Model training method and system, communication entity and storage medium

Abstract

The present application discloses a model training method and system, a communication entity and a storage medium. The method comprises: sending model training instruction information to a next-level communication entity, wherein the model training instruction information comprises initial model information, and the next-level communication entity comprises an intermediate-level server federated learning communication entity, or comprises an intermediate-level server federated learning communication entity and a client federated learning communication entity; receiving target intermediate model information determined by the next-level communication entity on the basis of the model training instruction information; and determining a target model on the basis of the target intermediate model information.

Description

A model training method, system, communication entity and storage medium

[0001] Cross-references

[0002] This application claims priority to the Chinese patent application filed with the China Patent Office on January 29, 2024, with application number 202410123831.1 and invention name “A model training method, system, communication entity and storage medium”. The entire contents of the application are incorporated by reference into this application. Technical Field

[0003] The present application relates to the field of communication technology, for example, to a model training method, system, communication entity and storage medium. Background Art

[0004] TS23.288 defines the Network Data Analytics Function (NWDAF). NWDAF can perform statistical data and machine learning tasks. NWDAF can interact with different entities for different purposes.

[0005] Although the TS23.288 protocol defines NWDAF for model training based on federated learning, it currently has only a two-layer structure: the first layer is a server, and the second layer is one or more clients.

[0006] However, the above-mentioned NWDAF federated learning process is limited by the two-layer structure of NWDAF, and the trained model is not flexible enough in use.

[0007] Summary of the Invention

[0008] An embodiment of the present application provides a model training method, which is applied to a main server federated learning communication entity. The method includes: sending model training indication information to a next-layer communication entity; wherein the model training indication information includes: initial model information, and the next-layer communication entity includes an intermediate-layer server federated learning communication entity, or includes an intermediate-layer server federated learning communication entity and a client federated learning communication entity; receiving target intermediate model information determined by the next-layer communication entity according to the model training indication information; and determining a target model according to the target intermediate model information.

[0009] An embodiment of the present application provides a model training method, which is applied to an intermediate-layer server federated learning communication entity or a client federated learning communication entity. The method includes: receiving model training indication information sent by an upper-layer communication entity; wherein the model training indication information includes: initial model information, and the upper-layer communication entity includes a main server federated learning communication entity and / or an intermediate-layer server federated learning communication entity; determining target intermediate model information based on the model training indication information; and sending the target intermediate model information to the upper-layer communication entity.

[0010] An embodiment of the present application provides a model training method, which is applied to a network data analysis function. The method includes: receiving a signaling storm prediction request or a model training request from a network function or an operation and maintenance management entity; collecting training data; wherein the training data includes: first training sub-data, second training sub-data, and third training sub-data; the first training sub-data includes at least one of the following: information on UEs in one or more areas, the number and time information of protocol data unit sessions in one or more areas, information on one or more network functions in one or more areas, and the number of user-plane data transmissions in one or more areas; the second training sub-data includes at least one of the following: label information corresponding to the first training sub-data; the third training sub-data includes at least one of the following: external factors affecting the information of UEs in the area; performing model training based on the training data to obtain a target model; wherein the target model is a model for predicting signaling storm related information; using the target model to perform inference and output signaling storm prediction related information, or providing the target model to a service customer.

[0011] An embodiment of the present application provides a model training system, including a main server federated learning communication entity, an intermediate layer server federated learning communication entity, and a client federated learning communication entity; the main server federated learning communication entity is used to execute the model training method executed by the main server federated learning communication entity in the above-mentioned model training method; the intermediate layer server federated learning communication entity is used to execute the model training method executed by the intermediate layer server federated learning communication entity in the above-mentioned model training method; the client federated learning communication entity is used to execute the model training method executed by the client federated learning communication entity in the above-mentioned model training method.

[0012] An embodiment of the present application provides a communication entity, including: a processor; the processor is used to implement the model training method of any of the above embodiments when executing a computer program.

[0013] An embodiment of the present application also provides a computer-readable storage medium storing a computer program, which implements the model training method of any of the above embodiments when the computer program is executed by a processor.

[0014] With respect to the above embodiments and other aspects of the present application and their implementation, further description is provided in the accompanying drawings, detailed description and claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] FIG1 is a schematic diagram of the architecture of a core network provided by an embodiment;

[0016] FIG2 is a schematic diagram of an architecture for collecting data by a federated learning communication entity provided by an embodiment;

[0017] FIG3 is a schematic diagram of another architecture of a federated learning communication entity collecting data provided by an embodiment;

[0018] FIG4 is a schematic diagram of an architecture for providing data by a federated learning communication entity according to an embodiment;

[0019] FIG5 is a schematic diagram of an architecture of another federated learning communication entity providing data according to an embodiment;

[0020] FIG6 is a schematic structural diagram of a model training system provided by an embodiment;

[0021] FIG7 is a schematic diagram of the structure of another model training system provided by an embodiment;

[0022] FIG8 is a flow chart of a model training method provided by an embodiment;

[0023] FIG9 is a flow chart of another model training method provided by an embodiment;

[0024] FIG10 is a schematic diagram of a flow chart of a model training method provided by an embodiment;

[0025] FIG11 is a schematic flow chart of another model training method provided by an embodiment;

[0026] FIG12 is a schematic diagram of signaling interaction of a model training method provided by an embodiment;

[0027] FIG13 is a schematic diagram of signaling interaction of another model training method provided by an embodiment;

[0028] FIG14 is a schematic diagram of a flow chart of a model training method provided by an embodiment;

[0029] FIG15 is a schematic diagram of signaling interaction of a model training method provided by an embodiment;

[0030] FIG16 is a schematic structural diagram of a communication entity provided by an embodiment. DETAILED DESCRIPTION

[0031] It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0032] The model training method provided in this application can be applied to fifth-generation mobile communication technology (5th-generation, 5G) systems, LTE and 5G hybrid architecture systems, 5G New Radio (New Radio, NR) systems, and new communication systems emerging in future communication developments, such as sixth-generation mobile communication technology (6th-generation, 6G) systems. In one embodiment, the model training method can be applied to the core network of a communication system.

[0033] Figure 1 is a schematic diagram of a core network architecture provided by one embodiment. As shown in Figure 1, user equipment (UE) is connected to the core network's network functions via the radio access network (RAN). The RAN manages radio resources, transmits user data received via the N3 interface to the UE, and transmits user data from the UE via the N3 interface. The RAN maps between dedicated radio bearers (DRBs) and Quality of Service (QoS) traffic in protocol data unit (PDU) sessions.

[0034] The Access and Mobility Management Function (AMF) includes the following functions: registration management, connection management, reachability management, and mobility management. It also performs access authentication and access authorization. The AMF is a non-access stratum (NAS) security terminal, forwarding SM NAS messages between the UE and the Session Management Function (SMF).

[0035] The SMF includes the following functions: session establishment, modification, and release; UE Internet Protocol (IP) address allocation and management (including optional authorization functions); selection and control of the User Plane Function (UPF); and downlink data notification. The SMF associates and controls the UPF via the N4 interface. The SMF provides the UPF with packet detection rules (PDRs), instructing it on how to detect user data traffic. It also provides forwarding action rules (FARs), QoS enforcement rules (QERs), and usage reporting rules (URRs), instructing the UPF on how to forward user data traffic, perform QoS processing, and perform usage reporting on user data traffic detected using the PDRs.

[0036] The UPF includes the following functions: serving as an anchor point for intra- and inter-Radio Access Technology (RAT) mobility, packet routing and forwarding, traffic usage reporting, user plane QoS processing, downlink packet buffering, and downlink data notification triggering. The GTP-U tunnel is used for the N3 interface between the RAN and the UPF. The GTP-U tunnel is per PDU session. For downlink traffic, the UPF binds the downlink traffic to the QoS traffic within the PDU session GTP-U tunnel by using the FAR received from the SMF. For uplink traffic, the RAN transfers the user plane traffic to the QoS flow identified by the UE.

[0037] The Policy Control Function (PCF) provides QoS policy rules to the control plane functions for enforcement. The PCF converts application function (AF) requests into policy control and charging control (PCC) rules applicable to the PDU session.

[0038] Unified Data Management (UDM) performs the generation of 3rd Generation Partnership Project (3GPP) authentication and key agreement (AKA) credentials, authorization of access based on subscription data, UE registration with the serving network function (NF) (e.g., provision of the UE storage service AMF and the UE PDU session storage service SMF), and subscription management. UDM accesses the Unified Data Repository (UDR) to retrieve UE subscription data and stores UE context in the UDR. UDM and UDR can be deployed together.

[0039] The model training method provided in the embodiments of the present application can be applied to a federated learning communication entity. The federated learning communication entity can be a NWDAF. The NWDAF is a 5G core network (5G Core, abbreviated as: 5GC) NF located in the control plane, which performs statistical data and machine learning related tasks in the 5GC.

[0040] NWDAF can interact with different communication entities for different purposes: 1. Collect data based on event subscriptions provided by AMF, SMF, UPF, PCF, UDM, Network Slice Admission Control Function (NSACF), AF (directly or through Network Exposure Function (NEF)) and Operation Administration and Maintenance (OAM); 2. (Optional) Perform analysis and data collection using the Data Collection Coordination Function (DCCF); 3. Retrieve information from the data repository, for example, retrieve user-related information UDR through UDM or retrieve PFD information through NEF (Packet Flow Descriptions Function (PFDF)); 4. Collect location information data from the Location Services (LCS) system; 5. (Optional) Store and retrieve information from the Analytic Data Repository Function (ADRF); 6. (Optional) Store and retrieve information from the Messaging Framework Adaptor Function (MFAF) 7. Retrieve information about NF, for example, retrieve NF-related information from Network Repository Function (NRF); 8. Provide analysis to consumers on demand; 9. Provide batch data related to analysis identity (ID); 10. Provide accuracy information of analysis ID; 11. Provide ML model accuracy information or ML model accuracy degradation indication for machine learning (ML) models.

[0041] A single instance or multiple instances of NWDAF can be deployed in a Public Land Mobile Network (PLMN). NWDAF can include the following logical functions: Analytics Logical Function (AnLF), which is a logical function in NWDAF used to perform reasoning, derive analytical information (i.e., derive statistics and / or predictions based on analytical consumer requests) and expose analytical services (i.e., Nnwdaf_AnalyticsSubscription or Nnwdaf_AnalyticsInfo); Model Training Logical Function (MTLF), which is another logical function in NWDAF used to train ML models and expose new training services (e.g., providing trained ML models). An NWDAF can include one MTLF, one AnLF, or both logical functions.

[0042] The DCCF is also a NF on the 5GC control plane. It coordinates the collection and distribution of data requested by NF consumers. It prevents data sources from processing multiple subscriptions to the same data and prevents multiple notifications containing the same information from being sent due to uncoordinated requests from data consumers. The DCCF applies to: NWDAFs requesting data from data sources (e.g., for computational analysis); NF consumers requesting analysis from NWDAF data sources; NF consumers requesting data from ADRF data sources; and ADRFs receiving data from NF data sources.

[0043] Figure 2 is a schematic diagram of the architecture for data collection by a federated learning communication entity, provided by one embodiment. Figure 3 is a schematic diagram of the architecture for data collection by another federated learning communication entity, provided by one embodiment. Figure 2 shows that NWDAF can collect data from any NF (also called a data source). Figure 3 shows that NWDAF can collect data from any NF through DCCF. In other words, Figures 2 and 3 present two data collection structures to facilitate data analysis and model training by NWDAF.

[0044] Figure 4 is a schematic diagram of an architecture for providing data by a federated learning communication entity, provided in one embodiment. Figure 5 is a schematic diagram of an architecture for providing data by another federated learning communication entity, provided in one embodiment. Figure 4 shows that NWDAF can provide data to any NF. Figure 5 shows that NWDAF can provide data to any NF through DCCF. The data here includes at least one of the following: analysis results, statistical results, and prediction information. In other words, Figures 4 and 5 show two possible network data analysis exposure architectures for use by any NF consumer that subscribes to or requests analysis.

[0045] In an embodiment of the present application, a model training method, system, communication entity and storage medium are provided, and a model training method based on layered federated learning is proposed, so that the target model and target intermediate model information trained based on the model training method can be flexibly used, and it is possible to flexibly establish a server (server) and client (client) relationship between federated learning communication entities.

