Network Optimization Method, Server, Client Device, Network Device, and Medium

Abstract

The present application proposes a network optimization method, a server, a client device, a network device, and a medium. The method includes: sending a session establishment request message to the client device to request the client device to perform measurement configuration on a specified network-side device and perform measurement configuration on a terminal device connected to the specified network-side device; receiving a measurement report message of the specified network-side device and a measurement report message of the terminal device; determining whether the client device needs to perform collaborative training according to the pre-acquired machine learning description information; based on whether the client device needs to perform collaborative training and the measurement data in the received measurement report message, performing specified model training processing for network optimization, and sending the result of the model training processing to the client device for instructing the client device to obtain network optimization operations according to the result of the model training processing.

Description

Technical Field

[0001] This application relates to the field of communications, and in particular, to a network optimization method, a server, a client device, a network device, and a medium. Background Art

[0002] The 5th Generation Wireless Systems Network is currently being deployed and is expected to develop beyond 5G (B5G) networks in the future.

[0003] When optimizing a network, there is no network intelligence solution on how to configure the Artificial Intelligence (AI) function for a communication network, so as to perform network intelligent optimization processes based on AI. Summary of the Invention

[0004] This application provides a network optimization method, a server, a client device, a network device, and a medium.

[0005] In a first aspect, an embodiment of this application provides a network optimization method, including: sending a session establishment request message to a client device to request the client device to perform measurement configuration on a specified network-side device and perform measurement configuration on a terminal device connected to the specified network-side device; receiving a measurement report message of the specified network-side device and a measurement report message of the terminal device; determining whether it is necessary for the client device to perform collaborative training according to the pre-obtained machine learning description information; based on whether it is necessary for the client device to perform collaborative training and the measurement data in the received measurement report message, performing specified model training processing for network optimization, and sending the result of the model training processing to the client device for instructing the client device to obtain network optimization operations according to the result of the model training processing.

[0006] In a second aspect, an embodiment of this application provides a network optimization method, including: in response to receiving a session establishment request message from a predetermined server, performing measurement configuration on a specified network-side device and a terminal device connected to the specified network-side device according to the measurement control information included in the session establishment request message; sending a measurement report message of the specified network-side device and a measurement report message of the terminal device to the predetermined server, and the measurement report of the specified network-side device and the measurement report of the terminal device are used for model training processing for network optimization in the predetermined server; in response to receiving the result of the model training processing from the predetermined server, obtaining network optimization operations according to the result of the model training processing.

[0007] In a third aspect, an embodiment of the present application provides a network optimization method, including: in response to receiving a measurement configuration request from a client device, performing measurement configuration according to the network-side measurement control information in the measurement configuration request and performing measurement configuration on a terminal device connected to the network-side device; sending the measurement report of the current network-side device obtained by measurement and the measurement report of the received terminal device to a predetermined server and a predetermined client device respectively, and the measurement report of the current network-side device and the measurement report of the terminal device are used in the predetermined server and the predetermined client device for model training processing for network optimization.

[0008] In a fourth aspect, an embodiment of the present application provides a server, including: a measurement configuration request module, configured to send a session establishment request message to a client device to request the client device to perform measurement configuration on a specified network-side device and perform measurement configuration on a terminal device connected to the specified network-side device; a measurement report receiving module, configured to receive a measurement report message of the specified network-side device and a measurement report message of the terminal device; a collaborative training determination module, configured to determine whether client device collaborative training is required according to pre-acquired machine learning description information; a model training processing module, configured to perform specified model training processing for network optimization based on whether client device collaborative training is required and the measurement data in the received measurement report message, and send the model training processing result to the client device for instructing the client device to obtain network optimization operations according to the model training processing result.

[0009] In a fifth aspect, an embodiment of the present application provides a client device, including: a measurement configuration module, configured to, in response to receiving a session establishment request message from a predetermined server, perform measurement configuration on a specified network-side device and a terminal device connected to the specified network-side device according to the measurement control information included in the session establishment request message; a measurement report sending module, configured to send a measurement report message of the specified network-side device and a measurement report message of the terminal device to the predetermined server, and the measurement report of the specified network-side device and the measurement report of the terminal device are used in the predetermined server for model training processing for network optimization; an optimization operation determination module, configured to, in response to receiving the model training processing result from the predetermined server, process the model training processing result to obtain network optimization operations.

[0010] Sixth aspect, an embodiment of the present application provides a network-side device, including: a measurement configuration module, configured to, in response to receiving a measurement configuration request from a client device, perform measurement configuration according to network-side measurement control information in the measurement configuration request and perform measurement configuration on a terminal device connected to the present network-side device; a measurement report sending module, configured to send the measurement report of the current network-side device obtained by measurement and the measurement report of the received terminal device to a predetermined server and a predetermined client device respectively, and the measurement report of the current network-side device and the measurement report of the terminal device are used in the predetermined server and the predetermined client device for model training processing for network optimization.

[0011] Seventh aspect, an embodiment of the present application provides a network optimization system, including: a server, configured to execute the network optimization method of the first aspect above; one or more client devices, configured to execute the network optimization method of the second aspect above; one or more network-side devices, configured to execute the network optimization method of the third aspect above.

[0012] Eighth aspect, an embodiment of the present application provides a network device, including: one or more processors; a memory, storing one or more programs thereon, when the one or more programs are executed by the one or more processors, enabling the one or more processors to implement any one of the network optimization methods in the embodiments of the present application.

[0013] Ninth aspect, an embodiment of the present application provides a storage medium, storing a computer program, which when executed by a processor implements any one of the network optimization methods in the embodiments of the present application.

[0014] According to the network optimization method, server, network device, and medium of the embodiments of the present application, by requesting the client device to perform measurement configuration on the network-side device and the terminal-side device, and determining whether to perform collaborative training with the client device according to the pre-acquired machine learning description information, different model training processes are performed according to the determination result, so as to implement the deployment of the machine learning training process through the server and the client device, perform distributed model training and processing, obtain network optimization operations, and thus perform in-depth analysis on the collected data through artificial intelligence and machine learning in the network system, providing a new optimization method and a network intelligent optimization process for operator network optimization.

[0015] The network optimization method, client device, network device, and medium according to the embodiments of the present application respond to the session establishment request message of the server, perform measurement configuration on the network-side device and the terminal-side device, and respond to the model training processing result from a predetermined server, process to obtain network optimization operations, and implement distributed model training processing with the server based on the model training processing result transmitted between the server and the client, and finally obtain network optimization operations, so that the data collected by the network-side device and the terminal device can be deeply analyzed in machine learning, and provide new optimization methods and network intelligent optimization processes for operator network optimization through distributed model training.

[0016] The network optimization method, network-side device, network device, and medium according to the embodiments of the present application can perform measurement configuration according to the received network-side measurement control information and perform measurement configuration on the connected terminal device, and send the measurement report obtained by performing the measurement and the measurement report of the received terminal-side device to a predetermined server and a client device. The measurement reports of the network-side device and the terminal device are used in the predetermined server and the predetermined client device for model training processing for network optimization, so that the data collected by the network-side device and the terminal device can be deeply analyzed in artificial intelligence and machine learning, and distributed model training processing is performed in the predetermined server and the predetermined client device, providing new optimization methods and network intelligent optimization processes for operator network optimization.

[0017] More descriptions about the above embodiments and other aspects of the present application and their implementation manners are provided in the drawings description, specific implementation manners, and claims. Description of the Drawings

[0018] Figure 1 Flow schematic diagram of the network optimization method showing an embodiment of the present application.

[0019] Figure 2 Flow schematic diagram of the network optimization method showing another embodiment of the present application.

[0020] Figure 3 Flow schematic diagram of the network optimization method showing still another embodiment of the present application.

[0021] Figure 4 Timing flow schematic diagram of the network optimization method showing an embodiment of the present application.

[0022] Figure 5 Timing flow schematic diagram of the network optimization method showing another embodiment of the present application.

[0023] Figure 6 Flow schematic diagram of establishing a communication interface between a server and a network-side device showing an embodiment of the present application.

[0024] Figure 7 Schematic flowchart of establishing a data channel between a server and a user device according to an embodiment of the present application.

[0025] Figure 8 Schematic structural diagram of a server according to an embodiment of the present application.

[0026] Figure 9 Schematic structural diagram of a client device according to an embodiment of the present application.

[0027] Figure 10 Schematic structural diagram of a network-side device according to an embodiment of the present application.

[0028] Figure 11 Schematic structural diagram of a network optimization system according to an embodiment of the present application.

[0029] Figure 12 It is a structural diagram showing an exemplary hardware architecture of a computing device capable of implementing the embodiments of the present application. Detailed implementation manners

[0030] To make the objectives, technical solutions and advantages of the present application more clear and understandable, the embodiments of the present application will be described in detail below with reference to the accompanying drawings. It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined arbitrarily with each other.

[0031] In the embodiments of the present application, while the 5G network brings fundamental changes in performance and flexibility, the complexity of network operation is also increasing significantly. Operators urgently need automated and intelligent means to reduce network deployment and operation and maintenance costs, and improve network performance and user experience. Self-Organized Network (SON) and Minimization of Drive Tests (MDT) in LTE and 5G play a great role in network optimization, but they do not achieve the goal of a flexible and intelligent network. The 5G network faces challenges in the joint optimization of many network Key Performance Indicator (KPI) indicators. These indicators include, for example, latency, reliability, connection density, user experience, etc.

[0032] Traditional network optimization methods are increasingly showing problems such as long response cycles, error-proneness, and high costs. The measurement quantities in the Remote Asynchronous Node (RAN) node devices and terminal devices can, on the one hand, be used by the network management system to monitor network KPIs and can also assist network manufacturers in optimizing radio resource management. The embodiments of the present application provide a network optimization solution. Through artificial intelligence / machine learning, by deeply analyzing the collected data, a new network optimization method is provided for operator network optimization, realizing the support for the AI-based network intelligent optimization process in the existing network architecture.

[0033] In the embodiments of the present application, the server device (AI Centralized Server), also referred to as an AI centralized server or an AI controller with AI capabilities, and one or more client devices (AI Distributed Client), which can also be referred to as AI distributed clients with AI capabilities. Each client device can be arranged on a RAN node device, and the RAN node device can include network-side node devices such as base stations.

[0034] Figure 1 The flowchart showing the network optimization method according to an embodiment of the present application is as follows. Figure 1 As shown, the network optimization method in the embodiments of the present application may include the following steps.

[0035] S110: Send a session establishment request message to the client device to request the client device to perform measurement configuration on the specified network-side device and on the terminal device connected to the specified network-side device.

[0036] S120: Receive the measurement report message of the specified network-side device and the measurement report message of the terminal device.

[0037] S130: Determine whether client device collaboration training is required according to the pre-obtained machine learning description information.

[0038] S140: Based on whether client device collaboration training is required and the measurement data in the received measurement report message, perform specified model training processing for network optimization, and send the model training processing result to the client device to instruct the client device to obtain network optimization operations according to the model training processing result.

[0039] According to the network optimization method of the embodiments of the present application, the client device is requested to perform measurement configuration on the network-side device and the terminal-side device, and it is determined whether to perform collaborative training with the client device according to the pre-acquired machine learning description information, and different model training processes are performed according to the determination result, so as to implement the deployment of the machine learning training process through the server and the client device, to implement distributed model training and processing, obtain network optimization operations, and thus perform in-depth analysis on the collected data through artificial intelligence and machine learning in the network system, providing a new optimization method and a network intelligent optimization process for operator network optimization.

