Model data management method, network element, network system and storage medium

Abstract

The invention provides a model data management method, a network element, a network system and a storage medium, and relates to the technical field of communication. The model data management method comprises the steps that a model storage network element obtains a model storage request from a first model training network element, and the model storage request comprises access condition information of a to-be-registered model; the access condition information of the to-be-registered model comprises at least one of the following information: whether the to-be-registered model is allowed to be discovered and acquired by network elements except the first model training network element; allowing discovery and acquisition of the address of the network element of the to-be-registered model; whether the to-be-registered model is allowed to be updated by network elements except the first model training network element is determined; the address of the network element of the to-be-registered model is allowed to be updated; or allowing to obtain or modify the identification of the manufacturer of the to-be-registered model.

Description

Technical Field

[0001] The present disclosure relates to the field of communication technology, and in particular to a model data management method, a network element, a network system and a storage medium. Background Art

[0002] In the 3GPP research, the key issue of using ADRF (The Analytics Data Repository Function) network elements to store ML (Machine Learning) models was proposed, and preliminary research has been completed.

[0003] In 5G networks, NWDAF (Network Data Analytics Function) (hereinafter referred to as MTLF) with MTLF (Model Training Logical Function) can perform model training and store the trained model in ADRF. The model stored in ADRF is extracted and used by model network elements for further training, and the further generated model is stored in ADRF. Summary of the invention

[0004] One object of the present disclosure is to improve the security of model data management.

[0005] According to one aspect of some embodiments of the present disclosure, a model data management method is proposed, including: a model storage network element obtains a model storage request from a first model training network element, the model storage request includes access condition information of the model to be registered, and the access condition information of the model to be registered includes at least one of the following information: whether the model to be registered is allowed to be discovered and acquired by network elements other than the first model training network element; the address of the network element that allows the model to be registered to be discovered and acquired; whether the model to be registered is allowed to be updated by network elements other than the first model training network element; the address of the network element that allows the model to be registered to be updated is allowed to be updated; or the manufacturer's identifier of the model to be registered is allowed to be acquired or modified.

[0006] In some embodiments, the method further includes: the model storage network element obtains the model to be registered from the first model training network element; and the model storage network element stores the model to be registered.

[0007] In some embodiments, the method further includes: the model storage network element sends a model registration request to the registration network element, wherein the model registration request includes a model identifier of the model to be registered and access condition information of the model to be registered.

[0008] In some embodiments, the method also includes: the model storage network element obtains a historical model request from the second model training network element, wherein the second model training network element sends the historical model request based on the model identifier of the target model from the registration network element, and the registration network element feeds back the model identifier of the target model to the second model training network element when determining based on the access condition information of the target model that the second model training network element is allowed to obtain the target model, and the historical model request includes the model identifier of the target model; the model storage network element feeds back the target model to the second model training network element based on the model identifier.

[0009] In some embodiments, the method also includes: the model storage network element obtains a model update request from the second model training network element, the model update request includes an updated model of the target model and updated access condition information; the model storage network element updates the currently stored target model with the updated model, and sends registration update information to the registration network element according to the model update request, so that the registration network element updates the associated information of the target model according to the registration update information, and the associated information includes access condition information.

[0010] According to one aspect of some embodiments of the present disclosure, a model management method is proposed, including: a model training network element sends a model storage request to a model storage network element, the model storage request includes access condition information of the model to be registered, and the access condition information of the model to be registered includes at least one of the following information: whether the model to be registered is allowed to be discovered and acquired by network elements other than the current model training network element; the address of the network element that allows the model to be registered to be discovered and acquired; whether the model to be registered is allowed to be updated by network elements other than the current model training network element; the address of the network element that allows the model to be registered to be updated is allowed to be updated; or the manufacturer's identifier of the model to be registered is allowed to be acquired or modified.

[0011] In some embodiments, the method also includes: the model training network element sends a model discovery request to the registration network element; the model training network element sends a historical model request to the model storage network element based on the model identifier of the target model from the registration network element, wherein the registration network element determines the target model based on the model discovery request, and when it is determined that the current model training network element is allowed to obtain the target model based on the access condition information of the target model, the registration network element feeds back the model identifier of the target model to the model training network element, and the historical model request includes the model identifier of the target model; the model training network element obtains the target model fed back by the model storage network element.

[0012] In some embodiments, the method also includes: the model training network element obtains update condition information of the target model from the registration network element, wherein, after receiving the model discovery request, the registration network element feeds back the update condition information of the target model to the current model training network element according to the access condition information; when the update condition information includes allowing update by the current model training network element, sending a model update request to the model storage network element so that the model storage network element updates the associated information of the target model; when the update condition information includes not allowing update by the current model training network element, sending a model registration request to the model storage network element, and the model identifier in the model registration request is different from the model identifier of the target model.

[0013] According to one aspect of some embodiments of the present disclosure, a model management method is proposed, including: a registration network element obtains a model registration request from a model storage network element and stores information carried in the model registration request, wherein the model registration request includes a model identifier and access condition information of the model to be registered, and the model storage network element sends a model registration request to the registration network element when obtaining a model storage request for the model to be registered from a first model training network element; stores information carried in the model registration request, wherein the access condition information of the model to be registered includes at least one of the following information: whether the model to be registered is allowed to be discovered and acquired by network elements other than the first model training network element; the address of the network element that allows the model to be registered to be discovered and acquired; whether the model to be registered is allowed to be updated by network elements other than the first model training network element; the address of the network element that allows the model to be registered to be updated is allowed to be updated; or the manufacturer's identifier of the model to be registered is allowed to be acquired or modified.

[0014] In some embodiments, the method also includes: the registration network element receives a model discovery request from the second model training network element; the registration network element determines the target model based on the model discovery request; and when it is determined based on the access condition information of the target model that the second model training network element is allowed to obtain the target model, the model identifier of the target model is fed back to the second model training network element so that the second model training network element requests the target model from the model storage network element.

[0015] In some embodiments, the method further includes: obtaining a registration update request from a model storage network element; and updating associated information of a model corresponding to the registration update request according to the registration update request, wherein the associated information includes access condition information of the updated model.

[0016] In some embodiments, the method also includes: the registration network element feeds back update condition information of the target model to the second model training network element based on the access condition information, so that the second model training network element determines the update strategy of the target model in the model storage network element based on the update condition information, wherein, when the update condition information includes not allowing update by the second model training network element, the update strategy includes adding a model to the model storage network element; when the update condition information includes allowing update by the second model training network element, the update strategy includes updating the target model in the model storage network element.

[0017] According to one aspect of some embodiments of the present disclosure, a model storage network element is proposed, including: a request acquisition unit, configured to obtain a model storage request from a first model training network element, the model storage request including access condition information of the model to be registered, the access condition information of the model to be registered including at least one of the following information: whether the model to be registered is allowed to be discovered and acquired by network elements other than the first model training network element; the address of the network element that allows the model to be registered to be discovered and acquired; whether the model to be registered is allowed to be updated by network elements other than the first model training network element; the address of the network element that allows the model to be registered to be updated is allowed to be updated; or the manufacturer's identifier of the model to be registered is allowed to be acquired or modified.

