WO2022236638A1 - 一种模型推理方法、模型推理装置及存储介质 - Google Patents

一种模型推理方法、模型推理装置及存储介质 Download PDF

Info

Publication number
WO2022236638A1
WO2022236638A1 PCT/CN2021/092900 CN2021092900W WO2022236638A1 WO 2022236638 A1 WO2022236638 A1 WO 2022236638A1 CN 2021092900 W CN2021092900 W CN 2021092900W WO 2022236638 A1 WO2022236638 A1 WO 2022236638A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
access network
radio access
network device
reasoning
Prior art date
Application number
PCT/CN2021/092900
Other languages
English (en)
French (fr)
Inventor
牟勤
洪伟
赵中原
熊可欣
Original Assignee
北京小米移动软件有限公司
北京邮电大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京小米移动软件有限公司, 北京邮电大学 filed Critical 北京小米移动软件有限公司
Priority to CN202180001523.6A priority Critical patent/CN115669030A/zh
Priority to PCT/CN2021/092900 priority patent/WO2022236638A1/zh
Priority to EP21941214.5A priority patent/EP4340300A1/en
Priority to US18/290,201 priority patent/US20240323099A1/en
Publication of WO2022236638A1 publication Critical patent/WO2022236638A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H01—ELECTRIC ELEMENTS
    • H01L—SEMICONDUCTOR DEVICES NOT COVERED BY CLASS H10
    • H01L25/00—Assemblies consisting of a plurality of individual semiconductor or other solid state devices ; Multistep manufacturing processes thereof
    • H01L25/16—Assemblies consisting of a plurality of individual semiconductor or other solid state devices ; Multistep manufacturing processes thereof the devices being of types provided for in two or more different main groups of groups H01L27/00 - H01L33/00, or in a single subclass of H10K, H10N, e.g. forming hybrid circuits
    • H01L25/167—Assemblies consisting of a plurality of individual semiconductor or other solid state devices ; Multistep manufacturing processes thereof the devices being of types provided for in two or more different main groups of groups H01L27/00 - H01L33/00, or in a single subclass of H10K, H10N, e.g. forming hybrid circuits comprising optoelectronic devices, e.g. LED, photodiodes
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/16—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/04—Network management architectures or arrangements
    • H04L41/044—Network management architectures or arrangements comprising hierarchical management structures
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/34—Signalling channels for network management communication
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/28—Flow control; Congestion control in relation to timing considerations
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W24/00—Supervisory, monitoring or testing arrangements
    • H04W24/02—Arrangements for optimising operational condition
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W88/00—Devices specially adapted for wireless communication networks, e.g. terminals, base stations or access point devices
    • H04W88/08—Access point devices
    • H04W88/085—Access point devices with remote components
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W88/00—Devices specially adapted for wireless communication networks, e.g. terminals, base stations or access point devices
    • H04W88/18—Service support devices; Network management devices

