Highway data sharing method and device based on federated learning and electronic equipment

Abstract

The present application relates to the technical field of intelligent transportation, in particular to a highway data sharing method and device based on federated learning and electronic equipment. The method is applied to any edge server in a multi-layer distributed intelligent transportation system, comprising: receiving a data sharing request sent by a roadside sensing device within the coverage range of the edge server; analyzing the data sharing request to obtain the data category; if the data category is safety, using a decentralized federated learning architecture for model training to obtain a trained first global model, and sending the first global model to the corresponding roadside sensing device; if the data category is not safety, using a centralized federated learning architecture for model training to obtain a trained second global model, and sending the second global model to the corresponding roadside sensing device. The present application can improve the efficiency of highway data sharing and ensure data security.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and more specifically, to a method, apparatus, and electronic device for highway data sharing based on federated learning. Background Technology

[0002] As the construction of smart highway projects accelerates across provinces, roadside sensing facilities have collected a large amount of vehicle and traffic operation data. This has led to a massive accumulation of data on edge computing servers and cloud servers. Data sharing is an indispensable link in realizing intelligent road traffic and improving road traffic and management efficiency. However, directly sharing raw data not only increases the burden on communication links but also poses the risk of data leakage. Therefore, an efficient method for highway data sharing is urgently needed. Summary of the Invention

[0003] The purpose of this invention is to provide a highway data sharing method, device, and electronic device based on federated learning, which solves the problems of data transmission redundancy and low information acquisition efficiency in smart highway data sharing through a federated learning architecture.

[0004] In a first aspect, the present invention provides a highway data sharing method based on federated learning, the method being applied to any edge server in a multi-layered distributed intelligent transportation system; the method comprising:

[0005] Receive data sharing requests sent by roadside sensing devices within the coverage area of ​​the edge server;

[0006] The data sharing request is parsed to obtain the data category;

[0007] If the data category is security, a decentralized federated learning architecture is used to train the model to obtain the first global model after training, and the first global model is sent to the corresponding roadside perception device.

[0008] If the data category is not a security category, a centralized federated learning architecture is used to train the model, resulting in a trained second global model, which is then sent to the corresponding roadside sensing device.

[0009] In some embodiments, the step of training the model using a decentralized federated learning architecture to obtain the trained first global model includes:

[0010] The first initial global model of the safety category is sent to the corresponding roadside perception device, so that the roadside perception device can train the first initial global model based on local data to obtain the first local model, and then send the first local model to the edge server.

[0011] The system receives a first local model sent by roadside sensing devices within the coverage area of ​​the edge server, and aggregates the first local models sent by each roadside sensing device to obtain a first updated global model.

[0012] If the first updated global model converges, then the first updated global model is determined as the first global model;

[0013] If the first updated global model fails to converge, the first updated global model is sent to the corresponding roadside perception device so that the roadside perception device can perform local training again.

[0014] As an example, the aggregation of the first local models sent by each roadside sensing device includes:

[0015] The model parameters corresponding to the first local model sent by each roadside sensing device are averaged.

[0016] In some embodiments, the step of training the model using a centralized federated learning architecture to obtain a trained second global model includes:

[0017] The second initial global model of the non-safety class is sent to the corresponding roadside perception device, so that the roadside perception device can train the second initial global model based on local data to obtain the trained second local model, and then send the second local model to the edge server.

[0018] The system receives a second local model sent by roadside sensing devices within the coverage area of ​​the edge server, and aggregates the second local models sent by each roadside sensing device to obtain an updated local model.

[0019] The updated local model is sent to the cloud server, which then aggregates the updated local models sent by each edge server to obtain a second updated global model. The second updated global model and its convergence status are then sent to the edge server.

[0020] Receive the second updated global model and its convergence status sent by the cloud server;

[0021] If the second updated global model converges, then the second updated global model is determined as the second global model;

[0022] If the second updated global model fails to converge, the second updated global model is sent to the corresponding roadside perception device so that the roadside perception device can perform local training again.

[0023] As an example, the cloud server aggregates the updated local models sent by each edge server in an averaging manner.

[0024] In some embodiments, the roadside sensing devices within the coverage area of ​​the edge server include roadside equipment with data acquisition, storage, computing, and communication functions.

