Model training method and device, network equipment, computer storage medium and program product

Abstract

The invention discloses a model training method and device, network equipment, a computer storage medium and a program product, and the method comprises the steps: receiving first information sent by at least one second network analysis function, and the first information comprises local model information and identification information of the second network analysis function; receiving an update request message sent by at least one second network analysis function, wherein the update request message carries identification information of the second network analysis function; second information of each second network analysis function is calculated according to the first information, the second information is sent to the corresponding second network analysis function based on the identification information of each second network analysis function, and the second information is used for updating a local model of the second network analysis function. And the problem that the overall efficiency of model training is reduced due to partial node faults is solved, so that the robustness of model training is improved.

Description

Technical Field

[0001] The present application relates to the field of core networks, and in particular to a model training method and apparatus, network equipment, computer storage media and program products. Background Art

[0002] With the rapid development of artificial intelligence (AI), the core network needs to integrate AI technology to improve network data analysis capabilities to meet complex business needs. For network services, a large amount of data is required for model training to achieve more accurate business implementation. However, there is a large communication overhead when a large amount of training data is reported to the network, and the pressure on data privacy protection on the network side is increased.

[0003] The existing technology proposes that the fifth generation mobile communication technology (5G) system supports vertical federated learning services to meet training needs. However, there will be a problem of partial node failure leading to a decrease in overall efficiency. Therefore, an effective model training method is urgently needed to solve this problem. Summary of the invention

[0004] In order to solve the above technical problems, the embodiments of the present invention provide a model training method and device, network equipment, computer storage medium and program product.

[0005] The model training method provided in the embodiment of the present application is applied to the first network analysis function, including:

[0006] receiving first information sent by at least one second network analysis function, wherein the first information includes local model information and identification information of the second network analysis function;

[0007] receiving an update request message sent by at least one second network analysis function, wherein the update request message carries identification information of the second network analysis function;

[0008] The second information of each second network analysis function is calculated according to the first information, and the second information is sent to the corresponding second network analysis function based on the identification information of each second network analysis function, where the second information is used to update the local model of the second network analysis function.

[0009] In some embodiments, the method further comprises:

[0010] Configure a first preset period, where the first preset period is used to determine the number T of second network analysis functions reported, where T is an integer greater than 1;

[0011] After receiving T first information respectively reported by T second network analysis functions, the local model of the first network analysis function is updated, and the local model of the first network analysis function and the local model of the second network analysis function constitute a first model.

[0012] In some implementations, after receiving the update request message sent by at least one second network analysis function, the method further includes:

[0013] After receiving the update request messages sent by the T second network analysis functions, calculating the second information of the T second network analysis functions;

[0014] Accordingly, the step of calculating the second information of each second network analysis function according to the first information, and sending the second information to the corresponding second network analysis function based on the identification information of each second network analysis function includes:

[0015] Querying corresponding first information based on identification information carried by T update request messages;

[0016] Second information of T second network analysis functions is calculated based on the corresponding first information, and the second information is sent to the T second network analysis functions based on the corresponding identification information.

[0017] In some embodiments, the method further comprises:

[0018] In each first preset period, first information reported by the T second network analysis functions is stored, where the first information includes identification information of the T second network analysis functions.

[0019] In some embodiments, the method further comprises:

[0020] Determining the activity of the second network analysis function based on the amount of identification information of the second network analysis function stored;

[0021] If the activity of the second network analysis function is lower than the first threshold, an activation request message is sent to the second network analysis function, and an activation response message sent by the second network analysis function is received, wherein the activation response message carries the identification information, health status report and training strategy of the second network analysis function.

[0022] In some implementations, before receiving the first information sent by the second network analysis function, the method further includes:

[0023] receiving a first model subscription request message sent by a consumer, and determining a second network analysis function participating in model training based on the first model subscription request message;

[0024] A first model training request message is sent to the second network analysis function, where the first model training request message carries initial model information and a model training method.

[0025] In some embodiments, the method further comprises:

[0026] Sending a first model training status message to the consumer;

[0027] Receive a second model subscription request message sent by the consumer, and update or terminate model training based on the second model subscription request message.

[0028] The model training method provided in the embodiment of the present application is applied to the second network analysis function, including:

[0029] Sending first information to the first network analysis function, the first information including local model information and identification information of the second network analysis function;

[0030] Sending an update request message to the first network analysis function, where the update request message carries identification information of the second network analysis function;

[0031] Second information sent by the first network analysis function based on the identification information is received, where the second information is calculated by the first network analysis function according to the first information.

[0032] In some embodiments, the method further comprises:

[0033] receiving an activation request message sent by the first network analysis function, and determining a health status report and a training strategy based on the activation request message;

[0034] An activation response message is sent to the first network analysis function, where the activation response message carries identification information of the second network analysis function, the health status report, and the training strategy.

[0035] In some implementations, before sending the first information to the first network analysis function, the method further includes:

[0036] Receiving a first model training request message sent by the first network analysis function, where the first model training request message carries initial model information and a model training method;

[0037] Collecting local data, and performing sample alignment processing on the local data;

[0038] The local model of the second network analysis function is trained based on the initial model information, the model training method and the local data after sample alignment processing, and the local model of the second network analysis function and the local model of the first network analysis function constitute a first model.

[0039] The model training device provided in the embodiment of the present application is applied to the first network analysis function, and the device includes:

[0040] A receiving unit, receiving first information sent by at least one second network analysis function, wherein the first information includes model information and identification information of the second network analysis function; receiving an update request message sent by at least one second network analysis function, wherein the update request message carries the identification information of the second network analysis function;

[0041] a calculation unit, configured to calculate second information of each second network analysis function according to the first information;

[0042] A sending unit is used to send the second information to the corresponding second network analysis function based on the identification information of each second network analysis function, where the second information is used to update the local model of the second network analysis function.