[0040] In one embodiment, before step S110, the network optimization method further includes: in response to the received first activation message, activating the machine learning function, and obtaining the machine learning description information carried in the activation message and the training hyperparameters of the corresponding machine learning model.

[0041] In this embodiment, the network management and maintenance system (Operation Administration And Maintenance, OAM) or the core network sends a machine learning activation message (Activation Message) through the interface of the AI central server to instruct the AI central server to activate or use the ML (Machine learning) function. Among them, the AI central server or the AI control node can be located inside or outside the core network.

[0042] Exemplarily, the activation message includes an indication indicating whether the ML function is activated; ML description information (ML Description Information); the training hyperparameters of the corresponding ML model, such as hyperparameters such as learning rate (Learning Rate), number of epochs (Epoch), and training batch size (Batchsize).

[0043] In one embodiment, the ML description message includes one or more of the following: one or more selected ML types such as supervised learning, unsupervised learning, reinforcement learning, deep learning, transfer learning, etc., and one or more selected ML models such as convolutional neural network (CNN), recurrent neural network (RNN), long short-term memory network (LSTM), support vector machine (SVM), autoregressive moving average model (ARIMA), decision tree, etc.

[0044] In this embodiment, the OAM or the core network can activate the AI function of the AI centralized server, and control the ML algorithms and ML models used for model training of the AI centralized server through the ML description information. Through the ML algorithms and ML models used for model training by the AI centralized server, the AI centralized server can determine whether it is necessary to cooperate with the AI distributed client for model training.

[0045] In one embodiment, the session establishment request message may include measurement control information for the client device, an item of computing processing capability information of the client device, and one or more machine learning models required to be supported by the client device, where the measurement configuration information is used to indicate the measurement quantities and measurement reporting methods that need to be configured by the specified network-side device.

[0046] In this embodiment, the AI centralized server sends an ML session establishment request message (ML SESSION SETUP REQUEST) to the AI distributed client in one or more base stations for configuring the radio-side measurement data required for a certain ML session.

[0047] In one embodiment, the session request message may include measurement control information for indicating what measurements the AI distributed client should perform and how to report them; the AI computing processing capability of the distributed client, which may include, for example, the memory size of the CPU of the AI distributed client, the video memory size of the GPU, the resource utilization rate of the current state of the CPU / GPU, etc.; and an indication that the distributed client reports one or more supported ML models.

[0048] In one embodiment, the session request message may further include an ML session identifier (ML Session ID) for uniquely identifying a certain machine learning session process. If the ML session ID is not included in the session request message, it indicates that the measurement control in the request message is for all ML processes. That is, for all machine learning processes, the measurement quantities to be collected and the measurement reporting methods can be indicated.

[0049] In the embodiments of the present application, the machine learning process can be used to represent the machine learning corresponding to the measurement control information carried in the session establishment request. Exemplarily, the machine learning process can be determined according to different optimization metrics. For example, different optimization metrics are set for different communication quality metrics, and different optimization metrics correspond to different machine learning processes; or, the machine learning process can be determined by the machine learning type, and the machine learning type includes but is not limited to any one of supervised learning, unsupervised learning, reinforcement learning, deep learning, and transfer learning; or, the machine learning process can be determined by the machine learning model, and the machine learning model includes but is not limited to any one of: convolutional neural network, recurrent neural network, long short-term memory network, support vector machine, autoregressive moving average model, and decision tree.

[0050] In one embodiment, after step S110, the network optimization method further includes: in response to the received session establishment response message, determining, according to the success flag carried in the session establishment response message, the information indicating that the client can meet the requirements for the computing and processing capabilities of the client device.

[0051] In this embodiment, it can be determined whether the measurement configuration for the network-side client and the terminal device is successful through the session establishment response message, and whether the client device meets one or more ML models supported by the client required in the session establishment request.

[0052] Exemplarily, if the measurements on the base station side and the terminal side are both configured successfully, and the client device located at the base station can meet the ML model in the establishment request message, the base station sends an ML session establishment response message to the AI centralized server, where the message carries a success indication. If the measurement on the base station side or the UE side is configured failed, or the ML model in the establishment request message cannot be met, the base station sends an ML session establishment response message to the AI centralized server, where the message carries a failure indication, and the ML session ID (ML SessionID) is optionally carried in the message. Exemplarily, the ML session ID can be carried in the ML session establishment response message.

[0053] In one embodiment, S120 may specifically include: receiving the measurement report of the specified network-side device and the measurement report of the terminal device sent by the specified network-side device.

[0054] In this embodiment, the AI centralized server is not directly connected to the terminal device. After the terminal device reports the measurement report to the network-side device, the network-side device forwards the measurement report of the terminal device, and the AI server can obtain the measurement report of the terminal device through the communication interface with the network-side device.

[0055] In one embodiment, the machine learning description information includes one or more selected machine learning types and one or more selected machine learning models; S130 may specifically include: S131, obtaining a preset training mode corresponding to the machine learning type and the machine learning model according to the selected machine learning model of the selected machine learning type; S132, determining whether client devices are required to perform collaborative training according to the preset training mode.

[0056] In this embodiment, the ML algorithm and the ML model used by the AI centralized server for model training are controlled through the ML description information to determine whether it is necessary to perform model training in cooperation with the AI distributed client. Exemplarily, a first correspondence relationship between a selected part of the ML algorithms and ML models and not requiring the cooperation of the AI distributed client for model training, and a second correspondence relationship between another selected part of the ML algorithms and ML models and requiring the cooperation of the AI distributed client for model training can be determined in advance. Then, according to the first correspondence relationship and the second correspondence relationship, it is determined whether it is necessary to perform model training in cooperation with the AI distributed client according to the selected ML algorithms and ML models. When it is not necessary to perform model training in cooperation with the AI distributed client, the AI distributed client can perform model training processes such as ML model derivation and model execution.

[0057] In one embodiment, when it is determined that client devices are not required to perform collaborative training, the model training processing result includes the trained machine learning model and the model parameter values; step S140 may specifically include the following steps.

[0058] S11, obtaining the computing and processing capabilities of the client device from the received measurement report of the specified network-side device, where the client device is deployed within the specified network-side device.

[0059] S12, selecting a machine learning algorithm for model training according to the received measurement data, the computing and processing capabilities of the client device, and the machine learning models supported by the client device obtained in advance, to obtain the trained machine learning model and the model parameter values.

[0060] S13, sending a first model deployment request message to the client device for generating a first network optimization operation instruction on the client device; where the first model deployment request message includes: a first model derivation configuration file, the trained machine learning model and the model parameter values, a first model performance reporting indication, and a first model performance reporting method.

[0061] In an embodiment of the present application, an ML Deployment Message is sent to distributed clients within one or more base stations for sending an ML model inference (or execution) configuration file.

[0062] Exemplarily, the model inference configuration file may include one or more of the following: GPU video memory usage size, input data size, output data size, etc.

[0063] Exemplarily, the trained ML model and ML model parameter values may include one or more of the following: Model graph of the ML model, trained weight values (Weights) of the ML model, trained bias values (Bias) of the ML model, etc.

[0064] Exemplarily, the model performance reporting indication information includes one or more of the following: Mean Squared Error (MSE), Mean Absolute Error (MAE), Accuracy; and the model performance reporting method, such as event-based reporting or periodic reporting, etc.

[0065] In this embodiment, when it is determined that client devices are not required for collaborative training, the AI central server selects a suitable ML algorithm for model training and model iterative update based on the measurement data, the computing processing capabilities of the AI distributed clients, and the supported ML models. The machine learning model and model parameter values obtained from the model training are sent to the distributed clients, enabling the distributed clients to perform model updates, model inferences, model instructions, and other model processing based on the received machine learning model and model parameter values, so as to obtain optimized operation instructions at the distributed clients.

[0066] In this embodiment, the AI server may notify the network-side device to perform relevant optimization operations (Action) through a RAN Action Request. The optimization operation action may be one or more operation instructions and the parameters required for the corresponding operations, for example, including but not limited to: UE handover indication, cell shutdown / activation indication, radio resource activation / deactivation indication, power adjustment indication, RRM parameter reconfiguration indication, traffic splitting operation indication, protocol layer parameter reconfiguration indication, etc.

[0067] During the model training process, the AI central server does not directly generate optimized operation instructions, but sends the trained model and model parameters to the distributed clients, enabling the distributed clients to perform targeted model training based on the measurement data in the received measurement reports, and obtaining optimized operations that are more suitable for the current distributed clients.

[0068] In one embodiment, after step S13, the network optimization method further includes: S14, in response to receiving a first model deployment response message from a client device, determining that the deployment of the trained machine learning model on the client device is successful according to the first model deployment success flag in the first model deployment response message; or, S15, determining that the deployment of the trained machine learning model on the client device fails according to the first model deployment failure flag in the first model deployment response message.

[0069] In this embodiment, the network side device (such as a base station) sends an ML model deployment response message (ML Deployment Response Message). If the ML model deployment is configured successfully, the base station sends an ML model deployment response message to the AI central server and carries a success indication. The ML session ID is optionally carried in this message. If the ML model deployment is configured fails, the base station can send an ML model deployment response message to the AI central server and carry a failure indication. The ML session ID is optionally carried in the message.

[0070] In one embodiment, after S13, the network optimization method further includes: S16, in response to receiving a first model performance report message from the network side device, determining whether to re-perform model training processing according to the model performance metric value carried in the first model performance report message.

[0071] In the embodiments of the present application, the model performance metrics include but are not limited to network KPI metrics and network energy saving performance, such as mean square error (MSE), mean absolute error (MAE), accuracy (Accuracy), etc. The model performance reporting method can be, for example, event reporting or periodic reporting and other various reporting methods. The embodiments of the present application do not make specific limitations.

[0072] In this embodiment, the AI central server determines whether to retrain and deploy the network according to the model performance report message reported by the network side device.

[0073] In one embodiment, when it is determined that the client device needs to perform collaborative training, step S140 may specifically include the following steps.

[0074] S21. Obtain the computing and processing capabilities reported by the client device from the received measurement report of the specified network-side device, where the client device is deployed on the specified network-side device; S23. Select a machine learning algorithm and configure the training hyperparameter values according to the received measurement data, the previously obtained computing and processing capabilities of the client device, and the machine learning models supported by the client device; S24. Send a model configuration establishment message to the client device to request the client device to perform model configuration according to the machine learning algorithm and the training hyperparameter values; S25. Receive the model configuration establishment message from the client device, where the model configuration establishment message includes the machine learning model information obtained by the client device through model training for the model configuration of the machine learning algorithm and the training hyperparameter values; S26. Perform model configuration and model training processing based on the trained machine learning model information, obtain the model training processing result, and send the model training processing result to the client device.

[0075] In this embodiment, through steps S21 - S25, the AI central server selects a suitable ML algorithm and configures the training hyperparameter values according to the measurement data, the AI computing and processing capabilities of the AI distributed client, and the supported ML models, and sends an ML model configuration establishment message (ML Model Configuration Message) to the AI distributed client in one or more base stations through the AI central server, for sending the selected ML model and the selected ML training hyperparameters.

[0076] In one embodiment, the machine learning model information may, for example, include the trained weight values (Weights) of the ML model, the trained bias values (Bias) of the ML model, etc. The message may optionally include an ML session ID (ML SessionID), which is used to uniquely identify the ID of a certain process.

[0077] Among them, the ML model configuration establishment message and the model configuration establishment message may optionally include an ML session ID (MLSession ID), which is used to uniquely identify the ID of a certain process.

