Model management method and apparatus, and computer device

Abstract

The present application relates to a model management method and apparatus, and a computer device. The method comprises: receiving a first configuration message sent by a base station, wherein the first configuration message includes user scheduling result indication information based on hierarchical federated learning, which user scheduling result indication information is obtained by the base station on the basis of user characteristic information of each terminal device, the user characteristic information comprising at least one of user data characteristic information and user channel state information; sending report information to a network, wherein the network stores a hierarchical federated learning model and a delivery policy, which have been generated by the base station, the report information at least comprising at least one of a terminal capability, an inference task type and model performance monitoring feedback information, and the report information being used for the network to generate a second configuration message, which is model inference policy delivery information or model update policy delivery information based on hierarchical federated learning; and on the basis of the second configuration information, obtaining a model inference result or a model update result.

Description

Model management method and device and computer device

[0001] Related applications

[0002] The present application claims priority to the Chinese patent application No. 2024113442757, filed on September 25, 2024, and entitled "Model management method and device based on hierarchical federated learning and computer device", the contents of which are hereby incorporated by reference in its entirety. TECHNICAL FIELD

[0003] The present application relates to the technical field of federated learning, and in particular to a model management method, device and computer device. BACKGROUND

[0004] With the rapid development of 3GPP radio access network (RAN) technology, the powerful representation and feature extraction capabilities of AI models can be combined with wireless communication scenarios to achieve time-frequency domain resource optimization and cost reduction. Among them, federated learning as a distributed framework can effectively improve the training efficiency of AI models, reduce transmission costs while protecting the privacy and security of local data, SUMMARY

[0005] In a first aspect, the present application provides a model management method applied to a terminal device in a mobile communication system, the mobile communication system further comprising a network, the network comprising a network node and a plurality of base stations, each base station corresponding to a plurality of terminal devices, the terminal device being a distributed node for performing hierarchical federated learning, and the base station being a central node for performing hierarchical federated learning; the method comprising:

[0006] receiving a first configuration message sent by the base station, the first configuration message containing user scheduling result indication information based on hierarchical federated learning, the user scheduling result indication information being obtained by the base station based on user feature information of each terminal device, the user feature information including at least one of user data feature information and user channel state information;

[0007] sending reporting information to the network, the network storing a hierarchical federated learning model generated by the base station and a downlink strategy, the reporting information including at least one of terminal capability, inference task type, model performance monitoring feedback information, the reporting information being used by the network to generate a second configuration message, the second configuration message being model inference strategy downlink information or model update strategy downlink information based on hierarchical federated learning;

[0008] receiving the second configuration message, and obtaining a model inference result or a model update result based on the second configuration message.

[0009] In one of the embodiments, the terminal capability includes one or more of a terminal computing capability, a terminal channel state, and an inference feature, the terminal computing capability is used to represent a model complexity supported by the terminal device, and the terminal channel state represents an uplink channel state of the terminal device.

[0010] In one of the embodiments, the sending of the reporting information to the network includes:

[0011] Based on a target transmission mode, the reporting information is sent to the base station, so that the base station sends the reporting information to the network node, and the target transmission mode includes at least one of a system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

[0012] In one of the embodiments, the model performance monitoring feedback information includes one or more of a system performance index, an accuracy index, a complexity index, and a generalization performance index.

[0013] In one of the embodiments, the method further includes:

[0014] If it is determined that the model performance monitoring feedback information meets a preset update condition with a preset index threshold, a model update request is sent to the base station, so that the network responds to the model update request, and the model update request includes one or more of a to-be-updated model, a to-be-updated function identifier, and reference information.

[0015] In one of the embodiments, the receiving of the second configuration message includes:

[0016] The second configuration message sent by the base station based on a target transmission mode is received, the second configuration message is sent by the network node to the base station, and the target transmission mode includes at least one of a system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

[0017] In one of the embodiments, the second configuration message is model update strategy issuing information, the model update strategy issuing information includes a to-be-updated model and an update strategy, and the obtaining of the model inference result or the model update result based on the second configuration message includes:

[0018] Based on the update strategy, the to-be-updated model is updated to obtain a model update result, the updated model includes a to-be-updated model / to-be-updated model group, and the update strategy includes one or more of a local update strategy, a global update strategy, a model replacement strategy, and a model deletion strategy.

[0019] In one of the embodiments, the updating process includes at least one of the following:

[0020] updating part of the model to obtain a model update result;

[0021] updating the model parameters of each of the to-be-updated models based on a target update algorithm to obtain a model update result, the target update algorithm including one or more of a momentum gradient descent algorithm, a stochastic gradient descent algorithm, and a mini-batch gradient descent algorithm;

[0022] updating the model parameters of each of the to-be-updated models based on the new model parameters to obtain a model update result;

[0023] deleting the model parameters and model content / feature identifiers of the updated model to obtain a model update result.

[0024] In one of the embodiments, the second configuration message is model inference strategy issuing information, the model inference strategy issuing information including a model identifier and a model file of a to-be-inferred model and identifier information of a terminal device to be executed for model inference; and obtaining a model inference result or a model update result based on the second configuration message includes:

[0025] performing online mode model inference or offline mode model inference based on the model identifier and the model file of the to-be-inferred model and the identifier information of the terminal device to be executed for model inference to obtain a model inference result.

[0026] In a second aspect, the embodiments of the present application provide a model management method applied to a network node in a mobile communication system, the network including a network node and a plurality of base stations, each of the base stations corresponding to a plurality of terminal devices, and the method including:

[0027] receiving report information sent by the terminal devices through the base stations, the report information including at least one of terminal capability, inference task type, and model performance monitoring feedback information;

[0028] generating a second configuration message based on the report information and sending the second configuration message to the terminal devices through the base stations, the second configuration message being model inference strategy issuing information or model update strategy issuing information based on hierarchical federated learning, and the second configuration message being used to make the terminal devices obtain a model inference result or a model update result.

[0029] In one of the embodiments, the terminal capability includes one or more of terminal computing capability, terminal channel state, and inference feature, the terminal computing capability being used to represent a model complexity supported by the terminal device, and the terminal channel state representing an uplink channel state on the terminal device side.

[0030] In one of the embodiments, the reporting information includes the terminal capability, the receiving the reporting information sent by the terminal device through the base station includes:

[0031] Based on the target transmission mode, receiving the terminal capability reported by the terminal device through the base station, and the target transmission mode includes at least one of system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

[0032] In one of the embodiments, the method further includes:

[0033] Receiving the model inference request signaling sent by the terminal device through the base station, and returning the response message corresponding to the inference request, so that the terminal device sends the reporting information based on the response message.

[0034] In one of the embodiments, the second configuration message includes model inference strategy issuing information or model update strategy issuing information based on hierarchical federated learning; and the generating the second configuration message based on the reporting information includes:

[0035] Based on the reporting information, determining at least one to-be-inferred model in a plurality of first models that meets a preset matching condition in a target dimension;

[0036] Based on the model identifier of the to-be-inferred model, the model file, and the identifier information of the terminal device to be executed for model inference, generating model inference strategy issuing information based on hierarchical federated learning.

[0037] In one of the embodiments, the determining at least one to-be-inferred model in a plurality of first models that meets a preset matching condition in a target dimension based on the reporting information includes:

[0038] Based on the reporting information, determining a preset number of to-be-inferred models in a plurality of first models that meet a preset matching condition in model structure, model complexity, and model accuracy.

[0039] In one of the embodiments, the sending the second configuration message to the terminal device through the base station includes:

[0040] Sending the second configuration message to the base station through a user plane interface, so that the base station sends the second configuration message to the terminal device through the target transmission mode, and the target transmission mode includes at least one of system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

[0041] In one of the embodiments, the model performance monitoring feedback information includes one or more of system performance indicators, accuracy indicators, complexity indicators, and generalization performance indicators.

[0042] In one of the embodiments, the reporting information comprises model performance monitoring feedback information, and the generating the second configuration message based on the reporting information comprises:

[0043] receiving the model performance monitoring feedback information;

[0044] receiving a model update request, the model update request comprising a to-be-updated model / to-be-updated function identifier and reference information, the model update request being generated in a case where the model performance monitoring feedback information of the terminal device and a preset index threshold satisfy a preset update condition;

[0045] responding to the model update request to generate the second configuration message.

[0046] In one of the embodiments, the reporting information comprises model performance monitoring feedback information, and the generating the second configuration message based on the reporting information comprises:

[0047] receiving the model performance monitoring feedback information;

[0048] triggering a model update event based on a preset time period, or triggering an update event if the model performance monitoring feedback information and a preset index threshold satisfy a preset update condition, the preset time period being associated with terminal capability and auxiliary information;

[0049] generating the second configuration message based on the triggered update event.

[0050] In one of the embodiments, the second configuration message comprises model inference strategy distribution information based on hierarchical federated learning, and the generating the second configuration message comprises:

[0051] determining at least one to-be-updated model;

[0052] generating model inference strategy distribution information based on hierarchical federated learning based on the to-be-updated model, identifier information of a terminal device to be executed for model update, and an update strategy.

[0053] In one of the embodiments, the update strategy comprises one or more of a local update strategy, a global update strategy, a model replacement strategy, and a model deletion strategy.

[0054] In one of the embodiments, the method further comprises:

[0055] receiving model data sent by each of the base stations, the model data comprising a model file and model content / function identifier;

[0056] Determine each first model based on each model data, and arrange each first model according to an evaluation index of each first model to obtain a first model list.

[0057] In one of the embodiments, the method further comprises:

[0058] Receive a model capability and an identification request sent by the base station, wherein the identification request comprises a model identification request and a function identification request, and the model capability is used to indicate a model supported by the base station or a function supported by the base station.

[0059] In one of the embodiments, the method further comprises:

[0060] Send an identification request to each base station to make each base station report a model capability, wherein the model capability is used to indicate a model supported by the base station or a function supported by the base station.

[0061] In a third aspect, the embodiments of the present application provide a model management method based on hierarchical federated learning, applied to a base station in a mobile communication system, wherein the mobile communication system comprises a network and a plurality of base stations, each base station corresponds to a plurality of terminal devices, the terminal devices are distributed nodes performing hierarchical federated learning, and the base station is a central node performing hierarchical federated learning; the method comprises:

[0062] Obtain user scheduling result indication information corresponding to each terminal device based on user feature information of each terminal device, and send a first configuration message to each terminal device, wherein the user feature information comprises at least one of user data feature information and user channel state information, and the first configuration message comprises the user scheduling result indication information.

[0063] Receive report information sent by the terminal device, wherein the report information comprises at least one of terminal capability, inference task type, and model performance monitoring feedback information.

[0064] Send the report information to the network node, so that the network node obtains a second configuration message; the second configuration message is model inference strategy issuing information or model update strategy issuing information based on hierarchical federated learning, and the second configuration message is used to make the terminal device obtain a model inference result or a model update result.

[0065] In one of the embodiments, the method further comprises:

[0066] Send model data to the network node, wherein the model data comprises a model file and a model content / function identifier, so that the network node determines each first model based on each model data and obtains a first model list.

[0067] In one of the embodiments, the method further comprises:

[0068] sending an identification request and a model capability to the network node, the identification request comprising a model identification request and a function identification request, and the model capability being used to indicate a model / function supported by the base station.

[0069] In one of the embodiments, the method further comprises:

[0070] receiving an identification request sent by the network node, and reporting a model capability to the network node based on the identification request, the model capability being used to indicate a model / function supported by the base station.

[0071] In one of the embodiments, the method further comprises:

[0072] obtaining a global model trained based on the model training results sent by the terminal devices.

[0073] In one of the embodiments, the terminal device is a terminal device of a scheduled user, and the obtaining of the global model trained based on the model training results sent by the terminal devices comprises:

[0074] for the i-th round of training, obtaining model parameters of the global model of the i-th round based on the training results sent by the terminal devices of the (i-1)-th round;

[0075] in a case where a preset training completion condition is not met, broadcasting the model parameters of the global model of the i-th round to the terminal devices, so that the terminal devices train based on the model parameters of the global model of the i-th round until the base station determines that the preset training completion condition is met, and obtains the global model trained.

[0076] In one of the embodiments, the preset training completion condition comprises one or more of the following: a number of training rounds reaches a preset upper limit value, a loss function of the global model is less than or equal to a preset value, and a difference between loss functions of adjacent two rounds is less than or equal to a preset value.

[0077] In one of the embodiments, the method further comprises:

[0078] obtaining user scheduling result indication information corresponding to each of the terminal devices based on user feature information of each of the terminal devices, the user feature information comprising at least one of user data feature information and user channel state information;

[0079] sending a first configuration message to each of the terminal devices, the first configuration message containing the user scheduling result indication information.

[0080] In one of the embodiments, the user scheduling result indication information corresponding to each terminal is obtained based on the user feature information of each terminal device, and the user scheduling result indication information includes:

[0081] obtaining a pre-configured number of scheduled users;

[0082] In all users within the coverage range of the base station, based on at least one of the user data feature information and the user channel state information of each terminal device, the terminal devices of the number of scheduled users are screened out, and the user scheduling result indication information is determined as the scheduled indication information.

[0083] In a fourth aspect, the embodiments of the present application provide a model inference device applied to a terminal device in a mobile communication system, the mobile communication system further comprising a network, the network comprising a network node and a plurality of base stations, each of the base stations corresponding to a plurality of terminal devices, the terminal devices being distributed nodes performing hierarchical federated learning, and the base stations being central nodes performing hierarchical federated learning; the device comprising:

[0084] a first receiving module configured to receive a first configuration message sent by the base station, the first configuration message containing user scheduling result indication information based on hierarchical federated learning, the user scheduling result indication information being obtained by the base station based on user feature information of each terminal device, the user feature information including at least one of user data feature information and user channel state information;

[0085] a first sending module configured to send reporting information to the network, the network storing a hierarchical federated learning model generated by the base station and a distribution policy, the reporting information including at least one of terminal capability, inference task type, and model performance monitoring feedback information, the reporting information being used by the network to generate a second configuration message, the second configuration message being model inference policy distribution information or model update policy distribution information based on hierarchical federated learning;

[0086] a second receiving module configured to receive the second configuration message and obtain a model inference result or a model update result based on the second configuration message.

[0087] In a fifth aspect, the embodiments of the present application provide a model management device applied to a network node in a mobile communication system, the network comprising a network node and a plurality of base stations, each of the base stations corresponding to a plurality of terminal devices, the device comprising:

[0088] a third receiving module configured to receive reporting information sent by the terminal device through the base station, the reporting information including at least one of terminal capability, inference task type, and model performance monitoring feedback information;

[0089] The first generating module is configured to generate a second configuration message based on the reporting information, and send the second configuration message to the terminal device through the base station; the second configuration message is model inference strategy issuing information or model update strategy issuing information based on hierarchical federated learning, and the second configuration message is used to make the terminal device obtain a model inference result or a model update result.

[0090] In a sixth aspect, an embodiment of the present application provides a model management device applied to a base station in a mobile communication system, the mobile communication system comprising a network and a plurality of base stations, each of the base stations corresponding to a plurality of terminal devices, the terminal devices being distributed nodes performing hierarchical federated learning, and the base station being a central node performing hierarchical federated learning; the device comprises:

[0091] The second sending module is configured to obtain user scheduling result indication information corresponding to each of the terminal devices based on user feature information of each of the terminal devices, and send a first configuration message to each of the terminal devices, wherein the user feature information comprises at least one of user data feature information and user channel state information, and the first configuration message comprises the user scheduling result indication information.