[0018] In some embodiments, the request acquisition unit is also configured to obtain the model to be registered from the first model training network element; the model storage network element also includes: a model storage unit, configured to store the model to be registered.

[0019] In some embodiments, the request acquisition unit is further configured to acquire a historical model request from a second model training network element, wherein the second model training network element sends a historical model request based on a model identifier of a target model from a registration network element, and the registration network element, when determining based on access condition information of the target model that the second model training network element is allowed to acquire the target model, feeds back the model identifier of the target model to the second model training network element, and the historical model request includes the model identifier of the target model; the model storage network element also includes: a model feedback unit configured to determine the target model based on the model identifier, and feed it back to the second model training network element.

[0020] According to one aspect of some embodiments of the present disclosure, a model training network element is proposed, comprising: a request sending unit, configured to send a model storage request to a model storage network element, the model storage request including access condition information of the model to be registered; the access condition information of the model to be registered includes at least one of the following information: whether the model to be registered is allowed to be discovered and acquired by network elements other than the current model training network element; the address of the network element that allows the model to be registered to be discovered and acquired; whether the model to be registered is allowed to be updated by network elements other than the current model training network element; the address of the network element that allows the model to be registered to be updated is allowed to be updated; or the manufacturer's identifier of the model to be registered is allowed to be acquired or modified.

[0021] In some embodiments, the request sending unit is also configured to send a model discovery request to the registration network element, and send a historical model request to the model storage network element based on the model identifier of the target model from the registration network element; the model training network element also includes: a model acquisition unit, configured to obtain the target model fed back by the model storage network element.

[0022] According to one aspect of some embodiments of the present disclosure, a registration network element is proposed, comprising: a request receiving unit, configured to obtain a model registration request from a model storage network element, wherein the model registration request includes a model identifier and access condition information of the model to be registered, and the model storage network element stores the model to be registered and sends the model registration request to the registration network element when obtaining the model storage request for the model to be registered from the first model training network element; a storage unit, configured to store information carried in the model registration request, wherein the access condition information of the model to be registered includes at least one of the following information: whether the model to be registered is allowed to be discovered and acquired by network elements other than the first model training network element; the address of the network element that allows the model to be registered to be discovered and acquired; whether the model to be registered is allowed to be updated by network elements other than the first model training network element; the address of the network element that allows the model to be registered to be updated; or the manufacturer's identifier of the model to be registered is allowed to be acquired or modified.

[0023] In some embodiments, the request receiving unit is also configured to receive a model discovery request from the second model training network element; the registration network element also includes: a model determination unit, configured to determine the target model according to the model discovery request; and an information feedback unit, configured to feedback the model identifier of the target model to the second model training network element when it is determined according to the access condition information of the target model that the second model training network element is allowed to obtain the target model, so that the second model training network element requests the target model from the model storage network element.

[0024] According to one aspect of some embodiments of the present disclosure, a network element is proposed, comprising: a memory; and a processor coupled to the memory, wherein the processor is configured to execute any one of the model data management methods described above based on instructions stored in the memory.

[0025] According to one aspect of some embodiments of the present disclosure, a computer-readable storage medium is provided, on which computer program instructions are stored, and when the instructions are executed by a processor, the steps of any one of the above model data management methods are implemented.

[0026] According to one aspect of some embodiments of the present disclosure, a network system is proposed, comprising: a registration network element, configured to execute any one of the model data management methods described above and executed by the registration network element; a model storage network element, configured to execute any one of the model data management methods described above and executed by the model storage network element; and a plurality of model training network elements, each model training network element being configured to execute any one of the model data management methods described above and executed by the model training network element. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The drawings described herein are used to provide further understanding of the present disclosure and constitute a part of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute improper limitations on the present disclosure.

[0028] Figure 1 The present invention is a flowchart of some embodiments of the model data management method performed by the model training network element and the model storage network element.

[0029] Figure 2 The present invention is a flowchart of some embodiments of the model data management method performed by the registration network element.

[0030] Figure 3A -C is a flowchart of other embodiments of the model data management method disclosed in the present invention.

[0031] Figure 4 The present invention provides a signaling flow chart of some embodiments of the interaction between network elements in the model data management method disclosed herein.

[0032] Figure 5 Schematic diagram of some embodiments of the registration network element of the present disclosure.

[0033] Figure 6 Schematic diagram of some embodiments of the model storage network element of the present disclosure.

[0034] Figure 7 Schematic diagram of some embodiments of the model training network element disclosed in the present invention.

[0035] Figure 8 Schematic diagrams of some embodiments of network elements of the present disclosure.

[0036] Fig. 9 Schematic diagrams of other embodiments of the network element of the present disclosure.

[0037] Fig.10 Schematic diagrams of some embodiments of the network system disclosed herein. DETAILED DESCRIPTION

[0038] The technical solution of the present disclosure is further described in detail below through the accompanying drawings and embodiments.

[0039] In some private networks or some specific networks, the ML models used by AI (Artificial Intelligence) services may have certain privacy. When a model training network element (such as MTLF) stores a model in a model storage network element (such as ADRF), it may not want other model training network elements to discover and obtain the model or only allow some model training network elements to discover and obtain the model. At the same time, some model training network elements do not want other model training network elements or only allow some model training network elements to modify or update the stored ML models.

[0040] To address the above problems, a model data management method, network element, network system and storage medium are proposed to improve the efficiency and security of core network AI network element model storage and retrieval.

[0041] The flowcharts of some embodiments of the model data management method disclosed in the present invention are as follows: Figure 1 In some embodiments, Figure 1 Step 111 on the left side is executed by the model training network element, and steps 121 and 122 are executed by the model storage network element.

[0042] In step 111, when the model training network element (hereinafter referred to as the first model training network element) wants to store the model trained or stored by itself in the model storage network element, it sends a model storage request to the model storage network element. In some embodiments, the model storage request includes storage condition information of the model to be stored (or referred to as to be registered).

[0043] In some embodiments, the access condition information includes at least one of the following: whether the model to be registered is allowed to be discovered and acquired by network elements other than the first model training network element (i.e., the network element that issues the model storage request); whether the address of the network element that allows the model to be registered to be discovered and acquired; whether the model to be registered is allowed to be updated by network elements other than the first model training network element; whether the address of the network element that allows the model to be registered to be updated; or whether the manufacturer's identifier of the model to be registered is allowed to be acquired or modified.

[0044] In some embodiments, the access condition information includes whether the model to be registered is allowed to be discovered and acquired by network elements other than the first model training network element, or at least one of the addresses of network elements that allow the model to be registered to be discovered and acquired, thereby limiting the sharing of the model between network elements.

