Model acquisition method, and related apparatus

Abstract

A model acquisition method, and a related apparatus. The method comprises: a first network element receiving a model acquisition request from a second network element, wherein the model acquisition request is used for requesting the acquisition of a model of the first network element, and the model acquisition request comprises first instruction information, the first instruction information being used for instructing the second network element to request the acquisition of a model generated by the first network element by using a transfer learning method, or the first instruction information being used for instructing the second network element to request the acquisition of a model in the first network element that satisfies transfer learning requirements; and the first network element sending a model acquisition response to the second network element. Therefore, the present invention is conducive to obtaining required models by using a transfer learning method, so that the model training efficiency is improved.

Description

Model acquisition method and related device

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on November 30, 2023, with application number 202311631741.5 and application name “Model Acquisition Method and Related Devices”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present application relates to the field of communication technology, and in particular to a model acquisition method and related devices. Background Art

[0003] The intelligent network architecture based on the network data analytics function (NWDAF) aims to collect massive amounts of information in the network and utilize existing big data and artificial intelligence technologies to utilize this massive amount of information, outputting some valuable information to assist operators in formulating strategies and adjusting network resources to improve user experience and reduce network load. NWDAF needs to collect data to provide corresponding analysis results. For example, in order to provide service experience analysis results, NWDAF needs to collect service-related information such as service identification and service experience from application functions (AF) / user equipment (UE), and collect information such as signal reception power and signal reception quality from operations, administration and management (OAM). NWDAF trains artificial intelligence (AI) models based on the collected information, and then obtains inference analysis results based on the AI ​​model, such as the predicted service experience for a certain time period in the future.

[0004] NWDAFs can request models from each other. For example, when a consumer NWDAF requests a model from a producer NWDAF, the producer NWDAF may not have the model that the consumer NWDAF wants. In this case, the producer NWDAF needs to retrain a model according to the requirements of the consumer NWDAF, resulting in low model training efficiency. Summary of the Invention

[0005] This application can provide a model acquisition method and related devices, which are conducive to improving model training efficiency.

[0006] In the first aspect, the present application provides a model acquisition method, which can be applied to a first network element, or a chip or chip module in the first network element, or to a module or unit that can realize all or part of the functions of the first network element, etc., and is described below using the first network element as an example. In this method, the first network element receives a model acquisition request from a second network element, and the model acquisition request is used to request to obtain the model of the first network element, and the model acquisition request includes first indication information; wherein the first indication information is used to instruct the second network element to request to obtain the model generated by the first network element using the transfer learning method, or to instruct the second network element to request to obtain the model in the first network element that meets the transfer learning requirements; the first network element sends a model acquisition response to the second network element.

[0007] It can be seen that the first indication information is used to indicate the situation where the second network element requests to obtain the model generated by the first network element using the transfer learning method. The first network element can use the transfer learning method to generate the model requested by the second network element, thereby avoiding the problem of low model training efficiency caused by retraining the corresponding model. The first indication information is used to indicate the situation where the second network element requests to obtain the model that meets the transfer learning requirements in the first network element. The first network element can provide the second network element with a model that meets the transfer learning requirements, thereby facilitating the second network element to obtain the required model based on the model using the transfer learning method, thereby improving the model training efficiency.

[0008] Optionally, the first network element may be an NWDAF network element including a model training logical function (MTLF), or an NWDAF network element including an MTLF and an analytics logical function (AnLF). Optionally, the second network element is an NWDAF network element service consumer (Service consumer), and the second network element may be an NWDAF network element including an MTLF, or an NWDAF network element including an AnLF, or an NWDAF network element including both an MTLF and an AnLF.

[0009] Optionally, the model acquisition request does not include the first indication information. The first network element determines, based on local configuration, to use transfer learning to generate a model that meets the requirements of the model acquisition request, and then returns the model generated using transfer learning to the second network element through a model acquisition response.

[0010] Optionally, the model acquisition request includes first indication information, and the first indication information is used to instruct the second network element to request the acquisition of the model generated by the first network element using the transfer learning method. In this case, the first network element can only generate the model using the transfer learning method. If the first network element cannot train the model using the transfer learning method, the model acquisition request of the second network element will be rejected, that is, the model acquisition response is used to inform the second network element that the model acquisition request has failed. For example, the first network element cannot train the model using the transfer learning method because the first network element finds that the similarity between the training data set information of all trained models and the training data set information in the model acquisition request of the second network element is lower than a threshold.

[0011] Optionally, the model acquisition request includes first indication information, and the first indication information is used to instruct the second network element to request to obtain the model generated by the first network element using the transfer learning method. Then, the first network element first considers using the transfer learning method to generate the model. If the first network element cannot train the model using the transfer learning method, it can retrain a model that meets the model acquisition request.

[0012] In an optional implementation, the model acquisition request also includes first training data set information, which is the training data set information required to be used by the model requested. That is, the first training data set information indicates the training data set information that the second network element expects to use. Optionally, the first training data set information includes a training data set and / or descriptive information of the training data set. Optionally, the first training data set information is used by the first network element to fine-tune the first model in the first network element, or to calculate the similarity with the training data set information used by the first model in the first network element. Optionally, if the model acquisition request does not include the first training data set information, the first network element may further request and obtain the first training data set information from the second network element.

[0013] In an optional embodiment, the model acquisition request also includes first threshold information, which is used to indicate the range of similarity that must be met between the training dataset information used by the first model in the first network element and the first training dataset information. The first model is a model in the first network element that can generate the model requested in the model acquisition request using transfer learning, or a model in the first network element that meets the transfer learning requirements that the second network element expects to obtain. In other words, only models whose similarity between the training dataset information in the first network element and the first training dataset information exceeds a threshold indicated by the first threshold information, or falls within a threshold range indicated by the first threshold information, are used to generate the model requested in the model acquisition request using transfer learning. Optionally, if the model acquisition request does not include the first indication information, the first network element may also determine whether to use transfer learning to generate the model requested in the model acquisition request based on local configuration. For example, a threshold may be locally configured, and the first network element uses transfer learning to generate the model requested in the model acquisition request for models whose similarity between the training dataset information and the first training dataset information exceeds the threshold.

[0014] In an optional embodiment, the model acquisition request also includes first similarity calculation method information, where the first similarity calculation method information is used to indicate a similarity calculation method used to calculate the similarity between the training dataset information used by the first model and the first training dataset information. Optionally, if the model acquisition request does not include the first similarity calculation method information, the first network element may determine the similarity calculation method to be used.

[0015] In an optional implementation, the model acquisition request also includes first transfer learning method information, where the first transfer learning method information is used to indicate the transfer learning method used by the first network element to generate the model requested by the second network element. In the case where the first indication information is used to indicate that the second network element requests to obtain a model generated by the first network element using a transfer learning method, the model acquisition request may include the first transfer learning method information, such as for indicating a sample-based transfer learning method or a model-based transfer learning method.

[0016] In one optional implementation, the model acquisition response includes result indication information and a first reason, where the result indication information indicates that the second network element's model acquisition request failed, and the first reason indicates the reason for the failure of the model acquisition request. Optionally, if the model acquisition response does not include the result indication information and the first reason, it indicates that the first network element can successfully respond to the second network element's model acquisition request, i.e., the first network element can send the model requested by the second network element to the second network element.

[0017] In another optional embodiment, the model acquisition response includes result indication information, where the result indication information is used to indicate whether the model acquisition request of the second network element is successful. Optionally, if the model acquisition response does not include result indication information, it indicates that the first network element can successfully respond to the model acquisition request of the second network element, that is, the first network element can send the model requested by the second network element to the second network element.

[0018] In another optional embodiment, the model acquisition response includes a first reason, where the first reason is used to indicate the reason why the model acquisition request failed. Optionally, if the model acquisition response does not include the first reason, it indicates that the first network element can successfully respond to the second network element's model acquisition request, that is, the first network element can send the model requested by the second network element to the second network element.

[0019] In an optional embodiment, the model acquisition response includes second indication information, and the second indication information is used to indicate that the model of the response feedback obtained by the first network element through the model is generated through transfer learning, or used to indicate that the model of the response feedback obtained by the first network element through the model is a model that meets the requirements of transfer learning.

[0020] In an optional embodiment, the model acquisition response further includes second similarity calculation value information, where the second similarity calculation value information is used to indicate a calculated result value of the similarity between the training dataset information used by the first model and the first training dataset information. The first model is a model in the first network element that can generate the model desired to be obtained in the model acquisition request using a transfer learning method, or a model in the first network element that meets the transfer learning requirements desired to be obtained by the second network element.

[0021] In an optional embodiment, the model acquisition response further includes second similarity calculation method information, where the second similarity calculation method information is used to indicate a similarity calculation method used to calculate the similarity between the training dataset information used by the first model and the first training dataset information.

[0022] In an optional implementation, the model acquisition response also includes second transfer learning method information, where the second transfer learning method information is used to indicate the transfer learning method used by the first network element to generate the model requested by the second network element.

[0023] In an optional embodiment, after the first network element receives a model acquisition request from the second network element, the method further includes: the first network element determines the first model based on the model acquisition request; the first network element sends a model retrieval request to the analysis data storage function network element, and the model retrieval request is used to request the analysis data storage function network element to retrieve the first model; the model retrieval request includes third indication information, and the third indication information is used to indicate the training data set information requested for the first model; the first network element receives a model retrieval response from the analysis data storage function network element, and the model retrieval response is used to feedback the retrieved first model and the training data set information used by the first model; wherein the first model is a model in the first network element that can use a transfer learning method to generate the model expected to be obtained by the model acquisition request, or a model in the first network element that meets the transfer learning requirements expected to be obtained by the second network element.

[0024] In an optional embodiment, before the first network element sends a model retrieval request to the analysis data storage function network element, the method also includes: the first network element sends a model or data storage request to the analysis data storage function network element, the model or data storage request includes the model identifier of the model requested to be stored, the model file address of the model and the training data set information used by the model; the first network element receives a model or data storage response from the analysis data storage function network element, the model or data storage response is used to indicate whether the information requested to be stored by the model or data storage request is successfully stored.

[0025] In an optional embodiment, the method also includes: the first network element sends a network element registration request to the network storage function network element, the network element registration request includes fourth indication information, and the fourth indication information is used to indicate whether the first network element supports the use of the transfer learning method to generate the model; the first network element receives a network element registration response from the network storage function network element, and the network element registration response is used to indicate whether the registration is successful.

[0026] In an optional implementation manner, the network element registration request further includes second training data set information, where the second training data set information is used to indicate training data set information corresponding to the model supported and provided by the first network element.

[0027] On the second aspect, the present application also provides a model acquisition method, which can be applied to a second network element, or a chip or chip module in the second network element, or to a module or unit that can realize all or part of the functions of the second network element, etc., and is described below using the second network element as an example. In this method, the second network element sends a model acquisition request to the first network element, and the model acquisition request is used to request to obtain the model of the first network element. The model acquisition request includes first indication information; wherein the first indication information is used to instruct the second network element to request to obtain the model generated by the first network element using the transfer learning method, or to instruct the second network element to request to obtain the model in the first network element that meets the transfer learning requirements; the second network element receives the model acquisition response from the first network element.

[0028] It can be seen that the first indication information is used to indicate the situation where the second network element requests to obtain the model generated by the first network element using the transfer learning method. The first network element can use the transfer learning method to generate the model requested by the second network element, thereby avoiding the problem of low model training efficiency caused by retraining the corresponding model. The first indication information is used to indicate the situation where the second network element requests to obtain the model that meets the transfer learning requirements in the first network element. The first network element can provide the second network element with a model that meets the transfer learning requirements, thereby facilitating the second network element to obtain the required model based on the model using the transfer learning method, thereby improving the model training efficiency.

[0029] In an optional implementation, the model acquisition request further includes first training data set information, where the first training data set information is training data set information that is expected to be used by the model requested for acquisition.

[0030] In an optional embodiment, the model acquisition request also includes first threshold information, and the first threshold information is used to indicate the range of similarity that must be met between the training data set information used by the first model in the first network element and the first training data set information; wherein, the first model is a model in the first network element that can use transfer learning to generate the model that the model acquisition request expects to obtain, or a model in the first network element that meets the transfer learning requirements that the second network element expects to obtain.

[0031] In an optional implementation, the model acquisition request further includes first similarity calculation method information, where the first similarity calculation method information is used to indicate a similarity calculation method used to calculate the similarity between the training dataset information used by the first model and the first training dataset information.

[0032] In an optional implementation, the model acquisition request also includes first transfer learning method information, and the first transfer learning method information is used to indicate the transfer learning method used by the first network element to generate the model requested by the second network element.