Definitions

  • the present disclosure relates to the technical field of wireless communication, and in particular to a model reasoning method, a model reasoning device and a storage medium.
  • the terminal requests the subscription model from the Operation Administration and Maintenance (OAM) network element through the wireless access device, and all the inference work of the model is performed by the Operation Administration and Maintenance (OAM) network element.
  • OAM Operation Administration and Maintenance
  • To complete the model inference work all model inference data needs to be uploaded to OAM.
  • OAM also needs to train the model based on the model inference data. Therefore, when OAM receives multiple subscription model requests at the same time, OAM cannot satisfy multiple subscription requests and provide model inference results, which will increase the feedback delay of model inference results and reduce system work efficiency.
  • the present disclosure provides a model reasoning method, a model reasoning device and a storage medium.
  • a model reasoning method which is applied to an OAM entity, and the method includes:
  • the control radio access network device In response to receiving the model subscription request information sent by the control radio access network device, determine a first model corresponding to the model subscription request information; divide the first model to obtain a first number of model segmentation blocks, and Distributing the first number of model segmentation blocks to a first number of control radio access network devices.
  • each model segmentation block in the first number of model segmentation blocks corresponds to allocation information
  • the allocation information includes an inference order of the first number of model partition blocks, and the control radio access network device corresponding to each model partition block.
  • the first number of control radio access network devices includes a first control radio access network device, and the first control radio access network device is a control radio access network device accessed by a terminal;
  • the distributing the first number of model segmentation blocks to the first number of control radio access network devices includes:
  • control radio access network equipment adjacent to the first control radio access network equipment determine a plurality of auxiliary control radio access network equipment
  • auxiliary control radio access network devices based on the computing power occupation state and load of each auxiliary control radio access network device, determine a second number of control radio access network devices; the second The number of controlled radio access network devices is other controlled radio access network devices in the first quantity except the first controlled radio access device;
  • the model reasoning method further includes:
  • the access network device sends the updated model parameters of the first model.
  • the model reasoning method further includes:
  • the first model analysis subscription update request In response to receiving the first model analysis subscription update request, update the distributed wireless access network device accessed by the terminal; wherein, the first model analysis subscription update request instructs the terminal to switch the distributed wireless access network device without switching Control wireless access network equipment;
  • the second model analysis subscription update request In response to receiving the second model analysis subscription update request, update the distributed wireless access network equipment accessed by the terminal, and re-segment the first model; wherein, the second model analysis subscription update request instructs the terminal to switch Distributed radio access network equipment, and switching control radio access network equipment.
  • a model reasoning method which is applied to control radio access network equipment, and the method includes:
  • model analysis subscription request In response to receiving the model analysis subscription request sent by the distributed wireless access network device, process the model analysis subscription request to obtain model subscription request information, and send the model subscription request information to OAM; receive the model segmentation sent by OAM block; the model segmentation block is a model segmentation block determined by dividing the first model; the first model is determined by the OAM based on the model subscription request information.
  • the method further includes:
  • model reasoning data request Sending a model reasoning data request to the distributed wireless access network device, where the model reasoning data request is used to obtain model reasoning data; based on the model reasoning data, reasoning is performed on the model segmentation block to obtain reasoning intermediate information of the model segmentation block .
  • the model segmentation blocks correspond to allocation information;
  • the allocation information includes the reasoning sequence of the first number of model segmentation blocks, and the control radio access network device corresponding to each model segmentation block;
  • the model reasoning method also includes:
  • the device is the last control radio access network device. After the model inference is completed, determine the first inference result corresponding to the first model, and send the first inference result to the first control radio access network device.
  • the first The control radio access network device is the control radio access network device for terminal access.
  • the method further includes:
  • the controlling radio access network device is a first controlling radio access network device, receiving the first reasoning result; sending the first reasoning result to the first distributed radio access network device, the first
  • the distributed radio access network device is a distributed radio access network device accessed by a terminal.
  • the model inference method further includes:
  • sending model subscription request information to OAM includes:
  • control radio access network device being the first control radio access network device
  • sending model subscription request information to the OAM wherein the first control radio access network device is the first distributed wireless access network accessed by the terminal The corresponding control radio access network equipment of the network equipment.
  • the model reasoning method further includes:
  • control radio access network device being the first control radio access network device
  • determine the second distributed radio access network device that resends the model analysis subscription request if the model analysis subscription request is received again, determine the second distributed radio access network device that resends the model analysis subscription request;
  • the second distributed wireless access network device is the distributed wireless access network device re-accessed after the terminal switches the distributed wireless access network device; the first reasoning result is sent to the second distributed wireless access network device, And send a model subscription update request to OAM.
  • the model reasoning method further includes:
  • the second control radio access network device In response to the control radio access network device being the second control radio access network device, if the model analysis subscription request is received again, determine the second distributed radio access network device that resends the model analysis subscription request, and The second control radio access network device that receives the model analysis subscription request again, the second control radio access network device is the control radio access network device corresponding to the second distributed radio access network device; the second The distributed radio access network device is a distributed radio access network device re-accessed after the terminal switches the distributed radio access network device; the first reasoning result is sent to the second control radio access network device, and sent to the OAM A model subscribes to an update request.
  • a model reasoning method is provided, which is applied to a distributed radio access network device, and the method includes:
  • the model analysis subscription request In response to receiving the model analysis subscription request sent by the terminal, send the model analysis subscription request to the control radio access network device; wherein, the model analysis subscription request is used to obtain a first model from OAM; the first model includes A first number of model partition blocks.
  • the method further includes:
