processing heterogeneous data of Internet of Vehicles

By dynamically allocating computing resources in the Internet of Vehicles and splitting the hidden layers of deep learning models, the problem of insufficient computing resources caused by the heterogeneity of smart bicycles is solved, and the efficiency and resource utilization of Internet of Vehicles data processing are improved.

CN118504660BActive Publication Date: 2024-11-22湖南红普创新科技发展有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410962502.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-22
Estimated Expiration
2044-07-18

AI Technical Summary

Technical Problem

In the Internet of Vehicles, the heterogeneity of smart bicycles leads to insufficient computing resources, resulting in the inability to train the initialization model issued by the server, and resource limitations are required when performing deep neural network training computing-intensive tasks locally, affecting the efficiency and effectiveness of federated learning.

Method used

The global model is initialized and aggregated and updated through the central server, and the computing resources are dynamically allocated in each round of training, using the computing power of each intelligent bicycle, achieving balanced utilization of resources, splitting the hidden layer of the deep learning model, and using heterogeneous structure Carry out intelligent bicycles and sort them according to their capabilities to achieve dynamic resource allocation.

Benefits of technology

The efficiency of heterogeneous data processing in the Internet of Vehicles is improved, ensuring that all smart bicycles can effectively participate in model training, making full use of the computing resources of each smart bicycle, and reducing the problem of unbalanced computing load.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118504660B_ABST
    Figure CN118504660B_ABST
Patent Text Reader

Abstract

The present invention discloses a method, device, computer equipment and storage medium for processing heterogeneous data of an Internet of Vehicles, including initializing a basic global model to obtain an initial global model, determining an initial global unloading point set and an initial global unloading relationship diagram, and broadcasting the global information to each single-vehicle node; receiving local information sent by each single-vehicle node, the local information including local calculation time, local model, local unloading point and local unloading relationship diagram; based on each local information, aggregating and updating the global information, and iteratively training to obtain a target global model; updating the unloading point set and the unloading relationship diagram based on the target global model to obtain updated global information; broadcasting the updated global information to each single-vehicle node, and using the updated global information to perform heterogeneous data task processing; the present invention is beneficial to improving the efficiency of processing heterogeneous data of the Internet of Vehicles.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a method, device, computer equipment and medium for processing heterogeneous data in an Internet of Vehicles. Background Art

[0002] As urban transportation pursues higher efficiency, cost, safety and intelligence, artificial intelligence has become an important field of cross-research between information technology and urban smart transportation. In the ubiquitous intelligence of future 6G vehicle networks, deep model training is crucial because it can mine knowledge from vehicle data to improve the quality of many vehicle services driven by artificial intelligence (AI). However, this centralized deep model training method faces some challenges.

[0003] Traditional centralized machine learning requires mobile devices to upload raw data to cloud computing centers or edge servers for deep model training, which not only leads to high data transmission costs, but also poses the risk of privacy leakage. To solve these problems, Google proposed the concept of Federated Learning (FL) in 2016. As a promising distributed learning paradigm, federated learning enables each mobile device to perform model training locally and upload the updated local model parameters to the edge server for aggregation, completing the model training required by the edge server in a distributed manner. In this process, mobile devices only need to share model parameters without disclosing raw data, making the data "available but invisible, computable but unrecognizable, and available and traceable", thereby protecting the user's data privacy security to a certain extent. At the same time, since the model parameter size is smaller than the original data size, federated learning greatly reduces the cost of data transmission. Especially in vehicle networks, federated learning enables smart bikes to train local models with private data, and further upload local gradients to the global server for aggregation to achieve global model training. In this process, smart bikes retain private data and only upload smaller local gradients, which greatly reduces communication overhead and mitigates the risk of privacy leakage.

[0004] In the process of implementing the present invention, the inventors found that in the vehicle-connected network, the actual deployment of federated learning still faces some new challenges, especially the problem of insufficient computing resources caused by the heterogeneity of smart bicycles. 1) Due to the diversity of smart bicycles, it may not be possible to train the initialization model sent by the server. This heterogeneity of capabilities may arise from hardware limitations such as processing power, storage capacity or battery life, or due to differences in operating system, driver or other software configurations. For example, although all smart bicycles may need to perform a task, not all smart bicycles have sufficient computing power to train complex deep neural network models. That is, different smart bicycles may require models of different complexity. Therefore, there is an urgent need to develop a federated learning framework that can adapt to the heterogeneity of device capabilities to ensure that all smart bicycles can effectively participate in model training. 2) Smart bicycles need to undertake a large amount of local intensive task calculations, such as deeper deep neural network training. However, due to the limitations of smart bicycle capabilities, it may not be practical to perform these computationally intensive tasks locally. This further highlights the impact of capability heterogeneity on the implementation of federated learning. Therefore, it is necessary to continue to find a method that can balance the computing load between smart bikes to ensure effective model training even on resource-constrained smart bikes. 3) Due to the large differences in the computing resource utilization of each smart bike, there is an imbalance in the distribution of local computing tasks between smart bikes. For example, some smart bikes with strong computing power may have a large amount of remaining computing resources, while some smart bikes with weaker computing power may encounter insufficient resources when performing computing tasks. This situation may affect the efficiency and effectiveness of federated learning. Therefore, there is an urgent need for a method for processing heterogeneous data in the Internet of Vehicles to make full use of the computing resources of each smart bike and improve the efficiency of heterogeneous data processing in the Internet of Vehicles. Summary of the invention

[0005] Embodiments of the present invention provide a method, apparatus, computer equipment and storage medium for processing heterogeneous data in an Internet of Vehicles to improve the efficiency of processing heterogeneous data in an Internet of Vehicles.