[0078] In one embodiment, the model training processing result includes a second network optimization operation instruction; step S26 may specifically include: S2601. Continue to perform model training according to the received measurement data and the trained machine learning model information to obtain a second network optimization operation instruction; S2602. Send a second network operation request message to the network-side device, where the second network operation request message includes the second network optimization operation instruction, a second model performance reporting indication, and a second model performance reporting method.

[0079] In this embodiment, the machine learning model information of the AI centralized server that has been trained continues to be trained to obtain a network optimization operation instruction. After step S2602, the network optimization method further includes: S2603, in response to receiving a second model performance report message from a network-side device, determining whether to re-perform model training processing according to the model performance metric value carried in the second model performance report message.

[0080] In this embodiment, the second model performance report message optionally carries an ML session ID (ML sessionID). Whether to re-perform model training processing is determined through the model performance metric value in the second model performance report message, so as to obtain a performance metric that meets the requirements of the network side through model training processing.

[0081] In one embodiment, the model training processing result includes the machine learning model after continued training and the model parameter values; step S26 may specifically include: S2611, continuing to perform model training according to the received measurement data and machine learning model information to obtain the machine learning model after continued training and the model parameter values; S2612, sending a second model deployment request message to the client device for generating a third network optimization operation instruction on the client device; where the second model deployment request message includes: a second model derivation configuration file, the machine learning model after continued training and the model parameter values, a second model performance reporting indication, and a second model performance reporting method.

[0082] In this embodiment, the AI server continues to perform model training based on the model and parameters trained by the AI distributed client to obtain the machine learning model after continued training and the model parameter values, and sends the machine learning model after continued training and the model parameter values to the AI distributed client, so that the AI distributed client performs model deployment and model instructions based on the machine learning model after continued training and the model parameter values to obtain a network optimization instruction.

[0083] In one embodiment, after S2612, the network optimization method further includes: S2613, in response to receiving a second model deployment response message from the client device, determining that the deployment of the machine learning model after continued training on the client device is successful according to the second model deployment success flag carried in the second model deployment response message; or, S2614, determining that the deployment of the machine learning model after continued training on the client device fails according to the second model deployment failure flag carried in the second model deployment response message.

[0084] In this embodiment, the second model deployment response message optionally includes an ML session ID, which is used to uniquely identify the ID of a certain process.

[0085] In one embodiment, after S2612, the network optimization method further includes: S2615, in response to receiving a second model performance report message from a network-side device, determining whether to re-perform model training processing according to the model performance metric values carried in the second model performance report message.

[0086] In this embodiment, the second model performance report message optionally includes an ML session ID (ML SessionID) for uniquely identifying the ID of a certain process.

[0087] In one embodiment, the current server communicates with a specified network-side device and a client device located within the specified network-side device through a predetermined interface; before step S110, the network optimization method may further include: S31, in response to receiving a control plane interface establishment request message, establishing a control plane interface between the current server and the specified network-side device as the predetermined interface; S32, sending a control plane interface establishment response message to the specified network-side device to indicate that the predetermined interface has been successfully established.

[0088] In one embodiment, if the control plane interface establishment request message includes data plane channel information of the specified network-side device, when performing step S32 of sending a control plane interface establishment response message to the specified network-side device, the data plane channel address of the current server is carried in the control plane interface establishment response message.

[0089] In this embodiment, the AI centralized server communicates with the network-side device by establishing a control plane interface. Between the network-side device where the client is located and other network-side devices, communication can be performed through the existing communication transmission interfaces between the network-side devices.

[0090] According to the network optimization method described in the above embodiments, the AI centralized server can determine whether to perform collaborative training with the client device according to the pre-acquired machine learning description information, and perform different model training processes according to the determination result, so as to implement the deployment of the machine learning training process through the server and the client device, perform distributed model training and processing, obtain network optimization operations, and provide a new optimization method and network intelligent optimization process for operator network optimization.

[0091] Figure 2 The flowchart showing the network optimization method according to another embodiment of the present application. As Figure 2 shown, the network optimization method in the embodiments of the present application may include the following steps.

[0092] S210. In response to receiving a session establishment request message from a predetermined server, perform measurement configuration on a specified network-side device and a terminal device connected to the specified network-side device according to the measurement control information included in the session establishment request message.

[0093] S220. Send the measurement report message of the specified network-side device and the measurement report message of the terminal device to the predetermined server. The measurement reports of the specified network-side device and the terminal device are used in the predetermined server for model training processing for network optimization.

[0094] S230. In response to receiving the model training processing result from the predetermined server, process to obtain a network optimization operation according to the model training processing result.

[0095] In this embodiment, a client device, such as an AI distributed client, in response to a session establishment request message from a server, performs measurement configuration on the network-side device and the terminal-side device, and in response to the model training processing result from the predetermined server, processes to obtain a network optimization operation. Based on the model training processing result transmitted between the client device and the server, a distributed model training process is implemented with the server, and finally a network optimization operation is obtained, enabling the data collected by the network-side device and the terminal device to be deeply analyzed in machine learning, and providing a new optimization method and a network intelligent optimization process for operator network optimization through distributed model training.

[0096] In one embodiment, step S210 may include: S211. In response to the session establishment request message, perform measurement configuration on the network-side device where the client device is located according to the network-side measurement control information; S212. Send a measurement configuration request to the network-side device where the client device is not deployed, to request the network-side device where the client device is not deployed to perform measurement configuration according to the network-side measurement control information.

[0097] In one embodiment, the session establishment request message further includes one or more machine learning models that the client device is required to support; after S210, the network optimization method further includes: if the measurement configuration of the network-side device and the terminal device is successful, and the client device supports the machine learning model, send a session establishment response message carrying a configuration success flag to the predetermined server; if the measurement configuration of the network-side device fails or the measurement configuration of the terminal device fails, or the client device does not support the machine learning model, send a session establishment response message carrying a configuration failure flag to the predetermined server.

[0098] In this embodiment, if the measurement configuration is successful and supports the machine learning model, a measurement configuration success message is fed back to the AI central server. If the measurement configuration fails or does not support the machine learning model, a measurement configuration failure message is fed back to the AI central server. The measurement configuration failure includes the measurement configuration failure of the network-side device or the measurement failure of the terminal device. Among them, the AI session ID is optionally carried in the session establishment response message.

[0099] In one embodiment, the session establishment request message further includes a computing processing capability information item; when performing step S220, the computing processing capability of the client device is carried in the measurement report of the network-side device where the client is located.

[0100] In this embodiment, the AI distributed client feeds back the AI computing processing capability of the client device to the AI central server through the session establishment request message. Among them, the AI session ID is optionally carried in the session establishment request message.

[0101] In one embodiment, before step S230, the network optimization method further includes: in response to the received second activation message, activating the machine learning function and obtaining the machine learning description information and the policy information that the network needs to meet carried in the second activation message.

[0102] In this embodiment, the network management OAM or the core network sends an activation message (MLActivation Message) through the interface with the base station to instruct the AI distributed client to activate (use) the ML function. The message includes an indication of whether the ML function is activated; an ML description message (ML Description Message); the message includes policy information (Policy Information) indicating that the RAN network-side device needs to meet. The ML description message in the AI distributed client is basically the same as the ML description message in the AI central server in the above embodiment. The ML description message in the AI distributed client can be used for the AI distributed client to select an appropriate ML type and ML model when performing model training.

[0103] In one embodiment, the policy information that the network needs to satisfy includes one or more of the following object identification information: one or more terminal device identities (User Equipment Identity, UE ID), one or more Quality of Service (QoS) flow identities (QoS FLOW Identity, QFI), one or more cell identities, one or more network slice identities, one or more public land mobile network identities (Public Land Mobile Network, PLMN Identity), one or more private network identities, one or more base station identities, and one or more tracking area identities (Tracking Area Code Identity, TAC ID).

[0104] In this embodiment, if the entity identified by the object identification information is one or more terminal devices, the designated network-side device that needs to perform measurement can be the network-side device to which the identified terminal devices are connected; if the entity identified by the object identification information is at least one of one or more QoS flows, one or more cells, one or more network slices, one or more public networks, and one or more private networks, the network-side device that needs to perform measurement can be the network-side device involved within the communication range of the identified entity.

[0105] In this embodiment, the network-side device that needs to perform measurement can also be flexibly selected according to the requirements of actual machine learning, and the embodiments of the present application do not make specific limitations.

[0106] In one embodiment, the communication quality metric information can be used to indicate the communication quality that the entity identified by the corresponding object identification information needs to achieve. Exemplarily, the communication quality metric information can include one or more of the following metric items: network energy saving metric, network key performance indicator (KPI), service quality of service (QoS) metric, quality of experience (QOE) metric, key quality indicator (KQI) metric for service perception, and traffic steering preference metric.

[0107] Exemplarily, network energy-saving metrics may include, for example, one or more of: energy-saving efficiency, percentage of energy consumption saved metric, and energy consumption saved value. As an example, network KPIs may include, for example, one or more of: handover success rate, call drop rate, access success rate, user throughput rate, cell throughput rate, cell load, network load, radio resource utilization rate, and network coverage rate. As an example, service quality of service metrics may include, for example, one or more of: service guaranteed rate, service maximum / minimum rate, service delay, service priority, delay jitter, and packet loss rate. As an example, user experience quality metrics may include, for example, one or more of: Mean Opinion Score (MOS) for measuring the voice quality of a communication system, streaming media open cache time, streaming media re-cache time, and streaming media re-cache count.

[0108] In one embodiment, the model training processing result is a trained machine learning model; step S230 may specifically include: S231, in response to receiving a first model deployment request message from a predetermined server, deploying and executing the trained machine learning model according to the model derivation configuration file, the trained machine learning model, and the model parameter values in the first model deployment request message to obtain a first network optimization operation instruction.

[0109] In this embodiment, the AI distributed client only performs ML model derivation and execution, and deploys and executes the trained machine learning model according to the trained model and parameters received from the AI central server to obtain a network optimization operation instruction.

[0110] In one embodiment, after S231, the network optimization method further includes: S232, if the deployment of the trained machine learning model is successful, sending a first model deployment response message to the predetermined server, where the first model deployment response message carries a first model deployment success flag; S233, if the deployment of the trained machine learning model fails, sending a first model deployment response message to the predetermined server, where the first model deployment response message carries a first model deployment failure flag.

[0111] In this embodiment, when the AI distributed client deploys according to the trained model and parameters received from the AI central server, it can feedback the model deployment result to the AI central server.

[0112] In one embodiment, the model deployment request message further includes a first model performance reporting indication and a first model performance reporting method. After step S230, the network optimization method further includes: S234, executing a first network optimization operation instruction on the network-side device where the client device is located; S235, generating a first network optimization operation request to request the network-side device that has not deployed the client device to specify the first network optimization operation instruction; S236, sending a first model performance report message to a predetermined server according to the first model performance reporting method, where the first model performance report message carries the corresponding model performance metric values.

[0113] In this embodiment, the first network optimization operation request optionally carries a session ID, which is used to uniquely identify the ID of a certain process. The AI distributed client can execute the first network optimization operation instruction on the network-side device where it is located and report the model performance metrics to the AI central server.

[0114] In one embodiment, step S230 may specifically include the following steps.