[0043] The model training device provided in the embodiment of the present application is applied to the second network analysis function, and the device includes:

[0044] a sending unit, configured to send first information to the first network analysis function, wherein the first information includes model information and identification information of the second network analysis function; and send an update request message to the first network analysis function, wherein the update request message carries the identification information of the second network analysis function;

[0045] A receiving unit is used to receive second information sent by the first network analysis function based on the identification information, where the second information is calculated by the first network analysis function according to the first information.

[0046] The network device provided in an embodiment of the present application includes: a processor and a memory, the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to execute any one of the above-mentioned model training methods.

[0047] The computer-readable storage medium provided in an embodiment of the present application is used to store a computer program, and the computer program enables a computer to execute any one of the above-mentioned model training methods.

[0048] The computer program product provided in an embodiment of the present application includes computer program instructions, which enable a computer to execute any one of the model training methods described above.

[0049] In the technical solution of the embodiment of the present application, the first network analysis function receives first information sent by at least one second network analysis function, the first information includes local model information and identification information of the second network analysis function; receives an update request message sent by at least one second network analysis function, the update request message carries the identification information of the second network analysis function; calculates second information of each second network analysis function according to the first information, and sends the second information to the corresponding second network analysis function based on the identification information of each second network analysis function, and the second information is used to update the local model of the second network analysis function. In this way, the first network analysis function and the second network analysis function can asynchronously trigger subsequent calculations and model updates, and the first network analysis function can solve the problem of reduced overall efficiency of model training due to partial node failures without waiting to receive the local calculation results of all second network analysis functions, thereby improving the robustness of model training. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flow chart of a model training method in an embodiment of the present application. Figure 1 ;

[0051] Figure 2 This is a flow chart of a model training method in an embodiment of the present application. Figure 2 ;

[0052] Figure 3 This is a flow chart of a model training method in an embodiment of the present application. Figure 3 ;

[0053] Figure 4 This is a flow chart of a model training method in an embodiment of the present application. Figure 4 ;

[0054] Figure 5 This is a schematic diagram of the structure of a model training device provided in an embodiment of the present application. Figure 1 ;

[0055] Figure 6 This is a schematic diagram of the structure of a model training device provided in an embodiment of the present application. Figure 2 ;

[0056] Figure 7 It is a schematic structural diagram of a network device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0057] The following will describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0058] It should be noted that the terms "system" and "network" are often used interchangeably in this article. The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there may be three relationships, for example, A and / or B, which can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the objects associated with each other are in an "or" relationship. The terms "first", "second", etc. in this article are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features, for example, the first object may be one or more. In the description of the present disclosure, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0059] To facilitate understanding of the technical solutions of the embodiments of the present application, the relevant technologies of the embodiments of the present application are described below. The following related technologies can be arbitrarily combined with the technical solutions of the embodiments of the present application as optional solutions, and they all belong to the protection scope of the embodiments of the present application.

[0060] Figure 1 This is a flow chart of a model training method in an embodiment of the present application. Figure 1 ,like Figure 1 As shown, the model training steps of the network data analysis function (NWDAF) in the 5G system to implement vertical federated learning (VFL) are shown, wherein the first network element can be represented as VFL Server NWDAF, and the second network element can be represented as VFL Client NWDAF. The specific steps are as follows:

[0061] Step 101: The consumer sends a model subscription request to .

[0062] Step 102: After receiving the model subscription request, send initial model information configuration to at least one.

[0063] Step 103: After receiving the initial model information configuration, at least one second network element performs data collection.

[0064] Step 104: Each second network element performs sample alignment.

[0065] Step 105: Each second network element reports local model information to the first network element.

[0066] Step 106: After receiving the local model information reported by each second network element, the first network element performs model calculation.

[0067] Step 107: The first network element sends a vertical federated learning status report to the consumer.

[0068] Step 108: The consumer sends information to the first network element to continue or stop the vertical federated learning process.

[0069] Step 109: The first network element selects to update or terminate the model training process according to the information sent by the consumer.

[0070] Step 110: The first network element sends a reverse local model information report to each second network element.

[0071] Step 111: Each second network element updates local model parameters after receiving the reverse local model information report.

[0072] Step 112: Repeat steps 103 to 108 until the termination condition is met.

[0073] Here, the first network element serves as the central node, and each second network element uploads the intermediate results after local training to the first network element. The first network element performs calculations based on the intermediate results of all second network elements received, and calculates the backward model information of each second network element. Each second network element updates the local parameters based on the backward model information. Each iteration of vertical federated learning requires synchronous communication between the first network element and the second network element. The first network element must receive the intermediate results of all the second network elements involved in the calculation before it can perform subsequent calculations. The calculation delay, communication delay, etc. of any second network element node will lead to a decrease in the overall efficiency of the training process, and the system robustness is not high. To this end, the following technical solution of this application is proposed.

[0074] To facilitate understanding of the technical solutions of the embodiments of the present application, the technical solutions of the present application are described in detail below through specific embodiments. The above related technologies can be combined arbitrarily with the technical solutions of the embodiments of the present application as optional solutions, and they all belong to the protection scope of the embodiments of the present application. The embodiments of the present application include at least part of the following contents.

[0075] Figure 2 This is a schematic diagram of the process of the model training method provided in the embodiment of the present application. Figure 2 ,like Figure 2 As shown, the model training method is applied to the first network analysis function, specifically comprising the following steps:

[0076] Step 201: Receive first information sent by at least one second network analysis function, where the first information includes local model information and identification information of the second network analysis function.

[0077] In the embodiment of the present application, the first network analysis function may be a network element or device in the communication network that can provide machine learning functions, such as a network data analysis function NWDAF. Exemplarily, the first network analysis function may be represented as a service NWDAF under a vertical federated learning framework, namely, VFL Server NWDAF.

[0078] In some implementations, the first network analysis function may also be a management data analysis service (MDAS), a management data analysis function (MDAF), or other network elements or devices dedicated to providing network intelligent services. Alternatively, the first network analysis function may also be a network element or device that provides other communication-related services, such as mobile management services and session management services, and also has intelligent functions, such as access and mobility management functions (AMF), session management functions (SMF), application functions (AF), etc. The AF may be an AF deployed by a communication operator or a third-party AF.