[0046] When the target model is used to predict network-related states such as signaling storms, the target model trained based on the model training method can flexibly predict network-related states such as signaling storms in multiple tracking areas (TAs) or larger TAs.

[0047] The following describes the model training method, system, communication entity and its technical effects.

[0048] Figure 6 is a structural diagram of a model training system provided by an embodiment. As shown in Figure 6, a model training system based on layered federated learning is provided. The system includes multiple layers of NWDAF. Figure 6 takes a 4-layer NWDAF as an example for illustration. NWDAF 1 at level 0 is the main server of the entire architecture (can be called main server). NWDAF 2 and NWDAF 4 at level 1 can be regarded as clients compared to NWDAF 1, but can be regarded as servers (can be called server and client) compared to NWDAF at level 2. NWDAF 8 at level 2 can be regarded as the client of NWDAF 2 (can be called client), and so on. The layer-related parameters involved in this architecture include: the layer in which the NWDAF is located. Model transmission can be performed between NWDAFs connected at adjacent layers.

[0049] FIG7 is a schematic structural diagram of another model training system provided by an embodiment. As shown in FIG7 , another model training system based on layered federated learning is provided. The system includes multiple layers of NWDAF. FIG7 is taken as an example of a system including three layers of NWDAF. In TA1, NWDAF 3 and NWDAF 4 serve different NFs (FIG7 is taken as an example of NF including SMF, AMF and PCF): NWDAF 3 serves SMF 1, AMF 2 and PCF 1, and NWDAF 4 serves SMF 2, AMF 3 and PCF 2, so NWDAF 3 cannot collect data from PCF 2, AMF 3 or SMF 2. Furthermore, NWDAF 1 cannot establish a direct model transmission connection with NWDAF 4, that is, it is impossible to directly establish a server and client connection, probably because the two are far away and are not in each other's service area. At the beginning of hierarchical federated learning, NWDAF 1 can send the initial model to NWDAF 2 and NWDAF 5, which then send it to NWDAF 3 and NWDAF 4. NWDAF 3 and NWDAF 4 perform training based on the initial model. After training, NWDAF 4 and NWDAF 3 each send the trained target sub-model information to NWDAF 2, which aggregates the information and sends the aggregated model to NWDAF 1. NWDAF 5 performs training based on the initial model. After training, NWDAF 5 sends the trained target sub-model to NWDAF 1. NWDAF 1 aggregates the models of NWDAF 2 and NWDAF 5 to form the final, complete target model.

[0050] Therefore, based on the system shown in FIG7 , assuming that the target model is used to predict signaling storm-related information, if you want to obtain a model for predicting whether a signaling storm will occur in TA1 in the future, you can contact NWDAF 2 to obtain it. If you want to obtain a model for predicting whether a signaling storm will occur in TA2, you can obtain it through NWDAF 5. If you want to obtain signaling storm information for a larger area, you can contact NWDAF 1 to obtain it. That is, signaling storms in small areas can be flexibly predicted through the models of mid-layer communication entities or bottom-layer communication entities (referred to as target intermediate models in this embodiment), and signaling storm predictions in large areas can be achieved through the global model (referred to as the target model in this embodiment).

[0051] It should be understood that Figure 7 is merely an example. If signaling storm prediction is required for more or larger areas, NWDAF 1 (the master server) can use the NRF to find more NWDAFs as clients. At the same time, the master server will also need to aggregate more models. As can be seen, the models trained by this model training system can be flexibly used, enabling flexible prediction of network-related conditions such as signaling storms in multiple or larger TAs. Furthermore, flexible server-client relationships can be established between NWDAFs.

[0052] Figure 8 is a schematic flow chart of a model training method provided by one embodiment. The model training method provided in this embodiment is applicable to a master server federated learning communication entity. The master server federated learning communication entity in this embodiment refers to the communication entity at the highest level of layered federated learning, such as NWDAF 1 in Figures 6 and 7. As shown in Figure 8, the method includes the following steps.

[0053] Step 801: Send model training instruction information to the next layer communication entity.

[0054] The model training instruction information includes: initial model information. The next layer communication entity includes an intermediate layer server federated learning communication entity, or includes an intermediate layer server federated learning communication entity and a client federated learning communication entity.

[0055] The middle-layer server federated learning communication entity in this embodiment refers to a communication entity located in the middle layer in the layered federated learning. The middle-layer server federated learning communication entity is connected to both the main server federated learning communication entity or the middle-layer server federated learning communication entity, and is also connected to the middle-layer server federated learning communication entity or the client federated learning communication entity. For example, NWDAF 2, NWDAF 4, and NWDAF 5 in Figure 6, and NWDAF 2 in Figure 7 are all middle-layer server federated learning communication entities. The client federated learning communication entity is connected to the main server federated learning communication entity or the middle-layer server federated learning communication entity, and is no longer connected to the communication entity downward. For example, NWDAF 3, NWDAF 8, NWDAF 6, and NWDAF 7 in Figure 6, and NWDAF 3, NWDAF 4, and NWDAF 5 in Figure 7 are all client federated learning communication entities.

[0056] In one implementation, the next layer of communication entities in this embodiment includes an intermediate layer server federated learning communication entity. In this scenario, the communication entities connected to the main server federated learning communication entity will also connect to the intermediate layer server federated learning communication entity or the client federated learning communication entity.

[0057] In another implementation, the next layer of communication entities in this embodiment includes: an intermediate layer server federated learning communication entity and a client federated learning communication entity. In this scenario, the communication entity connected to the main server federated learning communication entity can be connected to the intermediate layer server federated learning communication entity or the client federated learning communication entity, or it can no longer be connected to the communication entity downward. Figures 6 and 7 both belong to this scenario. For example, the next layer of communication entities of NWDAF 1 in Figure 6 includes: intermediate layer server federated learning communication entities NWDAF 2 and NWDAF 4, and client federated learning communication entity NWDAF 3.

[0058] In this embodiment, the master server federated learning communication entity and the next-level communication entity are used to implement layered federated learning. In both implementations, the number of layers of layered federated learning is greater than 2. The master server federated learning communication entity is located at the highest layer of the layered federated learning. In this embodiment, layer 0 can represent the highest layer.

[0059] In one embodiment, the main server federated learning communication entity sends Nnwdaf_MLModelTraining_Subscribe or Nnwdaf_MLModelTrainingInfo_Request to send model training instruction information to the next layer communication entity.

[0060] The model training instruction information in this embodiment includes initial model information. The initial model information in this embodiment refers to information related to the initial model, such as model parameters of the initial model. The initial model in this embodiment includes at least one of the following: an ML model pre-installed in the master server federated learning communication entity, a previously trained model obtained through layered federated learning, etc.

[0061] In one embodiment, before step 801, the model training method provided in this embodiment further includes the following steps: receiving a model subscription request message sent by a consumer-end communication entity, wherein the model subscription request message includes at least one of the following: an analysis identifier, a model metric, an accuracy reporting interval, and a training termination condition; and determining model training indication information based on the model subscription request message.

[0062] In this embodiment, the consumer-side communication entity includes at least one of the following: a NF and an OAM in a 5GC. The NF in this context can include an NWDAF with AnLF or an NWDAF with MTLF. The consumer-side communication entity can send a model subscription request message to the master server federated learning communication entity via the Nnwdaf_MLModelProvision service to retrieve the ML model.

[0063] The model metric in this embodiment refers to the standard used to measure the model, such as accuracy, precision, and recall. The accuracy reporting interval in this embodiment refers to the time interval for reporting accuracy to the consumer-side communication entity. The training termination conditions in this embodiment may include: an ML model accuracy threshold or the time when the ML model is required. The ML model accuracy threshold can be used to indicate the accuracy of the trained target model. When the ML model accuracy threshold is reached during the training process, the client federated learning communication entity can stop the training process. If the consumer-side communication entity provides the time when the ML model is required, the main server federated learning communication entity can consider this information to determine the maximum response time of its next-level communication entity.

[0064] In one embodiment, the model training instruction information further includes at least one of the following: a model metric standard and a maximum response time for reporting model information. The maximum response time for reporting model information in this embodiment is used to indicate the maximum time for the next layer communication entity to feedback model information.

[0065] When determining model training indication information based on a model subscription request message, the following steps may be included: determining initial model information based on an analysis identifier; determining a maximum response time for reporting model information based on training termination conditions and / or an accuracy reporting interval; and generating model training indication information based on the maximum response time for reporting model information and at least one of the model metrics, as well as the initial model information.

[0066] After receiving the model training instruction information sent by the master server federated learning communication entity, the next layer communication entity determines the target intermediate model information based on the model training instruction information and sends the target intermediate model information to the master server federated learning communication entity. The specific process will be described in detail in subsequent embodiments.

[0067] Step 802: Receive target intermediate model information determined by the next layer communication entity according to the model training instruction information.

[0068] The main server federated learning communication entity receives the target intermediate model information determined by the next layer communication entity according to the model training instruction information.

[0069] It should be noted that when the next-layer communication entity is an intermediate-layer server federated learning communication entity, the target intermediate model information is the model information obtained by the intermediate-layer server federated learning communication entity after aggregating the model information reported by its next-layer communication entity. When the next-layer communication entity is a client federated learning communication entity, the target intermediate model information is the model information obtained after the client federated learning communication entity trains the initial model based on the collected training data. Taking Figure 6 as an example, the target intermediate model information reported by NWDAF 4 is the model information obtained after NWDAF 4 aggregates the model information reported by NWDAF 6 and the model information reported by NWDAF 5. The target intermediate model information reported by NWDAF 3 is the model information obtained after NWDAF 3 trains the initial model based on the collected training data.

[0070] The target intermediate model information in this embodiment includes information required to establish an aggregation model.

[0071] Step 803: Determine the target model according to the target intermediate model information.

[0072] In step 803, the target intermediate model information is aggregated to obtain the target model.

[0073] The target model in this embodiment may be a model for predicting network status.

[0074] In one embodiment, in step 803, the target model information may also be determined based on the target intermediate model information.

[0075] The target model in this embodiment is a model for predicting information related to signaling storms. The output information of the target model includes at least one of the following: information about the area where the signaling storm occurred, the time when the signaling storm occurred, the network functions affected by the signaling storm, and the user equipment (UE) affected by the signaling storm. The output information of the target model may also include the volume of the signaling storm, i.e., the size of the signaling storm.

[0076] In one embodiment, after determining the target model, the main server federated learning communication entity may further perform the following steps: determining the model measurement information of the target model according to the model measurement standard; and sending the model measurement information of the target model to the consumer-end communication entity.

[0077] In one embodiment, the main server federated learning communication entity may send a Nnwdaf_MLModelProvision_Notify message to dynamically update the model metric information of the model in the training process to the consumer-end communication entity periodically (such as a certain number of training rounds or every 10 minutes) or under certain preset circumstances, including the following two implementation methods: periodic update (for example, a certain number of training rounds or update every 10 minutes), or dynamic update when certain predetermined states are reached (for example, dynamic update when certain predetermined states are reached or the ML model accuracy threshold is reached or the training time expires).

[0078] The consumer communication entity determines whether the target model meets the requirements and decides whether to stop or continue the training process. The consumer communication entity can call the Nnwdaf_MLModelProvision_Subscribe service operation to indicate whether to stop or continue the training process.

[0079] When the main server federated learning communication entity receives the continuing training instruction information sent by the consumer communication entity based on the model metric information of the target model, it determines the target model as the new initial model and returns to the step of executing step 801. Training stops until the consumer communication entity determines that the target model meets the requirements.

[0080] The model trained by the model training method provided in this embodiment can flexibly predict signaling storms in small areas through the target intermediate model of the middle-layer communication entity, and realize signaling storm prediction in large areas through the target model.

[0081] The model training method provided in this embodiment is applied to a master server federated learning communication entity and includes: sending model training instruction information to a next-layer communication entity, wherein the model training instruction information includes: initial model information, and the next-layer communication entity includes an intermediate-layer server federated learning communication entity, or includes an intermediate-layer server federated learning communication entity and a client federated learning communication entity; receiving target intermediate model information determined by the next-layer communication entity based on the model training instruction information; and determining a target model based on the target intermediate model information. This model training method implements hierarchical federated learning, allowing the target model and target intermediate model information trained based on the model training method to be flexibly used, and can flexibly establish server-client relationships between federated learning communication entities.

