Multi-factor based incentive method for federated learning in Internet of Vehicles and related equipment

Abstract

The application discloses a multi-factor-based Internet of Vehicles federated learning incentive method and related equipment thereof. The method comprises the following steps: receiving a to-be-trained federated model corresponding to a training task published by a task publisher; sending the to-be-trained federated model to participating vehicles in a corresponding coverage range to participate in training, so that the participating vehicles perform federated training on the to-be-trained federated model by using their own databases, and upload model-related information obtained through training to a roadside unit. After the training of the to-be-trained federated model is completed, the obtained model is a target model. According to the model-related information uploaded by the participating vehicles, the contribution degrees of the participating vehicles to the target model are calculated, and corresponding incentives are sent to the participating vehicles. The contribution degrees include the contribution degrees of the participating vehicles to the target model in terms of quality factors, time factors and credit factors. The application aims to improve the rationality of incentives.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication, in particular to a multi-factor-based Internet of Vehicles federated learning incentive method and related equipment thereof. BACKGROUND

[0002] In recent years, the Internet of Vehicles (IoV) technology has developed greatly. Specifically, the Internet of Vehicles can realize intelligent interconnection between vehicles, people, roads and service platforms, and can also realize information interconnection between vehicles, that is, the Internet of Vehicles technology can enable people to monitor vehicles through remote terminals, and enable vehicles to monitor road conditions in real time and select appropriate driving routes.

[0003] It should be noted that a large amount of training data is required to realize the many functions of the Internet of Vehicles, and driving data usually involves user privacy, so the Internet of Intelligent Vehicles cannot be separated from federated learning. In the process of training a model by federated learning in the Internet of Intelligent Vehicles, the vehicles participating in the training process consume memory, communication and other resources. Without incentives and compensation, the enthusiasm of vehicles participating in training is not high, and they cannot obtain the data information required for training. Therefore, it is crucial to develop a reasonable incentive policy, and the incentive policy needs to be fair and reasonable, that is, to accurately reflect the contribution of participants. In related technologies, only the size of the local database of the vehicle and the training energy consumption are considered when determining the contribution of the participants, and factors such as data timeliness, security and reliability are ignored, resulting in unreasonable incentives. Unreasonable incentives can result in long training cycles for model training and difficulty in identifying unreliable vehicle data. SUMMARY

[0004] Therefore, the embodiments of the present application provide a multi-factor-based Internet of Vehicles federated learning incentive method and related equipment thereof, aiming to encourage vehicles to participate in Internet of Vehicles model training, taking into account factors such as data timeliness, security and reliability, improving the rationality of incentives, and avoiding technical problems such as long training cycles for model training and difficulty in identifying unreliable vehicle data caused by unreasonable incentives.

[0005] The embodiments of the present application provide a multi-factor-based Internet of Vehicles federated learning incentive method applied to a roadside unit, the method comprising:

[0006] receiving a training task published by a task publisher, and determining a to-be-trained federated model corresponding to the training task;

[0007] sending the to-be-trained federated model to participating vehicles in a corresponding coverage range that are selected to participate in training, so that the participating vehicles perform federated training on the to-be-trained federated model using their own databases, and upload model-related information obtained by training to the roadside unit after the training is completed, wherein the model obtained after the to-be-trained federated model is trained is a target model.

[0008] Based on the model-related information uploaded by the participating vehicles, the contribution of the participating vehicles to the target model is calculated, and the incentive corresponding to the contribution is sent to the participating vehicles, wherein the contribution includes the degree of contribution of the participating vehicles to the target model in terms of quality factors, time factors and credit factors.

[0009] In a possible implementation manner of the present application, the contribution degree includes the contribution degree of the participating vehicles to the target model in terms of quality factors;

[0010] The model-related information includes the change in the loss function after training and the model parameters;

[0011] The step of calculating the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles includes:

[0012] After determining the first change amount of the loss function corresponding to the global model of the federated model to be trained for the participating vehicles;

[0013] Determine a second change in the overall loss function of the global model corresponding to the roadside unit in the process of obtaining the target model;

[0014] determining a mass contribution of the participating vehicle to the target model according to the first change amount and the second change amount;

[0015] Among them, the quality contribution is in,

[0016]

[0017] Where F(ω) is the second change amount, and ΔFi(W) is the first change amount of the vehicle.

[0018] In a possible implementation manner of the present application, the contribution degree includes the contribution degree of the participating vehicles to the target model in terms of time factor;

[0019] The step of calculating the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles includes:

[0020] The time T at which the participating vehicles upload the model-related information i , calculating the time contribution of the participating vehicles to the target model;

[0021] Among them, the time contribution Where α represents a constant coefficient.

[0022] In a possible implementation of the present application, the contribution degree includes the contribution degree of the participating vehicle to the target model in terms of the first credit factor; the model-related information includes the number of samples D in the local database of the participating vehicle. i The time for the participating vehicles to upload the model related information is T i ;

[0023] The step of calculating the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles includes:

[0024] According to T i and the number of participating vehicle datasets D i Is the ratio of less than the credibility warning line β, calculate the first credit contribution of the participating vehicle to the target model, wherein the first credit contribution Hi satisfies

[0025]

[0026] Among them, for H i = 0 participating vehicles are marked, and the H i =0 participating vehicles are dishonest users.

