Federal model training effect prediction method and device, electronic equipment and storage medium

Abstract

Embodiments of the present application provide a federal model training effect prediction method and device, electronic equipment and storage medium, relating to the technical field of computer. The method comprises: receiving a network quality parameter prediction request sent by an application server; obtaining each network quality parameter prediction result of each first terminal according to the identification information of each first terminal and the model training information of each first terminal; sending each network quality parameter prediction result of each first terminal to the application server, so that the application server determines each second terminal according to the model transmission delay of each first terminal, and obtains a federal model training effect prediction result according to each network quality parameter prediction result of each second terminal. Before the federal model training starts, the network data is analyzed, and the training effect of the model is predicted before the federal model training starts, so that the personnel in the field can adjust the training setting of the model according to the predicted training effect, and the training efficiency and quality of the federal model can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computers, in particular, the present application relates to a federated model training effect prediction method and device, electronic equipment and storage medium. BACKGROUND

[0002] Federated model training refers to each terminal participating in federated model training deriving a local model according to local data, and then each terminal sending the local model derived by itself to an application server. The application server combines the local models of each terminal to train a federated model. Using a federated model can enable each terminal to train data jointly without sharing data resources, and the data does not leave the local terminal. Therefore, the federated model ensures good data security and is widely deployed and applied in various industries.

[0003] Since the local computing capabilities and network communication capabilities of each terminal participating in federated model training are different, they can affect the training efficiency during the entire federated model training process. Specifically, the difference between the local computing latency of different terminals and the latency of sending their own models to the application server can be large. Since the models of all terminals need to be sent to the application server before the federated model can be obtained, this can result in longer overall training delay, which in turn can result in poor training effect of the federated model (e.g., unable to complete training within the expected time length). To solve the above problem, the prior art generally optimizes the training process of the federated model, such as reducing the number of iterations, grouping the terminals for training, and the like, to reduce the training time of the federated model and optimize the training effect of the federated model.

[0004] However, the prior art solution only optimizes the training process of the federated model, but in actual operation, sometimes even if the training process of the model is optimized, the overall training effect is still not ideal (e.g., when the overall training delay is too long, even if the training time of the model is reduced, the training of the federated model still cannot be completed within the expected time length). Therefore, a new way is needed to predict the training effect of the federated model to further improve the federated learning process. SUMMARY

[0005] The present application aims to at least solve one of the above technical defects, and the technical solution provided by the embodiments of the present application is as follows:

[0006] In a first aspect, the embodiments of the present application provide a federated model training effect prediction method, comprising:

[0007] receive, by the network exposure function module, a network quality parameter prediction request for at least one first terminal sent by an application server; wherein the network quality parameter prediction request contains identification information of each first terminal and model training information of each first terminal;

[0008] obtain, according to the identification information of each first terminal and the model training information of each first terminal, a network quality parameter prediction result of each first terminal; wherein the network quality parameter contains a model transmission delay;

[0009] send, by the network exposure function module, the network quality parameter prediction result of each first terminal to the application server, so that the application server determines at least one second terminal according to the model transmission delay of each first terminal, and obtains a federated model training effect prediction result according to the network quality parameter prediction result of each second terminal.

[0010] In an optional embodiment of the present application, obtaining, according to the identification information of each first terminal and the model training information of each first terminal, the network quality parameter prediction result of each first terminal specifically includes:

[0011] For each first terminal, obtaining, by the operation maintenance management network element, the signal receiving power of the first terminal according to the identification information of the first terminal, obtaining, by the policy control network element, the network quality of the first terminal according to the identification information of the first terminal, and obtaining, by the access and mobility management function network element, the registered slice information of the first terminal according to the identification information of the first terminal;

[0012] obtaining, according to the signal receiving power of the first terminal, the network quality of the first terminal, the registered slice information of the first terminal and the model training information of each first terminal, the network quality parameter prediction result of the first terminal.

[0013] In an optional embodiment of the present application, the network quality parameter further contains uplink and downlink bandwidth demand and traffic data in the federated model training process; and the model training information contains the local model size of the first terminal.

[0014] obtaining, according to the signal receiving power of the first terminal, the network quality of the first terminal, the registered slice information of the first terminal and the model training information of each first terminal, the network quality parameter prediction result of the first terminal specifically includes:

[0015] For each first terminal, determining the model transmission delay of the first terminal according to the local model size of the first terminal, determining the uplink and downlink bandwidth demand of the first terminal according to the signal receiving power of the first terminal, and determining the traffic data of the first terminal in the federated model training process according to the network quality of the first terminal and the registered slice information of the first terminal.

[0016] In a second aspect, the embodiments of the present application provide a method for predicting training effect of a federated model, comprising:

[0017] sending, by the network exposure function module, a network quality parameter prediction request for at least one first terminal to the network data analysis function module, wherein the network quality parameter prediction request comprises identification information of each first terminal and model training information of each first terminal;

[0018] receiving each network quality parameter prediction result of each first terminal sent by the network data analysis function module through the network exposure function module, wherein each network quality parameter prediction result is obtained by the network data analysis function module according to the identification information of each first terminal and the model training information of each first terminal, and the network quality parameter comprises a model transmission delay;

[0019] determining at least one second terminal according to the model transmission delay of each first terminal, and obtaining a federated model training effect prediction result according to each network quality parameter prediction result of each second terminal.

[0020] In an optional embodiment of the present application, determining at least one second terminal according to the model transmission delay of each first terminal specifically comprises:

[0021] obtaining a model transmission delay standard interval according to the average value of the prediction result of the model transmission delay of each first terminal;

[0022] taking the first terminal whose prediction result of the model transmission delay is within the model transmission delay standard interval as the second terminal.

[0023] In an optional embodiment of the present application, obtaining a federated model training effect prediction result according to each network quality parameter prediction result of each second terminal comprises:

[0024] for each network quality parameter, calculating an average value of the network quality parameter prediction result according to the network quality parameter prediction result of each second terminal;

[0025] calculating a first difference value according to the average value of the network quality parameter prediction result and a standard value of the network quality parameter prediction result, and calculating a second difference value according to the first difference value and a preset weight of the network quality parameter;

[0026] obtaining the federated model training effect prediction result according to the second difference value of each network quality parameter.