[0006] In order to solve the above technical problems, an embodiment of the present application provides a method for processing heterogeneous data in an Internet of Vehicles, which is applied to a distributed cluster, wherein the distributed cluster includes a central server and multiple single-vehicle nodes. The method for processing heterogeneous data in an Internet of Vehicles includes the following steps performed by the central server:

[0007] Initializing the basic global model to obtain an initial global model, and determining an initial global unloading point set and an initial global unloading relationship graph, and using the global model, the initial unloading point set and the initial unloading relationship graph as global information;

[0008] Broadcasting the global information to each of the bicycle nodes;

[0009] Receiving local information sent by each of the bicycle nodes, the local information including local computing time, local model, local unloading point and local unloading relationship graph;

[0010] Based on each of the local information, the global information is aggregated and updated, and the global information is broadcasted back to each of the bicycle nodes to continue the execution until the global model converges or reaches a preset accuracy, and the global model obtained at this time is used as the target global model;

[0011] Based on the target global model, an unloading point set and an unloading relationship graph are updated to obtain updated global information;

[0012] The updated global information is broadcasted to each of the bicycle nodes, and the updated global information is used to perform heterogeneous data task processing.

[0013] Optionally, the initializing the basic global model to obtain an initial global model, and determining an initial global unloading point set and an initial global unloading relationship graph includes:

[0014] In round When , the operating points of all bicycle nodes are located in the last layer of the hidden layer of the neural network, and the set of initialized unloading points is , , The set of uninstall points is The position parameter of the operation point of the k-th bicycle node in the initial round; n is the total number of bicycle nodes, k is the operation point of a bicycle node in a specific round t, which is located in the k-th layer of the neural network; M is the depth of the hidden layer of the model; as the round increases As the number of nodes increases, the position of the operating point of each bicycle node changes dynamically. In the bicycle node The operating point is moved to layer, , at this time, for round The global unloading point set of the operating point is .

[0015] Optionally, the initializing the basic global model to obtain an initial global model, and determining an initial global unloading point set and an initial global unloading relationship graph further includes:

[0016] In round When the initial uninstallation relationship graph is a The matrix ; Where n is the total number of bicycle nodes;

[0017] With the round As the number of unloading nodes increases, the unloading relationship of each bicycle node changes dynamically. In the bicycle node Choose to offload the task to the client , then for round The uninstall relationship diagram in is ;

[0018] in, Represents a bicycle node To the bicycle node The number of unloading parameters for each bicycle node When acting as an unloading source, it can unload to at most one target bicycle node, satisfying , , each bicycle node As an unloading target, it can receive unloading from multiple bicycle nodes, satisfying , .

[0019] Optionally, updating the unloading point set and the unloading relationship graph based on the target global model to obtain updated global information includes:

[0020] Rollback parameter set from actual send Take out the elements one by one , and actually receive the rollback parameter set Search in, is the actual rollback parameter with a positive sign, i is the index, representing the i-th bicycle node;

[0021] If the rollback parameter amount is actually received There are elements in satisfy , then according to Update the unloading point, where n is the total number of bicycle nodes; is the actual rollback parameter amount of the vehicle-mounted node i in the actual sent rollback parameter amount set G at round t+1; is the average rollback parameter of all on-board nodes participating in the calculation; It is the element in the set G of rollback parameters actually sent. The size of the local dataset;

[0022] Build elements based on the updated uninstall point With elements New uninstall diagram ;

[0023] like , the element With elements Add the relationship to , from the actual send rollback parameter quantity set Remove elements from , and calculate the remaining actual rollback parameters , and the remaining actual rollback parameter amount Add the actual amount of received rollback parameters middle;

[0024] If the rollback parameter amount is actually received No elements satisfying the condition were found , then according to Update the uninstall point and update the uninstall diagram accordingly , The maximum actual rollback parameter amount in the actually sent rollback parameter amount set G in round t+1.

[0025] In order to solve the above technical problems, an embodiment of the present application provides a method for processing heterogeneous data in an Internet of Vehicles, and further includes the following steps performed by the single vehicle node:

[0026] Receiving global information broadcasted by a central server, wherein the global information includes a global model, an initial unloading point set, and an initial unloading relationship graph;

[0027] Based on the global information, the local model is trained and updated according to the forward propagation of parameter unloading and the back propagation based on resource unloading to obtain an updated local model, and local information in the training process is determined, wherein the local information includes local computing time, local model, local unloading point and local unloading relationship diagram;

[0028] Sending local information to the central server;

[0029] The updated global information broadcasted by the central server is received, and the updated global information is used to perform task processing of heterogeneous data.

[0030] Optionally, the training and updating of the local model based on the global information in a forward propagation method based on parameter unloading and a back propagation method based on resource unloading to obtain the updated local model includes:

[0031] In the local forward propagation phase before the operation point, execution is performed from the input layer to the layer where the operation point is located. The forward propagation of the layer obtains fragmented data ,symbol It represents the function combination between layers in a neural network, which converts the output of the previous layer into the input of the next layer. Represents a bicycle node The model in round Middle The forward propagation function of the layer;

[0032] In the forward propagation phase unloaded between the operation point and the output layer, if the unloading relationship graph Elements in , then the bicycle node Break it down Send to the bicycle node ;

[0033] Bicycle Node from The layer continues to perform forward propagation and obtains fragmented data , Represents a bicycle node Receive from the bicycle node Broken data And in its model Layer to The result of forward propagation between layers;

[0034] If each bicycle node uses fragmented data received from other clients , continue to perform forward propagation to the output layer and get the final output , M is the depth of the hidden layer of the model.