[0027] In a possible implementation of the present application, the model-related information includes the number of samples D in the local database of participating vehicles. i ;

[0028] The contribution degree includes the contribution degree of the participating vehicle to the target model in terms of the second credit factor;

[0029] The step of calculating the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles includes:

[0030] Calculating a second credit contribution of the participating vehicle to the target model based on the roadside unit local model and samples in the local database; wherein the second credit contribution;

[0031]

[0032] Among them, (x k ,y k ) represents the local database D of participating vehicle i i The kth sample in x k Represents the input of the kth sample, y k represents the actual label value of the kth sample, f(x k ) represents the predicted value output after the k-th sample is input into the model obtained by training the corresponding participating vehicle.

[0033] In a possible implementation of the present application, the step of sending the incentive corresponding to the contribution to the participating vehicle includes:

[0034] Based on the contribution of participating vehicles, the reward weight of participating vehicles is determined, where the reward weight

[0035] Among them, a and b are constant weighting factors, is the quality contribution, S t (T i ) is the time contribution, S q (W i ) is the second credit contribution;

[0036] According to the reward weight, the incentive of the participating vehicle is determined and the incentive is sent to the participating vehicle, wherein the incentive

[0037]

[0038] Where P represents the total budget of the task publisher, N represents the total number of participating vehicles, and p i For motivation.

[0039] The present application also provides a multi-factor based vehicle network federated learning incentive method, which is applied to participating vehicles. The multi-factor based vehicle network federated learning incentive method includes:

[0040] Receive the federated model to be trained sent by the roadside unit;

[0041] Based on the local database, the federated model to be trained is federated trained, and after the training is completed, the model related information obtained by the training is uploaded to the roadside unit, wherein after the training of the federated model to be trained is completed, the model obtained by the roadside unit is the target model;

[0042] Receive local incentives determined by a roadside unit based on the model-related information, wherein the incentives are determined after the roadside unit determines the local contribution to the target model based on the model-related information, wherein the contribution includes the degree of local contribution to the target model in terms of quality factors, time factors, and credit factors.

[0043] The present application also provides a multi-factor based vehicle network federated learning incentive device, which is applied to a roadside unit, and the device includes:

[0044] A receiving module is used to receive a training task issued by a task publisher and determine the federated model to be trained corresponding to the training task;

[0045] The asynchronous federated training module is configured to send the to-be-trained federated model to the participating vehicles in the corresponding coverage range to participate in the training, so that the participating vehicles perform asynchronous federated training on the to-be-trained federated model by using the database of the participating vehicles, and upload the model-related information obtained after the training to the roadside unit. After the training of the to-be-trained federated model is completed, the obtained model is the target model.

[0046] The incentive module is configured to calculate the contribution degree of the participating vehicles to the target model according to the model-related information uploaded by the participating vehicles, and send the participating vehicles corresponding incentives according to the contribution degree. The contribution degree includes the contribution degree of the participating vehicles to the target model in the quality factor, the time factor and the credit factor.

[0047] The application also provides a multi-factor-based federated learning incentive device for Internet of Vehicles, which is an entity node device. The multi-factor-based federated learning incentive device for Internet of Vehicles comprises a memory, a processor, and a program of the multi-factor-based federated learning incentive method for Internet of Vehicles stored in the memory and executable on the processor. When the program of the multi-factor-based federated learning incentive method for Internet of Vehicles is executed by the processor, the steps of the multi-factor-based federated learning incentive method for Internet of Vehicles can be realized.

[0048] To achieve the above-mentioned purpose, a storage medium is also provided, which stores a multi-factor-based federated learning incentive program for Internet of Vehicles. When the multi-factor-based federated learning incentive program for Internet of Vehicles is executed by the processor, the steps of the multi-factor-based federated learning incentive method for Internet of Vehicles can be realized.

[0049] The present application provides a multi-factor based vehicle network federated learning incentive method and related equipment. Compared with the related technology in which the contribution of participants is determined by only considering the size of the vehicle's local database and training energy consumption, and ignoring factors such as data timeliness, security, and reliability, resulting in a long training cycle and difficulty in identifying unreliable vehicle data, in the present application, a training task issued by a task publisher is received, and the federated model to be trained corresponding to the training task is determined; the federated model to be trained is sent to participating vehicles that choose to participate in the training within the corresponding coverage area, so that the participating vehicles can use their own databases to perform federal training on the federated model to be trained, and upload the model related information obtained from the training to the roadside unit after the training is completed, wherein, after the training of the federated model to be trained is completed, the model obtained is the target model; based on the model related information uploaded by the participating vehicles, the participating vehicles' response to the target model is calculated. The contribution of the model is calculated, and the corresponding incentive of the contribution is sent to the participating vehicles, wherein the contribution includes the contribution of the participating vehicles to the target model in terms of quality factors, time factors and credit factors. It can be understood that in this application, after the roadside unit sends the federal model to be trained to the participating vehicles that choose to participate in the training within the corresponding coverage area, the contribution of the participating vehicles to the target model is calculated according to the model-related information uploaded by the participating vehicles, and the incentive is determined based on the contribution. When calculating the contribution of the participating vehicles to the target model, the contribution of the participating vehicles to the target model in terms of quality factors, time factors and credit factors is taken into account, rather than just the size of the vehicle's local database and training energy consumption. Therefore, the rationality of the incentive can be improved, and technical problems such as unreasonable incentives causing long training cycles for model training and difficulty in identifying unreliable vehicle data can be avoided. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flowchart of the first embodiment of the multi-factor-based vehicle network federated learning incentive method of this application;