[0045] In some embodiments, the access condition information includes whether the model to be registered is allowed to be updated by network elements other than the first model training network element, or at least one of the addresses of network elements that allow updating of the model to be registered, thereby restricting the update of the model in the model storage network element.

[0046] In some embodiments, the storage condition information includes an identifier of a manufacturer that allows acquisition or modification of the model to be registered, thereby using the manufacturer as a restriction condition to improve the convenience of management and control.

[0047] In step 121, the model storage network element obtains the model storage request from the first model training network element, especially the model storage condition information therein. In some embodiments, the model storage network element can store this information so as to serve as a restriction condition for subsequently providing models to other model training network elements or updating the models stored by itself. In some embodiments, the model storage network element can register with the registration network element using the storage condition information to improve the centralization of management and the convenience of model query.

[0048] Based on the method in the embodiment shown above, the model training network element can provide access condition information to the model stored in the model storage network element to limit the model from being acquired or updated by other network elements, thereby improving the controllability of the model generation network element over the network element and the security of the model.

[0049] In some embodiments, the model storage request also includes the model to be stored. Figure 1 As shown, in step 122, the model storage network element stores the model acquired in step 121. In some embodiments, the model storage network element may associate storage condition information with the model, thereby improving the accuracy of subsequent queries.

[0050] In some embodiments, after the registration network element obtains a model discovery request from a model training network element, based on the requirements in the relevant standards, the model storage network element associated with the model training network element is determined, and the identifier or address of the model storage network element is fed back to the model training network element, so as to facilitate the model training network element to obtain the model from the target model storage network element. In some embodiments, the model storage network element associated with the model training network element can be determined based on the area, distance or pre-setting. In some embodiments, the model training network element sends a historical model request to the model storage network element, and the model storage network element can determine whether the target model can be provided to the model training network element based on the access condition information, thereby improving the security of model management.

[0051] In some embodiments, the model storage network element can determine whether to allow the model training network element that initiates the historical model request to update the target model stored in the model storage network element and acquired by the model training network element based on the access condition information, thereby improving the reliability of the model storage network element for maintaining the models stored in itself.

[0052] The flowcharts of some embodiments of the model data management method disclosed in the present invention are as follows: Figure 2 In some embodiments, Figure 2 The model data management method shown in is executed by the registration network element.

[0053] In step 211, the registration network element obtains a model registration request from the model storage network element and stores the information carried in the model registration request. In some embodiments, the model registration request includes a model identifier and access condition information of the model to be registered. When the model storage network element obtains the model storage request of the model to be registered from the first model training network element, the model storage network element sends the model registration request to the registration network element.

[0054] In some embodiments, the access condition information includes at least one of the following: whether the model to be registered is allowed to be discovered and acquired by network elements other than the first model training network element (i.e., the network element that issues the model storage request); whether the address of the network element that allows the model to be registered to be discovered and acquired; whether the model to be registered is allowed to be updated by network elements other than the first model training network element; whether the address of the network element that allows the model to be registered to be updated; or whether the manufacturer's identifier of the model to be registered is allowed to be acquired or modified.

[0055] In some embodiments, the access condition information includes whether the model to be registered is allowed to be discovered and acquired by network elements other than the first model training network element, or at least one of the addresses of network elements that allow the model to be registered to be discovered and acquired, thereby limiting the sharing of the model between network elements.

[0056] In some embodiments, the access condition information includes whether the model to be registered is allowed to be updated by network elements other than the first model training network element, or at least one of the addresses of network elements that allow updating of the model to be registered, thereby restricting the update of the model in the model storage network element.

[0057] In some embodiments, the storage condition information includes an identifier of a manufacturer that allows acquisition or modification of the model to be registered, thereby using the manufacturer as a restriction condition to improve the convenience of management and control.

[0058] In step 212, the registration network element stores the information carried in the model registration request.

[0059] Based on the method in the embodiment shown above, the registration network element can store the access condition information of each model in the model storage network element, so that the model access authority of the model training network element can be restricted on the registration network element side, thereby improving the model security while reducing the processing burden of the model storage network element.

[0060] The flowcharts of some embodiments of the model data management method disclosed in the present invention are as follows: Figure 3A As shown, it includes steps 311 to 314. In some embodiments, the model data management method is executed by a registration network element, such as an NRF (Network Repository Function) network element.

[0061] In step 311, the registration network element receives a model discovery request from the second model training network element. In some embodiments, the second model training network element sends a model discovery request to the registration network element when receiving a model request from the AI ​​analysis network element and does not have the model required by the AI ​​analysis network element. In some embodiments, the model discovery request includes at least one of a target analysis identifier, filtering information, target performance, and target serialization information.

[0062] In step 312, the registering network element determines the target model according to the model discovery request.

[0063] In some embodiments, the registration network element determines the target model based on the model discovery request and the information carried in the stored model registration request. In some embodiments, when the model storage network element initiates model registration to the registration network element, the model registration request is sent to the registration network element. In some embodiments, the registration request includes a model identifier of the model to be registered, and in some embodiments, it also includes at least one of an analysis identifier of the model to be registered, a valid area of ​​the model, a valid time, model serialization information, and model performance information.

[0064] In step 313, it is determined whether the second model training network element is allowed to obtain the target model according to the access condition information of the target model. If the second model training network element is allowed to obtain the target model, step 314 is executed; otherwise, the model discovery fails and the process ends. In some embodiments, the registration network element can feedback the failure information to the second model training network element.

[0065] In some embodiments, the target model is provided by the first model training network element to the model storage network element for storage, and the access condition information includes whether the model to be registered is allowed to be discovered and acquired by network elements other than the first model training network element, or the address of the network element that allows the model to be registered to be discovered and acquired, or at least one of the manufacturer IDs that allow the model to be discovered and acquired. If the target model is allowed to be acquired by model training network elements other than the first model training network element, or the address of the second model training network element is in the address of the network element that allows the model to be registered to be discovered and acquired, or the manufacturer ID of the second model training network element is in the manufacturer ID specified in the storage condition, then the ID of the target model is determined, otherwise, this model discovery fails.

[0066] In step 314, the registration network element feeds back the model identifier of the target model to the second model training network element, so that the second model training network element requests the target model from the model storage network element.

[0067] Based on the method in the embodiment shown above, when the model training network element needs to obtain the model, it can initiate a request to the registration network element, which stores the access condition information of each model. Based on the stored information, it determines whether the model training network element is allowed to obtain the corresponding model, thereby improving the security protection of the model while enabling different model training network elements to share and use the model.

[0068] In some embodiments, the registration network element can also obtain a model registration request from the model storage network element and store the information carried in the model registration request. In some embodiments, the model registration request includes a model identifier of the model to be registered and access condition information of the model to be registered. When the model storage network element obtains a model storage request for the model to be registered from the first model training network element, the model storage network element stores the model to be registered and sends a model registration request to the registration network element.