[0115] S41, in response to receiving a model configuration establishment message from a predetermined server, perform model configuration and hyperparameter configuration according to the machine learning algorithm and training hyperparameter values in the model configuration establishment message; S42, if the model configuration and hyperparameter configuration are successful, perform machine model training according to the configured model and hyperparameters to obtain the information of the trained machine learning model; S43, send a distributed training request message to the predetermined server, where the distributed training request message includes the information of the trained machine learning model; S44, in response to receiving the model training processing result for the information of the trained machine learning model from the predetermined server, obtain the network optimization operation according to the model training processing result.

[0116] In this embodiment, the AI distributed client and the AI central server perform collaborative training, and can obtain the network optimization operation according to the model training processing result in response to receiving the model training processing result for the information of the trained machine learning model from the predetermined server.

[0117] In one embodiment, the model training processing result includes a second network optimization operation instruction; step S44 may specifically include: in response to receiving a second network optimization operation request from the predetermined server, obtaining the second network optimization operation instruction carried in the second network optimization operation request.

[0118] In this embodiment, a method for distributed training between an AI distributed client and an AI central server is provided. That is, after the AI distributed client sends the information of the trained machine learning model to the AI central server, the AI central server continues to train the information of the trained machine learning model reported by each AI distributed client to obtain network optimization operations.

[0119] In one embodiment, the second network optimization operation request further includes a second model performance reporting indication and a second model performance reporting method. After step S44, the network optimization method further includes: S441, executing the second network optimization operation instruction on the network side device where the client device is located; S442, generating a second network operation request message to request the network side device that does not deploy the client device to specify the second network optimization operation instruction; S443, sending a second model performance report message to a predetermined server according to the second model performance reporting method, where the second model performance report message carries the second model performance index value.

[0120] In this embodiment, the network side device where the AI distributed client is located needs to report the model performance index for the second network optimization operation instruction.

[0121] In one embodiment, the model training processing result includes the machine learning model after continued training and the model parameter values; step S44 may specifically include: S444, in response to receiving the second model deployment request message, deploying and executing the machine learning model after continued training according to the model derivation configuration file, the machine learning model after continued training, and the model parameter values in the second model deployment request message to obtain a third network optimization operation instruction.

[0122] In this embodiment, another method for distributed training between an AI distributed client and an AI central server is provided. That is, after the AI distributed client sends the information of the trained machine learning model to the AI central server, the AI central server continues to train the information of the trained machine learning model reported by each AI distributed client to obtain the information of the machine learning model after continued training, and the AI distributed client performs model training processing according to the information of the machine learning model after continued training to obtain network optimization operations.

[0123] In one embodiment, after S230, the network optimization method further includes: if the deployment of the machine learning model after continued training is successful, sending a second model deployment response message to a predetermined server, where the second model deployment response message carries a second model deployment success flag; if the deployment of the machine learning model after continued training fails, sending a second model deployment response message to a predetermined server, where the second model deployment response message carries a second model deployment failure flag.

[0124] In this embodiment, the second model deployment response message is used to feedback the model deployment result and optionally carry a session ID, which is used to uniquely identify a machine learning session process. If the ML session ID in the session request message is not included, it indicates that the measurement control in the request message is for all ML processes.

[0125] In one embodiment, the second model deployment request message further includes a second model performance reporting indication and a second model performance reporting method. After S444, the network optimization method further includes: S445, performing a third network optimization operation instruction on the network side device where the client device is located; S446, generating a third network optimization operation request to request the network side device that has not deployed the client device to perform the third network optimization operation instruction; S447, sending a second model performance report message to a predetermined server according to the second model performance reporting method, and the second model performance report message carries the corresponding model performance metric values.

[0126] In this embodiment, the third network optimization operation instruction optionally carries a session ID. The AI distributed client can feedback the corresponding model performance report message to the AI centralized server when performing the network optimization operation instruction on the network side device where the client device is located.

[0127] According to the network optimization method of the embodiments of the present application, the AI distributed client responds to the session establishment request message from the server, performs measurement configuration on the network side device and the terminal side device, and responds to the model training processing result from the predetermined server, processes to obtain a network optimization operation, and realizes distributed model training processing with the server based on the model training processing result transmitted between the client and the server, and finally obtains a network optimization operation, so that the data collected by the network side device and the terminal device can be deeply analyzed in machine learning, and provides a new optimization method and network intelligent optimization process for operator network optimization through distributed model training.

[0128] Figure 3 The flowchart of the network optimization method showing another embodiment is as follows Figure 3 As shown, in one embodiment, the network optimization method includes the following steps.

[0129] S310, in response to receiving a measurement configuration request from a client device, perform measurement configuration according to the network side measurement control information in the measurement configuration request and perform measurement configuration on the terminal device connected to the network side device.

[0130] S320. Send the measurement report of the current network-side device obtained through measurement and the measurement report of the received terminal device to a predetermined server and a predetermined client device respectively. The measurement reports of the current network-side device and the terminal device are used in the predetermined server and the predetermined client device for model training processing for network optimization.

[0131] In this embodiment, the RAN network-side device may perform measurement configuration according to the received network-side measurement control information, perform measurement configuration on the connected terminal device, and send the measurement report obtained by performing the measurement and the measurement report of the received terminal-side device to the predetermined server and the client device. The measurement reports of the network-side device and the terminal device are used in the predetermined server and the predetermined client device for model training processing for network optimization, so that the data collected by the network-side device and the terminal device can be deeply analyzed in artificial intelligence and machine learning, and distributed model training processing is performed in the predetermined server and the predetermined client device, providing a new optimization method and a network intelligent optimization process for operator network optimization.

[0132] In one embodiment, S310 may specifically include: S311. In response to receiving a measurement configuration request, perform measurement configuration according to the network-side measurement control information; S312. According to the network-side measurement control information, determine the measurement quantity and measurement reporting method that need to be configured for the terminal device connected to the current network-side device as the terminal-side measurement control information; S313. Send a radio resource control message to the terminal device to instruct the terminal device to perform measurement configuration according to the terminal-side measurement control information.

[0133] In this embodiment, the RAN network-side device may perform measurement configuration according to the network-side measurement control information and perform measurement configuration on the connected terminal device.

[0134] In one embodiment, after step S320, the network optimization method further includes: S330. Receive a network optimization operation request from the predetermined server or the client device, obtain and execute the network optimization operation instruction in the received network optimization operation request, and send a corresponding model performance report message to the predetermined server.

[0135] In this embodiment, when the RAN network-side device executes the network optimization operation instruction, it needs to send a corresponding model performance report message to the predetermined server.

[0136] In one embodiment, the network-side device communicates with a predetermined server through a predetermined interface. Before step S310, the network optimization method further includes: S51. According to the address of the predetermined server obtained in advance, send a control-plane interface establishment request message to the predetermined server to request the predetermined server to establish a control-plane interface between the current network-side device and the predetermined server as the predetermined interface.

[0137] In one embodiment, the control-plane interface establishment request message includes one or more of the following information items: measurements supported by the current network-side device, reporting methods supported by the current network-side device, network optimization operations supported by the current network-side device, the data-plane channel address of the current network-side device, computing capabilities supported by the deployed client device, and machine learning models supported by the deployed client device.

[0138] In one embodiment, the network optimization method further includes: S340. In response to receiving the control-plane interface establishment response message, determine that the control-plane interface between the local network-side device and the predetermined server is successfully established; if the control-plane interface establishment request message includes the data-plane channel address of the current network-side device, the received control-plane interface establishment response message includes the data-plane channel address of the predetermined server.

[0139] In this embodiment, through the established communication interface between the RAN node device and the AI central server, through the established communication interface, the received network-side measurement control information is measured and configured, and the connected terminal device is measured and configured, and the measurement reports obtained by performing measurements and the measurement reports of the received terminal-side devices are sent to the predetermined server and the client device. The measurement reports of the network-side device and the terminal device are used in the predetermined server and the predetermined client device for model training processing for network optimization, so that the data collected by the network-side device and the terminal device can be deeply analyzed in artificial intelligence and machine learning, and distributed model training processing is performed in the predetermined server and the predetermined client device, providing new optimization methods and network intelligent optimization processes for operator network optimization.

[0140] To better understand the present application, the following Figure 4 and Figure 5 are used to describe in detail the network optimization method of the embodiments of the present application. Figure 4 FIG. shows the timing process diagram of the network optimization method of one embodiment. Figure 5 FIG. shows the timing process diagram of the network optimization method of another embodiment.

[0141] As Figure 4 shown, in one embodiment, the network optimization method may include the following steps.

[0142] S401-1: The network management or the core network activates the ML function of the AI centralized server by sending an activation message.

[0143] S401-2: The network management or the core network activates the ML function of the AI distributed client by sending an activation message.

[0144] S401: The AI centralized server sends an ML session establishment request message to the AI distributed clients within one or more RAN node devices to configure the radio-side measurement data required for a certain ML session.

[0145] Among them, the ML session establishment request message includes measurement control information, the computing processing capability information item of the client device, and one or more machine learning models that require the client device to support. The session establishment request message may also contain an ML session ID to uniquely identify a certain machine learning process.

[0146] S403: The RAN node device configures what measurements need to be made on the RAN side and the reporting method according to the measurement control information in the received message, and configures what measurements need to be made on the connected terminal device and the reporting method.

[0147] S404: If the measurements on the RAN node device side and the UE side are both configured successfully and can meet the ML model in the establishment request message, the RAN node device sends an ML session establishment response message to the AI centralized server.

[0148] Among them, the ML session establishment response message optionally carries the ML session ID.

[0149] S405: If the measurements on the RAN node device side and the UE side are both configured successfully, the RAN node device and the UE both perform relevant measurements according to the specified measurement configuration.

[0150] S406: The RAN node device and the terminal respectively send measurement report messages to the AI distributed client and the AI centralized server.

[0151] Among them, the measurement report message sent carries the measured value of the measured quantity, and optionally carries the ML session ID to which the measurement belongs, and the AI computing processing capability of the AI distributed client.

[0152] S407: The AI centralized server selects a suitable ML algorithm for model training and model iteration update according to the measurement data, the AI computing processing capability of the AI distributed client, and the supported ML models.

[0153] S408: The AI centralized server sends an ML model deployment request message to the AI distributed clients in one or more RAN node devices. Optionally, the ML session ID is carried in the ML model deployment request message.

[0154] S409: The RAN node device sends an ML model deployment response message.

[0155] Optionally, the ML session ID can be carried in the ML model deployment response message.

[0156] S410: The AI distributed client performs model processing such as ML model inference, model update, and model execution based on the configuration file in the ML model deployment message, the trained ML model, and the ML model parameter values, and obtains the RAN operations required for optimization.

[0157] S4011: The AI distributed client notifies the RAN node device where it is located to perform relevant optimization operations through a RAN operation request message.

[0158] S4012: The RAN node device performs the relevant optimization operations in the RAN operation request message. If the relevant optimization operations involve one or more UEs, the base station sends an RRC reconfiguration message or an RRC release message to the UEs connected to this base station to notify the UEs to perform the relevant operations.

[0159] S4013: The RAN node device sends an ML model performance report message to the AI centralized server, which carries the metrics of the ML model performance.

[0160] Optionally, the ML session ID to which the measurement belongs is carried in the ML model performance report message. The AI CentralizedServer decides whether to retrain and deploy the network based on the reported model performance report.

[0161] In the embodiment of this application, the AI distributed client only performs ML model derivation and execution, trains a model for network optimization using the measurement data in the received measurement report according to the information of the machine learning model trained in the AI centralized server, and obtains network optimization operations.