[0092] The fourth receiving module is configured to receive reporting information sent by the terminal device, wherein the reporting information comprises at least one of terminal capability, inference task type, and model performance monitoring feedback information.

[0093] The third sending module is configured to send the reporting information to the network node, so that the network node obtains a second configuration message; the second configuration message is model inference strategy issuing information or model update strategy issuing information based on hierarchical federated learning, and the second configuration message is used to make the terminal device obtain a model inference result or a model update result.

[0094] In a seventh aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above embodiments when executing the computer program.

[0095] In an eighth aspect, the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the above embodiments.

[0096] In a ninth aspect, the present application further provides a computer program product comprising a computer program, and the computer program is executed by a processor to implement the steps of the above embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0097] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the following will briefly introduce the drawings needed to be used in the description of the embodiments of the present application or the related art. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other related drawings can also be obtained without creative labor.

[0098] Fig. 1 is an application environment diagram of a model management method based on hierarchical federated learning in an embodiment;

[0099] Fig. 2 is a flowchart of a model management method based on hierarchical federated learning in an embodiment;

[0100] Fig. 3 is a flowchart of a step of obtaining a model update result in an embodiment;

[0101] Fig. 4 is a flowchart of a step of sending in an embodiment;

[0102] Fig. 5 is a flowchart of a step of generating a second configuration message in an embodiment;

[0103] Fig. 6 is a flowchart of a step of generating a second configuration message in an embodiment;

[0104] Fig. 7 is a flowchart of a step of generating a second configuration message in an embodiment;

[0105] Fig. 8 is a flowchart of a step of obtaining a model list in an embodiment;

[0106] Fig. 9 is a flowchart of a step of reporting information transmission in an embodiment;

[0107] Fig. 10 is a flowchart of a step of training in an embodiment;

[0108] Fig. 11a is a flowchart of a model management method based on hierarchical federated learning in another embodiment;

[0109] Fig. 11b is a curve diagram of positioning accuracy of a scheme in an embodiment;

[0110] Fig. 11c is a diagram of a positioning error change trend of a federated learning algorithm in an embodiment;

[0111] Fig. 11d is a CDF curve diagram of a positioning error in an embodiment;

[0112] Fig. 12 is a structural block diagram of a model inference device based on hierarchical federated learning in an embodiment;

[0113] Fig. 13 is a structural block diagram of a model inference device based on hierarchical federated learning in an embodiment;

[0114] FIG. 14 is a structural block diagram of a model inference device based on hierarchical federated learning in an embodiment;

[0115] FIG. 15 is an internal structure diagram of a computer device in an embodiment. DETAILED DESCRIPTION

[0116] For the purpose, technical solutions and advantages of the present application to be clearer, further detailed description will be made to the present application in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0117] Based on how to realize the application of federated learning in 5G-A / 6G network is a technical problem to be solved, the present application provides a model management method, device, computer device, computer readable storage medium and computer program product capable of improving the coverage range of the algorithm and enriching the applicable scenarios of the generated federated learning model by using cloud-edge-end collaborative network architecture.

[0118] The model management method provided by the embodiments of the present application can be applied to a mobile communication system in an application environment as shown in FIG. 1. The model management method provided by the embodiments of the present application is, for example, a model management method based on hierarchical federated learning. The mobile communication system includes a network, and the network includes a network node and a plurality of base stations. Each base station corresponds to a plurality of terminal devices, each terminal device is a distributed node for performing hierarchical federated learning, and each base station is a central node for performing hierarchical federated learning. The network node can communicate with each base station, and the base station can communicate with the terminal device corresponding to the base station. The mobile communication system can be a distributed network architecture, the UE (User Equipment / terminal device) can be a terminal node, a terminal device or a local client, the base station can be an edge node or a server end, and specifically can be a CU / BBU under a gNB / CU-DU separation architecture, and the network node can be a cloud end, and specifically can be an LMF / OAM (Orchestration and Management) / network data analysis function NWDAF (NW side). The terminal device can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers and the like.

[0119] In an exemplary embodiment, as shown in FIG. 2, a model management method based on hierarchical federated learning is provided. Taking the terminal device in the mobile communication system in FIG. 1 as an example, the mobile communication system further includes a network, and the network includes a network node and a plurality of base stations. Each base station corresponds to a plurality of terminal devices, the terminal device is a distributed node for performing hierarchical federated learning, and the base station is a central node for performing hierarchical federated learning. The method includes:

[0120] Step 202, receiving a first configuration message sent by a base station.

[0121] The first configuration message contains user scheduling result indication information based on hierarchical federated learning, and the user scheduling result indication information is obtained by the base station based on user feature information of each terminal device, and the user feature information includes at least one of user data feature information and user channel state information. The user data feature information can be the feature information of the local update gradient of the terminal device, and the user channel state information can be information obtained based on user CSI-RS (Channel State Information Reference Signal) channel estimation feedback.

[0122] Specifically, the base station can determine all terminal devices within the coverage of the base station, and select each terminal device based on hierarchical federated learning among all terminal devices, and obtain the first configuration message based on the user scheduling indication information. The base station can send the first configuration message to each selected terminal device, and the user scheduling result indication information contained in the first configuration message can represent that the terminal device is selected by the base station.

[0123] In one example, the specific process of scheduling by the base station can include: the base station determines a first number of terminal devices that need to be scheduled (used) at present, and selects each terminal device based on user feature information of the terminal device among all terminal devices within the coverage of the base station, extracts the first number of terminal devices, and determines that the user scheduling result indication information of the first number of terminal devices is first indication information, which can represent that the terminal device is selected by the base station; the user scheduling indication information of other terminal devices can be empty or second indication information, which is used to represent that the terminal device is not selected by the base station. The specific form of the first indication information and the second indication information is not limited in the present disclosure, and can be determined by the person skilled in the art based on the actual application scenario.

[0124] Step 204, sending reporting information to a network.

[0125] The network stores the hierarchical federated learning model generated by the base station and the issued policy, and the reporting information includes at least one of terminal capability, inference task type, and model performance monitoring feedback information. The reporting information is used by the network to generate a second configuration message, which is model inference policy issuing information or model update policy issuing information based on hierarchical federated learning.

[0126] Specifically, the terminal device can send the reporting information to the network, and the network can generate the second configuration message based on the reporting information after receiving the reporting information. Based on this, the network can send the second configuration message to each terminal device. In the model inference process, the network sends the second configuration message to each terminal, which can contain the model inference strategy issuing information based on the hierarchical federated learning. In the model update process, the network sends the second configuration message to each terminal device, which can contain the model update strategy issuing information based on the hierarchical federated learning.

[0127] Step 206, receiving the second configuration message, and obtaining the model inference result or the model update result based on the second configuration message.

[0128] Specifically, the terminal device can receive the second configuration message, and perform model inference based on the model inference strategy issuing information in the second configuration message to obtain the model inference result. The terminal device can also update the model based on the model update strategy issuing information in the second configuration message to obtain the model update result.

[0129] In the above model management method based on hierarchical federated learning, the first configuration message sent by the base station is received, and the first configuration message contains user scheduling result indication information based on hierarchical federated learning. The user scheduling result indication information is obtained by the base station based on user feature information of each terminal device, and the user feature information includes at least one of user data feature information and user channel state information. The reporting information is sent to the network, and the network stores the hierarchical federated learning model and the issuing strategy generated by the base station. The reporting information includes at least one of terminal capability, inference task type, and model performance monitoring feedback information. The reporting information is used for the network to generate the second configuration message, which is model inference strategy issuing information or model update strategy issuing information based on hierarchical federated learning. The second configuration message is received, and the model inference result or the model update result is obtained based on the second configuration message. By using this method, data interaction between the base station and the terminal can be realized, the management efficiency and the generalization performance of the wireless intelligent federated learning model can be improved, the additional consumption of communication resources and the calculation overhead can be reduced, the to-be-allocated resources can be flexibly configured, and the terminal devices performing federated learning can be flexibly scheduled, and the application scenarios of the model are further expanded.

[0130] In one of the embodiments, the terminal capability includes one or more of terminal computing capability, terminal channel state, and inference feature, the terminal computing capability is used to represent the model complexity supported by the terminal device, and the terminal channel state represents the uplink channel state of the terminal device.

[0131] Specifically, the terminal computing capability is an index representing the strength of the computing capability of the terminal device, and the terminal computing capability can be used to determine the model complexity supported by the terminal device. The terminal computing capability of the terminal device can be in a positive correlation with the model complexity supported by the terminal device, that is, the stronger the terminal computing capability, the higher the model complexity that can be supported. The terminal computing capability can include the CPU frequency, thread number, core number of the terminal device, the bit width, and the display memory size of the GPU, and the like. The terminal channel state is used to represent the uplink channel state (CSI). The inference feature can be an intelligent feature used to represent the local data of the terminal device. For example, the inference feature can include the hyperparameters such as the number of dataset features, the number of samples, the dimension, the data type, the learning rate, and the data batch. The data type is used to represent whether it is a time series. The terminal computing capability and the inference feature can be optional items of the terminal device for sending the reporting information to the network. The terminal device can determine whether the terminal computing capability and the inference feature are included in the reporting information based on the actual application scenario. Alternatively, the terminal computing capability and the inference feature can be determined based on the network side configuration.

[0132] Optionally, the reporting information can include the terminal capability, and the reporting information can further include the inference task type, the auxiliary information, and the like. The inference task type can represent the type of the model inference task, that is, the type of the model or the type of the model inference result. The type can include one or more of a classification task, a prediction task, and a clustering task. The model used for performing the model inference can include one or more of a classification model, a prediction model, and a clustering model.

[0133] In this embodiment, the terminal capability carried in the reporting information can include multiple types, and the specific content of the terminal capability can be determined based on the actual needs of the terminal device, to ensure the flexibility of the information represented by the terminal capability.

[0134] In one of the embodiments, the specific implementation process of the step of “sending the reporting information to the network” can include:

[0135] Based on the target transmission mode, the reporting information is sent to the base station, so that the base station sends the reporting information to the network node.

[0136] The target transmission mode includes at least one of a system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

[0137] Specifically, the reporting information sent by the terminal device to the network can be terminal device capability reporting signaling. The terminal device can report the terminal device capability reporting signaling to the base station (edge node) through at least one of the following transmission modes: system message, RRC (Radio Resource Control) signaling, MAC CE (Media Access Control Control Element) signaling, DCI (Downlink Control Information) signaling, and NAS (Non Access Stratum) message, that is, through at least one of the following transmission modes: system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message, to the edge node. After receiving the reporting information, the edge node can send the reporting information to the network node corresponding to the edge node. For example, the edge node can send the reporting information to the network node through an NG-C interface.

[0138] In this embodiment, the terminal device can transmit data between the base station through multiple transmission modes, ensuring the diversity of data transmission and the reliability of data transmission.

[0139] In one of the embodiments, the model performance monitoring feedback information includes one or more of the following: system performance indicators, accuracy indicators, complexity indicators, and generalization performance indicators.

[0140] Specifically, the model performance monitoring feedback information can include multiple indicators. The system performance indicators can include one or more of the following: throughput, block error rate, and latency. The accuracy indicators can include one or more of the following: accuracy of model output, MSE / NMSE / cosine similarity between model input and output, and the like. The complexity indicators can include model size and computational complexity. The model size refers to the number of parameters included in the model. The computational complexity can be FLOPs (Floating Point Operations Per second). The model performance monitoring feedback information can also include generalization performance indicators, the specific content of which can be determined based on actual application scenarios.

[0141] Optionally, after the UE performs model inference, the model performance monitoring can be performed in real time to obtain a model performance monitoring result, and the model performance monitoring feedback information can be obtained based on the result. The UE can feed back the model performance monitoring feedback information to the edge node corresponding to the terminal device. After receiving the model performance monitoring feedback information, the edge node can send the model performance monitoring feedback information to the network node (NW side) corresponding to the edge node.

[0142] In the embodiment, the performance of the model can be monitored in real time by monitoring the performance indicators in the model inference process, and the timeliness of the model can be adjusted in time.

[0143] In one of the embodiments, the model management method based on the hierarchical federated learning further includes:

[0144] If it is determined that the model performance monitoring feedback information meets the preset update condition with the preset indicator threshold, a model update request is sent to the base station, so that the network responds to the model update request.

[0145] The model update request includes one or more of the to-be-updated model, the to-be-updated function identifier, and the reference information; the preset update condition is used to determine whether to trigger a model update event; if it is determined that the preset update condition is met, the update event is triggered; if it is determined that the preset update condition is not met, the update event is not triggered.

[0146] Specifically, the terminal device can obtain model performance monitoring feedback information based on the model performance monitoring result, for example, the model performance monitoring result can be used as the model performance monitoring feedback information; the terminal device can determine the model performance monitoring result as the reporting information, and send the reporting information to the base station to which the terminal device belongs; the base station can send the reporting information to the network node corresponding to the base station.

[0147] After the base station sends the reporting information to the network node corresponding to the base station, the model update process can be triggered, in one example, the triggering process of the model update can be triggered by the terminal device, the terminal device can compare the indicators in the model performance monitoring feedback information with the preset indicator threshold corresponding to the indicators, if it is determined that there is an indicator that meets the preset update condition, the terminal device can trigger a model update event, that is, the terminal device can generate a model update request and send the model update request to the base station corresponding to the terminal device; after receiving the model update request, the base station can report the model update request to the network node, the network node can respond to the model update request, the process of responding can include: the network node determines at least one to-be-updated model to be executed for model update, generates a model update strategy issuing information based on the at least one to-be-updated model, that is, obtains a second configuration message, the network node can send the model update strategy issuing information to the terminal device to be executed for model update through the base station, the terminal device can execute model update based on the model update strategy issuing information, and obtain a model update result, the specific update process will be described in detail in the following embodiments.

[0148] In another example, the triggering process of the model update can be actively triggered by the network side. After receiving the reported information, the network node can trigger the model update event based on the target triggering condition. The specific process of the target triggering condition can include periodic model update and aperiodic model update. In one possible implementation, the network node can trigger the update event of the to-be-updated model / to-be-updated model group according to a preset time period. The preset time period can be associated with the terminal capability or the change of the assistance information, such as scene, site, channel, or data sampling speed, etc. In another possible triggering manner, the network node can trigger the model update event based on the received model performance monitoring feedback information. Specifically, the network node can compare the index included in the model performance monitoring feedback information with a preset index threshold corresponding to the index. If it is determined that the index and the preset index threshold satisfy a preset update condition, the network node can trigger the model update event.

[0149] Optionally, the specific process of determining whether the preset update condition is met by the network node / UE can be that, in the case that the index included in the model performance monitoring feedback information is the time delay, if the network node / UE determines that the time delay is greater than or equal to a preset time delay threshold corresponding to the time delay, the network node / UE can determine that the preset update condition is currently met, and trigger the model update event. In the case that the index included in the model performance monitoring feedback information is the model accuracy, if the network node / UE determines that the model accuracy is less than or equal to a preset accuracy threshold corresponding to the model accuracy, the network node / UE can determine that the preset update condition is currently met, and trigger the model update event. The network node / UE can determine whether the preset update condition is met based on the meaning of the index, and the relationship between the index and the preset index threshold corresponding to the index.

[0150] In one example, the terminal device can report the model performance monitoring feedback information to the base station to which the terminal device belongs. After receiving the model performance monitoring feedback information, the base station can send the model performance monitoring feedback information to the network node (NW side) corresponding to the base station. The model update process is triggered by the terminal device side or the network side.