[0079] In an embodiment of the present application, the second network analysis function may be a network element or device in the communication network that has local machine learning capabilities, such as NWDAF. Exemplarily, the second network analysis function may be represented as a client NWDAF under a vertical federated learning framework, i.e., VFL Client NWDAF. It should be noted that the second network analysis function may be one or more.

[0080] In some implementations, the second network analysis function may also be an AI functional network element in a wireless access network domain, other AI functional network elements in a core network domain, a third-party AI application, an AI agent (client) in a UE, a local communication service device, etc. Alternatively, the second network analysis function may also be a network element or device that provides other communication-related services, such as a mobile management service, a session management service, and has intelligent functions, such as an AMF, SMF, AF, etc. The AF may be an AF deployed by a communication operator or a third-party AF.

[0081] In an embodiment of the present application, after the second network analysis function completes a round of local model training, it will report the first information to the first network analysis function, and the first information includes the local model information of the second network analysis function. Among them, the local model information of the second network analysis function can be the intermediate result or forward model information after a round of training of the local model calculation of the second network analysis function. The first information may also include identification information (Client ID) of the second network analysis function. It should be noted that the model training can be AI model training corresponding to data analysis, AI model training corresponding to communication robustness learning, or AI model training for model training of the mobile trajectory of the UE, etc., and the present application does not impose any restrictions on this.

[0082] In some embodiments, before the first network analysis function receives the first information sent by the second network analysis function, it also includes: the first network analysis function receives a first model subscription request message sent by a consumer, and determines the second network analysis function to participate in model training based on the first model subscription request message.

[0083] Here, the consumer is the model consumer in the model training system, which can be a network element or device in the wireless communication system that needs to consume the model, such as a third-party AI application, UE, network-side equipment, etc.

[0084] Specifically, the first network analysis function, the second network analysis function and the consumer constitute a model training system, and the model training process of the first network analysis function and the second network analysis function can be a federated model training process. The federated model training process can be triggered by the first network analysis function according to the model training task required by itself, or it can be triggered by the consumer. For example, if the consumer sends a first model subscription request message to the first network analysis function, the first model subscription request message can be a model request corresponding to the model training task sent by the consumer to the first network analysis function according to its own needs, so as to trigger the federated model training process. Among them, the federated model training process can also be understood as a vertical federated learning process, which is essentially a combination of sample features. By training the different sample features of the common samples of the second network analysis function through the joint model, the feature dimensions of the training data are increased to train a better model. In this way, when the first network analysis function receives the model training request sent by the consumer, it triggers the federated model training process, and then combines the second network analysis functions to perform local distributed model training to obtain the target model, thereby ensuring the security and privacy of the data on the second network analysis function while meeting the model training requirements.

[0085] In some embodiments, the first network analysis function determines the second network analysis function participating in the model training based on the first model subscription request message. There are many ways to determine the second network analysis function participating in the federal model training process. In one embodiment, the second network analysis function may be a network element that the first network analysis function obtains from NRF or UDM according to the first model training task and that can support the federal model training process. For example, the first network analysis function obtains the AMF instance (instance(s), RAN instance(s), AF instance(s), etc. participating in the federal model training process from the NRF query. In another embodiment, the first network analysis function obtains the UE instance (instance(s)) participating in the federal model training process from the UDM query, etc.

[0086] In some implementations, the first network analysis function sends a first model training request message to the second network analysis function, where the first model training request message carries initial model information and a model training method.

[0087] Here, after receiving the first model training request message, each second network analysis function can perform federated model training according to the first model training request message. Specifically, each second network analysis function collects local data for local model training, and each first network analysis function and second network analysis function participating in the vertical federated model training perform sample alignment. After that, each second network analysis function performs local model training to obtain intermediate data for federated model training, and sends the intermediate data for federated model training to the first network analysis function through the first information, that is, the first information at least includes the intermediate data for federated model training. Among them, when the second network analysis function performs model training according to the first model training request message, it can be based on a locally pre-configured initial model implementation, or it can be based on a local model implementation that has been trained before receiving the first model training request message, or it can be based on the model implementation obtained by training the initial model information and model training method carried in the first model training request message.

[0088] Step 202: Receive an update request message sent by at least one second network analysis function, where the update request message carries identification information of the second network analysis function.

[0089] Here, the first network analysis function receives an update request message sent by at least one second network analysis function. Since there are one or more second network analysis functions, the second network analysis functions need to report the identification information of their respective second network analysis functions at the same time when reporting the update request message to the first network analysis function.

[0090] Step 203: Calculate second information of each second network analysis function according to the first information, and send second information to the corresponding second network analysis function based on the identification information of each second network analysis function, where the second information is used to update the local model of the second network analysis function.

[0091] Here, after receiving the update request message sent by at least one second network analysis function, the first network analysis function calculates the corresponding second information based on the local model information reported by each second network analysis function. The second information may be backward model information, such as a gradient value, a backward gradient value, or backward local model information. After calculating the second information corresponding to each second network analysis function, the first network analysis function sends the second information to the corresponding second network analysis function according to the identification information of the second network analysis function. The second network analysis function updates the local model based on the second information.

[0092] From the above, it can be seen that in the model training method provided by the embodiment of the present application, the first network analysis function receives the first information sent by at least one second network analysis function, the first information carries the local model information and identification information of the second network analysis function; receives the update request message sent by at least one second network analysis function, the update request message carries the identification information of the second network analysis function; calculates the second information of each second network analysis function according to the first information, and sends the second information to the corresponding second network analysis function based on the identification information of each second network analysis function. In this way, the second network analysis function can asynchronously trigger subsequent calculations and model updates, and the first network analysis function can solve the problem of overall efficiency reduction caused by partial node failures without waiting to receive the local calculation results of all second network analysis functions, thereby improving the robustness of the system.