[0035] Optionally, the training and updating of the local model based on the global information according to the forward propagation of parameter unloading and the back propagation based on resource unloading to obtain the updated local model further includes:

[0036] In each round of communication, a backpropagation process is usually performed after forward propagation to calculate the gradient according to the loss function and update the weights of the network. The backpropagation process includes backpropagation of the local output layer, unloading backpropagation, and local backpropagation before the operation point.

[0037] In order to solve the above technical problems, the embodiment of the present application also provides a heterogeneous data processing device for an Internet of Vehicles, including a central server and multiple single-vehicle nodes; the central server includes an initialization module, a first broadcast module, a node information receiving module, a global model training module, a global information update module, and a second broadcast module, and the single-vehicle node includes a local model training module, a local information sending module, and a second receiving module; wherein:

[0038] An initialization module is used for the central server to initialize the basic global model, obtain the initial global model, and determine the initial global unloading point set and the initial global unloading relationship diagram, and use the global model, the initial unloading point set and the initial unloading relationship diagram as global information;

[0039] A first broadcast module, used for the central server to broadcast the global information to each of the bicycle nodes;

[0040] A first receiving module, configured for each of the bicycle nodes to receive global information broadcasted by the central server, wherein the global information includes a global model, an initial unloading point set, and an initial unloading relationship diagram;

[0041] A local model training module is used for each of the bicycle nodes to train and update the local model based on the passing information, according to the forward propagation of parameter unloading and the reverse propagation based on resource unloading, to obtain an updated local model, and to determine the local information in the training process, wherein the local information includes the local computing time, the local model, the local unloading point and the local unloading relationship diagram;

[0042] A local information sending module, used for each of the bicycle nodes to send local information to the central server;

[0043] A node information receiving module, used for the central server to receive local information sent by each of the single-vehicle nodes, wherein the local information includes local calculation time, local model, local unloading point and local unloading relationship diagram;

[0044] A global model training module, which is used for the central server to aggregate and update the global information based on each of the local information, and return to broadcast the global information to each of the bicycle nodes to continue execution until the global model converges or reaches a preset accuracy, and the global model obtained at this time is used as the target global model;

[0045] A global information updating module, used for the central server to update the unloading point set and the unloading relationship diagram based on the target global model to obtain updated global information;

[0046] A second broadcast module, configured for the central server to broadcast the updated global information to each of the bicycle nodes, and to use the updated global information to perform heterogeneous data task processing;

[0047] The second receiving module is used for each of the bicycle nodes to receive the updated global information broadcast by the central server, and use the updated global information to perform heterogeneous data task processing.

[0048] In order to solve the above technical problems, an embodiment of the present application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned vehicle network heterogeneous data processing method are implemented.

[0049] In order to solve the above technical problems, an embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned vehicle network heterogeneous data processing method are implemented.

[0050] The embodiment of the present invention provides a method, device, computer equipment and storage medium for processing heterogeneous data of the Internet of Vehicles. The central server initializes the basic global model to obtain an initial global model, and determines an initial global unloading point set and an initial global unloading relationship graph, and uses the global model, the initial unloading point set and the initial unloading relationship graph as global information; the central server broadcasts the global information to each single vehicle node; the single vehicle node receives the global information broadcast by the central server, wherein the global information includes the global model, the initial unloading point set and the initial unloading relationship graph; based on the global information, the single vehicle node trains and updates the local model according to the forward propagation of parameter unloading and the reverse propagation based on resource unloading to obtain an updated local model, and determines the local information in the training process, wherein the local information includes the local computing time, the local model, the local unloading point and the local unloading relationship graph. The central server receives the local information sent by each bicycle node, including local computing time, local model, local unloading point and local unloading relationship diagram; the central server aggregates and updates the global information based on each local information, and returns to broadcast the global information to each bicycle node to continue execution until the global model converges or reaches the preset accuracy, and the global model obtained at this time is used as the target global model; the central server updates the unloading point set and unloading relationship diagram based on the target global model to obtain the updated global information; the central server broadcasts the updated global information to each bicycle node, and uses the updated global information to process heterogeneous data tasks; the bicycle node receives the updated global information broadcast by the central server, and uses the updated global information to process heterogeneous data tasks. The hidden layers of the deep learning model are split, and these layers are held by heterogeneous smart bicycles. In each round of the training process, the central server ranks the capabilities of different smart bikes according to the time when smart bikes in the same area use their own original data and labels to train the sub-model in the first round, thereby realizing a dynamic resource allocation mechanism to fully utilize the computing resources of each smart bike and improve the efficiency of heterogeneous data processing in the Internet of Vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0052] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0053] Figure 2 It is a flowchart of an embodiment of the vehicle network heterogeneous data processing method of the present application;

[0054] Figure 3 is a structural schematic diagram of an embodiment of a vehicle networking heterogeneous data processing device according to the present application;

[0055] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0057] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] See also Figure 1 ,like Figure 1As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0060] Users can use terminal devices 101 , 102 , 103 to interact with server 105 via network 104 to receive or send messages, etc.

[0061] Terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV), laptop computers, desktop computers, etc.

[0062] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0063] It should be noted that the vehicle network heterogeneous data processing method provided in the embodiment of the present application is executed by a server, and accordingly, the vehicle network heterogeneous data processing device is arranged in the server.

[0064] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. According to the implementation requirements, there may be any number of terminal devices, networks and servers. The terminal devices 101, 102, 103 in the embodiment of the present application may specifically correspond to application systems in actual production.

[0065] See also Figure 2 , Figure 2 A method for processing heterogeneous data in an Internet of Vehicles provided by an embodiment of the present invention is shown. The method is applied in Figure 1 The server (central server) and terminal device (bicycle node) in the example are explained as follows:

[0066] S201: The central server initializes the basic global model to obtain an initial global model, and determines an initial global unloading point set and an initial global unloading relationship graph, and uses the global model, the initial unloading point set and the initial unloading relationship graph as global information.