[0025] As an example, the non-safety classes include early warning classes and / or planning classes.

[0026] A second aspect of the present invention provides a highway data sharing device based on federated learning, the device being configured on any edge server in a multi-layered distributed intelligent transportation system; the device includes:

[0027] The receiving module is used to receive data sharing requests sent by roadside sensing devices within the coverage area of ​​the edge server;

[0028] The parsing module is used to parse the data sharing request and obtain the data category;

[0029] The first training module is used to train the model using a decentralized federated learning architecture if the data category is security, to obtain the trained first global model, and to send the first global model to the corresponding roadside perception device.

[0030] The second training module is used to train the model using a centralized federated learning architecture if the data category is not a security category, to obtain a trained second global model, and then send the second global model to the corresponding roadside sensing device.

[0031] A third aspect of the present invention provides an electronic device including a processor and a memory storing a computer program, wherein the processor executes the program to implement the federated learning-based highway data sharing method described in the first aspect above.

[0032] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the highway data sharing method based on federated learning described in the first aspect.

[0033] The present invention provides a highway data sharing method, apparatus, and electronic device based on federated learning, which has the following beneficial effects:

[0034] In a multi-layered distributed intelligent transportation system, any edge server receives data sharing requests from roadside sensing devices within its coverage area. The server parses the data sharing requests to obtain the data category. If the data category is safety-related, a decentralized federated learning architecture is used to train the model, resulting in a first global model, which is then sent to the corresponding roadside sensing device. If the data category is not safety-related, a centralized federated learning architecture is used to train the model, resulting in a second global model, which is also sent to the corresponding roadside sensing device. This invention achieves efficient sharing of highway data through federated learning. By sharing data models instead of raw data, the scale of shared highway data is significantly compressed, reducing information redundancy and effectively improving data transmission and application efficiency between nodes, as well as security. Furthermore, for data sharing requests with a safety-related category, the decentralized federated learning architecture for model training further enhances data sharing security, reduces communication overhead, and meets the emergency handling needs of traffic safety incidents. For non-safety-related data sharing requests, a centralized federated learning architecture is used for model training, improving the model's generalization ability and ensuring the global nature of data sharing. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0036] Figure 1 This is a flowchart illustrating a highway data sharing method based on federated learning provided in an embodiment of the present invention.

[0037] Figure 2 This is a schematic diagram of the structure of a multi-layer distributed intelligent transportation system in an embodiment of the present invention;

[0038] Figure 3 A flowchart illustrating another highway data sharing method based on federated learning provided in an embodiment of the present invention;

[0039] Figure 4 A flowchart illustrating another highway data sharing method based on federated learning provided in an embodiment of the present invention;

[0040] Figure 5 A schematic diagram of the structure of a highway data sharing device based on federated learning, provided for an embodiment of the present invention;

[0041] Figure 6This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0042] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Figure 1 This is a flowchart illustrating a highway data sharing method based on federated learning, provided in an embodiment of the present invention. The method is applied to any edge server in a multi-layered distributed intelligent transportation system. In some embodiments, such as... Figure 2 As shown, the multi-layered distributed intelligent transportation system includes a center cloud server (CCS), multiple edge computing servers (ECS), and roadside sensing equipment (RSE) within the coverage area of ​​each edge server. The roadside sensing equipment includes devices equipped along the highway with data acquisition, storage, computing, and communication capabilities, such as smart cameras, radar, and radar-visual fusion devices. These devices can collect vehicle and road operation information and communicate with the ECS via wireless or wired transmission. Edge computing servers can be small servers deployed on the roadside or in the median strip, possessing storage, communication, and computing capabilities. They can accept data sharing requests initiated by the RSE and complete the model aggregation task of federated learning within their own coverage area. The center cloud server, deployed at the backend, possesses powerful storage and computing capabilities. It can receive updated models from the ECS and aggregate them, achieving global model synchronization within the framework.

[0044] like Figure 1 As shown, the highway data sharing method based on federated learning in this embodiment of the invention includes the following steps:

[0045] Step 101: Receive data sharing requests from roadside sensing devices within the coverage area of ​​the edge server.

[0046] Step 102: Parse the data sharing request to obtain the data category.