[0051] Figure 2 This is a flowchart of the second embodiment of the multi-factor based vehicle network federated learning incentive method of this application;

[0052] Figure 3 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present application;

[0053] Figure 4 This is a schematic diagram of the scenarios involved in the multi-factor-based vehicle network federated learning incentive method of this application. DETAILED DESCRIPTION

[0054] It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0055] The embodiment of the present application provides a method for stimulating federated learning of Internet of Vehicles based on multiple factors. In the first embodiment of the method for stimulating federated learning of Internet of Vehicles based on multiple factors, Figure 1 , applied to a roadside unit, the method comprises:

[0056] Step S10: receiving a training task issued by a task publisher, and determining a federated model to be trained corresponding to the training task;

[0057] Step S20: Send the federated model to be trained to participating vehicles within the corresponding coverage area that have chosen to participate in the training, so that the participating vehicles can use their own databases to perform federated training on the federated model to be trained. After the training is completed, the model related information obtained by the training is uploaded to the roadside unit. After the training of the federated model to be trained is completed, the model obtained is the target model.

[0058] Step S30, based on the model-related information uploaded by the participating vehicles, calculate the contribution of the participating vehicles to the target model, and send the corresponding incentives for the contribution to the participating vehicles, wherein the contribution includes the degree of contribution of the participating vehicles to the target model in terms of quality factors, time factors and credit factors.

[0059] In this embodiment, the research and development background is:

[0060] In recent years, the Internet of Vehicles (IoV) technology has developed greatly. Specifically, IoV can realize the intelligent interconnection between vehicles and people, roads, and service platforms, and can also enable information exchange and interconnection between vehicles. That is, IoV technology can enable people to monitor vehicles through remote terminals, enable vehicles to monitor road conditions in real time, and choose appropriate driving routes.

[0061] It should be noted that realizing the numerous functions of the Internet of Vehicles requires a large amount of training data, and driving data usually involves user privacy. Therefore, the smart Internet of Vehicles cannot do without federated learning. In the process of training models through federated learning, the vehicles participating in the training process will consume memory, communication and other resources. In the absence of incentives and compensation, the enthusiasm of vehicles participating in the training is not high and they cannot obtain the data information required for training. Therefore, it is crucial to formulate reasonable incentive policies.

[0062] Currently, federated learning faces the problem of the source of training data. In today's world where data is wealth, in order to obtain sufficient high-quality training data, it is necessary to formulate a reasonable incentive mechanism to mobilize users' enthusiasm for participating in federated learning.

[0063] And the incentive policy needs to be fair and reasonable, that is, accurately reflecting the contribution of participants. In related technologies, the contribution of participants is determined only by considering the size of the local database of the vehicle and the training energy consumption, ignoring factors such as data timeliness, security, and reliability, resulting in unreasonable incentives, which will cause the training cycle of model training to be long and it is difficult to identify unreliable vehicle data.

[0064] Overall, in the embodiment, an incentive mechanism is provided, which stimulates the enthusiasm of participating vehicles for the task publisher, and ensures the stability of the training data source. For participating vehicles, the mechanism fairly measures the contribution of each participating vehicle, and is fair and just, so as to avoid the technical problems that unreasonable incentives will cause the training cycle of model training to be long and it is difficult to identify unreliable vehicle data.

[0065] The specific steps are as follows:

[0066] Step S10, receiving the training task published by the task publisher, determining the to-be-trained federated model corresponding to the training task;

[0067] As an example, the multi-factor-based vehicle networking federated learning incentive method is applied to a multi-factor-based vehicle networking federated learning incentive device. The multi-factor-based vehicle networking federated learning incentive method device belongs to a multi-factor-based vehicle networking federated learning incentive equipment, which belongs to a multi-factor-based vehicle networking federated learning incentive system. The multi-factor-based vehicle networking federated learning incentive equipment can be a roadside unit (RSU). The multi-factor-based vehicle networking federated learning incentive system also includes participating vehicles and task publishers, as shown in Figure 4 .

[0068] As an example, as shown in Figure 4 , the participating vehicles and the roadside unit exchange information in the form of transactions, Tr up is the data packet determined after the participating vehicle training ends, including the size of the local database, the model parameters after training, the loss function, and the local vehicle identifier. The data packet is uploaded to the roadside unit in the form of a transaction, and Tr pay is the incentive, that is, the reward.

[0069] As an example, as shown in Figure 4 , the communication relationship between the roadside unit and the task publisher is described.

[0070] As an example, the task publisher publishes the total amount of rewards to the roadside unit.

[0071] As an example, the task publisher sends the training task to the roadside unit, the roadside unit receives the training task issued by the task publisher, and then the roadside unit determines the federated model to be trained corresponding to the training task.

[0072] As an example, the federated model to be trained can be an asynchronous federated learning model. This is because:

[0073] Federated learning can be categorized into synchronous and asynchronous federated learning. Because participating vehicles have varying computing power, communication bandwidth, and other factors, training completion times are often asynchronous. In synchronous federated learning, local training time is determined by the longest-training vehicle. Each global model iteration requires significant time to wait for slower-training vehicles. Asynchronous federated learning avoids this wasted waiting time and improves training efficiency.