[0069] Through the method in the embodiment shown above, after the model storage network element stores the model from the model training network element, it will initiate registration with the registration network element; the registration network element can obtain the model registration request issued by the model storage network element, and store the model identifier of the model to be registered and the access condition information of the model to be registered in association with each other, so as to facilitate the judgment of whether the model can be provided to the network element requesting the model based on the access condition information when the model is requested, thereby realizing model reuse and sharing while improving the security of the stored model.

[0070] In some embodiments, the access condition information also includes whether the model to be registered is allowed to be updated by network elements other than the first model training network element, or whether the address of the network element to be registered is allowed to be updated, or at least one of the manufacturer ID of the model to be registered is allowed to be modified. In the above step 313, it can also be determined whether the target model is allowed to be updated by the second model training network element. In some embodiments, the registration network element feeds back the update condition information of the target model to the second model training network element, so that the second model training network element determines the update strategy of the target model in the model storage network element according to the update condition information, wherein, when the update condition information includes not allowing the model to be updated by the second model training network element, the update strategy includes adding a model to the model storage network element; when the update condition information includes allowing the model to be updated by the second model training network element, the update strategy includes updating the target model in the model storage network element. In some embodiments, if the target model is allowed to be updated by a model training network element other than the model training network element that generates the target model, or the second model training network element is in the address of the network element that allows updating of the target model, or the manufacturer of the second model training network element belongs to the manufacturer that allows modification of the model to be registered, then it is determined that the target model is allowed to be updated by the second model training network element, and the update strategy for subsequent models is to replace the target model stored in the model storage network element with the model trained by the second model training network element; otherwise, it is determined that the target model is not allowed to be updated by the second model training network element, and the update strategy for subsequent models is to store the model trained by the second model training network element as a new model in the model storage network element.

[0071] Based on the method in the embodiment shown above, when storing the model, parameters are added to indicate whether other network elements are allowed to update the model and the address of the network elements that are allowed to update the model, thereby reducing the storage amount of different versions of models in the model storage network element and improving the utilization rate of storage resources.

[0072] In some embodiments, the registration network element can also obtain a model update request from the model storage network element, the model update request includes a model identifier of the corresponding model, and the model identifier is the same as the identifier of the updated model (the model obtained from the model storage network element based on which the model storage network element further trains), wherein when the model storage network element receives the model update request from the model training network element, it sends a registration update request to the registration network element. The registration network element updates the associated information of the model corresponding to the registration update request according to the registration update request, and the associated information includes access condition information. In some embodiments, the associated information also includes at least one of the analysis identifier, the effective area of ​​the model, the effective time, the model serialization information, and the model performance information.

[0073] Based on the method in the embodiment shown above, when the model stored in the model storage network element is updated, the corresponding information is also updated in the registration network element, thereby ensuring that the registration information can match the updated model version, further improving the security of the model.

[0074] Flow charts of other embodiments of the model data management method disclosed in the present invention are as follows: Figure 3B In some embodiments, the model data management method is executed by a model storage network element, such as ADRF.

[0075] In step 321, the model storage network element obtains a historical model request from the second model training network element. In some embodiments, the second model training network element sends a historical model request according to the model identifier of the target model from the registration network element. When the registration network element determines that the second model training network element is allowed to obtain the target model according to the access condition information of the target model, the registration network element feeds back the model identifier of the target model to the second model training network element, and the historical model request includes the model identifier of the target model.

[0076] In step 322, the model storage network element determines the target model according to the model identifier and feeds it back to the second model training network element.

[0077] Based on the method in the embodiment shown above, the model storage network element can determine the model identifier of its target model based on the request of the model training network element, and then feedback the corresponding model, so that the model training network element verified by the registration network element can obtain the existing model, thereby realizing model reuse and sharing while improving the security of the stored model.

[0078] In some embodiments, the model storage network element can also obtain a model storage request from the first model training network element, the model storage request including the model to be registered and the access condition information of the model to be registered. The model storage network element stores the model to be registered and sends a model registration request to the registration network element, the model registration request including the model identifier of the model to be registered and the access condition information of the model to be registered.

[0079] Through this method, the model storage network element can store the models generated by each model training network element, and generate a model identifier for searching and obtaining the model, and store the access conditions in the registration network element, so that other model training network elements can perform model search and authority verification in the registration network element, thereby improving the security of model storage and the reuse rate of the model.

[0080] In some embodiments, when the model storage network element obtains a model update request from the second model training network element, the model storage network element updates the currently stored target model with the updated model, and sends registration update information to the registration network element according to the model update request, so that the registration network element updates the associated information of the target model according to the registration update information, and the associated information includes access condition information, wherein the model update request includes the updated model of the target model and the updated access condition information.

[0081] Through this method, the model storage network element can update the version of the stored model and update the registration information in the registration network element, thereby reducing the storage pressure of the model storage network element, improving the matching degree between the registration information and the model, and improving security.

[0082] Flowcharts of some other embodiments of the model data management method disclosed in the present invention are as follows: Figure 3C In some embodiments, the model data management method is executed by a model training network element, such as MTLF.

[0083] In step 331, the model training network element sends a model discovery request to the registration network element. In some embodiments, the model discovery request includes at least one of a target analysis identifier, filtering information, target performance, and target serialization information. In some embodiments, the model training network element may generate a model discovery request based on the request of the AI ​​analysis network element and the target information carried in the request.

[0084] In step 332, the model training network element sends a historical model request to the model storage network element based on the model identifier of the target model from the registration network element, wherein the registration network element determines the target model based on the model discovery request, and when it is determined that the model training network element is allowed to obtain the target model based on the access condition information of the target model, the registration network element feeds back the model identifier of the target model to the model training network element, and the historical model request includes the model identifier of the target model.

[0085] In step 333, the model training network element obtains the target model fed back by the model storage network element.

[0086] Based on the method in the embodiment shown above, the model training network element can send a model discovery request to the registration network element in response to model needs, and then obtain the model from the model storage network element when the registration network element finds the corresponding model and confirms that the model training network element has the acquisition authority, thereby improving the security of the stored model while realizing model reuse and sharing.

[0087] In some embodiments, the model training network element may also send a model storage request to the model storage network element after generating the model, the model storage request including the model to be registered and access condition information of the model to be registered, so that the model storage network element stores the model to be registered, and sends a model registration request to the registration network element, the model registration request including the model identifier of the model to be registered and access condition information of the model to be registered.

[0088] Through this method, the generated model can be stored in the model storage network element, which is conducive to the reuse and sharing of the model and can also reduce the storage burden of the model training network element.

[0089] In some embodiments, after sending a model discovery request to the registration network element, the model training network element can also obtain update condition information of the target model from the registration network element. In some embodiments, after receiving the model discovery request, the registration network element feeds back the update condition information of the target model to the second model training network element according to the access condition information. When the model training network element completes acquiring the target model from the model storage network element and generates an updated version based on the target model, it is necessary to store the updated version to the model storage network element. In the case where the update condition information includes allowing update by the second model training network element, a model update request is sent to the model storage network element so that the model storage network element updates the associated information of the target model, wherein the model identifier of the updated version model is the same as the target model. In the case where the update condition information includes not allowing update by the second model training network element, a model registration request is sent to the model storage network element, and the model identifier in the model registration request is different from the model identifier of the target model.

