Model training method and model training device

Abstract

The invention relates to a model training method and a model training device. The method comprises the following steps: receiving a first model training demand sent by a model consumer, determining a first model training strategy according to the first model training demand, and training a to-be-trained model according to the first model training strategy and data required by the training model to obtain a trained target model. According to the method, the built-in intelligent capability in the RAN is realized, that is, the RAN has the capability of model training, the model training requirements of model consumers can be met, different model training strategies are formulated according to different model training requirements, model segmentation and integration of complex models can be performed, and the model training efficiency is improved. The network function module of the RAN side and the network function module of the core network side jointly complete model training, and the efficiency and the real-time performance of model training can be improved to a certain extent.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication, in particular to a model training method and a model training device. BACKGROUND

[0002] With the gradual maturity of AI technology, 6G network architecture is constantly enriched, and 6G network will be deeply integrated with artificial intelligence technology. It has been a consensus in the industry that intelligent endogeny is an important feature of 6G network.

[0003] In the current 6G network architecture prospect, the network function layer proposes connection function, computing function, AI function, perception function, data function, and security and trust, and management and arrangement. Among them, the AI function refers to the network itself operation and operation and maintenance, as well as the user or external AI service, through the endogenous construction of AI elements (computing power, algorithm, data) ability, under the arrangement management and control of integrated AI and connection elements, to realize the on-demand customization and high QoS right protection of AI service.

[0004] However, the industry has proposed 6G intelligent endogenous architecture, but only points out the basic function related to AI, and does not give specific description and design on how to train the AI model in the network. SUMMARY

[0005] The embodiments of the present application provide a model training method and a model training device, which provide multiple scenarios and possibilities for the AI model training method, and improve the training efficiency and real-time performance of the AI model.

[0006] In a first aspect, the embodiments of the present application provide a model training method, which is applied to RAN, and the method comprises:

[0007] receiving a first model training requirement sent by a model consumer;

[0008] determining a first model training strategy according to the first model training requirement;

[0009] training a to-be-trained model according to the first model training strategy and data required for training the model, to obtain a target model after training.

[0010] In one of the embodiments, the first model training strategy comprises any of the following:

[0011] training the model on the RAN;

[0012] training the model on a network function module on the RAN side;

[0013] jointly training the model with the network function module on the RAN side;

[0014] Collaborate with network function modules on the core network side for model training.

[0015] In one embodiment, the joint training of the model with the network functional modules on the RAN side includes:

[0016] The model can be trained collaboratively with the network functional modules on the RAN side, or incrementally trained with the network functional modules on the RAN side.

[0017] In one embodiment, the first model training strategy includes training the model on the RAN, and the step of training the model to be trained according to the first model training strategy and the data required for training the model to obtain the trained target model includes:

[0018] The data required for training the model is preprocessed;

[0019] The training model is trained based on the data required by the preprocessed training model to obtain the trained target model.

[0020] In one embodiment, the first model training strategy includes training the model using the network functional modules on the RAN side. The step of training the model to be trained according to the first model training strategy and the data required for training the model to obtain the trained target model includes:

[0021] A first model training request is sent to the network function module on the RAN side; the first model training request is used to instruct the network function module on the RAN side to train the model to be trained according to the first model training request to obtain the trained target model; the first model training request carries the model requirements for training the model to be trained, and also carries the data requirements for training the model to be trained or the model data for training the model to be trained.

[0022] Receive the target model sent by the network function module on the RAN side.

[0023] In one embodiment, the first model training strategy includes co-training the model with the network functional modules on the RAN side. The step of training the model to be trained according to the first model training strategy and the data required for training the model to obtain the trained target model includes:

[0024] The model to be trained is segmented to obtain a first RAN model and a first network function model;

[0025] The first RAN model is trained according to the data required for the training model to obtain the trained first RAN model;

[0026] Receive the trained first network function model sent by the network function module on the RAN side;

[0027] The first RAN model and the first network function model after training are integrated to obtain the target model.

[0028] In one embodiment, the method further includes:

[0029] Send a first model co-training request to the network function module on the RAN side;

[0030] The first model collaborative training request carries the model requirement for training the first network functional model, and also carries the data requirement for training the first network functional model or the model data for training the first network functional model.

[0031] The first model collaborative training request is used to instruct the network function module on the RAN side to train the first network function model according to the first model collaborative training request, so as to obtain the trained first network function model.

[0032] In one embodiment, the first model training strategy includes incremental model training with the network functional modules on the RAN side. The step of training the model to be trained according to the first model training strategy and the data required for training the model to obtain the trained target model includes:

[0033] The model to be trained is trained according to the data required for the training model to obtain the first intermediate model after training;

[0034] Receive the trained target model sent by the network function module on the RAN side.

[0035] In one embodiment, the method further includes:

[0036] Send a first model incremental training request to the network function module on the RAN side;

[0037] The first model incremental training request carries the first intermediate model, as well as the data requirement for training the first intermediate model or the model data for training the first intermediate model.

[0038] The first model incremental training request is used to instruct the network function module on the RAN side to train the first intermediate model according to the first model incremental training request, so as to obtain the trained target model.

[0039] In one embodiment, the first model training strategy includes incremental model training with the network functional modules on the RAN side. The step of training the model to be trained according to the first model training strategy and the data required for training the model to obtain the trained target model includes:

[0040] Receive the trained second intermediate model sent by the network function module on the RAN side;

[0041] The second intermediate model is trained using the data required for the training model to obtain the trained target model.

[0042] In one embodiment, the method further includes:

[0043] Send a second model incremental training request to the network function module on the RAN side;

[0044] The second model incremental training request carries the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained.

[0045] The second model incremental training request is used to instruct the network function module on the RAN side to train the model to be trained according to the second model incremental training request to obtain the second intermediate model.

[0046] In one embodiment, the first model training strategy includes co-training the model with network functional modules on the core network side. The step of training the model to be trained according to the first model training strategy and the data required for training the model to obtain the trained target model includes:

[0047] The model to be trained is segmented to obtain a second RAN model and a second network functional model;

[0048] The second RAN model is trained based on the data required for the training model to obtain the trained second RAN model;

[0049] Receive the trained second network function model sent by the network function module on the core network side;

[0050] The trained second RAN model and the trained second network function model are integrated to obtain the target model.

[0051] In one embodiment, the method further includes:

[0052] Send a second model collaborative training request to the network function module on the core network side;

[0053] The second model collaborative training request carries the model requirements for training the second network functional model, as well as the data requirements for training the second network functional model or the model data for training the second network functional model.

[0054] The second model collaborative training request is used to instruct the network function module on the core network side to train the second network function model according to the second model collaborative training request, so as to obtain the trained second network function model.

[0055] In one embodiment, determining the first model training strategy based on the first model training requirements includes:

[0056] The training requirements of the first model are analyzed to determine whether the computing resources required for the training of the first model exceed the computing resources of the RAN.

[0057] If the computing resources required for the training of the first model do not exceed the computing resources of the RAN, then the first model training strategy is determined to be to train the model on the RAN.

[0058] If the computing resources required for the training of the first model exceed the computing resources of the RAN, then the first model training strategy is determined to include any one of the following: training the model on the network function module on the RAN side, joint training the model with the network function module on the RAN side, and collaborative training the model with the network function module on the core network side.

[0059] In one embodiment, the method further includes:

[0060] Set and start the first timer, and acquire the data required for training the model;

[0061] When the first timer reaches its set time, the data required for training the model is sent to the network function module on the RAN side.

[0062] In one embodiment, the method further includes:

[0063] Upon receiving the first data report response, restart the first timer.

[0064] In one embodiment, the method further includes:

[0065] Receive the first data subscription request sent by the network function module on the RAN side;

[0066] Send the data required for training the model corresponding to the first data subscription request to the network function module on the RAN side.

[0067] In one embodiment, the method further includes:

[0068] Determine whether the data required for the training model has been updated;

[0069] If it is determined that the data required for the training model has been updated, the updated data required for the training model will be sent to the network function module on the RAN side.

[0070] In one embodiment, the method further includes:

[0071] The target model is deployed on the RAN.

[0072] Secondly, embodiments of this application provide a method for training a model, the method being applied to a Random Access Array (RAN), the method comprising:

[0073] Receive the second model training request sent by the model consumer;

[0074] Send the second model training requirement to the network function module on the RAN side to instruct the network function module on the RAN side to train the model to be trained according to the second model training requirement, so as to obtain the trained target model.

[0075] Receive the target model sent by the network function module on the RAN side.

[0076] In one embodiment, the method further includes:

[0077] The target model is deployed on the RAN.

[0078] Thirdly, embodiments of this application provide a model training method, which is applied to a network functional module on the RAN side, and the method includes:

[0079] Receive a first training request sent by the RAN; the first training request includes one of a first model training request, a first model co-training request, a first model incremental training request, and a second model incremental training request;

[0080] The model is trained according to the first training request to obtain the trained model.

[0081] The trained model is sent to the RAN.

[0082] In one embodiment, the step of training the model according to the first training request to obtain the trained model includes:

[0083] Obtain the model to be trained and the model data for training the model to be trained according to the first training request;

[0084] The model to be trained is trained based on the model data used to train the model to be trained, and the trained model is obtained.

[0085] In one embodiment, the first training request carries a model requirement for training the model to be trained, and the step of obtaining the model to be trained according to the first training request includes:

[0086] The model to be trained is obtained according to the model requirements for training the model to be trained.

[0087] In one embodiment, the first training request carries a model to be trained, and the step of obtaining the model to be trained according to the first training request includes:

[0088] Extract the model to be trained from the first training request.

[0089] In one embodiment, the first training request further carries a data requirement for training the model to be trained or model data for training the model to be trained, and the step of obtaining the model data for training the model to be trained according to the first training request includes:

[0090] Based on the data requirements for training the model to be trained, the model data of the model to be trained is collected, or the model data of the model to be trained is extracted from the first training request.

[0091] In one embodiment, the method further includes:

[0092] The model data for training the model to be trained is preprocessed;

[0093] The step of training the model to be trained based on the model data to obtain the trained model includes:

[0094] The model to be trained is trained based on the preprocessed model data to obtain the trained model.

[0095] In one embodiment, the method further includes:

[0096] Receive the data required for training the model sent by RAN and start the second timer;

[0097] Send a first data reporting response to the RAN;

[0098] When the second timer reaches its set time, expired training model data stored locally will be deleted, or training model data exceeding local storage limits will be deleted.

[0099] In one embodiment, the method further includes:

[0100] Send a first data subscription request to the RAN;

[0101] Receive the data required for training the model corresponding to the first data subscription request sent by the RAN.

[0102] Fourthly, embodiments of this application provide a model training method, which is applied to a network functional module on the RAN side, the method comprising:

[0103] Receive the second model training request sent by RAN;

[0104] The training model is trained according to the training requirements of the second model to obtain the trained target model.

[0105] The target model is sent to the RAN.

[0106] In one embodiment, training the model to be trained according to the second model training requirements to obtain the trained target model includes:

[0107] Determine the training strategy for the second model based on the training requirements of the second model;

[0108] The model to be trained is trained according to the second model training strategy and the data required for training the model, and the trained target model is obtained.

[0109] In one embodiment, the second model training strategy includes any of the following:

[0110] Model training is performed on the network functional modules on the RAN side;

[0111] Joint training of the model with RAN;

[0112] Collaborate with network function modules on the core network side for model training.

[0113] In one embodiment, the joint training of the model with the RAN includes:

[0114] Perform model co-training with RAN, or perform incremental model training with RAN.

[0115] Fifthly, embodiments of this application provide a method for training a model, the method being applied to an intelligent network element, the method comprising:

[0116] Receive third-model training requests sent by model consumers;

[0117] Determine the training strategy for the third model based on the training requirements of the third model;

[0118] The training model is trained according to the third model training strategy and the data required for training the model, and the trained target model is obtained.

[0119] In one embodiment, the third model training strategy includes any of the following:

[0120] Model training is performed on intelligent network elements;

[0121] Joint training of models is conducted with network functional modules on the core network side.

[0122] In one embodiment, the joint model training with the network functional modules on the core network side includes:

[0123] The model can be trained collaboratively with the network functional modules on the core network side, or incrementally trained with the network functional modules on the core network side.

[0124] In one embodiment, the third model training strategy includes training the model on the intelligent network element, wherein training the model to be trained according to the third model training strategy and the data required for training the model to obtain the trained target model includes:

[0125] The data required for training the model is preprocessed;

[0126] The training model is trained based on the data required by the preprocessed training model to obtain the trained target model.

[0127] In one embodiment, the third model training strategy includes co-training the model with network functional modules on the core network side. The step of training the model to be trained according to the third model training strategy and the data required for training the model to obtain the trained target model includes:

[0128] The model to be trained is segmented to obtain an intelligent network element model and a third network function model;

[0129] The intelligent network element model is trained based on the data required for the training model to obtain the trained intelligent network element model.

[0130] Receive the trained third network function model sent by the network function module on the core network side;

[0131] The trained third network function model and the trained intelligent network element model are integrated to obtain the target model.

[0132] In one embodiment, the method further includes:

[0133] Send a third model collaborative training request to the network function module on the core network side;

[0134] The third model collaborative training request carries the model requirements for training the third network functional model, as well as the data requirements for training the third network functional model or the model data for training the third network functional model.

[0135] The third model collaborative training request is used to instruct the network function modules on the core network side to train the third network function model according to the third model collaborative training request, so as to obtain the trained third network function model.

[0136] In one embodiment, the third model training strategy includes incremental model training with network functional modules on the core network side. The step of training the model to be trained according to the third model training strategy and the data required for training the model to obtain the trained target model includes:

[0137] Receive the trained third intermediate model sent by the network function module on the core network side;

[0138] The third intermediate model is trained using the data required for the training model to obtain the trained target model.

[0139] In one embodiment, the method further includes:

[0140] Send a third model incremental training request to the network function module on the core network side;

[0141] The third model incremental training request carries the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained.

[0142] The third model incremental training request is used to instruct the network function module on the core network side to train the model to be trained according to the third model incremental training request, so as to obtain the trained third intermediate model.

[0143] In one embodiment, the third model training strategy includes incremental model training with network functional modules on the core network side. The step of training the model to be trained according to the third model training strategy and the data required for training the model to obtain the trained target model includes:

[0144] The model to be trained is trained based on the data required for the training model to obtain the fourth intermediate model;

[0145] Receive the trained target model sent by the network function module on the core network side.

[0146] In one embodiment, the method further includes:

[0147] Send a fourth model incremental training request to the network function module on the core network side;

[0148] The fourth model incremental training request carries a fourth intermediate model, as well as the data requirement for training the fourth intermediate model or the model data for training the fourth intermediate model.

[0149] The fourth model incremental training request is used to instruct the network function module on the core network side to train the fourth intermediate model according to the fourth model incremental training request, so as to obtain the trained target model.

[0150] In one embodiment, determining the third model training strategy based on the third model training requirements includes:

[0151] The training requirements of the third model are analyzed to determine whether the computing resources required for the training of the third model exceed the computing resources of the intelligent network element.

[0152] If the computing resources required for the training of the third model do not exceed the computing resources of the network function module on the core network side, then the training strategy for the third model is determined to be to train the model on the intelligent network element.

[0153] If the computing resources required for training the third model exceed the computing resources of the network function module on the core network side, then the training strategy for the third model is determined to be joint training of the model with the network function module on the core network side.

[0154] In one embodiment, the method further includes:

[0155] Set and start the third timer, and acquire the data required for training the model;

[0156] When the third timer reaches its set time, the data required for training the model is sent to the network function module on the core network side.

[0157] In one embodiment, the method further includes:

[0158] Upon receiving the second data report response, the third timer is restarted.

[0159] In one embodiment, the method further includes:

[0160] Receive the second data subscription request sent by the network function module on the core network side;

[0161] Send the data required for training the model corresponding to the second data subscription request to the network function module on the core network side.

[0162] In one embodiment, the method further includes:

[0163] Determine whether the data required for the training model has been updated;

[0164] If it is determined that the data required for the training model has been updated, the updated data required for the training model will be sent to the network function module on the core network side.

[0165] In one embodiment, the method further includes:

[0166] The target model is deployed on the intelligent network element.

[0167] Sixthly, embodiments of this application provide a method for training a model, the method being applied to network functional modules on the core network side, the method comprising:

[0168] Receive a second training request sent by a target device; the target device is any one of RAN, RAN-side network function module, and intelligent network element; the second training request includes one of a third model collaborative training request, a third model incremental training request, and a fourth model incremental training request.

[0169] The model is trained according to the second training request to obtain the trained model;

[0170] The trained model is sent to the target device.

[0171] In one embodiment, the step of training the model according to the second training request to obtain the trained model includes:

[0172] Obtain the model to be trained and the model data for training the model to be trained according to the second training request;

[0173] The model to be trained is trained based on the model data used to train the model to be trained, and the trained model is obtained.

[0174] In one embodiment, the second training request carries a model requirement for training the model to be trained, and the step of obtaining the model to be trained according to the second training request includes:

[0175] The model to be trained is obtained according to the model requirements for training the model to be trained.

[0176] In one embodiment, the second training request carries a model to be trained, and obtaining the model to be trained according to the second training request includes:

[0177] The model to be trained is extracted from the second training request.

[0178] In one embodiment, the second training request further carries a data requirement for training the model to be trained or model data for training the model to be trained. The step of obtaining the model data for training the model to be trained according to the second training request includes:

[0179] Based on the data requirements for training the model to be trained, the model data of the model to be trained is collected, or the model data of the model to be trained is extracted from the second training request.

[0180] In one embodiment, collecting the model data for training the model to be trained according to the data requirements of the model to be trained includes:

[0181] Generate a model data collection request based on the data requirements for training the model to be trained;

[0182] Send the model data collection request to other intelligent network elements or data planes;

[0183] Receive model data returned by other intelligent network elements or data planes based on the model data collection request.

[0184] In one embodiment, the method further includes:

[0185] The model data for training the model to be trained is preprocessed;

[0186] The step of training the model to be trained based on the model data to obtain the trained model includes:

[0187] The model to be trained is trained based on the preprocessed model data to obtain the trained model.

[0188] In one embodiment, the method further includes:

[0189] Receive the data required for training the model sent by the intelligent network element and start the fourth timer;

[0190] Send a second data reporting response to the intelligent network element;

[0191] When the fourth timer reaches its set time, expired training model data stored locally will be deleted, or training model data exceeding local storage limits will be deleted.

[0192] In one embodiment, the method further includes:

[0193] Send a second data subscription request to the intelligent network element;

[0194] Receive the data required for training the model corresponding to the second data subscription request sent by the intelligent network element.

[0195] In a seventh aspect, embodiments of this application provide a training apparatus for a model, the apparatus comprising: a memory, a transceiver, and a processor; the memory for storing a computer program; the transceiver for transmitting and receiving data under the control of the processor; and the processor for reading the computer program from the memory and performing the following operations:

[0196] Receive the first model training request sent by the model consumer;

[0197] Determine the first model training strategy based on the first model training requirements;

[0198] The model to be trained is trained according to the first model training strategy and the data required for training the model, and the trained target model is obtained.

[0199] Eighthly, embodiments of this application provide a training apparatus for a model, the apparatus comprising: a memory, a transceiver, and a processor; the memory for storing a computer program; the transceiver for transmitting and receiving data under the control of the processor; and the processor for reading the computer program from the memory and performing the following operations:

[0200] Receive the second model training request sent by the model consumer;

[0201] Send the second model training requirement to the network function module on the RAN side to instruct the network function module on the RAN side to train the model to be trained according to the second model training requirement, so as to obtain the trained target model.

[0202] Receive the target model sent by the network function module on the RAN side.

[0203] Ninthly, embodiments of this application provide a model training apparatus, the apparatus comprising: a memory, a transceiver, and a processor.

[0204] A memory for storing computer programs; a transceiver for sending and receiving data under the control of the processor; and a processor for reading the computer programs from the memory and performing the following operations:

[0205] Receive a first training request sent by the RAN; the first training request includes one of a first model training request, a first model co-training request, a first model incremental training request, and a second model incremental training request;

[0206] The model is trained according to the first training request to obtain the trained model.

[0207] The trained model is sent to the RAN.

[0208] In a tenth aspect, embodiments of this application provide a training apparatus for a model, the apparatus comprising: a memory, a transceiver, and a processor; the memory for storing a computer program; the transceiver for transmitting and receiving data under the control of the processor; and the processor for reading the computer program from the memory and performing the following operations:

[0209] Receive the second model training request sent by RAN;

[0210] The training model is trained according to the training requirements of the second model to obtain the trained target model.

[0211] The target model is sent to the RAN.

[0212] In the eleventh aspect, embodiments of this application provide a model training apparatus, the apparatus comprising: a memory, a transceiver, and a processor.

[0213] A memory for storing computer programs; a transceiver for sending and receiving data under the control of the processor; and a processor for reading the computer programs from the memory and performing the following operations:

[0214] Receive third-model training requests sent by model consumers;

[0215] Determine the training strategy for the third model based on the training requirements of the third model;

[0216] The training model is trained according to the third model training strategy and the data required for training the model, and the trained target model is obtained.