[0162] As Figure 5 shown, in Figure 5 , the processing procedures of steps S501-1 to S506 are basically the same as those of steps S401-1 to S406 in Figure 4 , and will not be elaborated in the embodiment of this application. Figure 5 The network optimization processing method of Figure 4 is different from the network optimization processing method in

[0163] S507: The AI central server selects a suitable ML algorithm and configures the training hyperparameter values based on the measurement data, the AI computing processing capabilities of the AI distributed clients, and the supported ML models.

[0164] S508: The AI central server sends an ML model configuration establishment message to the AI distributed clients within one or more RAN node devices for sending the selected ML model and the selected ML training hyperparameters.

[0165] Among them, the ML model configuration establishment message optionally includes an ML session ID, which is used to uniquely identify the ID of a certain process.

[0166] S509: The RAN node device sends an ML session establishment response message to the AI central server.

[0167] Among them, the ML session establishment response message is used to indicate whether the ML model and the hyperparameter configuration are successful. The ML session establishment response message optionally carries the ML session ID.

[0168] S510: If the ML model configuration of the AI distributed client of the RAN node device is successful, the AI distributed client performs ML model training / ML model iterative update according to the specified training model and training hyperparameters.

[0169] S511: The AI distributed clients within one or more RAN node devices send an ML distributed model training request to the AI central server.

[0170] Among them, the ML distributed model training request carries the information of the ML model that has been trained well in the AI distributed client.

[0171] S5012: If the configuration of the information of the ML model that has been trained well is successful, the AI central server sends an ML distributed training response message to the AI distributed clients within one or more RAN node devices.

[0172] Among them, the ML distributed training response message is used to indicate whether the configuration of the ML model that has been trained well is successful or failed. The ML distributed training response message optionally carries the ML session ID.

[0173] Next, continue to describe the processing flow of collaborative training between the AI distributed client and the AI central server in this network optimization method through a model training processing flow Option A and another model training processing flow Option B.

[0174] In Option A, the processing flow of collaborative training may include the following steps.

[0175] S50A-1: The AI central server continues with model processing steps such as model update, model training, model inference, and model execution based on the measurement data and the ML training parameters reported by the AI distributed clients, and obtains the RAN operations required for optimization.

[0176] S50A-2: The AI central server notifies the RAN node to perform relevant optimization operations through a RAN network optimization operation request message.

[0177] In Option B, the processing flow of collaborative training may include the following steps.

[0178] S50B-1: The AI central server continues with model training and model iterative update based on the measurement data and the information of the trained ML model reported by the AI distributed clients.

[0179] S50B-2: The AI central server sends an ML model deployment request message to the AI distributed client within the RAN node device, so that the AI distributed client can perform model configuration and execution.

[0180] S50B-3: The AI distributed client sends an ML model deployment response message.

[0181] Among them, the model deployment response message is used to indicate whether the ML model deployment configuration is successful or failed.

[0182] The above model deployment request message and ML model deployment response message may carry an ML session ID.

[0183] S50B-4: The AI distributed client performs ML model inference and model execution based on the configuration file in the ML model deployment message, the trained ML model, and the ML model parameter values, and obtains the RAN operations required for optimization.

[0184] S5011: The AI distributed client notifies the RAN node device where it is located to perform relevant optimization operations.

[0185] S5012: The RAN node device performs the relevant optimization operations in the RAN operation request message. If the relevant optimization operations involve one or more terminal devices, the RAN node device sends an RRC reconfiguration message or an RRC release message to the terminal devices connected to this RAN node device to notify the terminal devices to perform relevant operations.

[0186] S5013: The RAN node device sends an ML model performance report message to the AI central server.

[0187] Among them, the ML model performance report message can carry the ML session ID to which the measurement belongs. The AI central server decides whether to retrain and deploy the network according to the reported model performance report.

[0188] In the embodiments of the present application, the AI distributed client and the AI central server perform centralized training and provide the above two distributed model training methods of Option A and Option B. Between the AI distributed client and the AI central server, model training for network optimization is performed using the measurement data in the received measurement report to obtain network optimization operations.

[0189] Figure 6 The flow diagram shows the process of establishing a communication interface between a server and a network-side device in an embodiment. As Figure 6 shown, in an embodiment, the interface establishment process may include the following steps.

[0190] S601, the network management configures the address of the AI central server for the base station through a configuration message or directly.

[0191] S602, the base station sends a communication interface establishment request according to the configured address of the AI central server.

[0192] In this step, if the interface between the AI central server and the RAN node is called the I1 interface, the RAN node sends an I1 interface establishment request message to the address of the AI central server to establish the I1 interface.

[0193] In an embodiment, the communication interface establishment request may include: measurements supported by the network-side device (such as a base station); the reporting method of the measurements supported by the network-side device; RAN optimization operations supported by the network-side device; the AI computing capabilities supported by the AI distributed client; the ML models supported by the AI distributed client.

[0194] In an embodiment, the communication interface establishment request may further include: the data plane channel address of the network-side device.

[0195] S603, the AI central server sends an interface establishment response message to the base station to indicate whether the interface is successfully established.

[0196] In this step, the interface establishment response message may be an I1 interface establishment response message to indicate whether the interface is successfully established. If the message received in step S602 carries the data plane channel address of the base station side, the data plane channel address of the AI central server is carried in the response message.

[0197] Through the above steps S501 to S503, a control plane interface for the RAN node can be established on the AI centralized server. The control plane interface can be used to transmit control messages and can also be used to transmit data required for ML, such as measurement data.

[0198] S604. Establish a data plane channel according to the data plane channel address configured by the AI centralized server and the data plane channel address on the base station side.

[0199] In the implementation of this application, the network-side device and the AI centralized server can transmit data required for machine learning on the data plane channel. For example, some data with a large data volume can be transmitted on the data plane channel, such as measurement data from the base station, parameters trained by the AI centralized server such as weights and biases, to relieve the data transmission pressure on the communication channel corresponding to the control plane interface and improve the data transmission efficiency. The data plane channel established in this embodiment is not limited to transmitting data of a specified user.

[0200] Figure 7 The flowchart shows the process of establishing a data channel related to a predetermined server and a user equipment in an embodiment. In one embodiment, the interface establishment process may include the following steps.

[0201] S701. Obtain the I1 control plane interface established between the AI centralized server and the base station.

[0202] Among them, the AI distributed client is deployed in the base station.

[0203] S702. The AI centralized server sends a channel establishment request message to the base station.

[0204] As an example, the channel establishment request message can be, for example, a UE AIContext Setup Request message, and carry the data plane channel address on the AI server side and the UE ID.

[0205] S703. The base station sends a response message to the AI centralized server, carrying the data plane channel address on the base station side.

[0206] As an example, the channel establishment response message can be, for example, a UE AIContext Setup Response message, and carry the data plane channel address on the base station side.

[0207] S704. Establish a data plane channel specified by the UEID and related to a specific user according to the data plane channel address configured by the AI server and the data plane channel address on the base station side.

[0208] As an example, the base station and the AI server can transmit ML - required data related to a specific user on the data plane channel. Generally, some data with a relatively large volume can be transmitted on the data plane channel, such as the measurement data of the user, the ML model trained based on the user data, and the parameters.

[0209] In the embodiments of the present application, the RAN node device and the AI server can transmit machine - learning - required data on the data plane channel. For example, some data with a relatively large volume can be transmitted on the data plane channel, such as the measurement data from the base station, so as to relieve the data transmission pressure on the communication channel corresponding to the control plane interface and improve the data transmission efficiency. The data plane channel established in this embodiment is not limited to transmitting data of a specified user.

[0210] Next, in conjunction with the accompanying drawings, a server according to the embodiments of the present application will be introduced in detail. Figure 8 The structural schematic diagram of the server provided by an embodiment of the present application is shown. In one embodiment, the server is an AI centralized server. As Figure 8 shown, the server may include the following modules.

[0211] A measurement configuration request module 810, configured to send a session establishment request message to a client device to request the client device to perform measurement configuration on a specified network - side device and perform measurement configuration on a terminal device connected to the specified network - side device.

[0212] A measurement report receiving module 820, configured to receive the measurement report messages of the specified network - side device and the terminal device.

[0213] A collaborative training determination module 830, configured to determine whether client device collaborative training is required according to the pre - obtained machine - learning description information.

[0214] A model training processing module 840, configured to perform specified model training processing for network optimization based on whether client device collaborative training is required and the measurement data in the received measurement report messages, and send the model training processing result to the client device to instruct the client device to obtain network optimization operations according to the model training processing result.

[0215] In one embodiment, the server further includes: a first activation module, configured to activate the machine - learning function in response to the received first activation message, and obtain the machine - learning description information carried in the activation message and the training hyperparameters of the corresponding machine - learning model.

[0216] In one embodiment, the session establishment request message includes measurement control information for the client device, an item of computing processing capability information of the client device, and one or more machine learning models that the client device is required to support, where the measurement configuration information is used to indicate the measurement quantities and measurement reporting methods that need to be configured by the specified network-side device.

[0217] In one embodiment, the server further includes: a session establishment response module, configured to, in response to the received session establishment response message, determine, according to the success flag carried in the session establishment response message, whether the client can meet the requirement information for the computing processing capability of the client device.

[0218] In one embodiment, the measurement report receiving module 820 is specifically configured to receive the measurement report of the specified network-side device and the measurement report of the terminal device sent by the specified network-side device.

[0219] In one embodiment, the machine learning description information includes one or more selected machine learning types and one or more selected machine learning models; the co-training determination module 830 includes: a training mode determination unit, configured to obtain a pre-set training mode corresponding to the machine learning type and the machine learning model according to the selected machine learning model of the selected machine learning type; and a co-training unit, configured to determine whether co-training of the client device is required according to the pre-set training mode.

[0220] In one embodiment, when it is determined that co-training of the client device is not required, the model training processing result includes the trained machine learning model and the model parameter values; the model training processing module 840 may include: a computing processing capability acquisition unit, configured to acquire the computing processing capability of the client device from the received measurement report of the specified network-side device, where the client device is deployed in the specified network-side device; a model training unit, configured to select a machine learning algorithm for model training according to the received measurement data, the computing processing capability of the client device, and the pre-acquired machine learning models supported by the client device, to obtain the trained machine learning model and the model parameter values; and a model deployment request sending unit, configured to send a first model deployment request message to the client device for generating a first network optimization operation instruction on the client device; where the first model deployment request message includes: a first model derivation configuration file, the trained machine learning model and the model parameter values, a first model performance reporting indication, and a first model performance reporting method.

[0221] In one embodiment, the server further includes: a model deployment response module, configured to, in response to receiving a first model deployment response message from a client device, determine that the deployment of the trained machine learning model on the client device is successful according to the first model deployment success flag in the first model deployment response message; or determine that the deployment of the trained machine learning model on the client device fails according to the first model deployment failure flag in the first model deployment response message.

[0222] In one embodiment, the server further includes: a model performance determination module, configured to, in response to receiving a first model performance report message from a network-side device, determine whether model training processing needs to be re-performed according to the model performance metric values carried in the first model performance report message.

[0223] In one embodiment, when it is determined that the client device needs to perform collaborative training, the model training processing module 840 includes: a computing processing capability acquisition unit, configured to acquire the computing processing capability reported by the client device from the measurement report of the specified network-side device received, where the client device is deployed on the specified network-side device; a model and parameter selection unit, configured to select a machine learning algorithm and configure training hyperparameter values according to the received measurement data, the pre-acquired computing processing capability of the client device, and the machine learning models supported by the client device; a model configuration establishment message sending unit, configured to send a model configuration establishment message to the client device to request the client device to perform model configuration according to the machine learning algorithm and the training hyperparameter values; a distributed training request receiving module, configured to receive the model configuration establishment message from the client device, where the model configuration establishment message includes the information of the trained machine learning model obtained by the client device through model training for the model configuration of the machine learning algorithm and the training hyperparameter values; a model performance determination module, further configured to perform model configuration and model training processing based on the information of the trained machine learning model to obtain a model training processing result, and send the model training processing result to the client device.