[0074] As an example, there may be multiple training tasks.

[0075] As an example, the federated models to be trained for different training tasks may be different.

[0076] As an example, after the roadside unit determines the federated model to be trained, it also initializes the model parameters of the model to be trained corresponding to the training task.

[0077] Step S20: Send the federated model to be trained to participating vehicles within the corresponding coverage area that have chosen to participate in the training, so that the participating vehicles can use their own databases to perform federated training on the federated model to be trained. After the training is completed, the model related information obtained by the training is uploaded to the roadside unit. After the training of the federated model to be trained is completed, the model obtained is the target model.

[0078] As an example, after receiving the training task, the roadside unit sends the training task to vehicles within the coverage area of ​​the roadside unit.

[0079] As an example, after receiving the training task, vehicles within the coverage area of ​​the roadside unit can choose to participate or not. The vehicles that choose to participate are participating vehicles, such as Figure 4 As shown, the participating vehicles communicate with the roadside unit, the roadside unit sends the federal model to be trained to the participating vehicles, and the participating vehicles use their own database (samples in the database) for training.

[0080] As an example, the participating vehicles communicate with the roadside unit, the roadside unit sends the federated model to be trained to the participating vehicles, and the participating vehicles use their own database (samples in the database) to perform asynchronous federated training.

[0081] As an example, after the roadside unit completes the training of the federated model, the model obtained is the target model.

[0082] As an example, the process of obtaining the target model by the roadside unit involves multiple interactions with participating vehicles.

[0083] As an example, after initially receiving the federated model to be trained, the participating vehicle uses its own database (samples in the database) to train it and trains it a certain number of times, and then obtains the intermediate model parameters of the intermediate model; the participating vehicle sends the intermediate model parameters of the intermediate model to the roadside unit (it can be understood that the roadside unit receives multiple intermediate model parameters corresponding to multiple participating vehicles), and then aggregates the multiple intermediate model parameters to obtain aggregated model parameters, and then sends the aggregated model parameters to each participating vehicle for each participating vehicle to train again until the target model is obtained.

[0084] As an example, the participating vehicles use their own databases to perform federal training on the federal model to be trained, and obtain model-related information.

[0085] Among them, model related information includes ω i ,F i (ω),D i ,id, where wi is the model parameter, Fi(w) is the loss function, Di is the size of the local database, and id is the identifier of the participating vehicle.

[0086] As an example, after the training of the participating vehicles is completed, the local database size, trained model parameters, loss function and identification of the participating vehicles are packaged to obtain a data packet, and then the data packet is uploaded to the roadside unit in the form of a transaction. Specifically, the data packet can be Tr up ,in,

[0087] Tr up ={D i ,ω i ,F i (ω),id}.

[0088] Step S30, based on the model-related information uploaded by the participating vehicles, calculate the contribution of the participating vehicles to the target model, and send the corresponding incentives for the contribution to the participating vehicles, wherein the contribution includes the degree of contribution of the participating vehicles to the target model in terms of quality factors, time factors and credit factors.

[0089] As an example, the roadside unit calculates the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles.

[0090] The contribution degree includes the contribution degree of the participating vehicles to the target model in terms of quality factors, time factors and credit factors.

[0091] As an example, the contribution degree also includes the contribution degree of memory, communication and other resources to the target model.

[0092] As an example, after the contribution is determined, an incentive corresponding to the contribution is determined, and the corresponding incentive is sent to the participating vehicles.

[0093] Before determining the incentive corresponding to the contribution, the reward weight Q(i) needs to be determined first.

[0094] Among them, the reward weight Q(i) is related to quality contribution, time contribution and credit contribution.

[0095] The contribution degree includes the contribution degree of the participating vehicles to the target model in terms of quality factors;

[0096] The model-related information includes the change in the loss function after training and the model parameters;

[0097] The step of calculating the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles includes:

[0098] Step S31, determining a first change in the loss function corresponding to the global model after the participating vehicle trains the federated model to be trained;

[0099] As an example, if the corresponding loss function of the participating vehicle before training is the first loss function value, and the corresponding loss function of the participating vehicle after training is the second loss function value, then the first change amount is the second loss function value minus the first loss function value.

[0100] Step S32, determining a second change in the overall loss function of the global model corresponding to the roadside unit in the process of obtaining the target model;

[0101] As an example, if before training, the loss function of the corresponding global model is the third loss function value, and after training, the corresponding global model is the fourth loss function value, then the first change amount is the fourth loss function value minus the third loss function value.

[0102] Step S33, determining the mass contribution of the participating vehicle to the target model according to the first change amount and the second change amount;

[0103] Among them, the quality contribution is in,

[0104]

[0105] Where F(ω) is the second change amount, and ΔFi(W) is the first change amount of the vehicle.

[0106] In this embodiment, the mass contribution of the participating vehicle to the target model is determined according to the first change amount and the second change amount.