[0082] Figure 9 is a flowchart of another model training method provided by one embodiment. Based on the embodiment shown in Figure 8 and various optional implementations, this embodiment provides a detailed description of how to determine the next-layer communication entity of the master server federated learning communication entity in this model training method. As shown in Figure 9, the model training method provided by this embodiment includes the following steps.

[0083] Step 901: Send the federated learning capability type to the network storage communication entity.

[0084] Among them, the federated learning capability type includes at least one of the following: whether it supports serving as a main server federated learning communication entity for layered federated learning, whether it supports serving as an intermediate-layer server federated learning communication entity for layered federated learning, whether it supports serving as a client federated learning communication entity for layered federated learning, and whether it supports serving as both an intermediate-layer server federated learning communication entity and a client federated learning communication entity for layered federated learning.

[0085] The network storage communication entity in this implementation may be an NRF.

[0086] The NRF can determine the master server federated learning communication entity based on the federated learning capability type reported by each federated learning communication entity. In this embodiment, the master server federated learning communication entity is a communication entity determined by the network storage communication entity based on a preset selection rule. In one embodiment, after receiving a request to discover the master server from the consumer communication entity, the NRF determines the master server federated learning communication entity based on the federated learning capability type reported by each federated learning communication entity.

[0087] Step 902: Send a communication entity discovery request to the network storage communication entity.

[0088] In one embodiment, the master server federated learning communication entity sends a communication entity discovery request by calling the Nnrf_NFDiscovery_Request service operation to determine the next layer of communication entity from the NRF.

[0089] Step 903: Receive communication entity discovery request response information fed back by the network storage communication entity.

[0090] The network storage communication entity can determine the initial candidate communication entities according to the preset selection rules, include the initial candidate communication entity list in the communication entity discovery request response information, and send the communication entity discovery request response information to the main server federated learning communication entity.

[0091] Step 904: Determine candidate communication entities according to the communication entity discovery request response information.

[0092] In one embodiment, step 904 can be implemented as follows: determining a candidate communication entity based on the communication entity discovery request response information and a preset selection rule. The communication entity included in the communication entity discovery request response information is the communication entity determined by the network storage communication entity based on the preset selection rule. That is, the candidate communication entity is determined based on the initial candidate communication entity and the preset selection rule.

[0093] Step 905: Send a federated learning preparation request to the candidate communication entity.

[0094] The federated learning preparation request includes at least one of the following: the level of the master server federated learning communication entity and the identifier of the master server federated learning communication entity.

[0095] In one embodiment, the main server federated learning communication entity may send a federated learning preparation request to the candidate communication entity through the Nnwdaf_MLModelTraining_Subscribe or Nnwdaf_MLModelTrainingInfo_Request service to check whether the candidate communication entity meets the model training requirements.

[0096] In one embodiment, the identifier of the master server federated learning communication entity may be an NF ID or a vendor ID.

[0097] In this embodiment, the federated learning preparation request includes the level of the master server federated learning communication entity and the identifier of the master server federated learning communication entity, which can prevent the master server federated learning communication entity from being selected as a server or client for the second time in the lower layer.

[0098] Step 906: Receive response information sent by the candidate communication entity in response to the federated learning preparation request.

[0099] Step 907: According to a preset selection rule, determine the next layer of communication entities from the candidate communication entities whose corresponding response information is the target response information.

[0100] The target response information is used to instruct the candidate communication entity to join the layered federated learning where the main server federated learning communication entity is located.

[0101] In one embodiment, the preset selection rules include at least one of the following: computing power information of the communication entity, whether the model corresponding to the analysis identifier needs to be trained through a layered federated learning process, model filtering information, the federated learning capability type of the communication entity, whether the communication entity is currently performing other layered federated learning, the time period of interest, the service area, the type of network function of the data source required for model training, the interoperability index of the model corresponding to the analysis identifier, the available data requirements, and the available time requirements.

[0102] Available data requirements include at least one of the following: a list of event IDs of local data used for training, dataset statistical properties, the time window of data samples, and the minimum number of data samples. Available time requirements include the time span required for the model training process.

[0103] It should be noted that the preset selection rules used by NRF in determining the main server federated learning communication entity, the preset selection rules used by NRF in determining the initial candidate communication entity, the preset selection rules used by the main server federated learning communication entity in determining the candidate communication entity, and the preset selection rules used by the main server federated learning communication entity in determining the next layer of communication entity may be the same or different.

[0104] After step 907, it is equivalent to establishing a layered federated learning architecture, based on which steps 801 to 803 in the above embodiment can be executed, which will not be described in detail in this embodiment.

[0105] The model training method provided in this embodiment provides a method for determining the next-level communication entity. The next-level communication entity determined based on this method can meet the model training requirements, thereby improving the efficiency and success rate of model training. In addition, the signaling interaction method for determining the next-level communication entity in this embodiment is simple and efficient, further improving the efficiency of model training. Moreover, based on this embodiment, it is possible to flexibly establish server-client relationships between federated learning communication entities.

[0106] Figure 10 is a flow chart of a model training method provided by an embodiment. The model training method provided in this embodiment is applicable to an intermediate-layer server federated learning communication entity or a client federated learning communication entity. The intermediate-layer server federated learning communication entity in this embodiment is connected to both the main server federated learning communication entity or the intermediate-layer server federated learning communication entity and the intermediate-layer server federated learning communication entity or the client federated learning communication entity. For example, NWDAF 2, NWDAF 4, and NWDAF 5 in Figure 6. For example, NWDAF 2 in Figure 7. The client federated learning communication entity in this embodiment is connected to the main server federated learning communication entity or the intermediate-layer server federated learning communication entity, and is no longer connected to the communication entity downward. For example, NWDAF 3, NWDAF 8, NWDAF 6, and NWDAF 7 in Figure 6. For example, NWDAF 3, NWDAF 4, and NWDAF 5 in Figure 7. As shown in Figure 10, the method includes the following steps.

[0107] Step 1001: Receive model training instruction information sent by the upper-layer communication entity.

[0108] The model training instruction information includes: initial model information. The upper layer communication entity includes the main server federated learning communication entity and / or the intermediate layer server federated learning communication entity.

[0109] In one embodiment, the model training instruction information further includes at least one of the following: a model metric standard and a maximum response time for reporting model information.

[0110] When the upper-layer communication entity is an intermediate-layer server federated learning communication entity, the model training instruction information may be the model training instruction information received by the intermediate-layer server federated learning communication entity of the upper layer, or may be information generated by the intermediate-layer server federated learning communication entity of the upper layer based on the model training instruction information received. For example, taking Figure 6 as an example, when the intermediate-layer server federated learning communication entity is NWDAF 5, NWDAF 5 receives the model training instruction information sent by the upper-layer communication entity NWDAF 4. The model training instruction information may be the model training instruction information received by NWDAF 4 from NWDAF 1, or may be information generated by NWDAF 4 based on the model training instruction information received from NWDAF 1.

[0111] Step 1002: Determine target intermediate model information according to model training instruction information.

[0112] In one scenario, when the method is applied to a client federated learning communication entity, the target intermediate model information includes target sub-model information. The implementation process of step 1002 includes: collecting training data; and training the initial model information based on the training data to obtain the target sub-model information.

[0113] In one embodiment, the training data includes at least one of the following: first training sub-data obtained from the first network function, second training sub-data obtained from the second network function, and third training sub-data obtained from the third network function.

[0114] In one embodiment, the first training sub-data includes at least one of the following: a historical number of UEs in one or more areas and time information related to the historical number of UEs, a current number of UEs in one or more areas and time information related to the current number of UEs, a change in the number of UEs in the one or more areas at different times, a movement trajectory of the UE, areas passed by the UE, time information of the areas passed by the UE, a historical number of UE registrations in the one or more areas and time information related to the historical number of registrations, a current number of UE registrations in the one or more areas and time information related to the current number of registrations, a historical registration type of the UE in the one or more areas and time information related to the historical registration type, a current registration type of the UE in the one or more areas and time information related to the current registration type, a historical number of protocol data unit sessions in the one or more areas and time information related to the historical number of protocol data unit sessions, a current number of protocol data unit sessions in the one or more areas and time information related to the current number of protocol data unit sessions, current bearer information of one or more network functions in the one or more areas and time information related to the current bearer information, historical bearer information of one or more network functions in the one or more areas and time information related to the historical bearer information, current bearer capacity of the one or more network functions in the one or more areas and time information related to the current bearer capacity information, historical bearer capacity of one or more network functions in one or more areas and time information related to the historical bearer capacity, current service area information, the current number of UEs served, the current number of network functions served, and time information related to the current service area information of one or more network functions in one or more areas, historical service area information, the historical number of UEs served, the historical number of network functions served, and time information related to the historical service area information of one or more network functions in one or more areas, change information of service area information of one or more network functions at different times, change in the number of UEs served, and change in the number of network functions served in one or more areas, current user plane data transmission volume and time information related to the current user plane data transmission volume, historical user plane data transmission volume and time information related to the historical user plane data transmission volume, historical user equipment categories and time information corresponding to the historical user equipment categories in one or more areas, current user equipment categories and time information corresponding to the current user equipment categories in one or more areas, historical user equipment group information in one or more areas and time information corresponding to the historical user equipment group information, time information and area information of current user equipment inventory in one or more areas, time information and area information of historical user equipment inventory in one or more areas.

[0115] In one embodiment, the second training sub-data includes: information related to whether a signaling storm occurs, corresponding time information, and location information.

[0116] In one embodiment, the third training sub-data includes at least one of the following: a historical number of users of the third network function and time information associated with the historical number of users; a current number of users of the third network function and time information associated with the current number of users; historical data traffic information of the third network function and time information associated with the historical data traffic information; current data traffic information of the third network function and time information associated with the current data traffic information; historical external factors affecting at least one of the number of UEs in the area, the number of registered UEs, and the amount of user-plane data transmission; and current external factors affecting at least one of the number of UEs in the area, the number of registered UEs, and the amount of user-plane data transmission. For example, the historical external factors and the current external factors may include concert information, flight information, etc.

[0117] In one embodiment, the first network function is a network function of a core network. The first network function is a network function other than an NWDAF in the core network. The second network function is an operation and maintenance management entity. The third network function is an application function.

[0118] In one embodiment, the training data further includes: fourth training sub-data obtained from other network elements. The fourth training sub-data includes at least one of the following: network function load information (NF load information) and related information, abnormal behavior of UE or 5GC NF (Abnormal behavior of UE or 5GC NF) and related information, suspected distributed denial of service (DDoS) attack (Suspicions of DDoS attack) and related information, UE session management congestion control experience (Session Management Congestion Control Experience) and related information, user data congestion analysis (User data congestion analytic) and related information, session unit congestion analysis (PDU session traffic analytics) and related information, end-to-end data volume transfer time analysis (E2E data volume transfer time analytics) and related information, packet filtering description decision analysis (PFD determination analytics) and related information.

[0119] In one embodiment, the fourth training sub-data includes at least one of the following: current load information of the network function, time information corresponding to the current load information, and relevant input and output information of network load analysis; historical load information of the network function, time information corresponding to the historical load information, and relevant input and output information of network load analysis; current abnormal behavior of the network function or UE, time information corresponding to the current abnormal behavior, and relevant input and output information of abnormal behavior analysis; historical abnormal behavior of the network function or UE, time information corresponding to the historical abnormal behavior, and relevant input and output information of historical abnormal behavior analysis; current suspected DDOS attack, time information corresponding to the current suspected DDOS attack; historical suspected DDOS attack, time information corresponding to the historical suspected DDOS attack; current session management congestion control experience of the UE, time information corresponding to the current session management congestion control experience, and relevant input and output information of UE session management congestion control experience analysis; historical session management congestion control experience of the UE, time information corresponding to the historical session management congestion control experience, and relevant input and output information of historical UE session management congestion control experience analysis; current user data congestion analysis, and time information corresponding to the current user data The present invention also includes the time information corresponding to the congestion analysis and the related input and output information of the user data congestion analysis, the historical user data congestion analysis and the time information corresponding to the historical user data congestion analysis and the input and output information of the related historical user data congestion analysis, the current session unit congestion analysis and the time information corresponding to the current session unit congestion analysis and the related input and output information of the session unit congestion analysis, the historical session unit congestion analysis and the time information corresponding to the historical session unit congestion analysis and the related input and output information of the historical session unit congestion analysis, the current end-to-end data volume transmission time analysis and the time information corresponding to the current end-to-end data volume transmission time analysis and the related input and output information of the end-to-end data volume transmission analysis, the historical end-to-end data volume transmission time analysis and the time information corresponding to the historical end-to-end data volume transmission time analysis and the input and output information of the related historical current end-to-end data volume transmission analysis, the current data packet filtering description decision analysis and the time information corresponding to the current data packet filtering description decision analysis and the related input and output information of the data packet filtering description decision analysis, the historical data packet filtering description decision analysis and the time information corresponding to the historical data packet filtering description decision analysis and the related input and output information of the historical data packet filtering description decision analysis.