[0093] In some embodiments, the first network analysis function is configured with a first preset period, and the first preset period is used to determine the number T of second network analysis functions to be reported, where T is an integer greater than 1; after receiving T first information reported by T second network analysis functions respectively, the local model of the first network analysis function is updated, and the local model of the first network analysis function and the local model of the second network analysis function constitute a first model.

[0094] Here, the first network analysis function side may set a local model, and the local model of the first network analysis function and the local model of the second network analysis function constitute a first model, wherein the first model is an overall training model. Specifically, the model training process of the first network analysis function and the second network analysis function may be a federated model training process, in which case the local model of the first network analysis function and the local model of the second network analysis function constitute a complete federated model. The first network analysis function may set a first preset period by local configuration or the like, and the first preset period is used to determine the number T of second network analysis functions reported, and within the first preset period, the local model of the first network analysis function is updated after receiving the first information reported by T second network analysis functions. Here, the first preset period is an asynchronous update period T, and the first network analysis function determines the number T of second network analysis functions reported based on the asynchronous update period T, that is, when the first network analysis function receives the first information reported by T second network analysis functions, the first network analysis function triggers the update of the local model of the first network analysis function. In some embodiments, if the first network analysis function is not configured with the first preset period, or the first preset period T is configured as 1, the first network analysis function updates the local model of the first network analysis function once after receiving the first information reported by the second network analysis function.

[0095] From the above, it can be seen that the model training method provided in the embodiment of the present application sets a local model through the first network analysis function, configures a first preset period, the first preset period is used to determine the number T of second network analysis functions reported, and after receiving T first information reported by T second network analysis functions respectively, updates the local model of the first network analysis function, and the local model of the first network analysis function and the local model of the second network analysis function constitute the first model; in this way, compared with the existing vertical federated learning process, each iterative update does not need to wait for the local calculation results of all second network analysis functions to be uploaded. The model training method provided in the embodiment of the present application can be independently triggered by a single or multiple second network analysis functions. Model update eliminates the barrel effect and can greatly improve the training efficiency when the performance of each second network analysis function is quite different.

[0096] In some implementations, after the first network analysis function receives an update request message sent by at least one second network analysis function, the method further includes: after receiving update request messages sent by T second network analysis functions, calculating second information of T second network analysis functions.

[0097] Correspondingly, step 203 includes: querying the corresponding first information based on the identification information carried by T update request messages; calculating the second information of T second network analysis functions based on the corresponding first information, and sending the second information to the T second network analysis functions based on the corresponding identification information.

[0098] Here, if the first network analysis function is set with a first preset period T, after receiving the update request message sent by T second network analysis functions, the first network analysis function triggers the calculation of the second information corresponding to each second network analysis function based on the first information reported by the T second network analysis functions. Then, the second information is sent to the T second network analysis functions based on the corresponding identification information. Among them, the second information can be a gradient, a backward gradient, and a backward local model information. In some embodiments, if the first network analysis function does not set the first preset period, the second information is calculated once after receiving an update request message sent by a second network analysis function by default, and then the second information is sent to the second network analysis function based on the corresponding identification information. In this way, the model training method provided in the embodiment of the present application, compared with the existing vertical federated learning process, does not need to wait for all local calculation results to be uploaded for each iterative update, and can be independently triggered by a single or multiple second network analysis functions, eliminating the barrel effect. When the performance of each second network analysis function is quite different, the training efficiency can be greatly improved.

[0099] In some implementations, the first network analysis function stores first information reported by T second network analysis functions in each first preset period, where the first information includes identification information of the T second network analysis functions.

[0100] Here, the first network analysis function sets a first preset period, that is, the first network analysis function will receive the first information reported by T second network analysis functions in each first preset period. At the same time, the first network analysis function will store the first information of T second network analysis functions in each first preset period, the first information including the local model information and identification information of the T second network analysis functions, and determine whether there is an inactive second network analysis function according to the number of identification information of the second network analysis functions stored in multiple first preset periods.

[0101] Specifically, the first network analysis function determines the activity of the second network analysis function based on the amount of identification information of the stored second network analysis function; if the activity of the second network analysis function is lower than the first threshold, an activation request message is sent to the second network analysis function, and an activation response message is received from the second network analysis function, wherein the activation response message carries the identification information, health status report and training strategy of the second network analysis function.

[0102] Here, the first network analysis function will store the identification information of the second network analysis function in each first preset period. The first network analysis function will determine whether there is an inactive second network analysis function based on the number of identification information of each second network analysis function stored in multiple first preset periods. If so, the first network analysis function will send an activation request message to the corresponding inactive second network analysis function. The second network analysis function that receives the activation request message will check its own status and send an activation response message to the first network analysis function. The activation response message carries a vertical federated model training strategy including but not limited to the identification information of the second network analysis function, a health status report, and whether to continue to participate in model training. In this way, the model training method provided in the embodiment of the present application, compared with the existing vertical federated learning process, does not need to wait for all local calculation results to be uploaded for each iterative update, and can be independently triggered by a single or multiple second network analysis functions, eliminating the barrel effect. When the performance differences of each second network analysis function are large, the training efficiency can be greatly improved.

[0103] In some embodiments, the first network analysis function sends a first model training status message to the consumer; the first network analysis function receives a second model subscription request message sent by the consumer, and updates or terminates model training based on the second model subscription request message.

[0104] Here, when the calculation result of the loss function of the target model meets the predetermined requirements, such as the accuracy value of the loss function reaches the predetermined value, that is, when the target model converges, the first network analysis function can send a first model training status message to the consumer, that is, relevant information of the target model, so that the consumer can perform corresponding tasks based on the target model. After receiving the first model training status message, the consumer determines whether the model meets the requirements and sends a second model subscription request message to the first network analysis function. The second model subscription request message can be a modification model subscription message. After receiving the second model subscription request message sent by the consumer, the first network analysis function updates or terminates the model training until the termination condition is triggered.