[0047] In some embodiments, when there is a need for data sharing, the roadside sensing device will send a data sharing request to the edge server in its area. The data sharing request may include the data category, as well as the type of the roadside sensing device, the coordinate information of the roadside sensing device, etc.

[0048] The data category is used to characterize the type of data to be shared. For example, it can include safety and non-safety categories. The safety category refers to data related to vehicle collision safety, merging area safety, etc. The non-safety category can include early warning and / or planning categories. The early warning category refers to data related to traffic incident early warning, traffic congestion early warning, etc., while the planning category refers to data related to traffic flow prediction, route planning, road network status simulation, etc.

[0049] Step 103: If the data category is security, a decentralized federated learning architecture is used to train the model to obtain the first global model after training, and the first global model is sent to the corresponding roadside perception device.

[0050] It is understandable that data sharing needs related to safety have the highest priority in order to ensure the emergency handling of traffic safety incidents. Therefore, for data sharing needs related to safety, a decentralized federated learning architecture is used for model training to improve sharing efficiency.

[0051] In some embodiments, a decentralized federated learning architecture refers to a model training process that does not require the participation of cloud servers, but is performed solely by edge servers and corresponding roadside perception devices to obtain a global model.

[0052] In some embodiments, roadside sensing devices can collect vehicle and road operating status information, including the number of vehicles, vehicle type, license plate number, traffic flow, average speed, and vehicle occupancy rate. In the decentralized federated learning architecture, the roadside sensing devices train their models locally based on the road data they collect, obtaining a trained local model. The edge server then aggregates the local models sent by each roadside sensing device to update the global model until the global model converges. Finally, the global model is sent to the corresponding roadside sensing device. In other words, the decentralized federated learning architecture only involves data model sharing between roadside sensing devices and edge servers, not the sharing of raw data. This improves sharing efficiency and avoids data leakage issues during raw data transmission.

[0053] Step 104: If the data category is not security, a centralized federated learning architecture is used to train the model to obtain the trained second global model, and the second global model is sent to the corresponding roadside sensing device.

[0054] Due to the need for non-security-related data sharing, it is necessary to ensure the global nature of data sharing in order to improve its efficiency and effectiveness. Therefore, for non-security-related data sharing needs, a centralized federated learning architecture is adopted to ensure the global nature of the shared data.

[0055] In some embodiments, the centralized federated learning architecture requires the participation of three layers of nodes: roadside sensing devices, edge servers, and cloud servers. The roadside sensing devices train the model locally based on the data they collect, and obtain the trained local model. The edge servers aggregate the local models sent by the roadside sensing devices and send the aggregated local models to the cloud servers. The cloud servers then aggregate the local models sent by each edge server to update the global model. This process is repeated until the final global model converges.

[0056] It is evident that the centralized federated learning architecture only involves the sharing of data models among roadside sensing devices, edge servers, and cloud servers, without involving the sharing of raw highway data. This can greatly improve data sharing efficiency, ensure the global nature of data sharing, and facilitate data application.

[0057] It should be noted that the global model obtained through the federated learning framework can be a vehicle collision safety warning model, a traffic event warning model, a traffic congestion warning model, a traffic flow prediction model, a route planning model, etc., to meet the corresponding data sharing needs. The global model can be any neural network model, machine learning model, etc.

[0058] According to the federated learning-based highway data sharing method provided by embodiments of the present invention, any edge server in a multi-layered distributed intelligent transportation system receives data sharing requests sent by roadside sensing devices within its coverage area; the data sharing request is parsed to obtain the data category; if the data category is safety-related, a decentralized federated learning architecture is used for model training to obtain a trained first global model, which is then sent to the corresponding roadside sensing device; if the data category is not safety-related, a centralized federated learning architecture is used for model training to obtain a trained second global model, which is then sent to the corresponding roadside sensing device. This invention achieves efficient sharing of highway data through federated learning. By sharing data models instead of raw data, the scale of shared highway data is significantly compressed, reducing information redundancy and effectively improving data transmission and application efficiency and security between nodes. Furthermore, for data sharing requests with a safety-related data category, the decentralized federated learning architecture for model training further enhances the security of data sharing, reduces communication overhead, and meets the emergency handling needs of traffic safety incidents. For non-safety-related data sharing requests, a centralized federated learning architecture for model training improves the model's generalization ability and ensures the globality of data sharing.