[0027] In an optional embodiment of the present application, the method specifically further comprises:

[0028] The federated model training effect prediction result is compared with a preset threshold value, if the federated model training effect prediction result is less than the preset threshold value, the training step of the federated model is delayed; wherein the preset threshold value represents an acceptable federated model training effect;

[0029] If the federated model training effect prediction result is greater than the preset threshold value, the training step of the federated model is started.

[0030] In an optional embodiment of the present application, after the training step of the federated model is delayed, specifically comprising:

[0031] At least one third terminal participating in the federated model training is obtained, and each third terminal and each first terminal are simultaneously used as a first terminal to send a network quality parameter prediction request for at least one first terminal to the network data analysis function module through the network exposure function module again.

[0032] In a third aspect, the embodiments of the present application provide a federated model training effect prediction device, comprising:

[0033] A prediction request receiving module is configured to receive a network quality parameter prediction request for at least one first terminal sent by an application server through a network exposure function module; wherein the network quality parameter prediction request contains identification information of each first terminal and model training information of each first terminal;

[0034] A network parameter prediction result obtaining module is configured to obtain each network quality parameter prediction result of each first terminal according to the identification information of each first terminal and the model training information of each first terminal; wherein the network quality parameter contains a model transmission delay;

[0035] A network parameter prediction result sending module is configured to send each network quality parameter prediction result of each first terminal to the application server through the network exposure function module, so that the application server determines at least one second terminal according to the model transmission delay of each first terminal, and obtains a federated model training effect prediction result according to each network quality parameter prediction result of each second terminal.

[0036] In an optional embodiment of the present application, the network parameter prediction result obtaining module is specifically configured to:

[0037] For each first terminal, the signal receiving power of the first terminal is obtained by the operation, maintenance and management network element according to the identification information of the first terminal, the network quality of the first terminal is obtained by the policy control network element according to the identification information of the first terminal, and the registered slice information of the first terminal is obtained by the access and mobile management function network element according to the identification information of the first terminal;

[0038] The network quality parameter prediction result of the first terminal is obtained according to the signal receiving power of the first terminal, the network quality of the first terminal, the registered slice information of the first terminal, and the model training information of each first terminal.

[0039] In an optional embodiment of the present application, the network quality parameter further includes uplink and downlink bandwidth demand and traffic data in the federated model training process; and the model training information includes the local model size of the first terminal.

[0040] The network parameter prediction result obtaining module is further configured to:

[0041] For each first terminal, the model transmission delay of the first terminal is determined according to the local model size of the first terminal; the uplink and downlink bandwidth demand of the first terminal is determined according to the signal receiving power of the first terminal; and the traffic data of the first terminal in the federated model training process is determined according to the network quality of the first terminal and the registered slice information of the first terminal.

[0042] In a fourth aspect, the embodiments of the present application provide a federated model training effect prediction device, which comprises:

[0043] The prediction request sending module is configured to send a network quality parameter prediction request for at least one first terminal to the network data analysis function module through the network exposure function module; wherein the network quality parameter prediction request includes the identification information of each first terminal and the model training information of each first terminal.

[0044] The network parameter prediction result receiving module is configured to receive each network quality parameter prediction result of each first terminal sent by the network data analysis function module through the network exposure function module; wherein each network quality parameter prediction result is obtained by the network data analysis function module according to the identification information of each first terminal and the model training information of each first terminal; and the network quality parameter includes the model transmission delay.

[0045] The training effect prediction module is configured to determine at least one second terminal according to the model transmission delay of each first terminal, and obtain a federated model training effect prediction result according to each network quality parameter prediction result of each second terminal.

[0046] In an optional embodiment of the present application, the training effect prediction module is specifically configured to:

[0047] Obtain a model transmission delay standard interval according to the average value of the prediction results of the model transmission delay of each first terminal.

[0048] The first terminal whose prediction result of the model transmission delay is within the model transmission delay standard interval is taken as the second terminal.

[0049] In an optional embodiment of the present application, the training effect prediction module can be further configured to:

[0050] For each network quality parameter, calculate a network quality parameter prediction result average value according to the network quality parameter prediction results of the second terminals;

[0051] Calculate a first difference value according to the network quality parameter prediction result average value and a network quality parameter prediction result standard value, and calculate a second difference value according to the first difference value and a preset weight of the network quality parameter;

[0052] Obtain a federated model training effect prediction result according to the second difference values of the network quality parameters.

[0053] In an optional embodiment of the present application, the device further comprises a model training module configured to:

[0054] Compare the federated model training effect prediction result with a preset threshold value, if the federated model training effect prediction result is less than the preset threshold value, delay the training step of the federated model; wherein the preset threshold value represents an acceptable federated model training effect;

[0055] If the federated model training effect prediction result is greater than the preset threshold value, start the training step of the federated model.

[0056] In an optional embodiment of the present application, the device further comprises a terminal obtaining module configured to:

[0057] Obtain at least one third terminal participating in the federated model training, and use each third terminal and each first terminal as a first terminal again to send a network quality parameter prediction request for at least one first terminal to the network data analysis function module through the network exposure function module.

[0058] In a fifth aspect, the embodiments of the present application provide an electronic device, comprising a memory, a processor and a computer program stored in the memory;

[0059] The processor executes the computer program to implement the method provided in the first aspect or any optional embodiment of the first aspect.

[0060] In a sixth aspect, the embodiments of the present application provide a computer readable storage medium, and the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the method provided in the first aspect or any optional embodiment of the first aspect.

[0061] The technical scheme provided by the embodiments of the present application has the beneficial effects that:

[0062] The scheme provided in the application sends a network quality parameter prediction request for each terminal to the network data analysis function module before the start of federal model training, and predicts the federal model training effect according to the returned prediction result. Compared with the prior art, the scheme provided in the application analyzes network data, can predict the training effect of the model before the start of federal model training, and enables the person skilled in the art to adjust the model training setting according to the predicted training effect, thereby effectively improving the training efficiency of the federal model and the quality of the federal model. BRIEF DESCRIPTION OF DRAWINGS

[0063] In order to more clearly illustrate the technical solutions in the embodiments of the application, the drawings needed to be used in the description of the embodiments of the application will be briefly introduced.