[0105] Figure 3 This is a schematic diagram of the process of the model training method provided in the embodiment of the present application. Figure 3 ,like Figure 3 As shown, the model training method is applied to the second network analysis function, and the specific steps include:

[0106] Step 301: The second network analysis function sends first information to the first network analysis function, where the first information includes local model information and identification information of the second network analysis function.

[0107] In an embodiment of the present application, the second network analysis function may be a network element or device in the communication network that has local machine learning capabilities, such as a network data analysis function NWDAF. Exemplarily, the second network analysis function may be represented as a client NWDAF under a vertical federated learning framework, i.e., VFL Client NWDAF. It should be noted that the second network analysis function may be one or more.

[0108] In some implementations, the second network analysis function may also be an AI function network element in a wireless access network domain, other AI function network elements in a core network domain, a third-party AI application, an AI agent (client) in a UE, a local communication service device, etc. Alternatively, the second network analysis function may also be a network element or device that provides other communication-related services, such as a mobility management service, a session management service, and has intelligent functions, such as an access and mobility management function AMF, a session management function SMF, an application function AF, etc. The AF may be an AF deployed by a communication operator or a third-party AF.

[0109] In the embodiment of the present application, the first network analysis function may be a network element or device in the communication network that can provide machine learning functions, such as NWDAF. Exemplarily, the first network analysis function may be represented as a service NWDAF under the vertical federated learning framework, namely VFL Server NWDAF.

[0110] In some implementations, the first network analysis function may also be a network element or device dedicated to providing network intelligent services, such as MDAS and MDAF. Alternatively, the first network analysis function may also be a network element or device that provides other communication-related services, such as mobile management services and session management services, and also has intelligent functions, such as AMF, SMF, AF, etc. The AF may be an AF deployed by a communication operator or a third-party AF.

[0111] In an embodiment of the present application, after the second network analysis function completes a round of local model training, it will report the first information to the first network analysis function, and the first information includes the local model information of the second network analysis function. Among them, the local model information of the second network analysis function can be the intermediate result or forward model information of the local model calculation of the second network analysis function after a round of training on this side. The first information may also include identification information (Client ID) of the second network analysis function. It should be noted that the model training can be AI model training corresponding to data analysis, AI model training corresponding to communication robustness learning, or AI model training for model training of UE's mobile trajectory, etc., and the present application does not impose any restrictions on this.

[0112] Step 302: The second network analysis function sends an update request message to the first network analysis function, where the update request message carries identification information of the second network analysis function.

[0113] Here, the first network analysis function receives an update request message sent by at least one second network analysis function. Since there are one or more second network analysis functions, the second network analysis functions need to report the ID information of their respective second network analysis functions at the same time when reporting the update request message to the first network analysis function.

[0114] Step 303: The second network analysis function receives second information sent by the first network analysis function based on the identification information, where the second information is calculated by the first network analysis function based on the first information.

[0115] Here, after receiving the update request message sent by at least one second network analysis function, the first network analysis function calculates the corresponding second information based on the local model information reported by each second network analysis function. The second information may be backward model information, such as a gradient value, a backward gradient value, or backward local model information. After calculating the second information corresponding to each second network analysis function, the first network analysis function sends the second information to the corresponding second network analysis function according to the identification information of the second network analysis function.

[0116] From the above, it can be seen that in the model training method provided by the embodiment of the present application, the second network analysis function sends the first information to the first network analysis function, and the first information carries the local model information and identification information of the second network analysis function; the second network analysis function sends an update request message to the first network analysis function, and the update request message carries the identification information of the second network analysis function; the second network analysis function receives the second information sent by the first network analysis function based on the identification information, and the second information is calculated by the first network analysis function according to the first information. In this way, the second network analysis function can asynchronously trigger subsequent calculations and model updates, and the first network analysis function can solve the problem of overall efficiency reduction caused by partial node failures without waiting to receive the local calculation results of all second network analysis functions, thereby improving the robustness of the system.

[0117] In some embodiments, the second network analysis function receives an activation request message sent by the first network analysis function, and determines a health status report and a training strategy based on the activation request message; the second network analysis function sends an activation response message to the first network analysis function, and the activation response message carries identification information, a health status report, and a training strategy of the second network analysis function.

[0118] Here, the first network analysis function sets an asynchronous update period T, that is, the first network analysis function will receive the first information reported by T second network analysis functions in each asynchronous update period. At the same time, the first network analysis function will store the identification information of T second network analysis functions in each first preset period, and determine whether there is an inactive second network analysis function based on the number of second network analysis function identification information stored in multiple first preset periods. If so, the first network analysis function will send an activation request message to the corresponding inactive second network analysis function. The second network analysis function that receives the activation request message will check its own status and send an activation response message to the first network analysis function. The activation response message carries including but not limited to the identification information of the second network analysis function, a health status report, and a longitudinal federated model training strategy for whether to continue participating in model training.

[0119] In some embodiments, before the second network analysis function sends the first information to the first network analysis function, the second network analysis function receives a first model training request message sent by the first network analysis function, and the first model training request message carries initial model information and a model training method; the second network analysis function collects local data and performs sample alignment on the local data; the second network analysis function trains the local model of the second network analysis function based on the initial model information, the model training method, and the local data after sample alignment.

[0120] Here, after receiving the first model training request message, each second network analysis function can perform federated model training according to the first model training request message. Specifically, each second network analysis function collects local data for local model training, and each first network analysis function and second network analysis function participating in the vertical federated model training perform sample alignment. After that, each second network analysis function performs local model training to obtain intermediate data for federated model training, and sends the intermediate data for federated model training to the first network analysis function through the first information, that is, the first information at least includes the intermediate data for federated model training. Among them, when the second network analysis function performs model training according to the first model training request message, it can be based on a locally pre-configured initial model implementation, or it can be based on a local model implementation that has been trained before receiving the first model training request message, or it can be based on the model implementation obtained by training the initial model information and model training method carried in the first model training request message.