[0224] In one embodiment, the model training processing result includes a second network optimization operation instruction; the model performance determination module further includes: an operation instruction generation unit, configured to continue model training according to the received measurement data and the information of the trained machine learning model to obtain a second network optimization operation instruction; an operation request sending module, configured to send a second network operation request message to the network-side device, where the second network operation request message includes the second network optimization operation instruction, a second model performance reporting indication, and a second model performance reporting method.

[0225] In one embodiment, the model performance determination module is further configured to, in response to receiving a second model performance report message from a network-side device, determine whether to re-perform model training processing according to the model performance metric values carried in the second model performance report message.

[0226] In one embodiment, the model training processing result includes the machine learning model after continued training and the model parameter values; performing model configuration and model training processing based on the machine learning model information, obtaining the model training processing result, and sending the model training processing result to the client device, including:

[0227] Continuing to perform model training according to the received measurement data and the machine learning model information, obtaining the machine learning model after continued training and the model parameter values;

[0228] Sending a second model deployment request message to the client device for generating a third network optimization operation instruction at the client device; wherein, the second model deployment request message includes: a second model derivation configuration file, the machine learning model after continued training and the model parameter values, a second model performance reporting indication, and a second model performance reporting method.

[0229] In one embodiment, the server further includes: a model deployment response module, further configured to, in response to receiving a second model deployment response message from the client device, determine that the deployment of the machine learning model after continued training at the client device is successful according to the second model deployment success flag carried in the second model deployment response message; or determine that the deployment of the machine learning model after continued training at the client device fails according to the second model deployment failure flag carried in the second model deployment response message.

[0230] In one embodiment, the server further includes: a model performance determination module, further configured to, in response to receiving a second model performance report message from a network-side device, determine whether to re-perform model training processing according to the model performance metric values carried in the second model performance report message.

[0231] In one embodiment, the current server communicates with a specified network-side device and a client device located within the specified network-side device through a predetermined interface; the server further includes: a predetermined interface establishment module, configured to, in response to receiving a control plane interface establishment request message, establish a control plane interface between the current server and the specified network-side device as the predetermined interface; sending a control plane interface establishment response message to the specified network-side device to indicate that the establishment of the predetermined interface is successful.

[0232] In one embodiment, if the control plane interface establishment request message includes the data plane channel information of a specified network-side device, when sending a control plane interface establishment response message to the specified network-side device, the data plane channel address of the current server is carried in the control plane interface establishment response message.

[0233] According to the server of the embodiments of the present application, by requesting the client device to perform measurement configuration on the network-side device and the terminal-side device, and determining whether to perform collaborative training with the client device according to the pre-acquired machine learning description information, different model training processes are performed according to the determination result, so as to implement the deployment of the machine learning training process through the server and the client device, to perform distributed model training and processing, obtain network optimization operations, and thus perform in-depth analysis on the collected data through artificial intelligence and machine learning in the network system, providing a new optimization method and network intelligent optimization process for operator network optimization.

[0234] Figure 9 The structural schematic diagram of the client device provided by an embodiment of the application is shown. In one embodiment, the client device is an AI distributed client and is deployed in a network-side node device such as a base station. As Figure 9 shown, the client device may include the following modules.

[0235] A measurement configuration module 910, configured to, in response to receiving a session establishment request message from a predetermined server, perform measurement configuration on a specified network-side device and a terminal device connected to the specified network-side device according to the measurement control information included in the session establishment request message.

[0236] A measurement report sending module 920, configured to send a measurement report message of the specified network-side device and a measurement report message of the terminal device to the predetermined server, and the measurement report of the specified network-side device and the measurement report of the terminal device are used in the predetermined server for model training processing for network optimization;

[0237] An optimization operation determination module 930, configured to, in response to receiving a model training processing result from a predetermined server, process the model training processing result to obtain a network optimization operation.

[0238] In one embodiment, the measurement configuration module 910 may specifically include, in response to the session establishment request message, performing measurement configuration on the network-side device where the client device is located according to the network-side measurement control information; sending a measurement configuration request to the network-side device where the client device is not deployed, so as to request the network-side device where the client device is not deployed to perform measurement configuration according to the network-side measurement control information.

[0239] In one embodiment, the session establishment request message further includes one or more machine learning models that the client device is required to support; the client device further includes: a session establishment response module, configured to, if the measurement configuration of the network-side device and the terminal device is successful and this client device supports the machine learning model, send a session establishment response message carrying a configuration success flag to a predetermined server; if the measurement configuration of the network-side device fails or the measurement configuration of the terminal device fails, or this client device does not support the machine learning model, send a session establishment response message carrying a configuration failure flag to a predetermined server.

[0240] In one embodiment, the session establishment request message further includes a computing processing capability information item; the client device further includes: a measurement report sending module 920, further configured to, when sending the measurement report messages of the specified network-side device and the terminal device to a predetermined server, carry the computing processing capability of this client device in the measurement report of the network-side device where this client is located.

[0241] In one embodiment, the client device further includes: an activation module, configured to, in response to a received second activation message, activate the machine learning function and obtain the machine learning description information and the policy information that the network needs to satisfy carried in the second activation message.

[0242] In one embodiment, the model training processing result is a trained machine learning model; the optimization operation determination module 930 includes: a model deployment module, configured to, in response to a received first model deployment request message from a predetermined server, deploy and execute the trained machine learning model according to the model derivation configuration file, the trained machine learning model, and the model parameter values in the first model deployment request message, and obtain a first network optimization operation instruction.

[0243] In one embodiment, the client device further includes: a model deployment response module, configured to, if the deployment of the trained machine learning model is successful, send a first model deployment response message to a predetermined server, and the first model deployment response message carries a first model deployment success flag; if the deployment of the trained machine learning model fails, send a first model deployment response message to a predetermined server, and the first model deployment response message carries a first model deployment failure flag.

[0244] In one embodiment, the model deployment request message further includes a first model performance reporting indication and a first model performance reporting method. The client device further includes: an operation instruction execution module, configured to execute a first network optimization operation instruction on the network-side device where the present client device is located; an operation request module, configured to generate a first network optimization operation request to request the network-side device where the present client device is not deployed to specify the first network optimization operation instruction; a performance reporting module, configured to send a first model performance report message to a predetermined server according to the first model performance reporting method, where the first model performance report message carries the corresponding model performance metric value.

[0245] In one embodiment, the optimization operation determination module 930 includes: a model deployment module, configured to, in response to receiving a model configuration establishment message from a predetermined server, perform model configuration and hyperparameter configuration according to the machine learning algorithm and training hyperparameter values in the model configuration establishment message; a model training module, configured to, if the model configuration and hyperparameter configuration are successful, perform machine model training according to the configured model and hyperparameters to obtain information on the trained machine learning model; a distributed training request module, configured to send a distributed training request message to the predetermined server, where the distributed training request message includes information on the trained machine learning model; the optimization operation determination module 930 is further configured to, in response to receiving a model training processing result for the information on the trained machine learning model from the predetermined server, obtain a network optimization operation according to the model training processing result.

[0246] In one embodiment, the model training processing result includes a second network optimization operation instruction; the optimization operation determination module 930 is further configured to, in response to receiving a second network optimization operation request from the predetermined server, obtain the second network optimization operation instruction carried in the second network optimization operation request.

[0247] In one embodiment, the second network optimization operation request further includes a second model performance reporting indication and a second model performance reporting method. The operation instruction execution module is further configured to execute a second network optimization operation instruction on the network-side device where the present client device is located; the operation request module is further configured to generate a second network operation request message to request the network-side device where the present client device is not deployed to specify the second network optimization operation instruction; the performance reporting module is further configured to send a second model performance report message to the predetermined server according to the second model performance reporting method, where the second model performance report message carries the second model performance metric value.

[0248] In one embodiment, the model training processing result includes the machine learning model after continued training and the model parameter values; the optimization operation determination module 930 is further configured to, in response to receiving a second model deployment request message, deploy and execute the machine learning model after continued training according to the model derivation configuration file, the machine learning model after continued training, and the model parameter values in the second model deployment request message, so as to obtain a third network optimization operation instruction.

[0249] In one embodiment, the model deployment response module is further configured to, if the deployment of the machine learning model after continued training is successful, send a second model deployment response message to a predetermined server, where the second model deployment response message carries a second model deployment success flag; if the deployment of the machine learning model after continued training fails, send a second model deployment response message to a predetermined server, where the second model deployment response message carries a second model deployment failure flag.

[0250] In one embodiment, the second model deployment request message further includes a second model performance reporting indication and a second model performance reporting method. The operation instruction execution module is further configured to execute the third network optimization operation instruction on the network side device where the client device is located; the operation request module is further configured to generate a third network optimization operation request to request the network side device that has not deployed the client device to execute the third network optimization operation instruction; the operation request module is further configured to send a second model performance report message to a predetermined server according to the second model performance reporting method, where the second model performance report message carries the corresponding model performance metric values.

[0251] According to the client device of the embodiment of the present application, in response to a session establishment request message from a server, measurement configuration is performed on the network side device and the terminal side device, and in response to the model training processing result from a predetermined server, a network optimization operation is processed. Based on the model training processing result transmitted between the client device and the server, distributed model training processing is implemented with the server, and finally a network optimization operation is obtained, so that the data collected by the network side device and the terminal device can be deeply analyzed in machine learning, and new optimization methods and network intelligent optimization processes are provided for operator network optimization through distributed model training.

[0252] Figure 10 The structural schematic diagram of the network side device provided by an embodiment of the present application is shown. In one embodiment, the network side device is a RAN node device. As Figure 10 shown, the network side node device may include the following modules.

[0253] The measurement configuration module 1010 is configured to, in response to receiving a measurement configuration request from a client device, perform measurement configuration according to the network side measurement control information in the measurement configuration request and perform measurement configuration on the terminal device connected to the network side device.

[0254] A measurement report sending module 1020 is configured to send the measurement report of the current network-side device obtained through measurement and the measurement report of the received terminal device to a predetermined server and a predetermined client device respectively. The measurement report of the current network-side device and the measurement report of the terminal device are used for model training processing for network optimization in the predetermined server and the predetermined client device.

[0255] In one embodiment, the measurement configuration module 1010 is further configured to, in response to receiving a measurement configuration request, perform measurement configuration according to network-side measurement control information; determine the measurement quantity and measurement reporting method that need to be configured for the terminal device connected to the current network-side device according to the network-side measurement control information as terminal-side measurement control information; and send a radio resource control message to the terminal device to instruct the terminal device to perform measurement configuration according to the terminal-side measurement control information.

[0256] In one embodiment, the network-side device further includes: a performance report sending module, configured to receive a network optimization operation request from a predetermined server or a client device, obtain and execute the network optimization operation instruction in the received network optimization operation request, and send a corresponding model performance report message to the predetermined server.

[0257] In one embodiment, the designated network-side communicates with a predetermined server through a predetermined interface. The network-side device further includes: an interface establishment request module, configured to send a control plane interface establishment request message to the predetermined server according to the address of the predetermined server obtained in advance, to request the predetermined server to establish a control plane interface between the current network-side device and the predetermined server as the predetermined interface.