[0059] Next, we will provide a detailed introduction to the model training process of the decentralized federated learning architecture.

[0060] Figure 3 This is a flowchart illustrating another highway data sharing method based on federated learning provided in an embodiment of the present invention. Figure 3 As shown, based on the above embodiments, Figure 1 Step 103, which uses a decentralized federated learning architecture to train the model and obtain the first global model after training, includes the following specific implementation process:

[0061] Step 301: The first initial global model of the safety category is sent to the corresponding roadside perception device, so that the roadside perception device can train the first initial global model based on local data to obtain the first local model, and then send the first local model to the edge server.

[0062] The first initial global model can be a training model initialized by the cloud server, while the first initial global model stored on the edge server is issued by the cloud server.

[0063] Step 302: Receive the first local model sent by the roadside sensing devices within the coverage area of ​​the edge server, and aggregate the first local models sent by each roadside sensing device to obtain the first updated global model.

[0064] In some embodiments, the edge server may set up a first model cache queue. After receiving the first local model sent by the roadside sensing device, the edge server puts the first local model into the first model cache queue; determines whether the current first model cache queue is overflowing. If the current first model cache queue is overflowing, it aggregates all the first local models in the first model cache queue and clears the first model cache queue; if the current first model cache queue is not overflowing, it continues to wait for other roadside sensing devices to send the first local model.

[0065] As one possible implementation, aggregating the first local models sent by each roadside sensing device includes averaging the model parameters corresponding to the first local models sent by each roadside sensing device.

[0066] Step 303: If the first updated global model converges, then the first updated global model is determined as the first global model.

[0067] In some embodiments, convergence can be determined based on the mean squared error. As an example, when the roadside sensing device sends the first local model, the data sent actually includes the model parameters of the first local model and the mean squared error of the first local model. When the edge server aggregates the various first local models, it also calculates the average mean squared error of each first local model to obtain the mean mean squared error of the first updated global model. If the mean squared error of the first updated global model is less than a threshold, it is determined that the first updated global model has converged.

[0068] Step 304: If the first updated global model fails to converge, the first updated global model is sent to the corresponding roadside perception device so that the roadside perception device can perform local training again.

[0069] In other words, if the first updated global model fails to converge, the first updated global model is sent to the corresponding roadside perception device to start the next round of iterative training. Steps 301 to 304 are introduced as an example of a round of training in a decentralized federated learning framework.

[0070] It should be noted that if the first updated global model does not converge, the process returns to step 301, but the first initial global model in step 301 is replaced with the first updated global model, and the next round of model training begins.

[0071] In other words, the training process in each round of the decentralized federated learning framework includes: sending the first updated global model after the i-th round of training to the corresponding roadside perception device, so that the roadside perception device can train the first updated global model after the i-th round of training based on local data to obtain the first local model after the (i+1)-th round of training, and sending the first local model after the (i+1)-th round of training to the edge server; receiving the first local models after the (i+1)-th round of training sent by the roadside perception devices within the coverage area of ​​the edge server, and aggregating the first local models after the (i+1)-th round of training sent by each roadside perception device to obtain the first updated global model after the (i+1)-th round of training; if the first updated global model after the (i+1)-th round of training converges, then the first updated global model after the (i+1)-th round of training is determined as the first global model; if the first updated global model after the (i+1)-th round of training does not converge, the first updated global model after the (i+1)-th round of training is sent to the corresponding roadside perception device to start the model training for the (i+2)-th round. Wherein, when i=0, the first updated global model after the i-th round of training is the first initial global model.

[0072] According to an embodiment of the federated learning-based highway data sharing method of the present invention, a first initial global model of safety categories is sent to the corresponding roadside sensing devices, which then train the first initial global model based on local data to obtain a trained first local model, which is then sent to an edge server. The edge server receives first local models sent by roadside sensing devices within its coverage area and aggregates these models to obtain a first updated global model. If the first updated global model converges, it is determined as the first global model. If the first updated global model does not converge, it is sent to the corresponding roadside sensing devices for further local training. This invention, through a decentralized federated learning framework, enables timely feedback on safety-related data sharing needs, further improving the efficiency of safety-related highway data sharing.