[0121] Embodiment 1

[0122] Figure 4 This is a schematic diagram of the process of the model training method provided in the embodiment of the present application. Figure 4 ,like Figure 4As shown, an asynchronous vertical intelligent joint learning method is proposed. Compared with the existing vertical federated learning process, each iterative update does not need to wait for all local calculation results to be uploaded. It can be independently triggered by a single or multiple second network analysis functions, eliminating the barrel effect. When the performance of each second network analysis function varies greatly, it can greatly improve the training efficiency. The specific steps of the model training method are as follows:

[0123] Step 401: The consumer initiates a model subscription request to the first network analysis function.

[0124] Step 402: The first network analysis function performs member selection.

[0125] Here, the first network analysis function selects the second network analysis function to participate in the model training.

[0126] Step 403: The first network analysis function sends initial model information to at least one second network analysis function.

[0127] Here, the first network analysis function sends a request to the second network analysis function selected to participate in the model training, requesting to start the model training, and the request information includes but is not limited to the initial model information and the model training method (method / algorithm / type).

[0128] Step 404: The second network analysis function performs data collection.

[0129] Here, each second network analysis function collects local data for local model training.

[0130] Step 405: The first network analysis function and the second network analysis function perform sample alignment.

[0131] Here, members participating in model training perform sample alignment.

[0132] Step 406: The second network analysis function performs local model training.

[0133] Here, each second network analysis function performs model training locally.

[0134] Step 407: The second network analysis function reports the local model information to the first network analysis function.

[0135] Here, after completing a round of local training, the second network analysis function reports to the first network analysis function, and the reported information includes but is not limited to the identification information of the second network analysis function and the local model information of the second network analysis function.

[0136] Step 408: The first network analysis function updates the local model.

[0137] Here, if the first network analysis function configures a local model, the local model of the first network analysis function is updated based on the local model information reported by the second network analysis function. The first network analysis function can obtain an asynchronous update period T through local configuration and other methods. When the local model information reported by T second network analysis functions is obtained, the local model update of the first network analysis function is triggered; if there is no setting, the default is to update once each time the first information reported by the second network analysis function is received. At the same time, the first network analysis function will record the identification information of each reported second network analysis function in each update period.

[0138] Step 409: The second network analysis function sends an update request message to the first network analysis function.

[0139] Here, the second network analysis function initiates a backward model information request to the first network analysis function, and the request information includes but is not limited to the identification information of the second network analysis function.

[0140] Step 410: The first network analysis function performs model calculation.

[0141] Here, the first network analysis function calculates the backward model information (gradient / backward gradient / backward local model information) corresponding to each second network analysis function based on the reported local model information. If the asynchronous update period T is set, the calculation is triggered after receiving the update request message sent by T second network analysis functions; if there is no setting, the calculation is performed once for each update request message received by default.

[0142] Step 411: The first network analysis function sends second information to the second network analysis function.

[0143] Here, the first network analysis function distributes corresponding backward model information based on the identification information of the second network analysis function.

[0144] Step 412: The second network analysis function updates the local model.

[0145] Here, the second network analysis function performs local model update after receiving the backward model information.

[0146] Step 413: The first network analysis function reactivates the inactive second network analysis function.

[0147] Here, the first network analysis function records the identification information of each reported second network analysis function in each update cycle. After multiple update cycles, the first network analysis function determines the inactive second network analysis function based on the number of second network analysis function identification information and performs activation processing. Specifically, the first network analysis function sends an activation request to the corresponding second network analysis function. The second network analysis function that receives the request will check its own status and send a response to the first network analysis function. The response information includes but is not limited to the identification information of the second network analysis function, the health status report, and the training strategy.

[0148] Step 414: The first network analysis function reports the training information to the consumer.

[0149] Here, the first network analysis function sends the training status to the consumer.

[0150] Step 415: The consumer sends a message to the first network analysis function to continue or terminate the training process.

[0151] Here, the consumer determines whether the model meets the requirements and sends a modification model subscription to the first network analysis function to update or terminate model training.

[0152] Step 416: The first network analysis function updates or terminates the training process.

[0153] Step 417 and steps 406 to 416 are iteratively executed until the termination condition is triggered.

[0154] Figure 5 This is a schematic diagram of the structure of the model training device provided in the embodiment of the present application. Figure 1 , applied to the first network analysis function, such as Figure 5 As shown, the model training device 500 includes:

[0155] The receiving unit 501 is used to receive first information sent by at least one second network analysis function, wherein the first information includes local model information and identification information of the second network analysis function; and receive an update request message sent by at least one second network analysis function, wherein the update request message carries the identification information of the second network function.

[0156] The calculation unit 502 is used to calculate the second information of each second network analysis function according to the first information.

[0157] The sending unit 503 is used to send the second information to the corresponding second network analysis function based on the identification information of each second network analysis function, where the second information is used to update the local model of the second network analysis function.

[0158] In some implementations, the model training device 500 further includes: a configuration unit 504 .

[0159] The configuration unit 504 is used to configure a first preset period, where the first preset period is used to determine the number T of second network analysis functions to be reported, where T is an integer greater than 1.

[0160] The receiving unit 501 is further used to update the local model of the first network analysis function after receiving T first information reported by T second network analysis functions respectively, and the local model of the first network analysis function and the local model of the second network analysis function constitute a first model.

[0161] In some embodiments, the computing unit 502 is further used to calculate the second information of the T second network analysis functions after receiving update request messages sent by the T second network analysis functions; query the corresponding first information based on the identification information carried by the T update request messages; and calculate the second information of the T second network analysis functions respectively based on the corresponding first information.

[0162] The sending unit 503 is further configured to send the second information to the T second network analysis functions respectively based on the corresponding identification information.

[0163] In some implementations, the model training device 500 further includes: a storage unit 505. The storage unit 505 is configured to store first information reported by the T second network analysis functions in each first preset period, wherein the first information includes identification information of the T second network analysis functions.