[0064] Figure 1 A flowchart of a federal model training effect prediction method provided by the embodiments of the application;

[0065] Figure 2 A flowchart of how to obtain network data in an example of the embodiments of the application;

[0066] Figure 3 A flowchart of a federal model training effect prediction method provided by the embodiments of the application;

[0067] Figure 4 A flowchart of a federal model training prediction process in an example of the embodiments of the application;

[0068] Figure 5 A structure block diagram of a federal model training effect prediction device provided by the embodiments of the application;

[0069] Figure 6 A structure block diagram of a federal model training effect prediction device provided by the embodiments of the application;

[0070] Figure 7 A structure diagram of an electronic device provided by the embodiments of the application. DETAILED DESCRIPTION

[0071] The embodiments of the application will be described below in conjunction with the drawings in the application. It should be understood that the embodiments described below in conjunction with the drawings are exemplary descriptions of the technical solutions of the embodiments of the application, and do not limit the technical solutions of the embodiments of the application.

[0072] Those skilled in the art can understand that the singular forms "a", "an" and "the" used herein include plural forms unless specifically stated otherwise. It should be further understood that the terms "comprise" and "include" used in the embodiments of the present application mean that the corresponding features can be implemented as the presented features, information, data, steps, operations, elements and / or components, but do not exclude implementation as other features, information, data, steps, operations, elements, components and / or combinations thereof supported by the present technology. It should be understood that when we say that an element is "connected" or "coupled" to another element, the element can be directly connected or coupled to the other element, or it can mean that the element and the other element establish a connection relationship through an intermediate element. In addition, "connected" or "coupled" used herein can include wireless connection or wireless coupling. The term "and / or" used herein indicates that at least one of the items defined by the term, for example, "A and / or B" can be implemented as "A", or as "B", or as "A and B".

[0073] For the purpose, technical solutions and advantages of the present application to be clearer, the embodiments of the present application will be described in further detail below with reference to the accompanying drawings.

[0074] The technical solutions of the embodiments of the present application and the technical effects produced by the technical solutions of the present application will be described below through the description of several exemplary embodiments. It should be noted that the following embodiments can be mutually referenced, borrowed or combined, and the same terms, similar features and similar implementation steps in different embodiments will not be described repeatedly.

[0075] Figure 1 A flowchart of a federal model training effect prediction method is provided for the embodiments of the present application. The execution subject of the method can be a network data analysis function module (a module in a 5GS (5th Generation Mobile Communication Technology System) that can provide network analysis services according to request data of network services), as shown in Figure 1 The method can include the following steps:

[0076] In step S101, a network exposure function module receives a network quality parameter prediction request for at least one first terminal sent by an application server; wherein the network quality parameter prediction request contains identification information of each first terminal and model training information of each first terminal.

[0077] The network exposure function module (Network Exposure Function, NEF for short) is a module for exposing network functions to third-party applications. The application server is a server for training a federated model. Each terminal can send a locally derived local model to the application server, and the application server integrates each local model to obtain the federated model. The first terminal can be a terminal participating in federated model training. The identification information of the first terminal can be the ID (Identity document) information of the first terminal, and the model training information of the first terminal can be the information of the local model derived by the first terminal (such as the size of the local model).

[0078] Specifically, before the federated model training, the training effect of the federated model needs to be predicted. In the embodiment of the present application, the network data of each first terminal can be collected with the assistance of the 5GS communication system, and the network quality parameters can be predicted. Then, the training effect of the federated model is predicted according to the prediction result of the network quality parameters. With the assistance of the 5G communication system, the application server can send a network quality parameter prediction request for all first terminals to the network data analysis function module in the NEF of the 5G communication system, and the identification information and model training information of each first terminal contained in the request can be collected by the application server in advance.

[0079] In step S102, each network quality parameter prediction result of each first terminal is obtained according to the identification information of each first terminal and the model training information of each first terminal. The network quality parameters include model transmission delay.

[0080] The network quality parameter prediction result can be a corresponding result obtained by the network data analysis function module according to the read network data or the model training information of the first terminal. The model transmission delay represents the time consumed from the start of sending the local model by the first terminal to the complete reception of the local model by the application server.

[0081] Specifically, the network data analysis function module can access each network element in the 5G communication system according to the identification information of each first terminal to read each network data of each first terminal, and then predict each network quality parameter prediction result according to the read network data, or calculate the model transmission delay of each first terminal according to the model training information of each first terminal, such as the size of the local model.

[0082] It should be noted that the network quality parameters include but are not limited to the model transmission delay, and other factors that can represent the network quality can also be used as the network quality parameters in the embodiment of the present application.

[0083] In step S103, the network quality parameter prediction result of each first terminal is sent to the application server through the network exposure function module, so that the application server determines at least one second terminal according to the model transmission time delay of each first terminal, and obtains the federated model training effect prediction result according to the network quality parameter prediction result of each second terminal.

[0084] The second terminal can be selected from the first terminals, and specifically, the first terminal whose model transmission time delay meets the condition can be taken as the second terminal. The federated model training effect prediction result can be a numerical value, and the size of the numerical value can represent the good or bad of the federated model training effect. For example, the larger the numerical value is, the better the predicted federated model training effect is.

[0085] Specifically, the network data analysis function module returns the network quality parameter result of each first terminal to the application server through the NEF, and the application server selects at least one second terminal from the first terminals according to the model transmission time delay in the network quality parameter of each first terminal, and finally calculates the federated model training effect prediction result according to the network quality parameter prediction result of each second terminal.

[0086] The scheme provided in the present application sends a network quality parameter prediction request for each terminal to the network data analysis function module before the federated model training starts, and predicts the federated model training effect according to the returned prediction result. Compared with the prior art, the scheme in the present application analyzes the network data, can predict the training effect of the model before the federated model training starts, and enables the person skilled in the art to adjust the model training setting according to the predicted training effect, which can effectively improve the training efficiency of the federated model and the quality of the federated model.

[0087] In an optional embodiment of the present application, the network quality parameter prediction result of each first terminal is obtained according to the identification information of each first terminal and the model training information of each first terminal, and specifically includes:

[0088] For each first terminal, the signal receiving power of the first terminal is obtained by the operation, maintenance and management network element according to the identification information of the first terminal, the network quality of the first terminal is obtained by the policy control network element according to the identification information of the first terminal, and the registered slice information of the first terminal is obtained by the access and mobility management function network element according to the identification information of the first terminal.

[0089] The network quality parameter prediction result of the first terminal is obtained according to the signal receiving power of the first terminal, the network quality of the first terminal, the registered slice information of the first terminal and the model training information of each first terminal.