[0067] Specifically, this embodiment is based on the federated learning and segmented learning frameworks and focuses on optimizing heterogeneous resource allocation in a vehicle-road collaborative environment. The model aims to increase the diversity of federated datasets by increasing the participation of single-vehicle nodes with weaker computing power, thereby reducing the risk of bias and expanding the scope of application. The system consists of a server located in a regional traffic management center and a collection of vehicles (single-vehicle nodes) distributed in different areas of the city. The central server is responsible for the initialization and aggregation of the global model, while the bicycle node is responsible for the training of the local model. Equipped with local datasets and computing power , and have the basic requirements to meet the federated learning task. , bicycle node Maintain a Local model composed of 10 layers of deep neural network , whose output is obtained through the recursive function It is calculated that The network layer. The bicycle nodes communicate with each other through device-to-device (D2D) communication to exchange data and model information. It is worth noting that the tag data is still retained locally and does not participate in the transmission. The core operations of the system include model initialization , Uninstallation point collection and uninstallation diagram Specifically, at the beginning of each round of training, the server initializes the global model , and set the initial set of unloading points and uninstallation diagram The central server then broadcasts this information to all bicycle nodes in the area. Each bicycle node receives , uninstall point and uninstallation diagram After that, perform local training to generate an updated local model The bicycle node that has completed the training will calculate the time , local model , Uninstall point , and uninstall relationship diagram Send it back to the server, and the central server will dynamically update it after receiving this information and , and perform global model aggregation. This process continues until the global model converges or reaches the preset accuracy.

[0068] Among them, in round When , the server initializes the global model as , Represents the initial parameters of the neural network model. This initialization step sets the starting point for the subsequent federated learning training process.

[0069] In a specific optional implementation, in step S201, the basic global model is initialized to obtain an initial global model, and an initial global unloading point set and an initial global unloading relationship diagram are determined, including:

[0070] In round When , the operating points of all bicycle nodes are located in the last layer of the hidden layer of the neural network, and the set of initialized unloading points is , , The set of uninstall points is The position parameter of the operation point of the k-th bicycle node in the initial round; n is the total number of bicycle nodes, k is the operation point of a bicycle node in a specific round t, which is located in the k-th layer of the neural network; M is the depth of the hidden layer of the model; as the round increases As the number of nodes increases, the position of the operating point of each bicycle node changes dynamically. In the bicycle node The operating point is moved to layer, , at this time, for round The global unloading point set of the operating point is .

[0071] In a specific optional implementation, in step S201, initializing the basic global model to obtain an initial global model, and determining an initial global unloading point set and an initial global unloading relationship diagram further includes:

[0072] In round When the initial uninstallation relationship graph is a The matrix ;

[0073] With the round As the number of unloading nodes increases, the unloading relationship of each bicycle node changes dynamically. In the bicycle node Choose to offload the task to the client , then for round The uninstall relationship diagram in is ;

[0074] in, Represents a bicycle node To the bicycle node The number of unloading parameters for each bicycle node When acting as an unloading source, it can unload to at most one target bicycle node, satisfying , , each bicycle node As an unloading target, it can receive unloading from multiple bicycle nodes, satisfying , .

[0075] S202: The central server broadcasts global information to each bicycle node;

[0076] S203: The bicycle node receives the global information broadcasted by the central server, wherein the global information includes a global model, an initial unloading point set, and an initial unloading relationship graph;

[0077] In each round of communication, forward propagation and back propagation will be performed multiple times, and the training process of different bicycle nodes will be carried out in parallel. The forward propagation process can be divided into three stages: the local forward propagation stage before the operating point (OP), the forward propagation unloaded between the operating point and the output layer, and the forward propagation of the local output layer.

[0078] S204: Based on the global information, the bicycle node trains and updates the local model according to the forward propagation of parameter unloading and the back propagation based on resource unloading to obtain an updated local model, and determines the local information in the training process. The local information includes local computing time, local model, local unloading point and local unloading relationship diagram.

[0079] Specifically, in round In the client The local model owned is represented as , the depth of the hidden layer of this model is For local datasets Given a small batch of data sampled from , the output of the model is obtained through the recursive function

[0080] calculate.

[0081] in, and Respectively represent The weight matrix and bias vector of the layer, is the activation function. The final output of the model is Back propagation also takes into account the model Layer structure,

[0082] ;

[0083] in, Indicates The error term of the layer, is the loss function, is the derivative of the activation function.

[0084] In a specific optional implementation, in step S204, based on global information, the local model is trained and updated according to the forward propagation of parameter unloading and the back propagation based on resource unloading, and the updated local model includes:

[0085] In the local forward propagation phase before the operation point, the execution is from the input layer to the layer where the operation point (OP) is located. The forward propagation of the layer obtains fragmented data .symbol It represents the function combination between layers in a neural network, which converts the output of the previous layer into the input of the next layer. Represents a bicycle node The model in round Middle The forward propagation function of the layer;

[0086] In the forward propagation phase unloaded between the operation point and the output layer, if the unloading relationship graph Elements in , then the bicycle node Break it down Send to the bicycle node ;

[0087] Bicycle Node from The layer continues to perform forward propagation and obtains fragmented data , Represents a bicycle node Receive from the bicycle node Broken data And in its model Layer to The result of forward propagation between layers;

[0088] If each bicycle node uses fragmented data received from other clients , continue to perform forward propagation to the output layer and get the final output , M is the depth of the hidden layer of the model.