[0164] In some embodiments, the calculating unit 502 is further used to determine the activity of the second network analysis function based on the amount of identification information of the second network analysis function stored; if the activity of the second network analysis function is lower than a first threshold, the sending unit 503 is further used to send an activation request message to the second network analysis function, and the receiving unit 501 is further used to receive an activation response message sent by the second network analysis function, wherein the activation response message carries the identification information, health status report and training strategy of the second network analysis function.

[0165] In some implementations, the receiving unit 501, before receiving the first information sent by the second network analysis function, is further used to receive a first model subscription request message sent by a consumer, and determine the second network analysis function that participates in model training based on the first model subscription request message. The sending unit 503 is further used to send a first model training request message to the second network analysis function, wherein the first model training request message carries initial model information and a model training method.

[0166] In some implementations, the sending unit 503 is further configured to send a first model training status message to the consumer.

[0167] In some implementations, the receiving unit 501 is further used to receive a second model subscription request message sent by a consumer, and update or terminate model training based on the second model subscription request message.

[0168] Those skilled in the art should understand that Figure 5 The implementation functions of each unit in the model training device shown can be understood by referring to the relevant description of the aforementioned method. Figure 5 The functions of each unit in the model training device shown can be implemented by a program running on a processor, or by a specific logic circuit.

[0169] Figure 6 This is a schematic diagram of the structure of the model training device provided in the embodiment of the present application. Figure 2 , applied to the second network analysis function, such as Figure 6 As shown, the model training device 600 includes:

[0170] The sending unit 601 is used to send first information to the first network analysis function, wherein the first information includes model information and identification information of the second network analysis function; and send an update request message to the first network analysis function, wherein the update request message carries the identification information of the second network analysis function.

[0171] The receiving unit 602 is used to receive second information sent by the first network analysis function based on the identification information, where the second information is calculated by the first network analysis function according to the first information; the second information is used to update the local model of the second network analysis function.

[0172] In some implementations, the receiving unit 602 is further configured to receive an activation request message sent by the first network analysis function, and determine a health status report and a training strategy based on the activation request message.

[0173] In some implementations, the sending unit 601 is further configured to send an activation response message to the first network analysis function, wherein the activation response message carries identification information of the second network analysis function, the health status report, and the training strategy.

[0174] In some implementations, the receiving unit 602 is further used to receive a first model training request message sent by the first network analysis function, where the first model training request message carries initial model information and a model training method.

[0175] In some implementations, the model training device 600 further includes a collection unit 603 and a training unit 604 .

[0176] In some implementations, the acquisition unit 603 is used to acquire local data and perform sample alignment processing on the local data.

[0177] In some implementations, the training unit 604 trains a local model of the second network analysis function based on the initial model information, the model training method, and the local data after sample alignment processing.

[0178] Those skilled in the art should understand that Figure 6 The implementation functions of each unit in the model training device shown can be understood by referring to the relevant description of the aforementioned method. Figure 6 The functions of each unit in the model training device shown can be implemented by a program running on a processor, or by a specific logic circuit.

[0179] Figure 7 It is a schematic structural diagram of a network device 700 provided in an embodiment of the present application. Figure 7 The network device 700 shown includes a processor 710, which can call and run a computer program from a memory to implement the method in the embodiment of the present application.

[0180] Alternatively, if Figure 7 As shown, the network device 700 may further include a memory 720. The processor 710 may call and run a computer program from the memory 720 to implement the method in the embodiment of the present application.

[0181] The memory 720 may be a separate device independent of the processor 710 , or may be integrated into the processor 710 .

[0182] Alternatively, if Figure 7 As shown, the network device 700 may further include a transceiver 730, and the processor 710 may control the transceiver 730 to communicate with other devices, specifically, may send information or data to other devices, or receive information or data sent by other devices.

[0183] The transceiver 730 may include a transmitter and a receiver. The transceiver 730 may further include an antenna, and the number of the antennas may be one or more.

[0184] Optionally, the network device 700 may specifically be the network device of the embodiment of the present application, and the network device 700 may implement the corresponding processes implemented by the network device in each method of the embodiment of the present application, which will not be described in detail here for the sake of brevity.

[0185] Optionally, the network device 700 may specifically be a mobile terminal / terminal device of an embodiment of the present application, and the network device 700 may implement the corresponding processes implemented by the mobile terminal / terminal device in each method of the embodiment of the present application, which will not be described in detail here for the sake of brevity.

[0186] It should be understood that the processor of the embodiment of the present application may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method embodiment can be completed by the hardware integrated logic circuit or software instructions in the processor. The above processor can be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or the hardware and software modules in the decoding processor are combined and performed. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, and other mature storage media in the art. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0187] It can be understood that the memory in the embodiments of the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0188] It should be understood that the above-mentioned memory is exemplary but not restrictive. For example, the memory in the embodiments of the present application may also be static random access memory (static RAM, SRAM), dynamic random access memory (dynamic RAM, DRAM), synchronous dynamic random access memory (synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (double data rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (synch link DRAM, SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DR RAM), etc. That is to say, the memory in the embodiments of the present application is intended to include but not limited to these and any other suitable types of memory.

[0189] An embodiment of the present application also provides a computer-readable storage medium for storing a computer program.

[0190] Optionally, the computer-readable storage medium can be applied to the network device in the embodiments of the present application, and the computer program enables the computer to execute the corresponding processes implemented by the network device in the various methods of the embodiments of the present application. For the sake of brevity, they are not repeated here.

[0191] Optionally, the computer-readable storage medium can be applied to the mobile terminal / terminal device in the embodiments of the present application, and the computer program enables the computer to execute the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of the present application. For the sake of brevity, they are not repeated here.

[0192] An embodiment of the present application also provides a computer program product, including computer program instructions.