[0090] The OAM (Operation Administration and Maintenance) network element stores the signal receiving power of each first terminal in the embodiment of the application. The PCF (Policy Control Function) network element stores the Qos (Quality of Service) configuration of each first terminal in the embodiment of the application, and the configuration can represent the network quality of each first terminal. The AMF (Access and Mobility Management Function) network element stores the registered slice information of each first terminal in the embodiment of the application, and the slice information represents the performance parameters that can be provided to the first terminal.

[0091] Specifically, the network data analysis function module obtains the network data of each first terminal through each network element in the 5G communication system. Specifically, the network data stored in each network element can be read according to the identification information of each first terminal. Then, the prediction result of each network quality parameter of each first terminal is predicted according to the read network data.

[0092] It should be noted that, in the embodiment of the application, before the network data is collected through the above network element, it can also be determined through the UDM (Unified Data Management) network element whether each first terminal agrees to collect, distribute and analyze the related data of the first terminal. If the first terminal does not agree to collect the related data, the related network data of the first terminal will not be collected through other network elements.

[0093] The process of obtaining each network data is briefly introduced below, which can be specifically divided into the following steps. Figure 2As shown, 1, the AF (Application Function, application server) first sends a network quality parameter prediction request to the NEF. 2, the NEF forwards the request to the NWDAF (Network Data Analytics Function). 3.1, the NWDAF determines whether each first terminal agrees to collect, distribute and analyze relevant data of the first terminal by accessing the UDM network element. 3.2, for the first terminal that agrees to collect, the NWDAF collects the RSRP (Reference Signal Receiving Power) of each first terminal through the OAM network element. 3.3, the NWDAF collects the network quality information of each first terminal through the PCF network element. 3.4, the NWDAF collects the registered slice information of each first terminal through the AMF network element. 4, the NWDAF performs data analysis on the collected network data, and returns the analysis result of the network quality parameter prediction to the application server through the NEF.

[0094] In an optional embodiment of the present application, the network quality parameter further comprises uplink and downlink bandwidth requirement, and traffic data in the federated model training process; the model training information comprises the local model size of the first terminal;

[0095] The network quality parameter prediction result of the first terminal is obtained according to the signal receiving power of the first terminal, the network quality of the first terminal, the registered slice information of the first terminal, and the model training information of each first terminal, specifically comprising:

[0096] For each first terminal, the model transmission delay of the first terminal is determined according to the local model size of the first terminal; the uplink and downlink bandwidth requirement of the first terminal is determined according to the signal receiving power of the first terminal; and the traffic data of the first terminal in the federated model training process is determined according to the network quality of the first terminal and the registered slice information of the first terminal.

[0097] The uplink and downlink bandwidth requirement represents the speed requirement of uploading data to the Internet and the speed requirement of downloading data from the Internet. The traffic data in the federated model training process represents the total traffic consumed by the first terminal in the process of transmitting the local model to the application server.

[0098] Specifically, in the embodiment of the present application, the network quality parameters can include model transmission delay, uplink and downlink bandwidth requirements, and traffic data in the federated model training process. The model transmission delay can be determined by the local model size in the model training information; the uplink and downlink bandwidth requirements can be determined according to the signal received power obtained in the above steps; the traffic data in the federated model training process can be determined according to the network quality and the registered slice information obtained in the above steps. The specific determination method is not limited in the embodiment of the present application.

[0099] Figure 3 A flowchart of a federated model training effect prediction method is provided in the embodiment of the present application, and the execution subject of the method can be an application server, as shown in the figure, the method can include: Figure 3

[0100] Step S201, sending a network quality parameter prediction request for at least one first terminal to a network data analysis function module through a network exposure function module; wherein the network quality parameter prediction request contains the identification information of each first terminal and the model training information of each first terminal.

[0101] The network exposure function module (Network Exposure Function, hereinafter referred to as NEF) is a module that can be used to open network functions to third-party applications. The application server is a server used to train the federated model, and each terminal will send the locally derived local model to the application server, and the application server will integrate each local model to obtain the federated model. The first terminal can be a terminal participating in the federated model training. The identification information of the first terminal can be the ID (Identity document, number) information of the first terminal, and the model training information of the first terminal can be the information of the local model derived by the first terminal (such as the size of the local model, etc.).

[0102] Specifically, before the federated model training, the training effect of the federated model needs to be predicted, in the embodiment of the present application, the network data of each first terminal can be collected by means of 5GS communication system and the network quality parameters are predicted, and then the training effect of the federated model is predicted according to the prediction result of the network quality parameters. The process assisted by 5G communication system can be that the application server sends a network quality parameter prediction request for all first terminals to the network data analysis function module through the NEF in the 5G communication system, and the identification information and model training information of each first terminal contained in the request can be collected by the application server in advance.

[0103] ​In step S202, the network data analysis function module receives the network quality parameter prediction result of each first terminal sent by the network exposure function module. The network quality parameter prediction result is obtained by the network data analysis function module according to the identification information of each first terminal and the model training information of each first terminal. The network quality parameter includes the model transmission delay.

[0104] The network quality parameter prediction result can be a corresponding result obtained by the network data analysis function module according to the read network data or the model training information of the first terminal. The model transmission delay represents the time consumed from the start of sending the local model by the first terminal to the complete reception of the local model by the application server.

[0105] Specifically, the network data analysis function module can access each network element in the 5G communication system according to the identification information of each first terminal to read each network data of each first terminal, and then predict each network quality parameter prediction result according to the read network data, or calculate the model transmission delay of each first terminal according to the model training information of each first terminal, such as the size of the local model.

[0106] It should be noted that the network quality parameter includes but is not limited to the model transmission delay, and other factors that can represent the network quality can also be used as the network quality parameter in the embodiments of the present application.

[0107] In step S203, at least one second terminal is determined according to the model transmission delay of each first terminal, and a federated model training effect prediction result is obtained according to the network quality parameter prediction result of each second terminal.

[0108] The second terminal can be selected from the first terminal. Specifically, the first terminal whose model transmission delay meets the condition can be used as the second terminal. The federated model training effect prediction result can be a numerical value, and the size of the numerical value can represent the good or bad of the federated model training effect. For example, the larger the numerical value is, the better the predicted federated model training effect is.

[0109] Specifically, the network data analysis function module returns the network quality parameter result of each first terminal to the application server through the NEF, and the application server selects at least one second terminal from each first terminal according to the model transmission delay in the network quality parameter of each first terminal, and finally calculates the federated model training effect prediction result according to the network quality parameter prediction result of each second terminal.