[0089] In a specific optional implementation, in step S204, based on the global information, the local model is trained and updated according to the forward propagation of parameter unloading and the back propagation based on resource unloading, and the updated local model is obtained, which further includes:

[0090] In each round of communication, the backpropagation process is usually performed after the forward propagation to calculate the gradient according to the loss function and update the weights of the network. The backpropagation process includes backpropagation of the local output layer, unloading backpropagation, and local backpropagation before the operation point.

[0091] It should be noted that this condition is used to emphasize that not all clients will receive fragmented data from other clients. In some cases, a client may not need to receive data from other clients to complete its forward propagation. For example, if a client does not offload its computation to other clients, or it is the client that ultimately receives offloaded data, then it will directly use its own local data to complete the forward propagation.

[0092] It should be noted that in each round of updates and , the client status is rolled back and the training time is re-set according to the training time of this round. Specifically, this embodiment is based on the previous round of as well as , the amount of parameters sent by the client as a sender or the amount of parameters received by the client as a receiver are fully recovered and re-evaluated.

[0093] Sub-step 1: Parameter rollback.

[0094] (1) Calculate the Actual parameters of the wheel:

[0095] ;

[0096] in, Is an indicator function used to decide whether to offload the computational load. When In this round, the client is the receiving device. When In the round, the client (i.e., bicycle node) is in the sending state. For Clients The collection of receiving clients.

[0097] (2) Calculate the actual parameter quantity after rollback .

[0098] If the client The receiver in the previous round is also the receiver in this round, then If the sign of the impact of this round is negative, then The sign is negative.

[0099] ;

[0100] Among them, this expression calculates the actual parameter change after the client changes its role (from a receiver to a sender). represents the parameter status of client i in round t+1, and Indicates the amount of parameters received from other clients. The entire expression reflects the update of the parameter amount in the new round. It is the percentage of the client's uninstallation parameters in round t+1, indicating the proportion of this client uninstalling in this round. Its sign (positive or negative) indicates whether the client is a sender or a receiver. A positive sign usually means that the client is a sender, and a negative sign indicates a receiver. is a constant representing the normalized or average amount of parameters involved in the unloading process and is used as a benchmark for calculating the unloading impact.

[0101] If the client The previous round was the receiver, and this round is the sender, then If the sign of the impact of this round is positive, then When the sign is positive, it is sent, and when the sign is negative, it is received.

[0102] ;

[0103] in, is the actual parameter amount of client i after round t+1, including parameter changes received or sent due to data offloading. This summation expression calculates the amount of data offloaded to the client by all clients in round t, where is the set of receiving clients of client i (bicycle node i) in round t.

[0104] If the client The sender in the previous round and the sender in this round, then If the sign of the impact of this round is positive, then The sign is positive, and the expression is .

[0105] If the client The previous round was the sender, and this round is the receiver, then If the sign of the impact of this round is negative, then When the sign is positive, it is sent, and when the sign is negative, it is received. The expression is: .

[0106] To sum up, the actual parameter amount after rollback is calculated as follows:

[0107] ;

[0108] in, is the uninstall parameter percentage, For the The actual number of parameters sent or received by the round.

[0109] Sub-step 2: Sorting of actual rollback parameter quantities.

[0110] The present invention patent defines two key parameter sets: the actual send rollback parameter set and the actual received rollback parameter set and .in, and Corresponding to the actual rollback parameter with positive and negative signs, that is, when hour, Indicates the actual amount of rollback parameters sent; on the contrary hour, Indicates the actual amount of rollback parameters received, and satisfies . In addition, it is worth noting that n is not necessarily equal to m.

[0111] S205: The bicycle node sends local information to the central server;

[0112] S206: The central server receives local information sent by each bicycle node, where the local information includes local computing time, local model, local unloading point and local unloading relationship graph;

[0113] S207: The central server aggregates and updates the global information based on each local information, and returns to broadcast the global information to each bicycle node to continue execution until the global model converges or reaches a preset accuracy, and the global model obtained at this time is used as the target global model;

[0114] S208: The central server updates the unloading point set and the unloading relationship graph based on the target global model to obtain updated global information;

[0115] In a specific optional implementation manner, the unloading point set and the unloading relationship graph are updated based on the target global model, and the updated global information obtained includes:

[0116] Rollback parameter set from actual send Take out the elements one by one , and actually receive the rollback parameter set Search in, is the actual rollback parameter with a positive sign, i is the index, representing the i-th bicycle node;

[0117] If the rollback parameter amount is actually received There are elements in satisfy , then according to Update the unloading point, where n is the total number of bicycle nodes; is the actual rollback parameter amount of the bicycle node i in the actual sent rollback parameter amount set G in round t+1; is the average rollback parameter of all on-board nodes participating in the calculation; It is the element in the set G of rollback parameters actually sent. The size of the local dataset;

[0118] Build elements based on the updated uninstall point With elements New uninstall diagram ;

[0119] like , the element With elements Add the relationship to , from the actual send rollback parameter quantity set Remove elements from , and calculate the remaining actual rollback parameters , and the remaining actual rollback parameter amount Add the actual amount of received rollback parameters middle;

[0120] If the rollback parameter amount is actually received No elements satisfying the condition were found , then according to Update the uninstall point and update the uninstall diagram accordingly , The maximum actual rollback parameter amount in the actually sent rollback parameter amount set G in round t+1.

[0121] In each round of communication, the backpropagation process is usually performed after the forward propagation to calculate the gradient according to the loss function and update the weights of the network. The backpropagation process is similar to the forward propagation and can also be divided into three stages: backpropagation of the local output layer (P4), unloading backpropagation (P5), and local backpropagation before OP (P6).