[0217] In a twelfth aspect, embodiments of this application provide a model training apparatus, the apparatus comprising: a memory, a transceiver, and a processor; the memory for storing a computer program; the transceiver for transmitting and receiving data under the control of the processor; and the processor for reading the computer program from the memory and performing the following operations:

[0218] Receive a second training request sent by a target device; the target device is any one of RAN, RAN-side network function module, and intelligent network element; the second training request includes one of a third model collaborative training request, a third model incremental training request, and a fourth model incremental training request.

[0219] The model is trained according to the second training request to obtain the trained model;

[0220] The trained model is sent to the target device.

[0221] In a thirteenth aspect, embodiments of this application provide a model training apparatus, the apparatus comprising:

[0222] The first receiving module is used to receive the first model training request sent by the model consumer.

[0223] The first determining module is used to determine the first model training strategy based on the first model training requirements.

[0224] The first training module is used to train the model to be trained according to the first model training strategy and the data required for training the model, so as to obtain the trained target model.

[0225] In a fourteenth aspect, embodiments of this application provide a model training apparatus, the apparatus comprising:

[0226] The second receiving module is used to receive the second model training request sent by the model consumer.

[0227] The first sending module is used to send the second model training requirement to the network function module on the RAN side, so as to instruct the network function module on the RAN side to train the model to be trained according to the second model training requirement, and obtain the trained target model.

[0228] The third receiving module is used to receive the target model sent by the network function module on the RAN side.

[0229] In a fifteenth aspect, embodiments of this application provide a model training apparatus, the apparatus comprising:

[0230] The fourth receiving module is used to receive a first training request sent by the RAN; the first training request includes one of a first model training request, a first model co-training request, a first model incremental training request, and a second model incremental training request.

[0231] The second training module is used to train the model according to the first training request to obtain the trained model.

[0232] The second sending module is used to send the trained model to the RAN.

[0233] In a sixteenth aspect, embodiments of this application provide a model training apparatus, the apparatus comprising:

[0234] The fifth receiving module is used to receive the second model training request sent by the RAN;

[0235] The third training module is used to train the model to be trained according to the training requirements of the second model, so as to obtain the trained target model.

[0236] The third sending module is used to send the target model to the RAN.

[0237] In a seventeenth aspect, embodiments of this application provide a model training apparatus, the apparatus comprising:

[0238] The sixth receiving module is used to receive the third model training request sent by the model consumer;

[0239] The second determining module is used to determine the training strategy of the third model based on the training requirements of the third model.

[0240] The fourth training module is used to train the model to be trained according to the third model training strategy and the data required for training the model, so as to obtain the trained target model.

[0241] Eighteenthly, embodiments of this application provide a model training apparatus, the apparatus comprising:

[0242] The seventh receiving module is used to receive a second training request sent by the target device; the target device is any one of RAN, RAN-side network function module, and intelligent network element; the second training request includes one of the third model collaborative training request, the third model incremental training request, and the fourth model incremental training request.

[0243] The fifth training module is used to train the model according to the second training request to obtain the trained model;

[0244] The fourth sending module is used to send the trained model to the target device.

[0245] In a nineteenth aspect, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any one of the first, second, third, fourth, fifth, and sixth aspects described above.

[0246] In a twentieth aspect, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements any one of the first, second, third, fourth, fifth, and sixth aspects described above.

[0247] The training method, training device, and storage medium described above involve receiving a first model training request from a model consumer, determining a first model training strategy based on the request, and training the model to be trained using the first training strategy and the required data to obtain the trained target model. This method incorporates built-in intelligent capabilities within the RAN (Radio Network Array), enabling the RAN to train models, meet the model training needs of model consumers, and formulate different model training strategies based on different training requirements. It can also perform model segmentation and integration for complex models, collaborating with both the RAN-side network function modules and the core network-side network function modules to complete model training, thereby improving the efficiency and real-time performance of model training to a certain extent. Attached Figure Description

[0248] Figure 1 This is a diagram illustrating the application environment of the model training method in one embodiment.

[0249] Figure 2 This is a schematic diagram of the structure of AIF in one embodiment;

[0250] Figure 3 This is one of the flowcharts illustrating the training method of the RAN model in one embodiment;

[0251] Figure 4 This is the second flowchart illustrating the training method of the RAN model in one embodiment;

[0252] Figure 5 This is the third flowchart illustrating the training method for the RAN model in one embodiment;

[0253] Figure 6 This is the fourth flowchart illustrating the training method of the RAN model in one embodiment;

[0254] Figure 7 This is the fifth flowchart illustrating the training method of the RAN model in one embodiment;

[0255] Figure 8 This is the sixth flowchart illustrating the training method of the RAN model in one embodiment;

[0256] Figure 9 This is the seventh flowchart illustrating the training method of the RAN model in one embodiment;

[0257] Figure 10 This is the eighth flowchart illustrating the training method of the RAN model in one embodiment;

[0258] Figure 11 This is the ninth flowchart illustrating the training method of the RAN model in one embodiment;

[0259] Figure 12 This is the tenth flowchart illustrating the training method of the RAN model in one embodiment;

[0260] Figure 13 This is eleventh of a flowchart illustrating the training method for the RAN model in one embodiment;

[0261] Figure 14 This is one of the flowcharts illustrating a training method for a model of a network functional module on the RAN side in one embodiment;

[0262] Figure 15 This is the second flowchart illustrating the training method of the network functional module model on the RAN side in one embodiment;

[0263] Figure 16 This is the third flowchart illustrating the training method for the network functional module model on the RAN side in one embodiment;

[0264] Figure 17 This is the fourth flowchart illustrating the training method for the network functional module model on the RAN side in one embodiment.

[0265] Figure 18 This is the fifth flowchart illustrating the training method for the network functional module model on the RAN side in one embodiment.

[0266] Figure 19 This is the sixth flowchart illustrating the training method for the network functional module model on the RAN side in one embodiment;

[0267] Figure 20 This is one of the interactive schematic diagrams of a model training method in one embodiment;

[0268] Figure 21 This is the second interactive schematic diagram of the model training method in one embodiment;

[0269] Figure 22 This is the third interactive schematic diagram of the model training method in one embodiment;

[0270] Figure 23 This is the third interactive schematic diagram of the model training method in one embodiment;

[0271] Figure 24 This is one of the flowcharts illustrating the training method of the model on the intelligent network element side in one embodiment;

[0272] Figure 25 This is the second flowchart illustrating the training method of the model on the intelligent network element side in one embodiment;

[0273] Figure 26This is the third flowchart illustrating the training method of the model on the intelligent network element side in one embodiment;

[0274] Figure 27 This is the fourth flowchart illustrating the training method of the model on the intelligent network element side in one embodiment;

[0275] Figure 28 This is the fifth flowchart illustrating the training method of the model on the intelligent network element side in one embodiment;

[0276] Figure 29 This is the sixth flowchart illustrating the training method of the model on the intelligent network element side in one embodiment;

[0277] Figure 30 This is the seventh flowchart illustrating the training method of the model on the intelligent network element side in one embodiment.

[0278] Figure 31 This is the eighth flowchart illustrating the training method of the model on the intelligent network element side in one embodiment.

[0279] Figure 32 This is one of the flowcharts illustrating the training method of the model on the core network side in one embodiment;

[0280] Figure 33 This is the second flowchart illustrating the training method of the model on the core network side in one embodiment;

[0281] Figure 34 This is the eighth flowchart illustrating the training method of the model on the core network side in one embodiment.

[0282] Figure 35 This is one of the flowcharts illustrating the training method of the model on the core network side in one embodiment;

[0283] Figure 36 This is the second flowchart illustrating the training method of the model on the core network side in one embodiment;

[0284] Figure 37 This is the fourth interactive schematic diagram of the model training method in one embodiment;

[0285] Figure 38 This is one of the structural block diagrams of a training device for a model in one embodiment;

[0286] Figure 39 This is a second structural block diagram of the training device for the model in one embodiment;

[0287] Figure 40 This is the third structural block diagram of the training device for the model in one embodiment;

[0288] Figure 41This is the fourth structural block diagram of the training device for the model in one embodiment;

[0289] Figure 42 This is the fifth structural block diagram of the training device for the model in one embodiment;

[0290] Figure 43 This is the sixth structural block diagram of the training device for the model in one embodiment;

[0291] Figure 44 This is the seventh structural block diagram of the training device for the model in one embodiment. Detailed Implementation

[0292] In the embodiments of this application, the term "and / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0293] In the embodiments of this application, the term "multiple" refers to two or more, and other quantifiers are similar.

[0294] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0295] Figure 1 This is a schematic diagram illustrating an application scenario of a model training method provided in an embodiment of this application. For example... Figure 1As shown, this scenario depicts a 6G AI network architecture, comprising User Equipment (UE) 100, Radio Access Network (RAN) 200, User Plane Function (UPF) 300, Data Network Name (DN) 400, and Core Network 500. RAN 200 includes a newly added Artificial Intelligence Feedback (AIF) 2001, which can be referred to as the RAN-side AIF. This AIF can be independently configured, forming a trusted domain with RAN 200, and interacting with RAN 200 to train and deploy intelligent models, thus providing targeted AI services. Core Network 500 includes Insertion Policy Control Function (iPCF), Insertion Network Exposure Function (iNEF), Insertion Unified Data Management (iUDM), and Insertion Application Function. The network functions include iAF (Insertion Access and Mobility Management Function), iAMF (Insertion Session Management Function), iSMF (Insertion Session Management Function), and iNRF (Insertion Network Repository Function). A new network function module, AIF5001, is added to the core network. This AIF can be independently configured, forming a trusted domain with the core network 500, and interacting with the core network 500 to train and deploy intelligent models, thus providing targeted AI services. Specifically, UE100 interacts with the core network 500 through the N1 interface, RAN200 through the N2 interface, and UPF300 through the N4 interface, and also interacts with DN400 through the N6 interface.It should be noted that the AIF2001 on the RAN side and the AIF5001 on the core network side possess computing and storage resources, and integrate all stages of the AI ​​lifecycle. They support the collection of data from both internal and external networks, and have built-in data preprocessing programs and scripts to achieve intelligent data delivery. They support AI model training, validation, inference, and updates. When managing AI model training, AIF considers the characteristics of the training and corresponding actions, such as the storage address after data collection, the storage address after data preprocessing, model evaluation, and training instructions. Furthermore, it needs to consider that algorithms written in different architectures (languages) require different versions of dependent environments (software versions, dependent library versions, etc.), and different algorithms need to be scheduled to run in different environments. Figure 2 As shown. Simultaneously, each network element evolves into an intelligent intrinsic network element, and the RAN evolves into an intelligent RAN. Each intelligent intrinsic network element and RAN, in addition to possessing basic functions, inherently possesses AI capabilities. Different network elements and RANs, tailored to specific needs, inherently deploy corresponding AI models and their respective dependent environments. Any network element in the core network can be an intelligent network element.

[0296] The equipment in RAN200 can be a base station (BTS) in Global System for Mobile communication (GSM) or Code Division Multiple Access (CDMA), a base station (NodeB, NB) in Wideband Code Division Multiple Access (WCDMA), an evolved Node B (eNB or eNodeB) in LTE, a relay station or access point, a base station in a 5G network, or a base station in a 6G network, etc., and is not limited here.

[0297] User terminal UE100 can be a device that provides voice and / or data connectivity to a user, a handheld device with wireless connectivity, or other processing devices connected to a wireless modem. The name of the terminal device may differ in different systems; for example, in a 5G system, the terminal device can be called User Equipment (UE). Wireless terminal devices can be USB storage devices, other personal computer memory devices, and dongles. They can also communicate with one or more core networks (CNs) via the RAN. Wireless terminal devices can be mobile terminal devices, such as mobile phones (or "cellular" phones) and computers with mobile terminal devices, for example, portable, pocket-sized, handheld, computer-embedded, or vehicle-mounted mobile devices that exchange voice and / or data with the radio access network. Examples of such devices include Personal Communication Service (PCS) telephones, cordless phones, Session Initiated Protocol (SIP) phones, Wireless Local Loop (WLL) stations, Personal Digital Assistants (PDAs), personal computers, tablets, and Machine-type Communication (MTC) terminal devices. Wireless terminal devices can also be referred to as systems, subscriber units, subscriber stations, mobile stations, mobile devices, remote stations, access points, remote terminals, access terminals, user terminals, user agents, user devices, and wireless access devices and routers / modems that meet the limitations of this definition; however, this application does not limit the scope of the embodiments.

[0298] Currently, artificial intelligence encompasses the training of computers to perform human-like behaviors such as autonomous learning, judgment, and decision-making. The deep integration of 6G networks with artificial intelligence technology, and the inherent intelligence of 6G networks as a key characteristic, has become a consensus in the industry. Currently, various 6G network architectures have been proposed for 6G mobile communication networks, most of which adopt a layered and faceted approach, adding intelligent, computing, data, and security planes to the 5G architecture.

[0299] As AI technology matures and 6G network architecture becomes increasingly sophisticated, AI services will be provided in an endogenous and distributed manner. Traditional 5G Network Data Analytics Functions (NWDAFs) perform simple intelligent predictions or analyses on the network through centralized external modules. However, these external NWDAFs are inefficient in data acquisition and model application, and are not direct consumers of the network's internal AI models (only the AnLF requests model training and inference from the MTLF, involving simple intelligent predictions or analyses). Furthermore, they lack the conditions and mechanisms for embedding and training AI models, thus failing to function as producers of AI models. In the vision of 6G network architecture, the network functional layer proposes connectivity, computing, AI, sensing, data, and security, trustworthiness, management, and orchestration. Among these, the AI ​​function in 6G networks refers to providing AI services to users or external parties, both for network operation and maintenance and through the endogenous construction of AI elements (computing power, algorithms, and data). Under the integrated orchestration, management, and control of AI and connectivity elements, it enables on-demand customization of AI services and high QoS + rights protection. Native AI capabilities will be a core feature of 6G networks, specifically including functions such as training, inference, and model validation. The network side will provide lifecycle management functions for AI models and data, such as AI data (e.g., training data reporting, gradient reporting), AI models (e.g., model generation, model storage, model selection, model distribution, model updates, model replacement, model activation and deactivation, model performance monitoring), and AI computing power (e.g., allocation of computing resources for training execution, inference execution, etc.).

[0300] Furthermore, the inherent AI capabilities in 6G networks will further enhance the intelligent communication experience, transforming "symbol-level" communication services focused on bit-level information transmission into "meaningful" communication services oriented towards high-level semantic delivery, thus breaking through the traditional "Shannon limit." By combining AI task orchestration and distributed computing power scheduling across edge, cloud, and device layers, multimodal, barrier-free communication among natural persons, digital persons, and intelligent and non-intelligent devices can ultimately be achieved. Consequently, the industry has proposed various 6G intelligent endogenous architectures, but these only specify basic AI-related functions, lacking concrete descriptions and designs regarding how to achieve intelligent endogenous transmission, training methods for AI models within the network, and the efficient collection and flow of data required for model training.

[0301] In view of this, for the intelligent endogenous network architecture proposed by 6G, this application proposes a training method for a model based on multi-source data collection, which provides a variety of scenarios and possibilities for AI model training methods, such as embedding AI models and model training capabilities in intelligent RAN and intelligent network elements, joint training or incremental training on the RAN side and the core network side, etc., to improve the training efficiency and real-time performance of AI models.

[0302] It should be noted that the beneficial effects or technical problems solved by the embodiments of this application are not limited to this one, but may also be other implicit or related problems. For details, please refer to the description of the embodiments below.

[0303] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0304] In one exemplary embodiment, such as Figure 3 As shown, a method for training a model is provided, which can be applied to... Figure 1 Taking RAN as an example, the explanation includes the following steps:

[0305] S201 receives the first model training request sent by the model consumer.

[0306] The first model training requirement is used to characterize the computing resources required for model training.

[0307] In this embodiment of the application, when a model consumer has a model training requirement, the data type, data volume, model type, and model resources required for model training can be evaluated to determine the computing power resources required for model training, thereby generating a first model training requirement, and then sending the first model training requirement to the RAN so that the RAN can then perform model training according to the first model training requirement.

[0308] S202, Determine the training strategy for the first model based on the training requirements of the first model.

[0309] The first model training strategy is used to characterize the model training method. The first model training strategy includes any of the following: model training on the RAN; model training on the network functional modules of the RAN side; joint model training with the network functional modules of the RAN side; and collaborative model training with the network functional modules of the core network side. The aforementioned joint model training with the network functional modules of the RAN side includes collaborative model training with the network functional modules of the RAN side, or incremental model training with the network functional modules of the RAN side.

[0310] In this embodiment, when the RAN receives a first model training request, it can analyze the first model training request and assess its own capabilities to determine the model training method, i.e., the first model training strategy. For example, the model training method may include the following: First, model training is performed on the RAN, i.e., the model is trained on the RAN; Second, model training is performed on the network functional modules on the RAN side, i.e., the model is trained on the network functional modules on the RAN side; Third, joint model training is performed with the network functional modules on the RAN side (e.g., the AIF on the RAN side), i.e., the model is split, with part trained on the RAN and part trained on the network functional modules on the RAN side, or part trained on the RAN and the remaining part incrementally trained by the AIF based on the model; Fourth, collaborative model training is performed with the network functional modules on the core network side, i.e., the model is split, with part trained on the RAN and part trained on the network functional modules on the RAN side.

[0311] S203, Train the model to be trained according to the first model training strategy and the data required for training the model, and obtain the trained target model.

[0312] In this embodiment of the application, when the RAN determines the first model training strategy based on the aforementioned steps, it can collect the data required for training the model and construct the model to be trained according to the model training method indicated by the model training strategy. Then, after preprocessing the collected data required for training the model, the model to be trained is trained to obtain the trained target model. For example, if the first model training strategy includes model training on the RAN, the corresponding RAN collects the data required for training the model and constructs the model to be trained. Then, after preprocessing the collected data required for training the model, it trains the model to obtain the trained target model. If the first model training strategy includes model training on the network functional modules on the RAN side, the corresponding RAN-side network functional modules can obtain the data required for training the model from the RAN and construct the model to be trained. Then, after preprocessing the collected data required for training the model, it trains the model to obtain the trained target model. If the first model training strategy includes joint model training with the network functional modules on the RAN side, the corresponding RAN-side network functional modules can obtain the data required for training the model or the model to be trained from the RAN, preprocess the collected data required for training the model, and train the model to obtain the trained target model. If the first model training strategy includes collaborative model training with the network functional modules on the core network side, the corresponding core network-side network functional modules can collect the data required for training the model themselves, preprocess the collected data required for training the model, and train the model to obtain the trained target model.

[0313] The training method described above involves the RAN receiving a first model training request from a model consumer, determining a first model training strategy based on this request, and training the target model according to the first model training strategy and the required data. This method leverages the RAN's built-in intelligent capabilities, enabling it to train models, meet the model consumer's training needs, and formulate different training strategies for different requirements. It can also perform model segmentation and integration for complex models, collaborating with both the RAN-side and core network functional modules to complete model training, thereby improving training efficiency and real-time performance to some extent.

[0314] The following examples will specifically illustrate the corresponding model training methods for different first model training strategies.

[0315] In an exemplary embodiment, when the first model training strategy includes model training on the RAN, a model training method is provided, namely, an implementation of the above-described S203 "training the model to be trained according to the first model training strategy and the data required for training the model to obtain the trained target model", such as... Figure 4 As shown, it includes:

[0316] S301, preprocesses the data required for training the model.

[0317] S302, Train the model to be trained based on the data required by the preprocessed training model to obtain the trained target model.

[0318] In this embodiment, the RAN can first preprocess the data required for training the model, and then train the model to be trained based on the preprocessed data to obtain the trained target model. The preprocessing may include data cleaning, data normalization, data feature extraction, and other processes.

[0319] In an exemplary embodiment, when the first model training strategy includes model training in the network function module on the RAN side, a model training method is provided, namely, an implementation of the above-mentioned S203 "training the model to be trained according to the first model training strategy and the data required for training the model, to obtain the trained target model", such as... Figure 5 As shown, it includes:

[0320] S401, send the first model training request to the network function module on the RAN side.

[0321] The first model training request instructs the network functional module on the RAN side to train the model to be trained according to the first model training request, thereby obtaining the trained target model. The first model training request carries the model requirements for training the model to be trained, as well as the data requirements or model data for training the model to be trained. For example, the model requirements are used to construct or obtain the model to be trained, and these requirements include algorithm type, model capabilities, input data dimensions, and output result requirements. The data requirements are used to collect the model data for training the model to be trained, and these requirements include information such as data dimensions, data type, data validity period, and data size. The RAN and the network functional module on the RAN side form a mutual trust domain.

[0322] In this embodiment of the application, when the RAN determines that the first model training strategy is to train the model on the network function module on the RAN side, the RAN can collect model data for training the model to be trained, or generate a first model training request in combination with the model requirements for training the model to be trained; or, the RAN can generate a first model training request in combination with the data requirements for training the model to be trained and the model requirements for training the model to be trained, and then send the first model training request to the network function module on the RAN side.

[0323] S402, Receive the target model sent by the network function module on the RAN side.