[0107] The formula is

[0108] The present application provides a multi-factor based vehicle network federated learning incentive method and related equipment. Compared with the related technology in which the contribution of participants is determined by only considering the size of the vehicle's local database and training energy consumption, and ignoring factors such as data timeliness, security, and reliability, resulting in a long training cycle and difficulty in identifying unreliable vehicle data, in the present application, a training task issued by a task publisher is received, and the federated model to be trained corresponding to the training task is determined; the federated model to be trained is sent to participating vehicles that choose to participate in the training within the corresponding coverage area, so that the participating vehicles can use their own databases to perform federal training on the federated model to be trained, and upload the model related information obtained from the training to the roadside unit after the training is completed, wherein, after the training of the federated model to be trained is completed, the model obtained is the target model; based on the model related information uploaded by the participating vehicles, the participating vehicles' response to the target model is calculated. The contribution of the model is calculated, and the corresponding incentive of the contribution is sent to the participating vehicles, wherein the contribution includes the contribution of the participating vehicles to the target model in terms of quality factors, time factors and credit factors. It can be understood that in this application, after the roadside unit sends the federal model to be trained to the participating vehicles that choose to participate in the training within the corresponding coverage area, the contribution of the participating vehicles to the target model is calculated according to the model-related information uploaded by the participating vehicles, and the incentive is determined based on the contribution. When calculating the contribution of the participating vehicles to the target model, the contribution of the participating vehicles to the target model in terms of quality factors, time factors and credit factors is taken into account, rather than just the size of the vehicle's local database and training energy consumption. Therefore, the rationality of the incentive can be improved, and technical problems such as unreasonable incentives causing long training cycles for model training and difficulty in identifying unreliable vehicle data can be avoided.

[0109] Furthermore, based on the first embodiment of the present application, another embodiment of the present application is provided, in which the contribution degree includes the contribution degree of the participating vehicles to the target model in terms of time factor;

[0110] The step of calculating the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles includes:

[0111] Step S34: according to the time T of the participating vehicles uploading the model related information i , calculating the time contribution of the participating vehicles to the target model;

[0112] Among them, the time contribution Where α represents a constant coefficient.

[0113] As an example, the roadside unit obtains a time factor for incentive measurement based on the time it receives information uploaded by participating vehicles.

[0114] Among them, the time contribution Where α represents the constant coefficient

[0115] Where T i The time when the information related to the model is uploaded for the participating vehicles.

[0116] The contribution degree includes the contribution degree of the participating vehicle to the target model in terms of the first credit factor; the model-related information includes the number of samples D in the local database of the participating vehicle. i The time for the participating vehicles to upload the model related information is T i ;

[0117] The step of calculating the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles includes:

[0118] Step S35, according to T i and the number of participating vehicle datasets D i Is the ratio of β to β less than the credibility warning line β, calculate the first credit contribution of the participating vehicle to the target model, wherein the first credit contribution H i satisfy

[0119]

[0120] Among them, for H i = 0 participating vehicles are marked, and the H i =0 participating vehicles are dishonest users.

[0121] If the number of participating vehicle datasets is too large and T i If H is too small, i = 0, which is understandable (lots of data, but results are produced too quickly). In this case, the training is not accurate, meaning that the participating vehicles may be fraudulent. Therefore, H i = 0 are marked to avoid affecting the accuracy of model training.

[0122] The model-related information includes the number of samples in the local database of participating vehicles D i ;

[0123] The contribution degree includes the contribution degree of the participating vehicle to the target model in terms of the second credit factor;

[0124] The step of calculating the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles includes:

[0125] Step S36, calculating the second credit contribution of the participating vehicle to the target model based on the roadside unit local model and the samples in the local database; wherein the second credit contribution

[0126]

[0127] Among them, (x k ,y k ) represents the local database D of participating vehicle i i The kth sample in x k Represents the input of the kth sample, y k represents the actual label value of the kth sample, f(x k ) represents the predicted value output after the k-th sample is input into the model obtained by training the corresponding participating vehicle.

[0128] In this embodiment, the second credit contribution of the participating vehicle to the target model is calculated based on the local model of the roadside unit and samples in the local database, and the second credit contribution reflects the accuracy of its prediction.

[0129] In this embodiment, when calculating the contribution of the participating vehicles to the target model, the contribution of the participating vehicles to the target model in terms of quality factors, time factors and credit factors is taken into consideration, rather than just the size of the vehicle's local database and training energy consumption. Therefore, the rationality of the incentives can be improved, and technical problems such as unreasonable incentives causing long training cycles for model training and difficulty in identifying unreliable vehicle data can be avoided.

[0130] Furthermore, based on the first and second embodiments of the present application, another embodiment of the present application is provided. In this embodiment, the step of sending the incentive corresponding to the contribution to the participating vehicle includes:

[0131] Step A1: Determine the reward weight of the participating vehicle based on the contribution of the participating vehicle, where the reward weight

[0132] Among them, a and b are constant weighting factors, is the quality contribution, S t (T i ) is the time contribution, S q (W i ) is the second credit contribution;

[0133] In this embodiment, the incentive of the participating vehicle is determined according to the reward weight Q(i), wherein the reward weight where a and b represent constant weighting factors.

[0134] Step A2: Determine the incentive for the participating vehicle based on the reward weight, and send the incentive to the participating vehicle.

[0135]

[0136] Where P represents the total budget of the task publisher, N represents the total number of participating vehicles, and p i For motivation.