[0033] In one optional embodiment, the model acquisition response includes result indication information and / or a first reason, where the result indication information is used to indicate whether the model acquisition request for the second model training function is successful, and the first reason is used to indicate the reason why the model acquisition request failed. Optionally, this embodiment can also refer to the relevant content described in the first aspect above and will not be described in detail here.

[0034] In an optional embodiment, the model acquisition response includes second indication information, and the second indication information is used to indicate that the model of the response feedback obtained by the first network element through the model is generated through transfer learning, or used to indicate that the model of the response feedback obtained by the first network element through the model is a model that meets the requirements of transfer learning.

[0035] In an optional implementation, the model acquisition response further includes second similarity calculation value information, where the second similarity calculation value information is used to indicate a calculation result value of the similarity between the training dataset information used by the first model and the first training dataset information.

[0036] In an optional embodiment, the model acquisition response further includes second similarity calculation method information, where the second similarity calculation method information is used to indicate a similarity calculation method used to calculate the similarity between the training dataset information used by the first model and the first training dataset information.

[0037] In an optional implementation, the model acquisition response also includes second transfer learning method information, where the second transfer learning method information is used to indicate the transfer learning method used by the first network element to generate the model requested by the second network element.

[0038] In an optional embodiment, before the second network element sends a model acquisition request to the first network element, the method also includes: the second network element sends a network element discovery request to the network storage function network element, the network element discovery request includes fifth indication information, and the fifth indication information is used to indicate a request to discover a model training function network element with transfer learning capabilities; the second network element receives a network element discovery response from the network storage function network element.

[0039] In an optional implementation, the network element discovery request further includes third training data set information, where the third training data set information is training data set information that is expected to be used by the model supported by the model training function network element requested for discovery.

[0040] In an optional embodiment, the network element discovery request also includes third threshold information, and the third threshold information is used to indicate the range of similarity that must be met between the training data set information used by the model in the model training function network element and the third training data set information; the model training function network element is the model training function network element that the network element discovery request in the network storage function network element expects to obtain.

[0041] In an optional embodiment, the network element discovery request also includes third similarity calculation method information, and the third similarity calculation method information is used to indicate the similarity calculation method used to calculate the similarity between the training data set information used by the model in the model training function network element and the third training data set information.

[0042] In an optional embodiment, the network element discovery response includes the identifiers of one or more model training function network elements. For each model training function network element identified by the identifier, the network element discovery response may also include one or more of the following information: model identifier, the value of the similarity between the training data set information used by the model and the third training data set information, and the similarity calculation method used to calculate the similarity.

[0043] On the third aspect, the present application also provides a network element discovery method, which can be applied to a network storage function network element, or a chip or chip module in a network storage function network element, or to a module or unit that can realize all or part of the functions of the network storage function network element, etc. The following description is taken as an example of a network storage function network element. In the network element discovery method, the network storage function network element receives a network element discovery request from a second network element, wherein the network element discovery request includes fifth indication information, and the fifth indication information is used to indicate a request to discover a model training function network element with transfer learning capability; the network storage function network element discovers a model training function network element with transfer learning capability, such as the first network element, based on the network element discovery request. The network storage function network element sends a network element discovery response to the second network element, and the network element discovery response includes the identification and / or address information of one or more model training function network elements.

[0044] It can be seen that this method enables consumer network elements, such as the second network element being the consumer NWDAF, to discover model training function network elements that have or support transfer learning capabilities through the network element storage function network element, which is conducive to requesting the model obtained through transfer learning from the discovered model training function network element, thereby helping to improve model training efficiency.

[0045] Alternatively, the fifth indication information is used to indicate a request to discover a model training function network element having a model that meets the transfer learning requirements. In this way, the consumer NWDAF can discover models that meet the transfer learning requirements through the network element storage function network element, which is conducive to utilizing these models and fine-tuning them through transfer learning to obtain the required models, thereby helping to improve model training efficiency.

[0046] In an optional implementation, the network element discovery request further includes third training data set information, where the third training data set information is training data set information that is expected to be used by the model supported by the model training function network element requested for discovery.

[0047] In an optional embodiment, the network element discovery request further includes third threshold information, which is used to indicate the range of similarity that must be satisfied between the training dataset information used by the model in the model training function network element and the third training dataset information. The model training function network element is the model training function network element that the network element discovery request expects to obtain in the network storage function network element. This embodiment is advantageous for the network storage function network element to return model training function network elements with a high degree of similarity between the model training dataset information and the third training dataset information.

[0048] In an optional embodiment, the network element discovery request also includes third similarity calculation method information, and the third similarity calculation method information is used to indicate the similarity calculation method used to calculate the similarity between the training data set information used by the model in the model training function network element and the third training data set information.

[0049] In an optional embodiment, for each model training function network element identified by an identifier and / or address, the network element discovery response further includes one or more of the following information: a model identifier, a similarity value between the training dataset information used by the model and the third training dataset information, and a similarity calculation method for calculating the similarity. This embodiment facilitates the second network element selecting a model training function network element as a first network element based on the network element discovery response, and sending a model acquisition request to the first network element.

[0050] Optionally, the network element discovery method may further include: the network storage function network element receiving a network element registration request from the first network element, wherein the network element registration request includes fourth indication information, and the fourth indication information is used to indicate whether the first network element supports using the transfer learning method to generate a model. Furthermore, the network storage function network element sends a network element registration response to the first network element, wherein the network element registration response is used to indicate whether the registration is successful. Optionally, the network element registration response includes result indication information, and the result indication information is used to indicate whether the first network element is successfully registered.

[0051] In an optional implementation manner, the network element registration request further includes second training data set information, where the second training data set information is training data set information of a model supported and provided by the first network element.

[0052] Fourthly, the present application also provides a model retrieval method, which can be applied to an analysis data storage functional network element, or a chip or chip module in the analysis data storage functional network element, or to a module or unit that can realize all or part of the functions of the analysis data storage functional network element, etc. The following description is taken as an example of the analysis data storage functional network element. In the model retrieval method, the analysis data storage functional network element receives a model retrieval request from a first network element, and the model retrieval request is used to request the analysis data storage functional network element to retrieve a first model; the first model is a model in the first network element that can generate a model using a transfer learning method, or a model in the first network element that meets the transfer learning requirements; the analysis data storage functional network element sends a model retrieval response to the first network element, and the model retrieval response is used to feedback the first model.

[0053] Optionally, the model retrieval request also includes third indication information, and the third indication information is used to indicate the training data set information used by the first model; in this way, the model retrieval response also includes the training data set information used by the first model.

[0054] It can be seen that this method enables the first network element to obtain the required first model and the training data set information used by the first model by analyzing the storage function network element after deleting the local model and training data set information, thereby avoiding the situation where the first network element cannot distinguish which data is the training data associated with the model after deleting the local model and training data set information, resulting in the inability to determine the source domain of the model (that is, the training data set information used by the model in the first network element), and the inability to calculate the similarity between the source domain and the destination domain of the model (that is, the requested training data set information, such as the first training data set information, etc.) to determine whether the model meets the transfer learning requirements.

[0055] Optionally, the method also includes: analyzing the data storage function network element receiving a model or data storage request from the first network element, the model or data storage request including a model identifier of the model requested to be stored, a model file address of the model, and training data set information used by the model; analyzing the data storage function network element sending a model or data storage response to the first network element, the model or data storage response being used to indicate whether the information requested to be stored by the model or data storage request is successfully stored.

[0056] In a fifth aspect, the present application also provides a model acquisition method, which is explained from the perspective of the interaction between the first network element and the second network element. The model acquisition method includes: the second network element sends a model acquisition request to the first network element, and the first network element receives the model acquisition request accordingly, wherein the model acquisition request is used to request the acquisition of the model of the first network element, and the model acquisition request includes first indication information; wherein the first indication information is used to instruct the second network element to request the acquisition of the model generated by the first network element using the transfer learning method, or to instruct the second network element to request the acquisition of the model in the first network element that meets the transfer learning requirements; the first network element sends a model acquisition response to the second network element.

[0057] It can be seen that the first indication information is used to indicate the situation where the second network element requests to obtain the model generated by the first network element using the transfer learning method. The first network element can use the transfer learning method to generate the model requested by the second network element, thereby avoiding the problem of low model training efficiency caused by retraining the corresponding model. The first indication information is used to indicate the situation where the second network element requests to obtain the model that meets the transfer learning requirements in the first network element. The first network element can provide the second network element with a model that meets the transfer learning requirements, thereby facilitating the second network element to obtain the required model based on the model using the transfer learning method, thereby improving the model training efficiency.

[0058] The following describes optional implementations of parameters that may be included in a model acquisition request and a model acquisition response.

[0059] In an optional implementation, the model acquisition request includes first training data set information, where the first training data set information is training data set information that the requested model expects to use.

[0060] In an optional embodiment, the model acquisition request also includes first threshold information, and the first threshold information is used to indicate the range of similarity between the training data set information used by the first model in the first network element and the first training data set information; wherein, the first model is a model in the first network element that can use transfer learning to generate the model that the model acquisition request expects to obtain, or a model in the first network element that meets the transfer learning requirements that the second network element expects to obtain.

[0061] In an optional implementation, the model acquisition request further includes first similarity calculation method information, where the first similarity calculation method information is used to indicate a similarity calculation method used to calculate the similarity between the training dataset information used by the first model and the first training dataset information.

[0062] In an optional implementation, the model acquisition request also includes first transfer learning method information, and the first transfer learning method information is used to indicate the transfer learning method used by the first network element to generate the model requested by the second network element.

[0063] In an optional embodiment, the model acquisition response includes result indication information and a first reason, the result indication information is used to indicate whether the model acquisition request of the second model training function is successful, and the first reason is used to indicate the reason why the model acquisition request failed.

[0064] In an optional embodiment, the model acquisition response includes second indication information, and the second indication information is used to indicate that the model of the response feedback obtained by the first network element through the model is generated through transfer learning, or used to indicate that the model of the response feedback obtained by the first network element through the model is a model that meets the requirements of transfer learning.

[0065] In an optional implementation, the model acquisition response further includes second similarity calculation value information, where the second similarity calculation value information is used to indicate a calculation result value of the similarity between the training dataset information used by the first model and the first training dataset information.

[0066] In an optional embodiment, the model acquisition response further includes second similarity calculation method information, where the second similarity calculation method information is used to indicate a similarity calculation method used to calculate the similarity between the training dataset information used by the first model and the first training dataset information.

[0067] In an optional implementation, the model acquisition response also includes second transfer learning method information, where the second transfer learning method information is used to indicate the transfer learning method used by the first network element to generate the model requested by the second network element.

[0068] Optionally, after receiving the model acquisition request, the first network element determines a first model based on the model acquisition request. The first model is a model in the first network element that can generate the model desired by the model acquisition request using transfer learning, or a model in the first network element that meets the transfer learning requirements desired by the second network element. If the first network element stores the first model in the analysis data storage function network element and deletes the local model, the first network element may send a model retrieval request to the analysis data storage function network element. Accordingly, the analysis data storage function network element receives the model retrieval request from the first network element. The model retrieval request is used to request the data storage function network element to retrieve the first model. The model retrieval request includes third indication information, which is used to indicate the request for training dataset information used by the first model. The analysis data storage function network element sends a model retrieval response to the first network element. Accordingly, the first network element receives the model retrieval response. The model retrieval response is used to feedback the retrieved first model and the training dataset information used by the first model. Accordingly, the first network element may send the first model and the training dataset information used by the first model to the second network element via the model acquisition response.

[0069] Optionally, before the first network element sends a model retrieval request to the analysis data storage function network element, the method further includes: the first network element sends a model or data storage request to the analysis data storage function network element, and the analysis data storage function network element receives the model or data storage request accordingly. The model or data storage request includes the model identifier of the model requested to be stored, the model file address of the model, and the training data set information used by the model. The analysis data storage function network element sends a model or data storage response to the first network element, and the first network element receives the model or data storage response accordingly. The model or data storage response is used to indicate whether the information requested to be stored by the model or data storage request is successfully stored.

[0070] Optionally, before the second network element sends a model acquisition request to the first network element, the method further includes: the second network element sending a network element discovery request to the network storage function network element, and the network storage function network element accordingly receives the network element discovery request. The network element discovery request includes fifth indication information, which indicates a request to discover a model training function network element with transfer learning capabilities. Based on the network element discovery request, the network storage function network element discovers a model training function network element with transfer learning capabilities, such as the first network element. The network storage function network element sends a network element discovery response to the second network element, and the second network element accordingly receives the network element discovery response from the network storage function network element. The network element discovery response is returned in response to the network element discovery request. Optionally, the network element discovery response may include identifiers of one or more model training function network elements. This embodiment facilitates the second network element selecting a first network element from one or more model training function network elements and then sending a model acquisition request to the first network element.