[0151] In one example, the network node can configure the preset index threshold through a target transmission manner, and the target transmission manner includes at least one of a system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message. The network node can send an index threshold configuration message to the terminal device through the base station. The index threshold configuration message at least includes the specific value of the index threshold and the index identifier. The index threshold configuration message can be at least one of a system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

[0152] Optionally, the model update request sent by the terminal device specifically includes a model to be updated, a function identifier to be updated, and reference information, which can be model update reference information fed back by the terminal device, such as one or more of gradient reference information, loss function reference information, and a log.

[0153] Optionally, the base station to which the terminal device belongs can be within the coverage of the terminal device, the base station serving as a center node of hierarchical federated learning, and each terminal device within the coverage of the base station serving as a distributed node of hierarchical federated learning. The network node corresponding to the base station can be a network node for managing each base station.

[0154] In this embodiment, the terminal device / network node can actively determine whether the model update process is needed based on the monitored model indicator information, improve the control degree of the terminal device / network node on the model update process, and update the model in time when the model does not meet the requirements.

[0155] In one of the embodiments, the specific implementation process of the step of "receiving the second configuration message" can include:

[0156] The second configuration message sent by the base station based on the target transmission mode is received.

[0157] The second configuration message is sent by the network node to the base station, and the target transmission mode includes at least one of a system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

[0158] Specifically, after receiving the report information uploaded by the terminal device, the network node can process the report information to obtain the second configuration message and send the second configuration message to the base station. After receiving the second configuration message, the base station can send the second configuration message to each terminal device corresponding to the base station.

[0159] In one example, the network node generates the second configuration message based on the report information reported by the terminal device, and can send the second configuration message to a base station to which the terminal device belongs, so that the base station sends the second configuration message to the terminal device; in another example, the network node can send the second configuration message to each base station managed by the network node, so that each base station can send the second configuration message to each terminal device within the range of the base station; in another example, the network node can screen each base station managed by the network node, and send the second configuration message to the screened base station, each base station can perform scheduling of the terminal device based on the method described in the above embodiment, obtain the scheduled terminal device, and send the second configuration message to each scheduled terminal device, wherein each base station can also send the second configuration message to the terminal device corresponding to the identifier of the terminal device based on the identifier of the terminal device carried in the second configuration message.

[0160] In the embodiment, the network node, the base station, and the terminal device can realize data transmission through various transmission methods, ensuring the reliability, diversity, and flexibility of data transmission.

[0161] In one of the embodiments, the second configuration message is model update policy distribution information, and the model update policy distribution information includes a to-be-updated model and an update policy. Specifically, in the model updating process, the second configuration message generated by the network node can include the model update policy distribution information; in the model inference process, the second configuration message generated by the network node can include model inference policy distribution information.

[0162] In one example, the second configuration message is model update policy distribution information; the process of generating the second configuration message by the network node can be that the network node determines at least one to-be-updated model to be executed for model updating based on the report information, encapsulates the to-be-updated model, a user identifier of a terminal device to be executed for model updating, and an update policy to obtain the model update policy distribution information; the network node can send the model update policy distribution information to the base station through a user plane interface, and the base station can send the second configuration message to the terminal device corresponding to the user identifier based on the user identifier in the received model update policy distribution information. The user plane interface can be an NG-U interface. The to-be-updated model determined by the network node can be one to-be-updated model or a group of to-be-updated models.

[0163] Correspondingly, the specific implementation process of the step "obtaining the model inference result or the model updating result based on the second configuration message" can include:

[0164] updating the to-be-updated model based on the update policy to obtain a model updating result.

[0165] The updating model includes a to-be-updated model or a to-be-updated model group, and the updating strategy includes one or more of a local updating strategy, a global updating strategy, a model replacement strategy, and a model deletion strategy. The updating strategy can be a model updating method used by the terminal device when performing model updating.

[0166] Specifically, after receiving the second configuration message, the terminal device can obtain the model updating strategy issuing information carried in the second configuration message, and perform updating processing on the to-be-updated model based on the updating strategy in the model updating strategy issuing information, to obtain a model updating result. The model updating result can be the to-be-updated model after the updating processing. That is, the terminal device currently receiving the second configuration message is a terminal device selected by the base station based on the user identifier of the to-be-updated model carried in the second configuration message. The terminal device is a terminal device to be updated. The terminal device can perform model updating processing based on the to-be-updated model or the to-be-updated model group in the model updating strategy issuing information and the updating strategy indicated by the issuing information, to obtain a model updating result. The obtained model updating result can be an updated model or model group obtained by the terminal device after performing the model updating method selected by the network node. The model updating strategy issuing information can be model updating strategy issuing signaling.

[0167] In this embodiment, the base station can send the model updating strategy issuing information to the specific terminal device based on the indication of the network node, so that the selected terminal device performs model updating processing.

[0168] In one of the embodiments, as shown in FIG. 3, the specific implementation process of the step of “performing updating processing” can include at least one of the following:

[0169] In step 302, part of the modules of the to-be-updated model is updated to obtain a model updating result.

[0170] The part of the model can be a specific module or a module pre-selected by the network node.

[0171] Specifically, in the case where the updating strategy includes a local updating strategy, the terminal device can determine the to-be-updated model or the to-be-updated model group based on the model updating strategy issuing information, determine part of the modules in the to-be-updated model or the to-be-updated model group, and update the part of the modules in the to-be-updated model or the to-be-updated model group based on the collected data, to obtain an updated model or model group. The terminal device can also update the model parameters of the part of the modules in the to-be-updated model or the to-be-updated model group, to obtain an updated model or model group, and determine the updated model or model group as a model updating result.

[0172] At step 304, the model parameters of each to-be-updated model are updated based on a target updating algorithm to obtain a model updating result.

[0173] The target updating algorithm includes one or more of a momentum gradient descent algorithm, a stochastic gradient descent algorithm, and a mini-batch gradient descent algorithm.

[0174] Specifically, the terminal device can determine the to-be-updated model / to-be-updated model group and the global updating strategy and the target updating algorithm included in the updating strategy based on the model updating strategy distribution information, and update the collected data and all model parameters of the to-be-updated model / to-be-updated model group based on the target updating algorithm to obtain an updated model / model group. Based on this, the terminal device can determine the updated model / model group as the model updating result. The target updating algorithm included in the updating strategy includes at least one of a momentum gradient descent algorithm, a stochastic gradient descent algorithm, and a mini-batch gradient descent algorithm.

[0175] At step 306, the model parameters of each to-be-updated model are updated based on new model parameters to obtain a model updating result.

[0176] Specifically, the terminal device can determine the to-be-updated model / to-be-updated model group and the model replacement strategy included in the updating strategy based on the model updating strategy distribution information, determine new model parameters based on the model replacement strategy, and replace the model parameters of the to-be-updated model / to-be-updated model group with the new model parameters to obtain an updated model / model group. Based on this, the terminal device can determine the updated model / model group as the model updating result.

[0177] At step 308, the model parameters and model content / functional identifier of the updated model are deleted to obtain a model updating result.

[0178] Specifically, if the terminal device determines that the updating strategy is a model deletion strategy, the terminal device can determine the to-be-updated model / to-be-updated model group, delete the model parameters of the to-be-updated model / to-be-updated model group, and delete the model content / functional identifier corresponding to the to-be-updated model / to-be-updated model group to obtain a model updating result.

[0179] It should be noted that the disclosure does not limit the order and range of execution of the above steps 302, 304, 306, and 308. Based on the specific content of the updating strategy included in the second configuration message, a person skilled in the art can determine the specific steps to be executed.

[0180] In the embodiment, the network node can configure the specific process of model updating performed on the terminal device through transmission of the second configuration message, so as to ensure the degree of information synchronization between the network node and the terminal device.

[0181] In one of the embodiments, the second configuration message is model inference strategy distribution information, and the model inference strategy distribution information includes a model identifier of a model to be inferred, a model file, and identifier information of a terminal device to be executed for model inference. Specifically, in the model inference process, the model inference strategy distribution information can be included in the second configuration message generated by the network node.

[0182] In one example, the second configuration message is model inference strategy distribution information; the process of generating the second configuration message by the network node can be that the terminal device to be executed for model inference can send a model inference request signaling to the network node through the base station, the network node can respond to the model inference request signaling, generate a model inference response message, and send the model inference response message to the terminal device to be executed for model inference through the base station; the terminal device receiving the model inference response message can send reporting information to the base station (edge node) to which the terminal device belongs, the base station can send the reporting information to the network node through an NG-C interface, and the network node, after receiving the reporting information, can determine, based on the reporting information, at least one model to be inferred that meets a preset matching condition in terms of a target dimension from a plurality of first models stored in the network node, and generate model inference strategy distribution information based on hierarchical federated learning based on a model identifier of the model to be inferred, a model file, and identifier information of the terminal device to be executed for model inference, that is, generate the second configuration message, and transmit the second configuration message to the edge node through an NG-U interface, and transmit the second configuration message to the terminal device to be executed for model inference by the edge node.

[0183] In the process of determining the model to be inferred / model group to be inferred by the network node, the plurality of models stored locally in the network node can be a plurality of first models, in which the network node can match a preset number of models to be inferred / model groups to be inferred by maximizing model accuracy, minimizing model complexity, or other dimensions based on the reporting information and the preset number configured in advance, and the specific implementation process will be described in detail in the following embodiments and will not be described here. The first model can be a model that has completed model identification or a model that has completed function identification.

[0184] Correspondingly, the specific implementation process of the step "obtaining a model inference result or a model updating result based on the second configuration message" can include:

[0185] performing online mode model inference or offline mode model inference based on the model identifier of the model to be inferred, the model file, and the identifier information of the terminal device to be executed for model inference, to obtain a model inference result.

[0186] The model identifier of the to-be-reasoned model can be a to-be-reasoned model group list identifier, for example, a to-be-reasoned model / to-be-reasoned model group list.

[0187] Specifically, the network node can encapsulate the to-be-reasoned model / to-be-reasoned model group, the model file of each to-be-reasoned model, and the identifier information of the terminal device to be executed model reasoning to obtain the model reasoning strategy distribution information. The network node can transmit the model reasoning strategy distribution information to the base station as a second configuration message through the NG-U interface. The base station can transmit the second configuration message to each terminal device to be executed model reasoning based on the identifier of the terminal device to be executed model reasoning in the second configuration message. After receiving the second configuration message, the terminal device can

[0188] In one example, the base station can transmit the second configuration message to each terminal device to be executed model reasoning through a target transmission mode. The base station can transmit the second configuration message to the terminal device to be executed model reasoning in at least one of the following message transmission modes: system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message. The form of model transmission can include one or more of the following: sent in a private format, sent in a public format, and the user equipment knows the model structure, and sent in a public format, and the user equipment does not know the model structure.

[0189] Based on this, the terminal device can perform model reasoning based on the model reasoning strategy distribution information to obtain a model reasoning result. For an application scenario with a time delay requirement greater than a first time delay threshold, the terminal device can use an online model reasoning strategy. After receiving the model reasoning strategy distribution information, the terminal device can collect real-time data and perform real-time model reasoning based on the collected real-time data and the second configuration message to obtain an online model reasoning result. For an application scenario with a time delay requirement less than a second time delay threshold, the terminal device can obtain data stored in the terminal device locally in advance and perform model reasoning based on the data and the second configuration message to obtain an offline model reasoning result.

[0190] In this embodiment, the terminal device can perform online real-time model reasoning or offline model reasoning based on pre-collected data based on the configuration requirements of the network node, thereby expanding the scenarios applicable to model reasoning.

[0191] In one exemplary embodiment, as shown in FIG. 4, a model management method, for example, a model management method based on hierarchical federated learning, is provided, which is applied to a network node in a mobile communication system in FIG. 1. The network includes a network node and multiple base stations. Each base station corresponds to multiple terminal devices. The terminal devices corresponding to each base station can include multiple terminal devices within the communication coverage of each base station. The method includes:

[0192] At step 402, the terminal device receives the reporting information sent by the base station.

[0193] The reporting information at least includes one of terminal capability, inference task type, and model performance monitoring feedback information.

[0194] Specifically, the terminal device can send the reporting information to the base station to which the terminal device belongs, and the base station can send the received reporting information to the network node.

[0195] At step 404, a second configuration message is generated based on the reporting information, and the second configuration message is sent to the terminal device by the base station.

[0196] The second configuration message is model inference strategy issuing information or model update strategy issuing information based on hierarchical federated learning, and the second configuration message is used to make the terminal device obtain a model inference result or a model update result.

[0197] Specifically, based on this, the network node can send the second configuration message to each terminal device. In the model inference process, the network node sends the second configuration message to each terminal device, and the second configuration message can include model inference strategy issuing information based on hierarchical federated learning. In the model update process, the network node sends the second configuration message to each terminal device, and the second configuration message can include model update strategy issuing information based on hierarchical federated learning. The terminal device can perform model inference based on the model inference strategy issuing information in the received second configuration message to obtain a model inference result. The terminal device can also update the model based on the model update strategy issuing information in the second configuration message to obtain a model update result.

[0198] In this embodiment, data interaction between the base station and the terminal can be implemented, the management efficiency and the generalization performance of the wireless intelligent federated learning model can be improved, the additional consumption of communication resources and the calculation overhead can be reduced, the to-be-allocated resources can be flexibly configured, and the terminal devices performing federated learning can be flexibly scheduled, and the application scenarios of the model are further expanded.

[0199] In one of the embodiments, the terminal capability includes one or more of terminal computing capability, terminal channel state, and inference feature, the terminal computing capability is used to represent the model complexity supported by the terminal device, and the terminal channel state represents the uplink channel state on the terminal device side.

[0200] Specifically, the terminal computing capability is an index representing the strength of the computing capability of the terminal device, and the terminal computing capability can be used to determine the model complexity supported by the terminal device. The terminal computing capability of the terminal device can be in a positive correlation with the model complexity supported by the terminal device, that is, the stronger the terminal computing capability, the higher the model complexity that can be supported. The terminal computing capability can include the CPU frequency, thread number, core number of the terminal device, the bit width, and the display memory size of the GPU, and the like. The terminal channel state is used to represent the uplink channel state (CSI). The inference feature can be an intelligent feature used to represent the local data of the terminal device. For example, the inference feature can include the number of dataset features, the number of samples, the dimension, the data type, the learning rate, the data batch, and the like. The data type is used to represent whether it is a time series. The terminal computing capability and the inference feature can be optional items of the terminal device for sending the reporting information to the network. The terminal device can determine whether the terminal computing capability and the inference feature are included in the reporting information based on the actual application scenario. Alternatively, the terminal computing capability and the inference feature can be determined based on the network side configuration.

[0201] Optionally, the reporting information can include the terminal capability, and the reporting information can further include the inference task type, the auxiliary information, and the like. The inference task type can represent the type of the model inference task, that is, the type of the model or the type of the model inference result. The type can include one or more of a classification task, a prediction task, and a clustering task. The model used for performing the model inference can include one or more of a classification model, a prediction model, and a clustering model.

[0202] In this embodiment, the terminal capability carried in the reporting information can include multiple types, and the specific content of the terminal capability can be determined based on the actual needs of the terminal device, to ensure the flexibility of the information represented by the terminal capability.

[0203] In one of the embodiments, the reporting information includes the terminal capability, and the specific implementation process of the step of “receiving the reporting information sent by the terminal device through the base station” can include the following steps.