[0137] As an example, based on the reward weight, the incentive of the participating vehicle is determined, and the incentive is sent to the participating vehicle, wherein the incentive (reward amount) is:

[0138]

[0139] Where P represents the total budget of the task publisher, N represents the total number of participating vehicles, and p i For motivation.

[0140] Finally, the roadside unit returns the incentive to the participants in the form of transactions:

[0141] Tr pay ={id,p i}

[0142] Among them, id represents the identification of the participating vehicle.

[0143] In this embodiment, the reward weight of the participating vehicle is determined based on the contribution of the participating vehicle, wherein the reward weight Among them, a and b are constant weighting factors, is the quality contribution, S t (T i ) is the time contribution, S q (W i ) is the second credit contribution; according to the reward weight, determine the incentive of the participating vehicle, and send the incentive to the participating vehicle, wherein the incentive Where P represents the total budget of the task publisher, N represents the total number of participating vehicles, and p i In this embodiment, the incentive is accurately determined based on the reward weight, thereby improving the rationality of the incentive and avoiding technical problems such as unreasonable incentives causing long training cycles for model training and difficulty in identifying unreliable vehicle data.

[0144] Further, if Figure 2 As shown, based on the above embodiment of the present application, another embodiment of the present application is provided, which is applied to participating vehicles. The multi-factor based vehicle network federated learning incentive method includes:

[0145] Step B1: receiving the federated model to be trained sent by the roadside unit;

[0146] Step B2: asynchronously perform federated training on the federated model to be trained based on the local database, and upload the trained model-related information to the roadside unit after the training is completed. After the training of the federated model to be trained is completed, the model obtained by the roadside unit is the target model.

[0147] Step B3, receiving the local incentive determined by the roadside unit based on the model-related information, wherein the incentive is determined by the roadside unit after determining the local contribution to the target model based on the model-related information, wherein the contribution includes the degree of local contribution to the target model in terms of quality factors, time factors and credit factors.

[0148] The specific implementation methods of the multi-factor based vehicle network federated learning incentive method of the present application are basically the same as the above-mentioned multi-factor based vehicle network federated learning incentive method embodiments, and will not be repeated here.

[0149] Reference Figure 3 , Figure 3 It is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present application.

[0150] like Figure 3 As shown, the multi-factor based vehicle network federated learning incentive device may include: a processor 1001, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to realize the connection and communication between the processor 1001 and the memory 1005.

[0151] Optionally, the multi-factor-based IoV federated learning incentive device may also include a user interface, a network interface, a camera, an RF (Radio Frequency) circuit, a sensor, a WiFi module, and the like. The user interface may include a display and an input submodule such as a keyboard. The optional user interface may also include a standard wired interface and a wireless interface. The network interface may include a standard wired interface and a wireless interface (such as a WiFi interface).

[0152] Those skilled in the art will understand that Figure 3The structure of the multi-factor based vehicle network federated learning incentive device shown in the does not constitute a limitation of the multi-factor based vehicle network federated learning incentive device, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0153] like Figure 3 As shown, memory 1005, a storage medium, may include an operating system, a network communication module, and a multi-factor-based vehicle network federated learning incentive program. The operating system is a program that manages and controls the hardware and software resources of the multi-factor-based vehicle network federated learning incentive device and supports the operation of the multi-factor-based vehicle network federated learning incentive program and other software and / or programs. The network communication module is used to enable communication between the various components within memory 1005, as well as communication with other hardware and software in the multi-factor-based vehicle network federated learning incentive system.

[0154] exist Figure 3 In the multi-factor based vehicle network federated learning incentive device shown, the processor 1001 is used to execute the multi-factor based vehicle network federated learning incentive program stored in the memory 1005 to implement the steps of the multi-factor based vehicle network federated learning incentive method described in any one of the above items.

[0155] The specific implementation of the multi-factor based vehicle network federated learning incentive device of this application is basically the same as the above-mentioned multi-factor based vehicle network federated learning incentive method embodiments, and will not be repeated here.

[0156] The present application also provides a multi-factor based vehicle network federated learning incentive device, which is applied to a roadside unit, and the device includes:

[0157] A receiving module is used to receive a training task issued by a task publisher and determine the federated model to be trained corresponding to the training task;

[0158] An asynchronous federated training module is configured to send the federated model to be trained to participating vehicles within the corresponding coverage area that have chosen to participate in the training, so that the participating vehicles can use their own databases to perform asynchronous federated training on the federated model to be trained, and upload the trained model-related information to the roadside unit after the training is completed. After the training of the federated model to be trained is completed, the model obtained is the target model;

[0159] An incentive module is used to calculate the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles, and send incentives corresponding to the contribution to the participating vehicles, wherein the contribution includes the degree of contribution of the participating vehicles to the target model in terms of quality factors, time factors and credit factors.

[0160] In a possible implementation manner of the present application, the contribution degree includes the contribution degree of the participating vehicles to the target model in terms of quality factors;

[0161] The model-related information includes the change in the loss function after training and the model parameters;

[0162] The excitation module is used to implement:

[0163] After determining the first change amount of the loss function corresponding to the global model of the federated model to be trained for the participating vehicles;

[0164] Determine a second change in the overall loss function of the global model corresponding to the roadside unit in the process of obtaining the target model;

[0165] determining a mass contribution of the participating vehicle to the target model according to the first change amount and the second change amount;

[0166] Among them, the quality contribution is in,

[0167]

[0168] Where F(ω) is the second change amount, and ΔFi(W) is the first change amount of the vehicle.