[0258] In one embodiment, the control plane interface establishment request message includes one or more of the following information items: the measurements supported by the current network-side device, the reporting methods supported by the current network-side device, the network optimization operations supported by the current network-side device, the data plane channel address of the current network-side device, the computing capabilities supported by the deployed client device, and the machine learning models supported by the deployed client device.

[0259] In one embodiment, the interface establishment request module is further configured to, in response to receiving a control plane interface establishment response message, determine that the control plane interface between the present network-side device and the predetermined server is successfully established; if the control plane interface establishment request message includes the data plane channel address of the current network-side device, the received control plane interface establishment response message includes the data plane channel address of the predetermined server.

[0260] The network-side device according to the embodiment of the present application can perform measurement configuration according to the received network-side measurement control information, perform measurement configuration on the connected terminal device, and send the measurement reports obtained by performing the measurements and the measurement reports of the received terminal-side device to a predetermined server and a client device. The measurement reports of the network-side device and the measurement reports of the terminal device are used in the predetermined server and the predetermined client device for model training processing for network optimization, so that the data collected by the network-side device and the terminal device can be deeply analyzed in artificial intelligence and machine learning, and distributed model training processing is performed in the predetermined server and the predetermined client device, providing new optimization methods and network intelligent optimization processes for operator network optimization.

[0261] Figure 11 The structural schematic diagram of the network optimization system provided by an embodiment of the present application is shown. In one embodiment, the network-side device is a RAN node device. As Figure 11 shown, the network optimization system may include a server 1110, one or more client devices 1120, and one or more network-side devices 1130. Among them, each client device is located in one of the one or more network-side devices;

[0262] The server 1110 is used to execute the network optimization method of the above embodiment; one or more client devices 1120 are used to execute the network optimization method of the above embodiment; one or more network-side devices 1130 are used to execute the network optimization method of the above embodiment.

[0263] In this embodiment, the server 1110 has the same or equivalent structure as the AI centralized server described in the above embodiment and can execute the network optimization method applied to the AI centralized server described in the above embodiment; the client device 1120 has the same or equivalent structure as the AI distributed client described in the above embodiment and can execute the network optimization method applied to the network-side device described in the above embodiment; the network-side device 1130 has the same or equivalent structure as the RAN node device described in the above embodiment and can execute the network optimization method applied to the RAN node device described in the above embodiment.

[0264] It should be clear that the present application is not limited to the specific configurations and processes described in the above embodiments and shown in the figures. For the convenience and conciseness of description, the detailed description of known methods is omitted here, and the specific working processes of the systems, modules, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be repeated here.

[0265] Figure 12 It is a structural diagram showing an exemplary hardware architecture of a computing device capable of implementing the embodiment of the present application.

[0266] As shown Figure 12 in FIG. 1, the computing device 1200 includes an input device 1201, an input interface 1202, a central processing unit 1203, a memory 1204, an output interface 1205, and an output device 1206. Among them, the input interface 1202, the central processing unit 1203, the memory 1204, and the output interface 1205 are interconnected through a bus 1210. The input device 1201 and the output device 1206 are respectively connected to the bus 1210 through the input interface 1202 and the output interface 1205, and then connected to other components of the computing device 1200.

[0267] Specifically, the input device 1201 receives input information from the outside and transmits the input information to the central processing unit 1203 through the input interface 1202; the central processing unit 1203 processes the input information based on the computer-executable instructions stored in the memory 1204 to generate output information, temporarily or permanently stores the output information in the memory 1204, and then transmits the output information to the output device 1206 through the output interface 1205; the output device 1206 outputs the output information to the outside of the computing device 1200 for user use.

[0268] In one embodiment, Figure 11 the computing device shown in FIG. 1 can be implemented as a server, which may include: a memory configured to store programs; a processor configured to run the programs stored in the memory to execute the network optimization method applied to the AI centralized server described in the above embodiments.

[0269] In one embodiment, Figure 11 the computing device shown in FIG. 1 can be implemented as a client device, which may include: a memory configured to store programs; a processor configured to run the programs stored in the memory to execute the network optimization method applied to the AI distributed client described in the above embodiments.

[0270] In one embodiment, Figure 11 the computing device shown in FIG. 1 can be implemented as a network-side device, which may include: a memory configured to store programs; a processor configured to run the programs stored in the memory to execute the network optimization method applied to the network-side device described in the above embodiments.

[0271] As described above, these are only exemplary embodiments of the present application and are not intended to limit the protection scope of the present application. Generally, various embodiments of the present application can be implemented in hardware or dedicated circuits, software, logic, or any combination thereof. For example, some aspects can be implemented in hardware, while other aspects can be implemented in firmware or software that can be executed by a controller, a microprocessor, or other computing devices, although the present application is not limited thereto.

[0272] Embodiments of the present application can be implemented by a data processor of a mobile device executing computer program instructions, for example, in a processor entity, or by hardware, or by a combination of software and hardware. The computer program instructions can be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state-setting data, or source code or object code written in any combination of one or more programming languages.

[0273] Any block diagram of a logical process in the drawings of the present application can represent program steps, or can represent interconnected logical circuits, modules, and functions, or can represent a combination of program steps and logical circuits, modules, and functions. The computer program can be stored in a memory. The memory can have any type suitable for the local technical environment and can be implemented using any suitable data storage technology, such as but not limited to read-only memory (ROM), random access memory (RAM), optical memory devices and systems (digital versatile disc DVD or CD disc), etc. The computer-readable medium can include non-transitory storage media. The data processor can be any type suitable for the local technical environment, such as but not limited to a general-purpose computer, a dedicated computer, a microprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), and a processor based on a multi-core processor architecture.

[0274] Through exemplary and non-limiting examples, a detailed description of the exemplary embodiments of the present application has been provided above. However, considering the accompanying drawings and the claims, various modifications and adjustments to the above embodiments will be obvious to those skilled in the art without departing from the scope of the present application. Therefore, the proper scope of the present application will be determined according to the claims.

Claims

1. A network optimization method, applied to a server, where the network is a wireless communication network and the server is an artificial intelligence (AI) server, characterized in that The method includes: Sending a session establishment request message to a client device to request the client device to perform measurement configuration on a specified network-side device and perform measurement configuration on a terminal device connected to the specified network-side device; the client device is an AI distributed client; Receiving a measurement report message of the specified network-side device and a measurement report message of the terminal device; Determining whether the client device needs to perform collaborative training according to the pre-acquired machine learning description information; Based on whether the client device needs to perform collaborative training and the measurement data in the received measurement report message, performing specified model training processing for network optimization, and sending the model training processing result to the client device for instructing the client device to obtain network optimization operations according to the model training processing result; When it is determined that the client device needs to perform collaborative training, the performing specified model training processing for network optimization based on whether the client device needs to perform collaborative training and the measurement data in the received measurement report message, and sending the model training processing result to the client device includes: Obtaining the computing processing capability reported by the client device from the received measurement report of the specified network-side device, where the client device is deployed on the specified network-side device; Selecting a machine learning algorithm and configuring training hyperparameter values according to the received measurement data, the pre-acquired computing processing capability of the client device, and the machine learning model supported by the client device; Sending a model configuration establishment message to the client device to request the client device to perform model configuration according to the machine learning algorithm and training hyperparameter values; Receiving a distributed training request message of the client device, where the distributed training request message includes information of the trained machine learning model obtained by the client device through model training for the model configuration according to the machine learning algorithm and training hyperparameter values; Performing model configuration and model training processing based on the information of the trained machine learning model to obtain a model training processing result, and sending the model training processing result to the client device.

2. The method according to claim 1, characterized in that, Before sending the session establishment request message to the client device, the method further includes: Responding to the received first activation message, activating the machine learning function, and obtaining the machine learning description information carried in the activation message and the training hyperparameters of the corresponding machine learning model.

3. The method according to claim 1, wherein The session establishment request message includes measurement control information for the client device, an information item of the computing processing capability of the client device, and one or more machine learning models that the client device is required to support, where the measurement control information is used to indicate the measurement quantities and measurement reporting methods that need to be configured for the specified network-side device.

4. The method according to claim 3, characterized in that After sending the session establishment request message to the client device, the method further includes: In response to the received session establishment response message, determine that the client can meet the requirement information for the computing and processing capabilities of the client device according to the success flag carried in the session establishment response message.

5. The method according to claim 1, wherein The receiving the measurement report messages of the specified network-side device and the measurement report messages of the terminal device includes: Receiving the measurement report of the specified network-side device and the measurement report of the terminal device sent by the specified network-side device.

6. The method according to claim 1, characterized in that, The machine learning description information includes one or more selected machine learning types and one or more selected machine learning models; the determining whether the client device needs to perform collaborative training according to the pre-acquired machine learning description information includes: According to the selected machine learning model of the selected machine learning type, obtain the pre-set training mode corresponding to the machine learning type and the machine learning model. According to the pre-set training mode, determine whether the client device needs to perform collaborative training.

7. The method according to claim 1, characterized in that, When it is determined that the client device does not need to perform collaborative training, the model training processing result includes the trained machine learning model and the model parameter values; the performing specified model training processing for network optimization based on whether the client device needs to perform collaborative training and the measurement data in the received measurement report message, and sending the model training processing result to the client device includes: Obtain the computing and processing capabilities of the client device from the received measurement report of the specified network-side device, where the client device is deployed in the specified network-side device. Select a machine learning algorithm for model training according to the received measurement data, the computing and processing capabilities of the client device, and the pre-acquired machine learning models supported by the client device, to obtain the trained machine learning model and the model parameter values. Send a first model deployment request message to the client device for generating a first network optimization operation instruction on the client device; where the first model deployment request message includes: a first model derivation configuration file, the trained machine learning model and the model parameter values, a first model performance reporting indication, and a first model performance reporting method.

8. The method according to claim 7, characterized in that After sending the first model deployment request message to the client device, the method further includes: In response to receiving the first model deployment response message of the client device, determine that the deployment of the trained machine learning model on the client device is successful according to the first model deployment success flag in the first model deployment response message; or, Determine that the deployment of the trained machine learning model on the client device fails according to the first model deployment failure flag in the first model deployment response message.

9. The method according to claim 7, wherein After sending the first model deployment request message to the client device, the method further includes: In response to receiving the first model performance report message of the network-side device, determine whether it is necessary to re-perform model training processing according to the model performance metric value carried in the first model performance report message.

10. The method according to claim 1, characterized in that, The model training processing result includes a second network optimization operation instruction; the performing model configuration and model training processing based on the trained machine learning model information, obtaining a model training processing result, and sending the model training processing result to the client device includes: Continuing to perform model training according to the received measurement data and the trained machine learning model information to obtain a second network optimization operation instruction; Sending a second network operation request message to the network side device, where the second network operation request message includes the second network optimization operation instruction, a second model performance reporting indication, and a second model performance reporting method.

11. The method according to claim 1, wherein After sending the second network operation request message to the network side device, the method further includes: Responding to receiving a second model performance report message from the network side device, and determining whether to re-perform model training processing according to the model performance metric value carried in the second model performance report message.

12. The method according to claim 1, wherein, The model training processing result includes a continuously trained machine learning model and model parameter values; the performing model configuration and model training processing based on the trained machine learning model information, obtaining a model training processing result, and sending the model training processing result to the client device includes: Continuing to perform model training according to the received measurement data and the machine learning model information to obtain a continuously trained machine learning model and model parameter values; Sending a second model deployment request message to the client device for generating a third network optimization operation instruction at the client device; wherein, the second model deployment request message includes: a second model derivation configuration file, the continuously trained machine learning model and model parameter values, a second model performance reporting indication, and a second model performance reporting method.