[0204] Based on the target transmission mode, the terminal capability reported by the terminal device through the base station is received.

[0205] The target transmission mode includes at least one of a system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

[0206] Specifically, the terminal device can report the terminal device capability reporting signaling to the base station (edge node) through at least one of the following transmission manners: system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message, that is, the reporting information is reported to the edge node through at least one of the following transmission manners: system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message. After receiving the reporting information, the edge node can send the reporting information to the network node corresponding to the edge node. For example, the edge node can send the reporting information to the network node through an NG-C interface.

[0207] In this embodiment, the terminal device can transmit data between the base station through multiple transmission manners, ensuring the diversity of data transmission and the reliability of data transmission.

[0208] In one of the embodiments, the model management method based on hierarchical federated learning further includes:

[0209] The model inference request signaling sent by the terminal device through the base station is received, and a response message corresponding to the inference request is returned to enable the terminal device to send reporting information based on the response message; the response message can be a model inference response message.

[0210] Specifically, in the model inference process, the terminal device can be a terminal device to be executed for model inference. The terminal device can generate a model inference request signaling and send the model inference request signaling to the base station to which the terminal device belongs. After receiving the model inference request signaling, the base station can send the model inference request signaling to the network node. The network node can respond to the model inference request signaling, generate a model inference response message, and return the model inference response message to the terminal device to be executed for model inference through the base station. The terminal device receiving the model inference response message can send reporting information to the base station (edge node) to which the terminal device belongs. The base station can send the reporting information to the network node through an NG-C interface. The network node can receive the reporting information and obtain a second configuration message.

[0211] In this embodiment, the terminal device can actively trigger the model inference process and timely receive the related information of the model inference issued by the network node, improving the model inference efficiency.

[0212] In one of the embodiments, as shown in FIG. 5, the second configuration message includes model inference strategy issuing information or model update strategy issuing information based on hierarchical federated learning. The specific implementation process of the step of “generating a second configuration message based on the reporting information” can be:

[0213] In step 502, based on the reporting information, at least one to-be-inferred model satisfying a preset matching condition in the target dimension is determined from the plurality of first models.

[0214] The plurality of models stored locally by the network node can be a plurality of first models, which can be models that have completed model identification or models that have completed function identification. The target dimensions can include model structure dimensions, model complexity dimensions, model accuracy, and the like; and the preset matching conditions can be least complex, most complex, highest accuracy, lowest accuracy, or most moderate, and the like, which can be determined based on actual application scenarios.

[0215] Specifically, the network node can filter the plurality of first models based on the received report information and the preset number of to-be-inferred models, extract the model structure, model complexity, and accuracy of the to-be-inferred models that are most matched with the terminal capabilities carried in the report information, and determine the to-be-inferred models as the to-be-inferred model or the to-be-inferred model group, or select the preset number of models as the to-be-inferred model or the to-be-inferred model group.

[0216] Step 504, based on the model identifier of the to-be-inferred model, the model file, and the identifier information of the terminal device to be executed for model inference, generate model inference strategy distribution information based on layered federated learning.

[0217] Specifically, the network node can encapsulate the model identifier of the to-be-inferred model / to-be-inferred model group, the model file of each to-be-inferred model, and the identifier information of the terminal device to be executed for model inference, obtain an encapsulation result, and determine the encapsulation result as the model inference strategy distribution information, which can also be model inference strategy distribution signaling. The network node can transmit the model inference strategy distribution information to the base station through the NG-U, and the base station can transmit the model inference strategy distribution information to each terminal device to be executed for model inference.

[0218] In this embodiment, the network node can determine the to-be-inferred model to be executed for model inference based on the report information, and timely synchronize data with the base station and the terminal device to be executed for model inference, thereby ensuring the timeliness of the data and improving the efficiency of model inference.

[0219] In one of the embodiments, the specific implementation process of the step of "determining at least one to-be-inferred model that meets the preset matching condition in the target dimensions from the plurality of first models based on the report information" can include:

[0220] Based on the report information, determine a preset number of to-be-inferred models that meet the preset matching conditions in the model structure, model complexity, and model accuracy from the plurality of first models.

[0221] Specifically, the network node can filter among the plurality of first models based on the received report information and a preset number of to-be-inferred models, extract a model that is most matched with the terminal capability carried in the report information in terms of model structure, model complexity, and precision as the to-be-inferred model / to-be-inferred model group, or select a preset number of models as the to-be-inferred model / to-be-inferred model group.

[0222] In the embodiment, the network node can filter the to-be-inferred model from multiple dimensions based on the report information, to ensure comprehensiveness and flexibility of the inference model filtering.

[0223] In one of the embodiments, the specific implementation process of the step of "sending, by the base station, the second configuration message to the terminal device" can include:

[0224] The second configuration message is sent to the base station through the user plane interface, so that the base station sends the second configuration message to the terminal device through the target transmission mode.

[0225] The target transmission mode includes at least one of system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message. The user plane interface can be an NG-U interface.

[0226] Specifically, after receiving the report information sent by the terminal device, the network node can process the report information to obtain the second configuration message and send the second configuration message to the base station. After receiving the second configuration message, the base station can send the second configuration message to each terminal device corresponding to the base station. The network node can transmit the second configuration message to each base station through the NG-U interface, and the base station can transmit the second configuration message to a plurality of terminal devices corresponding to the base station in the form of a target transmission mode.

[0227] Optionally, the base station can transmit the second configuration message to the terminal device in at least one of the following message transmission modes: system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

[0228] In one example, the network node generates a second configuration message based on the received report information reported by the terminal device, and can send the second configuration message to a base station to which the terminal device belongs, so that the base station sends the second configuration message to the terminal device; in another example, the network node can send the second configuration message to each base station managed by the network node, so that each base station can send the second configuration message to each terminal device within the base station; in another example, the network node can screen each base station managed by the network node, and send the second configuration message to the screened base station, each base station can perform scheduling of the terminal device based on the method described in the above embodiment, obtain the scheduled terminal device, and send the second configuration message to each scheduled terminal device, wherein each base station can also send the second configuration message to the terminal device corresponding to the identifier of the terminal device based on the identifier of the terminal device carried in the second configuration message.

[0229] In the embodiment, the network node, the base station and the terminal device can realize data transmission through various transmission modes, ensuring the reliability, diversity and flexibility of data transmission.

[0230] In one of the embodiments, the model performance monitoring feedback information includes one or more of system performance indicators, accuracy indicators, complexity indicators and generalization performance indicators.

[0231] Specifically, the model performance monitoring feedback information can include various indicators, the system performance indicators can include one or more of throughput, block error rate and latency, the accuracy indicators can include one or more of the accuracy of the model output, the MSE / NMSE / cosine similarity between the model input and output, etc., the complexity indicators can include the model size and the calculation complexity, the model size refers to the number of parameters included in the model, and the calculation complexity can be FLOPs; the model performance monitoring feedback information can also include generalization performance indicators, and the specific content can be determined based on the actual application scenario.

[0232] Optionally, after the UE performs model inference, the model performance monitoring can be performed in real time to obtain the model performance monitoring result, and the model performance monitoring feedback information is obtained based on the result. The UE can feed back the model performance monitoring feedback information to the edge node corresponding to the terminal device, and the edge node can send the model performance monitoring feedback information to the network node (NW side) corresponding to the edge node after receiving the model performance monitoring feedback information.

[0233] In the embodiment, by monitoring the performance indicators in the model inference process, the model performance can be monitored in real time, and the timeliness of the model can be adjusted in time.

[0234] In one of the embodiments, the reporting information comprises model performance monitoring feedback information.

[0235] Correspondingly, as shown in FIG. 6, the specific implementation process of the step of “generating a second configuration message based on the reporting information” can comprise:

[0236] Step 602: receiving model performance monitoring feedback information.

[0237] Specifically, the terminal device can report the model performance monitoring feedback information to the base station to which the terminal device belongs, and the base station can send the model performance monitoring feedback information to the network node (NW side) corresponding to the base station after receiving the model performance monitoring feedback information. The network node can receive the model performance monitoring feedback information sent by the base station.

[0238] Step 604: receiving a model update request, the model update request comprising a to-be-updated model / to-be-updated function identifier, reference information.

[0239] The model update request is generated in the case that the terminal device model performance monitoring feedback information and the preset index threshold satisfy the preset update condition.

[0240] Specifically, the terminal device can compare the index contained in the model performance monitoring feedback information with the preset index threshold corresponding to the index, and if it is determined that the index and the preset index threshold satisfy the preset update condition, a model update event is triggered, and a model update request is generated. The terminal device can send the model update request to the network node through the base station.

[0241] The specific process in which the UE determines whether the preset update condition is satisfied can be that, in the case that the index contained in the model performance monitoring feedback information is a time delay, if the UE determines that the time delay is greater than or equal to the preset time delay threshold corresponding to the time delay, the UE can determine that the preset update condition is currently satisfied, and a model update event is triggered; in the case that the index contained in the model performance monitoring feedback information is model accuracy, if the UE determines that the model accuracy is less than or equal to the preset time delay threshold corresponding to the time delay, the UE can determine that the preset update condition is currently satisfied, and a model update event is triggered; the UE can determine whether the preset update condition is satisfied based on the meaning of the index and the relationship between the index and the preset index threshold corresponding to the index.

[0242] Step 606: responding to the model update request to generate a second configuration message.

[0243] Specifically, the network node can respond to the model update request, and the responding process can include: the network node determines at least one to-be-updated model to be executed for model update, generates model update strategy issuing information based on the at least one to-be-updated model, that is, obtains the second configuration message, and the network node can send the model update strategy issuing information to the terminal device to be executed for model update through the base station, and the terminal device can execute model update based on the model update strategy issuing information to obtain a model update result.

[0244] In this embodiment, the terminal device / network node can actively determine whether the model update process is needed based on the monitored model indicator information, improve the control degree of the terminal device / network node on the model update process, and timely perform model update in the case that the model does not meet the requirements.

[0245] In one of the embodiments, the reporting information includes model performance monitoring feedback information.

[0246] Correspondingly, as shown in FIG. 7, the specific implementation process of the step “generating the second configuration message based on the reporting information” can include:

[0247] Step 702, receiving model performance monitoring feedback information.

[0248] Specifically, the terminal device can obtain the model performance monitoring feedback information based on the model performance monitoring result, for example, the model performance monitoring result can be taken as the model performance monitoring feedback information; the terminal device can determine the model performance monitoring result as the reporting information, and send the reporting information to the base station to which the terminal device belongs, and the base station can send the reporting information to the network node corresponding to the base station.

[0249] Step 704, triggering a model update event based on a preset time period. Or, if the model performance monitoring feedback information and the preset indicator threshold meet a preset update condition, triggering an update event.

[0250] The preset time period is associated with terminal capability or auxiliary information.

[0251] Specifically, the triggering process of the model update can be actively triggered by the network side. After receiving the reported information, the network node can trigger a model update event based on a target triggering condition. The specific process of the target triggering condition can include periodic model update and aperiodic model update. In one possible implementation, the network node triggers an update event of a to-be-updated model or a to-be-updated model group according to a preset time period, and generates a second configuration message. The preset time period can be associated with a terminal capability or a change in auxiliary information, such as a scene, a site, a channel, or a data sampling speed.

[0252] Optionally, the specific process in which the network node determines whether the preset update condition is met can be as follows. In a case where the index included in the model performance monitoring feedback information is a time delay, if the network node determines that the time delay is greater than or equal to a preset time delay threshold corresponding to the time delay, the network node can determine that the preset update condition is currently met, and trigger a model update event. In a case where the index included in the model performance monitoring feedback information is model accuracy, if the network node determines that the model accuracy is less than or equal to a preset accuracy threshold corresponding to the model accuracy, the network node can determine that the preset update condition is currently met, and trigger a model update event. The network node can determine whether the preset update condition is met based on a meaning of the index, and a relationship between the index and a preset index threshold corresponding to the index.

[0253] At step 706, a second configuration message is generated based on the triggered update event.

[0254] Specifically, the network node determines at least one to-be-updated model to be executed for model update, generates model update strategy distribution information based on the at least one to-be-updated model, that is, obtains the second configuration message. The network node can send the model update strategy distribution information to a terminal device to be executed for model update through a base station. The terminal device can execute model update based on the model update strategy distribution information, and obtain a model update result.

[0255] In this embodiment, the terminal device / network node can actively determine whether a model update process is needed based on monitored model index information, improve the control degree of the terminal device / network node on the model update process, and perform model update in a timely manner when the model does not meet the requirements.

[0256] In one of the embodiments, the second configuration message comprises model inference strategy under layer federated learning issuing information.

[0257] Correspondingly, the specific implementation process of the step of "generating the second configuration message" can comprise:

[0258] Determine at least one model to be updated. Based on the model to be updated, the identification information of the terminal device to be executed for model updating, and the updating strategy, generate the model inference strategy under layer federated learning issuing information.

[0259] Specifically, the second configuration message is model inference strategy issuing information; the process of the network node generating the second configuration message can be that the terminal device to be executed for model inference can send a model inference request signaling to the network node through the base station, the network node can respond to the model inference request signaling, generate a model inference response message, and send the model inference response message to the terminal device to be executed for model inference through the base station; the terminal device receiving the model inference response message can send reporting information to the base station (edge node) to which the terminal device belongs, the base station can send the reporting information to the network node through the NG-C interface, and the network node receiving the reporting information can determine at least one model to be inferred whose target dimension satisfies the preset matching condition among the plurality of first models stored in the network node based on the reporting information, and generate the model inference strategy under layer federated learning issuing information based on the model identification, model file of the model to be inferred, and identification information of the terminal device to be executed for model inference, that is, generate the second configuration message, and transmit it to the edge node through the NG-U interface, and transmit it to the terminal device to be executed for model inference by the edge node.

[0260] In one example, the network node can encapsulate the model identification, model file of the model to be inferred, and identification information of the terminal device to be executed for model inference to obtain an encapsulation result, and determine that the encapsulation result is the model inference strategy under layer federated learning issuing information.

[0261] Wherein, the process of the network node determining the model to be inferred / model group to be inferred can be that the plurality of models stored locally by the network node can be a plurality of first models, among the plurality of first models, the network node can match by maximizing model accuracy, minimizing model complexity, or other dimensions based on the reporting information and the preset number configured in advance to obtain the preset number of matched models to be inferred / model groups to be inferred.

[0262] In this embodiment, the network node can screen suitable models to be inferred for model inference processing in the model inference process, and ensure the efficiency of model inference.

[0263] In one of the embodiments, the update strategy includes one or more of a local update strategy, a global update strategy, a model replacement strategy, and a model deletion strategy.

[0264] Specifically, the specific meaning of the update strategy and the specific process of obtaining the model update result by the terminal device based on the update strategy for model update processing have been described in detail in the corresponding embodiments of steps 302 to 308, and will not be repeated here.

[0265] In one of the embodiments, as shown in FIG. 8, the model management method based on hierarchical federated learning further includes:

[0266] Step 802, receiving the model data sent by each base station.

[0267] The model data includes a model file, model content / function identifier.

[0268] Specifically, the base station can determine the identified federated learning model indicated by the base station, and send the model data of the identified federated learning model to the network node through the NG-U interface. The network node can receive the model data sent by each base station, which represents the federated learning model identified by the base station. The model data includes a model file, model content, or a function identifier.