[0120] In this embodiment, other network elements refer to other NWDAFs except the client federated learning communication entity.

[0121] In one embodiment, in this scenario, the following steps are further included: determining the model measurement information of the target sub-model information according to the model measurement standard; and sending the model measurement information of the target sub-model information to the upper-layer communication entity.

[0122] In this scenario, the following steps are also included: sending information about the training data to the upper-layer communication entity, such as the coverage area of ​​the training data, the sampling rate, the maximum / minimum value of each dimension of the data, etc.

[0123] In another scenario, when the method is applied to an intermediate-layer server federated learning communication entity, the target intermediate model information includes: current aggregated model information of the layer at which the intermediate-layer server federated learning communication entity resides. The implementation process of step 1002 includes: sending model training instruction information to a next-layer communication entity, where the next-layer communication entity includes the intermediate-layer server federated learning communication entity and / or the client federated learning communication entity; receiving target sub-model information and / or next-level aggregated model information determined by the next-layer communication entity based on the model training instruction information; and aggregating the target sub-model information and / or next-level aggregated model information to obtain current aggregated model information.

[0124] For example, when the intermediate-layer server federated learning communication entity is NWDAF 2 in Figure 7, NWDAF 2 sends model training instruction information to the next-layer communication entities, namely, NWDAF 3 and NWDAF 4. NWDAF 3 trains target sub-model information based on the initial model information, and NWDAF 4 trains target sub-model information based on the initial model information. NWDAF 2 aggregates the target sub-model information trained by NWDAF 3 and NWDAF 4 to obtain the current aggregated model information.

[0125] For example, when the intermediate-layer server federated learning communication entity is NWDAF 4 in Figure 6 , NWDAF 4 sends model training instruction information to the next-layer communication entities, namely, NWDAF 6 and NWDAF 5. NWDAF 6 trains target sub-model information based on the initial model information. NWDAF 5 summarizes the target sub-model information reported by NWDAF 7 to obtain the next-level summarized model information. NWDAF 4 summarizes the target sub-model information trained by NWDAF 6 and the next-level summarized model information summarized by NWDAF 5 to obtain the current summarized model information.

[0126] In this scenario, the following steps are also included: determining the model measurement information of the current aggregated model information according to the model measurement standard; and sending the model measurement information of the current aggregated model information to the upper-layer communication entity.

[0127] In this scenario, the following steps are also included: sending information about the training data to the upper-layer communication entity, such as the coverage area of ​​the training data, the sampling rate, the maximum / minimum value of each dimension of the data, etc.

[0128] In this scenario, in one implementation, receiving target sub-model information and / or next-level summary model information determined by a next-level communication entity based on model training indication information includes: receiving target sub-model information and / or next-level summary model information determined by all next-level communication entities connected to the intermediate-level server federated learning communication entity based on the model training indication information. That is, the intermediate-level server federated learning communication entity receives the target sub-model information and / or next-level summary model information determined by all next-level communication entities connected to it, and then summarizes the information to obtain the current summary model information.

[0129] In another implementation, receiving target sub-model information and / or next-level summary model information determined by a next-layer communication entity based on model training indication information includes: receiving target sub-model information and / or next-level summary model information determined by some next-layer communication entities connected to the intermediate-layer server federated learning communication entity based on the model training indication information. That is, after receiving the target sub-model information and / or next-level summary model information determined by the next-layer communication entity based on the model training indication information, the intermediate-layer server federated learning communication entity performs aggregation to obtain the current aggregated model information, without having to wait for the target sub-model information and / or next-level summary model information determined by all next-layer communication entities to be received before aggregation is performed.

[0130] Step 1003: Send target intermediate model information to the upper-layer communication entity.

[0131] After determining the target intermediate model information, the target intermediate model information is sent to the upper layer communication entity.

[0132] For example, in Figure 6, NWDAF 7 sends its determined target intermediate model information, i.e., target submodel information, to NWDAF 5. NWDAF 5 and NWDAF 6 then send their respective determined target intermediate model information to NWDAF 4. This process continues in this way, with target intermediate model information reported layer by layer.

[0133] When the upper-layer communication entity is the middle-layer server federated learning communication entity, it aggregates the target intermediate model information after receiving it and then sends it to the upper-layer communication entity.

[0134] When the upper-layer communication entity is the master server federated learning communication entity, it receives the target intermediate model information and aggregates it to obtain the target model.

[0135] It should be noted that in the scenario where the model training indication information includes the maximum response time for reporting model information, when sending the target intermediate model information to the upper-level communication entity, the model information needs to be reported to the upper-level communication entity within the maximum response time.

[0136] The target model in this embodiment is used to predict network status. The target model in this embodiment is a model for predicting information related to signaling storms. The output information of the target model includes at least one of the following: information about the area where the signaling storm occurred, the time when the signaling storm occurred, the network functions affected by the signaling storm, and the user equipment (UE) affected by the signaling storm. The output information of the target model may also include the volume of the signaling storm, i.e., the size of the signaling storm.

[0137] The model training method provided in this embodiment is applied to an intermediate-layer server federated learning communication entity or a client federated learning communication entity. The method includes: receiving model training instruction information sent by a communication entity in an upper layer, wherein the model training instruction information includes: initial model information, and the upper layer communication entity includes a main server federated learning communication entity and / or an intermediate-layer server federated learning communication entity; determining target intermediate model information based on the model training instruction information; and sending the target intermediate model information to the upper layer communication entity. This model training method implements hierarchical federated learning, allowing the target model and target intermediate model information trained based on the model training method to be flexibly used, and can flexibly establish server-client relationships between federated learning communication entities.

[0138] Figure 11 is a flowchart of another model training method provided by one embodiment. Based on the embodiment shown in Figure 10 and various optional implementations, this embodiment provides a detailed description of how to determine the next-layer communication entity of the intermediate-layer server federated learning communication entity in this model training method. As shown in Figure 11, the model training method provided by this embodiment includes the following steps.

[0139] Step 1101: Send the federated learning capability type to the network storage communication entity.

[0140] Among them, the federated learning capability type includes at least one of the following: whether it supports serving as a main server federated learning communication entity for layered federated learning, whether it supports serving as an intermediate-layer server federated learning communication entity for layered federated learning, whether it supports serving as a client federated learning communication entity for layered federated learning, and whether it supports serving as both an intermediate-layer server federated learning communication entity and a client federated learning communication entity for layered federated learning.

[0141] Step 1102: Receive a federated learning preparation request sent by an upper-layer communication entity.

[0142] Step 1103: Send response information to the upper-layer communication entity according to the federated learning preparation request.

[0143] The response information is used to indicate whether to join the federated learning system where the upper-layer communication entity is located. When the response information is used to indicate not to join the federated learning system where the upper-layer communication entity is located, the response information includes reason information.

[0144] In one embodiment, the reason information may be used to indicate that the computing power does not meet the conditions, etc.

[0145] In one embodiment, after step 1103, indication information sent by an upper-layer communication entity may also be received, where the indication information is used to indicate that the intermediate-layer server federated learning communication entity or the client federated learning communication entity be used as its client.

[0146] When the method is applied to a middle-layer server federated learning communication entity, the middle-layer server federated learning communication entity needs to determine its next-layer communication entity, so the method may further include the following steps.

[0147] Step 1104: Send a communication entity discovery request to the network storage communication entity.

[0148] The implementation process and technical principles of step 1104 are similar to those of step 902 and will not be repeated here.

[0149] Step 1105: Receive communication entity discovery request response information fed back by the network storage communication entity.

[0150] The implementation process and technical principle of step 1105 are similar to those of step 903 and will not be repeated here.

[0151] Step 1106: Determine a candidate communication entity according to the communication entity discovery request response information.

[0152] In one embodiment, the implementation process of step 1106 may include: determining a candidate communication entity based on the communication entity discovery request response information and a preset selection rule. The communication entity included in the communication entity discovery request response information is a communication entity determined by the network storage communication entity based on the preset selection rule.

[0153] The implementation process and technical principle of step 1106 are similar to those of step 904 and will not be repeated here.

[0154] Step 1107: Send a federated learning preparation request to the candidate communication entity.

[0155] The federated learning preparation request includes at least one of the following: the level of the intermediate layer server federated learning communication entity, the identifier of the intermediate layer server federated learning communication entity, and the identifier of the upper layer communication entity of the intermediate layer server federated learning communication entity.

[0156] In one embodiment, the intermediate layer server federated learning communication entity may send a federated learning preparation request to the candidate communication entity through the Nnwdaf_MLModelTraining_Subscribe or Nnwdaf_MLModelTrainingInfo_Request service to check whether the candidate communication entity meets the model training requirements.

[0157] To prevent the higher-level NWDAF from being reselected as an NWDAF client or NWDAF server at a lower level, the federated learning preparation request in this embodiment includes: the level of the federated learning communication entity of the middle-level server, the identifier of the federated learning communication entity of the middle-level server, and the identifier of the communication entity above the middle-level server federated learning communication entity. In one embodiment, the identifier of the federated learning communication entity can be an NF ID or a vendor ID.

[0158] Step 1108: Receive response information sent by the candidate communication entity in response to the federated learning preparation request.

[0159] Step 1109: According to a preset selection rule, determine the next layer of communication entities from the candidate communication entities whose corresponding response information is the target response information.

[0160] The target response information is used to instruct the candidate communication entity to join the layered federated learning where the intermediate layer server federated learning communication entity is located.

[0161] In one embodiment, the preset selection rules include at least one of the following: computing power information of the communication entity, whether the model corresponding to the analysis identifier needs to be trained through a layered federated learning process, model filtering information, the federated learning capability type of the communication entity, whether the communication entity is currently performing other layered federated learning, the time period of interest, the service area, the type of network function of the data source required for model training, the interoperability index of the model corresponding to the analysis identifier, the available data requirements, and the available time requirements.

[0162] Available data requirements include at least one of the following: a list of event IDs of local data used for training, dataset statistical properties, the time window of data samples, and the minimum number of data samples. Available time requirements include the time span required for the model training process.

[0163] It should be noted that the preset selection rules used by the NRF in determining the initial candidate communication entity, the preset selection rules used by the intermediate-layer server federated learning communication entity in determining the candidate communication entity, and the preset selection rules used by the intermediate-layer server federated learning communication entity in determining the next-layer communication entity may be the same or different.

[0164] After step 1109, steps 1001 to 1003 in the embodiment shown in Figure 10 may also be performed, which will not be described in detail in this embodiment.

[0165] The model training method provided in this embodiment provides a method for determining the next-level communication entity. The next-level communication entity determined based on this method can meet the model training requirements, thereby improving the efficiency and success rate of model training. In addition, the signaling interaction method for determining the next-level communication entity in this embodiment is simple and efficient, further improving the efficiency of model training. Moreover, based on this embodiment, it is possible to flexibly establish server-client relationships between federated learning communication entities.

[0166] Figure 12 is a schematic diagram of signaling interactions for a model training method provided in one embodiment. Building on the previous embodiments, this embodiment illustrates how the model training method constructs a layered federated learning process. This process can also be referred to as the layered federated learning participant selection process. As shown in Figure 12, the process includes the following steps.

[0167] Step 1201a: The server NWDAF sends the federated learning capability type to the NRF.

[0168] Step 1201b: The client NWDAF sends the federated learning capability type to the NRF.

[0169] The server NWDAF in this embodiment includes a master server federated learning communication entity and an intermediate server federated learning communication entity. The client NWDAF in this embodiment includes an intermediate server federated learning communication entity and a client federated learning communication entity. FIG12 illustrates the example with the number of client NWDAFs being n. n is an integer greater than 1.