[0193] Optionally, the computer program product can be applied to the network device in the embodiments of the present application, and the computer program instructions enable the computer to execute the corresponding processes implemented by the network device in the various methods of the embodiments of the present application. For the sake of brevity, they are not repeated here.

[0194] Optionally, the computer program product can be applied to the mobile terminal / terminal device in the embodiments of the present application, and the computer program instructions enable the computer to execute the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of the present application. For the sake of brevity, they are not repeated here.

[0195] The embodiment of the present application also provides a computer program.

[0196] Optionally, the computer program can be applied to the network device in the embodiments of the present application. When the computer program runs on a computer, the computer executes the corresponding processes implemented by the network device in the various methods in the embodiments of the present application. For the sake of brevity, they are not described here.

[0197] Optionally, the computer program can be applied to the mobile terminal / terminal device in the embodiments of the present application. When the computer program runs on the computer, the computer executes the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of the present application. For the sake of brevity, they are not repeated here.

[0198] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0199] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0200] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0201] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0202] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0203] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0204] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A model training method, characterized in that: Applied to a first network analysis function, the method comprises: receiving first information sent by at least one second network analysis function, wherein the first information includes local model information and identification information of the second network analysis function; receiving an update request message sent by at least one second network analysis function, wherein the update request message carries identification information of the second network analysis function; The second information of each second network analysis function is calculated according to the first information, and the second information is sent to the corresponding second network analysis function based on the identification information of each second network analysis function, where the second information is used to update the local model of the second network analysis function.

2. The method according to claim 1, characterized in that: The method further comprises: Configure a first preset period, where the first preset period is used to determine the number T of second network analysis functions reported, where T is an integer greater than 1; After receiving T first information respectively reported by T second network analysis functions, the local model of the first network analysis function is updated, and the local model of the first network analysis function and the local model of the second network analysis function constitute a first model.

3. The method according to claim 2, characterized in that After receiving the update request message sent by at least one second network analysis function, the method further includes: After receiving the update request messages sent by the T second network analysis functions, calculating the second information of the T second network analysis functions; Accordingly, the step of calculating the second information of each second network analysis function according to the first information, and sending the second information to the corresponding second network analysis function based on the identification information of each second network analysis function includes: Querying corresponding first information based on identification information carried by T update request messages; Second information of T second network analysis functions is calculated based on the corresponding first information, and the second information is sent to the T second network analysis functions based on the corresponding identification information.

4. The method according to claim 2, characterized in that: The method further comprises: In each first preset period, first information reported by the T second network analysis functions is stored, where the first information includes identification information of the T second network analysis functions.

5. The method according to claim 4, characterized in that The method further comprises: Determining the activity of the second network analysis function based on the amount of identification information of the second network analysis function stored; If the activity of the second network analysis function is lower than the first threshold, an activation request message is sent to the second network analysis function, and an activation response message sent by the second network analysis function is received, wherein the activation response message carries the identification information, health status report and training strategy of the second network analysis function.

6. The method according to any one of claims 1 to 5, characterized in that Before receiving the first information sent by the second network analysis function, the method further includes: receiving a first model subscription request message sent by a consumer, and determining a second network analysis function participating in model training based on the first model subscription request message; A first model training request message is sent to the second network analysis function, where the first model training request message carries initial model information and a model training method.

7. The method according to any one of claims 1 to 5, characterized in that The method further comprises: Sending a first model training status message to the consumer; Receive a second model subscription request message sent by the consumer, and update or terminate model training based on the second model subscription request message.

8. A model training method, characterized in that: Applied to the second network analysis function, the method comprises: Sending first information to the first network analysis function, the first information including local model information and identification information of the second network analysis function; Sending an update request message to the first network analysis function, where the update request message carries identification information of the second network analysis function; Receive second information sent by the first network analysis function based on the identification information, where the second information is calculated by the first network analysis function according to the first information; the second information is used to update a local model of the second network analysis function.

9. The method according to claim 8, characterized in that The method further comprises: receiving an activation request message sent by the first network analysis function, and determining a health status report and a training strategy based on the activation request message; An activation response message is sent to the first network analysis function, where the activation response message carries identification information of the second network analysis function, the health status report, and the training strategy.

10. The method according to claim 8, characterized in that Before sending the first information to the first network analysis function, the method further includes: Receiving a first model training request message sent by the first network analysis function, where the first model training request message carries initial model information and a model training method; Collecting local data, and performing sample alignment processing on the local data; The local model of the second network analysis function is trained based on the initial model information, the model training method and the local data after sample alignment processing, and the local model of the second network analysis function and the local model of the first network analysis function constitute a first model.

11. A model training device, characterized in that: Applied to the first network analysis function, the device comprises: A receiving unit, receiving first information sent by at least one second network analysis function, wherein the first information includes model information and identification information of the second network analysis function; receiving an update request message sent by at least one second network analysis function, wherein the update request message carries the identification information of the second network analysis function; a calculation unit, configured to calculate second information of each second network analysis function according to the first information; A sending unit is used to send the second information to the corresponding second network analysis function based on the identification information of each second network analysis function, where the second information is used to update the local model of the second network analysis function.

12. A model training device, characterized in that: Applied to the second network analysis function, the device comprises: a sending unit, configured to send first information to the first network analysis function, wherein the first information includes model information and identification information of the second network analysis function; and send an update request message to the first network analysis function, wherein the update request message carries the identification information of the second network analysis function; A receiving unit is used to receive second information sent by the first network analysis function based on the identification information, where the second information is calculated by the first network analysis function according to the first information.

13. A network device, characterized in that: include: A processor and a memory, the memory being used to store a computer program, the processor being used to call and run the computer program stored in the memory to execute the method according to any one of claims 1 to 10.

14. A computer-readable storage medium, characterized in that: Used to store a computer program, wherein the computer program causes a computer to execute the method according to any one of claims 1 to 10.

15. A computer program product, characterized in that The method comprises computer program instructions which cause a computer to execute the method as claimed in any one of claims 1 to 10.

Images