[0122] P4: At this stage, each client First, back propagation is performed from its output layer. The purpose of this process is to calculate the distance from the output layer to the layer where the operation point is located. The gradient of the layer, i.e. .in, Represents the client The model in round No. The back propagation function of the layer, Is the client The output result is:

[0123] P5: If you uninstall the relationship diagram Elements in , then the client Its gradient Send to client , Client Then from The layer continues to perform back propagation and obtains . This means that the client Receive from client The gradient of Back propagation is performed between the layers to the input layer.

[0124] P6: Each client Using gradients received from other clients (if present), this condition is used to emphasize that not all clients participate in the offloading process. For example, some clients may have enough computing power to complete the entire forward and backward propagation process independently without relying on the computing resources of other clients}) Continue to perform backpropagation to the input layer, that is, This stage is to complete the gradient calculation of the entire network and prepare for weight update.

[0125] S209: The central server broadcasts the updated global information to each bicycle node, and uses the updated global information to perform heterogeneous data task processing.

[0126] S210: The bicycle node receives the updated global information broadcasted by the central server, and uses the updated global information to perform task processing of heterogeneous data.

[0127] In this embodiment, the central server initializes the basic global model to obtain the initial global model, and determines the initial global unloading point set and the initial global unloading relationship graph, and uses the global model, the initial unloading point set and the initial unloading relationship graph as global information; the central server broadcasts the global information to each bicycle node; the bicycle node receives the global information broadcast by the central server, wherein the global information includes the global model, the initial unloading point set and the initial unloading relationship graph; based on the global information, the bicycle node trains and updates the local model according to the forward propagation of parameter unloading and the back propagation of resource unloading to obtain the updated local model, and determines the local information in the training process, wherein the local information includes the local computing time, the local model, the local unloading point and the local unloading relationship graph; the bicycle node sends the global information to the central server. Send local information; the central server receives the local information sent by each bicycle node, and the local information includes local computing time, local model, local unloading point and local unloading relationship diagram; the central server aggregates and updates the global information based on each local information, and returns to broadcast the global information to each bicycle node to continue execution until the global model converges or reaches the preset accuracy, and the global model obtained at this time is used as the target global model; the central server updates the unloading point set and unloading relationship diagram based on the target global model to obtain the updated global information; the central server broadcasts the updated global information to each bicycle node, and uses the updated global information to process heterogeneous data tasks; the bicycle node receives the updated global information broadcast by the central server, and uses the updated global information to process heterogeneous data tasks. The hidden layers of the deep learning model are split, and these layers are held by heterogeneous smart bicycles. In each round of the training process, the central server ranks the capabilities of different smart bicycles according to the time when smart bicycles in the same area use their own original data and labels to train sub-models in the first round, thereby realizing a dynamic resource allocation mechanism to make full use of the computing resources of each smart bicycle and improve the efficiency of heterogeneous data processing in the Internet of Vehicles.

[0128] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0129] Figure 3 The schematic block diagram of the vehicle networking heterogeneous data processing device corresponding to the vehicle networking heterogeneous data processing method in the above embodiment is shown. Figure 3As shown, the vehicle network heterogeneous data processing device includes a central server and multiple single-vehicle nodes; the central server includes an initialization module 301, a first broadcast module 302, a node information receiving module 303, a global model training module 304, a global information update module 305, and a second broadcast module 306, and the single-vehicle node includes a first receiving module 307, a local model training module 308, a local information sending module 309, and a second receiving module 310. The functional modules are described in detail as follows:

[0130] Initialization module 301, used for the central server to initialize the basic global model, obtain the initial global model, and determine the initial global unloading point set and the initial global unloading relationship diagram, and use the global model, the initial unloading point set and the initial unloading relationship diagram as global information;

[0131] A first broadcast module 302, configured for the central server to broadcast the global information to each of the bicycle nodes;

[0132] A node information receiving module 303 is used for the central server to receive local information sent by each bicycle node, wherein the local information includes local calculation time, local model, local unloading point and local unloading relationship diagram;

[0133] The global model training module 304 is used for the central server to aggregate and update the global information based on each of the local information, and return to broadcast the global information to each of the bicycle nodes to continue execution until the global model converges or reaches a preset accuracy, and the global model obtained at this time is used as the target global model;

[0134] A global information updating module 305, configured for the central server to update the unloading point set and the unloading relationship diagram based on the target global model to obtain updated global information;

[0135] A second broadcast module 306, configured for the central server to broadcast the updated global information to each of the bicycle nodes, and to use the updated global information to perform heterogeneous data task processing;

[0136] A first receiving module 307, configured for each of the bicycle nodes to receive global information broadcasted by the central server, wherein the global information includes a global model, an initial unloading point set, and an initial unloading relationship diagram;

[0137] The local model training module 308 is used for each of the bicycle nodes to train and update the local model based on the global information, according to the forward propagation of parameter unloading and the reverse propagation based on resource unloading, to obtain an updated local model, and to determine the local information in the training process, wherein the local information includes the local computing time, the local model, the local unloading point and the local unloading relationship diagram;

[0138] A local information sending module 309, configured for each bicycle node to send local information to the central server;

[0139] The second receiving module 310 is used for each of the bicycle nodes to receive the updated global information broadcasted by the central server, and use the updated global information to perform heterogeneous data task processing.

[0140] For the specific definition of the vehicle network heterogeneous data processing device, please refer to the definition of the vehicle network heterogeneous data processing method above, which will not be repeated here. Each module in the above-mentioned vehicle network heterogeneous data processing 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.

[0141] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0142] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components connected to the memory 41, the processor 42, and the network interface 43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0143] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0144] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or D interface display memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device 4. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as the program code of the heterogeneous data processing method of the Internet of Vehicles, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0145] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run the program code stored in the memory 41 or process data, such as running the program code of the vehicle network heterogeneous data processing method.