[0170] In step 1201a and step 1201b, the federated learning capability type may be sent via the Nnrf_NFManagement_NFRegister_request service operation. The specific content of the federated learning capability type is as described in the above embodiment and will not be repeated here.

[0171] Step 1202: The NRF stores the received federated learning capability type.

[0172] Step 1203a: The NRF sends a registration response message to the client NWDAF.

[0173] Step 1203b: The NRF sends a registration response message to the server NWDAF.

[0174] In one embodiment, in step 1203a and step 1203b, the NRF may send a registration response message through the Nnrf_NFManagement_NFRegister_response service operation.

[0175] The process from step 1201a to step 1203b may also be referred to as a NWDAF registration process.

[0176] The NWDAF including the MTLF determines, based on the analysis ID (e.g., signaling storm analysis ID) and the service area / data network access identifier (DNAI), that the ML model needs to be trained through a layered federated learning process or that data cannot be directly obtained from the data producer NF (e.g., due to privacy reasons).

[0177] If the NWDAF with MTLF cannot serve as the master server (i.e., the master federated learning communication entity) in the layered federated learning process, the NWDAF with MTLF first discovers and selects the master NWDAF (i.e., the master federated learning communication entity) for federated learning from the NRF by calling the Nnrf_NFDiscovery_Request service operation. The NRF may use at least one of the following criteria to determine the master NWDAF: the computing power information of the NWDAF, the analysis ID of the required ML model, the model filtering information defined in TS23.288, the federated learning capability type (whether it supports serving as the master server for layered federated learning), whether the selected federated learning server NWDAF is currently performing other multi-layer federated learning, the time period of interest, the service area, etc.

[0178] In one embodiment, a consumer-side communication entity can discover a master server through an NRF. The consumer-side communication entity can send a master server discovery request to the NRF. After receiving the master server discovery request, the NRF determines the master server federated learning communication entity based on the federated learning capability type reported by each federated learning communication entity and initiates the hierarchical federated learning process.

[0179] After selecting the NWDAF server at layer 0, the NWDAFs at the lower layers are selected layer by layer. The master server NWDAF can send the client NWDAF with appropriate federated learning capabilities through NRF for local model training.

[0180] Step 1204: The server NWDAF sends a communication entity discovery request to the NRF.

[0181] In one embodiment, the server NWDAF sends a communication entity discovery request by calling the Nnrf_NFDiscovery_Request service operation.

[0182] Step 1205: The NRF authorizes the communication entity to discover the service.

[0183] Step 1206: The NRF sends a communication entity discovery request response message to the server NWDAF.

[0184] In one embodiment, the NRF sends a communication entity discovery request response message via the Nnrf_NFDiscovery_RequestResponse service operation.

[0185] The communication entity discovery request response information includes: the initial candidate communication entity determined by the NRF according to the preset selection rules. The preset selection rules include at least one of the following information: NWDAF computing power information, the analysis ID of the required ML model, whether an additional NWDAF client is required for layered federated learning, model filtering information defined in TS23.288, federated learning capability type (whether it supports serving as a server in the middle layer of layered federated learning, or a client in the middle layer or the bottom layer), the selected server NWDAF, whether the NWDAF is currently performing other layered federated learning, the time period of interest, the service area, the NF type of the data source of the federated learning training data, the time period of interest, and the ML model interoperability indicator.

[0186] After receiving the communication entity discovery request response information, the server NWDAF determines candidate communication entities based on the communication entity discovery request response information. Specifically, the candidate communication entities can be determined based on the communication entity discovery request response information and preset selection rules. It will be understood that in this embodiment, the candidate communication entities are several of the client NWDAFs.

[0187] The process from step 1204 to step 1206 may also be referred to as a discovery process of NWDAF.

[0188] Step 1207: The server NWDAF sends a federated learning preparation request to the candidate communication entity in the client NWDAF.

[0189] In one embodiment, the server NWDAF can send a federated learning preparation request to the candidate communication entity through the Nnwdaf_MLModelTraining_Subscribe or Nnwdaf_MLModelTrainingInfo_Request service to check whether the candidate communication entity in the client meets the model training requirements. The model training requirements here include at least one of the following: analysis ID, ML model interoperability information, available data requirements (including at least one of the following: event ID list of local data for training, dataset statistical properties, time window of data samples and minimum number of data samples), NWDAF computing power information, available time requirements (such as the time span required for the federated learning process), etc.

[0190] The candidate communication entity in this embodiment is the NWDAF including the MTLF.

[0191] To prevent a higher-level NWDAF from being reselected as an NWDAF client or NWDAF server at a lower level, the federated learning preparation request in this embodiment includes the server NWDAF's level, its identifier, and the identifier of its parent server NWDAF. The server NWDAF can be either an NF ID or a vendor ID.

[0192] Step 1208: The candidate communication entity in the client NWDAF decides whether to join the joint learning process.

[0193] The candidate communication entities in the client NWDAF check whether they can meet the ML model training requirements and / or successfully download the model if the model information is provided in the federated learning preparation request, and decide whether to join the federated learning (i.e., layered federated learning) process based on the execution results. The standard examples used by the client NWDAF can be based on their availability, computing and communication capabilities, and ML model interoperability information.

[0194] Step 1209: The candidate communication entity in the client NWDAF sends a response message to the server NWDAF according to the federated learning preparation request.

[0195] In one embodiment, the candidate communication entity in the client NWDAF calls the Nnwdaf_MLModelTraining_Subscribe response service operation or the Nnwdaf_MLModelTrainingInfo_Request response service operation to instruct the server NWDAF whether to join the layered federated learning process. If it cannot join the layered federated learning process, the reason (such as insufficient computing power, etc.) can be stated in the response message.

[0196] Step 12010: The server NWDAF selects its client NWDAF.

[0197] It can be understood that the client NWDAF of the server NWDAF is the next layer communication entity.

[0198] The specific implementation process of step 12010 may be: according to a preset selection rule, determine the next layer of communication entities from the candidate communication entities whose corresponding response information is the target response information. The target response information is used to instruct the candidate communication entity to join the layered federated learning where the master server federated learning communication entity is located.

[0199] Steps 1204 to 12010 may be repeated until all NWDAFs with server capabilities no longer need to select clients for model training.

[0200] Taking Figure 6 as an example, the above process is explained. The NRF first selects NWDAF 1 as the master server NWDAF. Then, NWDAF 1 executes steps 1204 to 12010 to select its client NWDAF. In Figure 6, the client NWDAFs selected by NWDAF 1 are NWDAF 2, NWDAF 3, and NWDAF 4. Then, NWDAF 2 determines that a client NWDAF needs to be selected, and NWDAF 4 also determines that a client NWDAF needs to be selected. NWDAF 2 (in this case, NWDAF 2 is the server NWDAF in Figure 12) executes steps 1204 to 12010 to select its client NWDAF. In Figure 6, the client NWDAF selected by NWDAF 2 is NWDAF 8. NWDAF 4 (in this case, NWDAF 4 is the server NWDAF in Figure 12) executes steps 1204 to 12010 to select its client NWDAF. In Figure 6 , NWDAF 4 selects NWDAF 6 and NWDAF 5 as client NWDAFs. NWDAF 5 then determines that it needs to select a client NWDAF. NWDAF 5 (which is the server NWDAF in Figure 12 ) executes steps 1204 through 12010 to select its client NWDAF. In Figure 6 , NWDAF 5 selects NWDAF 7 as the client NWDAF. This completes NWDAF selection, establishing a layered federated learning architecture.

[0201] The model training method provided in this embodiment provides a method for determining the next-level communication entity. The next-level communication entity determined based on this method can meet the model training requirements, thereby improving the efficiency and success rate of model training. In addition, the signaling interaction method for determining the next-level communication entity in this embodiment is simple and efficient, further improving the efficiency of model training. Furthermore, based on this embodiment, it is possible to flexibly establish a server-client relationship between NWDAFs.

[0202] Figure 13 is a schematic diagram of signaling interactions for another model training method provided by one embodiment. This embodiment, based on the above embodiment, illustrates how to implement the above model training method through signaling interactions. This process can also be referred to as a layered federated learning process. As shown in Figure 13, the process includes the following steps.

[0203] Step 1300: The consumer-side communication entity sends a model subscription request message to the main server NWDAF.

[0204] In one embodiment, the consumer communication entity uses the Nnwdaf_MLModelProvision service to send a model subscription request to the master server NWDAF to retrieve the ML model. The model subscription request message includes at least one of the following: an analysis identifier, a model metric, an accuracy reporting interval, and a training termination condition.

[0205] Step 1301: Each NWDAF performs the process of selecting participants for layered federated learning.

[0206] The process of selecting participants in layered federated learning is the same as the process in the embodiment shown in Figure 12, which will not be repeated here.

[0207] Step 1302a: The main server NWDAF sends model training instruction information to the next layer communication entity.

[0208] In one embodiment, the main server NWDAF sends Nnwdaf_MLModelTraining_Subscribe or Nnwdaf_MLModelTrainingInfo_Request to the selected NWDAF (client NWDAF or server NWDAF in the middle layer) participating in the layered federated learning and including MTLF to send model training instruction information to the next layer communication entity.

[0209] The client NWDAF or server NWDAF in the middle layer performs local model training and determines the local ML model information based on the input parameters (including model metrics and initial model information) in the model training instruction information sent by the master server NWDAF. The model training instruction information may also include the maximum response time that the client NWDAF must report the temporary local ML model information to the server NWDAF.

[0210] In this embodiment, the master server NWDAF includes a master server federated learning communication entity. The server NWDAF includes an intermediate server federated learning communication entity. The client NWDAF in this embodiment includes an intermediate server federated learning communication entity and a client federated learning communication entity.

[0211] Step 1302b: The server NWDAF in the middle layer sends model training instruction information to the client NWDAF in the lower layer.

[0212] In one embodiment, the intermediate-layer server NWDAF sends Nnwdaf_MLModelTraining_Subscribe or Nnwdaf_MLModelTrainingInfo_Request to the lower-layer client NWDAF to send model training instruction information (also known as a request to participate in federated learning training). The model training instruction information includes: initial model information. The model training instruction information may also include at least one of the following: model metrics and the maximum response time for reporting model information.

[0213] Step 1303: The client NWDAF that needs to perform model training collects training data.

[0214] The client NWDAF that needs to perform model training in this embodiment refers to the client federated learning communication entity in the above embodiment.

[0215] If the client NWDAF that needs to perform model training does not have available local data, each client NWDAF that needs to perform model training collects training data using the current mechanism. Each client NWDAF that needs to perform model training collects its own training data. In one embodiment, the client NWDAF that needs to perform model training can collect training data from the NRF or various NFs that provide data. The specific implementation and acquisition methods of training data are the same as those in the embodiment shown in Figure 10 and are not further described here.

[0216] Step 1304: Send the target intermediate model information to the upper-layer communication entity.

[0217] During the layered federated learning training process, each client NWDAF that needs to perform model training further trains the ML model provided by the server NWDAF based on its own data and reports the local ML model information (i.e., the target intermediate model information) to the server NWDAF in Nnwdaf_MLModelTraining_Notify or Nnwdaf_MLModelTrainingInfo_Response. Nnwdaf_MLModelTraining_Notify or Nnwdaf_MLModelTrainingInfo_Response may also include the local ML model metrics calculated by the client NWDAF that needs to perform model training and the training input data information in the client NWDAF (such as the area covered by the dataset, sampling rate, maximum / minimum values ​​of each dimension, etc.).

[0218] During the training process of layered federated learning, the ML model sent from the client NWDAF to the server NWDAF is the information required by the server NWDAF to build an aggregate model based on the local ML model.

[0219] If the client NWDAF that needs to perform model training cannot complete the training of the temporary local ML model within the maximum response time provided by the server NWDAF, the NWDAF that needs to perform model training should send a delay event notification, which includes a delay event indication, an optional reason code (such as local ML model training failed, local ML model training requires more time), and the expected time to complete the training before the end of the maximum response time (if provided by the NWDAF that needs to perform model training).

[0220] The client NWDAF that needs to perform model training uses the data collected in step 1303 to perform model training. After the training is completed, the client NWDAF reports the local ML model information and the above-mentioned other information to the upper-level server NWDAF or the main server NWDAF.

[0221] Step 1305: The server NWDAF aggregates the model information reported by the lower-layer NWDAF.