[0324] In this embodiment, when the network function module on the RAN side receives a first model training request, if the first model training request includes model data for training the model to be trained and model requirements for training the model to be trained, the network function module on the RAN side can extract the model requirements for training the model to be trained from the first model training request, obtain the model to be trained according to the model requirements for training the model to be trained, then extract the model data for training the model to be trained from the first model training request, and then train the model to be trained according to the model data for training the model to be trained to obtain a target model, and return the target model to the RAN; optionally, if the first model training request includes data requirements for training the model to be trained and model requirements for training the model to be trained, the network function module on the RAN side can extract the model requirements for training the model to be trained from the first model training request, obtain the model to be trained according to the model requirements for training the model to be trained, then extract the data requirements for training the model to be trained from the first model training request, collect the model data for training the model to be trained according to the data requirements for training the model to be trained, and then train the model to be trained according to the model data for training the model to be trained to obtain a target model, and return the target model to the RAN.

[0325] In an exemplary embodiment, when the first model training strategy includes model co-training with the network function modules on the RAN side, a model training method is provided, namely, an implementation of the above-mentioned S203 "training the model to be trained according to the first model training strategy and the data required for training the model, to obtain the trained target model", such as... Figure 6 As shown, it includes:

[0326] S501, the model to be trained is segmented to obtain the first RAN model and the first network function model.

[0327] In this embodiment, when the RAN determines that the first model training strategy includes collaborative model training with the network function modules on the RAN side, it can first obtain the model to be trained and segment it into a model trained on the RAN, i.e., the first RAN model, and a model trained on the network function modules on the RAN side, i.e., the first network function model. Then, the RAN sends the first network function model to the network function modules on the RAN side so that the network function modules on the RAN side can train the first network function model.

[0328] S502, train the first RAN model according to the data required for training the model, and obtain the trained first RAN model.

[0329] In this embodiment of the application, when the RAN obtains the first RAN model based on the above steps, it can obtain the data required for training the model based on the data it generates, and then train the first RAN model according to the data required for training the model to obtain the trained first RAN model.

[0330] S503 sends the first model collaborative training request to the network function module on the RAN side.

[0331] The first model collaborative training request carries the model requirements for training the first network functional model, as well as the data requirements or model data for training the first network functional model. The first model collaborative training request instructs the network functional modules on the RAN side to train the first network functional model according to the first model collaborative training request, thereby obtaining the trained first network functional model.

[0332] In this embodiment, when the RAN obtains the first network function model based on the above steps, it can obtain the model data for training the first network function model based on its own generated data, determine the model requirements for training the first network function model according to the first network function model, and then generate a first model co-training request by combining the model data for training the first network function model and the model requirements for training the first network function model. Optionally, when the RAN obtains the first network function model based on the above steps, it determines the data requirements and model requirements for training the first network function model according to the first network function model, and then generates a first model co-training request based on these two. After generating the first model co-training request, it can send the first model co-training request to the network function module on the RAN side.

[0333] S504 receives the trained first network function model sent by the network function module on the RAN side.

[0334] In this embodiment of the application, when the network function module on the RAN side obtains the first network function model based on the above steps, it can obtain the data required for model training from the RAN according to the first model collaborative training request or collect the data required for model training itself. Then, it trains the first network function model according to the data required for training the model to obtain the trained first network function model, and sends the trained first network function model to the RAN.

[0335] S505 integrates the first trained RAN model and the first trained network functional model to obtain the target model.

[0336] In this embodiment, when the RAN obtains the trained first RAN model and the trained first network function model based on the above steps, it can integrate the trained first RAN model and the trained first network function model to obtain the target model, thereby completing model training. It should be noted that the trained first RAN model and the trained first network function model belong to the same version of the model, so the RAN can integrate models of the same version to obtain the final trained target model.

[0337] In an exemplary embodiment, the first model training strategy includes providing a model training method when performing incremental model training with the network functional modules on the RAN side. This is an implementation of the above-mentioned S203, "training the model to be trained according to the first model training strategy and the data required for training the model, to obtain the trained target model," as shown below. Figure 7 As shown, it includes:

[0338] S601, train the model to be trained according to the data required for training the model, and obtain the first intermediate model after training.

[0339] In this embodiment of the application, when the RAN determines that the first model training strategy includes incremental model training with the network function modules on the RAN side, it can first obtain the model to be trained and the data required for training the model, and perform preliminary or partial training on the model to be trained according to the data required for training the model to obtain the first intermediate model after training.

[0340] S602, sends the first model incremental training request to the network function module on the RAN side.

[0341] The first model incremental training request carries a first intermediate model, as well as the data requirements for training the first intermediate model or the model data for training the first intermediate model. The first model incremental training request is used to instruct the network function module on the RAN side to train the first intermediate model according to the first model incremental training request to obtain the trained target model.

[0342] In this embodiment, when the RAN obtains the first intermediate model based on the above steps, it can obtain the model data for training the first intermediate model based on its own generated data, and combine the model data for training the first intermediate model with the first intermediate model to generate a first model incremental training request. Optionally, when the RAN obtains the first intermediate model based on the above steps, it determines the data requirements for training the first intermediate model and the first intermediate model based on the first intermediate model, and then generates a first model incremental training request based on these two. After generating the first model incremental training request, it can send the first model incremental training request to the network function module on the RAN side.

[0343] S603 receives the trained target model sent by the network function module on the RAN side.

[0344] In this embodiment of the application, when the network function module on the RAN side can obtain the first intermediate model according to the first model incremental training request, and obtain the data required for model training or collect the data required for model training itself, then train the remaining part of the first intermediate model according to the data required for training the model, obtain the trained target model, and send the trained target model to the RAN.

[0345] In an exemplary embodiment, the first model training strategy, when performing incremental model training with the network functional modules on the RAN side, also provides another model training method, namely, an implementation of the above-mentioned S203 "training the model to be trained according to the first model training strategy and the data required for training the model, to obtain the trained target model", such as... Figure 8 As shown, it includes:

[0346] S701 sends a second model incremental training request to the network function module on the RAN side.

[0347] The second model incremental training request carries the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained. The second model incremental training request is used to instruct the network function module on the RAN side to train the model to be trained according to the second model incremental training request to obtain the second intermediate model.

[0348] In this embodiment, when the RAN obtains the model to be trained, it can acquire the model data for training the model based on its own generated data, and combine the model data and the model to be trained to generate a second model incremental training request. Optionally, when the RAN obtains the model to be trained, it determines the data requirements and the model to be trained based on the model, and then generates a second model incremental training request based on these two. After generating the second model incremental training request, it can send the second model incremental training request to the network function module on the RAN side.

[0349] S702 receives the trained second intermediate model sent by the network function module on the RAN side.

[0350] In this embodiment, the network function module on the RAN side can obtain the model to be trained according to the second model incremental training request, and obtain the data required for model training according to the second model incremental training request or collect the data required for model training itself. Then, it can perform partial training on the model to be trained according to the data required for training the model to obtain the trained second intermediate model, and send the trained second intermediate model to the RAN.

[0351] S703 trains the second intermediate model based on the data required for training the model, and obtains the trained target model.

[0352] In this embodiment of the application, when the RAN obtains the trained second intermediate model based on the aforementioned steps, it can collect the data required for training the model and train the second intermediate model according to the data required for training the model to obtain the trained target model and complete the entire model training.

[0353] In an exemplary embodiment, the first model training strategy includes providing a model training method when co-training the model with the network function modules on the core network side. This is an implementation of the above-mentioned S203, "training the model to be trained according to the first model training strategy and the data required for training the model, to obtain the trained target model," as shown below. Figure 9 As shown, it includes:

[0354] S801, the model to be trained is segmented to obtain the second RAN model and the second network function model.

[0355] In this embodiment, when the RAN determines that the first model training strategy includes collaborative model training with the network function modules on the core network side, it can first obtain the model to be trained and segment it into a model trained on the RAN, i.e., the second RAN model, and a model trained on the network function modules on the core network side, i.e., the second network function model. Then, the RAN sends the second network function model to the network function modules on the core network side so that the network function modules on the core network side can train the second network function model.

[0356] S802, train the second RAN model according to the data required for training the model, and obtain the trained second RAN model.

[0357] In this embodiment of the application, when the RAN obtains the second RAN model based on the above steps, it can obtain the data required for training the model based on the data it generates, and then train the second RAN model according to the data required for training the model to obtain the trained second RAN model.

[0358] S803 sends a second model collaborative training request to the network function modules on the core network side.

[0359] The second model collaborative training request carries the model requirements for training the second network functional model, as well as the data requirements for training the second network functional model or the model data for training the second network functional model. The second model collaborative training request is used to instruct the network functional modules on the core network side to train the second network functional model according to the second model collaborative training request, so as to obtain the trained second network functional model.

[0360] In this embodiment, when the RAN obtains the second network function model based on the above steps, it can obtain the model data for training the second network function model based on its own generated data, determine the model requirements for training the second network function model according to the second network function model, and then generate a second model co-training request by combining the model data for training the second network function model and the model requirements for training the second network function model. Optionally, when the RAN obtains the second network function model based on the above steps, it determines the data requirements and model requirements for training the second network function model according to the second network function model, and then generates a second model co-training request based on these two. After generating the second model co-training request, it can send the second model co-training request to the network function module on the core network side.

[0361] S804 receives the trained second network function model sent by the network function module on the core network side.

[0362] In this embodiment of the application, when the network function module on the core network side receives the second model collaborative training request sent by the network function module on the core network side, it can obtain the second network function model according to the model requirements of the second network function model in the second model collaborative training request, and obtain the data required for model training according to the second model collaborative training request or collect the data required for model training itself. Then, it trains the second network function model according to the data required for training the model to obtain the trained second network function model, and sends the trained second network function model to the RAN.

[0363] S805 integrates the trained second RAN model and the trained second network function model to obtain the target model.

[0364] In this embodiment, when the RAN obtains the trained second RAN model and the trained second network function model based on the above steps, it can integrate the trained second RAN model and the trained second network function model to obtain the target model, thereby completing model training. It should be noted that the trained second RAN model and the trained second network function model belong to the same version of the model, so the RAN can integrate models of the same version to obtain the final trained target model.

[0365] In an exemplary embodiment, a method for determining a model training strategy is also provided, namely, the above-described S202 "determining a first model training strategy based on the first model training requirements", such as... Figure 10 As shown, it includes:

[0366] S901, Analyze the training requirements of the first model to determine whether the computing resources required for the training of the first model exceed the computing resources of the RAN. If the computing resources required for the training of the first model do not exceed the computing resources of the RAN, proceed to step S902; if the computing resources required for the training of the first model exceed the computing resources of the RAN, proceed to step S903.

[0367] S902, the first model training strategy is determined to be to train the model on RAN.

[0368] S903, the first model training strategy is determined to include any one of the following: training the model on the network functional modules on the RAN side, jointly training the model with the network functional modules on the RAN side, or co-training the model with the network functional modules on the core network side.

[0369] In this embodiment, when the RAN receives a first model training request, it can analyze the request to determine the required computing resources and assess its own capabilities to determine its current computing resources. Then, it compares the computing resources required by the first model training request with the current RAN computing resources. If the required computing resources exceed the current RAN computing resources, it indicates that the RAN is insufficient to support model training and requires assistance from network function modules on the RAN side, or joint or collaborative training with network function modules on the RAN side, or joint or collaborative training with network function modules on the core network side. If the required computing resources do not exceed the current RAN computing resources, it indicates that the RAN is sufficient to support model training, and in this case, model training can be performed on the RAN.

[0370] In one exemplary embodiment, a method for collecting data required for RAN model training is also provided, i.e., a scenario where the data required for model training needs to be reported periodically, such as... Figure 11 As shown, the method includes:

[0371] S1001, Set and start the first timer, and acquire the data required for training the model.

[0372] S1002, when the first timer reaches its set time, sends the data required for training the model to the network function module on the RAN side.

[0373] S1003, upon receiving the first data report response, restart the first timer.

[0374] In this embodiment, the RAN can periodically report the data required for model training to the network function module on the RAN side. Specifically, a first timer and a timeout period can be set on the RAN. When the first timer reaches its timeout period, the data required for training the model is sent to the network function module on the RAN side. This data requires data information and applicable model information. The data information includes data dimension, data type, data validity period, etc.; the applicable model information includes model type, model name, model capabilities, model version, model ID, and other related content. When the network function module on the RAN side receives the data required for training the model, it can send a first data reporting response to the RAN to inform the RAN that it has received the data reported by the RAN. When the RAN receives the first data reporting response, it can restart the first timer so that it can periodically report data to the network function module on the RAN side thereafter.

[0375] In one exemplary embodiment, another method for collecting the data required for RAN model training is also provided, namely, a scenario where the data required for model training is actively acquired and subscribed to by the AIF, such as... Figure 12 As shown, the method includes:

[0376] S1101, Receive the first data subscription request sent by the network function module on the RAN side.

[0377] In this embodiment of the application, the network function module on the RAN side can request the data required for model training from the RAN, that is, generate a first data subscription request based on the data information, and then send the first data subscription request to the RAN with the data information or data requirements in the first data subscription request.

[0378] S1102, send the data required for training the model corresponding to the first data subscription request to the network function module on the RAN side.

[0379] In this embodiment of the application, when the RAN receives the first data subscription request, it can collect the data required for training the model corresponding to the first data subscription request according to the data information or data requirements in the first data subscription request, and send the data required for training the model to the network function module on the RAN side.

[0380] S1103, determine whether the data required for training the model has been updated. If it is determined that the data required for training the model has been updated, send the updated data required for training the model to the network function module on the RAN side.

[0381] In this embodiment of the application, when the data required for training the model on the RAN is updated, the updated data needs to be automatically reported to the network function module on the RAN side. That is, first determine whether the data required for training the model has been updated, and if it is determined that the data required for training the model has been updated, send the updated data required for training the model to the network function module on the RAN side.

[0382] In an exemplary embodiment, once the RAN has completed training and obtained the trained target model, the target model can be further deployed on the RAN and applied.

[0383] In one exemplary embodiment, such as Figure 13 As shown, another method for training the model on the RAN side is provided, which includes:

[0384] S1201 receives the second model training request sent by the model consumer.

[0385] The second model training requirement is used to characterize the computing resources needed for model training.

[0386] In this embodiment of the application, when a model consumer has a model training requirement, the data type, data volume, model type, and model resources required for model training can be evaluated to determine the computing power resources required for model training, thereby generating a second model training requirement, and then sending the second model training requirement to the RAN so that the RAN can then perform model training according to the second model training requirement.

[0387] S1202, send the second model training requirement to the network function module on the RAN side, instructing the network function module on the RAN side to train the model to be trained according to the second model training requirement, and obtain the trained target model.

[0388] In this embodiment of the application, when the RAN receives the second model training request, it can forward the second model training request to the network function module on the RAN side. When the network function module on the RAN side receives the second model training request, it can perform model training according to the second model training request.

[0389] S1203 receives the target model sent by the network function module on the RAN side.

[0390] In this embodiment, after the network function module on the RAN side completes model training according to the second model training requirements and obtains the trained target model, it can send the trained target model to the RAN. It should be noted that the network function module on the RAN side can, according to... Figures 3-12 The method described in this embodiment for model training differs only in that the execution entity changes from the RAN to a network functional module on the RAN side. This enables interaction with network functional modules on both the RAN and core network sides, allowing for individual model training, collaborative model training, and incremental model training. For details, please refer to the foregoing. Figures 3-12 The method described in the embodiments can also be found in the subsequent description of the model training method for the network functional modules on the RAN side, which will not be repeated here.

[0391] S1204, Deploy the target model on the RAN.

[0392] In this embodiment of the application, after the RAN receives the target model, it can deploy the target model on the RAN.

[0393] The above Figures 3-13 The method described in this embodiment is a model training method on the RAN side. The following embodiment will specifically describe the model training method for the corresponding network functional module on the RAN side.

[0394] In one exemplary embodiment, such as Figure 14 As shown, a method for training a model is provided, which can be applied to... Figure 1Taking the RAN-side network function module (RAN-side AIF) as an example, the explanation includes the following steps:

[0395] S1301 receives the first training request sent by the RAN.

[0396] The first training request can be used to request independent model training, collaborative training, or joint incremental training. The first training request includes one of the following: a first model training request, a first model collaborative training request, a first model incremental training request, and a second model incremental training request. Specifically, the first model training request requests the network functional modules on the RAN side to perform independent model training; the first model collaborative training request requests the network functional modules on the RAN side to collaborate with the RAN for model training, i.e., splitting the model, with some parts trained on the RAN and others on the network functional modules on the RAN side; the first model incremental training request requests the network functional modules on the RAN side to jointly perform one type of incremental model training, i.e., performing preliminary training on the RAN and then training the remaining parts on the network functional modules on the RAN side; the second model incremental training request requests the network functional modules on the RAN side to jointly perform another type of incremental model training, i.e., performing preliminary training on the network functional modules on the RAN and then training the remaining parts on the RAN. Furthermore, the first model training request and the aforementioned... Figure 5 The first model co-training request in the embodiment corresponds to the aforementioned first model co-training request. Figure 6 The first model collaborative training request in the embodiment corresponds to the aforementioned first model incremental training request. Figure 7 The first model incremental training request in the embodiment corresponds to the second model incremental training request mentioned above. Figure 8 The first model incremental training request in the embodiment corresponds to this.

[0397] S1302, Train the model according to the first training request to obtain the trained model.

[0398] In this embodiment, when the RAN-side network function module receives a first training request, it can perform model training according to the training type requested in the first training request to obtain a trained model. The trained model can be a complete trained model, a partial trained model, or a segmented trained model. For example, if the first training request is a first model training request, and the corresponding training type is independent model training, the RAN-side network function module can obtain the data required for model training according to the first training request or collect the data itself, determine the model to be trained according to the model requirements in the first training request, and train the model to be trained according to the required data to obtain the trained model. This trained model can correspond to the aforementioned... Figure 5 The target model in this embodiment. If the first training request is a first model collaborative training request, and the training type of the corresponding first training request is collaborative training, then the network function module on the RAN side can obtain the data required for model training based on the first training request or collect the data required for model training itself, determine the model to be trained based on the model requirements in the first training request, and train the model to be trained based on the data required for model training to obtain the trained model. The model to be trained can correspond to the aforementioned... Figure 6 The first network functional model in the embodiment, and the trained model, can correspond to the aforementioned Figure 6 The first network functional model trained in this embodiment. If the first training request is a first model incremental training request, and the training type of the corresponding first model incremental training request is joint incremental training, then the network functional module on the RAN side can obtain the data required for model training according to the first training request or collect the data required for model training itself, and obtain the model to be trained according to the first training request, and train the model to be trained according to the data required for model training to obtain the trained model. The model to be trained can correspond to the aforementioned... Figure 7 The first intermediate model in the embodiment, and the trained model, can correspond to the aforementioned Figure 7 The target model in this embodiment. If the first training request is a second model incremental training request, and the training type of the corresponding first model incremental training request is joint incremental training, then the network function module on the RAN side can obtain the data required for model training according to the first training request or collect the data required for model training itself, and obtain the model to be trained according to the first training request, and train the model to be trained according to the data required for model training to obtain the trained model. The model to be trained can correspond to the aforementioned... Figure 8 The model to be trained in the embodiment, and the trained model, can correspond to the aforementioned Figure 8 The second intermediate model in the embodiment.

[0399] S1303 sends the trained model to the RAN.

[0400] In this embodiment of the application, when the network function module on the RAN side completes model training and obtains the trained model, it can send the trained model to the RAN so that the RAN can continue training, integrate models, or deploy models based on the model sent by the network function module on the RAN side.

[0401] In an exemplary embodiment, a method for training a model according to a first training request is provided, namely, the above-described S1302 "training the model according to the first training request to obtain the trained model", as shown below. Figure 15 As shown, the method includes:

[0402] S1401, Obtain the model to be trained and the model data for training the model to be trained according to the first training request.

[0403] The first training request carries either the model requirement for training the model to be trained, or the model to be trained; the first training request also carries either the data requirement for training the model to be trained or the model data for training the model to be trained.

[0404] In this embodiment, when the first training request carries model requirements for training the model to be trained, the network functional module on the RAN side can extract the model requirements from the first training request and construct or obtain the model to be trained based on the model requirements. Optionally, when the first training request carries a model to be trained, the network functional module on the RAN side can directly extract the model to be trained from the first training request. Optionally, when the first training request also carries data requirements for training the model to be trained or model data for training the model to be trained, the network functional module on the RAN side can extract the data requirements for training the model to be trained from the first training request and collect the model data for training the model to be trained based on the data requirements; or, the network functional module on the RAN side can directly extract the model data for training the model to be trained from the first training request.

[0405] S1402, Train the model to be trained based on the model data used to train the model to be trained, and obtain the trained model.

[0406] In this embodiment of the application, when the network function module on the RAN side obtains the model data and the model to be trained, it can train the model to be trained based on the model data to obtain the trained model.

[0407] In an exemplary embodiment, the network function module on the RAN side can first preprocess the model data for training the model to be trained, and then train the model to be trained based on the preprocessed model data to obtain the trained model.

[0408] In one exemplary embodiment, a method is also provided for the network function module on the RAN side to collect the data required for model training, i.e., a scenario where the data required for model training needs to be reported periodically, such as... Figure 16 As shown, the method includes:

[0409] S1501 receives the data required for training the model sent by the RAN and starts the second timer.

[0410] S1502, send the first data reporting response to the RAN.