[0073] Next, we will provide a detailed introduction to the model training process of the centralized federated learning framework.

[0074] Figure 4 This is a flowchart illustrating another highway data sharing method based on federated learning provided in an embodiment of the present invention. Figure 4 As shown, based on the above embodiments, Figure 1 Step 104, which uses a centralized federated learning architecture to train the model and obtain the trained second global model, includes the following specific implementation process:

[0075] Step 401: Send the second initial global model (non-safety type) to the corresponding roadside sensing device so that the roadside sensing device can train the second initial global model based on local data to obtain the trained second local model, and then send the second local model to the edge server.

[0076] The second initial global model can be a training model initialized by the cloud server, while the second initial global model stored on the edge server is issued by the cloud server.

[0077] Step 402: Receive the second local model sent by the roadside sensing devices within the coverage area of ​​the edge server, and aggregate the second local models sent by each roadside sensing device to obtain the updated local model.

[0078] In some embodiments, the edge server can set up a second model cache queue. After receiving the second local model sent by the roadside sensing device, the edge server puts the second local model into the second model cache queue; it determines whether the current second model cache queue is overflowing. If the current second model cache queue is overflowing, it aggregates all the second local models in the second model cache queue and clears the second model cache queue; if the current second model cache queue is not overflowing, it continues to wait for other roadside sensing devices to send second local models.

[0079] As one possible implementation, aggregating the second local models sent by each roadside sensing device includes averaging the model parameters corresponding to each second local model sent by the roadside sensing device. As an example, the second local models sent by each roadside sensing device include model parameters and the mean squared error of the model. The edge server averages the model parameters of each second local model and averages the mean squared errors of each second local model to obtain the updated model parameters and their mean squared errors for the local model.

[0080] Step 403: The updated local model is sent to the cloud server, so that the cloud server can aggregate the updated local models sent by each edge server to obtain the second updated global model, and send the second updated global model and its convergence status to the edge server.

[0081] In some embodiments, the cloud server can set up a model cache queue. After receiving the updated local model sent by the edge server, the cloud server puts the updated local model into the model cache queue. It then determines whether the current model cache queue has overflowed. If the current model cache queue has overflowed, all the updated local models in the model cache queue are aggregated and the model cache queue is cleared. If the current model cache queue has not overflowed, it continues to wait for other edge servers to send updated local models.

[0082] As one possible implementation, the cloud server aggregates the updated local models sent by each edge server by averaging the model parameters corresponding to each updated local model sent by the edge server. As an example, each edge server sends the updated local model parameters and mean squared error to the cloud server. The cloud server averages the model parameters and mean squared errors of each updated local model to obtain the model parameters and mean squared error of the second updated global model.

[0083] The convergence of the second updated global model can be determined based on its mean square error. If the mean square error of the second updated global model is less than a threshold, then the second updated global model is considered to have converged.

[0084] Step 404: Receive the second updated global model and its convergence status sent by the cloud server.

[0085] Step 405: If the second updated global model converges, then the second updated global model is determined as the second global model.

[0086] Step 406: If the second updated global model fails to converge, the second updated global model is sent to the corresponding roadside perception device for local training again by the roadside perception device.

[0087] In other words, if the second updated global model fails to converge, the second updated global model is sent to the corresponding roadside perception device to start the next round of iterative training. Steps 401 to 406 are introduced as an example of a round of training in a centralized federated learning framework.

[0088] It should be noted that if the second updated global model does not converge, the process returns to step 401, but the second initial global model in step 401 is replaced with the second updated global model, thus starting the next round of model training.

[0089] In other words, the training process in each round of the decentralized federated learning framework includes: sending the second updated global model after the i-th round of training to the corresponding roadside perception device, so that the roadside perception device can train the second updated global model after the i-th round of training based on local data to obtain the second local model after the (i+1)-th round of training, and sending the second local model after the (i+1)-th round of training to the edge server; receiving the second local models after the (i+1)-th round of training sent by the roadside perception devices within the coverage area of ​​the edge server, and aggregating the second local models after the (i+1)-th round of training sent by each roadside perception device to obtain the second local model after the (i+1)-th round of training. The local model after training for the (i+1)th round is sent to the cloud server. The cloud server aggregates the local models from each edge server to obtain the second updated global model after the (i+1)th round of training. If the second updated global model after the (i+1)th round of training converges, it is designated as the second global model. If the second updated global model after the (i+1)th round of training does not converge, it is sent to the corresponding roadside perception device to begin the (i+2)th round of model training. When i=0, the second updated global model after the i-th round of training is the second initial global model.