[0071] In an optional implementation, the network element discovery request further includes third training data set information, where the third training data set information is training data set information that is expected to be used by the model supported by the model training function network element requested for discovery.

[0072] In an optional embodiment, the network element discovery request further includes third threshold information, which is used to indicate the range of similarity that must be satisfied between the training dataset information used by the model in the model training function network element and the third training dataset information. The model training function network element is the model training function network element that the network element discovery request expects to obtain in the network storage function network element. This embodiment is advantageous for the network storage function network element to return model training function network elements with a high degree of similarity between the model training dataset information and the third training dataset information.

[0073] In an optional embodiment, the network element discovery request also includes third similarity calculation method information, and the third similarity calculation method information is used to indicate the similarity calculation method used to calculate the similarity between the training data set information used by the model in the model training function network element and the third training data set information.

[0074] In an optional embodiment, the network element discovery response includes the identifiers of one or more model training function network elements. For each model training function network element identified by the identifier, the network element discovery response may also include one or more of the following information: a model identifier, a similarity value between the training data set information used by the model and the third training data set information, and a similarity calculation method for calculating the similarity. This embodiment facilitates the second network element to select a model training function network element as the first network element based on the network element discovery response, and to send a model acquisition request to the first network element.

[0075] Optionally, before the second network element sends a network element discovery request to the network storage function network element, the first network element sends a network element registration request to the network storage function network element, and accordingly, the network storage function network element receives the network element registration request. The network element registration request includes fourth indication information, and the fourth indication information is used to indicate whether the first network element supports using a transfer learning method to generate a model. Furthermore, the network storage function network element sends a network element registration response to the first network element, and accordingly, the first network element receives a network element registration response from the network storage function network element. The network element registration response is used to indicate whether the registration is successful. Optionally, the network element registration response includes result indication information, and the result indication information is used to indicate whether the first network element is successfully registered.

[0076] In an optional implementation manner, the network element registration request further includes second training data set information, where the second training data set information is training data set information of a model supported and provided by the first network element.

[0077] In a sixth aspect, embodiments of the present application further provide a communication device. The communication device is a first network element, or a device of the first network element, or a device capable of being used in conjunction with the first network element. In one possible implementation, the communication device includes a functional module, which is implemented as a hardware circuit, or software, or a combination of a hardware circuit and software.

[0078] In one possible embodiment, the communication device includes one or more functional units, such as a communication unit, wherein the communication unit is used to receive a model acquisition request from a second network element, the model acquisition request is used to request to obtain the model of the first network element, and the model acquisition request includes first indication information; wherein the first indication information is used to instruct the second network element to request to obtain the model generated by the first network element using a transfer learning method, or to instruct the second network element to request to obtain the model in the first network element that meets the transfer learning requirements; the communication unit is also used to send a model acquisition response to the second network element.

[0079] Optionally, the communication device may also include a processing unit, which is used to generate the model requested by the model acquisition request using a transfer learning method, or determine a model that meets the transfer learning requirements, and then the communication unit sends the generated or determined model to the second network element through a model acquisition response.

[0080] Optionally, in this implementation, the model acquisition request, model acquisition response and other related contents and possible implementation methods of the communication device can be found in the relevant description of the first aspect and will not be described in detail here.

[0081] In a seventh aspect, embodiments of the present application further provide a communication device. The communication device is a second network element, or a device of the second network element, or a device capable of being used in conjunction with the second network element. In one possible implementation, the communication device includes a functional module, which is implemented as a hardware circuit, or software, or a combination of a hardware circuit and software.

[0082] In one possible embodiment, the communication device includes one or more functional units, such as a communication unit, wherein the communication unit is used to send a model acquisition request to the first network element, the model acquisition request is used to request to obtain the model of the first network element, and the model acquisition request includes first indication information; wherein the first indication information is used to instruct the second network element to request to obtain the model generated by the first network element using the transfer learning method, or to instruct the second network element to request to obtain the model in the first network element that meets the transfer learning requirements; the communication unit is also used to receive a model acquisition response from the first network element.

[0083] Optionally, the communication device further includes a processing unit configured to determine a first network element so that the communication unit can send a model acquisition request to the first network element. Optionally, the processing unit can also generate a desired model using transfer learning based on the model returned in the model acquisition response.

[0084] Optionally, in this implementation, the relevant contents such as the model acquisition request and the model acquisition response and possible implementation methods of the communication device can be found in the relevant description of the second aspect and will not be described in detail here.

[0085] In an eighth aspect, embodiments of the present application further provide a communication device. The communication device is a network storage function network element, or a device of a network storage function network element, or a device capable of being used in conjunction with a network storage function network element. In one possible implementation, the communication device includes a functional module, which is implemented as a hardware circuit, or software, or a combination of a hardware circuit and software.

[0086] In one possible embodiment, the communication device includes one or more functional units, such as a communication unit, wherein the communication unit receives a network element discovery request from a second network element, wherein the network element discovery request includes fifth indication information, and the fifth indication information is used to indicate a request to discover a model training function network element with transfer learning capability; the communication unit is also used to send a network element discovery response to the second network element, and the network element discovery response includes the identifiers of one or more model training function network elements.

[0087] Optionally, the communication device further includes a processing unit, which is used to discover a model training functional network element with transfer learning capability based on the network element discovery request.

[0088] Optionally, possible implementations of the communication device can be found in the relevant description of the third aspect and will not be described in detail here.

[0089] Ninthly, embodiments of the present application further provide a communication device. The communication device is an analysis data storage functional network element, or a device for an analysis data storage functional network element, or a device capable of being used in conjunction with an analysis data storage functional network element. In one possible implementation, the communication device includes a functional module, which is implemented as a hardware circuit, or software, or a combination of a hardware circuit and software.

[0090] In one possible embodiment, the communication device includes one or more functional units, such as a communication unit, wherein the communication unit is used to receive a model retrieval request from the first network element, and the model retrieval request is used to request the analysis data storage function network element to retrieve the first model; the first model is a model in the first network element that can be generated using a transfer learning method, or a model in the first network element that meets the transfer learning requirements; the communication unit is also used to send a model retrieval response to the first network element, and the model retrieval response is used to feedback the first model.

[0091] Optionally, the communication device further includes a processing unit, configured to retrieve the first model based on the model retrieval request.

[0092] Optionally, possible implementations of the communication device can be found in the relevant description of the fourth aspect and will not be described in detail here.

[0093] For the sixth to ninth aspects, as an example, the processing unit can also be embodied as a processing circuit or a logic circuit; the transceiver unit can be an input / output interface, interface circuit, output circuit, input circuit, pin or related circuit on the chip or chip system.

[0094] During implementation, the processor can be used to perform, for example, but not limited to, baseband-related processing, and the transceiver or communication interface can be used to perform, for example, but not limited to, radio frequency transceiver. The above-mentioned devices can be provided on separate chips, or at least partially or entirely on the same chip. For example, the processor can be further divided into an analog baseband processor and a digital baseband processor. The analog baseband processor can be integrated with the transceiver (or communication interface) on the same chip, while the digital baseband processor can be provided on a separate chip. With the continuous development of integrated circuit technology, more and more devices can be integrated on the same chip. For example, a digital baseband processor can be integrated with multiple application processors (such as, but not limited to, a graphics processor, a multimedia processor, etc.) on the same chip. Such a chip can be called a system on a chip (SoC). Whether each device is provided independently on different chips or integrated on one or more chips often depends on the needs of the product design. The embodiments of the present application do not limit the implementation form of the above-mentioned devices.

[0095] In the tenth aspect, the embodiment of the present application further provides a processor for executing the method described in any one of the first to fourth aspects above, or the method described in any one of the possible implementation methods of any one of the first to fourth aspects above. In the process of executing these methods, the process of sending the above-mentioned signal and receiving the above-mentioned signal can be understood as the process of outputting the above-mentioned signal by the processor, and the process of the above-mentioned signal input by the processor. When outputting the above-mentioned signal, the processor outputs the above-mentioned signal to the transceiver so that it is transmitted by the transceiver (or communication interface). After being output by the processor, the above-mentioned signal may also need to be processed otherwise before it reaches the transceiver (or communication interface). Similarly, when the processor receives the above-mentioned signal input, the transceiver (or communication interface) receives the above-mentioned signal and inputs it into the processor. Furthermore, after the transceiver (or communication interface) receives the above-mentioned signal, the above-mentioned signal may need to be processed otherwise before it is input into the processor.

[0096] For the sending and receiving operations involved in the processor, unless otherwise specified, or unless they conflict with their actual functions or internal logic in the relevant descriptions, they can be more generally understood as processor output, reception, input and other operations, rather than sending and receiving operations directly performed by the RF circuit and antenna.

[0097] During implementation, the processor may be a processor specifically configured to execute these methods, or may be a processor that executes computer instructions in a memory to execute these methods, such as a general-purpose processor. The memory may be a non-transitory memory, such as a read-only memory (ROM), which may be integrated with the processor on the same chip or disposed on separate chips. The embodiments of the present application do not limit the type of memory or the configuration of the memory and the processor.

[0098] In an eleventh aspect, embodiments of the present application further provide a communication device, comprising: a processor configured to invoke a computer program stored in a memory, and, through a transceiver, enable the communication device to implement the method described in any one of the first to fourth aspects or any possible implementation manner. Optionally, the communication device further includes a memory, and the processor and the memory are coupled.

[0099] In a twelfth aspect, the present application further provides a communication system, which includes at least one first network element that performs the method of the first aspect or any optional implementation method of the first aspect and at least one second network element that performs the method of the second aspect or any possible implementation method of the second aspect; and / or, the system includes a network storage function network element that performs the third aspect or any implementation method of the third aspect, and an analysis data storage function network element that performs any implementation method of the fourth aspect. In another possible design, the system may also include other devices that interact with the first network element and / or the second network element in the solution provided in the embodiment of the present application.

[0100] In the thirteenth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed, the computer device implements the method of any aspect or any possible implementation method of the above-mentioned first to fourth aspects through a transceiver and a processor.

[0101] In the fourteenth aspect, the present application also provides a computer program product comprising instructions, the computer program product comprising: computer program code, which, when the computer program code is run in parallel, enables a computer device to implement any one of the above-mentioned first to fourth aspects or any possible implementation method through a transceiver and a processor.

[0102] In a fifteenth aspect, the present application provides a chip system, which includes a processor and an interface, the interface is used to obtain a program or instruction, the processor is used to call the program or instruction, and the interface is used to implement the method of any aspect or any possible implementation method of the above-mentioned first to fourth aspects. In one possible design, the chip system also includes a memory, which is used to store program instructions and data necessary for the terminal. The chip system can be composed of a chip, or it can include a chip and other discrete devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] FIG1 is a schematic diagram of a network architecture of a fifth generation (5G) communication system based on a service-oriented interface;

[0104] FIG2 is a schematic diagram of a traditional machine learning method and a transfer learning method;

[0105] FIG3 is a flow chart of a model acquisition method provided in an embodiment of the present application;

[0106] FIG4 is a flow chart of another model acquisition method provided in an embodiment of the present application;

[0107] FIG5 is a flow chart of a network element registration and discovery method provided in an embodiment of the present application;

[0108] FIG6 is a flow chart of a model / data storage and retrieval method provided in an embodiment of the present application;

[0109] FIG7 is a schematic structural diagram of a communication device provided in an embodiment of the present application;

[0110] FIG8 is a schematic structural diagram of another communication device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0111] This application is based on the idea of ​​generating a model using a transfer learning method, and provides a model acquisition method and related devices, which are conducive to improving the efficiency of model training. Optionally, this application can be applied to a network architecture diagram of a fifth generation (5G) communication system based on a service-oriented interface as shown in Figure 1. As shown in Figure 1, the 5G communication system is divided into an access network and a core network, wherein the access network implements wireless access-related functions through a radio access network (RAN) device. The network functions are based on modular disassembly, and the decoupled network functions (NFs) can be independently expanded, independently evolved, and deployed on demand. Service-oriented interfaces are used between all NFs in the control plane. The same service can be called by multiple NFs, reducing the coupling degree of the interface definition between NFs, and ultimately achieving on-demand customization of the entire network function, flexibly supporting different business scenarios and requirements. In the architecture shown in Figure 1, the network elements in the dotted box are service-oriented NF network elements, the interfaces between NF network elements are service-oriented interfaces, and the interactive messages are service-oriented messages. The architecture may include an access network and a core network, and optionally, may also include user equipment (UE).