[0146] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0147] The present application also provides another implementation, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores an interface display program, and the interface display program can be executed by at least one processor so that the at least one processor performs the steps of the above-mentioned vehicle network heterogeneous data processing method.

[0148] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0149] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A method for processing heterogeneous data in an Internet of Vehicles, characterized in that: Applied to a distributed cluster, the distributed cluster includes a central server and multiple single-vehicle nodes, and the vehicle network heterogeneous data processing method includes the following steps performed by the central server: Initializing the basic global model to obtain an initial global model, and determining an initial global unloading point set and an initial global unloading relationship graph, and using the global model, the initial unloading point set and the initial unloading relationship graph as global information; Broadcasting the global information to each bicycle node; Receiving local information sent by each of the bicycle nodes, the local information including local computing time, local model, local unloading point and local unloading relationship graph; Based on each of the local information, the global information is aggregated and updated, and the global information is broadcasted back to each of the bicycle nodes to continue the execution until the global model converges or reaches a preset accuracy, and the global model obtained at this time is used as the target global model; Based on the target global model, an unloading point set and an unloading relationship graph are updated to obtain updated global information; Broadcasting the updated global information to each of the bicycle nodes, and using the updated global information to perform heterogeneous data task processing; Wherein, the updating of the unloading point set and the unloading relationship graph based on the target global model to obtain updated global information includes: Rollback parameter set from actual send Take out the elements one by one , and actually receive the rollback parameter set Search in, is the actual rollback parameter with a positive sign, i is the index, representing the i-th bicycle node; If the rollback parameter amount is actually received There are elements in satisfy , then according to Update the unloading point, k is the index, representing the kth bicycle node; is the actual rollback parameter amount of the bicycle node i in the actual sent rollback parameter amount set G in round t+1; is the average rollback parameter of all on-board nodes participating in the calculation; It is the element in the set G of rollback parameters actually sent. The size of the local data set, n is the total number of bicycle nodes; Build elements based on the updated uninstall point With elements New uninstall diagram ; like , the element With elements Add the relationship to , from the actual send rollback parameter quantity collection Remove elements from , and calculate the remaining actual rollback parameters , and the remaining actual rollback parameter amount Add the actual amount of received rollback parameters middle; If the rollback parameter amount is actually received No elements satisfying the condition were found , then according to Update the uninstall point and update the uninstall diagram accordingly , The maximum actual rollback parameter amount in the actually sent rollback parameter amount set G in round t+1.

2. The method for processing heterogeneous data in an Internet of Vehicles according to claim 1, characterized in that: The step of initializing the basic global model to obtain an initial global model and determining an initial global unloading point set and an initial global unloading relationship diagram includes: In round When , the operating points of all bicycle nodes are located in the last layer of the hidden layer of the neural network, and the set of initialized unloading points is , The set of uninstall points is The position parameter of the operating point of the kth bicycle node in the initial round; n is the total number of bicycle nodes, and M is the depth of the hidden layer of the model; With the round As the number of nodes increases, the position of the operating point of each bicycle node changes dynamically. In the example, the operation point of the bicycle node moves to the layer, , at this time, for round The global unloading point set of the operating point is , The global uninstall point set is The kth element in .

3. The method for processing heterogeneous data in an Internet of Vehicles according to claim 1, characterized in that: The step of initializing the basic global model to obtain an initial global model and determining an initial global unloading point set and an initial global unloading relationship diagram further includes: In round When the initial uninstallation relationship graph is a The matrix , where n is the total number of bicycle nodes; With the round As the number of unloading nodes increases, the unloading relationship of each bicycle node changes dynamically. In the bicycle node Choose to offload the task to the bicycle node , then for round The uninstall relationship diagram in is in, Represents a bicycle node To the bicycle node The unloading parameter quantity of each bicycle node When acting as an unloading source, it can unload to at most one target bicycle node, satisfying , , each bicycle node As the unloading target, it receives unloading from multiple bicycle nodes, satisfying .

4. A method for processing heterogeneous data in an Internet of Vehicles, characterized in that: Applied to a distributed cluster, the distributed cluster includes a central server and multiple single-vehicle nodes, and the vehicle network heterogeneous data processing method includes the following steps performed by the single-vehicle node: Receiving global information broadcasted by a central server, wherein the global information includes a global model, an initial unloading point set, and an initial unloading relationship graph; Based on the global information, the local model is trained and updated according to the forward propagation of parameter unloading and the back propagation based on resource unloading to obtain an updated local model, and local information in the training process is determined, wherein the local information includes local computing time, local model, local unloading point and local unloading relationship diagram; Sending local information to the central server, so that the central server aggregates and updates the global information based on each of the local information to obtain a target global model, and updates the unloading point set and the unloading relationship graph based on the target global model to obtain updated global information; Receiving updated global information broadcasted by the central server, and using the updated global information to perform task processing of heterogeneous data; Wherein, the updating of the unloading point set and the unloading relationship graph based on the target global model to obtain updated global information includes: Rollback parameter set from actual send Take out the elements one by one , and actually receive the rollback parameter set Search in, is the actual rollback parameter with a positive sign, i is the index, representing the i-th bicycle node; If the rollback parameter amount is actually received There are elements in satisfy , then according to Update the unloading point, k is the index, representing the kth bicycle node; is the actual rollback parameter amount of the bicycle node i in the actual sent rollback parameter amount set G in round t+1; is the average rollback parameter of all on-board nodes participating in the calculation; It is the element in the set G of rollback parameters actually sent. The size of the local data set, n is the total number of bicycle nodes; Build elements based on the updated uninstall point With elements New uninstall diagram ; like , the element With elements Add the relationship to , from the actual send rollback parameter quantity collection Remove elements from , and calculate the remaining actual rollback parameters , and the remaining actual rollback parameter amount Add the actual amount of received rollback parameters middle; If the rollback parameter amount is actually received No elements satisfying the condition were found , then according to Update the uninstall point and update the uninstall diagram accordingly , The maximum actual rollback parameter amount in the rollback parameter amount set G in round t+1 is actually sent; wherein, based on the global information, the local model is trained and updated according to the forward propagation of parameter unloading and the reverse propagation based on resource unloading, and the updated local model is obtained, including: In the local forward propagation phase before the operation point, execution is performed from the input layer to the layer where the operation point is located. The forward propagation of the layer obtains fragmented data , Represents a bicycle node The model in round Middle The forward propagation function of the layer, symbol It represents the function combination between layers in a neural network, which converts the output of the previous layer into the input of the next layer. In the forward propagation phase unloaded between the operation point and the output layer, if the unloading relationship graph Elements in , then the bicycle node Break it down Send to the bicycle node ; Bicycle Node from The layer continues to perform forward propagation and obtains fragmented data , Represents a bicycle node Receive from the bicycle node Broken data And in its model Layer to The result of forward propagation between layers; If each bicycle node uses fragmented data received from other clients , continue to perform forward propagation to the output layer and get the final output M is the depth of the model’s hidden layer.