[0411] S1503, when the second timer reaches its set time, delete the expired training model data stored locally, or delete the training model data that exceeds the local storage limit.

[0412] In this embodiment, a second timer and a timing period are set on the network function module on the RAN side. This application's embodiment is similar to the one described above. Figure 11 The steps described in the embodiment correspond to the following: When the RAN sends the data required for training the model to the RAN-side network function module, and the RAN-side network function module receives the data, it can immediately start a second timer and send a first data reporting response to the RAN to inform the RAN that it has received the data required for training the model. Then, when the second timer reaches its set time, it deletes the expired data required for training the model stored locally, or, based on the AIF configuration and performance, sets a data storage occupancy limit to delete the data required for training the model that exceeds the local storage limit. It should be noted that the time settings of the first and second timers, as well as the AIF storage limit settings, should consider that the storage resources and data volume at the AIF should meet the model training requirements.

[0413] In one exemplary embodiment, a further provision is also provided Figure 12 Another method for collecting the data required for model training corresponding to the example is a scenario where the data required for model training is actively acquired and subscribed to by AIF, such as... Figure 17 As shown, the method includes:

[0414] S1601, send the first data subscription request to the RAN.

[0415] S1602, Receive the data required for training the model corresponding to the first data subscription request sent by the RAN.

[0416] The method described in the embodiments of this application is the same as the one described above. Figure 12The method described in the embodiments is the same; for detailed explanation, please refer to the foregoing. Figure 12 The methods described herein will not be elaborated here.

[0417] In an exemplary embodiment, another method for training a model of the network functional modules on the RAN side is also provided, which is similar to the method described above. Figure 13 The RAN method shown is consistent with that, such as Figure 18 As shown, the method includes:

[0418] S1701 receives the second model training request sent by the RAN.

[0419] S1702, Train the model to be trained according to the training requirements of the second model to obtain the trained target model.

[0420] S1703, send the target model to the RAN.

[0421] The method described in the embodiments of this application is the same as the one described above. Figure 13 The method described in the embodiments is the same; for detailed explanation, please refer to the foregoing. Figure 13 The methods described herein will not be elaborated here.

[0422] In one exemplary embodiment, a method for training a model according to a second model training requirement is also provided, such as... Figure 19 As shown, the method includes:

[0423] S1801, Determine the training strategy for the second model based on the training requirements of the second model.

[0424] The second model training strategy includes any of the following: model training on the RAN side network functional modules; joint model training with the RAN; and collaborative model training with the core network side network functional modules. Joint model training with the RAN includes: collaborative model training with the RAN, or incremental model training with the RAN.

[0425] The method described in the embodiments of this application is the same as the one described above. Figure 3 The steps in S202 are the same, only the execution subject is different. In this embodiment, the execution subject is the network function module on the RAN side. For details, please refer to the foregoing. Figure 3 The methods described herein will not be elaborated here.

[0426] S1802, the model to be trained is trained according to the second model training strategy and the data required for training the model, and the trained target model is obtained.

[0427] The method described in the embodiments of this application is the same as the one described above. Figure 3The steps in S203 are the same as those described above, only the execution subject is different. In this embodiment, the execution subject is the network function module on the RAN side. For details, please refer to the foregoing. Figure 3 The methods described herein will not be elaborated here.

[0428] The following will describe in detail the model training methods corresponding to different second model training strategies. The second model training strategy includes training the model to be trained according to the second model training strategy and the data required for training the model in the above embodiment to obtain the trained target model. This includes: receiving a second model training request sent by the RAN; the second model training request carries the model requirements for training the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained; and training the model to be trained according to the second model training request and the data required for training the model to obtain the trained target model.

[0429] Optionally, the training of the model to be trained is performed according to the second model training request and the data required for training the model to obtain the trained target model, including: obtaining the model to be trained according to the model requirements for training the model to be trained; obtaining the data required for training the model according to the second model training request; and training the model to be trained according to the data required for training the model to obtain the trained target model.

[0430] Optionally, obtaining the data required for training the model according to the second model training request includes: collecting the data required for training the model according to the data requirements of training the model to be trained; or, extracting the model data for training the model to be trained from the second model training request as the data required for training the model.

[0431] Optionally, the method described above for training the model to be trained based on the second model training request and the data required for training the model to obtain the trained target model further includes: preprocessing the model data for training the model to be trained; correspondingly, training the model to be trained based on the data required for training the model to obtain the trained target model includes: training the model to be trained based on the preprocessed model data to obtain the trained target model.

[0432] Optionally, the second model training strategy includes co-training the model with the RAN, training the model to be trained according to the second model training strategy and the data required for training the model, and obtaining the trained target model, including: segmenting the model to be trained to obtain the corresponding RAN model and network function model; receiving the trained RAN model sent by the RAN; training the network function model according to the data required for training the model, and obtaining the trained network function model; receiving the trained RAN model sent by the RAN, and integrating the trained RAN model and the trained network function model to obtain the target model.

[0433] Optionally, the method described above for training the model to be trained according to the second model training strategy and the data required for training the model to obtain the trained target model further includes: sending a model co-training request to the RAN; wherein the model co-training request carries the model requirements for training the RAN model, and also carries the data requirements for training the RAN model or the model data for training the RAN model; the model co-training request is used to instruct the RAN to train the RAN model according to the model co-training request to obtain the trained RAN model.

[0434] Optionally, the second model training strategy includes incremental model training with the RAN, training the model to be trained according to the second model training strategy and the data required for training the model, and obtaining the trained target model, including: training the model to be trained according to the data required for training the model, obtaining the trained intermediate model; and receiving the trained target model sent by the RAN.

[0435] Optionally, the method described above for training the model to be trained according to the second model training strategy and the data required for training the model to obtain the trained target model further includes: sending a third model incremental training request to the RAN; wherein the third model incremental training request carries an intermediate model, and also carries the data requirements for training the intermediate model or the model data for training the intermediate model; the above model incremental training request is used to instruct the RAN to train the intermediate model according to the model incremental training request to obtain the trained target model.

[0436] Optionally, the second model training strategy includes incremental model training with the RAN, training the model to be trained according to the second model training strategy and the data required for training the model, and obtaining the trained target model, including: receiving the trained intermediate model sent by the RAN; training the fourth intermediate model according to the data required for training the model, and obtaining the trained target model.

[0437] Optionally, the method described above for training the model to be trained according to the second model training strategy and the data required for training the model to obtain the trained target model further includes: sending a model incremental training request to the RAN; wherein the model incremental training request carries the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained; the model incremental training request is used to instruct the RAN to train the model to be trained according to the model incremental training request to obtain the trained intermediate model.

[0438] Optionally, the second model training strategy includes collaborative model training with the network function modules on the core network side, training the model to be trained according to the second model training strategy and the data required for training the model, and obtaining the trained target model, including: segmenting the model to be trained to obtain a RAN model and a network function model; training the RAN model according to the data required for training the model to obtain a trained RAN model; receiving the trained network function model sent by the network function modules on the core network side; and integrating the trained network function model and the trained RAN model to obtain the target model.

[0439] Optionally, the method described above for training the model to be trained according to the second model training strategy and the data required for training the model to obtain the trained target model further includes: sending a model co-training request to the network function module on the core network side; wherein, the model co-training request carries the model requirements for training the network function model, and also carries the data requirements for training the network function model or the model data for training the network function model; the model co-training request is used to instruct the network function module on the core network side to train the network function model according to the third model co-training request to obtain the trained network function model.

[0440] Optionally, the second model training strategy is determined based on the second model training requirements, including: analyzing the second model training requirements to determine whether the computing resources required for the second model training requirements exceed the computing resources of the network functional modules on the RAN side; if the computing resources required for the second model training requirements do not exceed the computing resources of the network functional modules on the RAN side, then the second model training strategy is determined to be model training on the network functional modules on the RAN side; if the computing resources required for the second model training requirements exceed the computing resources of the network functional modules on the RAN side, then the second model training strategy is determined to include joint model training with the RAN or collaborative model training with the network functional modules on the core network side.

[0441] Based on all the above embodiments, a method is provided for model training through interaction between the RAN, the network functional modules on the RAN side, and the network functional modules on the core network side. Figure 20 As shown, the method includes:

[0442] S1, RAN receives the first model training request sent by the model consumer.

[0443] S2, the RAN determines the first model training strategy based on the first model training requirements. If the first model training strategy includes model training on the RAN, then step S3 is executed; if the first model training strategy includes model training on the network functional modules on the RAN side, then steps S4-S6 are executed; if the first model training strategy includes model co-training with the network functional modules on the RAN side, then steps S7-S9 are executed; if the first model training strategy includes incremental model training with the network functional modules on the RAN side, then steps S10-S12 are executed; if the first model training strategy includes model co-training with the network functional modules on the core network side, then steps S13-S16 are executed.

[0444] S3, RAN preprocesses the data required for training the model; it then trains the model to be trained based on the preprocessed data, obtains the trained target model, and deploys the target model.

[0445] S4, the RAN sends the first model training request to the network function module on the RAN side.

[0446] S5, the network function module on the RAN side performs model training, obtains the target model, and sends the target model to the RAN.

[0447] S6, the RAN receives the target model sent by the network function module on the RAN side.

[0448] S7, the RAN splits the model to be trained to obtain a first RAN model and a first network function model. The first RAN model is trained according to the data required for training the model to obtain the trained first RAN model. The first model co-training request is sent to the network function module on the RAN side.

[0449] S8, the network function module on the RAN side trains the model of the first network function module to obtain the trained first network function model, and sends the trained first network function model to the RAN.

[0450] S9, the RAN receives the trained first network function model sent by the network function module on the RAN side, and integrates the trained first RAN model and the trained first network function model to obtain the target model.

[0451] S10: Train the model to be trained according to the data required for training the model, obtain the first intermediate model after training, and send the first model incremental training request to the network function module on the RAN side.

[0452] S11, the network function module on the RAN side trains the first intermediate model to obtain the trained target model, and sends the trained target model to the RAN.

[0453] S12, RAN receives the target model.

[0454] S13, the RAN splits the model to be trained into a second RAN model and a second network function model. The second RAN model is trained using the data required for the training model, resulting in the trained second RAN model. A second model co-training request is then sent to the network function modules on the core network side.

[0455] S14, the network function modules on the core network side perform model training to obtain the trained second network function model, and send the trained second network function model to the RAN.

[0456] S15, the RAN receives the trained second network function model sent by the network function module on the core network side, and integrates the trained second RAN model and the trained second network function model to obtain the target model.

[0457] S16, RAN deployment target model.

[0458] For a detailed implementation of the above embodiments, please refer to [link / reference]. Figures 3-12 Implementation examples Figures 14-17 The methods described in the embodiments are not repeated here.

[0459] Based on all the above embodiments, another method is provided for model training through interaction between the RAN, the RAN-side network functional modules, and the core network-side network functional modules, such as... Figure 21 As shown, the method includes:

[0460] S1, RAN receives the second model training request sent by the model consumer.

[0461] S2, RAN sends the second model training request to the network function module on the RAN side.

[0462] S3, the network function module on the RAN side trains the model to be trained according to the training requirements of the second model, obtains the trained target model, and sends the target model to the RAN.

[0463] S4, the RAN receives the target model sent by the network function module on the RAN side.

[0464] For a detailed implementation of the above embodiments, please refer to [link / reference]. Figure 13 Implementation examples Figures 18-19 The methods described in the embodiments are not repeated here.

[0465] In summary, all the above embodiments provide a method for interaction between the RAN and network functional modules on the RAN side to collect model data, that is, a method in which the data required for model training needs to be reported periodically, such as...Figure 22 As shown, the method includes:

[0466] K1, RAN sets and starts the first timer, and acquires the data required to train the model.

[0467] K2, when the first timer reaches its set time, sends the data required for training the model to the network function module on the RAN side.

[0468] K3, the network function module on the RAN side sets and starts the second timer.

[0469] K4, the network function module on the RAN side sends the first data reporting response to the RAN.

[0470] When K5 and RAN receive the first data report response, they restart the first timer.

[0471] K6, the network function module on the RAN side, deletes expired training model data or training model data that exceeds local storage limits when the second timer reaches its set time.

[0472] For a detailed implementation of the above embodiments, please refer to [link / reference]. Figure 11 , Figure 16 The methods described in the embodiments are not repeated here.

[0473] Alternatively, the data required for model training can be actively acquired and subscribed to by AIF, such as... Figure 23 As shown, the method includes:

[0474] K1, the network function module on the RAN side sends the first data subscription request to the RAN.

[0475] K2, RAN obtains the data required for training the model corresponding to the first data subscription request.

[0476] K3, RAN sends the data required for training the model to the network function module on the RAN side.

[0477] K4 determines whether the data required for training the model has been updated. If it is determined that the data required for training the model has been updated, it sends the updated data required for training the model to the network function module on the RAN side.

[0478] For a detailed implementation of the above embodiments, please refer to [link / reference]. Figure 12 The methods described in the embodiments are not repeated here.

[0479] The above Figures 14-19The method described in this embodiment is a training method for a model of a network functional module on the RAN side. In one embodiment, a method for training a model through interaction between the intelligent network element and the core network is also provided. The following embodiments will specifically describe the training method for the model of the network functional module on the corresponding intelligent network element side.

[0480] In one exemplary embodiment, such as Figure 24 As shown, a method for training a model is provided, which can be applied to... Figure 1 Taking the intelligent network element in the example, the explanation includes the following steps:

[0481] S1901 receives the third model training request sent by the model consumer.

[0482] S1902, Determine the training strategy for the third model based on the training requirements of the third model.

[0483] The third model training strategy includes any of the following: training the model on intelligent network elements; or conducting joint model training with network functional modules on the core network side. Joint model training with network functional modules on the core network side includes: collaborative model training with network functional modules on the core network side, or incremental model training with network functional modules on the core network side.

[0484] S1903, the model to be trained is trained according to the third model training strategy and the data required for training the model, and the trained target model is obtained.

[0485] The method described in this application embodiment is similar to the aforementioned... Figure 3 The steps in the embodiments are basically the same, only the execution subject is different. In this embodiment, the execution subject is an intelligent network element. For details, please refer to the foregoing. Figure 3 The methods described herein will not be elaborated here.

[0486] The following examples will specifically illustrate the corresponding model training methods for different third-model training strategies.

[0487] In an exemplary embodiment, when the third model training strategy includes model training on intelligent network elements, a model training method is provided, namely, the above-mentioned S1903 "training the model to be trained according to the third model training strategy and the data required for training the model, to obtain the trained target model", as shown. Figure 25 As shown, the method includes:

[0488] S2001, preprocesses the data required for training the model.

[0489] S2002, train the model to be trained based on the data required by the preprocessed training model to obtain the trained target model.

[0490] The method described in this application embodiment is similar to the aforementioned... Figure 4 The steps in the embodiments are basically the same, only the execution subject is different. In this embodiment, the execution subject is an intelligent network element. For details, please refer to the foregoing. Figure 4 The methods described herein will not be elaborated here.

[0491] In an exemplary embodiment, when the third model training strategy includes model co-training with network function modules on the core network side, a model training method is provided, namely, an implementation of the above-mentioned S1903 "training the model to be trained according to the third model training strategy and the data required for training the model, to obtain the trained target model", such as... Figure 26 As shown, it includes:

[0492] S2101, the model to be trained is segmented to obtain the intelligent network element model and the third network function model.

[0493] S2102, Train the intelligent network element model according to the data required for training the model, and obtain the trained intelligent network element model.

[0494] S2103, sends a third model collaborative training request to the network function module on the core network side.

[0495] S2104 receives the trained third network function model sent by the network function module on the core network side.

[0496] S2105 integrates the trained third network function model and the trained intelligent network element model to obtain the target model.

[0497] The method described in this application embodiment is similar to the aforementioned... Figure 6 The steps in the embodiments are basically the same, only the execution subject is different. In this embodiment, the execution subject is an intelligent network element. For details, please refer to the foregoing. Figure 6 The methods described herein will not be elaborated here.

[0498] In an exemplary embodiment, the third model training strategy includes providing a model training method when performing incremental model training with the network functional modules on the core network side. This is an implementation of the above-mentioned S1903, "training the model to be trained according to the third model training strategy and the data required for training the model, to obtain the trained target model," as shown below. Figure 27 As shown, it includes:

[0499] S2201 sends a third model incremental training request to the network function module on the core network side.

[0500] S2202 receives the trained third intermediate model sent by the network function module on the core network side.

[0501] S2203, Train the third intermediate model according to the data required for training the model to obtain the trained target model.

[0502] The method described in this application embodiment is similar to the aforementioned... Figure 8 The steps in the embodiments are basically the same, only the execution subject is different. In this embodiment, the execution subject is an intelligent network element. For details, please refer to the foregoing. Figure 8 The methods described herein will not be elaborated here.

[0503] In an exemplary embodiment, the third model training strategy includes incremental model training with the network functional modules on the core network side. It also provides another model training method, namely, an implementation of the above-mentioned S1903 "training the model to be trained according to the third model training strategy and the data required for training the model to obtain the trained target model," such as... Figure 28 As shown, it includes:

[0504] S2301, Train the model to be trained according to the data required for training the model, and obtain the fourth intermediate model.

[0505] S2302, sends the fourth model incremental training request to the network function module on the core network side.

[0506] S2302 receives the trained target model sent by the network function module on the core network side.

[0507] The method described in this application embodiment is similar to the aforementioned... Figure 7 The steps in the embodiments are basically the same, only the execution subject is different. In this embodiment, the execution subject is an intelligent network element. For details, please refer to the foregoing. Figure 7 The methods described herein will not be elaborated here.

[0508] In an exemplary embodiment, a method for determining a model training strategy is also provided, namely, the above-described S1902 "determining the third model training strategy according to the training requirements of the third model", such as... Figure 29 As shown, it includes:

[0509] S2401, Analyze the training requirements of the third model to determine whether the computing resources required for the training of the third model exceed the computing resources of the intelligent network element. If the computing resources required for the training of the third model do not exceed the computing resources of the network function module on the core network side, then proceed to step S2402; if the computing resources required for the training of the third model exceed the computing resources of the network function module on the core network side, then proceed to step S2403.

[0510] S2402, The third model training strategy is determined to be to train the model on the intelligent network element.

[0511] S2403, the third model training strategy is determined to be joint training of the model with the network functional modules on the core network side.

[0512] The method described in this application embodiment is similar to the aforementioned... Figure 10 The steps in the embodiments are basically the same, only the execution subject is different. In this embodiment, the execution subject is an intelligent network element. For details, please refer to the foregoing. Figure 10 The methods described herein will not be elaborated here.

[0513] In one exemplary embodiment, a method for collecting data required for training intelligent network element models is also provided, i.e., in scenarios where the data required for model training needs to be reported periodically, such as... Figure 30 As shown, the method includes:

[0514] S2501: Set and start the third timer, and acquire the data required for training the model.

[0515] S2502 sends the data required for training the model to the network function module on the core network side when the third timer reaches its set time.

[0516] S2503, upon receiving the second data report response, restarts the third timer.

[0517] The method described in this application embodiment is similar to the aforementioned... Figure 11 The steps in the embodiments are basically the same, only the execution subject is different. In this embodiment, the execution subject is an intelligent network element. For details, please refer to the foregoing. Figure 11 The methods described herein will not be elaborated here.

[0518] In one exemplary embodiment, another method for intelligent network elements to collect the data required for model training is also provided, namely, a scenario where the data required for model training is actively acquired and subscribed to by the AIF, such as... Figure 31 As shown, the method includes:

[0519] S2601, Receive the second data subscription request sent by the network function module on the core network side;

[0520] S2602, send the data required for training the model corresponding to the second data subscription request to the network function module on the core network side.

[0521] S2603, determine whether the data required for training the model has been updated, and if it is determined that the data required for training the model has been updated, send the updated data required for training the model to the network function module on the core network side.

[0522] The method described in this application embodiment is similar to the aforementioned... Figure 12 The steps in the embodiments are basically the same, only the execution subject is different. In this embodiment, the execution subject is an intelligent network element. For details, please refer to the foregoing.Figure 12 The methods described herein will not be elaborated here.

[0523] In an exemplary embodiment, after the intelligent network element completes training and obtains the trained target model, the target model can be further deployed on the intelligent network element and applied.

[0524] In one exemplary embodiment, such as Figure 32 As shown, a method for training a model is provided, which can be applied to... Figure 1 Taking the network function module (AIF) on the core network side as an example, the explanation includes the following steps:

[0525] S2701, Receive the second training request sent by the target device; the target device is any one of RAN, RAN-side network function module, or intelligent network element.

[0526] S2702, Train the model according to the second training request to obtain the trained model.

[0527] The second training request is used to request collaborative training, and may also request joint incremental training. The second training request includes one of the following: a third model collaborative training request, a third model incremental training request, and a fourth model incremental training request. Specifically, the third model collaborative training request requests the network functional modules on the core network side to collaborate with the RAN for model training, i.e., model splitting, with some models trained on the RAN and others trained on the network functional modules on the core network side. The third model incremental training request requests the network functional modules on the core network side to jointly perform a model incremental training with intelligent network elements, i.e., preliminary training of the model to be trained is performed on the intelligent network elements, and the remaining training is performed on the network functional modules on the core network side. The fourth model incremental training request requests the network functional modules on the core network side to jointly perform another type of model incremental training, i.e., preliminary training of the model to be trained is performed on the network functional modules on the core network side, and the remaining training is performed on the intelligent network elements. Furthermore, the aforementioned third model collaborative training request and the aforementioned... Figure 26 The third model collaborative training request in the embodiment corresponds to the aforementioned third model incremental training request. Figure 27 The third model incremental training request in the embodiment corresponds to the fourth model incremental training request mentioned above. Figure 28 The fourth model incremental training request in the embodiment corresponds to this.