[0112] A UE is a device with wireless transceiver capabilities that can be deployed on land, indoors or outdoors, as a handheld, wearable, or vehicle-mounted device; on water (such as ships); or in the air (for example, on airplanes, balloons, and satellites). A UE can be a mobile phone, tablet, computer with wireless transceiver capabilities, virtual reality (VR) terminal device, augmented reality (AR) terminal device, wireless terminal for industrial control, vehicle-mounted terminal device, wireless terminal for self-driving, wireless terminal for remote medical care, wireless terminal for smart grids, wireless terminal for transportation safety, wireless terminal for smart cities, wireless terminal for smart homes, wearable terminal device, and so on. A UE is sometimes also referred to as a terminal, terminal device, access terminal device, vehicle-mounted terminal, industrial control terminal, UE unit, UE station, mobile station, mobile station, remote station, remote terminal device, mobile device, UE agent, or UE device. The UE may also be fixed or mobile.

[0113] The access network implements access-related functions, providing network access to authorized users in a specific area and enabling the use of different quality transmission tunnels for user data based on user level and service requirements. The access network forwards control signals and user data between terminal devices and the core network. The access network can include access network equipment. For example, in this application, information exchanged between tag devices and core network elements can be forwarded via access network equipment.

[0114] Access network equipment provides access to terminal devices and can include both RAN and AN devices. (R)AN devices are primarily responsible for air interface functions such as radio resource management, quality of service (QoS) management, data compression, and encryption. RAN devices can include various base stations, such as macro base stations, micro base stations (also known as small base stations), relay stations, access points, and balloon stations. In systems using different radio access technologies, the names of devices with base station functions may vary. For example, in fifth-generation (5G), sixth-generation (6G), and even seventh-generation (7G) systems, network devices may be called: RAN or next-generation Node basestation (gNB), evolved NodeB (eNB or eNodeB), base station controller (BSC), base transceiver station (BTS), home network device (for example, home evolved Node B or home Node B, HNB), baseband unit (BBU), access point (AP) in wireless fidelity (WIFI) systems, wireless relay node, wireless backhaul node, transmission and reception point (TRP), transmission point (TP) point, TP), etc.; or one or a group of (including multiple antenna panels) antenna panels of a network device in a 5G system, or it can also be a network node constituting a gNB or transmission point, such as a baseband unit (BBU), or a distributed unit (DU), or a road side unit (RSU) in a vehicle to everything (V2X) or intelligent driving scenario.

[0115] In some deployments, a gNB or transmission point may include a centralized unit (CU) and a DU. A gNB or transmission point may also include a radio unit (RU). The CU implements some of the gNB or transmission point's functions, while the DU implements some of the gNB or transmission point's functions. For example, the CU implements radio resource control (RRC) and packet data convergence protocol (PDCP) layer functions, while the DU implements radio link control (RLC), media access control (MAC), and physical (PHY) layer functions. Because RRC layer information ultimately becomes physical layer information, or is converted from physical layer information, in this architecture, higher-layer signaling, such as RRC layer signaling or PDCP layer signaling, can also be considered to be sent by the DU, or by both the DU and the RU. It is understood that a network device can be a CU node, a DU node, or a device that includes both a CU node and a DU node. Optionally, the network device can also be an auxiliary communication device, such as a satellite.

[0116] The core network is responsible for maintaining mobile network subscription data and providing UE with session management, mobility management, policy management, and security authentication. The core network may include the following network elements: network exposure function (NEF), network repository function (NRF), network data analytics function (NWDAF), analytics data repository function (ADRF), application function (AF), and policy control function (PCF).

[0117] NEF network elements are mainly used to support the opening of capabilities and events. NRF network elements mainly provide service registration, discovery and authorization, and maintain available network function (NF) instance information, which can realize on-demand configuration of network functions and services and interconnection between NFs. Among them, service registration means that NF network elements need to register with NRF network elements before they can provide services. Service discovery means that when NF network elements need other NF network elements to provide services for them, they must first perform service discovery through NRF network elements to discover the desired NF network elements that provide services for them. For example, when NF network element 1 needs NF network element 2 to provide services for it, it must first perform service discovery through NRF network elements to discover NF network element 2.

[0118] NWDAF network elements have functions such as data collection, model training, data analysis, and model reasoning. They can be used to collect relevant data from network elements, third-party service servers, terminal devices, or network management systems, perform data analysis or model training based on the relevant data, and provide data analysis results to network elements, third-party service servers, terminal devices, or network management systems, or provide trained models to other data analysis network elements. NWDAF network elements can be divided into analytics logical functions (AnLF) and model training logical functions (MTLF) based on their functions. Among them, AnLF is the logical reasoning function in NWDAF, which is used to perform model reasoning, derive analysis results (i.e., derive statistical or predictive analysis results based on the analysis consumer's request), and make analysis results available. MTLF is the model training function in NWDAF, which is used to train models and make training services available (for example, providing trained models). An NWDAF network element may contain only AnLF, only MTLF, or both AnLF and MTLF.

[0119] The AF network element primarily supports interaction with the 3GPP core network to provide services, such as influencing data routing decisions, policy control functions, or providing third-party services to the network side. The PCF network element primarily supports providing a unified policy framework to control network behavior, providing policy rules to the control layer network functions, and is responsible for obtaining user subscription information related to policy decisions. The PCF network element can provide policies to the AMF network element and SMF network element, such as quality of service (QoS) policies and slice selection policies.

[0120] ADRF network element is used to store data collected by NWDAF, analysis results generated by NWDAF, and models trained by NWDAF. SMF network element is mainly responsible for session management in mobile networks, such as session establishment, modification, and release. Specific functions include allocating Internet Protocol (IP) addresses to users, selecting UPF network elements that provide message forwarding functions, etc. AMF network element is mainly responsible for mobility management in mobile networks, such as user location updates, user registration networks, user switching, etc. UPF network element is mainly responsible for forwarding and receiving user data. It can receive user data from the data network and transmit it to the UE through the access network device; it can also receive user data from the UE through the access network device and forward it to the data network. Operations, administration and management (OAM) are mainly used to monitor and manage the status and performance of network equipment.

[0121] In addition, in the communication system network architecture shown in Figure 1, the AN and AMF can interact via the N2 interface, the N3 interface supports selective activation / deactivation of user plane connections, and the SMF and UPF network elements interact via the N4 interface. All NFs in the control plane can interact using service-based interfaces. For example, the NEF network element can interact with other network function network elements via the service-based interface Nnef. The NRF network element can interact with other network function network elements via the service-based interface Nnrf. The NWDAF network element can interact with other network function network elements via the service-based interface Nnwdaf. The AF network element can interact with other network function network elements via the service-based interface Naf. The PCF network element can interact with other network function network elements via the service-based interface Npcf. The ADRF network element can interact with other network function network elements via the service-based interface Nadrf. The AMF network element can interact with other network function network elements via the service-based interface Namf. The SMF network element can interact with other network function network elements via the service-based interface Nsmf.

[0122] It should be noted that the above only lists 5G communication systems to which the communication method provided in the embodiment of the present application is applicable. The communication method provided in the embodiment of the present application can also be applied to future communication systems such as the sixth generation (6G) or even the seventh generation (7G) system. The network architecture and service scenarios described in the embodiment of the present application are intended to more clearly illustrate the technical solutions of the embodiment of the present application and do not constitute a limitation on the technical solutions provided in the embodiment of the present application. It is known to those skilled in the art that with the evolution of the communication network architecture and the emergence of new service scenarios, the technical solutions provided in the embodiment of the present application are also applicable to similar technical problems.

[0123] In order to facilitate the clear description of the technical solutions of the embodiments of the present application, in the embodiments of the present application, words such as "first", "second", "third", "fourth", "fifth", "sixth", "seventh", and "eighth" are used to distinguish between identical or similar items with basically the same functions and effects. Those skilled in the art will understand that words such as "first" and "second" have no specific technical meanings, do not limit the quantity and execution order, and words such as "first" and "second" do not necessarily limit them to be different. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent three situations: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship.

[0124] The intelligent network architecture based on NWDAF aims to collect massive amounts of information from the network and utilize existing big data and artificial intelligence technologies to leverage this information, outputting valuable information to assist operators in policy formulation and network resource adjustment, thereby improving user experience and reducing network load. The NWDAF needs to collect data to provide corresponding analysis results. Table 1 shows an example of the analysis results provided by the NWDAF and the data that needs to be collected. As shown in Table 1, to provide service experience analysis results, the NWDAF needs to collect service-related information such as service identification and service experience from the application function (AF) / user equipment (UE), and collect information such as signal reception power and signal reception quality from the OAM. The NWDAF trains an artificial intelligence (AI) model based on the collected information. Then, based on the AI ​​model, it obtains inference analysis results, such as a predicted service experience analysis for a certain time period in the future. As shown in Table 1, in order to provide network element load analysis results, NWDAF needs to collect relevant information such as network element resource usage from NRF, and collect UE speed or direction, such as UE minimization of drive tests (MDT) data, from OAM. NWDAF trains an artificial intelligence (AI) model based on the collected information, and then obtains inference analysis results based on the AI ​​model, for example, a predicted network element load analysis for a certain time period in the future. As shown in Table 1, in order to provide mobile edge computing (MEC) service experience analysis results, NWDAF needs to collect relevant information such as UE identification, UE location, application identification, application location, uplink or downlink transmission performance data from AF or UE. NWDAF trains an artificial intelligence (AI) model based on the collected information, and then obtains inference analysis results based on the AI ​​model, for example, a predicted MEC service experience analysis for a certain time period in the future.

[0125] Table 1. Analysis results provided by NWDAF and examples of data to be collected

[0126] In addition, NWDAFs can request models from each other. For example, when a consumer NWDAF requests a model from a producer NWDAF, the producer NWDAF may not have the model that the consumer NWDAF wants. In this case, the producer NWDAF needs to retrain a model according to the requirements of the consumer NWDAF, resulting in low model training efficiency.

[0127] This application is based on the idea of ​​generating a model through transfer learning, and provides a model acquisition method and related devices, which are conducive to improving the efficiency of model training. To facilitate understanding of this application, some concepts involved in this application are briefly explained.

[0128] 1. Definition of Transfer Learning Method

[0129] Transfer learning is a model generation method that is different from traditional machine learning methods. In the basic concept of transfer learning, the domain is the subject of learning, consisting of data and the probability distribution that generates this data. For example, a dataset consisting of pictures of cats and tigers can be called a domain. The task is the goal of learning. A task usually refers to building a prediction function f(·) that generates labels through learning. For example, classifying pictures of cats and tigers can be called a task. Let D S is the source domain, T S The source domain D S A learning task on T is the target domain, T T The target domain D T Transfer learning is a learning task on D S and T S The knowledge on T Task T in T of learning, where D S ≠D T or T S ≠T T .

[0130] Figure 2 is a schematic diagram of traditional machine learning methods and transfer learning methods. The left side of Figure 2 shows the traditional machine learning process. Traditional machine learning can only train and use different machine learning models for different domains / tasks (respectively, learning systems are obtained as shown in Figure 2). However, if transfer learning technology is used, as shown in the right side of Figure 2, the source task in the source domain can be fine-tuned through knowledge to obtain a model (such as a learning system) and applied to the target task in the target domain, greatly improving model training efficiency. It can also generate a high-performance model even when the target domain is smaller. For example, Task A (source task) is to classify cats and tigers in pictures, and Task B (target task) is to distinguish the length of cats and tigers in pictures. Traditional machine learning methods can only train two different models for these two completely different tasks. However, using transfer learning, the model of Task A can be fully utilized and fine-tuned based on the model of Task A to obtain a model suitable for Task B.

[0131] 2. Classification of transfer learning methods

[0132] Transfer learning can be divided into the following four categories according to the learning method:

[0133] (1) Sample-based transfer learning: Find data in the source domain that is similar to the target domain, adjust the weights of this data so that the weighted data matches the data in the target domain, and then train and learn to obtain a model suitable for the target domain. This method is simple and easy to implement, but the choice of weights and the measurement of similarity rely on experience, and the data distribution of the source and target domains is often different.

[0134] (2) Feature-based transfer learning: When the source and target domains share some common cross-features, feature transformation can be used to transform the features of the source and target domains into the same space, so that the data in the source and target domains have similar data distributions in this space. Traditional machine learning can then be performed. This approach is suitable for most situations, but it requires screening for good common features, which is difficult to find and prone to overfitting.