[0110] The scheme provided in the application sends a network quality parameter prediction request for each terminal to the network data analysis function module before the start of federal model training, and predicts the federal model training effect according to the returned prediction result. Compared with the prior art, the scheme of the application analyzes network data to predict the training effect of the model before the start of federal model training, so that the personnel in the field can adjust the model training setting process according to the predicted training effect, thereby effectively improving the training efficiency of the federal model and the quality of the federal model.

[0111] In an optional embodiment of the application, the at least one second terminal is determined according to the model transmission delay of each first terminal, specifically comprising:

[0112] The model transmission delay standard interval is obtained according to the average value of the prediction results of the model transmission delay of each first terminal.

[0113] The first terminal whose prediction result of the model transmission delay is within the model transmission delay standard interval is taken as the second terminal.

[0114] The model transmission delay standard interval can be obtained according to the average value of the prediction results of the model transmission delay of each first terminal. Specifically, a constant t can be set, and then [average value of prediction results-t, average value of prediction results+t] is taken as the model transmission delay standard interval.

[0115] Specifically, since the difference in model transmission delay of different terminals can be large, each first terminal needs to be screened before the model starts training, and the first terminal whose model transmission delay is greatly different from other first terminals is excluded from the model training process. In the embodiment of the application, the average value of the prediction results of the model transmission delay of each first terminal is taken as a reference factor, and the first terminal whose model transmission delay is close to the average value of the prediction results will be selected as the second terminal to participate in subsequent federal model training. Specifically, a related constant t can be set according to the experience of historical training data, and the model transmission delay standard interval can be determined according to the constant t and the average value of the prediction results, and then the first terminal whose model transmission delay belongs to the model transmission delay standard interval is taken as the second terminal.

[0116] In an optional embodiment of the application, the federal model training effect prediction result is obtained according to the network quality parameter prediction result of each second terminal, comprising:

[0117] For each network quality parameter, the average value of the network quality parameter prediction result is calculated according to the network quality parameter prediction result of each second terminal.

[0118] The first difference value is calculated according to the average value of the network quality parameter prediction result and the standard value of the network quality parameter prediction result, and the second difference value is calculated according to the first difference value and a preset weight of the network quality parameter;

[0119] The federated model training effect prediction result is obtained according to the second difference values of the network quality parameters.

[0120] The standard value of the network quality parameter prediction result can be obtained through historical data and is used to represent the specific value of the network quality parameter when the federated model training effect is good. The first difference value can be obtained by subtracting the average value of the network quality parameter prediction result of each second terminal from the standard value of the network quality parameter prediction result. The second difference value can be calculated by multiplying the first difference value of the network quality parameter by a preset weight of the network quality parameter. The preset weight can also be obtained through historical data and is used to represent the influence of the network quality parameter on the federated model training effect.

[0121] Specifically, after the second terminals are selected, the federated model training effect can be predicted according to the network quality parameter prediction results of each second terminal. Specifically, after the second terminals are determined in the previous step, the average value of each network quality parameter prediction result can be calculated according to the network quality parameter prediction results of each second terminal. Then, the first difference value is calculated by subtracting the average value from the standard value corresponding to the network quality parameter. The second difference value corresponding to the network quality parameter is obtained by multiplying the first difference value by the preset weight corresponding to the network quality parameter. Finally, the federated model training effect prediction result can be determined according to the second difference values. For example, the federated model training effect prediction result can be calculated by adding the second difference values.

[0122] It should be noted that in the embodiments of the present application, since the local data of each second terminal is different, the local data of each second terminal can also be obtained, and the quality of the local data of each second terminal is also used as a reference factor for the federated model training effect.

[0123] In an optional embodiment of the present application, the method can further include:

[0124] The federated model training effect prediction result is compared with a preset threshold value. If the federated model training effect prediction result is less than the preset threshold value, the training step of the federated model is delayed. The preset threshold value represents an acceptable federated model training effect.

[0125] If the federated model training effect prediction result is greater than the preset threshold value, the training step of the federated model is started.

[0126] The preset threshold can be obtained according to historical data, and is used to represent a minimum value acceptable for a prediction result of the federated model training effect. When the prediction result is not less than the preset threshold, it indicates that the predicted federated model training effect is good. When the prediction result is less than the preset threshold, it indicates that the predicted federated model training effect is poor.

[0127] Specifically, the prediction result of the federated model training effect calculated in the previous step is compared with the preset threshold. If the prediction result is less than the preset threshold, it is considered that the predicted federated model training effect cannot achieve the expected effect. At this time, the training step of the federated model can be delayed to further adjust the training setting of the federated model. If the prediction result is greater than the preset threshold, it is considered that the predicted federated model training effect can achieve the expected effect. At this time, the second terminal screened in the previous step can be started as a federated model training participating terminal to start the training process of the federated model.

[0128] In an optional embodiment of the present application, delaying the training step of the federated model specifically includes:

[0129] At least one third terminal participating in the federated model training is obtained, and each third terminal and each first terminal are simultaneously used as a first terminal to send a network quality parameter prediction request for at least one first terminal to the network data analysis function module through the network exposure function module again.

[0130] The third terminal can be a new terminal, that is, a terminal newly added to the federated model training in addition to the first terminal.

[0131] Specifically, in the case where the prediction result is less than the preset threshold, the training setting of the current federated model needs to be adjusted again. In the embodiment of the present application, the members participating in the federated model training can be modified. Specifically, a new third terminal can be added, that is, a new member participating in the federated model training is added. Then, the current newly added third terminal and the first terminal participating in the federated model training are both used as a first terminal to send a network quality parameter prediction request to the network data analysis function module again, that is, the federated model training effect of each first terminal is predicted again by the application server and the network data analysis function module.

[0132] In summary, the scheme provided by the present application is described as a whole below, such as Figure 4As shown, before the application server starts the federated model training process, the application server can request the 5GS for assistance (i.e., send a network quality parameter prediction request to the network analysis function module), obtain the network quality parameter prediction result of each first terminal through the 5GS, and then the application server predicts the training effect of the federated model according to the network quality parameter prediction result of each first terminal. If the prediction result is less than the preset threshold, the training setting of the federated model can be adjusted and then the 5GS is requested again for assistance; if the prediction result is not less than the preset threshold, the training process of the federated model is directly started.