[0169] In a possible implementation manner of the present application, the contribution degree includes the contribution degree of the participating vehicles to the target model in terms of time factor;

[0170] The incentive module is used to implement:

[0171] The time T at which the participating vehicles upload the model-related information i , calculating the time contribution of the participating vehicles to the target model;

[0172] Among them, the time contribution Where α represents a constant coefficient.

[0173] In a possible implementation of the present application, the contribution degree includes the contribution degree of the participating vehicle to the target model in terms of the first credit factor; the model-related information includes the number of samples D in the local database of the participating vehicle. i The time for the participating vehicles to upload the model related information is T i ;

[0174] The excitation module is used to implement:

[0175] According to T i and the number of participating vehicle datasets D iIs the ratio of β to β less than the credibility warning line β, calculate the first credit contribution of the participating vehicle to the target model, wherein the first credit contribution H i satisfy

[0176]

[0177] Among them, for H i = 0 participating vehicles are marked, the H i =0 participating vehicles are dishonest users.

[0178] In a possible implementation of the present application, the model-related information includes the number of samples D in the local database of participating vehicles. i ;

[0179] The contribution degree includes the contribution degree of the participating vehicle to the target model in terms of the second credit factor;

[0180] The step of calculating the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles includes:

[0181] Calculating a second credit contribution of the participating vehicle to the target model based on the roadside unit local model and samples in the local database; wherein the second credit contribution;

[0182]

[0183] Among them, (x k ,y k ) represents the local database D of participating vehicle i i The kth sample in x k Represents the input of the kth sample, y k represents the actual label value of the kth sample, f(x k ) represents the predicted value output after the k-th sample is input into the model obtained by training the corresponding participating vehicle.

[0184] In a possible implementation of the present application, the incentive module is used to implement:

[0185] Based on the contribution of participating vehicles, the reward weight of participating vehicles is determined, where the reward weight

[0186] Among them, a and b are constant weighting factors, is the quality contribution, S t (T i ) is the time contribution, S q (W i ) is the second credit contribution;

[0187] According to the reward weight, the incentive of the participating vehicle is determined and the incentive is sent to the participating vehicle, wherein the incentive

[0188]

[0189] Where P represents the total budget of the task publisher, N represents the total number of participating vehicles, and p i For motivation.

[0190] The specific implementation of the multi-factor based vehicle network federated learning incentive device of the present application is basically the same as the various embodiments of the multi-factor based vehicle network federated learning incentive method described above, and will not be repeated here.

[0191] The present application also provides a multi-factor based vehicle network federated learning incentive device, which is applied to participating vehicles. The multi-factor based vehicle network federated learning incentive device is used to achieve:

[0192] Receive the federated model to be trained sent by the roadside unit;

[0193] Based on the local database, the federated model to be trained is federated trained, and after the training is completed, the model related information obtained by the training is uploaded to the roadside unit, wherein after the training of the federated model to be trained is completed, the model obtained by the roadside unit is the target model;

[0194] Receive local incentives determined by a roadside unit based on the model-related information, wherein the incentives are determined after the roadside unit determines the local contribution to the target model based on the model-related information, wherein the contribution includes the degree of local contribution to the target model in terms of quality factors, time factors, and credit factors.

[0195] The specific implementation of the multi-factor based vehicle network federated learning incentive device of the present application is basically the same as the various embodiments of the multi-factor based vehicle network federated learning incentive method described above, and will not be repeated here.

[0196] An embodiment of the present application provides a storage medium, and the storage medium stores one or more programs, and the one or more programs can also be executed by one or more processors to implement the steps of the multi-factor-based vehicle network federated learning incentive method described above.

[0197] The specific implementation methods of the storage medium of this application are basically the same as the above-mentioned embodiments of the multi-factor-based vehicle network federated learning incentive method, and will not be repeated here.

[0198] The application also provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of the above-mentioned multi-factor-based federated learning incentive method for Internet of Vehicles.

[0199] The specific implementation of the computer program product of the application is basically the same as that of each embodiment of the above-mentioned multi-factor-based federated learning incentive method for Internet of Vehicles, and will not be described here.

[0200] It should be noted that in this document, the terms "comprising", "containing" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or device. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0201] The above-mentioned serial numbers of the embodiments of the application are only for description, and do not represent the advantages or disadvantages of the embodiments.

[0202] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of a software and hardware platform, or by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solutions of the application or the parts that contribute to the related art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes a plurality of instructions for making a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) execute the methods described in each embodiment of the application.

[0203] The above is only the preferred embodiment of the application, and does not limit the patent scope of the application, and any equivalent structure or equivalent process transformation, or direct or indirect application in other related technical fields based on the content of the specification and drawings of the application, are also included in the patent protection scope of the application.

Claims

1. A multi-factor based incentive method for federated learning in Internet of Vehicles, characterized by: Applied to a roadside unit, the method comprises: Receive the training task issued by the task publisher, and determine the federated model to be trained corresponding to the training task; The federated model to be trained is sent to participating vehicles that choose to participate in the training within the corresponding coverage area, so that the participating vehicles can use their own databases to conduct federated training on the federated model to be trained, and upload the model-related information obtained after the training to the roadside unit. After the training of the federated model to be trained is completed, the model obtained is the target model, and the model-related information includes the change in the loss function and the model parameters after training, and the number of samples in the local database of the participating vehicles. ; Based on the model-related information uploaded by the participating vehicles, the contribution of the participating vehicles to the target model is calculated, and the incentive corresponding to the contribution is sent to the participating vehicles, wherein the contribution includes the degree of contribution of the participating vehicles to the target model in terms of quality factors, time factors and credit factors.