[0528] S2703 sends the trained model to the target device.

[0529] The method described in this application embodiment is similar to the aforementioned... Figures 3-23The embodiments differ only in the executing entity; in this embodiment, the executing entity is the network function module on the core network side. For details, please refer to the aforementioned figures. Figures 3-23 The methods described herein will not be elaborated upon here. Furthermore, it should be noted that: for example, if the second training request is a third model collaborative training request, and the training type of the corresponding second training request is collaborative training, then the network function module on the core network side can obtain the data required for model training based on the second training request or collect the data required for model training itself. It can also determine the model to be trained based on the model requirements in the second training request, and train the model to be trained using the required data to obtain the trained model. This model to be trained can correspond to the aforementioned... Figure 26 The third network functional model in the embodiment, and the trained model, can correspond to the aforementioned Figure 26 The trained third network functional model in this embodiment. If the second training request is a third model incremental training request, and the training type of the corresponding third model incremental training request is joint incremental training, then the network functional module on the core network side can obtain the data required for model training according to the second training request or collect the data required for model training itself, and obtain the model to be trained according to the second training request, and train the model to be trained according to the data required for model training to obtain the trained model. The model to be trained can correspond to the aforementioned... Figure 27 The model to be trained in the embodiment, and the trained model, can correspond to the aforementioned Figure 27 The third intermediate model in the embodiment. If the second training request is a fourth model incremental training request, and the training type of the corresponding fourth model incremental training request is joint incremental training, then the network function module on the core network side can obtain the data required for model training according to the second training request or collect the data required for model training itself, and obtain the model to be trained according to the second training request, and train the model to be trained according to the data required for model training to obtain the trained model. The model to be trained can correspond to the aforementioned Figure 28 The fourth intermediate model in the embodiment, and the trained model, can correspond to the aforementioned Figure 8 The target model in the embodiment.

[0530] In one exemplary embodiment, a method for training a model based on a second training request is provided, such as... Figure 33 As shown, the method includes:

[0531] S2801, Obtain the model to be trained and the model data for training the model to be trained according to the second training request.

[0532] The second training request includes the model requirement for training the model to be trained; the second training request includes the model to be trained; and the second training request also includes the data requirement for training the model to be trained or the model data for training the model to be trained.

[0533] In this embodiment, when the second training request carries model requirements for training the model to be trained, the network function module on the RAN side can extract the model requirements from the second training request and construct or obtain the model to be trained based on the model requirements. Optionally, when the second training request carries a model to be trained, the network function module on the core network side can directly extract the model to be trained from the second training request. Optionally, when the second training request also carries data requirements for training the model to be trained or model data for training the model to be trained, the network function module on the core network side can extract the data requirements for training the model to be trained from the second training request and collect the model data for training the model to be trained based on the data requirements; or, the network function module on the core network side can directly extract the model data for training the model to be trained from the second training request.

[0534] S2802, Train the model to be trained based on the model data used to train the model to be trained, and obtain the trained model.

[0535] The method described in this application embodiment is similar to the aforementioned... Figure 15 The embodiments are similar, differing only in the executing entity. In this embodiment, the executing entity is the network function module on the core network side. For detailed explanation, please refer to the foregoing. Figure 15 The methods described herein will not be elaborated here.

[0536] In an exemplary embodiment, a method for collecting model data by the network function module of the core network is provided, namely, a method for collecting model data for training the model to be trained according to the data requirements for training the model to be trained, such as... Figure 34 As shown, the method includes:

[0537] S2901, Generate a model data collection request based on the data requirements for training the model to be trained.

[0538] In this embodiment of the application, when the network function module on the core network side extracts the data requirements for training the model to be trained from the second training request, it can generate a model data collection request based on the data requirements for training the model to be trained.

[0539] S2902 sends a model data collection request to other intelligent network elements or data planes.

[0540] In this embodiment of the application, when the network function module on the core network side generates a model data collection request based on the aforementioned steps, it can send the model data collection request to other intelligent network elements or data planes.

[0541] S2903 receives model data returned by other intelligent network elements or data planes based on model data collection requests.

[0542] In this embodiment of the application, when other intelligent network elements or data planes receive a model data collection request, they can determine the data requirements for training the model to be trained based on the model data collection request, collect model data according to the data requirements for training the model to be trained, and return the collected model data to the network function module on the core network side.

[0543] In an exemplary embodiment, the network function module on the core network side can first preprocess the model data for training the model to be trained, and then train the model to be trained based on the preprocessed model data to obtain the trained model.

[0544] In one exemplary embodiment, a method for the network function module on the core network side to collect the data required for model training is also provided, i.e., a scenario where the data required for model training needs to be reported periodically, such as... Figure 35 As shown, the method includes:

[0545] S3001 receives the data required for training the model sent by the intelligent network element and starts the fourth timer;

[0546] S3002 sends a second data reporting response to the intelligent network element;

[0547] S3003: When the fourth timer reaches its set time, delete the expired training model data stored locally, or delete the training model data that exceeds the local storage limit.

[0548] The method described in this application embodiment is similar to the aforementioned... Figure 30 The embodiments are similar, differing only in the executing entity. In this embodiment, the executing entity is the network function module on the core network side. For detailed explanation, please refer to the foregoing. Figure 30 The methods described herein will not be elaborated here.

[0549] In one exemplary embodiment, a further provision is also provided Figure 31 Another method for collecting the data required for model training corresponding to the example is a scenario where the data required for model training is actively acquired and subscribed to by AIF, such as... Figure 36 As shown, the method includes:

[0550] S3101 sends a second data subscription request to the intelligent network element;

[0551] S3102 receives the data required for training the model corresponding to the second data subscription request sent by the intelligent network element.

[0552] The method described in this application embodiment is similar to the aforementioned... Figure 31 The embodiments are similar, differing only in the executing entity. In this embodiment, the executing entity is the network function module on the core network side. For detailed explanation, please refer to the foregoing. Figure 31 The methods described herein will not be elaborated here.

[0553] Based on all the above embodiments, a method is provided for model training through interaction between intelligent network elements and network function modules on the core network side, such as... Figure 37 As shown, the method includes:

[0554] L1, the intelligent network element receives the third model training request sent by the model consumer.

[0555] L2, the intelligent network element determines the third model training strategy according to the training requirements of the third model. If the third model training strategy includes training the model on the intelligent network element, then step L3 is executed; if the third model training strategy includes co-training the model with the network function modules on the core network side, then steps L4-L9 are executed; if the third model training strategy includes incremental training of the model with the network function modules on the core network side, then steps L9-L12 are executed.

[0556] L3 intelligent network elements preprocess the data required for training the model; based on the preprocessed data required for training the model, the model to be trained is trained to obtain the trained target model.

[0557] L4: The intelligent network element segments the model to be trained to obtain the intelligent network element model and the third network function model; it trains the intelligent network element model according to the data required for training the model to obtain the trained intelligent network element model; and sends a third model collaborative training request to the network function module on the core network side.

[0558] L5, the network function module on the core network side sends model data collection requests to other intelligent network elements or data planes.

[0559] L6, other intelligent network elements or data planes collect model data according to the model data collection request, and send the model data to the network function modules on the core network side.

[0560] L7, the network function module on the core network side receives model data returned by other intelligent network elements or data planes based on model data collection requests.

[0561] L8, the network function module on the core network side trains the model based on the model data to obtain the trained third network function model, and sends the trained third network function model to the intelligent network element.

[0562] L9, the intelligent network element receives the trained third network function model sent by the network function module on the core network side, and integrates the trained third network function model and the trained intelligent network element model to obtain the target model.

[0563] L10: The intelligent network element trains the model to be trained based on the data required for training the model, obtains the fourth intermediate model, and sends the fourth model incremental training request to the network function module on the core network side.

[0564] L11, the network function module on the core network side trains the fourth intermediate model to obtain the trained target model, and sends the trained target model to the intelligent network element.

[0565] L12, Intelligent Network Element Target Receiving Model.

[0566] The training method for the model provided in any of the above embodiments enables the integration of intelligent capabilities into the RAN and core network intelligent network elements, allowing for the training of simple models. After training, the model is directly deployed in the RAN or intelligent network elements, eliminating the need for data collection and model distribution processes. Simultaneously, it allows for the segmentation and integration of complex models, and can collaboratively complete model training with both the RAN-side AIF and the core network-side AIF, improving the efficiency and real-time performance of model training.

[0567] The training method for the model provided in any of the above embodiments enables collaborative training of models that the RAN cannot train when training the model on the RAN-side AIF. Collaborative training can train the complete model, or the model can be trained separately, with one part trained by the RAN and another part trained by the RAN-side AIF, and then the models of the same version can be integrated into the final model, which can improve the efficiency and real-time performance of model training. To ensure data privacy, collaborative training of the model can also be performed between the RAN-side AIF and the core network-side AIF.

[0568] The training methods for the models provided in any of the above embodiments, when training the model on the core network side AIF, can perform collaborative training or incremental training on models that intelligent network elements cannot train. Collaborative training considers the computing resources of network elements and the data required for model training. The model can be trained entirely on the AIF, or a portion can be trained on the network element and another portion on the AIF, and then the models of the same version can be integrated into the final model. Incremental training involves the core network side AIF performing incremental training on the model trained on the intelligent network element. New training data features (other network data required for model training) are used to update the model parameters to obtain the final model, which can improve the efficiency and real-time performance of model training.

[0569] In summary, this application proposes a multi-source data collection model training method for 6G intelligent endogenous architecture. It integrates the intelligent capabilities of the RAN and various network elements, and provides design descriptions for specific methods of intelligent endogenous model training and different data source situations. This method brings model training closer to data producers and model deployment closer to model consumers, filling the current technical gap in model training and improving the efficiency of model training and the flow efficiency of the data required in model training.

[0570] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0571] In one embodiment, this application provides a model training apparatus, such as... Figure 38 As shown, the device includes a memory, a transceiver, and a processor: the memory stores computer programs; the transceiver transmits and receives data under the control of the processor; and the processor reads the computer program from the memory and performs the following operations:

[0572] Receive the first model training request sent by the model consumer;

[0573] Determine the first model training strategy based on the first model training requirements;

[0574] The model to be trained is trained according to the first model training strategy and the data required for training the model, and the trained target model is obtained.

[0575] Among them, Figure 38In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits together, represented by one or more processors (processor 3800) and memory (memory 3820). The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. The transceiver 3810 can be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. The processor 3800 is responsible for managing the bus architecture and general processing, and the memory 3820 can store data used by the processor 3800 during operation.

[0576] The processor 3800 can be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a complex programmable logic device (CPLD). The processor can also adopt a multi-core architecture.

[0577] The processor 3800 is responsible for managing the bus architecture and general processing, while the memory 3820 can store the data used by the processor 3000 during operation.

[0578] The processor 3800 is also configured to read the computer program in the memory and perform the following operations:

[0579] The data required for training the model is preprocessed;

[0580] The training model is trained based on the data required by the preprocessed training model to obtain the trained target model.

[0581] The processor 3800 is also configured to read the computer program in the memory and perform the following operations:

[0582] A first model training request is sent to the network function module on the RAN side; the first model training request is used to instruct the network function module on the RAN side to train the model to be trained according to the first model training request to obtain the trained target model; the first model training request carries the model requirements for training the model to be trained, and also carries the data requirements for training the model to be trained or the model data for training the model to be trained.

[0583] Receive the target model sent by the network function module on the RAN side.

[0584] The processor 3800 is also configured to read the computer program in the memory and perform the following operations:

[0585] The model to be trained is segmented to obtain a first RAN model and a first network function model;

[0586] The first RAN model is trained according to the data required for the training model to obtain the trained first RAN model;

[0587] Receive the trained first network function model sent by the network function module on the RAN side;

[0588] The first RAN model and the first network function model after training are integrated to obtain the target model.

[0589] The processor 3800 is also configured to read the computer program in the memory and perform the following operations:

[0590] Send a first model co-training request to the network function module on the RAN side;

[0591] The first model collaborative training request carries the model requirement for training the first network functional model, and also carries the data requirement for training the first network functional model or the model data for training the first network functional model.

[0592] The first model collaborative training request is used to instruct the network function module on the RAN side to train the first network function model according to the first model collaborative training request, so as to obtain the trained first network function model.

[0593] The processor 3800 is also configured to read the computer program in the memory and perform the following operations:

[0594] The model to be trained is trained according to the data required for the training model to obtain the first intermediate model after training;

[0595] Receive the trained target model sent by the network function module on the RAN side.

[0596] The processor 3800 is also configured to read the computer program in the memory and perform the following operations:

[0597] Send a first model incremental training request to the network function module on the RAN side;

[0598] The first model incremental training request carries the first intermediate model, as well as the data requirement for training the first intermediate model or the model data for training the first intermediate model.

[0599] The first model incremental training request is used to instruct the network function module on the RAN side to train the first intermediate model according to the first model incremental training request, so as to obtain the trained target model.

[0600] The processor 3800 is also configured to read the computer program in the memory and perform the following operations:

[0601] Receive the trained second intermediate model sent by the network function module on the RAN side;

[0602] The second intermediate model is trained using the data required for the training model to obtain the trained target model.

[0603] The processor 3800 is also configured to read the computer program in the memory and perform the following operations:

[0604] Send a second model incremental training request to the network function module on the RAN side;

[0605] The second model incremental training request carries the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained.

[0606] The second model incremental training request is used to instruct the network function module on the RAN side to train the model to be trained according to the second model incremental training request to obtain the second intermediate model.

[0607] The processor 3800 is also configured to read the computer program in the memory and perform the following operations:

[0608] The model to be trained is segmented to obtain a second RAN model and a second network functional model;

[0609] The second RAN model is trained based on the data required for the training model to obtain the trained second RAN model;

[0610] Receive the trained second network function model sent by the network function module on the core network side;

[0611] The trained second RAN model and the trained second network function model are integrated to obtain the target model.

[0612] The processor 3800 is also configured to read the computer program in the memory and perform the following operations:

[0613] Send a second model collaborative training request to the network function module on the core network side;

[0614] The second model collaborative training request carries the model requirements for training the second network functional model, as well as the data requirements for training the second network functional model or the model data for training the second network functional model.

[0615] The second model collaborative training request is used to instruct the network function module on the core network side to train the second network function model according to the second model collaborative training request, so as to obtain the trained second network function model.

[0616] The processor 3800 is also configured to read the computer program in the memory and perform the following operations:

[0617] The training requirements of the first model are analyzed to determine whether the computing resources required for the training of the first model exceed the computing resources of the RAN.

[0618] If the computing resources required for the training of the first model do not exceed the computing resources of the RAN, then the first model training strategy is determined to be to train the model on the RAN.

[0619] If the computing resources required for the training of the first model exceed the computing resources of the RAN, then the first model training strategy is determined to include any one of the following: training the model on the network function module on the RAN side, joint training the model with the network function module on the RAN side, and collaborative training the model with the network function module on the core network side.

[0620] The processor 3800 is also configured to read the computer program in the memory and perform the following operations:

[0621] Set and start the first timer, and acquire the data required for training the model;

[0622] When the first timer reaches its set time, the data required for training the model is sent to the network function module on the RAN side.

[0623] The processor is also configured to read the computer program in the memory and perform the following operations:

[0624] Upon receiving the first data report response, restart the first timer.

[0625] The processor 3800 is also configured to read the computer program in the memory and perform the following operations:

[0626] Determine whether the data required for the training model has been updated;

[0627] If it is determined that the data required for the training model has been updated, the updated data required for the training model will be sent to the network function module on the RAN side.

[0628] The processor 3800 is also configured to read the computer program in the memory and perform the following operations:

[0629] The target model is deployed on the RAN.

[0630] In one embodiment, this application provides a training apparatus for another model, which is similar to... Figure 38 The device shown contains the same components; see details for example. Figure 38 The schematic diagram shown illustrates a processor configured to read a computer program from the memory and perform the following operations:

[0631] Receive the second model training request sent by the model consumer;

[0632] Send the second model training requirement to the network function module on the RAN side to instruct the network function module on the RAN side to train the model to be trained according to the second model training requirement, so as to obtain the trained target model.

[0633] Receive the target model sent by the network function module on the RAN side.

[0634] The processor is also configured to read the computer program in the memory and perform the following operations:

[0635] The target model is deployed on the RAN.

[0636] In one embodiment, this application provides a training apparatus for another model, which is similar to... Figure 38 The device shown contains the same components; see details for example. Figure 38 The schematic diagram shown illustrates a processor configured to read a computer program from the memory and perform the following operations:

[0637] Receive a first training request sent by the RAN; the first training request includes one of a first model training request, a first model co-training request, a first model incremental training request, and a second model incremental training request;

[0638] The model is trained according to the first training request to obtain the trained model.

[0639] The trained model is sent to the RAN.

[0640] The processor is also configured to read the computer program in the memory and perform the following operations:

[0641] Obtain the model to be trained and the model data for training the model to be trained according to the first training request;

[0642] The model to be trained is trained based on the model data used to train the model to be trained, and the trained model is obtained.

[0643] The processor is also configured to read the computer program in the memory and perform the following operations:

[0644] The model to be trained is obtained according to the model requirements for training the model to be trained.

[0645] The processor is also configured to read the computer program in the memory and perform the following operations:

[0646] Extract the model to be trained from the first training request.

[0647] The processor is also configured to read the computer program in the memory and perform the following operations:

[0648] Based on the data requirements for training the model to be trained, the model data of the model to be trained is collected, or the model data of the model to be trained is extracted from the first training request.

[0649] The processor is also configured to read the computer program in the memory and perform the following operations:

[0650] The model data for training the model to be trained is preprocessed;

[0651] The step of training the model to be trained based on the model data to obtain the trained model includes:

[0652] The model to be trained is trained based on the preprocessed model data to obtain the trained model.

[0653] The processor is also configured to read the computer program in the memory and perform the following operations:

[0654] Receive the data required for training the model sent by RAN and start the second timer;

[0655] Send a first data reporting response to the RAN;

[0656] When the second timer reaches its set time, expired training model data stored locally will be deleted, or training model data exceeding local storage limits will be deleted.

[0657] The processor is also configured to read the computer program in the memory and perform the following operations:

[0658] Send a first data subscription request to the RAN;

[0659] Receive the data required for training the model corresponding to the first data subscription request sent by the RAN.

[0660] In one embodiment, this application provides a training apparatus for another model, which is similar to... Figure 38 The device shown contains the same components; see details for example. Figure 38 The schematic diagram shown illustrates a processor configured to read a computer program from the memory and perform the following operations:

[0661] Receive the second model training request sent by RAN;

[0662] The training model is trained according to the training requirements of the second model to obtain the trained target model.

[0663] The target model is sent to the RAN.

[0664] The processor is also configured to read the computer program in the memory and perform the following operations:

[0665] Determine the training strategy for the second model based on the training requirements of the second model;

[0666] The model to be trained is trained according to the second model training strategy and the data required for training the model, and the trained target model is obtained.

[0667] In one embodiment, this application provides a training apparatus for another model, which is similar to... Figure 38 The device shown contains the same components; see details for example. Figure 38 The schematic diagram shown illustrates a processor configured to read a computer program from the memory and perform the following operations:

[0668] Receive third-model training requests sent by model consumers;

[0669] Determine the training strategy for the third model based on the training requirements of the third model;

[0670] The training model is trained according to the third model training strategy and the data required for training the model, and the trained target model is obtained.

[0671] The processor is also configured to read the computer program in the memory and perform the following operations:

[0672] The data required for training the model is preprocessed;

[0673] The training model is trained based on the data required by the preprocessed training model to obtain the trained target model.

[0674] The processor is also configured to read the computer program in the memory and perform the following operations:

[0675] The model to be trained is segmented to obtain an intelligent network element model and a third network function model;

[0676] The intelligent network element model is trained based on the data required for the training model to obtain the trained intelligent network element model.

[0677] Receive the trained third network function model sent by the network function module on the core network side;

[0678] The trained third network function model and the trained intelligent network element model are integrated to obtain the target model.

[0679] The processor is also configured to read the computer program in the memory and perform the following operations:

[0680] Send a third model collaborative training request to the network function module on the core network side;

[0681] The third model collaborative training request carries the model requirements for training the third network functional model, as well as the data requirements for training the third network functional model or the model data for training the third network functional model.

[0682] The third model collaborative training request is used to instruct the network function modules on the core network side to train the third network function model according to the third model collaborative training request, so as to obtain the trained third network function model.

[0683] The processor is also configured to read the computer program in the memory and perform the following operations:

[0684] Receive the trained third intermediate model sent by the network function module on the core network side;

[0685] The third intermediate model is trained using the data required for the training model to obtain the trained target model.

[0686] The processor is also configured to read the computer program in the memory and perform the following operations:

[0687] Send a third model incremental training request to the network function module on the core network side;

[0688] The third model incremental training request carries the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained.