[0269] Step 804, determining each first model based on each model data, and arranging according to the evaluation index of each first model to obtain a first model list.

[0270] The evaluation index can include one or more of accuracy, convergence speed, model complexity, or other evaluation indexes.

[0271] Specifically, the network node obtains a plurality of identified models, i.e., first models, based on the model file in each model data. The network node can sort each first model according to the evaluation index of each first model to obtain a first model list. The first model list can be a model list or a function list.

[0272] In this embodiment, the network node can efficiently store the functions / models identified by each base station, improving the storage efficiency of the model.

[0273] In one of the embodiments, the model management method based on hierarchical federated learning further includes:

[0274] Receiving the identification request and the model capability sent by the base station, the identification request including a model identification request and a function identification request, and the model capability being used to indicate the supported model / function supported by the base station.

[0275] Specifically, after the federal learning model training is completed, an identification event can be triggered, which can be a model identification event or a function identification event; the identification event can be triggered by the base station, and the specific identification process can be: the base station can send an identification request to the network node through the NG-C interface, the base station reports the model capability to the network node, the model capability reported by the base station is associated with a specific configuration / condition, and the model capability reported by the base station to the network node can also be associated with an auxiliary condition. The specific configuration / condition can be a specific configuration / condition under the terminal capability in the area where the base station is located, and the area where the base station is located can be a cell / sector / cell group where the base station is located. The auxiliary condition can be, but is not limited to, a scene, a site, a channel, a data set, and the like.

[0276] In one example, the model capability reported by the base station can include a model type, an input and output description of the model, an AI (Artificial Intelligence) related feature involved in the model, an accuracy that can be achieved on a standard data set, a convergence speed, a model complexity, an application scenario, etc. The interaction of the model capability between the base station and the network node can be carried through the user plane signaling. The network node sorts and establishes a model / function list according to the accuracy, convergence speed, model complexity or other evaluation indexes for all the models reported by the base station, and assigns a model / function ID to each base station. The network node sends the model / function ID indication information to the corresponding base station through the NG-U interface. The model / function ID is unique on the network side, and the format of the ID includes at least one of the following:

[0277] Two parts of bit strings are concatenated, including: identification bits of the cell / sector / cell group to which the base station belongs, and bits for distinguishing different models / functions in the same cell / sector / cell group;

[0278] Other bit encoding methods for distinguishing different models / functions.

[0279] Based on this, the base station can determine the federal learning model indicated by the base station that has been identified, and send the model data of the federal learning model that has been identified to the network node through the NG-U interface. The network node can receive the model data sent by each base station, which represents the federal learning model identified by the base station, and the model data includes a model file, model content or a function identifier.

[0280] In this embodiment, the base station can actively trigger model identification, improving the flexibility of model identification.

[0281] In one of the embodiments, the model management method based on hierarchical federal learning further includes:

[0282] The identification request is sent to each base station to make each base station report model capability, which is used to indicate the model / function supported by the base station.

[0283] Specifically, after the federated learning model training is completed, an identification event can be triggered, which can be a model identification event or a function identification event; the identification event can be triggered by a network node, and the specific identification process can be: the network node can send an identification request to the base station through the NG-C interface, after the base station receives the identification request, in response to the model identification request, the network node can report the model capability, the model capability reported by the base station is associated with a specific configuration / condition, and the model capability reported by the base station to the network node can also be associated with an auxiliary condition, the specific configuration / condition can be a specific configuration / condition under the terminal capability in the area where the base station is located, and the area where the base station is located can be a cell / sector / cell group where the base station is located; the auxiliary condition can be but not limited to scene, site, channel, data set, etc.

[0284] The base station reports the model capability to the network node, the model capability reported by the base station is associated with a specific configuration / condition, and the model capability reported by the base station to the network node can also be associated with an auxiliary condition, the specific configuration / condition can be a specific configuration / condition under the terminal capability in the area where the base station is located, and the area where the base station is located can be a cell / sector / cell group where the base station is located; the auxiliary condition can be but not limited to scene, site, channel, data set, etc.

[0285] In one example, the model capability reported by the base station can include model type, input and output description of the model, AI-related features involved in the model, accuracy that can be achieved on the specification data set, convergence speed, model complexity, application scenario, etc. The interaction of model capability between the base station and the network node can be carried through the user plane signaling. The network node sorts and establishes a model / function list according to the accuracy, convergence speed, model complexity or other evaluation indexes for all the models reported by the base station, and assigns a model / function ID to each base station. The network node sends the model / function ID indication information to the corresponding base station through the NG-U interface. The model / function ID is unique on the network side, and the format of the ID includes at least one of the following:

[0286] Two parts of bit strings are concatenated, including: identification bits of the cell / sector / cell group to which the base station belongs, and bits used to distinguish different models / functions in the same cell / sector / cell group;

[0287] Other bit encoding methods used to distinguish different models / functions.

[0288] Based on this, the base station can determine the identified federated learning model indicated by the base station, and send the model data of the identified federated learning model to the network node through the NG-U interface. The network node can receive the model data sent by each base station, which represents the respective federated learning models identified by the base station, and the model data includes model files, model contents, or function identifiers.

[0289] In this embodiment, the network node can actively trigger model identification, improving the flexibility of model identification.

[0290] In one exemplary embodiment, as shown in FIG. 9, a model management method, such as a model management method based on hierarchical federated learning, is provided and applied to a base station in a mobile communication system as shown in FIG. 1. The mobile communication system includes a network and a plurality of base stations, each base station corresponding to a plurality of terminal devices, the terminal devices being distributed nodes performing hierarchical federated learning, and the base station being a central node performing hierarchical federated learning. The method includes:

[0291] Step 902: obtaining user scheduling result indication information corresponding to each terminal based on user feature information of each terminal, and sending a first configuration message to each terminal device.

[0292] The user feature information includes at least one of user data feature information and user channel state information, and the first configuration message contains the user scheduling result indication information.

[0293] Specifically, the base station can determine all terminal devices within the coverage range of the base station, and select each terminal device based on hierarchical federated learning among all terminal devices. The base station can obtain a first configuration message based on the user scheduling indication information, and send the first configuration message to each selected terminal device. The user scheduling result indication information contained in the first configuration message can represent that the terminal device is selected by the base station.

[0294] In one example, the specific process of scheduling by the base station can include: the base station determines a first number of terminal devices that need to be scheduled (used) at present, and selects a first number of terminal devices based on user feature information of each terminal device among all terminal devices within the coverage range of the base station. The user scheduling result indication information of the first number of terminal devices is first indication information, which can represent that the terminal device is selected by the base station. The user scheduling indication information of other terminal devices can be empty or second indication information, which is used to represent that the terminal device is not selected by the base station. The specific form of the first indication information and the second indication information is not limited in the present disclosure, and can be determined based on actual application scenarios by those skilled in the art.

[0295] Step 904, receiving the report information sent by the terminal device.

[0296] The report information at least includes at least one of terminal capability, inference task type, and model performance monitoring feedback information.

[0297] Step 906, sending the report information to the network node, so that the network node obtains a second configuration message.

[0298] The second configuration message is model inference strategy issuing information or model update strategy issuing information based on hierarchical federated learning, and the second configuration message is used to make the terminal device obtain a model inference result or a model update result.

[0299] Specifically, the terminal device can send the report information to the base station, the base station can send the received report information to the network node, and the network node can generate a second configuration message based on the report information after receiving the report information. Based on this, the network node can send the second configuration message to the base station, and the base station can send the received second configuration message to each terminal device. In the model inference process, the base station sends the second configuration message to each terminal device, which can contain model inference strategy issuing information based on hierarchical federated learning. In the model update process, the network node sends the second configuration message to each terminal device, which can contain model update strategy issuing information based on hierarchical federated learning. The terminal device can receive the second configuration message and perform model inference based on the model inference strategy issuing information in the second configuration message to obtain a model inference result. The terminal device can also update the model based on the model update strategy issuing information in the second configuration message to obtain a model update result.

[0300] In this embodiment, data interaction between the base station and the terminal can be realized, the management efficiency and generalization performance of the wireless intelligent federated learning model can be improved, the additional consumption of communication resources and the calculation overhead can be reduced, the to-be-allocated resources can be flexibly configured, and the terminal devices performing federated learning can be flexibly scheduled, and the application scenarios of the model are further expanded.

[0301] In one embodiment, the model management method based on hierarchical federated learning further includes:

[0302] Sending model data to the network node, the model data including model files and model content / function identifiers, so that the network node determines each first model based on each model data and obtains a first model list.

[0303] Specifically, the base station can determine the identified federated learning model indicated by the base station, and send model data of the identified federated learning model to the network node through the NG-U interface. The network node can receive the model data sent by each base station, the model data representing the respective federated learning models identified by the base station, and the model data including a model file, model content, or a function identifier.

[0304] The network node obtains a plurality of identified models, i.e., first models, based on the model files in the respective model data. The network node can sort the respective first models according to evaluation indexes of the respective first models to obtain a first model list, which can be a model list or a function list. The evaluation indexes can include one or more of accuracy, convergence speed, model complexity, or other evaluation indexes.

[0305] In this embodiment, the network node can efficiently store the functions / models identified by the respective base stations, thereby improving the storage efficiency of the models.

[0306] In one embodiment, the model management method based on hierarchical federated learning further includes:

[0307] The network node sends an identification request and a model capability to the network node, the identification request including a model identification request and a function identification request, and the model capability indicating the supported models / supported functions of the base station.

[0308] Specifically, after the federated learning model training is completed, an identification event can be triggered, which can be a model identification event or a function identification event. The identification event can be triggered by the base station. The specific identification process can be that the base station sends an identification request to the network node through the NG-C interface, the base station reports a model capability to the network node, the model capability reported by the base station is associated with a specific configuration / specific condition, and the model capability reported by the base station to the network node can also be associated with an auxiliary condition. The specific configuration / specific condition can be a specific configuration / specific condition under the terminal capability in the area where the base station is located, and the area where the base station is located can be a cell / sector / cell group where the base station is located. The auxiliary condition can be, but is not limited to, a scene, a site, a channel, a data set, and the like.

[0309] In one example, the model capability reported by the base station can include the model type, the input and output description of the model, the AI-related features involved in the model, the accuracy that can be achieved on the specification dataset, the convergence speed, the model complexity, the application scenario, etc. The interaction of model capability between the base station and the network node can be carried through the user plane signaling. The network node ranks and establishes a model / function list according to the accuracy, convergence speed, model complexity or other evaluation indexes for all the models reported by the base station, and assigns a model / function ID to each base station. The network node sends the model / function ID indication information to the corresponding base station through the NG-U interface. The model / function ID is unique at the network side, and the format of the ID includes at least one of the following:

[0310] Two parts of bit strings are concatenated, including: identification bits of the cell / sector / cell group to which the base station belongs, and bits used to distinguish different models / functions in the same cell / sector / cell group;

[0311] Other bit encoding methods used to distinguish different models / functions.

[0312] Based on this, the base station can determine the identified federated learning model indicated by the base station, and send the model data of the identified federated learning model to the network node through the NG-U interface. The network node can receive the model data sent by each base station, which represents the federated learning model identified by the base station, and the model data includes the model file, the model content or the function identifier.

[0313] In this embodiment, the base station can actively trigger model identification, improving the flexibility of model identification.

[0314] In one of the embodiments, the model management method based on hierarchical federated learning further includes:

[0315] Receiving the identification request sent by the network node, and reporting the model capability to the network node based on the identification request, the model capability being used to indicate the supported model / function of the base station.

[0316] Specifically, after the federated learning model training is completed, an identification event can be triggered, which can be a model identification event or a function identification event; the identification event can be triggered by a network node, and the specific identification process can be: the network node can send an identification request to the base station through the NG-C interface, after the base station receives the identification request, in response to the model identification request, the base station can report the model capability to the network node, the model capability reported by the base station is associated with a specific configuration / condition, and the model capability reported by the base station to the network node can also be associated with an auxiliary condition, the specific configuration / condition can be a specific configuration / condition under the terminal capability in the area where the base station is located, and the area where the base station is located can be a cell / sector / cell group where the base station is located; the auxiliary condition can be but not limited to scene, site, channel, data set, etc.

[0317] The base station reports the model capability to the network node, the model capability reported by the base station is associated with a specific configuration / condition, and the model capability reported by the base station to the network node can also be associated with an auxiliary condition, the specific configuration / condition can be a specific configuration / condition under the terminal capability in the area where the base station is located, and the area where the base station is located can be a cell / sector / cell group where the base station is located; the auxiliary condition can be but not limited to scene, site, channel, data set, etc.

[0318] In one example, the model capability reported by the base station can include model type, input and output description of the model, AI-related features involved in the model, accuracy that can be achieved on the specification data set, convergence speed, model complexity, application scenario, etc. The interaction of model capability between the base station and the network node can be carried through the user plane signaling. The network node sorts and establishes a model / function list according to the accuracy, convergence speed, model complexity or other evaluation indexes for all the model reported by the base station, and assigns a model / function ID to each base station. The network node sends the model / function ID indication information to the corresponding base station through the NG-U interface. The model / function ID is unique on the network side, and the format of the ID includes at least one of the following:

[0319] Two parts of bit strings are concatenated to form a bit string, including: identification bits of the cell / sector / cell group to which the base station belongs, and bits used to distinguish different models / functions in the same cell / sector / cell group;

[0320] Other bit encoding methods used to distinguish different models / functions.

[0321] Based on this, the base station can determine the identified federated learning model indicated by the base station, and send the model data of the identified federated learning model to the network node through the NG-U interface. The network node can receive the model data sent by each base station, which represents the respective federated learning models identified by the base station, and the model data includes model files, model contents, or function identifiers.

[0322] In this embodiment, the network node can actively trigger model identification, improving the flexibility of model identification.

[0323] In one of the embodiments, the model management method based on hierarchical federated learning further includes:

[0324] Based on the model training results sent by each terminal device, a global model trained is obtained.

[0325] Specifically, the base station can obtain the global model trained based on the model training results of each terminal device.

[0326] In one of the embodiments, the terminal device is a terminal device of a scheduled user.

[0327] Correspondingly, as shown in FIG. 10, the specific implementation process of the step "obtaining a global model trained based on the model training results sent by each terminal device" can include:

[0328] In step 1002, for the i-th round of training, the model parameters of the global model of the i-th round are obtained based on the training results sent by each terminal device of the (i-1)-th round.

[0329] In step 1004, in the case where the preset training completion condition is not met, the base station broadcasts the model parameters of the global model of the i-th round to each terminal device, so that each terminal device performs training based on the model parameters of the global model of the i-th round until the base station determines that the preset training completion condition is met, and a global model trained is obtained.

[0330] In one of the embodiments, each terminal device sends a local update gradient, the base station can perform model aggregation based on the received local update gradient sent by each terminal device, and obtain updated global model parameters. In the case where it is determined that the preset training completion condition is not met, the base station can broadcast the updated global model parameters to the terminal devices corresponding to the base station through a physical downlink shared channel (PDSCH), so that the terminal devices receiving the global model parameters perform training until the global model trained is obtained in the case where the preset training completion condition is met.