[0222] The server NWDAF in the middle layer aggregates the model information received in step 1304 from all lower-layer NWDAFs to update the ML model for this node. The server NWDAF may also calculate ML model metrics for this node or apply the model for this node to a validation dataset (if available). The server NWDAF may perform aggregation each time some lower-layer client NWDAFs provide updated local ML model information, or the server NWDAF may decide to wait for local ML model information from all lower-layer client NWDAFs before aggregating the global ML model for this node.

[0223] Step 1306a: After aggregating the model, the intermediate server NWDAF sends the aggregated model information to the main server NWDAF.

[0224] In one embodiment, the intermediate server NWDAF sends the aggregated model information (ie, target intermediate model information) to the main server NWDAF via Nnwdaf_MLModelTraining_Notify.

[0225] Step 1306b: The master server NWDAF performs model aggregation.

[0226] The master server NWDAF updates the global model and calculates the model metric information of the global model (i.e., the target model).

[0227] Step 1307a (optional step): The main server NWDAF sends the model measurement information of the target model to the consumer-end communication entity.

[0228] Based on the request of the consumer communication entity in step 1300, the main server NWDAF can send an Nnwdaf_MLModelProvision_Notify message to dynamically update the model metric information of the model in the training process to the consumer communication entity regularly (such as a certain number of training rounds or every 10 minutes) or under certain preset circumstances.

[0229] Step 1307b (optional step): The consumer-end communication entity sends a training continuation instruction message or a training stop instruction message to the main server NWDAF.

[0230] The consumer-side communication entity determines whether the current model meets the requirements and decides to stop or continue the training process. The consumer-side communication entity calls the Nnwdaf_MLModelProvision_Subscribe service operation used in step 1300 to stop or continue the training process.

[0231] According to the subscription request sent by the consumer in step 1307b, the master server NWDAF updates or terminates the current federated learning training process. If the master server NWDAF receives a request to stop the federated learning process in step 1307b, steps 1309 and 13010 are skipped.

[0232] Step 1308: The master server NWDAF terminates or continues the layered federated learning according to the request of the consumer communication entity.

[0233] Step 1309a: If the layered federated learning process continues, the master server NWDAF sends the updated global model information to the next layer of communication entities.

[0234] Step 1309b: The server NWDAF transparently transmits the updated global model information to its corresponding lower or bottom layer client NWDAF or server NWDAF.

[0235] Step 13010: The client NWDAF that needs to perform model training continues the federated learning training and updates the local model.

[0236] Steps 1303 to 13010 may be repeatedly executed until the termination condition in step 1307b is reached.

[0237] The model training method provided in this embodiment can realize layered federated learning, so that the target model and target intermediate model information trained based on the model training method can be flexibly used, and it can realize the flexible establishment of server and client relationships between federated learning communication entities.

[0238] An embodiment of the present application also provides a model training system, including: a main server federated learning communication entity, an intermediate layer server federated learning communication entity, and a client federated learning communication entity.

[0239] The master server federated learning communication entity is used to execute the steps in the model training method previously described by the master server federated learning communication entity. The intermediate server federated learning communication entity is used to execute the steps in the model training method previously described by the intermediate server federated learning communication entity. The client federated learning communication entity is used to execute the steps in the model training method previously described by the client federated learning communication entity.

[0240] FIG14 is a flow chart of a model training method provided by an embodiment. The model training method provided by this embodiment is applied to NWDAF. As shown in FIG14 , the model training method includes the following steps.

[0241] Step 1401: Receive a signaling storm prediction request or a model training request from a network function or an operation and maintenance management entity.

[0242] Step 1402: Collect training data.

[0243] The training data includes at least one of the following: first training sub-data, second training sub-data, and third training sub-data. The first training sub-data includes at least one of the following: information about UEs in one or more areas, the number and time information of protocol data unit sessions in one or more areas, information about one or more network functions in one or more areas, and the amount of user plane data transmission in one or more areas. The second training sub-data includes label information corresponding to the first training sub-data. The third training sub-data includes external factors that affect information about UEs in the area.

[0244] In one embodiment, the first training sub-data is data acquired from the first network function.The first training sub-data includes at least one of the following: a historical number of UEs in one or more areas and time information related to the historical number of UEs, a current number of UEs in one or more areas and time information related to the current number of UEs, a change in the number of UEs in one or more areas at different times, a movement trajectory of the UE, areas passed by the UE, time information of areas passed by the UE, a historical number of UE registrations in one or more areas and time information related to the historical number of registrations, a current number of UE registrations in one or more areas and time information related to the current number of registrations, a historical registration type of the UE in one or more areas and time information related to the historical registration type, a current registration type of the UE in one or more areas and time information related to the current registration type, a historical number of protocol data unit sessions in one or more areas and time information related to the historical number of protocol data unit sessions, a current number of protocol data unit sessions in one or more areas and time information related to the current number of protocol data unit sessions, current bearer information of one or more network functions in one or more areas and time information related to the current bearer information, historical bearer information of one or more network functions in one or more areas and time information related to the historical bearer information, current bearer capacity of one or more network functions in one or more areas and time information related to the current bearer capacity, Historical carrying capacity of one or more network functions in one or more areas and time information related to the historical carrying capacity; current service area information, the current number of UEs served, the current number of network functions served, and time information related to the current service area information of one or more network functions in one or more areas; historical service area information, the historical number of UEs served, the historical number of network functions served, and time information related to the historical service area information of one or more network functions in one or more areas; change information of service area information at different times, the change in the number of UEs served, and the change in the number of network functions served in one or more areas; current user plane data transmission volume and time information related to the current user plane data transmission volume in one or more areas; historical user plane data transmission volume and time information related to the historical user plane data transmission volume in one or more areas; historical user equipment categories and time information corresponding to the historical user equipment categories in one or more areas; current user equipment categories and time information corresponding to the current user equipment categories in one or more areas; historical user equipment group information in one or more areas and time information corresponding to the historical user equipment group information; time information and area information of current user equipment inventory in one or more areas; time information and area information of historical user equipment inventory in one or more areas.

[0245] In one embodiment, the second training sub-data is data acquired from the second network function, wherein the second training sub-data includes: information related to whether a signaling storm occurs, corresponding time information, and location information.

[0246] In one embodiment, the third training sub-data is data obtained from a third network function. The third training sub-data includes at least one of the following: the historical number of users of the third network function and time information related to the historical number of users; the current number of users of the third network function and time information related to the current number of users; historical data traffic information of the third network function and time information related to the historical data traffic information; current data traffic information of the third network function and time information related to the current data traffic information; historical external factors affecting at least one of the number of UEs in the area, the number of registered UEs, and the amount of user-plane data transmission; and current external factors affecting at least one of the number of UEs in the area, the number of registered UEs, and the amount of user-plane data transmission. For example, the historical external factors and the current external factors may include concert information, flight information, etc.

[0247] In one embodiment, the first network function is a network function of the core network. In another embodiment, the first network function is a network function other than the NWDAF in the core network. The second network function is an operation and maintenance management entity. The third network function is an application function.

[0248] In one embodiment, the training data further includes: fourth training sub-data obtained from other NWDAFs. The fourth training sub-data includes at least one of the following: network function load information (NF load information) and related information, abnormal behavior of UE or 5GC NF and related information, suspected distributed denial of service (DDoS) attack (Suspicions of DDoS attack) and related information, UE session management congestion control experience (SMS Congestion Control Experience) and related information, user data congestion analysis (User data congestion analytic) and related information, session unit congestion analysis (PDU session traffic analytics) and related information, end-to-end data volume transfer time analysis (E2E data volume transfer time analytics) and related information, packet filtering description decision analysis (PFD determination analytics) and related information.

[0249] In another embodiment, the fourth training sub-data includes at least one of the following: current load information of the network function and time information corresponding to the current load information, and relevant input and output information of network load analysis; historical load information of the network function and time information corresponding to the historical load information, and relevant input and output information of network load analysis; current abnormal behavior of the network function or UE and time information corresponding to the current abnormal behavior, and relevant input and output information of abnormal behavior analysis; historical abnormal behavior of the network function or UE and time information corresponding to historical abnormal behavior, and relevant input and output information of historical abnormal behavior analysis; current suspected DDOS attack and time information corresponding to the current suspected DDOS attack; historical suspected DDOS attack and time information corresponding to the historical suspected DDOS attack; current session management congestion control experience of the UE and time information corresponding to the current session management congestion control experience, and relevant input and output information of UE session management congestion control experience analysis; historical session management congestion control experience of the UE and time information corresponding to the historical session management congestion control experience, and relevant input and output information of historical UE session management congestion control experience analysis; current user data congestion analysis and time information corresponding to the current user data The present invention also includes the time information corresponding to the congestion analysis and the related input and output information of the user data congestion analysis, the historical user data congestion analysis and the time information corresponding to the historical user data congestion analysis and the input and output information of the related historical user data congestion analysis, the current session unit congestion analysis and the time information corresponding to the current session unit congestion analysis and the related input and output information of the session unit congestion analysis, the historical session unit congestion analysis and the time information corresponding to the historical session unit congestion analysis and the related input and output information of the historical session unit congestion analysis, the current end-to-end data volume transmission time analysis and the time information corresponding to the current end-to-end data volume transmission time analysis and the related input and output information of the end-to-end data volume transmission analysis, the historical end-to-end data volume transmission time analysis and the time information corresponding to the historical end-to-end data volume transmission time analysis and the input and output information of the related historical current end-to-end data volume transmission analysis, the current data packet filtering description decision analysis and the time information corresponding to the current data packet filtering description decision analysis and the related input and output information of the data packet filtering description decision analysis, the historical data packet filtering description decision analysis and the time information corresponding to the historical data packet filtering description decision analysis and the related input and output information of the historical data packet filtering description decision analysis.

[0250] In this embodiment, the other network elements may be NWDAFs. The NWDAFs herein refer to other NWDAFs other than the NWDAF currently executing the model training method.

[0251] Step 1403: Perform model training based on the training data to obtain a target model.

[0252] The target model is a model for predicting signaling storm related information.

[0253] In one embodiment, the output information of the target model includes at least one of the following: information about the area where the signaling storm occurs, the time when the signaling storm occurs, the network function affected by the signaling storm, and the user equipment UE affected by the signaling storm.

[0254] Step 1404: Use the target model to perform reasoning and output information related to signaling storm prediction, or provide the target model to the service client.

[0255] The service client in step 1404 may be a network function or an operation and maintenance management entity.

[0256] Step 1404 includes two scenarios. In the first scenario, the NWDAF uses the target model for inference and outputs information related to signaling storm prediction. This information can be output to the service client. The service client can take measures based on this information, such as notifying potentially affected NFs or UEs in advance to enable early prevention. In the second scenario, the target model is provided to the service client.

[0257] The model training method provided in this embodiment implements model training based on a single NWDAF, performs inference based on a target model, outputs information related to signaling storm prediction, or provides the target model to service customers, thereby implementing signaling storm prediction based on a single NWDAF.

[0258] Figure 15 is a schematic diagram of signaling interaction of a model training method provided by an embodiment. Based on the embodiment shown in Figure 14, this embodiment describes in detail the signaling interaction process of the model training method. As shown in Figure 15, the model training method provided by this embodiment includes the following steps.

[0259] Step 1500: The consumer-side communication entity subscribes to the NWDAF for related analysis of signaling storms.

[0260] In one embodiment, the consumer-side communication entity subscribes to the NWDAF for signaling storm analysis via AnalyticsSubscribe / request. The analysis ID is "signaling storm analytic".

[0261] The consumer-end communication entity in this embodiment includes at least one of the following: NF and OAM in 5GC.

[0262] Step 1501a: NWDAF subscribes to or requests data from 5GC NF.

[0263] To provide signaling storm analysis, train a model to predict signaling storms, or output signaling storm statistics, the NWDAF subscribes to or requests data from the 5GC NF. The requested data may be the first training sub-data or the fourth training sub-data in the embodiment shown in FIG14 .

[0264] Step 1501b: 5GC NF returns data to NWDAF.

[0265] Step 1502a: NWDAF subscribes to or requests data from OAM.

[0266] To provide signaling storm analysis, train models to predict signaling storms, or output signaling storm statistics, the NWDAF subscribes to or requests data from the OAM. The requested data can be the second training sub-data in the embodiment shown in Figure 14 , used as label data for machine learning or as ground truth data.