[0689] The third model incremental training request is used to instruct the network function module on the core network side to train the model to be trained according to the third model incremental training request, so as to obtain the trained third intermediate model.

[0690] The processor is also configured to read the computer program in the memory and perform the following operations:

[0691] The model to be trained is trained based on the data required for the training model to obtain the fourth intermediate model;

[0692] Receive the trained target model sent by the network function module on the core network side.

[0693] The processor is also configured to read the computer program in the memory and perform the following operations:

[0694] Send a fourth model incremental training request to the network function module on the core network side;

[0695] The fourth model incremental training request carries a fourth intermediate model, as well as the data requirement for training the fourth intermediate model or the model data for training the fourth intermediate model.

[0696] The fourth model incremental training request is used to instruct the network function module on the core network side to train the fourth intermediate model according to the fourth model incremental training request to obtain the trained target model.

[0697] The processor is also configured to read the computer program in the memory and perform the following operations:

[0698] The training requirements of the third model are analyzed to determine whether the computing resources required for the training of the third model exceed the computing resources of the intelligent network element.

[0699] If the computing resources required for the training of the third model do not exceed the computing resources of the network function module on the core network side, then the training strategy for the third model is determined to be to train the model on the intelligent network element.

[0700] If the computing resources required for training the third model exceed the computing resources of the network function module on the core network side, then the training strategy for the third model is determined to be joint training of the model with the network function module on the core network side.

[0701] The processor is also configured to read the computer program in the memory and perform the following operations:

[0702] Set and start the third timer, and acquire the data required for training the model;

[0703] When the third timer reaches its set time, the data required for training the model is sent to the network function module on the core network side.

[0704] The processor is also configured to read the computer program in the memory and perform the following operations:

[0705] Upon receiving the second data report response, the third timer is restarted.

[0706] The processor is also configured to read the computer program in the memory and perform the following operations:

[0707] Receive the second data subscription request sent by the network function module on the core network side;

[0708] Send the data required for training the model corresponding to the second data subscription request to the network function module on the core network side.

[0709] The processor is also configured to read the computer program in the memory and perform the following operations:

[0710] Determine whether the data required for the training model has been updated;

[0711] If it is determined that the data required for the training model has been updated, the updated data required for the training model will be sent to the network function module on the core network side.

[0712] The processor is also configured to read the computer program in the memory and perform the following operations:

[0713] The target model is deployed on the intelligent network element.

[0714] In one embodiment, this application provides a training apparatus for another model, which is similar to... Figure 38 The device shown contains the same components; see details for example. Figure 38 The schematic diagram shown illustrates a processor configured to read a computer program from the memory and perform the following operations:

[0715] Receive a second training request sent by a target device; the target device is any one of RAN, RAN-side network function module, and intelligent network element; the second training request includes one of a third model collaborative training request, a third model incremental training request, and a fourth model incremental training request.

[0716] The model is trained according to the second training request to obtain the trained model;

[0717] The trained model is sent to the target device.

[0718] The processor is also configured to read the computer program in the memory and perform the following operations:

[0719] Obtain the model to be trained and the model data for training the model to be trained according to the second training request;

[0720] The model to be trained is trained based on the model data used to train the model to be trained, and the trained model is obtained.

[0721] The processor is also configured to read the computer program in the memory and perform the following operations:

[0722] The model to be trained is obtained according to the model requirements for training the model to be trained.

[0723] The processor is also configured to read the computer program in the memory and perform the following operations:

[0724] The model to be trained is extracted from the second training request.

[0725] The processor is also configured to read the computer program in the memory and perform the following operations:

[0726] Based on the data requirements for training the model to be trained, the model data of the model to be trained is collected, or the model data of the model to be trained is extracted from the second training request.

[0727] The processor is also configured to read the computer program in the memory and perform the following operations:

[0728] Generate a model data collection request based on the data requirements for training the model to be trained;

[0729] Send the model data collection request to other intelligent network elements or data planes;

[0730] Receive model data returned by other intelligent network elements or data planes based on the model data collection request.

[0731] The processor is also configured to read the computer program in the memory and perform the following operations:

[0732] The model data for training the model to be trained is preprocessed;

[0733] The step of training the model to be trained based on the model data to obtain the trained model includes:

[0734] The model to be trained is trained based on the preprocessed model data to obtain the trained model.

[0735] The processor is also configured to read the computer program in the memory and perform the following operations:

[0736] Receive the data required for training the model sent by the intelligent network element and start the fourth timer;

[0737] Send a second data reporting response to the intelligent network element;

[0738] When the fourth timer reaches its set time, expired training model data stored locally will be deleted, or training model data exceeding local storage limits will be deleted.

[0739] The processor is also configured to read the computer program in the memory and perform the following operations:

[0740] Send a second data subscription request to the intelligent network element;

[0741] Receive the data required for training the model corresponding to the second data subscription request sent by the intelligent network element.

[0742] In one embodiment, a training apparatus for a model is provided, such as Figure 39 As shown, the device includes:

[0743] The first receiving module 10 is used to receive the first model training request sent by the model consumer;

[0744] The first determining module 11 is used to determine the first model training strategy according to the first model training requirements;

[0745] The first training module 12 is used to train the model to be trained according to the first model training strategy and the data required for training the model, so as to obtain the trained target model.

[0746] In one embodiment, the first model training strategy includes any one of the following:

[0747] Model training is performed on RAN;

[0748] Model training is performed on the network functional modules on the RAN side;

[0749] Joint training of the model with the network functional modules on the RAN side;

[0750] Collaborate with network function modules on the core network side for model training.

[0751] In one embodiment, the joint training of the model with the network functional modules on the RAN side includes:

[0752] The model can be trained collaboratively with the network functional modules on the RAN side, or incrementally trained with the network functional modules on the RAN side.

[0753] In one embodiment, the first model training strategy includes training the model on the RAN, and the first training module 12 includes:

[0754] The first preprocessing unit is used to preprocess the data required for the training model;

[0755] The first training unit is used to train the model to be trained based on the data required by the preprocessed training model, so as to obtain the trained target model.

[0756] In one embodiment, the first model training strategy includes training the model in the network function module on the RAN side, wherein the first training module 12 includes:

[0757] The first sending unit is used to send a first model training request to the network function module on the RAN side; the first model training request is used to instruct the network function module on the RAN side to train the model to be trained according to the first model training request to obtain the trained target model; the first model training request carries the model requirements for training the model to be trained, and also carries the data requirements for training the model to be trained or the model data for training the model to be trained.

[0758] The first receiving unit is used to receive the target model sent by the network function module on the RAN side.

[0759] In one embodiment, the first model training strategy includes co-training the model with the network functional modules on the RAN side, wherein the first training module 12 includes:

[0760] The first segmentation unit is used to segment the model to be trained to obtain a first RAN model and a first network function model;

[0761] The second training unit is used to train the first RAN model according to the data required by the training model, so as to obtain the trained first RAN model.

[0762] The second receiving unit is used to receive the trained first network function model sent by the network function module on the RAN side;

[0763] The first integration unit is used to integrate the trained first RAN model and the trained first network function model to obtain the target model.

[0764] In one embodiment, the apparatus further includes:

[0765] The fifth sending module is used to send a first model collaborative training request to the network function module on the RAN side;

[0766] The first model collaborative training request carries the model requirement for training the first network functional model, and also carries the data requirement for training the first network functional model or the model data for training the first network functional model.

[0767] The first model collaborative training request is used to instruct the network function module on the RAN side to train the first network function model according to the first model collaborative training request, so as to obtain the trained first network function model.

[0768] In one embodiment, the first model training strategy includes incremental model training with the network functional modules on the RAN side, wherein the first training module 12 includes:

[0769] The third training unit is used to train the model to be trained based on the data required by the training model, so as to obtain the first intermediate model after training.

[0770] The third receiving unit is used to receive the trained target model sent by the network function module on the RAN side.

[0771] In one embodiment, the apparatus further includes:

[0772] The sixth sending module is used to send a first model incremental training request to the network function module on the RAN side;

[0773] The first model incremental training request carries the first intermediate model, as well as the data requirement for training the first intermediate model or the model data for training the first intermediate model.

[0774] The first model incremental training request is used to instruct the network function module on the RAN side to train the first intermediate model according to the first model incremental training request, so as to obtain the trained target model.

[0775] In one embodiment, the first model training strategy includes incremental model training with the network functional modules on the RAN side, wherein the first training module 12 includes:

[0776] The fourth receiving unit is used to receive the trained second intermediate model sent by the network function module on the RAN side;

[0777] The fourth training unit is used to train the second intermediate model based on the data required by the training model to obtain the trained target model.

[0778] In one embodiment, the apparatus further includes:

[0779] The seventh sending module is used to send a second model incremental training request to the network function module on the RAN side;

[0780] The second model incremental training request carries the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained.

[0781] The second model incremental training request is used to instruct the network function module on the RAN side to train the model to be trained according to the second model incremental training request to obtain the second intermediate model.

[0782] In one embodiment, the first model training strategy includes co-training the model with network functional modules on the core network side. The first training module 12 includes:

[0783] The second segmentation unit is used to segment the model to be trained to obtain a second RAN model and a second network function model.

[0784] The fifth training unit is used to train the second RAN model based on the data required by the training model, so as to obtain the trained second RAN model.

[0785] The fifth receiving unit is used to receive the trained second network function model sent by the network function module on the core network side;

[0786] The second integration unit is used to integrate the trained second RAN model and the trained second network function model to obtain the target model.

[0787] In one embodiment, the apparatus further includes:

[0788] The eighth sending module is used to send a second model collaborative training request to the network function modules on the core network side;

[0789] The second model collaborative training request carries the model requirements for training the second network functional model, as well as the data requirements for training the second network functional model or the model data for training the second network functional model.

[0790] The second model collaborative training request is used to instruct the network function module on the core network side to train the second network function model according to the second model collaborative training request, so as to obtain the trained second network function model.

[0791] In one embodiment, the first determining module 11 includes:

[0792] The first analysis unit is used to analyze the training requirements of the first model and determine whether the computing resources required for the training of the first model exceed the computing resources of the RAN.

[0793] The first determining unit is configured to determine that the first model training strategy is to train the model on the RAN when the computing resources required for the first model training requirement do not exceed the computing resources of the RAN.

[0794] The second determining unit is used to determine, when the computing resources required for the first model training exceed the computing resources of the RAN, any one of the following: training the model on the network function module on the RAN side, jointly training the model with the network function module on the RAN side, or co-training the model with the network function module on the core network side.

[0795] In one embodiment, the apparatus further includes:

[0796] The first startup module is used to set and start the first timer and acquire the data required for the training model;

[0797] The ninth sending module is used to send the data required by the training model to the network function module on the RAN side when the first timer reaches the set time.

[0798] In one embodiment, the apparatus further includes:

[0799] The second startup module is used to restart the first timer when the first data report response is received.

[0800] In one embodiment, the apparatus further includes:

[0801] The eighth receiving module is used to receive the first data subscription request sent by the network function module on the RAN side;

[0802] The second sending unit is used to send the data required for training the model corresponding to the first data subscription request to the network function module on the RAN side.

[0803] In one embodiment, the apparatus further includes:

[0804] The third determining module is used to determine whether the data required for the training model has been updated.

[0805] The first update module is used to send the updated training model data to the network function module on the RAN side when it is determined that the data required for the training model has been updated.

[0806] In one embodiment, the apparatus further includes:

[0807] The first deployment module is used to deploy the target model on the RAN.

[0808] In one embodiment, a training apparatus for a model is provided, such as Figure 40 As shown, the device includes:

[0809] The second receiving module 20 is used to receive the second model training request sent by the model consumer.

[0810] The first sending module 21 is used to send the second model training requirement to the network function module on the RAN side, so as to instruct the network function module on the RAN side to train the model to be trained according to the second model training requirement, and obtain the trained target model.

[0811] The third receiving module 22 is used to receive the target model sent by the network function module on the RAN side.

[0812] In one embodiment, the training apparatus for the above model further includes:

[0813] The second deployment module is used to deploy the target model on the RAN.

[0814] In one embodiment, a training apparatus for a model is provided, such as Figure 41 As shown, the device includes:

[0815] The fourth receiving module 30 is used to receive a first training request sent by the RAN; the first training request includes one of a first model training request, a first model co-training request, a first model incremental training request, and a second model incremental training request.

[0816] The second training module 31 is used to train the model according to the first training request to obtain the trained model.

[0817] The second sending module 32 is used to send the trained model to the RAN.

[0818] In one embodiment, the second training module 31 includes:

[0819] The first acquisition unit is used to acquire the model to be trained and the model data for training the model to be trained according to the first training request.

[0820] The sixth training unit is used to train the model to be trained based on the model data of the training model to obtain the trained model.

[0821] In one embodiment, the first acquisition unit is specifically used to acquire the model to be trained according to the model requirements of the model to be trained when the first training request carries the model requirements of the model to be trained.

[0822] In one embodiment, the first acquisition unit is specifically used to extract the model to be trained from the first training request when the model to be trained is carried in the first training request.

[0823] In one embodiment, the first acquisition unit is specifically used to collect the model data of the model to be trained according to the data requirements of the model to be trained, or to extract the model data of the model to be trained from the first training request, when the first training request also carries the data requirements of the model to be trained or the model data of the model to be trained.

[0824] In one embodiment, the training apparatus for the above model further includes:

[0825] The first preprocessing module is used to preprocess the model data of the model to be trained.

[0826] The sixth training unit mentioned above is specifically used to train the model to be trained based on the preprocessed model data to obtain the trained model.

[0827] In one embodiment, the training apparatus for the above model further includes:

[0828] The third startup module is used to receive the data required for training the model sent by the RAN and start the second timer;

[0829] The tenth transmitting module is used to send a first data reporting response to the RAN;

[0830] The first deletion module is used to delete expired training model data stored locally, or to delete training model data that exceeds the local storage limit, when the second timer reaches its set time.

[0831] In one embodiment, the training apparatus for the model further includes:

[0832] The eleventh sending module is used to send a first data subscription request to the RAN;

[0833] The ninth receiving module is used to receive the data required for training the model corresponding to the first data subscription request sent by the RAN.

[0834] In one embodiment, a training apparatus for a model is provided, such as Figure 42 As shown, the device includes:

[0835] The fifth receiving module 40 is used to receive the second model training request sent by the RAN;

[0836] The third training module 41 is used to train the model to be trained according to the training requirements of the second model, so as to obtain the trained target model.

[0837] The third sending module 42 is used to send the target model to the RAN.

[0838] In one embodiment, the third training module 41 includes:

[0839] The third determining unit is used to determine the training strategy of the second model based on the training requirements of the second model.

[0840] The seventh training unit is used to train the model to be trained according to the second model training strategy and the data required for training the model, so as to obtain the trained target model.

[0841] In one embodiment, the second model training strategy includes any of the following:

[0842] Model training is performed on the network functional modules on the RAN side;

[0843] Joint training of the model with RAN;

[0844] Collaborate with network function modules on the core network side for model training.

[0845] In one embodiment, the joint training of the model with the RAN includes:

[0846] Perform model co-training with RAN, or perform incremental model training with RAN.

[0847] In one embodiment, a training apparatus for a model is provided, such as Figure 43 As shown, the device includes:

[0848] The sixth receiving module 50 is used to receive the third model training request sent by the model consumer;

[0849] The second determining module 51 is used to determine the third model training strategy according to the third model training requirements.

[0850] The fourth training module 52 is used to train the model to be trained according to the third model training strategy and the data required for training the model, so as to obtain the trained target model.

[0851] In one embodiment, the third model training strategy includes any of the following:

[0852] Model training is performed on intelligent network elements;

[0853] Joint training of models is conducted with network functional modules on the core network side.

[0854] In one embodiment, the joint training of the model with the network functional modules on the core network side includes:

[0855] The model can be trained collaboratively with the network functional modules on the core network side, or incrementally trained with the network functional modules on the core network side.

[0856] In one embodiment, the third model training strategy includes training the model on the intelligent network element, and the fourth training module 52 mentioned above includes:

[0857] The second preprocessing unit is used to preprocess the data required for the training model;

[0858] The eighth training unit is used to train the model to be trained based on the data required by the preprocessed training model, so as to obtain the trained target model.

[0859] In one embodiment, the third model training strategy includes co-training the model with the network functional modules on the core network side, and the aforementioned fourth training module 52 includes:

[0860] The third segmentation unit is used to segment the model to be trained to obtain the intelligent network element model and the third network function model.

[0861] The ninth training unit is used to train the intelligent network element model according to the data required by the training model, so as to obtain the trained intelligent network element model.

[0862] The sixth receiving unit is used to receive the trained third network function model sent by the network function module on the core network side;

[0863] The third integration unit is used to integrate the trained third network function model and the trained intelligent network element model to obtain the target model.

[0864] In one embodiment, the above-mentioned apparatus further includes:

[0865] The twelfth sending module is used to send a third model collaborative training request to the network function module on the core network side;

[0866] The third model collaborative training request carries the model requirements for training the third network functional model, as well as the data requirements for training the third network functional model or the model data for training the third network functional model.

[0867] The third model collaborative training request is used to instruct the network function modules on the core network side to train the third network function model according to the third model collaborative training request, so as to obtain the trained third network function model.

[0868] In one embodiment, the third model training strategy includes incremental model training with the network functional modules on the core network side, and the aforementioned fourth training module 52 includes:

[0869] The seventh receiving unit is used to receive the trained third intermediate model sent by the network function module on the core network side;

[0870] The tenth training unit is used to train the third intermediate model based on the data required by the training model to obtain the trained target model.

[0871] In one embodiment, the above-mentioned apparatus further includes:

[0872] The thirteenth sending module is used to send a third model incremental training request to the network function module on the core network side;

[0873] The third model incremental training request carries the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained.

[0874] The third model incremental training request is used to instruct the network function module on the core network side to train the model to be trained according to the third model incremental training request, so as to obtain the trained third intermediate model.

[0875] In one embodiment, the third model training strategy includes incremental model training with the network functional modules on the core network side, and the aforementioned fourth training module 52 includes:

[0876] The eleventh training unit is used to train the model to be trained based on the data required by the training model, so as to obtain the fourth intermediate model.

[0877] The eighth receiving unit is used to receive the trained target model sent by the network function module on the core network side.

[0878] In one embodiment, the above-mentioned apparatus further includes:

[0879] The fourteenth sending module is used to send a fourth model incremental training request to the network function module on the core network side;

[0880] The fourth model incremental training request carries a fourth intermediate model, as well as the data requirement for training the fourth intermediate model or the model data for training the fourth intermediate model.

[0881] The fourth model incremental training request is used to instruct the network function module on the core network side to train the fourth intermediate model according to the fourth model incremental training request, so as to obtain the trained target model.

[0882] In one embodiment, the second determining module 51 includes:

[0883] The second analysis unit is used to analyze the training requirements of the third model and determine whether the computing resources required for the training of the third model exceed the computing resources of the intelligent network element.

[0884] The fourth determining unit is used to determine that the third model training strategy is to train the model on the intelligent network element when the computing resources required for the training of the third model do not exceed the computing resources of the network function module on the core network side.

[0885] The fifth determining unit is used to determine the third model training strategy as joint training of the model with the network function module on the core network side when the computing resources required for the training of the third model exceed the computing resources of the network function module on the core network side.

[0886] In one embodiment, the above-mentioned apparatus further includes:

[0887] The fourth startup module is used to set and start the third timer and acquire the data required for the training model;

[0888] The fifteenth sending module is used to send the data required by the training model to the network function module on the core network side when the third timer reaches the set time.

[0889] In one embodiment, the above-mentioned apparatus further includes:

[0890] The fifth startup module is used to restart the third timer when the second data reporting response is received.

[0891] In one embodiment, the above-mentioned apparatus further includes:

[0892] The tenth receiving module is used to receive the second data subscription request sent by the network function module on the core network side;

[0893] The sixteenth sending module is used to send the data required for training the model corresponding to the second data subscription request to the network function module on the core network side.

[0894] In one embodiment, the above-mentioned apparatus further includes:

[0895] The fourth determining module is used to determine whether the data required for the training model has been updated;

[0896] The second update module is used to send the updated training model data to the network function module on the core network side when it is determined that the data required for the training model has been updated.

[0897] In one embodiment, the above-mentioned apparatus further includes:

[0898] The third deployment module is used to deploy the target model on the intelligent network element.

[0899] In one embodiment, a training apparatus for a model is provided, such as Figure 44 As shown, the device includes:

[0900] The seventh receiving module 60 is used to receive a second training request sent by the target device; the target device is any one of RAN, RAN-side network function module, and intelligent network element; the second training request includes one of the third model collaborative training request, the third model incremental training request, and the fourth model incremental training request.

[0901] The fifth training module 61 is used to train the model according to the second training request to obtain the trained model;

[0902] The fourth sending module 62 is used to send the trained model to the target device.

[0903] In one embodiment, the fifth training module 61 includes:

[0904] The second acquisition unit is used to acquire the model to be trained and the model data for training the model to be trained according to the second training request.

[0905] The twelfth training unit is used to train the model to be trained based on the model data of the training model to obtain the trained model.