[0090] It should be noted that if non-security-related data sharing needs include multiple data categories, models for each data category can be trained separately based on a centralized federated learning model.

[0091] According to the federated learning method for highway data sharing based on embodiments of the present invention, for non-safety-related data sharing needs, a centralized federated learning architecture is used for model training to obtain a second global model. This not only improves the efficiency of highway data sharing and avoids the leakage of original data, but also ensures the globality of data sharing and improves the availability of highway data sharing.

[0092] To implement the above embodiments, the present invention also provides a highway data sharing device based on federated learning.

[0093] Figure 5 This is a schematic diagram of a highway data sharing device based on federated learning, provided as an embodiment of the present invention. It should be noted that this device is configured on any edge server within a multi-layered distributed intelligent transportation system. Figure 5 As shown, the device includes a receiving module 510, a parsing module 520, a first training module 530, and a fourth training module 540.

[0094] The receiving module 510 is used to receive data sharing requests sent by roadside sensing devices within the coverage area of ​​the edge server.

[0095] Parsing module 520 is used to parse data sharing requests and obtain data categories;

[0096] The first training module 530 is used to train the model using a decentralized federated learning architecture if the data category is safety, to obtain the first global model after training, and to send the first global model to the corresponding roadside perception device.

[0097] The second training module 540 is used to train the model using a centralized federated learning architecture if the data category is not the safety category, to obtain the trained second global model, and then send the second global model to the corresponding roadside perception device.

[0098] In some embodiments, the first training module 530 is specifically used for:

[0099] The first initial global model of the safety category is sent to the corresponding roadside perception device, so that the roadside perception device can train the first initial global model based on local data to obtain the first local model, and then send the first local model to the edge server.

[0100] The system receives the first local model sent by the roadside sensing devices within the coverage area of ​​the edge server, and aggregates the first local models sent by each roadside sensing device to obtain the first updated global model.

[0101] If the first updated global model converges, then the first updated global model is determined as the first global model;

[0102] If the first updated global model fails to converge, the first updated global model will be sent to the corresponding roadside perception device for local training again.

[0103] As one possible implementation, the first training module 530 is also used for:

[0104] The model parameters corresponding to the first local model sent by each roadside sensing device are averaged.

[0105] In some embodiments, the second training module 540 is specifically used for:

[0106] The second initial global model of the non-safety class is sent to the corresponding roadside perception device, so that the roadside perception device can train the second initial global model based on local data to obtain the trained second local model, and then send the second local model to the edge server.

[0107] The system receives the second local model sent by the roadside sensing devices within the coverage area of ​​the edge server, and aggregates the second local models sent by each roadside sensing device to obtain the updated local model.

[0108] The updated local model is sent to the cloud server, which then aggregates the updated local models sent by each edge server to obtain the second updated global model. The second updated global model and its convergence status are then sent to the edge server.

[0109] Receive the second updated global model and its convergence status sent by the cloud server;

[0110] If the second updated global model converges, then the second updated global model is determined as the second global model;

[0111] If the second updated global model fails to converge, the second updated global model will be sent to the corresponding roadside perception device for local training again.

[0112] As an example, the cloud server aggregates the updated local models sent to each edge server by averaging them.

[0113] In some embodiments, roadside sensing devices include roadside equipment with data acquisition, storage, computing, and communication functions.

[0114] As an example, non-safety classes include early warning classes and / or planning classes.

[0115] It should be noted that the explanations and descriptions in the above embodiments of the highway data sharing method based on federated learning can also be applied to the highway data sharing device based on federated learning in the embodiments of the present invention, and will not be repeated here.

[0116] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6 As shown, the electronic device may include a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other via the communication bus 640. The processor 610 can call the computer program in the memory 630 to execute the steps of the federated learning-based highway data sharing method provided in the above embodiments.