[0090] Through this method, compared with the model training network element directly storing a model in the model storage network element each time it generates a model, the storage burden of the model storage network element can be reduced and the utilization rate of storage resources can be improved.

[0091] The signaling flow chart of some embodiments of the interaction between network elements in the model data management method of the present disclosure is as follows: Figure 4 The signaling interaction may include three stages, such as the model storage stage 41, the model request node 42 and the model update stage 43 as shown in the figure.

[0092] In the model storage phase, the model training network element needs to upload the trained data to the model storage network element. In some embodiments, the process of the model storage phase has no absolute order relationship with other phases and can occur independently of the steps of other phases.

[0093] In 411, the model training network element stores the model to the model storage network element. The parameters include analysis ID, model file (including model parameters, model topology and other information) or model address (URL or FQDN), model information (including the effective area and effective time of the model), model serialization information, model performance, whether other network elements are allowed to discover and obtain the model, the address of the network element that allows discovery and acquisition of the model, whether other network elements are allowed to update the model (mainly refers to whether the model is allowed to be updated in the model storage network element), the address of the network element that allows updating the model, the manufacturer ID that allows acquisition of the model, the manufacturer ID that allows modification of the model, etc.

[0094] In 412, the model storage network element stores the model and assigns a model ID etc. to the model.

[0095] In 413, the model storage network element registers the relevant information of the model with the registration network element, including analysis ID, model information (including the effective area and effective time of the model), model serialization information, model performance, model ID, whether other network elements are allowed to discover and obtain the model, the address list of network elements that are allowed to discover and obtain the model, whether other network elements are allowed to update the model, the address list of network elements that are allowed to update the model, the manufacturer ID that is allowed to obtain the model, the manufacturer ID that is allowed to modify the model, etc.

[0096] In the model request phase 42, the model training network element has a model acquisition requirement and hopes to extract the model from the model storage network element. Assume that the AI ​​analysis network element makes a model request to the model training network element.

[0097] In 421, the AI ​​analysis network element requests the model from the model training network element. The parameters include analysis ID, model filtering information, etc., which are consistent with the 3GPP process.

[0098] In 422, if the model training network element decides to obtain the historical model, the model training network element discovers the model storage network element through the registration network element, and the parameters sent to the registration network element include but are not limited to analysis ID, model filtering information (including but not limited to S-NSSAI, AoI, time range, etc.), model performance, model serialization information, etc. The registration network element determines whether the registration information of the model matches the request, and should match the following conditions: ① The model meets the requirements of the request, including analysis ID, filtering information, etc.; ② The model is allowed to be discovered and acquired by other network elements, or the model training network element is in the list of network element addresses where the model is allowed to be discovered and acquired. The registration network element returns the address of the model storage network element where the matching model is stored. In some embodiments, the registration network element can return the model ID corresponding to the matching model.

[0099] In 423, the model training network element sends a historical model request to the model storage network element. If the registration network element returns a model ID in step 422, the model training network element can directly request the model corresponding to the model ID; if the model ID is not returned, the model training network element includes but is not limited to analysis ID, model filtering information, model performance, model serialization information and other information in the request. The model storage network element returns the model corresponding to the request, and the parameters include but are not limited to: model file (including model parameters, model topology and other information) or model address (URL or FQDN), model information (including the effective area and effective time of the model), model serialization information, model performance, whether to allow the model to be updated in the model storage network element (if the model allows other network elements to update, and the model training network element is in the list of allowed updates, it is allowed, otherwise it is not allowed), etc.

[0100] In 424, the model training network element performs subsequent training until the requirements of the AI ​​analysis network element are met.

[0101] In 425, the model training network element sends the AI ​​analysis network element response model.

[0102] In the model updating stage 43, after completing the model training, the model training network element stores the model obtained after the training in the model storage network element.

[0103] In some embodiments, if the model training network element does not obtain the permission to update this model in the model storage network element in the preamble 423 (such as overwriting historical models to save storage resources), the trained model is stored as a new model in the model storage network element according to the process of the model storage stage 41; if the model training network element obtains the permission to update this model in the model storage network element in the preamble 423, execute 431-432.

[0104] In 431, after the model training network element subsequently trains the model, it updates the model to the model storage network element, and the parameters include the updated model file (including model parameters, model topology and other information) or the updated model address (URL or FQDN), updated effective area, effective time, model performance, serialization information, etc.

[0105] In 432 , the model storage network element updates the relevant registration information to the registration network element, and the updated parameters are similar to the operation in 413 .

[0106] Based on the method in the embodiment shown above, the model training network element adds new parameters when storing the model in the model storage network element, including whether other network elements are allowed to discover and obtain the model and the address of the network element that allows discovery and acquisition of the model, which can protect privacy and increase the security of model storage and acquisition; when storing the model, adding parameters such as whether other network elements are allowed to update the model and the address of the network element that allows updating the model can improve the efficiency of model storage and acquisition, and at the same time improve the utilization rate of storage resources to a certain extent.

[0107] Schematic diagrams of some embodiments of the registration network element disclosed in the present invention are as follows Figure 5 shown.

[0108] The request receiving unit 511 is capable of obtaining a model registration request from a model storage network element, wherein the model registration request includes a model identifier and access condition information of the model to be registered. When the model storage request of the model to be registered is obtained from the first model training network element, the model storage network element stores the model to be registered and sends the model registration request to the registration network element. In some embodiments, the access condition information includes at least one of the following: whether the model to be registered is allowed to be discovered and obtained by network elements other than the first model training network element (i.e., the network element that issues the model storage request); the address of the network element that allows the model to be registered to be discovered and obtained; whether the model to be registered is allowed to be updated by network elements other than the first model training network element; the address of the network element that allows the model to be registered to be updated; or the identification of the manufacturer that allows the model to be registered to be obtained or modified.

[0109] The storage unit 514 can store the information carried in the model registration request.

[0110] Such a registration network element can store the access condition information of each model in the model storage network element, so that the model access authority of the model training network element can be restricted on the registration network element side, thereby improving the model security and reducing the processing burden of the model storage network element.

[0111] In some embodiments, the request receiving unit 511 can also receive a model discovery request from the second model training network element. In some embodiments, the second model training network element sends a model discovery request to the registration network element when receiving a model request from the AI ​​analysis network element and does not have the model required by the AI ​​analysis network element. In some embodiments, the model discovery request includes at least one of a target analysis identifier, filtering information, target performance, and target serialization information.