[0267] Step 1502b: OAM returns data to NWDAF.

[0268] Step 1503: NWDAF performs model training based on the training data.

[0269] The NWDAF uses the collected data to train models for prediction or statistical purposes. In one embodiment, the NWDAF can also obtain data from the AF for model training. The specific data requested can be the third training sub-data in the embodiment shown in Figure 14.

[0270] Step 1504: The NWDAF returns signaling storm prediction related information to the consumer-end communication entity, or provides the target model to the consumer-end communication entity.

[0271] Afterwards, the consumer-side communication entity can take measures based on the signaling storm prediction-related information, for example, notifying the NF or UE that may be affected in advance so that it can take precautions as early as possible.

[0272] The model training method provided in this embodiment implements model training based on a single NWDAF, performs inference based on a target model, outputs information related to signaling storm prediction, or provides the target model to service customers, thereby implementing signaling storm prediction based on a single NWDAF.

[0273] An embodiment of the present application further provides a communication entity, including: a processor, configured to implement the model training method provided in any embodiment of the present application when executing a computer program. The communication entity in this embodiment can be a main server federated learning communication entity, an intermediate server federated learning communication entity, a client federated learning communication entity, or a network data analysis function.

[0274] Figure 16 is a schematic diagram of the structure of a communication entity provided by one embodiment. As shown in Figure 16, the communication entity includes a processor 60, a memory 61, and a communication interface 62. The number of processors 60 in the communication entity can be one or more, and Figure 16 uses one processor 60 as an example. The processor 60, memory 61, and communication interface 62 in the communication entity can be connected via a bus or other means, and Figure 16 uses a bus connection as an example. The bus represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus structures.

[0275] The memory 61, as a computer-readable storage medium, can be configured to store software programs, computer-executable programs, and modules, such as program instructions / modules corresponding to the methods in the embodiments of the present application. The processor 60 executes the software programs, instructions, and modules stored in the memory 61 to execute at least one functional application and data processing of the communication entity, thereby implementing the above-mentioned method.

[0276] Memory 61 may include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function; the data storage area may store data generated based on the use of the terminal. Furthermore, memory 61 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state memory device. In some instances, memory 61 may include memory remotely located relative to processor 60, and such remote memory may be connected to a communication entity via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a network, a mobile communication network, and combinations thereof.

[0277] The communication interface 62 can be configured to receive and send data.

[0278] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method provided in any embodiment of the present application is implemented.

[0279] The computer storage medium of the embodiment of the present application can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. Computer-readable storage media include (non-exhaustive list): an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a 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, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0280] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, the data signal carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0281] The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, radio frequency (RF), etc., or any suitable combination of the foregoing.

[0282] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination of multiple programming languages, including object-oriented programming languages ​​(such as Java, Smalltalk, C++, Ruby, Go), and conventional procedural programming languages ​​(such as "C" or similar programming languages). The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may 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 may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0283] It will be appreciated by those skilled in the art that the term user terminal covers any suitable type of wireless user equipment, such as a mobile phone, a portable data processing device, a portable web browser or a vehicle-mounted mobile station.

[0284] In general, various embodiments of the present application may be implemented in hardware or dedicated circuits, software, logic, or any combination thereof. For example, some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device, although the present application is not limited thereto.

[0285] Embodiments of the present application may be implemented by executing computer program instructions by a data processor of a mobile device, for example, in a processor entity, or by hardware, or by a combination of software and hardware. The computer program instructions may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages.

[0286] Any block diagram of a logical flow in the drawings of this application may represent program steps, or may represent interconnected logical circuits, modules and functions, or may represent a combination of program steps and logical circuits, modules and functions. A computer program may be stored on a memory. The memory may be of any type suitable for the local technical environment and may be implemented using any suitable data storage technology, such as, but not limited to, read-only memory (ROM), random access memory (RAM), optical memory devices and systems (digital versatile discs DVD or CD), etc. Computer-readable media may include non-transitory storage media. A data processor may be of any type suitable for the local technical environment, such as, but not limited to, a general-purpose computer, a special-purpose computer, a microprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a programmable logic device (FPGA), and a processor based on a multi-core processor architecture.

Claims

1. A model training method, wherein: Applied to a master server federated learning communication entity, the method includes: Sending model training instruction information to a next-layer communication entity; wherein the model training instruction information includes: initial model information, and the next-layer communication entity includes an intermediate-layer server federated learning communication entity, or includes an intermediate-layer server federated learning communication entity and a client federated learning communication entity; Receiving target intermediate model information determined by the next layer communication entity according to the model training instruction information; A target model is determined according to the target intermediate model information.

2. The method according to claim 1, wherein The method further comprises: Receiving a model subscription request message sent by a consumer communication entity; wherein the model subscription request message includes at least one of the following: an analysis identifier, a model metric, an accuracy reporting interval, and a training termination condition; Determine the model training instruction information according to the model subscription request message.

3. The method according to claim 2, wherein: The model training indication information also includes at least one of the following: the model metric standard and the maximum response time for reporting model information.

4. The method according to claim 2, wherein: The method further comprises: Determining model metric information of the target model according to the model metric standard; Sending model measurement information of the target model to the consumer-end communication entity.

5. The method according to claim 4, wherein The method further comprises: If a continue training instruction message is received from the consumer-side communication entity based on the model measurement information of the target model, the target model is determined as the new initial model, and the process returns to the step of "sending model training instruction message to the next-layer communication entity".

6. The method according to claim 2, wherein: The method further comprises: Send a federated learning capability type to the network storage communication entity; wherein the federated learning capability type includes at least one of the following: whether it supports serving as a master server federated learning communication entity for layered federated learning, whether it supports serving as an intermediate-layer server federated learning communication entity for layered federated learning, whether it supports serving as a client federated learning communication entity for layered federated learning, and whether it supports serving as both an intermediate-layer server federated learning communication entity and a client federated learning communication entity for layered federated learning.

7. The method according to claim 6, wherein: The method further comprises: Sending a communication entity discovery request to the network storage communication entity; Receiving communication entity discovery request response information fed back by the network storage communication entity; Determine a candidate communication entity according to the communication entity discovery request response information.

8. The method according to claim 7, wherein: The determining, according to the communication entity discovery request response information, a candidate communication entity includes: The candidate communication entity is determined according to the communication entity discovery request response information and a preset selection rule; wherein the communication entity included in the communication entity discovery request response information is the communication entity determined by the network storage communication entity according to the preset selection rule.

9. The method according to claim 8, wherein The method further comprises: Send a federated learning preparation request to the candidate communication entity; wherein the federated learning preparation request includes the following At least one item: the level of the master server federated learning communication entity and the identifier of the master server federated learning communication entity; receiving response information sent by the candidate communication entity in response to the federated learning preparation request; According to the preset selection rule, the next layer communication entity is determined from the candidate communication entities whose corresponding response information is the target response information; wherein the target response information is used to instruct the candidate communication entity to join the layered federated learning where the main server federated learning communication entity is located.

10. The method according to claim 2, wherein: The master server federated learning communication entity is a communication entity determined by the network storage communication entity according to a preset selection rule.

11. The method according to any one of claims 8 to 10, wherein: The preset selection rules include at least one of the following: computing power information of the communication entity, whether the model corresponding to the analysis identifier needs to be trained through a layered federated learning process, model filtering information, the federated learning capability type of the communication entity, whether the communication entity is currently performing other layered federated learning, the time period of interest, the service area, the type of network function of the data source required for model training, the interoperability index of the model corresponding to the analysis identifier, the available data requirements, and the available time requirements.

12. The method according to any one of claims 1 to 10, wherein: The target model is a model for predicting information related to a signaling storm, and the output information of the target model includes at least one of the following: information about the area where the signaling storm occurs, the time when the signaling storm occurs, the network function affected by the signaling storm, and the user equipment UE affected by the signaling storm.

13. The method according to any one of claims 1 to 10, wherein: The master server federated learning communication entity and the next layer communication entity are used to implement layered federated learning, the number of layers of the layered federated learning is greater than 2, and the master server federated learning communication entity is located at the highest layer of the layered federated learning.

14. A model training method, wherein: Applied to an intermediate server federated learning communication entity or a client federated learning communication entity, the method includes: Receiving model training instruction information sent by an upper-layer communication entity; wherein the model training instruction information includes: initial model information, and the upper-layer communication entity includes a main server federated learning communication entity and / or an intermediate-layer server federated learning communication entity; Determining target intermediate model information according to the model training instruction information; Send the target intermediate model information to the upper-layer communication entity.

15. The method according to claim 14, wherein The model training instruction information also includes at least one of the following: a model metric standard and a maximum response time for reporting model information.

16. The method according to claim 14, wherein When the method is applied to the client federated learning communication entity, the target intermediate model information includes: target sub-model information; The determining target intermediate model information according to the model training instruction information includes: Collect training data; The initial model information is trained according to the training data to obtain target sub-model information.

17. The method according to claim 16, wherein The training data includes at least one of the following: The first training sub-data obtained from the first network function; wherein the first training sub-data includes at least one of the following: the historical number of UEs in one or more areas and time information related to the historical number of UEs, the current number of UEs in one or more areas and time information related to the current number of UEs, the change in the number of UEs at different times, the movement trajectory of the UE, the areas passed by the UE, time information of the areas passed by the UE, the historical number of UE registrations in one or more areas and time information related to the historical number of registrations, the current number of UE registrations in one or more areas and time information related to the current number of registrations, the historical registration types of UEs in one or more areas and time information related to the historical registration types, the current registration types of UEs in one or more areas and time information related to the current registration types, the historical number of protocol data unit sessions in one or more areas and time information related to the historical number of protocol data unit sessions, the current number of protocol data unit sessions in one or more areas and time information related to the current number of protocol data unit sessions, current bearer information of one or more network functions in one or more areas and time information related to the current bearer information, historical bearer information of one or more network functions in one or more areas and time information related to the historical bearer information, current bearer capacity of one or more network functions in one or more areas and time information related to the current bearer capacity, historical bearer capacity of one or more network functions in one or more areas and time information related to the historical bearer capacity, Current service area information, the number of UEs currently served, the number of network functions currently served, and time information related to the current service area information for one or more network functions in the region; historical service area information, the number of UEs historically served, the number of network functions historically served, and time information related to the historical service area information for one or more network functions in the region; change information about service area information at different times for one or more network functions in the region, the amount of change in the number of UEs served, and the amount of change in the number of network functions served; current user plane data transmission volume and time information related to the current user plane data transmission volume in the region; historical user plane data transmission volume and time information related to the historical user plane data transmission volume in the region; historical user equipment categories and time information corresponding to the historical user equipment categories in the region; current user equipment categories and time information corresponding to the current user equipment categories in the region; historical user equipment group information and time information corresponding to the historical user equipment group information in the region; time information and region information of a current user equipment inventory in the region; time information and region information of a historical user equipment inventory in the region; Second training sub-data obtained from the second network function; wherein the second training sub-data includes: information related to whether a signaling storm has occurred, corresponding time information, and location information; Third training sub-data obtained from a third network function; wherein the third training sub-data includes at least one of the following: a historical number of users of the third network function and time information related to the historical number of users, a current number of users of the third network function and time information related to the current number of users, historical data traffic information of the third network function and time information related to the historical data traffic information, current data traffic information of the third network function and time information related to the current data traffic information, historical external factors affecting at least one of the number of UEs in the area, the number of registered UEs, and the amount of user-plane data transmission, and current external factors affecting at least one of the number of UEs in the area, the number of registered UEs, and the amount of user-plane data transmission.

18. The method according to claim 17, wherein The first network function is a network function of a core network, the second network function is an operation and maintenance management entity, and the third network function is an application function.

19. The method according to claim 15, wherein When the method is applied to the intermediate layer server federated learning communication entity, the target intermediate model information includes: current summary model information of the level where the intermediate layer server federated learning communication entity is located; The determining target intermediate model information according to the model training instruction information includes: Sending the model training instruction information to a next-layer communication entity; wherein the next-layer communication entity includes an intermediate-layer server federated learning communication entity and / or a client federated learning communication entity; Receiving target sub-model information and / or next-level summary model information determined by the next-layer communication entity according to the model training instruction information; The target sub-model information and / or the next-level summary model information are summarized to obtain the current summary model information.