5. The method for processing heterogeneous data in an Internet of Vehicles according to claim 4, characterized in that: The training and updating of the local model based on the global information according to the forward propagation of parameter unloading and the reverse propagation based on resource unloading to obtain the updated local model also includes: In each round of communication, a backpropagation process is performed after forward propagation to calculate the gradient according to the loss function and update the weights of the network. The backpropagation process includes backpropagation of the local output layer, unloading backpropagation, and local backpropagation before the operation point.

6. A vehicle network heterogeneous data processing device, characterized in that: The vehicle network heterogeneous data processing device includes a central server and multiple single-vehicle nodes; the central server includes an initialization module, a first broadcast module, a node information receiving module, a global model training module, a global information update module, and a second broadcast module, and each of the single-vehicle nodes includes a local model training module, a local information sending module, and a second receiving module; wherein, An initialization module is used for the central server to initialize the basic global model, obtain the initial global model, and determine the initial global unloading point set and the initial global unloading relationship diagram, and use the global model, the initial unloading point set and the initial unloading relationship diagram as global information; A first broadcast module, used for the central server to broadcast the global information to each bicycle node; A first receiving module, configured for each of the bicycle nodes to receive global information broadcasted by the central server, wherein the global information includes a global model, an initial unloading point set, and an initial unloading relationship diagram; A local model training module is used for each of the single-vehicle nodes to train and update the local model based on the global information, according to the forward propagation of parameter unloading and the reverse propagation based on resource unloading, to obtain an updated local model, and to determine the local information in the training process, wherein the local information includes the local computing time, the local model, the local unloading point and the local unloading relationship diagram; A local information sending module, used for each of the bicycle nodes to send local information to the central server; A node information receiving module, used for the central server to receive local information sent by each of the single-vehicle nodes, wherein the local information includes local calculation time, local model, local unloading point and local unloading relationship diagram; A global model training module, which is used for the central server to aggregate and update the global information based on each of the local information, and return to broadcast the global information to each of the bicycle nodes to continue execution until the global model converges or reaches a preset accuracy, and the global model obtained at this time is used as the target global model; A global information updating module, used for the central server to update the unloading point set and the unloading relationship diagram based on the target global model to obtain updated global information; A second broadcast module, configured for the central server to broadcast the updated global information to each of the bicycle nodes, and to use the updated global information to perform heterogeneous data task processing; A second receiving module is used for each of the bicycle nodes to receive the updated global information broadcasted by the central server, and use the updated global information to perform heterogeneous data task processing; Wherein, the global information updating module comprises: Rollback parameter set from actual send Take out the elements one by one , and actually receive the rollback parameter set Search in; If in There are elements in satisfy , then according to Update the uninstallation point; n represents the number of elements in the rollback parameter set G that is actually sent; is the actual rollback parameter amount of the vehicle-mounted node i in the set G at round t+1; is the average rollback parameter of all on-board nodes participating in the calculation; is the local dataset size of the vehicle-mounted node gi in the set G; Based on the updated uninstall point, build and New uninstall diagram ; like ,Will and The relationship is added to ,from Remove , and calculate the remaining actual rollback parameters , and join in middle; If in No results were found that met the criteria , then according to Update the uninstall point and update the uninstall diagram accordingly ; Wherein, the local model training module includes: In the local forward propagation phase before the operation point, execution is performed from the input layer to the layer where the operation point is located. The forward propagation of the layer obtains fragmented data Represents a bicycle node The model in round Middle The forward propagation function of the layer, symbol It represents the function combination between layers in a neural network, which converts the output of the previous layer into the input of the next layer. In the forward propagation phase unloaded between the operation point and the output layer, if the unloading relationship graph Elements in , then the bicycle node Break it down Send to the bicycle node ; Bicycle Node from The layer continues to perform forward propagation and obtains fragmented data Represents a bicycle node Receive from the bicycle node Broken data And in its model Layer to The result of forward propagation between layers; If each bicycle node uses fragmented data received from other clients , continue to perform forward propagation to the output layer and get the final output M is the depth of the model’s hidden layer.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for processing heterogeneous data in the Internet of Vehicles as described in any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for processing heterogeneous data in the Internet of Vehicles as described in any one of claims 1 to 5 is implemented.