[0133] Figure 5 A structural block diagram of a federated model training effect prediction device provided by an embodiment of the present application is shown in FIG. 5. Figure 5 As shown, the federated model training effect prediction device 500 can include a prediction request receiving module 501, a network parameter prediction result obtaining module 502, and a network parameter prediction result sending module 503, wherein,

[0134] The prediction request receiving module 501 is configured to receive, through the network exposure function module, a network quality parameter prediction request for at least one first terminal sent by the application server; wherein the network quality parameter prediction request contains the identification information of each first terminal and the model training information of each first terminal.

[0135] The network parameter prediction result obtaining module 502 is configured to obtain the network quality parameter prediction result of each first terminal according to the identification information of each first terminal and the model training information of each first terminal; wherein the network quality parameter contains the model transmission delay.

[0136] The network parameter prediction result sending module 503 is configured to send the network quality parameter prediction result of each first terminal to the application server through the network exposure function module, so that the application server determines at least one second terminal according to the model transmission delay of each first terminal, and obtains the federated model training effect prediction result according to the network quality parameter prediction result of each second terminal.

[0137] The scheme provided by the present application sends a network quality parameter prediction request for each terminal to the network data analysis function module before the federated model training starts, and predicts the training effect of the federated model according to the returned prediction result. Compared with the prior art, the scheme of the present application analyzes the network data, can predict the training effect of the model before the federated model training starts, and enables the person skilled in the art to adjust the training setting of the model according to the predicted training effect, which can effectively improve the training efficiency of the federated model and the quality of the federated model.

[0138] In an optional embodiment of the present application, the network parameter prediction result obtaining module is specifically configured to:

[0139] For each first terminal, the signal receiving power of the first terminal is obtained by the operation and maintenance management network element according to the identification information of the first terminal, the network quality of the first terminal is obtained by the policy control network element according to the identification information of the first terminal, and the registered slice information of the first terminal is obtained by the access and mobility management function network element according to the identification information of the first terminal;

[0140] The network quality parameter prediction result of the first terminal is obtained according to the signal receiving power of the first terminal, the network quality of the first terminal, the registered slice information of the first terminal, and the model training information of each first terminal.

[0141] In an optional embodiment of the present application, the network quality parameter further includes uplink and downlink bandwidth demand and traffic data in the federated model training process, and the model training information includes the local model size of the first terminal.

[0142] The network parameter prediction result obtaining module is further used for:

[0143] For each first terminal, the model transmission delay of the first terminal is determined according to the local model size of the first terminal, the uplink and downlink bandwidth demand of the first terminal is determined according to the signal receiving power of the first terminal, and the traffic data of the first terminal in the federated model training process is determined according to the network quality of the first terminal and the registered slice information of the first terminal.

[0144] Figure 6 A structural block diagram of a federated model training effect prediction device provided by an embodiment of the present application is shown in FIG. 6. Figure 6 As shown in FIG. 6, the federated model training effect prediction device 600 can include a prediction request sending module 601, a network parameter prediction result receiving module 602, and a training effect prediction module 603, wherein,

[0145] The prediction request sending module 601 is used for sending a network quality parameter prediction request for at least one first terminal to a network data analysis function module through a network exposure function module; wherein the network quality parameter prediction request includes identification information of each first terminal and model training information of each first terminal.

[0146] The network parameter prediction result receiving module 602 is used for receiving each network quality parameter prediction result of each first terminal sent by the network data analysis function module through the network exposure function module; wherein each network quality parameter prediction result is obtained by the network data analysis function module according to the identification information of each first terminal and the model training information of each first terminal; and the network quality parameter includes a model transmission delay.

[0147] The training effect prediction module 603 is configured to determine at least one second terminal according to the model transmission time delay of each first terminal, and obtain a federated model training effect prediction result according to the network quality parameter prediction result of each second terminal.

[0148] The scheme provided in the present application sends a network quality parameter prediction request for each terminal to the network data analysis function module before the start of federated model training, and predicts the training effect of the federated model according to the returned prediction result. Compared with the prior art, the scheme provided in the present application analyzes network data to predict the training effect of the model before the start of federated model training, so that personnel in the field can adjust the training settings of the model according to the predicted training effect, thereby effectively improving the training efficiency of the federated model and the quality of the federated model.

[0149] In an optional embodiment of the present application, the training effect prediction module is specifically configured to:

[0150] obtain a model transmission time delay standard interval according to the average value of the prediction results of the model transmission time delay of each first terminal;

[0151] regard the first terminal whose prediction result of the model transmission time delay is within the model transmission time delay standard interval as a second terminal.

[0152] In an optional embodiment of the present application, the training effect prediction module can also be configured to:

[0153] for each network quality parameter, calculate a network quality parameter prediction result average value according to the network quality parameter prediction results of each second terminal;

[0154] calculate a first difference value according to the network quality parameter prediction result average value and a network quality parameter prediction result standard value, and calculate a second difference value according to the first difference value and a preset weight of the network quality parameter;

[0155] obtain a federated model training effect prediction result according to the second difference value of each network quality parameter.

[0156] In an optional embodiment of the present application, the device further comprises a model training module configured to:

[0157] compare the federated model training effect prediction result with a preset threshold value, if the federated model training effect prediction result is less than the preset threshold value, delay the training step of the federated model; wherein the preset threshold value represents an acceptable federated model training effect;

[0158] if the federated model training effect prediction result is greater than the preset threshold value, start the training step of the federated model.

[0159] In an optional embodiment of the present application, the device further comprises a terminal obtaining module configured to:

[0160] obtain at least one third terminal participating in the federated model training, and send, by each third terminal, a network quality parameter prediction request for the at least one first terminal to the network data analysis function module again through the network exposure function module as a first terminal at the same time as each first terminal.

[0161] Reference is made below to Figure 7 which shows a structural schematic diagram of an electronic device (for example, a terminal device or a server) 700 suitable for being used to implement the method shown in Figure 1 or Figure 2 embodiments of the present application. The electronic device in the embodiments of the present application can include, but is not limited to, a mobile terminal such as a mobile phone, a notebook computer, a digital broadcast receiver, a PDA (Personal Digital Assistant), a PAD (Tablet Personal Computer), a PMP (Portable Multimedia Player), a vehicle terminal (for example, a vehicle navigation terminal), a wearable device, and the like, and a fixed terminal such as a digital TV, a desktop computer, and the like. Figure 7 The electronic device shown is merely an example, and should not bring any limitation to the functions and use range of the embodiments of the present application.