[0112] like Figure 5As shown, the registration network element also includes a model determination unit 512, which can determine the target model according to the model discovery request. In some embodiments, the model determination unit 512 determines the target model according to the model discovery request based on the information carried in the stored model registration request. In some embodiments, when the model storage network element initiates model registration to the registration network element, a model registration request is sent to the registration network element. In some embodiments, the registration request includes a model identifier of the model to be registered, and in some embodiments, it also includes at least one of the analysis identifier of the model to be registered, the effective area of ​​the model, the effective time, the model serialization information, and the model performance information.

[0113] The registration network element also includes an information feedback unit 513 capable of determining whether the target model is allowed to be acquired by the second model training network element. When it is determined that the second model training network element is allowed to acquire the target model based on the access condition information of the target model, the model identifier of the target model is fed back to the second model training network element so that the second model training network element can request the target model from the model storage network element.

[0114] Such a registration network element can initiate a request to the registration network element when the model training network element needs to obtain the model. The registration network element stores the access condition information of each model and determines whether the model training network element is allowed to obtain the corresponding model based on the stored information, thereby realizing model reuse and sharing while improving the security of the stored models.

[0115] In some embodiments, the information feedback unit 513 can also determine whether to allow the second model training network element to update the stored target model to the model storage network element based on the access condition information, and feedback the update condition information of the target model to the second model training network element, so that the second model training network element can determine the update strategy of the target model in the model storage network element based on the update condition information, wherein, when the update condition information includes not allowing update by the second model training network element, the update strategy includes adding a model to the model storage network element; when the update condition information includes allowing update by the second model training network element, the update strategy includes updating the target model in the model storage network element.

[0116] Such a registration network element reduces the storage amount of different versions of models in the model storage network element and improves the utilization rate of storage resources.

[0117] In some embodiments, the request receiving unit 511 can also obtain a model update request from a model storage network element, the model update request includes a model identifier of the corresponding model, and the model identifier is the same as the identifier of the updated model (the model obtained from the model storage network element based on which the model storage network element further trains), wherein when the model storage network element receives the model update request from the model training network element, it sends a registration update request to the registration network element. The storage unit 514 updates the associated information of the model corresponding to the registration update request according to the registration update request, and the associated information includes access condition information. In some embodiments, the associated information also includes at least one of the analysis identifier, the effective area of ​​the model, the effective time, the model serialization information, and the model performance information.

[0118] Such a registration network element can update corresponding information after the model stored in the model storage network element is updated, thereby ensuring that the registration information can match the updated model version, further improving the security of the model.

[0119] Schematic diagrams of some embodiments of the model storage network element disclosed in the present invention are as follows Figure 6 shown.

[0120] The request acquisition unit 611 is capable of acquiring a model storage request from the first model training network element, wherein the model storage request includes access condition information of the model to be registered, and the access condition information of the model to be registered includes at least one of the following information: whether the model to be registered is allowed to be discovered and acquired by network elements other than the first model training network element; the address of the network element that allows the model to be registered to be discovered and acquired; whether the model to be registered is allowed to be updated by network elements other than the first model training network element; the address of the network element that allows the model to be registered to be updated; or the manufacturer's identifier of the model to be registered is allowed to be acquired or modified.

[0121] Such a model storage network element can obtain access condition information of each model to be stored to restrict the model from being acquired or updated by other network elements, thereby improving the controllability of the model generation network element over the network element and the security of the model.

[0122] In some embodiments, the request acquisition unit 611 can also acquire the model to be stored. Figure 6 As shown, the model storage network element further includes a model storage unit 613, which can store the model obtained by the request obtaining unit 611. In some embodiments, the model storage unit 613 can associate storage condition information with the model, thereby improving the accuracy of subsequent queries.

[0123] In some embodiments, the request acquisition unit 611 can also obtain a historical model request from a second model training network element, wherein the second model training network element sends a historical model request based on the model identifier of the target model from the registration network element, and the registration network element, when determining that the second model training network element is allowed to obtain the target model based on the access condition information of the target model, feeds back the model identifier of the target model to the second model training network element, and the historical model request includes the model identifier of the target model.

[0124] The model storage network element also includes a model feedback unit 612, which can determine the target model according to the model identifier and feed back to the second model training network element.

[0125] Such a model storage network element can determine the model identifier of its target model based on the request of the model training network element, and then feedback the corresponding model, so that the model training network element verified by the registration network element can obtain the existing model, thereby improving the security of model storage.

[0126] In some embodiments, when the request acquisition unit 611 is also capable of acquiring a model update request from a second model training network element, the model storage network element 613 updates the currently stored target model with the updated model, and sends registration update information to the registration network element according to the model update request, so that the registration network element updates the associated information of the target model according to the registration update information, and the associated information includes access condition information, wherein the model update request includes the updated model of the target model and the updated access condition information.

[0127] Such a model storage network element can perform version updates on the stored models and update the registration information in the registration network element, thereby reducing the storage pressure of the model storage network element, improving the matching degree between the registration information and the model, and improving security.

[0128] Schematic diagrams of some embodiments of the model training network element disclosed in the present invention are as follows Figure 7 shown.

[0129] The request sending unit 711 can send a model storage request to the model storage network element, where the model storage request includes access condition information of the model to be registered.

[0130] In some embodiments, the access condition information includes at least one of the following: whether the model to be registered is allowed to be discovered and acquired by network elements other than the first model training network element (i.e., the network element that issues the model storage request); whether the address of the network element that allows the model to be registered to be discovered and acquired; whether the model to be registered is allowed to be updated by network elements other than the first model training network element; whether the address of the network element that allows the model to be registered to be updated; or whether the manufacturer's identifier of the model to be registered is allowed to be acquired or modified.

[0131] Such a model training network element can provide access condition information to the model stored in the model storage network element to restrict the model from being acquired or updated by other network elements, thereby improving the controllability of the model generation network element over the network element and the security of the model.

[0132] In some embodiments, in some embodiments, after the model training network element generates the model, the request sending unit 711 includes the model to be registered in the model storage request sent to the model storage network element, so that the model storage network element stores the model to be registered, thereby storing the generated model in the model storage network element, which is conducive to model reuse and sharing, and can also reduce the storage burden of the model training network element.

[0133] In some embodiments, the request sending unit 711 is also capable of sending a model discovery request to a registration network element, and sending a historical model request to a model storage network element based on the model identifier of the target model from the registration network element, wherein the registration network element determines the target model based on the model discovery request, and when it is determined that the model training network element is allowed to obtain the target model based on the access condition information of the target model, the registration network element feeds back the model identifier of the target model to the model training network element, and the historical model request includes the model identifier of the target model.

[0134] The model training network element also includes a model acquisition unit 712, which can acquire the target model fed back by the model storage network element.

[0135] Such a model training NE can send a model discovery request to the registration NE in response to model needs. When the registration NE finds the corresponding model and confirms that the model training NE has the access authority, it can obtain the model from the model storage NE, thereby improving the security of the stored model while achieving model reuse and sharing.