[0135] (3) Model-based transfer learning: Some model parameters are shared between the source and target domains. The model trained with a large amount of data in the source domain is transferred to the target domain for prediction. This method can fully utilize the similarities between models, but the model parameters are not easy to converge.

[0136] (4) Relationship-based transfer learning: When two domains are similar, they share a certain similarity relationship. The logical network relationship learned in the source domain is applied to the target domain for migration, for example, migrating the propagation rules of biological viruses to the propagation rules of computer viruses.

[0137] Please refer to Figure 3, which is a flow chart of a model acquisition method provided by an embodiment of the present application. In the model acquisition method described in Figure 3, the first network element and the second network element are respectively described as the first NWDAF and the second NWDAF. As shown in Figure 3, the model acquisition method may include but is not limited to the following steps:

[0138] S101: The second NWDAF sends a model acquisition request to the first NWDAF. Correspondingly, the first NWDAF receives the model acquisition request.

[0139] Optionally, the model acquisition request may be a model subscription request, such as NWDAF_MLModelProvision_Subscribe, that is, the second NWDAF may subscribe to the model supported by the first NWDAF through the model subscription request, or request the first NWDAF to train a machine learning (ML) model for the second NWDAF.

[0140] In an optional implementation, the model acquisition request may include, but is not limited to, one or more of the following parameters:

[0141] (1) One or more analytics ID(s) that identify the type of analysis the requested model is used for. For example, if the Analytics ID in the model acquisition request is an identifier for the analysis type used to analyze service experience, then the ML model requested by the second NWDAF is used for service experience analysis.

[0142] (2) ML model filter information, such as single network slice selection assistance information (S-NSSAI) and area of ​​interest, indicating that the requested model is for a specific slice or a specific area.

[0143] (3) Target of ML Model Reporting, such as specific UEs, a group of UE(s) or any UE (ieall UEs), indicates that the model requested by the second NWDAF is for a specific UE, a group of UE(s), or all UEs within the range of the model filtering information identifier (such as within a slice or a region).

[0144] (4) First indication information, used to instruct the second NWDAF to request the first NWDAF to obtain the model generated by the first NWDAF using transfer learning, such as the second NWDAF requesting the first NWDAF to generate the requested model using transfer learning. Optionally, the first indication information may be a transfer learning indication. Optionally, if the model acquisition request does not include the first indication information, the first NWDAF may also determine, based on local configuration, to use transfer learning to generate a model that meets the model acquisition request requirements of the second NWDAF; if the first indication information is included, it is used to indicate that the first NWDAF can only use transfer learning to generate the model, which means that if the first NWDAF cannot use the transfer learning method to train the model (for example, the first NWDAF finds that the similarity between the training dataset information of all trained models and the consumer's dataset is lower than a threshold), the consumer's request will be rejected; if this parameter is included, it can also be used to indicate that the first NWDAF can use the transfer learning method to generate the model, which means that after receiving the indication, the first NWDAF will first consider using the transfer learning method to generate the model. If the first NWDAF finds that the model cannot be trained using the transfer learning method (for example, the first NWDAF finds that the similarity between the training dataset information of all trained models and the training dataset information requested by the second NWDAF is lower than a threshold), it can also retrain a model that meets the requirements of the second NWDAF.

[0145] (5) First training data set information, used to indicate the training data set information that the model requested by the second NWDAF expects to use. The first training data set information may include a training data set, or include training data set description information, or may include both a training data set and a training data set description information. In one possible implementation, the training data set information is for each model of each analysis type, that is, at a granularity of per model ID per analytics ID for each analysis type identifier. This parameter can be used to calculate the similarity between the source domain and the target domain, and can also be used by the first NWDAF to fine-tune the model. If the second NWDAF does not include this parameter, the first NWDAF needs to further request and obtain this information from the second NWDAF.

[0146] (6) First threshold information, used to indicate the range that the similarity between the training dataset information used by the first model in the first NWDAF and the first training dataset information must meet; wherein the first model is a model in the first NWDAF that can use the transfer learning method to generate the model that the model acquisition request expects to obtain. The range indicated by the first threshold information can specify a lower limit of the value, or specify an upper limit of the value, or specify both a lower limit and an upper limit of the value. The model acquisition request includes the first threshold information, which can indicate that the value of the similarity between the training dataset information in the first NWDAF and the first training dataset information must meet the range indicated by the first threshold information before the transfer learning method can be used to generate the model requested by the model acquisition request. Optionally, if the model acquisition request does not include the first threshold information, the first NWDAF can determine whether the transfer learning method can be used to generate the model requested by the model acquisition request based on local configuration. For example, the first NWDAF can locally configure a range, and the value of the similarity between the training dataset information in the first NWDAF and the first training dataset information must meet the range before the transfer learning method can be used to generate the model requested by the model acquisition request.

[0147] (7) First similarity calculation method information, used to indicate the similarity calculation method used to calculate the similarity between the training dataset information used by the first model and the first training dataset information. Optionally, if the model acquisition request does not include the first similarity calculation method information, the first NWDAF determines the similarity calculation method to be used.

[0148] (8) First transfer learning method information, used to instruct the first NWDAF to generate the transfer learning method used by the model requested by the second NWDAF. Optionally, if the model acquisition request does not include the first transfer learning method information, the first NWDAF determines the transfer learning method to be used.

[0149] In another optional embodiment, the model acquisition request may include, but is not limited to, one or more of the following parameters: Analytics ID(s), ML model filter information, Target of ML Model Reporting, model information (such as ML Model ID(s), ML model address(es), etc.), first indication information, first training dataset information, and first transfer learning method information. The definitions of Analytics ID(s), ML model filter information, and Target of ML Model Reporting can be referred to the relevant descriptions of the above embodiment and are not described in detail here. The model identified by Model ID(s) is provided by the second NWDAF to the first NWDAF, and is used to instruct the first NWDAF to use the model to perform transfer learning and generate a model that meets the requirements of the second NWDAF. The first indication information is used to instruct the first NWDAF to use transfer learning to generate the model required by the second NWDAF. The first training dataset information is the training dataset information expected to be used by the model identified by Model ID(s). The first transfer learning method is used to indicate the transfer learning method used by the first NWDAF. Optionally, in this embodiment, Analytics ID(s) and model information are required parameters, and the remaining parameters may be optional.

[0150] The training dataset information may include the training dataset and / or the description information of the training dataset as described above. The description information of the training dataset may include but is not limited to the following parameters:

[0151] 1) ML Model Filter Information, such as S-NSSAI(s) and Area(s) of Interest, is used to describe which slices (S-NSSAI(s)) or areas (Area(s) of Interest) the training dataset is collected from;

[0152] 2) Target of ML Model Reporting, such as specific UEs, a group of UE(s) or any UE, is used to describe which UE or group of UEs the training dataset is collected from, or whether the training dataset is collected from all UEs within the specified range of the filtering information;

[0153] 3) Data sources, such as a list of NF instance (or NF set) IDs and corresponding NF Types, are used to describe which NF instances the training dataset is collected from. It can also specify the NF Type corresponding to the NF instance. For example, this parameter can specify that the training dataset is collected from AMF1, or this parameter can specify that the training dataset is collected from SMF1, or the parameter can specify that the training dataset is collected from AMF2, SMF2, and AF1.

[0154] 4) Event ID(s) per NF instance ID, used to describe the Event ID(s) by which the training dataset is collected from the corresponding NF instance. In one possible implementation, Event ID(s) is NF instance ID granularity, that is, for each data source specified by NF instance ID, information is given by which Event ID(s) the data source is collected from. For example<AMF1,Event ID=Location Report> It represents the data collected from AMF1 through the Location Report event (i.e., location data).<SMF2,Event ID=QFI allocation> It represents the data collected from SMF2 through the QFI allocation event (i.e., QFI, 5QI, DNN, S-NSSAI, etc.);

[0155] 5) Other filter information, such as App ID, is used to describe which application the dataset collects data related to. This filter information may be at the Event ID level, such as collecting data related to a specific application through a specific Event ID;

[0156] 6) The timestamp of the collected data (a specific moment), or the time period for collected data (Time period for collected data (per Event ID)). This parameter may also be at the Event ID granularity.

[0157] 7) Data metrics, such as a sampling ratio, the maximum number of input values ​​and / or the maximum time interval between samples of this input data, the data range including maximum and minimum values, and data distribution information such as mean, standard deviation, and data distribution.

[0158] S102: The first NWDAF determines a model acquisition response according to the model acquisition request.

[0159] Optionally, the model acquisition response may be a model subscription notification, such as NWDAF_MLModelProvision_Notify.

[0160] S103: The first NWDAF sends a model acquisition response to the second NWDAF. Correspondingly, the second NWDAF receives the model acquisition response.

[0161] In an optional implementation, when the model acquisition request does not include a model identifier (Model ID(s)), the first NWDAF determines a model acquisition response based on the model acquisition request, including: the first NWDAF determines whether there is a trained model that can meet the requirements of the second NWDAF (i.e., whether there is a trained model that meets the requirements of the model requested by the model acquisition request) based on information such as ML Model Filter Information and Target of ML Model Reporting in the model acquisition request; if so, directly returning the corresponding model through the model acquisition response, wherein the model acquisition response includes one or more ML model information, each ML model information includes an ML model identifier (ML Model identifier) ​​and an ML model file address information (such as a uniform resource locator (URL) or a fully qualified domain name (FQDN) of the model file), or each ML model information includes an ML model identifier (ML Model identifier) ​​and an ADRF (Set) ID information. If the ML model information includes ML model file address information, the second NWDAF can download the model according to the ML model file address information; if the ML model information includes ADRF (Set) ID information, the second NWDAF obtains the model based on the ML model identifier from the ADRF identified by the ADRF (Set) ID.If there is no trained model that can meet the requirements of the second NWDAF, the first NWDAF calculates the similarity between the training dataset information of the existing trained models (such as Model 1 and Model 2) and the training dataset information requested by the second NWDAF (that is, the model acquisition request includes the first training dataset information); if there is a model with a calculated similarity higher than the threshold (for example, the similarity between the training dataset information of Model 1 and the training dataset information requested by the second NWDAF is higher than the threshold), the first NWDAF performs a similarity check on the model (such as Model 1) based on the trained model (such as Model 1) and the training dataset information requested by the second NWDAF. 1) Fine-tune to obtain a model that meets the requirements of the second NWDAF, that is, a model that meets the requirements of the model requested by the model acquisition request, and then feedback the model through a model acquisition response, such as the model acquisition response including the ML model information of the fine-tuned model; if the similarity between the training dataset information of all trained models of the first NWDAF and the training dataset information requested by the second NWDAF is lower than the threshold, the first NWDAF needs to retrain a new model and feedback the model through the model acquisition response, or the first NWDAF can reject the model acquisition request of the second NWDAF, that is, inform the second NWDAF of the failure of the model acquisition request through the model acquisition response.

[0162] In another optional embodiment, if the model acquisition request includes a model identifier (Model ID(s)) (for example, the second NWDAF includes Model ID(s) information in the model acquisition request, that is, the model identified by the Model ID(s) information is a trained model of the first NWDAF, and the model meets the requirements of transfer learning), then the first NWDAF determines a model acquisition response based on the model acquisition request, including: the first NWDAF directly fine-tunes the model using transfer learning based on the model and the training data set information provided by the second NWDAF to obtain a model that meets the requirements of the second NWDAF. The model acquisition response includes ML model information of the fine-tuned model.

[0163] Optionally, the first NWDAF can determine the first model based on the model acquisition request. For example, the first model can be a trained model of the first NWDAF that meets the requirements of the second NWDAF, or a model retrained by the first NWDAF to meet the requirements of the second NWDAF, or a model fine-tuned by the first NWDAF. In this case, the ML model information in the model acquisition response is a required parameter. Optionally, the model acquisition response may also include but is not limited to one or more of the following optional parameters: (1) result indication information and / or a first reason. The result indication information is used to indicate that the model acquisition request of the second NWDAF failed, and the first reason is used to indicate the reason for the failure of the model acquisition request. Optionally, if the model acquisition response does not include the result indication information and the first reason, it means that the first NWDAF can successfully respond to the model acquisition request of the second NWDAF. Optionally, the model acquisition response includes result indication information, and the result indication information is used to indicate whether the model acquisition request of the second NWDAF is successful. Optionally, if the model acquisition response does not include result indication information, it means that the first NWDAF can successfully respond to the model acquisition request of the second NWDAF. Optionally, the model acquisition response includes a first reason, which is used to indicate the reason why the model acquisition request failed. Optionally, if the model acquisition response does not include the first reason, it means that the first NWDAF can successfully respond to the model acquisition request of the second NWDAF; (2) second indication information (such as Transfer Learning indication), which is used to indicate that the model fed back by the first NWDAF through the model acquisition response is generated through transfer learning; (3) second similarity calculation value information, which is used to indicate the calculation result value (or calculation value) of the similarity between the training data set information used by the first model and the first training data set information. Among them, the first model is a model in the first NWDAF that can use transfer learning to generate the model that the model acquisition request expects to obtain. (4) second similarity calculation method information, which is used to indicate the similarity calculation method used to calculate the similarity between the training data set information used by the first model and the first training data set information. (5) second transfer learning method information, which is used to indicate the transfer learning method used by the first NWDAF to generate the model requested by the second NWDAF.