[0162] The electronic device includes a memory and a processor, the memory is used to store a program for executing the methods described in each of the above method embodiments, and the processor is configured to execute the program stored in the memory. Wherein, the processor here can be referred to as the processing apparatus 701 described below, and the memory can include at least one of a read-only memory (ROM) 702, a random access memory (RAM) 703 and a storage device 708, as shown below:

[0163] As shown in Figure 7 , the electronic device 700 can include a processing apparatus (for example, a central processing unit, a graphics processing unit, etc.) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage device 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the electronic device 700 are also stored. The processing apparatus 701, the ROM 702 and the RAM 703 are connected to each other through a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0164] In general, the following devices can be connected to the I / O interface 705: input devices 706 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, and the like; output devices 707 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, and the like; storage devices 708 including, for example, a magnetic tape, a hard disk, and the like; and communication devices 709. The communication devices 709 can allow the electronic device 700 to communicate wirelessly or wired with other devices to exchange data. Although Figure 7 An electronic device having various devices is shown, but it is understood that all of the shown devices are not required to be implemented or possessed. More or less devices can be alternatively implemented or possessed.

[0165] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to embodiments of the present application. For example, embodiments of the present application include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for executing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network through the communication devices 709, or installed from the storage devices 708, or installed from the ROM 702. When the computer program is executed by the processing devices 701, the above-described functions defined in the methods of embodiments of the present application are performed.

[0166] It should be noted that the computer-readable storage medium in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or component. In the present application, the computer-readable signal medium can include a data signal carried in a baseband or as a part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to an electromagnetic signal, an optical signal or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, device or component. The program code contained in the computer-readable medium can be transmitted by any suitable medium, including but not limited to a wire, a cable, an RF (radio frequency) or the like, or any suitable combination of the above.

[0167] In some embodiments, the client, server can communicate using any currently known or future developed network protocol, such as HTTP (HyperText Transfer Protocol), and can be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include local area networks ("LAN"), wide area networks ("WAN"), the Internet, and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future developed networks.

[0168] The above computer-readable medium can be contained in the above electronic device; or can exist separately without being assembled into the electronic device.

[0169] The above computer-readable medium carries one or more programs, when the above one or more programs are executed by the electronic device, the electronic device:

[0170] The network exposure function module receives a network quality parameter prediction request for at least one first terminal sent by the application server; wherein the network quality parameter prediction request contains the identification information of each first terminal and the model training information of each first terminal; the network quality parameter prediction results of each first terminal are obtained according to the identification information of each first terminal and the model training information of each first terminal; wherein the network quality parameter contains the model transmission delay; the network quality parameter prediction results of each first terminal are sent to the application server through the network exposure function module, so that the application server determines at least one second terminal according to the model transmission delay of each first terminal, and obtains the federated model training effect prediction result according to the network quality parameter prediction results of each second terminal.

[0171] Or,

[0172] The network exposure function module sends a network quality parameter prediction request for at least one first terminal to the network data analysis function module; wherein the network quality parameter prediction request contains the identification information of each first terminal and the model training information of each first terminal; the network data analysis function module sends the network quality parameter prediction results of each first terminal through the network exposure function module; wherein the network data analysis function module obtains the network quality parameter prediction results of each first terminal according to the identification information of each first terminal and the model training information of each first terminal; the network quality parameter contains the model transmission delay; at least one second terminal is determined according to the model transmission delay of each first terminal, and the federated model training effect prediction result is obtained according to the network quality parameter prediction results of each second terminal.

[0173] Computer program code for carrying out operations of the present disclosure can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0174] The computer program product of the first aspect can include one or more non-transitory computer-readable media storing instructions that, when executed, cause one or more processors to perform the operations of the first aspect. The one or more non-transitory computer-readable media can include, for example, magnetic media such as one or more magnetic disks, magnetic tapes or cassettes; optical media such as one or more compact discs, optical discs or Blu-ray discs; magneto-optical media such as one or more floptical discs; solid state media such as one or more solid state drives or other flash memory arrays; or any suitable combination of these. The one or more non-transitory computer-readable media can be encoded with instructions that, when executed, cause one or more processors to perform the operations of the first aspect.

[0175] The modules or units described in the embodiments of the present application can be implemented by software or by hardware. In some cases, the name of the module or unit does not constitute a limitation on the module or unit itself. For example, the first constraint obtaining module can also be described as a "module that obtains a first constraint".

[0176] The functions described above in the specification can be implemented in part or in whole by one or more hardware logic components. For example, and without limitation, illustrative types of hardware logic components that can be used include Field-programmable Gate Arrays (FPGAs), Program-specific Integrated Circuits (ASICs), Program-specific Standard Products (ASSPs), System-on-a-chip systems (SOCs), Complex Programmable Logic Devices (CPLDs), etc.

[0177] In the context of the present application, a machine-readable medium can be a tangible medium that contains or stores a program (for use by or in connection with an instruction execution system, apparatus, or device). The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium will include one or more of: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0178] It should be understood that although the steps in the flowcharts of the drawings are shown in sequence according to the indication of the arrows, the steps are not necessarily executed in sequence according to the indication of the arrows. Unless explicitly stated herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other sequences. Moreover, at least part of the steps in the flowcharts of the drawings can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or sub-steps or stages of other steps.

[0179] The above only describes some embodiments of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should be considered as the protection scope of the present application.

Claims

1. A method for predicting the training effect of a federated model, characterized in that, include: The network open function module receives network quality parameter prediction requests for at least one first terminal sent by the application server; wherein the network quality parameter prediction requests include the identification information of each first terminal and the model training information of each first terminal. The prediction results of each network quality parameter of each first terminal are obtained based on the identification information and model training information of each first terminal; wherein, the network quality parameter includes model transmission delay; The prediction results of each network quality parameter of each first terminal are sent to the application server through the network open function module, so that the application server determines at least one second terminal based on the model transmission latency of each first terminal, and obtains the prediction results of the federated model training effect based on the prediction results of each network quality parameter of each second terminal. The step of obtaining the prediction results of each network quality parameter of each first terminal based on the identification information and model training information of each first terminal specifically includes: For each first terminal, the operation and maintenance management network element obtains the signal receiving power of the first terminal based on the identification information of the first terminal, the policy control network element obtains the network quality of the first terminal based on the identification information of the first terminal, and the access and mobility management function network element obtains the registered slice information of the first terminal based on the identification information of the first terminal. The network quality parameter prediction results of the first terminal are obtained based on the signal receiving power of the first terminal, the network quality of the first terminal, the registered slice information of the first terminal, and the model training information of each first terminal.