[0331] In one example, the network can contain M cells / sectors / cell groups (with the same cell ID), each of which is provided with an edge node (base station) and N user equipment (terminal equipment), and there is one network node in the network, and for each cell / sector / cell group, the following steps are independently performed:

[0332] Step 1, the user equipment and the edge node start model training, the process of model training at least includes one or more of periodic model training, aperiodic model training:

[0333] Periodic model training, the user equipment and the edge node start model training process according to a preset time period value, wherein the time period value is associated with specific configurations / conditions under the capability of each UE in the cell / sector / cell group, and the time period value can also be associated with auxiliary conditions, which can include scenarios, sites, channels, data sampling speed, typical parameters or other requirements. Optionally, the network node can configure the time period value through at least one of system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

[0334] Aperiodic model training, the user equipment and the edge node start model training process in a non-periodic manner (triggered by the UE side or the NW side).

[0335] The triggering mode of model training at least includes one of the following:

[0336] UE side triggering, the UE side feeds back a model training request signaling to the edge node for one / more AI related features, and the edge node responds to the request and starts model training;

[0337] NW side triggering, the NW side issues a model training request signaling to the user equipment through the edge node for one / more AI related features, and the user equipment responds to the request and starts model training.

[0338] Step 2, the model training is performed for T rounds, and in the i-th round of model training process, 0≤i≤T, the user equipment trains a local model to obtain a local model update result.

[0339] Step 3, the edge node can determine the user set to be scheduled in the current model training process based on the user scheduling method. Only the scheduled user equipment needs to perform the local model training process in step 102. Specifically, the edge node pre-sets the number of users to be scheduled in the current model training, determines the user set to be scheduled from all users in the cell / sector / cell group in its coverage based on the channel state information fed back by the user CSI-RS channel estimation, the feature information between the user local update gradient, or randomly selects a part of the user, and broadcasts the ID set of the scheduled user to all users through the downlink physical broadcast channel PBCH in the cell / sector / cell group.

[0340] Step 4, the scheduled user equipment uploads the local model update result (training result) to the edge node through the physical uplink shared channel (PUSCH, Physical Uplink Shared Channel, Physical Uplink Shared Channel).

[0341] Step 5, the edge node performs model aggregation according to the local update gradient in the training result uploaded by all scheduled users to obtain the model parameter of the updated global model.

[0342] Step 6, the edge node broadcasts the model parameter of the updated global model to all user equipment through the physical downlink shared channel (PDSCH, Physical Downlink Shared Channel, Physical Downlink Shared Channel).

[0343] If it is determined that the preset training completion condition is not met, steps 1 to 6 are repeatedly executed until the global model converges, that is, the preset training completion condition is met. The preset training completion condition can be a convergence condition, and determining whether the model meets the preset convergence condition can include at least one of the following:

[0344] The number of training rounds reaches a preset upper limit value, the loss function of the global model on the validation set is less than or equal to a preset value, and the difference between the loss functions of the global model on the validation set in adjacent two rounds is less than or equal to a preset value.

[0345] Specifically, in the same cell / sector / cell group, one or more model training processes can be performed, and the training mode includes at least one of the following: parallel mode: multiple model training is simultaneous or overlapping in time; serial mode: multiple model training is sequential and non-overlapping in time.

[0346] In one embodiment, the preset training completion condition includes one or more of the number of training rounds reaching a preset upper limit value, the loss function of the global model being less than or equal to a preset value, and the difference between the loss functions of the global model in adjacent two rounds being less than or equal to a preset value.

[0347] Specifically, the preset training completion condition can be a convergence condition, and determining whether the model meets the preset convergence condition can at least include one of the following: the number of training rounds reaches a preset upper limit value, the loss function of the global model on the validation set is less than or equal to a preset value, and the difference between the loss functions of the global model on the validation set at adjacent two rounds is less than or equal to a preset value.

[0348] In one of the embodiments, the model management method based on hierarchical federated learning further includes:

[0349] Based on the user feature information of each terminal device, the user scheduling result indication information corresponding to each terminal is obtained, and the user feature information includes at least one of user data feature information and user channel state information. The first configuration message is sent to each terminal device, and the first configuration message includes the user scheduling result indication information.

[0350] In one of the embodiments, the specific implementation process of the step "obtaining the user scheduling result indication information corresponding to each terminal based on the user feature information of each terminal device" can include:

[0351] The number of scheduled users is obtained. Based on at least one of the user data feature information and the user channel state information of each terminal device, the terminal devices of the number of scheduled users are filtered out from all users in the coverage range of the base station, and the user scheduling result indication information is determined as the scheduled indication information.

[0352] Specifically, the base station can determine all terminal devices in the coverage range of the base station, and filter out the selected terminal devices based on hierarchical federated learning in the all terminal devices. The first configuration message is obtained based on the user scheduling indication information, and the base station can send the first configuration message to the selected terminal devices. The user scheduling result indication information included in the first configuration message can represent that the terminal device is selected by the base station.

[0353] In one example, the specific process of scheduling by the base station can include: the base station determines a first number of terminal devices that need to be scheduled (used) at present, and filters out the first number of terminal devices based on the user feature information of each terminal device from all terminal devices in the coverage range of the base station. The user scheduling result indication information of the first number of terminal devices is determined as first indication information, which can represent that the terminal device is selected by the base station. The user scheduling indication information of other terminal devices can be empty or second indication information, which is used to represent that the terminal device is not selected by the base station. The specific forms of the first indication information and the second indication information are not limited in the disclosure, and can be determined based on actual application scenarios by those skilled in the art.

[0354] In this embodiment, the base station can perform user scheduling, guarantee the reliability of the terminal device performing model training, and improve the efficiency of model training.

[0355] In the following, the specific implementation process of the above-mentioned model management method based on hierarchical federated learning will be described in detail in combination with a specific embodiment. Before introducing the specific implementation process, the professional terms involved in this embodiment are explained first:

[0356] 6G: 6th Generation Mobile Communication Technology, 6th generation mobile communication technology

[0357] IoT: Internet of Things, Internet of Things

[0358] LCM: Life Cycle Management, Life Cycle Management

[0359] LMF: Location Management Function, Location Management Function

[0360] SINR: Signal to Interference plus Noise Ratio, Signal to Interference plus Noise Ratio

[0361] SGD: Stochastic Gradient Descent, Stochastic Gradient Descent

[0362] SSB: Synchronization Signal Block, Synchronization Signal Block

[0363] RSRP: Reference Signal Receiving Power, Reference Signal Receiving Power

[0364] KPI: Key Performance Indicators, Key Performance Indicators

[0365] MSE / NMSE: (Normalized)Mean Squared error, (Normalized)Mean Squared error

[0366] The method provided in the embodiment is a life cycle management method supporting flexible user scheduling method, which can improve the management efficiency and generalization performance of the intelligent air interface federated learning model, reduce the communication and calculation overhead, effectively resist the influence of data heterogeneity, improve the convergence speed and precision performance of model training, consider the strategy of reducing energy consumption, reasonably reduce the cost of deploying the federated learning algorithm in the communication system, consider the model generalization performance, so that the federated learning model can be well adapted or deployed on the newly added network node. The method provided in the embodiment is based on the close relationship between the convergence of the federated learning model and the data distribution of each user equipment, which can effectively schedule users and accelerate the convergence speed of the federated learning model.

[0367] The life cycle management (LCM) involved in the method provided in the embodiment is an important concept proposed in the current 3GPP standardization research based on the wireless AI framework. As a full-process management scheme covering the AI model, the LCM includes data collection, model training, model inference, model management (selection based on functionality / model ID, activation / deactivation, model switching / rollback, model monitoring, model updating), model transmission, etc. Therefore, the method provided in the embodiment is designed and optimized under the condition of meeting the life cycle management framework defined by 3GPP.

[0368] Specifically, the distributed network architecture in the method provided in the embodiment includes a network node, a base station, and a terminal device. The network node can be a central node / cloud, for example, LMF / OAM / NWDAF (NW side). The base station can be an edge node / server end, for example, CU / BBU under the CU-DU separation architecture of gNB. The terminal device can be a terminal node / user equipment / local client, for example, UE (UE side). As shown in FIG. 11a, the life cycle management based on hierarchical federated learning in the method provided in the embodiment includes at least a model training process based on hierarchical federated learning, a model identification process based on hierarchical federated learning, a model inference process based on hierarchical federated learning, and a model updating process based on hierarchical federated learning. Specifically, initialization, initializing the model parameters for all user equipments; model training based on hierarchical federated learning, for M cells / sectors / cell groups in the network, in each cell / sector / cell group, the user equipment or the edge node triggers to start the model training process, any scheduled user equipment updates the local model parameters, and transmits the model parameters to the edge node through the air interface, the edge node performs model aggregation to update the global model parameters, and through several rounds of iteration, the global model parameters are updated until convergence, forming M clustering federated learning models; model / function identification based on hierarchical federated learning, the network node identifies the trained federated learning models on all cells / sectors / cell groups, all edge nodes send the identified federated learning models to the network node through the NG interface for model storage, and the network node allocates model / function IDs to all identified models for model deployment / inference / update operations; model inference based on hierarchical federated learning, the user equipment to be executed sends a model inference request signaling to the NW side, reports UE capability and inference task related information, the network node decides the inference model / model group matched with the UE side according to the reported information, generates a signaling containing inference model / model group configuration information, and transmits it to the user equipment to be executed, the user equipment to be executed performs model inference and obtains inference results, and performs real-time model performance monitoring; model updating based on hierarchical federated learning, the model updating process is triggered by the UE side or the NW side, and the network node performs the model updating process.

[0369] The following describes the model training process based on hierarchical federated learning, which includes:

[0370] The network can include M cells / sectors / cell groups (with the same cell ID), each of which is provided with an edge node (base station) and N user equipments (terminal devices). The network has one network node, and for each cell / sector / cell group, the following steps are independently performed:

[0371] Step 1: User devices and edge nodes initiate model training. The model training process includes at least one or more of the following: periodic model training and non-periodic model training.

[0372] Periodic model training involves user equipment and edge nodes initiating the model training process according to a preset time period. This time period is associated with specific configurations / conditions within the cell / sector / cell group and the capabilities of each UE. The time period can also be associated with auxiliary conditions, which may include scenarios, sites, channels, data sampling rates, typical parameters, or other requirements. Optionally, network nodes can configure this time period through at least one of system messages, RRC signaling, MAC CE signaling, DCI signaling, and NAS messages.

[0373] Non-periodic model training involves the user equipment and edge nodes initiating the model training process in a non-periodic manner (triggered by the UE side or NW side).

[0374] The triggering methods for model training include at least one of the following:

[0375] The UE side triggers the signal to send a model training request to the edge node for one or more AI-related features. The edge node responds to the request and begins model training.

[0376] When triggered on the NW side, the NW side sends a model training request signaling to the user device through the edge node for one or more AI-related features. The user device responds to the request and begins model training.

[0377] Step 2: The model training is carried out in T rounds. During the i-th round of model training, 0≤i≤T, the user device trains the local model and obtains the local model update result.

[0378] Specifically, the process of "user device training a local model and obtaining local model update results" can be as follows: the user device can be a terminal device, the local model update result (training result) can be the local update gradient, and the terminal device can obtain the local update gradient based on local data using the stochastic gradient descent (SGD) algorithm. For example, the local update gradient g can be obtained using the following formula. i (w i,t ):

[0379] Among them, w i,t For the local model parameters of the i-th user, This refers to data batches obtained by random sampling based on the user's local dataset. For w i,t The gradient of the loss function at the data point (x,y).

[0380] Step 3, the edge node can determine the user set to be scheduled in the current model training process based on the user scheduling method. Only the scheduled user equipment needs to perform the local model training process in step 102. Specifically, the edge node pre-sets the number of users to be scheduled in the current model training, determines the user set to be scheduled from all users in the cell / sector / cell group in its coverage based on the channel state information fed back by the user CSI-RS channel estimation, the feature information between the user local update gradient, or randomly selects a part of the users, and broadcasts the ID set of the scheduled users to all users in the cell / sector / cell group through the downlink physical broadcast channel PBCH.

[0381] Wherein, the number of scheduled users pre-configured by the base station is less than or equal to the number of all users in RRC active state in the cell / sector / cell group.

[0382] Specifically, the base station can define the utility function of each terminal device as the intra-cell benefit minus the inter-cell benefit, the intra-cell benefit is the average value of the cosine similarity between the local update gradient of the user and the local update gradient of other users in the cell / sector / cell group, the local update gradient of each user equipment can be fed back to the base station through the uplink data channel, for example, the local update gradient of the user equipment can be compressed to obtain the compression result, and the compression result is fed back to the base station through the uplink data channel, realizing the reduction of transmission overhead; the inter-cell benefit is the average value of the cosine similarity between the local update gradient of the user and the local update gradient of other users in the non-cell / sector / cell group.

[0383] The base station can also define the optimization problem as maximizing the sum of the weighted utility functions of all users in the cell / sector / cell group, and the weight is the data size ratio of each user. The base station can schedule users based on the proportional fair algorithm to determine the scheduled user equipment, for example, the real utility of each user equipment can be defined as the ratio of the utility function value of the current training round to the time average utility function value of all past training rounds, and the first K users with the maximum real utility are selected from all users in the cell / sector / cell group as the scheduled users in the current training round, that is, the user scheduling result indication information corresponding to the scheduled user equipment is scheduled, wherein K is specific to the cell / sector / cell group and is configured by the NW side, which can be included in the system message or the NAS message.

[0384] Step 4, the scheduled user equipment uploads the local model update result (training result) to the edge node through the physical uplink shared channel (PUSCH, Physical Uplink Shared Channel).

[0385] Specifically, the physical uplink shared channel can be a flat Rayleigh fading channel, and a signal-to-interference-and-noise ratio (SINR) can be calculated by the following formula:

[0386] where P i,t is the transmit power of the i-th user (user equipment), h i,t is used to represent small-scale flat Rayleigh channel fading, d i is the distance from the user equipment to the base station, Φ o,i represents a set of users that interfere with the i-th user, α represents a path loss coefficient, and σ 2 represents the power of additive white Gaussian noise.

[0387] Step 5, the edge node performs model aggregation according to the local update gradient in the training result uploaded by all scheduled users to obtain the model parameters of the updated global model.

[0388] Specifically, the base station updates the global model parameters based on a weighted summation, and the weight is the data size of each scheduled user equipment, as shown in the following formula.

[0389] where, represents the global model parameters of the m-th cell / sector / cell group, η represents the learning rate, and β i represents the ratio of the data size of the i-th user to the data size of all users, and λ i,t represents whether the i-th user successfully transmits the local update gradient in the t-th round of model training, if the transmission is successful, λ i,t = 1, otherwise λ i,t = 0.

[0390] Step 6, the edge node broadcasts the model parameters of the updated global model to all user equipment through a physical downlink shared channel (PDSCH, Physical Downlink Shared Channel).

[0391] If it is determined that the preset training completion condition is not met at present, steps 1 to 6 are repeatedly executed until the global model converges, i.e., the preset training completion condition is met. The preset training completion condition can be a convergence condition, and determining whether the model meets the preset convergence condition can at least include one of the following:

[0392] The number of training rounds reaches a preset upper limit value, the loss function of the global model on the validation set is less than or equal to a preset value, and the difference between the loss functions of the global model on the validation set in adjacent two rounds is less than or equal to a preset value.

[0393] Specifically, in the same cell / sector / cell group, the training process of one or more models can be carried out, and the training mode at least includes one of the following: parallel mode: the training of multiple models is simultaneous or overlapping in time; serial mode: the training of multiple models is sequential and non-overlapping in time.