[0164] As can be seen, in the model acquisition method shown in Figure 3, first indication information and related parameters can be added to the model acquisition request. Therefore, when the second NWDAF acquires the model, it can include the first indication information and related parameters in the model acquisition request, allowing the first NWDAF to use transfer learning to generate the required model and return the migrated model through the model acquisition response, thereby improving the efficiency of model training. In addition, this method can add second indication information and related parameters to the model acquisition response, which can also enable the second NWDAF to know whether the acquired model was obtained through transfer learning.

[0165] Please refer to Figure 4, which is a flowchart of another model acquisition method provided in an embodiment of the present application. The model acquisition method shown in Figure 4 differs from the model acquisition method shown in Figure 3 in that, in the model acquisition method shown in Figure 4, the first indication information in the model acquisition request is used to instruct the second NWDAF to request the first NWDAF to obtain a model that meets the transfer learning requirements. This embodiment is described below in conjunction with the accompanying drawings. As shown in Figure 4, the model acquisition method may include but is not limited to the following steps:

[0166] S201. The second NWDAF sends a model acquisition request to the first NWDAF. Correspondingly, the first NWDAF receives the model acquisition request.

[0167] Optionally, the model acquisition request may be a model subscription request, that is, the second NWDAF may subscribe to the model supported and provided by the first NWDAF through the model subscription request.

[0168] In an optional embodiment, the model acquisition request may include but is not limited to one or more of the following parameters: (1) Analytics ID(s), which is the same as the definition in the embodiment described in Figure 3 and is not described in detail here. (2) ML model filter information, which is the same as the definition in the embodiment described in Figure 3 and is not described in detail here. (3) Target of ML Model Reporting, which is the same as the definition in the embodiment described in Figure 3 and is not described in detail here. (4) First indication information, the first indication information is used to instruct the second NWDAF to request to obtain the model in the first NWDAF that meets the transfer learning requirements. For the sake of convenience, the model in the first NWDAF that meets the transfer learning requirements is referred to as the model to be migrated. The second NWDAF can use the model to be migrated and local data for fine-tuning to obtain the final model. Optionally, if the model acquisition request includes the first indication information, it can be used to indicate that the first NWDAF can only provide the model to be migrated, which means that if the first NWDAF cannot provide the model to be migrated (for example, the first NWDAF finds that the similarity between the training dataset information of all trained models and the consumer's dataset is lower than a threshold), the model acquisition request of the second NWDAF will be rejected. Optionally, if the model acquisition request includes the first indication information, it can be used to indicate that the first NWDAF can provide the model to be migrated, which means that after receiving the indication, the first NWDAF first considers providing the model to be migrated. When the first NWDAF finds that the model to be migrated can be provided (for example, the first NWDAF finds that the similarity between the training dataset information of all trained models and the training dataset information requested by the second NWDAF is lower than a threshold, wherein the training dataset information requested by the second NWDAF can be provided through the model acquisition request, such as the first training dataset information described below), it can also retrain a model that meets the request of the second NWDAF. (5) First training dataset information, used to indicate the training dataset information that the model requested by the second NWDAF expects to use. This parameter has the same definition as that described in FIG3 and will not be described in detail here. (6) First threshold information, used to indicate the range that the similarity between the training data set information used by the first model in the first NWDAF and the first training data set information must meet; wherein, the first model is the model in the first NWDAF that meets the requirements of transfer learning that the second NWDAF expects to obtain, that is, the model to be transferred. The range indicated by the first threshold information can specify a lower limit, or an upper limit, or both a lower limit and an upper limit. The model acquisition request includes the first threshold information, which indicates that the value of the similarity between the training data set information in the first NWDAF and the first training data set information must meet the range indicated by the first threshold information before it can be used as a model to be transferred.Optionally, if the model acquisition request does not include the first threshold information, the first NWDAF may determine whether the trained model can be used as the model to be migrated based on the local configuration. For example, the first NWDAF may locally configure a range, and the model whose similarity value between the training dataset information in the first NWDAF and the first training dataset information must meet the range can be fed back to the second NWDAF as the model to be migrated. (7) First similarity calculation method information, used to indicate the similarity calculation method used to calculate the similarity between the training dataset information used by the first model and the first training dataset information. Optionally, if the model acquisition request does not include the first similarity calculation method information, the first NWDAF determines the similarity calculation method to be used.

[0169] S202: The first NWDAF determines a model acquisition response according to the model acquisition request.

[0170] S203: The first NWDAF sends a model acquisition response to the second NWDAF. Correspondingly, the second NWDAF receives the model acquisition response.

[0171] The first NWDAF determines a model acquisition response based on the model acquisition request, including: the first NWDAF determines whether there is a trained model that can meet the requirements of the second NWDAF (that is, whether there is a trained model that meets the requirements of the model requested by the model acquisition request) based on information such as ML Model Filter Information and Target of ML Model Reporting in the model acquisition request; if so, directly returns the corresponding model through the model acquisition response, where the model acquisition response includes ML model information. If there is no trained model that can meet the requirements of the second NWDAF, the first NWDAF calculates the similarity between the training dataset information of the existing trained models (such as Model 1 and Model 2) and the training dataset information requested by the second NWDAF (that is, the first training dataset information is included in the model acquisition request); if there is a model with a calculated similarity higher than the threshold (for example, the similarity between the training dataset information of Model 1 and the training dataset information requested by the second NWDAF is higher than the threshold), the first NWDAF uses it as the model to be migrated and feeds back the model to be migrated through a model acquisition response, such as the model acquisition response including the ML model information of the model to be migrated; if the similarity between the training dataset information of all trained models of the first NWDAF and the training dataset information requested by the second NWDAF is lower than the threshold, the first NWDAF needs to retrain a new model and feed back the model through the model acquisition response, or the first NWDAF can reject the model acquisition request of the second NWDAF, that is, inform the second NWDAF of the failure of the model acquisition request through the model acquisition response.

[0172] Optionally, the first NWDAF can determine the first model based on the model acquisition request. For example, the first model can be a trained model of the first NWDAF that meets the requirements of the second NWDAF, or a model retrained by the first NWDAF to meet the requirements of the second NWDAF, or a model to be migrated determined by the first NWDAF. In this case, the ML model information in the model acquisition response is a required parameter. Optionally, the model acquisition response may also include but is not limited to one or more of the following optional parameters: (1) result indication information and / or a first reason. The result indication information is used to indicate that the model acquisition request of the second NWDAF failed, and the first reason is used to indicate the reason for the failure of the model acquisition request. Optionally, the model acquisition response does not include the result indication information and the first reason, which indicates that the first NWDAF can successfully respond to the model acquisition request of the second NWDAF. Optionally, the model acquisition response includes result indication information, and the result indication information is used to indicate whether the model acquisition request of the second NWDAF is successful. Optionally, the model acquisition response does not include result indication information, which indicates that the first NWDAF can successfully respond to the model acquisition request of the second NWDAF. Optionally, the model acquisition response includes a first reason, which is used to indicate the reason why the model acquisition request failed. Optionally, if the model acquisition response does not include the first reason, it means that the first NWDAF can successfully respond to the model acquisition request of the second NWDAF; (2) second indication information (such as Transfer Learning indication), the second indication information is used to indicate that the model fed back by the first NWDAF through the model acquisition response is a model that meets the transfer learning requirements; 3) second similarity calculation value information, the second similarity calculation value information is used to indicate the calculation result value (or calculation value) of the similarity between the training data set information used by the first model and the first training data set information. Among them, the first model is the model to be migrated determined by the first NWDAF. 4) second similarity calculation method information, the second similarity calculation method information is used to indicate the similarity calculation method used to calculate the similarity between the training data set information used by the first model and the first training data set information.

[0173] Optionally, if the model acquisition request includes the first indication information and the model fed back in the model acquisition response is the model to be migrated, then the method, as shown in FIG3 , further includes the following steps:

[0174] S204: The second NWDAF uses the model to obtain the model to be transferred in response to the feedback, and uses the transfer learning method to perform fine-tuning to obtain the required model.

[0175] Since the second NWDAF needs to fine-tune the model, or in other words, needs to retrain the model, the second NWDAF is an NWDAF with MTLF function.

[0176] As can be seen, in the model acquisition method shown in Figure 4, first indication information and related parameters can be added to the model acquisition request. Therefore, when the second NWDAF acquires the model, it can include the first indication information and related parameters in the model acquisition request, allowing the first NWDAF to provide the model to be migrated and return the model to be migrated through the model acquisition response. In this way, the second NWDAF can fine-tune the model to be migrated to obtain the required model, thereby improving the efficiency of model training. In addition, this method can add second indication information and related parameters to the model acquisition response, which can also enable the second NWDAF to know that the acquired model is the model to be migrated.

[0177] In another embodiment, the present application also provides a network element registration and discovery method, in which the first NWDAF can register the migration learning capability information with the NRF, so that the second NWDAF can discover the first NWDAF that supports the migration learning capability through the NRF, and then send a model acquisition request to the first NWDAF. Please refer to Figure 5, which is a flow chart of a network element registration and discovery method provided in an embodiment of the present application. As shown in Figure 5, the network element registration and discovery method may include but is not limited to the following steps:

[0178] S301. The first NWDAF sends a network element registration request to the NRF. The network element registration request includes fourth indication information, and the fourth indication information is used to indicate whether the first NWDAF supports using a transfer learning method to generate a model.

[0179] Alternatively, the fourth indication information is used to indicate whether the first NWDAF has transfer learning capability.

[0180] Optionally, the network element registration request is used to register capability information of the first NWDAF, and the network element registration request may be an NRF_NF Management_NF Registration Request (Nnrf_NFManagement_NFRegister Request).

[0181] The fourth indication information is used to indicate whether the first NWDAF has the ability to transfer learning. The network element registration request may also include Analytics ID, which indicates the analysis identifier supported by the first NWDAF, such as the identifier of the service experience analysis mentioned above, indicating that the first NWDAF supports providing a model for service experience analysis. Optionally, the network element registration request also includes one or more of the following parameters: (1) Second training data set information. If the first NWDAF supports transfer learning, then optionally, the training data set information is registered. The third training data set information can be per Model ID per Analytics ID granularity, that is, the first NWDAF registers the training data set information of each trained model. The training data set information may include the description information of the training data set and / or the training data set. The description information of the training data set can be found in the above text and will not be described in detail here. (2) NWDAF service area information (NWDAF Serving Area information), for example, it can be an area composed of one or more tracking areas (TA) and / or cells. (3) ML model filter information (such as S-NSSAI(s) and Area(s) of Interest) for the trained ML model(s) per Analytics ID(s), such as S-NSSAI(s) or Area(s) of Interest, indicating the scope of application of the NWDAF trained model. (4) ML Model Interoperability indicator per Analytics ID(s), which consists of a list of Vendor ID(s), indicating that the NWDAF supports providing models to the vendors identified by these Vendor IDs. (5) NF Set ID and / or NF Type of the NF data sources.

[0182] S302 : The NRF sends a network element registration response to the first NWDAF. Correspondingly, the first NWDAF receives the network element registration response. The network element registration response is used to indicate whether the registration is successful.

[0183] Optionally, the network element registration response may be an NRF_NF Management_NF Registration Response (Nnrf_NFManagement_NFRegister Response), wherein the network element registration response may include result indication information to indicate whether the registration is successful.

[0184] S303. The second NWDAF sends a network element discovery request (Discovery Request) to the NRF. Correspondingly, the NRF receives the network element discovery request, wherein the network element discovery request includes fifth indication information, and the fifth indication information is used to indicate a request to discover MTLF network elements with transfer learning capabilities.