2. The method according to claim 1, characterized in that, The network quality parameters also include uplink and downlink bandwidth requirements and traffic data during the federated model training process; the model training information includes the local model size of the first terminal; The step of obtaining the network quality parameter prediction result of the first terminal based on the signal receiving power of the first terminal, the network quality of the first terminal, the registered slice information of the first terminal, and the model training information of each first terminal includes: For each first terminal, the model transmission delay of the first terminal is determined based on the local model size of the first terminal; the uplink and downlink bandwidth requirements of the first terminal are determined based on the signal receiving power of the first terminal; and the traffic data of the first terminal during the federated model training process is determined based on the network quality of the first terminal and the registered slice information of the first terminal.

3. A method for predicting the training effect of a federated model, characterized in that, include: The network open function module sends a network quality parameter prediction request for at least one first terminal to the network data analysis function module; wherein the network quality parameter prediction request includes the identification information of each first terminal and the model training information of each first terminal. The network data analysis function module receives prediction results of network quality parameters for each first terminal sent by the network open function module; wherein the prediction results of each network quality parameter are obtained by the network data analysis function module based on the identification information and model training information of each first terminal; the network quality parameters include model transmission latency. At least one second terminal is determined based on the model transmission latency of each first terminal, and the prediction results of the federated model training effect are obtained based on the prediction results of each network quality parameter of each second terminal. The network quality parameters are obtained in the following ways: For each first terminal, the network data analysis function module obtains the signal receiving power of the first terminal based on the identification information of the first terminal through the operation and maintenance management network element, obtains the network quality of the first terminal based on the identification information of the first terminal through the policy control network element, and obtains the registered slice information of the first terminal based on the identification information of the first terminal through the access and mobility management function network element. The network data analysis module obtains the network quality parameter prediction results of the first terminal based on the signal receiving power of the first terminal, the network quality of the first terminal, the registered slice information of the first terminal, and the model training information of each first terminal.

4. The method according to claim 3, characterized in that, Determining at least one second terminal based on the model transmission delay of each first terminal includes: The standard range of model transmission delay is obtained based on the average prediction results of the model transmission delay of each first terminal. The first terminal whose model transmission delay prediction result is within the model transmission delay standard range is used as the second terminal.

5. The method according to claim 3, characterized in that, The step of obtaining the federated model training effect prediction result based on the prediction results of each network quality parameter of each second terminal includes: For each network quality parameter, the average value of the network quality parameter prediction result is calculated based on the network quality parameter prediction result of each second terminal; A first difference is calculated based on the average value of the predicted network quality parameters and the standard value of the predicted network quality parameters, and a second difference is calculated based on the first difference and the preset weights of the network quality parameters. The training effect prediction result of the federated model is obtained based on the second difference of each network quality parameter.

6. The method according to claim 3, characterized in that, The method further includes: The predicted training effect of the federated model is compared with a preset threshold. If the predicted training effect of the federated model is less than the preset threshold, the training step of the federated model is delayed. The preset threshold represents the acceptable training effect of the federated model. If the predicted training effect of the federated model is greater than the preset threshold, then the training step of the federated model is started.

7. The method according to claim 6, characterized in that, The delay following the training step of the federated model includes: At least one third terminal participating in the training of the federated model is obtained, and each third terminal and each first terminal are simultaneously treated as the first terminal. The network quality parameter prediction request for at least one first terminal is sent again to the network data analysis function module through the network open function module.

8. A device for predicting the training effect of a federated model, characterized in that, include: The prediction request receiving module is used to receive a network quality parameter prediction request sent by the application server for at least one first terminal through the network open function module; wherein the network quality parameter prediction request includes the identification information of each first terminal and the model training information of each first terminal. The network parameter prediction result acquisition module acquires the prediction results of each network quality parameter of each first terminal based on the identification information of each first terminal and the model training information of each first terminal; wherein, the network quality parameter includes the model transmission delay; The network parameter prediction result sending module is used to send the network quality parameter prediction results of each first terminal to the application server through the network open function module, so that the application server can determine at least one second terminal based on the model transmission latency of each first terminal, and obtain the federated model training effect prediction result based on the network quality parameter prediction results of each second terminal. The network parameter prediction result sending module is specifically used for: For each first terminal, the operation and maintenance management network element obtains the signal receiving power of the first terminal based on the identification information of the first terminal, the policy control network element obtains the network quality of the first terminal based on the identification information of the first terminal, and the access and mobility management function network element obtains the registered slice information of the first terminal based on the identification information of the first terminal. The network quality parameter prediction results of the first terminal are obtained based on the signal receiving power of the first terminal, the network quality of the first terminal, the registered slice information of the first terminal, and the model training information of each first terminal.

9. A device for predicting the training effect of a federated model, characterized in that, include: The prediction request sending module is used to send a network quality parameter prediction request for at least one first terminal to the network data analysis function module through the network open function module; wherein, the network quality parameter prediction request includes the identification information of each first terminal and the model training information of each first terminal; The network parameter prediction result receiving module is used to receive the network quality parameter prediction results of each first terminal sent by the network data analysis function module through the network open function module; wherein, the network quality parameter prediction results are obtained by the network data analysis function module based on the identification information of each first terminal and the model training information of each first terminal; the network quality parameters include model transmission latency; The training effect prediction module is used to determine at least one second terminal based on the model transmission latency of each first terminal, and to obtain the training effect prediction result of the federated model based on the prediction results of each network quality parameter of each second terminal. The network quality parameters are obtained in the following ways: For each first terminal, the network data analysis function module obtains the signal receiving power of the first terminal based on the identification information of the first terminal through the operation and maintenance management network element, obtains the network quality of the first terminal based on the identification information of the first terminal through the policy control network element, and obtains the registered slice information of the first terminal based on the identification information of the first terminal through the access and mobility management function network element. The network data analysis module obtains the network quality parameter prediction results of the first terminal based on the signal receiving power of the first terminal, the network quality of the first terminal, the registered slice information of the first terminal, and the model training information of each first terminal.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-7.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-7.

Images