[0394] The following describes a model identification / function identification process based on hierarchical federated learning, which includes:

[0395] Step a, the network node performs model identification / function identification on the trained federated learning model on all cells / sectors / cell groups. The triggering mode of the identification process at least includes one of the following: edge node triggering or network node triggering, edge node triggering includes: the edge node sends a model / function identification request to the network node through the NG-C interface and reports the model capability, which is the capability of the model on the edge node, indicating the model / function supported by the edge node; network node triggering includes: the network node initiates a model / function identification request to the edge node through the NG-C interface, and the edge node responds to the request and reports the model capability, which is the capability of the model on the edge node, indicating the model / function supported by the model capability on the edge node.

[0396] Step b, all edge nodes send the federated learning model file indicated by the model / function ID to the network node through the NG-U interface for model deployment / inference / update / storage.

[0397] The following describes a specific embodiment in the model inference process based on hierarchical federated learning. The network node can obtain the model inference strategy distribution information by encapsulating the inference model / model group list and the model file information in the order of "model / function ID of model 1, model file of model 1, model / function ID of model 2, model file of model 2, …", and then determining the encapsulation result as the model inference strategy distribution information. The encapsulation format of the encapsulation result is IP packet, and the format of the model inference strategy distribution information can be a user device-specific binary executable file format.

[0398] In one embodiment, a 5G NR indoor positioning algorithm based on hierarchical federated learning is proposed. The user terminal is mainly responsible for updating the local model parameters, the base station is responsible for collecting the local model parameter information, using an aggregation algorithm to aggregate the local parameters to update the global model, and then distributing the new model parameters to the local terminal. The network node is responsible for the distribution of positioning strategy and resource control.

[0399] As an efficient positioning method for indoor scenes, AI technology can be applied to indoor industrial internet scenes to improve the problem of inaccurate positioning in traditional methods for indoor heavy NLOS scenes. The distributed federated learning training architecture has certain advantages, including high user privacy protection, increased positioning accuracy through collaborative positioning, reduced communication overhead, and accelerated training convergence speed.

[0400] As shown in Table 1, the positioning errors of different positioning schemes at the 50th, 67th, 80th, and 90th percentiles of positioning accuracy are recorded:

[0401] Table 1

[0402] As shown in FIG. 11b, the CDF of positioning accuracy can be the CDF corresponding to the positioning accuracy of the AI scheme. As can be seen from the results listed in Table 1 and FIG. 11b, the AI-based method can significantly improve the positioning accuracy. When using a traditional positioning algorithm to predict the coordinates of the UE, the positioning error exceeds 10 meters at the 90th percentile of the CDF. The AI-based method can reduce the positioning error to within 1 meter. When multiple measurement information is combined as input to the AI model, the positioning accuracy can be further improved. The simulation results show that when RSRP is used to obtain features, the positioning error of 90% of users can be reduced to within 0.38 meters.

[0403] In the case of 10 clients / base stations online in the simulation settings, the personalized federated learning algorithm FedProx and the federated average algorithm FedAvg are used as comparative algorithms, and the error variation trend of Per-FedAvg algorithm and centralized training is also compared, and the positioning error variation trend is shown in FIG. 11c. As can be seen from FIG. 11c, compared with the centralized training algorithm, the method of distributed training using federated learning has better convergence.

[0404] As shown in FIG. 11d, using the CDF of positioning error for comparison, the 90th percentile of the positioning error of Per-FedAvg corresponds to an error of 0.22m, and from the distribution function, no outstanding value appears, while in the case of centralized training, the 90th percentile of the positioning error corresponds to a value of 0.54m after 300 iterations. Compared with the centralized training case, distributed training significantly reduces the positioning error.

[0405] It should be understood that although each step in the flowchart involved in the embodiments described above is shown in sequence according to the direction of the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, there is no strict order limitation for the execution of these steps, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in the embodiments described above can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or steps or stages in other steps.

[0406] Based on the same idea, the embodiments of the present application also provide a model inference device for implementing the above-mentioned model management method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more model inference device embodiments provided below can refer to the limitations of the model management method described above, which will not be repeated here.

[0407] In one exemplary embodiment, as shown in FIG. 12, a model inference device 1200 is provided, which is applied to a terminal device in a mobile communication system, the mobile communication system further comprising a network, the network comprising a network node and a plurality of base stations, each of the base stations corresponding to a plurality of terminal devices, the terminal devices being distributed nodes performing hierarchical federated learning, and the base stations being central nodes performing hierarchical federated learning. The device comprises:

[0408] A first receiving module 1202 is configured to receive a first configuration message sent by the base station, wherein the first configuration message comprises user scheduling result indication information based on hierarchical federated learning, and the user scheduling result indication information is obtained by the base station based on user feature information of each terminal device, wherein the user feature information comprises at least one of user data feature information and user channel state information.

[0409] A first sending module 1204 is configured to send reporting information to the network, wherein the network stores a hierarchical federated learning model generated by the base station and a distribution policy, and the reporting information comprises at least one of terminal capability, inference task type, and model performance monitoring feedback information, and the reporting information is used by the network to generate a second configuration message, wherein the second configuration message is model inference strategy distribution information or model update strategy distribution information based on hierarchical federated learning.

[0410] A second receiving module 1206 is configured to receive the second configuration message, and obtain a model inference result or a model update result based on the second configuration message.

[0411] In one of the embodiments, the terminal capability includes one or more of a terminal computing capability, a terminal channel state, and an inference feature, the terminal computing capability is used to represent a model complexity supported by the terminal device, and the terminal channel state represents an uplink channel state of the terminal device.

[0412] In one of the embodiments, the first sending module is specifically used for:

[0413] Based on a target transmission mode, sending the reporting information to the base station, so that the base station sends the reporting information to the network node, and the target transmission mode includes at least one of a system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

[0414] In one of the embodiments, the model performance monitoring feedback information includes one or more of a system performance index, an accuracy index, a complexity index, and a generalization performance index.

[0415] In one of the embodiments, the apparatus further includes:

[0416] The first determining module is specifically used for: if it is determined that the model performance monitoring feedback information meets a preset update condition with a preset index threshold, sending a model update request to the base station, so that the network responds to the model update request, and the model update request includes one or more of a to-be-updated model, a to-be-updated function identifier, and reference information.

[0417] In one of the embodiments, the second receiving module is specifically used for:

[0418] Receiving a second configuration message sent by the base station based on a target transmission mode, the second configuration message is sent by the network node to the base station, and the target transmission mode includes at least one of a system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

[0419] In one of the embodiments, the second configuration message is model update strategy issuing information, and the model update strategy issuing information includes a to-be-updated model and an update strategy. The second receiving module is specifically used for:

[0420] Based on the update strategy, performing update processing on the to-be-updated model to obtain a model update result, the update model includes a to-be-updated model / to-be-updated model group, and the update strategy includes one or more of a local update strategy, a global update strategy, a model replacement strategy, and a model deletion strategy.

[0421] In one of the embodiments, the second receiving module is specifically used for at least one of:

[0422] updating part of the model to obtain a model update result.

[0423] updating the model parameters of each of the to-be-updated models based on a target update algorithm to obtain a model update result, the target update algorithm including one or more of a momentum gradient descent algorithm, a stochastic gradient descent algorithm, and a mini-batch gradient descent algorithm.

[0424] updating the model parameters of each of the to-be-updated models based on the new model parameters to obtain a model update result.

[0425] deleting the model parameters and model content / feature identifiers of the update model to obtain a model update result.

[0426] In one of the embodiments, the second configuration message is model inference strategy issuing information, and the model inference strategy issuing information includes a model identifier of a to-be-inferred model, a model file, and identifier information of a terminal device to be executed for model inference. The second receiving module is specifically configured to:

[0427] performing online mode model inference or offline mode model inference based on the model identifier of the to-be-inferred model, the model file, and the identifier information of the terminal device to be executed for model inference to obtain a model inference result.

[0428] In one exemplary embodiment, as shown in FIG. 13, a model inference apparatus 1300 is provided, which is applied to a network node in a mobile communication system, the network including a network node and a plurality of base stations, each of the base stations corresponding to a plurality of terminal devices, and the apparatus including:

[0429] The third receiving module 1302 is configured to receive report information sent by the terminal device through the base station, the report information including at least one of terminal capability, inference task type, and model performance monitoring feedback information.

[0430] The first generating module 1304 is configured to generate a second configuration message based on the report information, and send the second configuration message to the terminal device through the base station. The second configuration message is model inference strategy issuing information or model update strategy issuing information based on hierarchical federated learning, and the second configuration message is used to make the terminal device obtain a model inference result or a model update result.

[0431] In one of the embodiments, the terminal capability includes one or more of terminal computing capability, terminal channel state, and inference feature, the terminal computing capability is used to represent a model complexity supported by the terminal device, and the terminal channel state represents an uplink channel state on the terminal device side.

[0432] In one of the embodiments, the third receiving module is specifically configured to:

[0433] Based on the target transmission mode, receive terminal capability reported by the terminal device through the base station, and the target transmission mode includes at least one of system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

[0434] In one of the embodiments, the apparatus further includes:

[0435] The fourth receiving module is configured to receive a model inference request signaling sent by the terminal device through the base station, and return a response message corresponding to the inference request, so that the terminal device sends reporting information based on the response message.

[0436] In one of the embodiments, the second configuration message includes model inference strategy issuing information or model update strategy issuing information based on hierarchical federated learning. The first generating module is specifically configured to:

[0437] Based on the reporting information, determine at least one to-be-inferred model in a plurality of first models that meets a preset matching condition in a target dimension.

[0438] Based on the model identifier of the to-be-inferred model, the model file, and the identifier information of the terminal device to be executed for model inference, generate model inference strategy issuing information based on hierarchical federated learning.

[0439] In one of the embodiments, the first generating module is specifically configured to: based on the reporting information, determine a preset number of to-be-inferred models in a plurality of first models that meet a preset matching condition in model structure, model complexity, and model accuracy.

[0440] In one of the embodiments, the sending of the second configuration message to the terminal device through the base station includes:

[0441] Through a user plane interface, send the second configuration message to the base station, so that the base station sends the second configuration message to the terminal device through the target transmission mode, and the target transmission mode includes at least one of system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

[0442] In one of the embodiments, the model performance monitoring feedback information includes one or more of system performance indicators, accuracy indicators, complexity indicators, and generalization performance indicators.

[0443] In one of the embodiments, the reporting information includes model performance monitoring feedback information, and the first generating module is specifically configured to: receive the model performance monitoring feedback information.

[0444] receiving a model update request, the model update request comprising a to-be-updated model / to-be-updated function identifier, reference information, the model update request being generated in a case where the model performance monitoring feedback information of the terminal device and a preset index threshold satisfy a preset update condition.

[0445] responding to the model update request, generating a second configuration message.

[0446] In one of the embodiments, the reported information comprises model performance monitoring feedback information, and the first generating module is specifically configured to receive the model performance monitoring feedback information.

[0447] triggering a model update event based on a preset time period, or triggering an update event if the model performance monitoring feedback information and a preset index threshold satisfy a preset update condition, the preset time period being associated with terminal capability and auxiliary information.

[0448] generating a second configuration message based on the triggered update event.

[0449] In one of the embodiments, the second configuration message comprises model inference strategy under a hierarchical federated learning issuing information, and the first generating module is specifically configured to determine at least one to-be-updated model.

[0450] generating model inference strategy under a hierarchical federated learning issuing information based on the to-be-updated model, the identifier information of the terminal device to be executed for model update, and an update strategy.

[0451] In one of the embodiments, the update strategy comprises one or more of a local update strategy, a global update strategy, a model replacement strategy, and a model deletion strategy.

[0452] In one of the embodiments, the apparatus further comprises:

[0453] a fifth receiving module configured to receive model data sent by each of the base stations, the model data comprising a model file, model content / function identifier.

[0454] an arranging module configured to determine each of the first models based on each of the model data, and arrange each of the first models according to an evaluation index of each of the first models to obtain a first model list.

[0455] In one of the embodiments, the apparatus further comprises:

[0456] a sixth receiving module configured to receive a recognition request and model capability sent by a base station, the recognition request comprising a model recognition request and a function recognition request, the model capability being used to indicate a model / function supported by the base station.

[0457] In one of the embodiments, the apparatus further comprises:

[0458] The second sending module is configured to send an identification request to each of the base stations, so that each of the base stations reports a model capability, the model capability being used to indicate a model / function supported by the base station.

[0459] In one of the embodiments, as shown in FIG. 14, a model inference apparatus 1400 is provided, which is applied to a base station in a mobile communication system, and the apparatus comprises:

[0460] The third sending module 1402 is configured to obtain user scheduling result indication information corresponding to each of the terminal devices based on user feature information of each of the terminal devices, and send a first configuration message to each of the terminal devices, the user feature information comprising at least one of user data feature information and user channel state information, and the first configuration message comprising the user scheduling result indication information.

[0461] The seventh receiving module 1404 is configured to receive report information sent by the terminal device, the report information comprising at least one of terminal capability, inference task type, and model performance monitoring feedback information.

[0462] The fourth sending module 1406 is configured to send the report information to the network node, so that the network node obtains a second configuration message, the second configuration message being model inference strategy information or model update strategy information based on hierarchical federated learning, and the second configuration message being used to make the terminal device obtain a model inference result or a model update result.

[0463] In one of the embodiments, the apparatus further comprises:

[0464] The fifth sending module is configured to send model data to the network node, the model data comprising a model file and a model content / function identifier, so that the network node determines each first model based on each of the model data and obtains a first model list.

[0465] In one of the embodiments, the apparatus further comprises:

[0466] The sixth sending module is configured to send an identification request and a model capability to the network node, the identification request comprising a model identification request and a function identification request, and the model capability being used to indicate a model / function supported by the base station.

[0467] In one of the embodiments, the apparatus further comprises:

[0468] An eighth receiving module is configured to receive an identification request sent by the network node, and report model capability to the network node based on the identification request, wherein the model capability is used to indicate a model / function supported by the base station.

[0469] In one of the embodiments, the apparatus further comprises:

[0470] A seventh sending module is configured to obtain a global model trained based on the model training results sent by the terminal devices.

[0471] In one of the embodiments, the seventh sending module is specifically configured to, for the i-th round of training, obtain the model parameters of the global model of the i-th round based on the training results sent by the terminal devices of the (i-1)-th round.

[0472] In the case where the preset training completion condition is not met, the model parameters of the global model of the i-th round are broadcast to the terminal devices, so that the terminal devices train based on the model parameters of the global model of the i-th round until the base station determines that the preset training completion condition is met, and obtains the global model trained.

[0473] In one of the embodiments, the preset training completion condition comprises one or more of the following: the number of training rounds reaches a preset upper limit value, the loss function of the global model is less than or equal to a preset value, and the difference between the loss functions of the global model of adjacent two rounds is less than or equal to a preset value.

[0474] In one of the embodiments, the apparatus further comprises:

[0475] An indication information determining module is configured to obtain user scheduling result indication information corresponding to each of the terminal devices based on user feature information of each of the terminal devices, wherein the user feature information comprises at least one of user data feature information and user channel state information.

[0476] The first configuration message containing the user scheduling result indication information is sent to each of the terminal devices.

[0477] In one of the embodiments, the indication information determining module is specifically configured to obtain a preconfigured number of scheduled users.

[0478] Among all users in the coverage range of the base station, the terminal devices of the number of scheduled users are screened out based on at least one of the user data feature information and the user channel state information of each of the terminal devices, and the user scheduling result indication information is determined as scheduled indication information.

[0479] The model inference apparatus can also be referred to as a model inference apparatus based on hierarchical federated learning.