20. The method according to claim 19, wherein The method further comprises: Determining model metric information of the current summary model information according to the model metric standard; Sending model measurement information of the current aggregated model information to the upper-layer communication entity.

21. The method according to claim 19, wherein The receiving target sub-model information and / or next-level summary model information determined by the next-layer communication entity according to the model training instruction information includes: Receive target sub-model information and / or next-level summary model information determined by all next-layer communication entities connected to the intermediate-layer server federated learning communication entity according to the model training instruction information; or, Receive target sub-model information and / or next-level summary model information determined by some next-layer communication entities connected to the intermediate-layer server federated learning communication entity according to the model training instruction information.

22. The method according to claim 14, wherein The method further comprises: Send a federated learning capability type to the network storage communication entity; wherein the federated learning capability type includes at least one of the following: whether it supports serving as a master server federated learning communication entity for layered federated learning, whether it supports serving as an intermediate-layer server federated learning communication entity for layered federated learning, whether it supports serving as a client federated learning communication entity for layered federated learning, and whether it supports serving as both an intermediate-layer server federated learning communication entity and a client federated learning communication entity for layered federated learning.

23. The method according to claim 22, wherein When the method is applied to the intermediate layer server federated learning communication entity, the method further includes: Sending a communication entity discovery request to the network storage communication entity; Receiving communication entity discovery request response information fed back by the network storage communication entity; Determine a candidate communication entity according to the communication entity discovery request response information.

24. The method according to claim 23, wherein The determining, according to the communication entity discovery request response information, a candidate communication entity includes: The candidate communication entity is determined according to the communication entity discovery request response information and a preset selection rule; wherein the communication entity included in the communication entity discovery request response information is the communication entity determined by the network storage communication entity according to the preset selection rule.

25. The method according to claim 24, wherein The method further comprises: Sending a federated learning preparation request to the candidate communication entity; wherein the federated learning preparation request includes at least one of the following: the level of the intermediate-layer server federated learning communication entity, the identifier of the intermediate-layer server federated learning communication entity, and the identifier of the upper-layer communication entity of the intermediate-layer server federated learning communication entity; receiving response information sent by the candidate communication entity in response to the federated learning preparation request; According to the preset selection rule, the next layer communication entity is determined from the candidate communication entities whose corresponding response information is the target response information; wherein the target response information is used to instruct the candidate communication entity to join the intermediate In the layered federated learning, the layer server federated learning communication entity is located.

26. The method according to claim 24 or 25, wherein The preset selection rules include at least one of the following: computing power information of the communication entity, whether the model corresponding to the analysis identifier needs to be trained through a layered federated learning process, model filtering information, the federated learning capability type of the communication entity, whether the communication entity is currently performing other layered federated learning, the time period of interest, the service area, the type of network function of the data source required for model training, the interoperability index of the model corresponding to the analysis identifier, the available data requirements, and the available time requirements.

27. The method according to claim 14, wherein The method further comprises: Receive the federated learning preparation request sent by the upper layer communication entity; Sending response information to the upper-layer communication entity according to the federated learning preparation request; wherein the response information is used to indicate whether to join the federated learning system where the upper-layer communication entity is located, and when the response information is used to indicate not to join the federated learning system where the upper-layer communication entity is located, the response information includes reason information.

28. The method according to claim 17, wherein The training data also includes: Fourth training sub-data obtained from other network elements; wherein the fourth training sub-data includes at least one of the following: current load information of the network function and time information corresponding to the current load information, and relevant input and output information of network load analysis; historical load information of the network function and time information corresponding to the historical load information, and relevant input and output information of network load analysis; current abnormal behavior of the network function or UE and time information corresponding to the current abnormal behavior, and relevant input and output information of abnormal behavior analysis; historical abnormal behavior of the network function or UE and time information corresponding to historical abnormal behavior, and relevant input and output information of historical abnormal behavior analysis; current suspected DDOS attack and time information corresponding to the current suspected DDOS attack; historical suspected DDOS attack and time information corresponding to the historical suspected DDOS attack; current session management congestion control experience of the UE and time information corresponding to the current session management congestion control experience, and relevant input and output information of UE session management congestion control experience analysis; historical session management congestion control experience of the UE and time information corresponding to the historical session management congestion control experience, and relevant input and output information of historical UE session management congestion control experience analysis; current user data congestion analysis and and the time information corresponding to the current user data congestion analysis and the related input and output information of the user data congestion analysis, the historical user data congestion analysis and the time information corresponding to the historical user data congestion analysis and the input and output information of the related historical user data congestion analysis, the current session unit congestion analysis and the time information corresponding to the current session unit congestion analysis and the related input and output information of the session unit congestion analysis, the historical session unit congestion analysis and the time information corresponding to the historical session unit congestion analysis and the related input and output information of the historical session unit congestion analysis, the current end-to-end data volume transmission time analysis and the time information corresponding to the current end-to-end data volume transmission time analysis and the related input and output information of the end-to-end data volume transmission analysis, the historical end-to-end data volume transmission time analysis and the time information corresponding to the historical end-to-end data volume transmission time analysis and the input and output information of the related historical current end-to-end data volume transmission analysis, the current data packet filtering description decision analysis and the time information corresponding to the current data packet filtering description decision analysis and the related input and output information of the data packet filtering description decision analysis, the historical data packet filtering description decision analysis and the time information corresponding to the historical data packet filtering description decision analysis and the related input and output information of the historical data packet filtering description decision analysis.

29. A model training method, wherein: Applied to network data analysis functions, the method includes: receiving a signaling storm prediction request or a model training request from a network function or an operation and maintenance management entity; Collect training data; wherein, the training data includes at least one of the following: first training sub-data, second training sub-data and third training sub-data; the first training sub-data includes at least one of the following: information about UEs in one or more areas, the number and time information of protocol data unit sessions in one or more areas, information about one or more network functions in one or more areas, and the amount of user plane data transmission in one or more areas; the second training sub-data includes: label information corresponding to the first training sub-data; the third training sub-data includes: external factors affecting the information of UEs in the area; Performing model training based on the training data to obtain a target model; wherein the target model is a model for predicting signaling storm related information; The target model is used for reasoning to output information related to signaling storm prediction, or the target model is provided to a service client.

30. The method according to claim 29, wherein The training data includes at least one of the following: First training sub-data obtained from the first network function; wherein the first training sub-data includes at least one of the following: a historical number of UEs in one or more areas and time information related to the historical number of UEs, a current number of UEs in one or more areas and time information related to the current number of UEs, a change in the number of UEs in the one or more areas at different times, a movement trajectory of the UE, an area passed by the UE, time information of the area passed by the UE, a historical number of UE registrations in the one or more areas and time information related to the historical number of registrations, a current number of UE registrations in the one or more areas and time information related to the current number of registrations, a historical registration type of the UE in the one or more areas and time information related to the historical registration type, a current registration type of the UE in the one or more areas and time information related to the current registration type, a historical number of protocol data unit sessions in the one or more areas and time information related to the historical number of protocol data unit sessions, a current number of protocol data unit sessions in the one or more areas and time information related to the current number of protocol data unit sessions, current bearer information of one or more network functions in the one or more areas and time information related to the current bearer information, historical bearer information of one or more network functions in the one or more areas and time information related to the historical bearer information, current bearer capacity of the one or more network functions in the one or more areas and time information related to the current bearer capacity time information related to bearer capacity, historical bearer capacity of one or more network functions in one or more areas and time information related to the historical bearer capacity, current service area information, the current number of UEs served, the current number of network functions served, and time information related to the current service area information of one or more network functions in one or more areas, historical service area information, the historical number of UEs served, the historical number of network functions served, and time information related to the historical service area information of one or more network functions in one or more areas, change information of service area information of one or more network functions at different times, the change in the number of UEs served, and the change in the number of network functions served in one or more areas, current user plane data transmission volume and time information related to the current user plane data transmission volume, historical user plane data transmission volume and time information related to the historical user plane data transmission volume, historical user equipment categories and time information corresponding to the historical user equipment categories in one or more areas, current user equipment categories and time information corresponding to the current user equipment categories in one or more areas, historical user equipment group information in one or more areas and time information corresponding to the historical user equipment group information, time information and area information of current user equipment inventory in one or more areas, time information and area information of historical user equipment inventory in one or more areas; Second training sub-data obtained from the second network function; wherein the second training sub-data includes: information related to whether a signaling storm has occurred, corresponding time information, and location information; Third training sub-data obtained from a third network function; wherein the third training sub-data includes at least one of the following: a historical number of users of the third network function and time information related to the historical number of users, a current number of users of the third network function and time information related to the current number of users, historical data traffic information of the third network function and time information related to the historical data traffic information, current data traffic information of the third network function and time information related to the current data traffic information, historical external factors affecting at least one of the number of UEs in the area, the number of registered UEs, and the amount of user-plane data transmission, and current external factors affecting at least one of the number of UEs in the area, the number of registered UEs, and the amount of user-plane data transmission.

31. The method according to claim 30, wherein The first network function is a network function of a core network, the second network function is an operation and maintenance management entity, and the third network function is an application function.

32. The method of claim 29, wherein: The output information of the target model includes at least one of the following: information about the area where the signaling storm occurs, the time when the signaling storm occurs, the network function affected by the signaling storm, and the user equipment UE affected by the signaling storm.

33. The method of claim 29, wherein: The training data also includes: Fourth training sub-data obtained from other network elements; wherein the fourth training sub-data includes at least one of the following: current load information of the network function and time information corresponding to the current load information, and relevant input and output information of network load analysis; historical load information of the network function and time information corresponding to the historical load information, and relevant input and output information of network load analysis; current abnormal behavior of the network function or UE and time information corresponding to the current abnormal behavior, and relevant input and output information of abnormal behavior analysis; historical abnormal behavior of the network function or UE and time information corresponding to historical abnormal behavior, and relevant input and output information of historical abnormal behavior analysis; current suspected DDOS attack and time information corresponding to the current suspected DDOS attack; historical suspected DDOS attack and time information corresponding to the historical suspected DDOS attack; current session management congestion control experience of the UE and time information corresponding to the current session management congestion control experience, and relevant input and output information of UE session management congestion control experience analysis; historical session management congestion control experience of the UE and time information corresponding to the historical session management congestion control experience, and relevant input and output information of historical UE session management congestion control experience analysis; current user data congestion analysis and and the time information corresponding to the current user data congestion analysis and the related input and output information of the user data congestion analysis, the historical user data congestion analysis and the time information corresponding to the historical user data congestion analysis and the input and output information of the related historical user data congestion analysis, the current session unit congestion analysis and the time information corresponding to the current session unit congestion analysis and the related input and output information of the session unit congestion analysis, the historical session unit congestion analysis and the time information corresponding to the historical session unit congestion analysis and the related input and output information of the historical session unit congestion analysis, the current end-to-end data volume transmission time analysis and the time information corresponding to the current end-to-end data volume transmission time analysis and the related input and output information of the end-to-end data volume transmission analysis, the historical end-to-end data volume transmission time analysis and the time information corresponding to the historical end-to-end data volume transmission time analysis and the input and output information of the related historical current end-to-end data volume transmission analysis, the current data packet filtering description decision analysis and the time information corresponding to the current data packet filtering description decision analysis and the related input and output information of the data packet filtering description decision analysis, the historical data packet filtering description decision analysis and the time information corresponding to the historical data packet filtering description decision analysis and the related input and output information of the historical data packet filtering description decision analysis.

34. The method according to claim 33, wherein The other network elements are network data analysis functions.

35. A model training system, wherein: Including the main server federated learning communication entity, the middle layer server federated learning communication entity and the client federated learning communication entity; The main server federated learning communication entity is used to perform the model training method according to any one of claims 1 to 13; The intermediate layer server federated learning communication entity or the client federated learning communication entity is used to execute the model training method as described in any one of claims 14 to 28.

36. A communication entity, wherein: include: processor; The processor is used to implement the model training method as described in any one of claims 1 to 13 when executing a computer program, or to implement the model training method as described in any one of claims 14 to 28, or to implement the model training method as described in any one of claims 29 to 33.

37. A computer-readable storage medium storing a computer program, wherein: When the computer program is executed by a processor, it implements the model training method as described in any one of claims 1 to 13, or implements the model training method as described in any one of claims 14 to 28, or implements the model training method as described in any one of claims 29 to 34.

Images