[0136] In some embodiments, the model acquisition unit 712 can also obtain update condition information of the target model from the registration network element. When the model training network element completes the acquisition of the target model from the model storage network element and generates an updated version based on the target model, it is necessary to store the updated version to the model storage network element. In the case where the update condition information includes allowing update by the second model training network element, the request sending unit 711 sends a model update request to the model storage network element so that the model storage network element updates the associated information of the target model, wherein the model identifier of the updated version of the model is the same as the target model. In the case where the update condition information includes not allowing update by the second model training network element, the request sending unit 711 sends a model registration request to the model storage network element, and the model identifier in the model registration request is different from the model identifier of the target model.

[0137] Compared with directly storing a model in a model storage network element each time a model is generated, such a model training network element can reduce the storage burden of the model storage network element and improve the utilization rate of storage resources.

[0138] A structural diagram of an embodiment of the network element disclosed in the present invention is shown in FIG. Figure 8 As shown. The network element may be any one of the registration network element, model storage network element or model training network element mentioned above, including a memory 801 and a processor 802. Wherein: the memory 801 may be a disk, a flash memory or any other non-volatile storage medium. The memory is used to store instructions in the corresponding embodiments of the model data management method executed by the registration network element, the model storage network element or the model training network element mentioned above. The processor 802 is coupled to the memory 801 and may be implemented as one or more integrated circuits, such as a microprocessor or a microcontroller. The processor 802 is used to execute instructions stored in the memory, which can improve the security of the model.

[0139] In one embodiment, it is also possible to Fig. 9 As shown, the network element 900 includes a memory 901 and a processor 902. The processor 902 is coupled to the memory 901 via a BUS 903. The network element 900 can also be connected to an external storage device 905 via a storage interface 904 to call external data, and can also be connected to a network or another computer system (not shown) via a network interface 906. No further detailed description will be given here.

[0140] In this embodiment, the security of the model can be improved by storing data instructions in a memory and processing the instructions through a processor.

[0141] In another embodiment, a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the steps of the method in the corresponding embodiment of the model data management method performed by a registration network element, a model storage network element, or a model training network element. It should be understood by those skilled in the art that the embodiments of the present disclosure may be provided as methods, devices, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable non-transitory storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0142] Schematic diagrams of some embodiments of the network system disclosed in the present invention are as follows Fig.10 shown.

[0143] The registration network element 1030 may be any one of the registration network elements mentioned above, and execute any one of the model data management methods executed by the registration network element mentioned above.

[0144] The model storage network element 1020 may be any one of the model storage network elements mentioned above, and may execute any one of the model data management methods executed by the model storage network element mentioned above.

[0145] There are multiple model training network elements 1010, each of which can be any one of the model training network elements mentioned above, and can execute any one of the model data management methods executed by the model training network element mentioned above.

[0146] In such a network system, the model training network element adds new parameters when storing the model in the model storage network element, including whether other network elements are allowed to discover and obtain the model and the address of the network element that allows the discovery and acquisition of the model. This can protect privacy and increase the security of model storage and acquisition. Furthermore, when storing the model, adding parameters such as whether other network elements are allowed to update the model and the address of the network element that allows the update of the model can improve the efficiency of model storage and acquisition, and at the same time improve the utilization of storage resources to a certain extent.

[0147] In some embodiments, taking the model training network element as MTLF, the model storage network element as ADRF, and the registration network element as NRF as an example, the connection relationship between the network elements is as follows: Fig.10 shown.

[0148] Consumer and AnLF may be based on 3GPP specifications in related technologies.

[0149] When MTLF has no available ML model locally, it can discover ADRF through NRF and obtain the historical model stored in ADRF, and use it directly or as the pre-trained model for this model request for subsequent training. After training a new ML model, it stores the model in ADRF or updates the corresponding historical model.

[0150] ADRF stores ML models, assigns a model ID to each model, and registers model-related information with NRF.

[0151] NRF determines whether the model meets the requirements of the MTLF model request and returns the ADRF address corresponding to the model that meets the MTLF request requirements.

[0152] The network system based on the above-mentioned embodiment can facilitate the NWDAF network element to act as an AI network element to perform transfer learning, continuous learning, and model collaboration and sharing between multiple NWDAFs; in addition, in other network environments, the AI ​​network elements in the network can also perform transfer learning, continuous learning, and knowledge sharing based on the network system disclosed in the present invention.

[0153] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0154] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0155] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0156] So far, the present disclosure has been described in detail. In order to avoid obscuring the concept of the present disclosure, some details known in the art are not described. Based on the above description, those skilled in the art can fully understand how to implement the technical solution disclosed here.

[0157] The method and apparatus of the present disclosure may be implemented in many ways. For example, the method and apparatus of the present disclosure may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present disclosure are not limited to the order specifically described above, unless otherwise specifically stated. In addition, in some embodiments, the present disclosure may also be implemented as a program recorded in a recording medium, which includes machine-readable instructions for implementing the method according to the present disclosure. Therefore, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.

[0158] It should be noted that the terms "first", "second", etc. in the specification, claims, and drawings of the present disclosure are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present disclosure rather than to limit it. Although the present disclosure has been described in detail with reference to the preferred embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present disclosure can still be modified or some technical features can be replaced by equivalents without departing from the spirit of the technical solution of the present disclosure, which should be included in the scope of the technical solution for protection of the present disclosure.

Claims

1. A model data management method, comprising: The model storage network element obtains a model storage request from the first model training network element, wherein the model storage request includes access condition information of the model to be registered. The access condition information of the model to be registered includes at least one of the following information: Whether to allow the model to be registered to be discovered and acquired by network elements other than the first model training network element; Allowing discovery and acquisition of the address of the network element of the model to be registered; Whether to allow the model to be registered to be updated by network elements other than the first model training network element; Allowing update of the address of the network element of the model to be registered; or Allows to obtain or modify the manufacturer's ID of the model to be registered.

2. The method according to claim 1, further comprising: The model storage network element obtains the model to be registered from the first model training network element; The model storage network element stores the model to be registered.

3. The method according to claim 1, further comprising: The model storage network element sends the model registration request to the registration network element, wherein the model registration request includes the model identifier of the model to be registered and access condition information of the model to be registered.

4. The method according to claim 1, further comprising: The model storage network element obtains a historical model request from a second model training network element, wherein the second model training network element sends the historical model request according to the model identifier of the target model from the registration network element, and the registration network element feeds back the model identifier of the target model to the second model training network element when determining according to the access condition information of the target model that the second model training network element is allowed to obtain the target model, and the historical model request includes the model identifier of the target model; The model storage network element feeds back the target model to the second model training network element according to the model identifier.

5. The method according to claim 4, further comprising: The model storage network element obtains a model update request from the second model training network element, wherein the model update request includes an updated model of the target model and updated access condition information; The model storage network element updates the currently stored target model with the update model, and sends registration update information to the registration network element according to the model update request, so that the registration network element updates the associated information of the target model according to the registration update information, wherein the associated information includes the access condition information.