[0480] Each module in the model inference apparatus can be implemented by software, hardware, and combinations thereof, in whole or in part. Each module can be embedded in or independent of a processor in a computer device in hardware form, or stored in a memory in a computer device in software form, so as to be invoked and executed by a processor to perform operations corresponding to each module.

[0481] In an example embodiment, a computer device, which can be a server, is provided, and an internal structure diagram of the computer device can be as shown in FIG. 15. The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store data of model lifecycle management. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with terminals outside through a network connection. The computer program is executed by the processor to implement a model management method.

[0482] Those skilled in the art can understand that the structure shown in FIG. 15 is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. Specifically, the computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0483] In an example embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps described in the above embodiments.

[0484] In an embodiment, a computer readable storage medium is provided, which stores a computer program. The computer program is executed by a processor to implement the steps described in the above embodiments.

[0485] In an embodiment, a computer program product is provided, which includes a computer program. The computer program is executed by a processor to implement the steps described in the above embodiments.

[0486] It should be noted that the user information (including but not limited to terminal device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0487] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments of each method. In the embodiments provided in the present application, any reference to memory, database or other medium can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (Resistive Random Access Memory, ReRAM), magnetoresistive random access memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric memory (Ferroelectric Random Access Memory, FRAM), phase change memory (Phase Change Memory, PCM), graphene memory, etc. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (Artificial Intelligence, AI) processor, etc., without being limited thereto.

[0488] Any technical features in the above embodiments can be combined, and for the sake of brevity, not all possible combinations are described above, however, any combination of these technical features is deemed to be within the scope of the present application.

[0489] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be pointed out that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A model management method applied to a terminal device in a mobile communication system, the mobile communication system further comprising a network including network nodes and a plurality of base stations, each of the base stations corresponding to a plurality of terminal devices, the terminal devices being distributed nodes performing hierarchical federated learning, and the base stations being central nodes performing hierarchical federated learning; the method comprising: receiving a first configuration message sent by the base station, the first configuration message containing user scheduling result indication information based on hierarchical federated learning, the user scheduling result indication information being obtained by the base station based on user feature information of each of the terminal devices, the user feature information including at least one of user data feature information and user channel state information; sending reporting information to the network, the network storing a hierarchical federated learning model generated by the base station and a distribution policy, the reporting information including at least one of terminal capability, inference task type, and model performance monitoring feedback information, the reporting information being used by the network to generate a second configuration message, the second configuration message being model inference policy distribution information or model update policy distribution information based on hierarchical federated learning; and receiving the second configuration message, and obtaining a model inference result or a model update result based on the second configuration message. The terminal capability includes one or more of terminal computing capability, terminal channel state, and inference feature, the terminal computing capability being used to represent a model complexity supported by the terminal device, and the terminal channel state representing an uplink channel state of the terminal device. The sending of the reporting information to the network comprises: sending the reporting information to the base station based on a target transmission mode, so that the base station sends the reporting information to the network node, the target transmission mode including at least one of system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message. The model performance monitoring feedback information includes one or more of system performance index, accuracy index, complexity index, and generalization performance index.

2. The method of claim 1, wherein, 5. The method of claim 4, further comprising: if it is determined that the model performance monitoring feedback information meets a preset update condition with a preset index threshold, sending a model update request to the base station, so that the network responds to the model update request, the model update request including one or more of a to-be-updated model, a to-be-updated function identifier, and reference information.

3. The method of claim 1, wherein, The receiving of the second configuration message comprises: receiving the second configuration message sent by the base station based on a target transmission mode, the second configuration message being sent by the network node to the base station, and the target transmission mode including at least one of system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message. The second configuration message is model update policy distribution information, the model update policy distribution information including a to-be-updated model and an update policy; and the obtaining of the model inference result or the model update result based on the second configuration message comprises:

4. The method of claim 1, wherein, ​ ​ ​ 6. The method of claim 1, wherein, ​ ​ 7. The method of claim 1, wherein, ​ The updating strategy is used to update the to-be-updated model, and a model updating result is obtained. The to-be-updated model includes a to-be-updated model or a to-be-updated model group. The updating strategy includes one or more of a local updating strategy, a global updating strategy, a model replacement strategy, and a model deletion strategy.

8. The method of claim 7, wherein, The updating process includes at least one of the following: updating part of the model to obtain a model updating result; updating the model parameters of each to-be-updated model based on a target updating algorithm to obtain a model updating result. The target updating algorithm includes one or more of a momentum gradient descent algorithm, a stochastic gradient descent algorithm, and a small batch gradient descent algorithm; updating the model parameters of each to-be-updated model based on new model parameters to obtain a model updating result; deleting the model parameters, model content, and function identifier of the updated model to obtain a model updating result.

9. The method of claim 1, wherein, The second configuration message is a model inference strategy delivery information. The model inference strategy delivery information includes a model identifier and a model file of a to-be-inferred model, and identifier information of a terminal device to be executed for model inference. The model inference result or the model updating result is obtained based on the second configuration message, including: performing online model inference or offline model inference based on the model identifier and the model file of the to-be-inferred model and the identifier information of the terminal device to be executed for model inference to obtain a model inference result.

10. A model management method applied to a network node in a mobile communication system. The network includes a network node and a plurality of base stations. Each base station corresponds to a plurality of terminal devices. The method includes: receiving report information sent by the terminal device through the base station. The report information includes at least one of terminal capability, inference task type, and model performance monitoring feedback information; generating a second configuration message based on the report information and sending the second configuration message to the terminal device through the base station. The second configuration message is a model inference strategy delivery information or a model updating strategy delivery information based on hierarchical federated learning. The second configuration message is used to obtain a model inference result or a model updating result by the terminal device.

11. The method of claim 10, wherein, The terminal capability includes one or more of terminal computing capability, terminal channel state, and inference feature. The terminal computing capability is used to represent the model complexity supported by the terminal device. The terminal channel state represents the uplink channel state on the terminal device side.

12. The method of claim 10, wherein, The report information includes the terminal capability. The receiving of the report information sent by the terminal device through the base station includes: receiving the terminal capability reported by the terminal device through the base station based on a target transmission mode. The target transmission mode includes at least one of system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

13. The method of claim 10, further comprising: receiving a model inference request signaling sent by the terminal device through the base station, and returning a response message corresponding to the inference request, so that the terminal device sends report information based on the response message.

14. The method of claim 10, wherein, The second configuration message comprises model inference strategy issuing information or model update strategy issuing information based on hierarchical federated learning; The second configuration message is generated based on the reported information, comprising: Based on the reported information, at least one to-be-inferred model in the plurality of first models is determined, which satisfies a preset matching condition of a target dimension; Based on the model identifier of the to-be-inferred model, the model file, and the identifier information of the terminal device to be executed for model inference, the model inference strategy issuing information based on hierarchical federated learning is generated.

15. The method of claim 11, wherein, The second configuration message is generated based on the reported information, comprising: Based on the reported information, a preset number of to-be-inferred models in the plurality of first models are determined, which satisfy a preset matching condition of model structure, model complexity, and model accuracy.

16. The method of claim 14, wherein, The second configuration message is transmitted to the terminal device by the base station, comprising: The second configuration message is transmitted to the base station through a user plane interface, so that the base station transmits the second configuration message to the terminal device through the target transmission mode, and the target transmission mode comprises at least one of system message, RRC signaling, MAC CE signaling, DCI signaling, and NAS message.

17. The method of claim 14, wherein, The model performance monitoring feedback information comprises one or more of system performance indicators, accuracy indicators, complexity indicators, and generalization performance indicators.

18. The method of claim 14, wherein, The reported information comprises model performance monitoring feedback information, and the second configuration message is generated based on the reported information, comprising: Receiving model performance monitoring feedback information; Receiving a model update request, the model update request comprising a to-be-updated model / to-be-updated function identifier, reference information, and the model update request being generated in a case where the model performance monitoring feedback information of the terminal device and a preset indicator threshold satisfy a preset update condition; Responding to the model update request to generate a second configuration message.

19. The method of claim 14, wherein, The reported information comprises model performance monitoring feedback information, and the second configuration message is generated based on the reported information, comprising: Receiving model performance monitoring feedback information; Based on a preset time period, triggering a model update event; or, if the model performance monitoring feedback information and a preset indicator threshold satisfy a preset update condition, triggering an update event; the preset time period is associated with terminal capability and auxiliary information; Based on the triggered update event, a second configuration message is generated.

20. The method of claim 18 or 19, wherein, The second configuration message comprises model inference strategy issuing information based on hierarchical federated learning; The second configuration message is generated, comprising: Determining at least one to-be-updated model; Based on the to-be-updated model, the identifier information of the terminal device to be executed for model update, and the update strategy, the model inference strategy issuing information based on hierarchical federated learning is generated.

21. The method of claim 20, wherein, The update strategy comprises one or more of a local update strategy, a global update strategy, a model replacement strategy, and a model deletion strategy.

22. The method of claim 18, further comprising: Receiving model data transmitted by each of the base stations, the model data comprising a model file, a model content / function identifier; Determine each first model based on each model data, and arrange each first model according to an evaluation index of each first model to obtain a first model list.

23. The method of claim 22, further comprising: receiving an identification request and model capability from the base station, wherein the identification request comprises a model identification request and a function identification request, and the model capability is used to indicate a model supported by the base station and a function supported by the base station.

24. The method of claim 22, further comprising: sending an identification request to each base station to make each base station report a model capability, wherein the model capability is used to indicate a model supported by the base station and a function supported by the base station.

25. A model management method, wherein, A base station applied to a mobile communication system, the mobile communication system comprising a network and a plurality of base stations, each of the base stations corresponding to a plurality of terminal devices, the terminal devices being distributed nodes performing hierarchical federated learning, and the base station being a central node performing hierarchical federated learning; the method comprising: obtaining user scheduling result indication information corresponding to each terminal device based on user feature information of each terminal device, and sending a first configuration message to each terminal device, wherein the user feature information comprises at least one of user data feature information and user channel state information, and the first configuration message comprises the user scheduling result indication information; receiving report information sent by the terminal device, wherein the report information comprises at least one of terminal capability, inference task type, and model performance monitoring feedback information; sending the report information to the network node to make the network node obtain a second configuration message; the second configuration message is a model inference strategy or a model update strategy under hierarchical federated learning, and the second configuration message is used to make the terminal device obtain a model inference result or a model update result.

26. The method of claim 25, further comprising: sending model data to the network node, wherein the model data comprises a model file and a model content / function identifier, so that the network node determines each first model based on each model data and obtains a first model list.

27. The method of claim 25, further comprising: sending an identification request and model capability to the network node, wherein the identification request comprises a model identification request and a function identification request, and the model capability is used to indicate a model supported by the base station and a function supported by the base station.

28. The method of claim 25, further comprising: receiving an identification request sent by the network node, and reporting a model capability to the network node based on the identification request, wherein the model capability is used to indicate a model supported by the base station and a function supported by the base station.

29. The method of claim 25, further comprising: obtaining a global model trained based on model training results sent by each terminal device.

30. The method of claim 29, wherein, The terminal device is a user-scheduled terminal device, and obtaining a global model trained based on model training results sent by each terminal device comprises: For the i-th round of training, based on the training results sent by each terminal device in the (i-1)-th round, the model parameters of the global model in the i-th round are obtained; In the case where the preset training completion condition is not met, the base station broadcasts the model parameters of the global model in the i-th round to each terminal device, so that each terminal device trains based on the model parameters of the global model in the i-th round until the base station determines that the preset training completion condition is met, and obtains the trained global model.

31. The method of claim 30, wherein, The preset training completion condition includes one or more of the following: the number of training rounds reaches a preset upper limit value, the loss function of the global model is less than or equal to a preset value, and the difference between the loss functions of the global model in adjacent two rounds is less than or equal to a preset value.

32. The method of claim 25, further comprising: obtaining user scheduling result indication information corresponding to each terminal device based on user feature information of each terminal device, the user feature information including at least one of user data feature information and user channel state information; sending a first configuration message to each terminal device, the first configuration message containing the user scheduling result indication information.

33. The method of claim 32, wherein, The user scheduling result indication information corresponding to each terminal device is obtained based on user feature information of each terminal device, including: obtaining a pre-configured number of scheduled users; Among all users in the coverage range of the base station, based on at least one of the user data feature information and the user channel state information of each terminal device, the terminal devices of the number of scheduled users are screened out, and the user scheduling result indication information is determined as the scheduled indication information.

34. A model inference device applied to a terminal device in a mobile communication system, the mobile communication system further comprising a network, the network including a network node and a plurality of base stations, each base station corresponding to a plurality of terminal devices, the terminal device being a distributed node performing hierarchical federated learning, and the base station being a central node performing hierarchical federated learning; the device comprising: a first receiving module configured to receive a first configuration message sent by the base station, the first configuration message containing user scheduling result indication information based on hierarchical federated learning, the user scheduling result indication information being obtained by the base station based on user feature information of each terminal device, the user feature information including at least one of user data feature information and user channel state information; a first sending module configured to send reporting information to the network, the network storing a hierarchical federated learning model generated by the base station and a distribution strategy, the reporting information including at least one of terminal capability, inference task type, and model performance monitoring feedback information, the reporting information being used by the network to generate a second configuration message, the second configuration message being model inference strategy distribution information or model update strategy distribution information based on hierarchical federated learning; a second receiving module configured to receive the second configuration message, and obtain a model inference result or a model update result based on the second configuration message. 35.A model management apparatus applied to a network node in a mobile communication system, the network comprising a network node and a plurality of base stations, each of the base stations corresponding to a plurality of terminal devices, the apparatus comprising: a third receiving module configured to receive report information sent by the terminal devices through the base stations, the report information comprising at least one of terminal capability, inference task type, and model performance monitoring feedback information; a first generating module configured to generate a second configuration message based on the report information, and send the second configuration message to the terminal devices through the base stations; the second configuration message being model inference strategy or model update strategy issuing information based on hierarchical federated learning, and the second configuration message being used to make the terminal devices obtain model inference results or model update results. 36.A model management apparatus applied to a base station in a mobile communication system, the mobile communication system comprising a network and a plurality of base stations, each of the base stations corresponding to a plurality of terminal devices, the terminal devices being distributed nodes performing hierarchical federated learning, and the base station being a central node performing hierarchical federated learning; the apparatus comprising: a second sending module configured to obtain user scheduling result indication information corresponding to each of the terminal devices based on user feature information of each of the terminal devices, and send a first configuration message to each of the terminal devices, the user feature information comprising at least one of user data feature information and user channel state information, and the first configuration message comprising the user scheduling result indication information; a fourth receiving module configured to receive report information sent by the terminal devices, the report information comprising at least one of terminal capability, inference task type, and model performance monitoring feedback information; a third sending module configured to send the report information to the network node, so that the network node obtains a second configuration message; the second configuration message being model inference strategy or model update strategy issuing information based on hierarchical federated learning, and the second configuration message being used to make the terminal devices obtain model inference results or model update results.

37. A computer device comprising a memory and a processor, the memory storing a computer program, wherein, The processor executes the computer program to implement the steps of the method in any one of claims 1 to 33.

38. A computer readable storage medium having stored thereon a computer program, wherein, The computer program is executed by the processor to implement the steps of the method in any one of claims 1 to 33.

39. A computer program product comprising a computer program, wherein, The computer program is executed by the processor to implement the steps of the method in any one of claims 1 to 33.

Images