[0906] In one embodiment, the second training request carries the model requirements for training the model to be trained, and the second acquisition unit is specifically used to acquire the model to be trained according to the model requirements for training the model to be trained.

[0907] In one embodiment, the second training request carries a model to be trained, and the second acquisition unit is specifically used to extract the model to be trained from the second training request.

[0908] In one embodiment, the second training request also carries the data requirement for training the model to be trained or the model data for training the model to be trained. The second acquisition unit is specifically used to collect the model data for training the model to be trained according to the data requirement for training the model to be trained, or to extract the model data for training the model to be trained from the second training request.

[0909] In one embodiment, the second acquisition unit is specifically used to generate a model data collection request based on the data requirements of the model to be trained; send the model data collection request to other intelligent network elements or data planes; and receive model data returned by the other intelligent network elements or data planes based on the model data collection request.

[0910] In one embodiment, the above-mentioned apparatus further includes:

[0911] The second preprocessing module is used to preprocess the model data of the model to be trained.

[0912] Correspondingly, the twelfth training unit mentioned above is specifically used to train the model to be trained based on the preprocessed model data to obtain the trained model.

[0913] In one embodiment, the above-mentioned apparatus further includes:

[0914] The sixth startup module is used to receive the data required for training the model sent by the intelligent network element and to start the fourth timer;

[0915] The seventeenth sending module is used to send a second data reporting response to the intelligent network element;

[0916] The second deletion module is used to delete expired training model data stored locally, or to delete training model data that exceeds the local storage limit, when the fourth timer reaches its set time.

[0917] In one embodiment, the above-mentioned apparatus further includes:

[0918] The eighteenth sending module is used to send a second data subscription request to the intelligent network element;

[0919] The eleventh receiving module is used to receive the data required for training the model corresponding to the second data subscription request sent by the intelligent network element.

[0920] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0921] Receive the first model training request sent by the model consumer;

[0922] Determine the first model training strategy based on the first model training requirements;

[0923] The model to be trained is trained according to the first model training strategy and the data required for training the model, and the trained target model is obtained.

[0924] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0925] The data required for training the model is preprocessed;

[0926] The training model is trained based on the data required by the preprocessed training model to obtain the trained target model.

[0927] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0928] A first model training request is sent to the network function module on the RAN side; the first model training request is used to instruct the network function module on the RAN side to train the model to be trained according to the first model training request to obtain the trained target model; the first model training request carries the model requirements for training the model to be trained, and also carries the data requirements for training the model to be trained or the model data for training the model to be trained.

[0929] Receive the target model sent by the network function module on the RAN side.

[0930] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0931] The model to be trained is segmented to obtain a first RAN model and a first network function model;

[0932] The first RAN model is trained according to the data required for the training model to obtain the trained first RAN model;

[0933] Receive the trained first network function model sent by the network function module on the RAN side;

[0934] The first RAN model and the first network function model after training are integrated to obtain the target model.

[0935] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0936] Send a first model co-training request to the network function module on the RAN side;

[0937] The first model collaborative training request carries the model requirement for training the first network functional model, and also carries the data requirement for training the first network functional model or the model data for training the first network functional model.

[0938] The first model collaborative training request is used to instruct the network function module on the RAN side to train the first network function model according to the first model collaborative training request, so as to obtain the trained first network function model.

[0939] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0940] The model to be trained is trained according to the data required for the training model to obtain the first intermediate model after training;

[0941] Receive the trained target model sent by the network function module on the RAN side.

[0942] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0943] Send a first model incremental training request to the network function module on the RAN side;

[0944] The first model incremental training request carries the first intermediate model, as well as the data requirement for training the first intermediate model or the model data for training the first intermediate model.

[0945] The first model incremental training request is used to instruct the network function module on the RAN side to train the first intermediate model according to the first model incremental training request, so as to obtain the trained target model.

[0946] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0947] Receive the trained second intermediate model sent by the network function module on the RAN side;

[0948] The second intermediate model is trained using the data required for the training model to obtain the trained target model.

[0949] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0950] Send a second model incremental training request to the network function module on the RAN side;

[0951] The second model incremental training request carries the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained.

[0952] The second model incremental training request is used to instruct the network function module on the RAN side to train the model to be trained according to the second model incremental training request to obtain the second intermediate model.

[0953] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0954] The model to be trained is segmented to obtain a second RAN model and a second network functional model;

[0955] The second RAN model is trained based on the data required for the training model to obtain the trained second RAN model;

[0956] Receive the trained second network function model sent by the network function module on the core network side;

[0957] The trained second RAN model and the trained second network function model are integrated to obtain the target model.

[0958] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0959] Send a second model collaborative training request to the network function module on the core network side;

[0960] The second model collaborative training request carries the model requirements for training the second network functional model, as well as the data requirements for training the second network functional model or the model data for training the second network functional model.

[0961] The second model collaborative training request is used to instruct the network function module on the core network side to train the second network function model according to the second model collaborative training request, so as to obtain the trained second network function model.

[0962] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0963] The training requirements of the first model are analyzed to determine whether the computing resources required for the training of the first model exceed the computing resources of the RAN.

[0964] If the computing resources required for the training of the first model do not exceed the computing resources of the RAN, then the first model training strategy is determined to be to train the model on the RAN.

[0965] If the computing resources required for the training of the first model exceed the computing resources of the RAN, then the first model training strategy is determined to include any one of the following: training the model on the network function module on the RAN side, joint training the model with the network function module on the RAN side, and collaborative training the model with the network function module on the core network side.

[0966] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0967] Set and start the first timer, and acquire the data required for training the model;

[0968] When the first timer reaches its set time, the data required for training the model is sent to the network function module on the RAN side.

[0969] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0970] Upon receiving the first data report response, restart the first timer.

[0971] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0972] Receive the first data subscription request sent by the network function module on the RAN side;

[0973] Send the data required for training the model corresponding to the first data subscription request to the network function module on the RAN side.

[0974] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0975] Determine whether the data required for the training model has been updated;

[0976] If it is determined that the data required for the training model has been updated, the updated data required for the training model will be sent to the network function module on the RAN side.

[0977] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0978] The target model is deployed on the RAN.

[0979] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0980] Receive the second model training request sent by the model consumer;

[0981] Send the second model training requirement to the network function module on the RAN side to instruct the network function module on the RAN side to train the model to be trained according to the second model training requirement, so as to obtain the trained target model.

[0982] Receive the target model sent by the network function module on the RAN side.

[0983] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0984] The target model is deployed on the RAN.

[0985] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0986] Receive a first training request sent by the RAN; the first training request includes one of a first model training request, a first model co-training request, a first model incremental training request, and a second model incremental training request;

[0987] The model is trained according to the first training request to obtain the trained model.

[0988] The trained model is sent to the RAN.

[0989] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0990] Obtain the model to be trained and the model data for training the model to be trained according to the first training request;

[0991] The model to be trained is trained based on the model data used to train the model to be trained, and the trained model is obtained.

[0992] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0993] The model to be trained is obtained according to the model requirements for training the model to be trained.

[0994] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0995] Extract the model to be trained from the first training request.

[0996] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0997] Based on the data requirements for training the model to be trained, the model data of the model to be trained is collected, or the model data of the model to be trained is extracted from the first training request.

[0998] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0999] The model data for training the model to be trained is preprocessed;

[1000] The step of training the model to be trained based on the model data to obtain the trained model includes:

[1001] The model to be trained is trained based on the preprocessed model data to obtain the trained model.

[1002] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1003] Receive the data required for training the model sent by RAN and start the second timer;

[1004] Send a first data reporting response to the RAN;

[1005] When the second timer reaches its set time, expired training model data stored locally will be deleted, or training model data exceeding local storage limits will be deleted.

[1006] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1007] Send a first data subscription request to the RAN;

[1008] Receive the data required for training the model corresponding to the first data subscription request sent by the RAN.

[1009] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[1010] Receive the second model training request sent by RAN;

[1011] The training model is trained according to the training requirements of the second model to obtain the trained target model.

[1012] The target model is sent to the RAN.

[1013] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1014] Determine the training strategy for the second model based on the training requirements of the second model;

[1015] The model to be trained is trained according to the second model training strategy and the data required for training the model, and the trained target model is obtained.

[1016] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[1017] Receive third-model training requests sent by model consumers;

[1018] Determine the training strategy for the third model based on the training requirements of the third model;

[1019] The training model is trained according to the third model training strategy and the data required for training the model, and the trained target model is obtained.

[1020] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1021] The data required for training the model is preprocessed;

[1022] The training model is trained based on the data required by the preprocessed training model to obtain the trained target model.

[1023] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1024] The model to be trained is segmented to obtain an intelligent network element model and a third network function model;

[1025] The intelligent network element model is trained based on the data required for the training model to obtain the trained intelligent network element model.

[1026] Receive the trained third network function model sent by the network function module on the core network side;

[1027] The trained third network function model and the trained intelligent network element model are integrated to obtain the target model.

[1028] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1029] Send a third model collaborative training request to the network function module on the core network side;

[1030] The third model collaborative training request carries the model requirements for training the third network functional model, as well as the data requirements for training the third network functional model or the model data for training the third network functional model.

[1031] The third model collaborative training request is used to instruct the network function modules on the core network side to train the third network function model according to the third model collaborative training request, so as to obtain the trained third network function model.

[1032] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1033] Receive the trained third intermediate model sent by the network function module on the core network side;

[1034] The third intermediate model is trained using the data required for the training model to obtain the trained target model.

[1035] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1036] Send a third model incremental training request to the network function module on the core network side;

[1037] The third model incremental training request carries the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained.

[1038] The third model incremental training request is used to instruct the network function module on the core network side to train the model to be trained according to the third model incremental training request, so as to obtain the trained third intermediate model.

[1039] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1040] The model to be trained is trained based on the data required for the training model to obtain the fourth intermediate model;

[1041] Receive the trained target model sent by the network function module on the core network side.

[1042] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1043] Send a fourth model incremental training request to the network function module on the core network side;

[1044] The fourth model incremental training request carries a fourth intermediate model, as well as the data requirement for training the fourth intermediate model or the model data for training the fourth intermediate model.

[1045] The fourth model incremental training request is used to instruct the network function module on the core network side to train the fourth intermediate model according to the fourth model incremental training request, so as to obtain the trained target model.

[1046] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1047] The training requirements of the third model are analyzed to determine whether the computing resources required for the training of the third model exceed the computing resources of the intelligent network element.

[1048] If the computing resources required for the training of the third model do not exceed the computing resources of the network function module on the core network side, then the training strategy for the third model is determined to be to train the model on the intelligent network element.

[1049] If the computing resources required for training the third model exceed the computing resources of the network function module on the core network side, then the training strategy for the third model is determined to be joint training of the model with the network function module on the core network side.

[1050] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1051] Set and start the third timer, and acquire the data required for training the model;

[1052] When the third timer reaches its set time, the data required for training the model is sent to the network function module on the core network side.

[1053] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1054] Upon receiving the second data report response, the third timer is restarted.

[1055] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1056] Receive the second data subscription request sent by the network function module on the core network side;

[1057] Send the data required for training the model corresponding to the second data subscription request to the network function module on the core network side.

[1058] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1059] Determine whether the data required for the training model has been updated;

[1060] If it is determined that the data required for the training model has been updated, the updated data required for the training model will be sent to the network function module on the core network side.

[1061] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1062] The target model is deployed on the intelligent network element.

[1063] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[1064] Receive a second training request sent by a target device; the target device is any one of RAN, RAN-side network function module, and intelligent network element; the second training request includes one of a third model collaborative training request, a third model incremental training request, and a fourth model incremental training request.

[1065] The model is trained according to the second training request to obtain the trained model;

[1066] The trained model is sent to the target device.

[1067] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1068] Obtain the model to be trained and the model data for training the model to be trained according to the second training request;

[1069] The model to be trained is trained based on the model data used to train the model to be trained, and the trained model is obtained.

[1070] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1071] The model to be trained is obtained according to the model requirements for training the model to be trained.

[1072] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1073] The model to be trained is extracted from the second training request.

[1074] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1075] Based on the data requirements for training the model to be trained, the model data of the model to be trained is collected, or the model data of the model to be trained is extracted from the second training request.

[1076] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1077] Generate a model data collection request based on the data requirements for training the model to be trained;

[1078] Send the model data collection request to other intelligent network elements or data planes;

[1079] Receive model data returned by other intelligent network elements or data planes based on the model data collection request.

[1080] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1081] The model data for training the model to be trained is preprocessed;

[1082] The step of training the model to be trained based on the model data to obtain the trained model includes:

[1083] The model to be trained is trained based on the preprocessed model data to obtain the trained model.

[1084] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1085] Receive the data required for training the model sent by the intelligent network element and start the fourth timer;

[1086] Send a second data reporting response to the intelligent network element;

[1087] When the fourth timer reaches its set time, expired training model data stored locally will be deleted, or training model data exceeding local storage limits will be deleted.

[1088] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[1089] Send a second data subscription request to the intelligent network element;

[1090] Receive the data required for training the model corresponding to the second data subscription request sent by the intelligent network element.

[1091] This application also provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the following steps:

[1092] Receive the first model training request sent by the model consumer;

[1093] Determine the first model training strategy based on the first model training requirements;

[1094] The model to be trained is trained according to the first model training strategy and the data required for training the model, and the trained target model is obtained.

[1095] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1096] The data required for training the model is preprocessed;

[1097] The training model is trained based on the data required by the preprocessed training model to obtain the trained target model.

[1098] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1099] A first model training request is sent to the network function module on the RAN side; the first model training request is used to instruct the network function module on the RAN side to train the model to be trained according to the first model training request to obtain the trained target model; the first model training request carries the model requirements for training the model to be trained, and also carries the data requirements for training the model to be trained or the model data for training the model to be trained.

[1100] Receive the target model sent by the network function module on the RAN side.

[1101] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1102] The model to be trained is segmented to obtain a first RAN model and a first network function model;

[1103] The first RAN model is trained according to the data required for the training model to obtain the trained first RAN model;

[1104] Receive the trained first network function model sent by the network function module on the RAN side;

[1105] The first RAN model and the first network function model after training are integrated to obtain the target model.

[1106] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1107] Send a first model co-training request to the network function module on the RAN side;

[1108] The first model collaborative training request carries the model requirement for training the first network functional model, and also carries the data requirement for training the first network functional model or the model data for training the first network functional model.

[1109] The first model collaborative training request is used to instruct the network function module on the RAN side to train the first network function model according to the first model collaborative training request, so as to obtain the trained first network function model.

[1110] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1111] The model to be trained is trained according to the data required for the training model to obtain the first intermediate model after training;

[1112] Receive the trained target model sent by the network function module on the RAN side.

[1113] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1114] Send a first model incremental training request to the network function module on the RAN side;

[1115] The first model incremental training request carries the first intermediate model, as well as the data requirement for training the first intermediate model or the model data for training the first intermediate model.

[1116] The first model incremental training request is used to instruct the network function module on the RAN side to train the first intermediate model according to the first model incremental training request, so as to obtain the trained target model.

[1117] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1118] Send a first model incremental training request to the network function module on the RAN side;

[1119] The first model incremental training request carries the first intermediate model, as well as the data requirement for training the first intermediate model or the model data for training the first intermediate model.

[1120] The first model incremental training request is used to instruct the network function module on the RAN side to train the first intermediate model according to the first model incremental training request, so as to obtain the trained target model.

[1121] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1122] Receive the trained second intermediate model sent by the network function module on the RAN side;

[1123] The second intermediate model is trained using the data required for the training model to obtain the trained target model.

[1124] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1125] Send a second model incremental training request to the network function module on the RAN side;

[1126] The second model incremental training request carries the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained.

[1127] The second model incremental training request is used to instruct the network function module on the RAN side to train the model to be trained according to the second model incremental training request to obtain the second intermediate model.

[1128] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1129] The model to be trained is segmented to obtain a second RAN model and a second network functional model;

[1130] The second RAN model is trained based on the data required for the training model to obtain the trained second RAN model;

[1131] Receive the trained second network function model sent by the network function module on the core network side;

[1132] The trained second RAN model and the trained second network function model are integrated to obtain the target model.

[1133] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1134] Send a second model collaborative training request to the network function module on the core network side;

[1135] The second model collaborative training request carries the model requirements for training the second network functional model, as well as the data requirements for training the second network functional model or the model data for training the second network functional model.

[1136] The second model collaborative training request is used to instruct the network function module on the core network side to train the second network function model according to the second model collaborative training request, so as to obtain the trained second network function model.

[1137] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1138] The training requirements of the first model are analyzed to determine whether the computing resources required for the training of the first model exceed the computing resources of the RAN.

[1139] If the computing resources required for the training of the first model do not exceed the computing resources of the RAN, then the first model training strategy is determined to be to train the model on the RAN.

[1140] If the computing resources required for the training of the first model exceed the computing resources of the RAN, then the first model training strategy is determined to include any one of the following: training the model on the network function module on the RAN side, joint training the model with the network function module on the RAN side, and collaborative training the model with the network function module on the core network side.

[1141] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1142] Set and start the first timer, and acquire the data required for training the model;

[1143] When the first timer reaches its set time, the data required for training the model is sent to the network function module on the RAN side.

[1144] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1145] Upon receiving the first data report response, restart the first timer.

[1146] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1147] Receive the first data subscription request sent by the network function module on the RAN side;

[1148] Send the data required for training the model corresponding to the first data subscription request to the network function module on the RAN side.

[1149] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1150] Determine whether the data required for the training model has been updated;

[1151] If it is determined that the data required for the training model has been updated, the updated data required for the training model will be sent to the network function module on the RAN side.

[1152] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1153] The target model is deployed on the RAN.

[1154] This application also provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the following steps:

[1155] Receive the second model training request sent by the model consumer;

[1156] Send the second model training requirement to the network function module on the RAN side to instruct the network function module on the RAN side to train the model to be trained according to the second model training requirement, so as to obtain the trained target model.

[1157] Receive the target model sent by the network function module on the RAN side.

[1158] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1159] The target model is deployed on the RAN.

[1160] This application also provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the following steps:

[1161] Receive a first training request sent by the RAN; the first training request includes one of a first model training request, a first model co-training request, a first model incremental training request, and a second model incremental training request;

[1162] The model is trained according to the first training request to obtain the trained model.

[1163] The trained model is sent to the RAN.

[1164] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1165] Obtain the model to be trained and the model data for training the model to be trained according to the first training request;

[1166] The model to be trained is trained based on the model data used to train the model to be trained, and the trained model is obtained.

[1167] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1168] The model to be trained is obtained according to the model requirements for training the model to be trained.

[1169] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1170] Extract the model to be trained from the first training request.

[1171] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1172] Based on the data requirements for training the model to be trained, the model data of the model to be trained is collected, or the model data of the model to be trained is extracted from the first training request.

[1173] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1174] The model data for training the model to be trained is preprocessed;

[1175] The step of training the model to be trained based on the model data to obtain the trained model includes:

[1176] The model to be trained is trained based on the preprocessed model data to obtain the trained model.

[1177] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1178] Receive the data required for training the model sent by RAN and start the second timer;

[1179] Send a first data reporting response to the RAN;

[1180] When the second timer reaches its set time, expired training model data stored locally will be deleted, or training model data exceeding local storage limits will be deleted.

[1181] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1182] Send a first data subscription request to the RAN;

[1183] Receive the data required for training the model corresponding to the first data subscription request sent by the RAN.

[1184] This application also provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the following steps:

[1185] Receive the second model training request sent by RAN;

[1186] The training model is trained according to the training requirements of the second model to obtain the trained target model.

[1187] The target model is sent to the RAN.

[1188] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1189] Determine the training strategy for the second model based on the training requirements of the second model;

[1190] The model to be trained is trained according to the second model training strategy and the data required for training the model, and the trained target model is obtained.

[1191] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1192] Receive third-model training requests sent by model consumers;

[1193] Determine the training strategy for the third model based on the training requirements of the third model;

[1194] The training model is trained according to the third model training strategy and the data required for training the model, and the trained target model is obtained.

[1195] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1196] The data required for training the model is preprocessed;

[1197] The training model is trained based on the data required by the preprocessed training model to obtain the trained target model.

[1198] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps: segmenting the model to be trained to obtain an intelligent network element model and a third network function model;

[1199] The intelligent network element model is trained based on the data required for the training model to obtain the trained intelligent network element model.

[1200] Receive the trained third network function model sent by the network function module on the core network side;

[1201] The trained third network function model and the trained intelligent network element model are integrated to obtain the target model.

[1202] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1203] Send a third model collaborative training request to the network function module on the core network side;

[1204] The third model collaborative training request carries the model requirements for training the third network functional model, as well as the data requirements for training the third network functional model or the model data for training the third network functional model.

[1205] The third model collaborative training request is used to instruct the network function modules on the core network side to train the third network function model according to the third model collaborative training request, so as to obtain the trained third network function model.

[1206] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1207] Receive the trained third intermediate model sent by the network function module on the core network side;

[1208] The third intermediate model is trained using the data required for the training model to obtain the trained target model.

[1209] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1210] Send a third model incremental training request to the network function module on the core network side;

[1211] The third model incremental training request carries the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained.

[1212] The third model incremental training request is used to instruct the network function module on the core network side to train the model to be trained according to the third model incremental training request, so as to obtain the trained third intermediate model.