[0185] Among them, the network element discovery request also includes Analytics ID, which indicates the analysis identifier requested by the second NWDAF, such as the identifier of the business experience analysis mentioned above, indicating that the second NWDAF requests to support the MTLF that provides the model for business experience analysis. Optionally, the network element discovery request may also include one or more of the following parameters: (1) third training data set information, which is the training data set information expected to be used by the model supported by the MTLF requested for discovery. (2) third threshold information, which is used to indicate the range of similarity that must be met between the training data set information used by the model in the MTLF requested for discovery and the third training data set information. When the network element discovery request includes the third training data set information, the NRF will compare the third training data set information with the training data set information registered by the MTLF (i.e., the second training data set information). If the similarity between the two is calculated, the NRF will feedback the MTLF instance whose similarity is higher than the threshold to the second NWDAF. For example, assuming that the similarity between the second training data set information registered by the first NWDAF and the third training data set information requested by the second NWDAF is higher than the threshold requested by the second NWDAF, the NRF can feedback the first NWDAF to the second NWDAF as the requested MTLF. (3) Third similarity calculation method information, the third similarity calculation method information is used to indicate the similarity calculation method used to calculate the similarity between the training dataset information used by the model in the MTLF (i.e., the training dataset information registered by the MTLF) and the third training dataset information. For example, if the training dataset information registered by the first NWDAF is training dataset A, and the third training dataset information is training dataset B, then the similarity between the training datasets can be calculated by some methods, for example, the normalized distance between the training dataset matrices A and B; if the training dataset information registered by the first NWDAF and the third training dataset information also include descriptive information of the training datasets, then the overlap of the descriptive information of the training datasets can also be calculated and normalized with the similarity between the training datasets to obtain a similarity value.

[0186] S304. The NRF discovers an MTLF network element with transfer learning capability, such as the first NWDAF, based on the network element discovery request.

[0187] Optionally, step S304 is: the NRF discovers, based on the network element discovery request, an MTLF (or NWDAF) that has transfer learning capability and has the highest similarity between the registered training dataset information and the third training dataset information or exceeds a threshold.

[0188] S305 . The NRF sends a network element discovery response (Discovery Response) to the second NWDAF. Correspondingly, the second NWDAF receives the network element discovery response.

[0189] Among them, the network element discovery response is returned for the network element discovery request. The network element discovery response includes the Analytics ID of the second NWDAF request and one or more NWDAF instances that meet the second NWDAF network element discovery request (such as the identifiers of one or more model training function network elements). Optionally, the network element discovery response also includes the identifiers of one or more model training function network elements. For each model training function network element identified by the identifier, the network element discovery response may also include one or more of the following information: model identifier, the value of the similarity between the training data set information used by the model and the third training data set information, and a similarity calculation method for calculating the similarity (a list of<Model ID,Domain similarity,Similarity measurement method> ). If the network element discovery request includes the third threshold information, for a specific Analytics ID and a specific NWDAF instance, the NWDAF (or MTLF) may have multiple models that can meet the similarity requirements of the third threshold information. For each model that meets the requirements, the NRF provides the Model ID of the model, the calculated similarity value corresponding to the Model ID, and the method used to calculate the similarity. Optionally, in addition to providing the Model ID, the network element discovery response can also provide the address information of the model file (egURL or FQDN) or the identifier of the analytical data storage network element (set) that stores the model (ADRF (Set) ID).

[0190] S306: The second NWDAF selects an NWDAF, such as the first NWDAF, from the network element discovery response.

[0191] S307: The second NWDAF sends a model acquisition request to the first NWDAF.

[0192] S308: The first NWDAF returns a model acquisition response to the second NWDAF.

[0193] Among them, the second NWDAF can obtain the model after transfer learning from the first NWDAF according to the embodiment described in Figure 3, or can obtain the model to be transferred from the first NWDAF according to the embodiment described in Figure 4, and then fine-tune the model using transfer learning to obtain a model that meets the requirements. This will not be described in detail here.

[0194] It can be seen that the embodiment of the present application registers the transfer learning capability information with NRF through NWDAF, so that the consumer can discover the MTLF that supports the transfer learning capability through NRF, avoiding the situation where NWDAF does not register the relevant information of the transfer learning capability, resulting in the MTLF discovered by the consumer not necessarily having the transfer learning capability and being unable to train the model using the transfer learning method, thereby helping to improve the efficiency of model training. The embodiment of the present application enables the consumer, such as the second NWDAF, to directly discover the MTLF with transfer learning capability, such as the first NWDAF. In addition, NWDAF can also register the training data set information of the trained model, so that the MTLF directly discovered is not only the MTLF with transfer learning capability but also the MTLF of the model that meets the transfer learning requirements (such as similarity requirements).

[0195] In another embodiment, if the NRF provides the second NWDAF with a Model ID that meets the transfer learning requirements in the network element discovery response, the second NWDAF can directly obtain the ML Model Information from the first NWDAF based on the Model ID (wherein the network element discovery response contains the address information of the model file (egURL or FQDN) or the ADRF (Set) ID storing the model), and then download the model based on the ML Model Information, or obtain the model from the ADRF, and fine-tune the model based on the transfer learning method to obtain a model that meets the requirements. If the NRF provides the second NWDAF with a Model ID that meets the transfer learning requirements in the network element discovery response and provides the ML Model Information, the second NWDAF can directly download the model based on the ML Model Information, or obtain the model from the ADRF, and fine-tune the model based on the transfer learning method to obtain a model that meets the requirements.

[0196] In another embodiment, the present application further provides a model / data storage and retrieval method, in which the first NWDAF can store model / data information to the ADRF, so that after the first NWDAF deletes the local model and training data, it can also retrieve the required model through the ADRF. Please refer to Figure 6, which is a flow chart of a model / data storage and retrieval method provided in an embodiment of the present application. As shown in Figure 6, the model / data storage and retrieval method may include but is not limited to the following steps:

[0197] S401: The first NWDAF sends a model or data storage request to the ADRF. Correspondingly, the ADRF receives the model or data storage request.

[0198] Optionally, a model or data storage request can be an ADRF_ML_MODEL_MANAGEMENT_STORAGE_REQUEST request.

[0199] (Nadrf_MLModelManagement_StorageRequest Request).

[0200] The model or data storage request includes the model identifier of the model requested for storage, the model file address of the model, and the training dataset information used by the model (list of<Model ID,address(e.g.URL or FQDN)of Model file,Training dataset Info> Optionally, the model or data storage request further includes an instance ID of the NF of the NWDAF containing MTLF (NF instance ID of the NWDAF containing MTLF), such as the ID of the first NWDAF, and an analytics ID (Analytics ID).

[0201] S402: The first NWDAF receives a model or data storage response from the ADRF. The model or data storage response is used to indicate whether the information requested to be stored by the model or data storage request is successfully stored.

[0202] Optionally, the model or data storage response can be a Nadrf_MLModelManagement_StorageRequest Response.

[0203] S403: The second NWDAF sends a model acquisition request to the first NWDAF. Correspondingly, the first NWDAF receives the model acquisition request.

[0204] The description of step S403 can refer to the relevant contents of the embodiments described in FIG. 3 and FIG. 4 , and will not be described in detail here.

[0205] S404: The first NWDAF determines the first model according to the model acquisition request.

[0206] Among them, the first model is a model that can use the transfer learning method to generate the model that the model acquisition request expects to obtain in the first NWDAF in the embodiment as described in Figure 3, or a model that meets the transfer learning requirements that the second NWDAF expects to obtain in the first NWDAF in the embodiment as described in Figure 4, that is, the model to be migrated.

[0207] After the first NWDAF determines that the trained model is the model requested by the model acquisition request in Figure 3 or Figure 4, if the first NWDAF stores the trained model in the ADRF and deletes the local model, the first NWDAF may execute step S405.

[0208] S405. The first NWDAF sends a model retrieval request to the ADRF. Correspondingly, the ADRF receives the model retrieval request, where the model retrieval request is used to request the ADRF to retrieve the first model. The model retrieval request includes third indication information, where the third indication information is used to indicate the training dataset information requested for the first model.

[0209] Optionally, the model retrieval request may be a Nadrf_MLModelManagement_Retrieval Request.

[0210] The model retrieval request includes the following parameters: Storage Transaction Identifier or one or more tuples of unique ML Model identifier(s). This means that the first NWDAF can retrieve a stored model from the ADRF using the Storage Transaction Identifier, or it can retrieve one or more models using the ML Model ID(s). Furthermore, the model retrieval request can also include the Training Dataset Info Indication parameter, which indicates that the training dataset information corresponding to the model must also be obtained.

[0211] S406. The ADRF returns a model retrieval response to the first NWDAF. Correspondingly, the first NWDAF receives the model retrieval response. The model retrieval response is used to feed back the retrieved first model and the training dataset information used by the first model.

[0212] Optionally, the model retrieval response may be Nadrf_MLModelManagement_Retrieval Response.

[0213] The model retrieval response also includes the following parameters: one or more tuples of unique ML Model identifiers and address (eg, URL or FQDN) of Model file stored in ADRF.

[0214] S407: The first NWDAF sends a model acquisition response to the second NWDAF. Correspondingly, the second NWDAF receives the model acquisition response.

[0215] The model acquisition response may include model information of the first model.

[0216] The description of step S407 can refer to the relevant contents of the embodiments described in FIG. 3 and FIG. 4 , and will not be described in detail here.

[0217] It can be seen that the embodiment of the present application saves the correspondence between the ML model and the model training dataset information in ADRF, so that the first NWDAF can retrieve the corresponding model training dataset information while retrieving the model from ADRF. The first NWDAF can determine whether the model meets the requirements of transfer learning based on the training dataset information of the model and the training dataset information provided by the second NWDAF.

[0218] In the embodiments provided by the present application, the scheme of each method provided by the embodiment of the present application is introduced from the perspective of each node itself and from the perspective of interaction between each node. It is understandable that each node, such as the first node, the second node, etc., in order to realize the above-mentioned functions, includes a hardware structure and / or software unit corresponding to each function. Those skilled in the art should be easily aware that, in conjunction with the units and algorithm steps of each example described in the embodiment disclosed in this application, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a certain function is executed in a way that hardware or computer software drives hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to realize the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0219] Please refer to FIG7 , which is a schematic diagram of the structure of a communication device provided in an embodiment of the present application. The communication device shown in FIG7 includes a transceiver module 701 and a processing module 702 .

[0220] In one design of this embodiment, the communication device is the first NWDAF or a related device in the first NWDAF:

[0221] Exemplarily, the transceiver module 701 is used to receive a model acquisition request from the second NWDAF, where the model acquisition request is used to request the acquisition of the model of the first NWDAF, and the model acquisition request includes first indication information; wherein the first indication information is used to instruct the second NWDAF to request the acquisition of the model generated by the first network element using the transfer learning method, or to instruct the second NWDAF to request the acquisition of the model in the first NWDAF that meets the transfer learning requirements; the transceiver module 701 is also used to send a model acquisition response to the second NWDAF.

[0222] Optionally, the communication device may also include a processing module 702, which is used to generate the model requested by the model acquisition request using a transfer learning method, or determine a model that meets the transfer learning requirements, and then the transceiver module 701 sends the generated or determined model to the second NWDAF through a model acquisition response.

[0223] Optionally, when the communication device is the first NWDAF or a device related to the first NWDAF, it is used to implement the functions and optional implementation methods of the first NWDAF in the embodiments shown in Figures 1 to 6.

[0224] In one design of this embodiment, the communication device is the second NWDAF or a related device in the second NWDAF:

[0225] The transceiver module 701 is used to send a model acquisition request to the first network element, where the model acquisition request is used to request the acquisition of the model of the first network element. The model acquisition request includes first indication information; wherein the first indication information is used to instruct the second network element to request the acquisition of the model generated by the first network element using the transfer learning method, or to instruct the second network element to request the acquisition of the model in the first network element that meets the transfer learning requirements; the transceiver module 701 is also used to receive a model acquisition response from the first network element.

[0226] Optionally, the communication device further includes a processing module 702 configured to determine a first NWDAF so that the communication unit can send a model acquisition request to the first NWDAF. Optionally, the processing module 702 can also generate a desired model using transfer learning based on the model returned in the model acquisition response.

[0227] Optionally, when the communication device is a second NWDAF or a device related to the second NWDAF, it is used to implement the functions and optional implementation methods of the second NWDAF in the embodiments shown in Figures 1 to 6.

[0228] Please refer to Figure 8, which is a schematic diagram of the structure of another communication device provided in an embodiment of the present application. The communication device shown in Figure 8 includes at least one processor 801, a memory 802, and optionally, a transceiver 803. The specific connection medium between the above-mentioned processor 801 and the memory 802 is not limited in the embodiment of the present application. In Figure 8, the connection between the memory 802 and the processor 801 via the bus 804 is taken as an example. The bus 804 is represented by a bold line in the figure. The connection method between other components is only for schematic illustration and is not limited. The bus 804 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one bold line is used in Figure 8, but it does not mean that there is only one bus or one type of bus.