[0117] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0118] On the other hand, embodiments of the present invention also provide a computer program product, the computer program product including a computer program that can be stored on a computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to perform the steps of the highway data sharing method based on federated learning provided in the above embodiments.

[0119] On the other hand, embodiments of the present invention also provide a non-transitory computer-readable storage medium having a computer program stored thereon, the computer program being used to cause a processor to execute the highway data sharing method based on federated learning provided in the above embodiments.

[0120] The non-transitory computer-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0121] The device 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 any creative effort.

[0122] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence 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 computer-readable 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 the various embodiments or some parts of the embodiments.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A highway data sharing method based on federated learning, characterized in that, The method is applied to any edge server in a multi-layered distributed intelligent transportation system; the method includes: Receive data sharing requests sent by roadside sensing devices within the coverage area of ​​the edge server; The data sharing request is parsed to obtain the data category; If the data category is security, a decentralized federated learning architecture is used to train the model to obtain the first global model after training, and the first global model is sent to the corresponding roadside perception device. If the data category is not a security category, a centralized federated learning architecture is used to train the model, resulting in a trained second global model, which is then sent to the corresponding roadside sensing device.

2. The method according to claim 1, characterized in that, The process of training the model using a decentralized federated learning architecture to obtain the first global model after training includes: The first initial global model of the safety category is sent to the corresponding roadside perception device, so that the roadside perception device can train the first initial global model based on local data to obtain the first local model, and then send the first local model to the edge server. The system receives a first local model sent by roadside sensing devices within the coverage area of ​​the edge server, and aggregates the first local models sent by each roadside sensing device to obtain a first updated global model. If the first updated global model converges, then the first updated global model is determined as the first global model; If the first updated global model fails to converge, the first updated global model is sent to the corresponding roadside perception device so that the roadside perception device can perform local training again.

3. The method according to claim 2, characterized in that, The aggregation of the first local models sent by each roadside sensing device includes: The model parameters corresponding to the first local model sent by each roadside sensing device are averaged.

4. The method according to claim 1, characterized in that, The method employs a centralized federated learning architecture for model training, resulting in a trained second global model, including: The second initial global model of the non-safety class is sent to the corresponding roadside perception device, so that the roadside perception device can train the second initial global model based on local data to obtain the trained second local model, and then send the second local model to the edge server. The system receives a second local model sent by roadside sensing devices within the coverage area of ​​the edge server, and aggregates the second local models sent by each roadside sensing device to obtain an updated local model. The updated local model is sent to the cloud server, which then aggregates the updated local models sent by each edge server to obtain a second updated global model. The second updated global model and its convergence status are then sent to the edge server. Receive the second updated global model and its convergence status sent by the cloud server; If the second updated global model converges, then the second updated global model is determined as the second global model; If the second updated global model fails to converge, the second updated global model is sent to the corresponding roadside perception device so that the roadside perception device can perform local training again.

5. The method according to claim 4, characterized in that, The cloud server aggregates the updated local models sent by each edge server by averaging them.

6. The method according to claim 1, characterized in that, Roadside sensing equipment includes devices equipped along highways that have data acquisition, storage, computing, and communication functions.

7. The method according to claim 1, characterized in that, The non-safety categories include early warning categories and / or planning categories.

8. A highway data sharing device based on federated learning, characterized in that, The device is configured on any edge server in a multi-layered distributed intelligent transportation system; the device includes: The receiving module is used to receive data sharing requests sent by roadside sensing devices within the coverage area of ​​the edge server; The parsing module is used to parse the data sharing request and obtain the data category; The first training module is used to train the model using a decentralized federated learning architecture if the data category is security, to obtain the first global model after training, and to send the first global model to the corresponding roadside perception device. The second training module is used to train the model using a centralized federated learning architecture if the data category is not a security category, to obtain a trained second global model, and then send the second global model to the corresponding roadside sensing device.

9. An electronic device, characterized in that, It includes a processor and a memory storing a computer program, wherein the processor executes the program to implement the highway data sharing method based on federated learning as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the highway data sharing method based on federated learning as described in any one of claims 1 to 7.

Images