13. The method according to claim 12, wherein After sending the second model deployment request message to the client device, the method further includes: Responding to receiving a second model deployment response message from the client device, and determining that the deployment of the continuously trained machine learning model at the client device is successful according to the second model deployment success flag carried in the second model deployment response message; or, Determining that the deployment of the continuously trained machine learning model at the client device fails according to the second model deployment failure flag carried in the second model deployment response message.

14. The method according to claim 12, wherein After sending the second model deployment request message to the client device, the method further includes: Responding to receiving a second model performance report message from the network side device, and determining whether to re-perform model training processing according to the model performance metric value carried in the second model performance report message.

15. The method according to claim 1, wherein The current server communicates with the specified network side device and the client device located within the specified network side device through a predetermined interface; Before sending a session establishment request message to the client device, the method further includes: Responding to a received control plane interface establishment request message, and establishing a control plane interface between the current server and the specified network side device as the predetermined interface; Send a control plane interface establishment response message to the specified network-side device to indicate that the predetermined interface has been successfully established.

16. The method according to claim 15, wherein if the data plane channel information of the specified network-side device is included in the control plane interface establishment request message, when sending the control plane interface establishment response message to the specified network-side device, carry the data plane channel address of the current server in the control plane interface establishment response message.

17. A network optimization method is applied to a client device. The network is a wireless communication network, and the client device is an AI distributed client. It is characterized in that, The method includes: In response to receiving a session establishment request message from a predetermined server, perform measurement configuration on a specified network-side device and a terminal device connected to the specified network-side device according to the measurement control information included in the session establishment request message; the predetermined server is an AI server; Send the measurement report message of the specified network-side device and the measurement report message of the terminal device to the predetermined server, and the measurement report of the specified network-side device and the measurement report of the terminal device are used in the predetermined server for model training processing for network optimization; In response to receiving the model training processing result from the predetermined server, obtain a network optimization operation according to the model training processing result; The step of in response to receiving the model training processing result from the predetermined server and obtaining a network optimization operation according to the model training processing result includes: In response to receiving a model configuration establishment message from a predetermined server, perform model configuration and hyperparameter configuration according to the machine learning algorithm and training hyperparameter values in the model configuration establishment message; If the model configuration and hyperparameter configuration are successful, perform machine model training according to the configured model and hyperparameters to obtain trained machine learning model information; Send a distributed training request message to the predetermined server, and the distributed training request message includes the trained machine learning model information; In response to receiving the model training processing result from the predetermined server for the trained machine learning model information, obtain a network optimization operation according to the model training processing result.

18. The method according to claim 17, characterized in that, The step of in response to receiving a session establishment request message from a predetermined server and performing measurement configuration on a specified network-side device and a terminal device connected to the specified network-side device according to the measurement control information included in the session establishment request message includes: In response to the session establishment request message, perform measurement configuration on the network-side device where the client device is located according to the measurement control information; Send a measurement configuration request to the network-side device where the client device is not deployed to request the network-side device where the client device is not deployed to perform measurement configuration according to the measurement control information.

19. The method according to claim 17, wherein The session establishment request message further includes one or more machine learning models that the client device is required to support; after performing measurement configuration on the specified network-side device and the terminal device connected to the specified network-side device, the method further includes: If the measurement configuration of the network-side device and the terminal device is successful, and this client device supports the machine learning model, then send a session establishment response message carrying a configuration success flag to the predetermined server; If the measurement configuration of the network-side device fails or the measurement configuration of the terminal device fails, or this client device does not support the machine learning model, then send a session establishment response message carrying a configuration failure flag to the predetermined server.

20. The method according to claim 17, wherein The session establishment request message further includes a computing processing capability information item; when sending the measurement report messages of the specified network-side device and the terminal device to the predetermined server, the method further includes: Carry the computing processing capability of this client device in the measurement report of the network-side device where this client is located.

21. The method according to claim 17, wherein, Before, in response to receiving a session establishment request message from a predetermined server, and according to the measurement control information included in the session establishment request message, performing measurement configuration on the specified network-side device and the terminal device connected to the specified network-side device, the method further includes: In response to the received second activation message, activate the machine learning function, and obtain the machine learning description information and the policy information that the network needs to satisfy carried in the second activation message.

22. The method according to claim 17, wherein The model training processing result is a trained machine learning model; In response to receiving the model training processing result from the predetermined server, processing according to the model training processing result to obtain network optimization operations, including: In response to receiving a first model deployment request message from a predetermined server, according to the model derivation configuration file, the trained machine learning model, and the model parameter values in the first model deployment request message, deploy and execute the trained machine learning model to obtain a first network optimization operation instruction.

23. The method according to claim 22, characterized in that, After deploying the trained machine learning model, the method further includes: If the deployment of the trained machine learning model is successful, send a first model deployment response message to the predetermined server, and the first model deployment response message carries a first model deployment success flag; If the deployment of the trained machine learning model fails, send a first model deployment response message to the predetermined server, and the first model deployment response message carries a first model deployment failure flag.

24. The method according to claim 22, characterized in that, The model deployment request message further includes a first model performance reporting indication and a first model performance reporting method. After executing the trained machine learning model to obtain a first network optimization operation instruction, the method further includes: Execute the first network optimization operation instruction on the network-side device where this client device is located; Generate a first network optimization operation request to request the network-side device that has not deployed this client device to specify the first network optimization operation instruction; According to the first model performance reporting method, send a first model performance report message to the predetermined server, and the first model performance report message carries the corresponding model performance index value.

25. The method according to claim 18, wherein The model training processing result includes a second network optimization operation instruction; the obtaining of a network optimization operation according to the model training processing result in response to receiving the model training processing result of the trained machine learning model information from the predetermined server includes: In response to receiving a second network optimization operation request from the predetermined server, obtaining the second network optimization operation instruction carried in the second network optimization operation request.

26. The method according to claim 25, wherein The second network optimization operation request further includes a second model performance reporting indication and a second model performance reporting method. After obtaining the second network optimization operation instruction carried in the second network optimization operation request, the method further includes: Executing the second network optimization operation instruction on the network side device where the client device is located; Generating a second network operation request message to request the network side device where the client device is not deployed to specify the second network optimization operation instruction; Sending a second model performance report message to the predetermined server according to the second model performance reporting method, where the second model performance report message carries a second model performance metric value.

27. The method according to claim 18, characterized in that, The model training processing result includes a machine learning model after continued training and model parameter values; the obtaining of a network optimization operation according to the model training processing result in response to receiving the model training processing result of the trained machine learning model information from the predetermined server includes: In response to receiving a second model deployment request message, deploying and executing the machine learning model after continued training according to the model derivation configuration file, the machine learning model after continued training, and the model parameter values in the second model deployment request message, to obtain a third network optimization operation instruction.

28. The method according to claim 27, wherein After deploying the machine learning model after continued training to obtain a third network optimization operation instruction, the method further includes: If the deployment of the machine learning model after continued training is successful, sending a second model deployment response message to the predetermined server, where the second model deployment response message carries a second model deployment success flag; If the deployment of the machine learning model after continued training fails, sending a second model deployment response message to the predetermined server, where the second model deployment response message carries a second model deployment failure flag.

29. The method according to claim 27, wherein The second model deployment request message further includes a second model performance reporting indication and a second model performance reporting method. After executing the machine learning model after continued training to obtain a third network optimization operation instruction, the method further includes: Executing the third network optimization operation instruction on the network side device where the client device is located; Generating a third network optimization operation request to request the network side device where the client device is not deployed to execute the third network optimization operation instruction; Sending a second model performance report message to the predetermined server according to the second model performance reporting method, where the second model performance report message carries the corresponding model performance metric value.

30. A server, where the server is an AI server in a network, and the network is a wireless communication network, characterized in that The server includes: A measurement configuration request module, which is used to send a session establishment request message to a client device to request the client device to perform measurement configuration on a specified network-side device and perform measurement configuration on a terminal device connected to the specified network-side device; the client device is an AI distributed client; A measurement report receiving module, which is used to receive the measurement report message of the specified network-side device and the measurement report message of the terminal device; A collaborative training determination module, which is used to determine whether the client device needs to perform collaborative training according to the pre-acquired machine learning description information; A model training processing module, which is used to perform specified model training processing for network optimization based on whether the client device needs to perform collaborative training and the measurement data in the received measurement report message, and send the model training processing result to the client device for instructing the client device to obtain network optimization operations according to the model training processing result; wherein, when it is determined that the client device needs to perform collaborative training, the performing specified model training processing for network optimization based on whether the client device needs to perform collaborative training and the measurement data in the received measurement report message, and sending the model training processing result to the client device includes: obtaining the computing processing capability reported by the client device from the measurement report of the specified network-side device received, wherein the client device is deployed on the specified network-side device; selecting a machine learning algorithm and configuring training hyperparameter values according to the received measurement data, the pre-acquired computing processing capability of the client device, and the machine learning model supported by the client device; sending a model configuration establishment message to the client device to request the client device to perform model configuration according to the machine learning algorithm and training hyperparameter values; receiving the distributed training request message of the client device, where the distributed training request message includes the information of the trained machine learning model obtained by the client device for model training according to the machine learning algorithm and training hyperparameter values model configuration; performing model configuration and model training processing based on the information of the trained machine learning model to obtain a model training processing result, and sending the model training processing result to the client device.

31. A client device, the client device being an AI distributed client in a network, the network being a wireless communication network, characterized in that, The client device includes: A measurement configuration module, which is used to respond to the session establishment request message received from a predetermined server, and perform measurement configuration on a specified network-side device and a terminal device connected to the specified network-side device according to the measurement control information included in the session establishment request message; the predetermined server is an AI server; A measurement report sending module, which is used to send the measurement report message of the specified network-side device and the measurement report message of the terminal device to the predetermined server, and the measurement report of the specified network-side device and the measurement report of the terminal device are used for model training processing for network optimization in the predetermined server; An optimization operation determination module, configured to, in response to receiving a model training processing result from the predetermined server, obtain a network optimization operation according to the model training processing result; wherein, the obtaining a network optimization operation according to the model training processing result in response to receiving the model training processing result from the predetermined server includes: in response to receiving a model configuration establishment message from the predetermined server, performing model configuration and hyperparameter configuration according to the machine learning algorithm and training hyperparameter values in the model configuration establishment message; if the model configuration and hyperparameter configuration are successful, performing machine model training according to the configured model and hyperparameters to obtain trained machine learning model information; sending a distributed training request message to the predetermined server, where the distributed training request message includes the trained machine learning model information; in response to receiving a model training processing result from the predetermined server for the trained machine learning model information, obtaining a network optimization operation according to the model training processing result.

32. A network optimization system, characterized in that, Including a server, one or more client devices, and one or more network-side devices, wherein each client device is located in one of the one or more network-side devices; The server is configured to execute the network optimization method according to any one of claims 1-16; The one or more client devices are configured to execute the network optimization method according to any one of claims 17-29; The one or more network-side devices.

33. A network device, characterized in that, Including: One or more processors; A memory storing one or more programs, which when executed by the one or more processors cause the one or more processors to implement the network optimization method according to any one of claims 1-16, or the network optimization method according to any one of claims 17-29.

34. A storage medium, characterized in that, The storage medium stores a computer program, which when executed by a processor implements the network optimization method according to any one of claims 1-16, or the network optimization method according to any one of claims 17-29.

Images