[1213] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1214] The model to be trained is trained based on the data required for the training model to obtain the fourth intermediate model;

[1215] Receive the trained target model sent by the network function module on the core network side.

[1216] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1217] Send a fourth model incremental training request to the network function module on the core network side;

[1218] The fourth model incremental training request carries a fourth intermediate model, as well as the data requirement for training the fourth intermediate model or the model data for training the fourth intermediate model.

[1219] The fourth model incremental training request is used to instruct the network function module on the core network side to train the fourth intermediate model according to the fourth model incremental training request, so as to obtain the trained target model.

[1220] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1221] The training requirements of the third model are analyzed to determine whether the computing resources required for the training of the third model exceed the computing resources of the intelligent network element.

[1222] If the computing resources required for the training of the third model do not exceed the computing resources of the network function module on the core network side, then the training strategy for the third model is determined to be to train the model on the intelligent network element.

[1223] If the computing resources required for training the third model exceed the computing resources of the network function module on the core network side, then the training strategy for the third model is determined to be joint training of the model with the network function module on the core network side.

[1224] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1225] Set and start the third timer, and acquire the data required for training the model;

[1226] When the third timer reaches its set time, the data required for training the model is sent to the network function module on the core network side.

[1227] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1228] Upon receiving the second data report response, the third timer is restarted.

[1229] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1230] Receive the second data subscription request sent by the network function module on the core network side;

[1231] Send the data required for training the model corresponding to the second data subscription request to the network function module on the core network side.

[1232] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1233] Determine whether the data required for the training model has been updated;

[1234] If it is determined that the data required for the training model has been updated, the updated data required for the training model will be sent to the network function module on the core network side.

[1235] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1236] The target model is deployed on the intelligent network element.

[1237] This application also provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the following steps:

[1238] Receive a second training request sent by a target device; the target device is any one of RAN, RAN-side network function module, and intelligent network element; the second training request includes one of a third model collaborative training request, a third model incremental training request, and a fourth model incremental training request.

[1239] The model is trained according to the second training request to obtain the trained model;

[1240] The trained model is sent to the target device.

[1241] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1242] Obtain the model to be trained and the model data for training the model to be trained according to the second training request;

[1243] The model to be trained is trained based on the model data used to train the model to be trained, and the trained model is obtained.

[1244] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1245] The model to be trained is obtained according to the model requirements for training the model to be trained.

[1246] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1247] The model to be trained is extracted from the second training request.

[1248] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1249] Based on the data requirements for training the model to be trained, the model data of the model to be trained is collected, or the model data of the model to be trained is extracted from the second training request.

[1250] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1251] Generate a model data collection request based on the data requirements for training the model to be trained;

[1252] Send the model data collection request to other intelligent network elements or data planes;

[1253] Receive model data returned by other intelligent network elements or data planes based on the model data collection request.

[1254] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1255] The model data for training the model to be trained is preprocessed;

[1256] The step of training the model to be trained based on the model data to obtain the trained model includes:

[1257] The model to be trained is trained based on the preprocessed model data to obtain the trained model.

[1258] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1259] Receive the data required for training the model sent by the intelligent network element and start the fourth timer;

[1260] Send a second data reporting response to the intelligent network element;

[1261] When the fourth timer reaches its set time, expired training model data stored locally will be deleted, or training model data exceeding local storage limits will be deleted.

[1262] In one embodiment, when a computer program product is run on a computer, the computer performs the following steps:

[1263] Send a second data subscription request to the intelligent network element;

[1264] Receive the data required for training the model corresponding to the second data subscription request sent by the intelligent network element.

[1265] It should be noted that the division of units in the embodiments of this application is illustrative and only represents one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units.

[1266] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[1267] This application also provides a processor-readable storage medium storing a program. When executed by a processor, this program implements the various processes of the above-described beamforming method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The readable storage medium can be any available medium or data storage device accessible to the processor, including but not limited to magnetic storage (e.g., floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO), etc.), optical storage (e.g., CDs, DVDs, BDs, HVDs, etc.), and semiconductor storage (e.g., ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs), etc.).

[1268] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[1269] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[1270] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for training a model, characterized in that, The method is applied to a radio access network (RAN), and the method includes: Receive the first model training request sent by the model consumer; Determine the first model training strategy based on the first model training requirements; The model to be trained is trained according to the first model training strategy and the data required for training the model, and the trained target model is obtained.

2. The method according to claim 1, characterized in that, The first model training strategy includes any one of the following: Model training is performed on RAN; Model training is performed on the network functional modules on the RAN side; Joint training of the model with the network functional modules on the RAN side; Collaborate with network function modules on the core network side for model training.

3. The method according to claim 2, characterized in that, The joint training of the model with the network functional modules on the RAN side includes: The model can be trained collaboratively with the network functional modules on the RAN side, or incrementally trained with the network functional modules on the RAN side.

4. The method according to claim 1, characterized in that, The first model training strategy includes training the model on the RAN, and the step of training the model to be trained according to the first model training strategy and the data required for training the model to obtain the trained target model includes: The data required for training the model is preprocessed; The training model is trained based on the data required by the preprocessed training model to obtain the trained target model.

5. The method according to claim 1, characterized in that, The first model training strategy includes training the model in the network functional modules on the RAN side. The step of training the model to be trained according to the first model training strategy and the data required for training the model to obtain the trained target model includes: A first model training request is sent to the network function module on the RAN side; the first model training request is used to instruct the network function module on the RAN side to train the model to be trained according to the first model training request to obtain the trained target model; the first model training request carries the model requirements for training the model to be trained, and also carries the data requirements for training the model to be trained or the model data for training the model to be trained. Receive the target model sent by the network function module on the RAN side.

6. The method according to claim 1, characterized in that, The first model training strategy includes co-training the model with the network functional modules on the RAN side. The step of training the model to be trained according to the first model training strategy and the data required for training the model to obtain the trained target model includes: The model to be trained is segmented to obtain a first RAN model and a first network function model; The first RAN model is trained according to the data required for the training model to obtain the trained first RAN model; Send a first model collaborative training request to the network function module on the RAN side; the first model collaborative training request carries the model requirements for training the first network function model, and also carries the data requirements for training the first network function model or the model data for training the first network function model; Receive the trained first network function model sent by the network function module on the RAN side; The first RAN model and the first network function model after training are integrated to obtain the target model.

7. The method according to claim 1, characterized in that, The first model training strategy includes incremental model training with the network functional modules on the RAN side. The step of training the model to be trained according to the first model training strategy and the data required for training the model to obtain the trained target model includes: The model to be trained is trained according to the data required for the training model to obtain the first intermediate model after training; Send a first model incremental training request to the network function module on the RAN side; the first model incremental training request carries the first intermediate model, and also carries the data requirements for training the first intermediate model or the model data for training the first intermediate model; Receive the trained target model sent by the network function module on the RAN side.

8. The method according to claim 1, characterized in that, The first model training strategy includes incremental model training with the network functional modules on the RAN side. The step of training the model to be trained according to the first model training strategy and the data required for training the model to obtain the trained target model includes: Send a second model incremental training request to the network function module on the RAN side; the second model incremental training request carries the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained; Receive the trained second intermediate model sent by the network function module on the RAN side; The second intermediate model is trained using the data required for the training model to obtain the trained target model.

9. The method according to claim 1, characterized in that, The first model training strategy includes collaborative model training with network functional modules on the core network side. The step of training the model to be trained according to the first model training strategy and the data required for training the model to obtain the trained target model includes: The model to be trained is segmented to obtain a second RAN model and a second network functional model; The second RAN model is trained based on the data required for the training model to obtain the trained second RAN model; Send a second model collaborative training request to the network function module on the core network side; the second model collaborative training request carries the model requirements for training the second network function model, as well as the data requirements for training the second network function model or the model data for training the second network function model; Receive the trained second network function model sent by the network function module on the core network side; The trained second RAN model and the trained second network function model are integrated to obtain the target model.

10. The method according to any one of claims 1-9, characterized in that, Determining the first model training strategy based on the first model training requirements includes: The training requirements of the first model are analyzed to determine whether the computing resources required for the training of the first model exceed the computing resources of the RAN. If the computing resources required for the training of the first model do not exceed the computing resources of the RAN, then the first model training strategy is determined to be to train the model on the RAN. If the computing resources required for the training of the first model exceed the computing resources of the RAN, then the first model training strategy is determined to include any one of the following: training the model on the network function module on the RAN side, joint training the model with the network function module on the RAN side, and collaborative training the model with the network function module on the core network side.

11. The method according to any one of claims 1-9, characterized in that, The method further includes: Set and start the first timer, and acquire the data required for training the model; When the first timer reaches its set time, the data required for training the model is sent to the network function module on the RAN side. Upon receiving the first data report response, restart the first timer.

12. The method according to any one of claims 1-9, characterized in that, The method further includes: Receive the first data subscription request sent by the network function module on the RAN side; Send the data required for training the model corresponding to the first data subscription request to the network function module on the RAN side.

13. The method according to any one of claims 1-9, characterized in that, The method further includes: Determine whether the data required for the training model has been updated; If it is determined that the data required for the training model has been updated, the updated data required for the training model will be sent to the network function module on the RAN side.

14. The method according to any one of claims 1-9, characterized in that, The method further includes: The target model is deployed on the RAN.

15. A method for training a model, characterized in that, The method is applied to a RAN, and the method includes: Receive the second model training request sent by the model consumer; Send the second model training requirement to the network function module on the RAN side to instruct the network function module on the RAN side to train the model to be trained according to the second model training requirement, so as to obtain the trained target model. Receive the target model sent by the network function module on the RAN side.

16. A method for training a model, characterized in that, The method is applied to the network function module on the RAN side, and the method includes: Receive a first training request sent by the RAN; the first training request includes one of a first model training request, a first model co-training request, a first model incremental training request, and a second model incremental training request; The model is trained according to the first training request to obtain the trained model. The trained model is sent to the RAN.

17. The method according to claim 16, characterized in that, The step of training the model according to the first training request to obtain the trained model includes: Obtain the model to be trained and the model data for training the model to be trained according to the first training request; The model to be trained is trained based on the model data used to train the model to be trained, and the trained model is obtained.

18. The method according to claim 17, characterized in that, The first training request carries the model requirement for training the model to be trained. The step of obtaining the model to be trained according to the first training request includes: The model to be trained is obtained according to the model requirements for training the model to be trained.

19. The method according to claim 17, characterized in that, The first training request carries a model to be trained, and the step of obtaining the model to be trained according to the first training request includes: Extract the model to be trained from the first training request.

20. The method according to claim 17, characterized in that, The first training request also carries the data requirement for training the model to be trained or the model data for training the model to be trained. The step of obtaining the model data for training the model to be trained according to the first training request includes: Based on the data requirements for training the model to be trained, the model data of the model to be trained is collected, or the model data of the model to be trained is extracted from the first training request.

21. The method according to claim 17, characterized in that, The method further includes: The model data for training the model to be trained is preprocessed; The step of training the model to be trained based on the model data to obtain the trained model includes: The model to be trained is trained based on the preprocessed model data to obtain the trained model.

22. The method according to any one of claims 16-21, characterized in that, The method further includes: Receive the data required for training the model sent by RAN and start the second timer; Send a first data reporting response to the RAN; When the second timer reaches its set time, expired training model data stored locally will be deleted, or training model data exceeding local storage limits will be deleted.

23. The method according to any one of claims 16-21, characterized in that, The method further includes: Send a first data subscription request to the RAN; Receive the data required for training the model corresponding to the first data subscription request sent by the RAN.

24. A method for training a model, characterized in that, The method is applied to the network function module on the RAN side, and the method includes: Receive the second model training request sent by RAN; The training model is trained according to the training requirements of the second model to obtain the trained target model. The target model is sent to the RAN.

25. The method according to claim 24, characterized in that, The step of training the model to be trained according to the second model training requirements to obtain the trained target model includes: Determine the training strategy for the second model based on the training requirements of the second model; The model to be trained is trained according to the second model training strategy and the data required for training the model, and the trained target model is obtained.

26. The method according to claim 25, characterized in that, The second model training strategy includes any one of the following: Model training is performed on the network functional modules on the RAN side; Joint training of the model with RAN; Collaborate with network function modules on the core network side for model training.

27. The method according to claim 26, characterized in that, The joint training of the model with RAN includes: Perform model co-training with RAN, or perform incremental model training with RAN.

28. A method for training a model, characterized in that, The method is applied to intelligent network elements, and the method includes: Receive third-model training requests sent by model consumers; Determine the training strategy for the third model based on the training requirements of the third model; The training model is trained according to the third model training strategy and the data required for training the model, and the trained target model is obtained.

29. The method according to claim 28, characterized in that, The third model training strategy includes any one of the following: Model training is performed on intelligent network elements; Joint training of models is conducted with network functional modules on the core network side.

30. The method according to claim 29, characterized in that, The joint training of the model with the network functional modules on the core network side includes: The model can be trained collaboratively with the network functional modules on the core network side, or incrementally trained with the network functional modules on the core network side.

31. The method according to claim 28, characterized in that, The third model training strategy includes training the model on intelligent network elements. The step of training the model to be trained according to the third model training strategy and the data required for training the model, to obtain the trained target model, includes: The data required for training the model is preprocessed; The training model is trained based on the data required by the preprocessed training model to obtain the trained target model.

32. The method according to claim 28, characterized in that, The third model training strategy includes collaborative model training with network functional modules on the core network side. The step of training the model to be trained according to the third model training strategy and the data required for training the model to obtain the trained target model includes: The model to be trained is segmented to obtain an intelligent network element model and a third network function model; The intelligent network element model is trained based on the data required for the training model to obtain the trained intelligent network element model. Send a third model collaborative training request to the network function module on the core network side; the third model collaborative training request carries the model requirements for training the third network function model, as well as the data requirements for training the third network function model or the model data for training the third network function model. Receive the trained third network function model sent by the network function module on the core network side; The trained third network function model and the trained intelligent network element model are integrated to obtain the target model.

33. The method according to any one of claims 28-32, characterized in that, The third model training strategy includes incremental model training with the network functional modules on the core network side. The step of training the model to be trained according to the third model training strategy and the data required for training the model to obtain the trained target model includes: Send a third model incremental training request to the network function module on the core network side; the third model incremental training request carries the model to be trained, as well as the data requirements for training the model to be trained or the model data for training the model to be trained; Receive the trained third intermediate model sent by the network function module on the core network side; The third intermediate model is trained using the data required for the training model to obtain the trained target model.

34. The method according to claim 28, characterized in that, The third model training strategy includes incremental model training with the network functional modules on the core network side. The step of training the model to be trained according to the third model training strategy and the data required for training the model to obtain the trained target model includes: The model to be trained is trained based on the data required for the training model to obtain the fourth intermediate model; Send a fourth model incremental training request to the network function module on the core network side; the fourth model incremental training request carries a fourth intermediate model, as well as the data requirements for training the fourth intermediate model or the model data for training the fourth intermediate model; Receive the trained target model sent by the network function module on the core network side.

35. The method according to any one of claims 28-32, characterized in that, The step of determining the third model training strategy based on the third model training requirements includes: The training requirements of the third model are analyzed to determine whether the computing resources required for the training of the third model exceed the computing resources of the intelligent network element. If the computing resources required for the training of the third model do not exceed the computing resources of the network function module on the core network side, then the training strategy for the third model is determined to be to train the model on the intelligent network element. If the computing resources required for training the third model exceed the computing resources of the network function module on the core network side, then the training strategy for the third model is determined to be joint training of the model with the network function module on the core network side.

36. The method according to any one of claims 28-32, characterized in that, The method further includes: Set and start the third timer, and acquire the data required for the training model; When the third timer reaches its set time, the data required for training the model is sent to the network function module on the core network side. Upon receiving the second data report response, the third timer is restarted.

37. The method according to any one of claims 28-32, characterized in that, The method further includes: Receive the second data subscription request sent by the network function module on the core network side; Send the data required for training the model corresponding to the second data subscription request to the network function module on the core network side.

38. The method according to any one of claims 28-32, characterized in that, The method further includes: Determine whether the data required for the training model has been updated; If it is determined that the data required for the training model has been updated, the updated data required for the training model will be sent to the network function module on the core network side.

39. The method according to any one of claims 28-32, characterized in that, The method further includes: The target model is deployed on the intelligent network element.

40. A method for training a model, characterized in that, The method is applied to network function modules on the core network side, and the method includes: Receive a second training request sent by a target device; the target device is any one of RAN, RAN-side network function module, and intelligent network element; the second training request includes one of a third model collaborative training request, a third model incremental training request, and a fourth model incremental training request. The model is trained according to the second training request to obtain the trained model; The trained model is sent to the target device.

41. The method according to claim 40, characterized in that, The step of training the model according to the second training request to obtain the trained model includes: Obtain the model to be trained and the model data for training the model to be trained according to the second training request; The model to be trained is trained based on the model data used to train the model to be trained, and the trained model is obtained.

42. The method according to claim 41, characterized in that, The second training request carries the model requirement for training the model to be trained. Obtaining the model to be trained according to the second training request includes: The model to be trained is obtained according to the model requirements for training the model to be trained.

43. The method according to claim 41, characterized in that, The second training request carries a model to be trained. Obtaining the model to be trained according to the second training request includes: The model to be trained is extracted from the second training request.

44. The method according to claim 41, characterized in that, The second training request also carries the data requirement for training the model to be trained or the model data for training the model to be trained. The step of obtaining the model data for training the model to be trained according to the second training request includes: Based on the data requirements for training the model to be trained, the model data of the model to be trained is collected, or the model data of the model to be trained is extracted from the second training request.

45. The method according to claim 44, characterized in that, The step of collecting model data for training the model to be trained according to the data requirements of the training model includes: Generate a model data collection request based on the data requirements for training the model to be trained; Send the model data collection request to other intelligent network elements or data planes; Receive model data returned by other intelligent network elements or data planes based on the model data collection request.

46. ​​The method according to claim 41, characterized in that, The method further includes: The model data for training the model to be trained is preprocessed; The step of training the model to be trained based on the model data to obtain the trained model includes: The model to be trained is trained based on the preprocessed model data to obtain the trained model.

47. The method according to any one of claims 40-46, characterized in that, The method further includes: Receive the data required for training the model sent by the intelligent network element and start the fourth timer; Send a second data reporting response to the intelligent network element; When the fourth timer reaches its set time, expired training model data stored locally will be deleted, or training model data exceeding local storage limits will be deleted.

48. The method according to any one of claims 40-46, characterized in that, The method further includes: Send a second data subscription request to the intelligent network element; Receive the data required for training the model corresponding to the second data subscription request sent by the intelligent network element.

49. A training device for a model, characterized in that, The device includes: a memory, a transceiver, and a processor. A memory for storing computer programs; a transceiver for sending and receiving data under the control of the processor; and a processor for reading the computer programs from the memory and performing the following operations: Receive the first model training request sent by the model consumer; Determine the first model training strategy based on the first model training requirements; The model to be trained is trained according to the first model training strategy and the data required for training the model, and the trained target model is obtained.

50. A training device for a model, characterized in that, The device includes: a memory, a transceiver, and a processor. A memory for storing computer programs; a transceiver for sending and receiving data under the control of the processor; and a processor for reading the computer programs from the memory and performing the following operations: Receive the second model training request sent by the model consumer; Send the second model training requirement to the network function module on the RAN side to instruct the network function module on the RAN side to train the model to be trained according to the second model training requirement, so as to obtain the trained target model. Receive the target model sent by the network function module on the RAN side.

51. A training device for a model, characterized in that, The device includes: a memory, a transceiver, and a processor. A memory for storing computer programs; a transceiver for sending and receiving data under the control of the processor; and a processor for reading the computer programs from the memory and performing the following operations: Receive a first training request sent by the RAN; the first training request includes one of a first model training request, a first model co-training request, a first model incremental training request, and a second model incremental training request; The model is trained according to the first training request to obtain the trained model. The trained model is sent to the RAN.

52. A training device for a model, characterized in that, The device includes: a memory, a transceiver, and a processor. A memory for storing computer programs; a transceiver for sending and receiving data under the control of the processor; and a processor for reading the computer programs from the memory and performing the following operations: Receive the second model training request sent by RAN; The training model is trained according to the training requirements of the second model to obtain the trained target model. The target model is sent to the RAN.

53. A training device for a model, characterized in that, The device includes: a memory, a transceiver, and a processor. A memory for storing computer programs; a transceiver for sending and receiving data under the control of the processor; and a processor for reading the computer programs from the memory and performing the following operations: Receive third-model training requests sent by model consumers; Determine the training strategy for the third model based on the training requirements of the third model; The training model is trained according to the third model training strategy and the data required for training the model, and the trained target model is obtained.

54. A training device for a model, characterized in that, The device includes: a memory, a transceiver, and a processor. A memory for storing computer programs; a transceiver for sending and receiving data under the control of the processor; and a processor for reading the computer programs from the memory and performing the following operations: Receive a second training request sent by a target device; the target device is any one of RAN, RAN-side network function module, and intelligent network element; the second training request includes one of a third model collaborative training request, a third model incremental training request, and a fourth model incremental training request. The model is trained according to the second training request to obtain the trained model; The trained model is sent to the target device.

Images