6. A model management method, comprising: The model training network element sends a model storage request to the model storage network element, wherein the model storage request includes access condition information of the model to be registered. The access condition information of the model to be registered includes at least one of the following information: Whether to allow the model to be registered to be discovered and acquired by network elements other than the current model training network element; Allowing discovery and acquisition of the address of the network element of the model to be registered; Whether to allow the model to be registered to be updated by network elements other than the current model training network element; The address of the network element that allows the model to be registered to be updated; or Allows to obtain or modify the manufacturer's ID of the model to be registered.

7. The method according to claim 6, further comprising: The model training network element sends a model discovery request to the registration network element; The model training network element sends the historical model request to the model storage network element according to the model identifier of the target model from the registration network element, wherein the registration network element determines the target model according to the model discovery request, and when it is determined according to the access condition information of the target model that the current model training network element is allowed to obtain the target model, the registration network element feeds back the model identifier of the target model to the model training network element, and the historical model request includes the model identifier of the target model; The model training network element obtains the target model fed back by the model storage network element.

8. The method according to claim 7, further comprising: The model training network element obtains the update condition information of the target model from the registration network element, wherein the registration network element feeds back the update condition information of the target model to the current model training network element according to the access condition information after receiving the model discovery request; In a case where the update condition information includes allowing update by the current model training network element, sending a model update request to the model storage network element so that the model storage network element updates the associated information of the target model; In a case where the update condition information includes that the model is not allowed to be updated by the current model training network element, a model registration request is sent to the model storage network element, and the model identifier in the model registration request is different from the model identifier of the target model.

9. A model management method, comprising: The registration network element obtains a model registration request from the model storage network element and stores information carried in the model registration request, wherein the model registration request includes the model identifier and access condition information of the model to be registered, and the model storage network element sends the model registration request to the registration network element when obtaining the model storage request of the model to be registered from the first model training network element; storing the information carried in the model registration request, The access condition information of the model to be registered includes at least one of the following information: Whether to allow the model to be registered to be discovered and acquired by network elements other than the first model training network element; Allowing discovery and acquisition of the address of the network element of the model to be registered; Whether to allow the model to be registered to be updated by network elements other than the first model training network element; Allowing update of the address of the network element of the model to be registered; or Allows to obtain or modify the manufacturer's ID of the model to be registered.

10. The method according to claim 9, further comprising: The registration network element receives a model discovery request from a second model training network element; The registration network element determines the target model according to the model discovery request; When it is determined that the second model training network element is allowed to obtain the target model based on the access condition information of the target model, the model identifier of the target model is fed back to the second model training network element so that the second model training network element requests the target model from the model storage network element.

11. The method according to claim 9, further comprising: Obtaining a registration update request from a model storage network element; The associated information of the model corresponding to the registration update request is updated according to the registration update request, wherein the associated information includes access condition information of the updated model.

12. The method according to claim 10, further comprising: The registration network element feeds back the update condition information of the target model to the second model training network element according to the access condition information, so that the second model training network element determines the update strategy of the target model in the model storage network element according to the update condition information. Wherein, when the update condition information includes not allowing update by the second model training network element, the update strategy includes adding a model to the model storage network element; when the update condition information includes allowing update by the second model training network element, the update strategy includes updating the target model in the model storage network element.

13. A model storage network element, comprising: a request acquisition unit, configured to acquire a model storage request from a first model training network element, wherein the model storage request includes access condition information of the model to be registered, The access condition information of the model to be registered includes at least one of the following information: Whether to allow the model to be registered to be discovered and acquired by network elements other than the first model training network element; Allowing discovery and acquisition of the address of the network element of the model to be registered; Whether to allow the model to be registered to be updated by network elements other than the first model training network element; Allowing update of the address of the network element of the model to be registered; or Allows to obtain or modify the manufacturer's ID of the model to be registered.

14. The model storage network element according to claim 13, wherein: The request acquisition unit is further configured to acquire the model to be registered from the first model training network element; The model storage network element also includes: The model storage unit is configured to store the model to be registered.

15. The model storage network element according to claim 13, wherein: The request acquisition unit is further configured to acquire a historical model request from a second model training network element, wherein the second model training network element sends the historical model request according to the model identifier of the target model from the registration network element, and the registration network element feeds back the model identifier of the target model to the second model training network element when determining according to the access condition information of the target model that the second model training network element is allowed to acquire the target model, and the historical model request includes the model identifier of the target model; The model storage network element also includes: The model feedback unit is configured to determine the target model according to the model identifier and feed back the target model to the second model training network element.

16. A model training network element, comprising: A request sending unit, configured to send a model storage request to a model storage network element, wherein the model storage request includes access condition information of the model to be registered; The access condition information of the model to be registered includes at least one of the following information: Whether to allow the model to be registered to be discovered and acquired by network elements other than the current model training network element; Allowing discovery and acquisition of the address of the network element of the model to be registered; Whether to allow the model to be registered to be updated by network elements other than the current model training network element; The address of the network element that allows the model to be registered to be updated; or Allows to obtain or modify the manufacturer's ID of the model to be registered.

17. The model training network element according to claim 16, wherein: The request sending unit is further configured to send a model discovery request to a registration network element, and send the historical model request to a model storage network element according to a model identifier of a target model from the registration network element; The model training network element also includes: a model acquisition unit configured to acquire the target model fed back by the model storage network element.

18. A registration network element, comprising: A request receiving unit is configured to obtain a model registration request from a model storage network element, wherein the model registration request includes the model identifier and access condition information of the model to be registered, and the model storage network element stores the model to be registered and sends the model registration request to the registration network element when obtaining the model storage request of the model to be registered from the first model training network element; a storage unit configured to store the information carried in the model registration request, The access condition information of the model to be registered includes at least one of the following information: Whether to allow the model to be registered to be discovered and acquired by network elements other than the first model training network element; Allowing discovery and acquisition of the address of the network element of the model to be registered; Whether to allow the model to be registered to be updated by network elements other than the first model training network element; The address of the network element that allows the model to be registered to be updated; or Allows to obtain or modify the manufacturer's ID of the model to be registered.

19. The registration network element according to claim 18, wherein: The request receiving unit is further configured to receive a model discovery request from a second model training network element; The registration network element also includes: a model determination unit, configured to determine a target model according to the model discovery request; and The information feedback unit is configured to feedback the model identifier of the target model to the second model training network element when it is determined that the second model training network element is allowed to obtain the target model based on the access condition information of the target model, so that the second model training network element requests the target model from the model storage network element.

20. A network element, comprising: Memory; as well as A processor coupled to the memory, the processor being configured to execute the method according to any one of claims 1 to 12 based on instructions stored in the memory.

21. A computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 12.

22. A network system comprising: A plurality of model training network elements, each of which is configured to execute the method according to any one of claims 6 to 8; A model storage network element, configured to execute the method according to any one of claims 1 to 5; and The registration network element is configured to execute the method according to any one of claims 9 to 12.

Images