[0229] The processor 801 may have a data transceiver function and may communicate with other devices. In the apparatus shown in FIG8 , an independent data transceiver module, such as a transceiver 803, may also be provided for transmitting and receiving data. When the processor 801 communicates with other devices, data may be transmitted through the transceiver 803.

[0230] In one example, when the first NWDAF adopts the form shown in Figure 8, the processor 801 in Figure 8 can call the computer execution instructions stored in the memory 802 to enable the first NWDAF to execute the method executed by the first NWDAF in any embodiment of Figures 1 to 6.

[0231] In one example, when the second NWDAF adopts the form shown in Figure 8, the processor 801 in Figure 8 can call the computer execution instructions stored in the memory 802 to enable the second NWDAF to execute the method executed by the second NWDAF in any embodiment of Figures 1 to 6.

[0232] The present application also provides a communication system, which may include the first NWDAF and at least two second NWDAFs shown in Figures 1 to 6, for details, see the method embodiments described above. Optionally, the system may also include the NRF and / or ADRF described above, for details, see the method embodiments described above.

[0233] The solutions described in this application can be implemented in various ways. For example, these techniques can be implemented in hardware, software, or a combination of hardware. For hardware implementation, the processing module used to execute these techniques at a communication device (e.g., a base station, a terminal, a network entity, or a chip) can be implemented in one or more general-purpose processors, digital signal processors (DSPs), digital signal processing devices, application-specific integrated circuits (ASICs), programmable logic devices, field programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, or any combination thereof. The general-purpose processor can be a microprocessor, and optionally, the general-purpose processor can also be any conventional processor, controller, microcontroller, or state machine. The processor can also be implemented by a combination of computing devices, such as a digital signal processor and a microprocessor, multiple microprocessors, one or more microprocessors combined with a digital signal processor core, or any other similar configuration.

[0234] It is understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus RAM (DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0235] The present application also provides a computer-readable medium having instructions stored thereon, which, when executed by a computer, enables the computer device to implement the functions of any of the above method embodiments through a transceiver and a processor.

[0236] The present application also provides a computer program product, which, when executed by a computer, enables the computer device to implement the functions of any of the above method embodiments through a transceiver and a processor.

[0237] The present application also provides a processing device, which may be a product such as a chip device, and the processing device is used to determine the relevant information required in the embodiments described in any of the diagrams in Figures 1 to 6. For example, the processing device is used to determine a model acquisition response. For another example, the processing device is used to use the model to be migrated, fine-tune it through transfer learning, and obtain the required model. Optionally, the processing device may be a baseband processing module, and the information determined by the processing device may be sent out through the radio frequency processing module, such as the migrated model or the model to be migrated. Alternatively, information sent by other devices is received through the radio frequency processing module, and the processing device determines other information based on the information.

[0238] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated therein. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a high-density digital video disc (DVD)), or a semiconductor medium (eg, a solid state disk (SSD)).

[0239] It is understandable that some optional features in the embodiments of the present application may, in certain scenarios, be implemented independently of other features, such as the solution on which they are currently based, to solve corresponding technical problems and achieve corresponding effects. They may also be combined with other features in certain scenarios as needed. Accordingly, the devices provided in the embodiments of the present application may also implement these features or functions accordingly, which will not be described in detail here.

[0240] Those skilled in the art will also appreciate that the various illustrative logical blocks and steps listed in the embodiments of the present application can be implemented by electronic hardware, computer software, or a combination of both. Whether such functions are implemented by hardware or software depends on the specific application and the design requirements of the entire system. Those skilled in the art may use various methods to implement the described functions for corresponding applications, but such implementation should not be construed as exceeding the scope of protection of the embodiments of the present application.

[0241] It will be understood that the “embodiment” mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, the various embodiments in the entire specification do not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It will be understood that in the various embodiments of the present application, the size of the sequence numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0242] It can be understood that in this application, "when", "if" and "if" all mean that the device will perform corresponding processing under certain objective circumstances, and do not limit the time, nor do they require the device to make judgments when it is implemented, nor do they mean that there are other limitations.

[0243] In this application, elements expressed in the singular are intended to mean "one or more" rather than "one and only one", unless otherwise specified. In this application, unless otherwise specified, "at least one" is intended to mean "one or more", and "more than one" is intended to mean "two or more". In the text description of this application, "including at least one of A, B and C" may mean: including A; including B; including C; including A and B; including A and C; including B and C; including A, B and C. In the text description of this application, "and / or" is merely a description of the association relationship of associated objects, indicating that there may be three relationships. For example, A and / or B may mean: A exists alone, A and B exist at the same time, and B exists alone, where A may be singular or plural, and B may be singular or plural.

[0244] It is understood that the various numbers involved in the embodiments of the present application are only for the convenience of description and are not intended to limit the scope of the embodiments of the present application. The size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic. In addition, the terms "system" and "network" are often used interchangeably in this article.

[0245] The predefined in this application may be understood as defined, predefined, stored, pre-stored, pre-negotiated, pre-configured, solidified, or pre-burned.

[0246] Those skilled in the art will appreciate that, for the sake of convenience and brevity of description, the specific working processes of the above-described systems, devices, and units may refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0247] The same or similar parts between the various embodiments in this application can refer to each other. In the various embodiments in this application, and the various implementation methods / implementation methods / implementation methods in each embodiment, if there is no special explanation and logical conflict, the terms and / or descriptions between different embodiments and the various implementation methods / implementation methods / implementation methods in each embodiment are consistent and can be referenced to each other. The technical features in different embodiments and the various implementation methods / implementation methods / implementation methods in each embodiment can be combined to form new embodiments, implementation methods, implementation methods, or implementation methods according to their inherent logical relationships. The above-described implementation methods of this application do not constitute a limitation on the scope of protection of this application.

[0248] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A model acquisition method, characterized in that: The method comprises: The first network element receives a model acquisition request from the second network element, where the model acquisition request is used to request to acquire a model of the first network element, and the model acquisition request includes first indication information; The first indication information is used to instruct the second network element to request to obtain a model generated by the first network element using a transfer learning method, or is used to instruct the second network element to request to obtain a model in the first network element that meets the transfer learning requirement; The first network element sends a model acquisition response to the second network element.

2. A model acquisition method, characterized in that: The method comprises: The second network element sends a model acquisition request to the first network element, where the model acquisition request is used to request to acquire a model of the first network element, and the model acquisition request includes first indication information; The first indication information is used to instruct the second network element to request to obtain a model generated by the first network element using a transfer learning method, or is used to instruct the second network element to request to obtain a model in the first network element that meets the transfer learning requirement; The second network element receives a model acquisition response from the first network element.

3. The method according to claim 1 or 2, characterized in that: The model acquisition request also includes first training data set information. The first training data set information is training data set information that the model requested to obtain expects to use.

4. The method according to claim 3, characterized in that The model acquisition request also includes first threshold information. The first threshold information is used to indicate a range that the similarity between the training data set information used by the first model in the first network element and the first training data set information must satisfy; Among them, the first model is a model in the first network element that can use transfer learning to generate the model that the model acquisition request expects to obtain, or a model in the first network element that meets the transfer learning requirements that the second network element expects to obtain.

5. The method according to claim 4, characterized in that The model acquisition request also includes first similarity calculation method information, The first similarity calculation method information is used to indicate a similarity calculation method used to calculate the similarity between the training data set information used by the first model and the first training data set information.

6. The method according to any one of claims 1 to 5, characterized in that The model acquisition request also includes first transfer learning method information. The first transfer learning method information is used to instruct the first network element to generate a transfer learning method used by the model requested by the second network element.

7. The method according to any one of claims 1 to 6, characterized in that The model acquisition response includes result indication information and / or a first reason, The result indication information is used to indicate that the model acquisition request of the second network element failed, and the first reason is used to indicate the reason for the failure of the model acquisition request.

8. The method according to any one of claims 1 to 7, characterized in that The model acquisition response includes second indication information, The second indication information is used to indicate that the model through which the first network element obtains response feedback is generated through transfer learning, or to indicate that the model through which the first network element obtains response feedback is a model that meets the requirements of transfer learning.

9. The method according to claim 4 or 5, characterized in that: The model acquisition response also includes second similarity calculation value information, The second similarity calculation value information is used to indicate a calculation result value of the similarity between the training data set information used by the first model and the first training data set information.

10. The method according to claim 9, characterized in that The model acquisition response also includes second similarity calculation method information, The second similarity calculation method information is used to indicate a similarity calculation method used to calculate the similarity between the training data set information used by the first model and the first training data set information.

11. The method according to any one of claims 1 to 10, characterized in that The model acquisition response also includes second transfer learning method information, The second transfer learning method information is used to instruct the first network element to generate a transfer learning method used by the model requested by the second network element.

12. The method according to claim 1, characterized in that After the first network element receives the model acquisition request from the second network element, the method further includes: The first network element determines a first model according to the model acquisition request; The first network element sends a model retrieval request to the analysis data storage function network element, where the model retrieval request is used to request the analysis data storage function network element to retrieve the first model; the model retrieval request includes third indication information, where the third indication information is used to indicate the training data set information used to request the first model; The first network element receives a model retrieval response from the analysis data storage function network element, where the model retrieval response is used to feed back information about the first model retrieved and a training data set used by the first model; Among them, the first model is a model in the first network element that can use transfer learning to generate the model that the model acquisition request expects to obtain, or a model in the first network element that meets the transfer learning requirements that the second network element expects to obtain.

13. The method according to claim 12, characterized in that Before the first network element sends the model retrieval request to the analysis data storage function network element, the method further includes: The first network element sends a model or data storage request to the analysis data storage function network element, wherein the model or data storage request includes a model identifier of the model requested to be stored, a model file address of the model, and training data set information used by the model; The first network element receives a model or data storage response from the analysis data storage function network element, where the model or data storage response is used to indicate whether the information requested to be stored by the model or data storage request is successfully stored.

14. The method according to claim 1, or 12, or 13, characterized in that: The method further comprises: The first network element sends a network element registration request to the network storage function network element, where the network element registration request includes fourth indication information, where the fourth indication information is used to indicate whether the first network element supports generating a model using a transfer learning method; The first network element receives a network element registration response from the network storage function network element, where the network element registration response is used to indicate whether the registration is successful.

15. The method according to claim 14, characterized in that The network element registration request also includes second training data set information, The second training data set information is training data set information of a model supported and provided by the first network element.

16. The method according to claim 2, characterized in that Before the second network element sends the model acquisition request to the first network element, the method further includes: The second network element sends a network element discovery request to the analysis data storage function network element, where the network element discovery request includes fifth indication information, where the fifth indication information is used to indicate a request to discover a model training function network element with transfer learning capability; The second network element receives a network element discovery response from the analysis data storage function network element.

17. The method according to claim 16, characterized in that The network element discovery request also includes third training data set information, where the third training data set information is training data set information that is expected to be used by a model supported by the model training function network element requested for discovery.

18. The method according to claim 17, characterized in that The network element discovery request further includes third threshold information, where the third threshold information is used to indicate a range of similarity between the training data set information used by the model in the model training function network element and the third training data set information that needs to be satisfied; The model training functional network element is the model training functional network element that the network element discovery request in the analysis data storage functional network element expects to obtain.

19. The method according to claim 18, characterized in that The network element discovery request also includes third similarity calculation method information, The third similarity calculation method information is used to indicate the similarity calculation method used to calculate the similarity between the training data set information used by the model in the model training function network element and the third training data set information.

20. The method according to any one of claims 17 to 19, characterized in that The network element discovery response includes the identifiers of one or more model training function network elements. For each model training function network element identified by the identifier, the network element discovery response may also include one or more of the following information: a model identifier, a similarity value between the training data set information used by the model and the third training data set information, and a similarity calculation method for calculating the similarity.

21. A communication system, characterized in that: The system comprises: At least one first network element for performing the method of any one of claims 1, or 3 to 15; and At least one second network element for executing the method according to any one of claims 2 to 11, or 16 to 20.

22. A communication device, characterized in that: The communication device comprises a processor and a transceiver, wherein the processor calls a computer program stored in a memory through the transceiver, so that the communication device implements the method as claimed in claim 1, or any one of claims 3 to 15, or implements the method as claimed in any one of claims 2 to 11, or 16 to 20.

23. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed, enables a computer device to implement the method as claimed in claim 1, or any one of claims 3 to 15, or any one of claims 2 to 11, or any one of claims 16 to 20 through a transceiver and a processor.

Images