2. The multi-factor based vehicle network federated learning incentive method according to claim 1, characterized in that: The contribution degree includes the contribution degree of the participating vehicles to the target model in terms of quality factors; The step of calculating the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles includes: After determining the first change amount of the loss function corresponding to the global model of the federated model to be trained for the participating vehicles; Determine a second change in the overall loss function of the global model corresponding to the roadside unit in the process of obtaining the target model; determining a mass contribution of the participating vehicle to the target model according to the first change amount and the second change amount; Among them, the quality contribution is ， in, in is the second change, It is the first change of the vehicle.

3. The multi-factor based vehicle network federated learning incentive method according to claim 1, characterized in that: The contribution degree includes the contribution degree of the participating vehicles to the target model in terms of time factor; The step of calculating the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles includes: Based on the time when the participating vehicles uploaded the relevant information of the model , calculating the time contribution of the participating vehicles to the target model; Among them, the time contribution ,in represents the constant coefficient.

4. The multi-factor based vehicle network federated learning incentive method according to claim 1, characterized in that: The contribution degree includes the contribution degree of the participating vehicle to the target model based on the first credit factor; the time when the participating vehicle uploads the model related information is ; The step of calculating the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles includes: according to and the number of participating vehicle datasets Is the ratio less than the credibility warning line? , calculate the first credit contribution of the participating vehicle to the target model, where the first credit contribution H i satisfy ； Among them, for The participating vehicles are marked. 's participating vehicles are from dishonest users.

5. The multi-factor based vehicle network federated learning incentive method according to claim 1, characterized in that: The contribution degree includes the contribution degree of the participating vehicle to the target model in terms of the second credit factor; The step of calculating the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles includes: Calculating a second credit contribution of the participating vehicle to the target model based on the roadside unit local model and samples in the local database; wherein the second credit contribution; ； in, Representative participating vehicles Local database Middle samples, Representative The input of samples, Representative The actual label value of the sample, Representative The predicted value output after the model obtained by training the corresponding participating vehicles is input.

6. The multi-factor based vehicle network federated learning incentive method according to claim 5, characterized in that: The step of sending the incentive corresponding to the contribution to the participating vehicles includes: Based on the contribution of participating vehicles, the reward weight of participating vehicles is determined, where the reward weight ； in, and are constant weighting factors, Contribution to quality, For time contribution, Second credit contribution; According to the reward weight, the incentive of the participating vehicle is determined and the incentive is sent to the participating vehicle, wherein the incentive ; in represents the total budget of the task publisher, represents the total number of participating vehicles, For motivation.

7. A multi-factor based incentive method for federated learning in Internet of Vehicles, characterized by: Applied to participating vehicles, the multi-factor-based vehicle network federated learning incentive method includes: Receive the federated model to be trained sent by the roadside unit; Based on the local database, the federated model to be trained is federated trained, and after the training, the model-related information obtained is uploaded to the roadside unit. After the training of the federated model to be trained is completed, the model obtained by the roadside unit is the target model. The model-related information includes the change in the loss function and the model parameters after training, the number of samples in the local database of the participating vehicles, and the number of samples in the local database of the participating vehicles. ; Receive local incentives determined by a roadside unit based on the model-related information, wherein the incentives are determined after the roadside unit determines the local contribution to the target model based on the model-related information, wherein the contribution includes the degree of local contribution to the target model in terms of quality factors, time factors, and credit factors.

8. A multi-factor based vehicle network federated learning incentive device, characterized in that: Applied to a roadside unit, the device comprises: A receiving module is used to receive a training task issued by a task publisher and determine the federated model to be trained corresponding to the training task; The federated training module is used to send the federated model to be trained to the participating vehicles that choose to participate in the training within the corresponding coverage area, so that the participating vehicles can use their own databases to conduct federated training on the federated model to be trained, and upload the model-related information obtained after the training to the roadside unit. After the training of the federated model to be trained is completed, the model obtained is the target model, and the model-related information includes the change in the loss function and the model parameters after training, and the number of samples in the local database of the participating vehicles. ; An incentive module is used to calculate the contribution of the participating vehicles to the target model based on the model-related information uploaded by the participating vehicles, and send incentives corresponding to the contribution to the participating vehicles, wherein the contribution includes the degree of contribution of the participating vehicles to the target model in terms of quality factors, time factors and credit factors.

9. A multi-factor based vehicle network federated learning incentive device, characterized in that: It includes a memory, a processor, and a multi-factor vehicle network federated learning incentive program stored in the memory and executable on the processor. When the processor executes the multi-factor vehicle network federated learning incentive program, it implements the steps of the multi-factor vehicle network federated learning incentive method described in any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium stores a multi-factor based vehicle network federated learning incentive program, which, when executed by a processor, implements the steps of the multi-factor based vehicle network federated learning incentive method as described in any one of claims 1 to 7.

Images