  • model reasoning data request sent by the control radio access network device, the model reasoning data request being used to obtain the model reasoning data; obtaining the model reasoning data from the terminal, and sending it to the control radio access network device.
  • the method further includes:
  • the distributed radio access network device In response to the distributed radio access network device being the first distributed radio access network device, receiving a first inference result sent by the first control radio access network device; and sending the first inference result to the terminal.
  • the method further includes:
  • the distributed radio access network device is the first distributed radio access network device, receiving performance data sent by the terminal; the performance data is real performance data after the terminal has adjusted an execution strategy based on the first model; A device for controlling the radio access network to send the performance data.
  • the method further includes:
  • the wireless access network device is the second distributed wireless access network device
  • the access network device sends a model analysis subscription request; wherein, the second distributed wireless access network device is a distributed wireless access network device re-accessed after the terminal switches the distributed wireless access network device.
  • a model reasoning device which is applied to an OAM entity, and the device includes:
  • the determining module is configured to determine the first model corresponding to the model subscription request information in response to receiving the model subscription request information sent by the control radio access network device; the sending module is configured to divide the first model, Obtain a first number of model segmentation blocks, and distribute the first number of model segmentation blocks to a first number of control radio access network devices.
  • each model segmentation block in the first number of model segmentation blocks corresponds to allocation information
  • the allocation information includes an inference order of the first number of model partition blocks, and the control radio access network device corresponding to each model partition block.
  • the first number of control radio access network devices includes a first control radio access network device, and the first control radio access network device is a control radio access network device accessed by a terminal;
  • control radio access network equipment adjacent to the first control radio access network equipment determine a plurality of auxiliary control radio access network equipment
  • auxiliary control radio access network devices based on the computing power occupation state and load of each auxiliary control radio access network device, determine a second number of control radio access network devices; the second The number of controlled radio access network devices is other controlled radio access network devices in the first quantity except the first controlled radio access device;
  • the model reasoning device further includes: a receiving module
  • the receiving module is configured to receive model performance update data sent by the first control radio access network device; update the first model based on the model performance update data, determine updated model parameters of the first model, and Send the updated model parameters of the first model to the first controlling radio access network device.
  • the receiving module is also used for:
  • the first model analysis subscription update request In response to receiving the first model analysis subscription update request, update the distributed wireless access network device accessed by the terminal; wherein, the first model analysis subscription update request instructs the terminal to switch the distributed wireless access network device without switching Control wireless access network equipment;
  • the second model analysis subscription update request In response to receiving the second model analysis subscription update request, update the distributed wireless access network equipment accessed by the terminal, and re-segment the first model; wherein, the second model analysis subscription update request instructs the terminal to switch Distributed radio access network equipment, and switching control radio access network equipment.
  • a model reasoning apparatus which is applied to control radio access network equipment, and the apparatus includes:
  • a sending module configured to process the model analysis subscription request to obtain model subscription request information in response to receiving the model analysis subscription request sent by the distributed wireless access network device, and send the model subscription request information to OAM; receive A module, configured to receive a model segmentation block sent by OAM; the model segmentation block is a model segmentation block determined by dividing a first model; the first model is determined by OAM based on the model subscription request information.
  • the sending module is also used for:
  • model reasoning data request Sending a model reasoning data request to the distributed wireless access network device, where the model reasoning data request is used to obtain model reasoning data; based on the model reasoning data, reasoning is performed on the model segmentation block to obtain reasoning intermediate information of the model segmentation block .
  • the model segmentation blocks correspond to allocation information;
  • the allocation information includes the reasoning sequence of the first number of model segmentation blocks, and the control radio access network device corresponding to each model segmentation block;
  • the sending module is also used for:
  • the device is the last control radio access network device.
  • the wireless access network control device is the wireless access network control device for terminal access.
  • the sending module is also used for:
  • the controlling radio access network device is a first controlling radio access network device, receiving the first reasoning result; sending the first reasoning result to the first distributed radio access network device, the first
  • the distributed radio access network device is a distributed radio access network device accessed by a terminal.
  • the receiving module is further configured to:
  • the sending module is also used for:
  • control radio access network device being the first control radio access network device
  • sending model subscription request information to the OAM wherein the first control radio access network device is the first distributed wireless access network accessed by the terminal The corresponding control radio access network equipment of the network equipment.
  • the sending module is also used for:
  • control radio access network device being the first control radio access network device
  • determine the second distributed radio access network device that resends the model analysis subscription request if the model analysis subscription request is received again, determine the second distributed radio access network device that resends the model analysis subscription request;
  • the second distributed wireless access network device is the distributed wireless access network device re-accessed after the terminal switches the distributed wireless access network device; the first reasoning result is sent to the second distributed wireless access network device, And send a model subscription update request to OAM.
  • the sending module is also used for:
  • the second control radio access network device In response to the control radio access network device being the second control radio access network device, if the model analysis subscription request is received again, determine the second distributed radio access network device that resends the model analysis subscription request, and The second control radio access network device that receives the model analysis subscription request again, the second control radio access network device is the control radio access network device corresponding to the second distributed radio access network device; the second The distributed radio access network device is a distributed radio access network device re-accessed after the terminal switches the distributed radio access network device; the first reasoning result is sent to the second control radio access network device, and sent to the OAM A model subscribes to an update request.
  • a model reasoning apparatus which is applied to a distributed radio access network device, and the apparatus includes:
  • the sending module is configured to send the model analysis subscription request to the control radio access network device in response to receiving the model analysis subscription request sent by the terminal; wherein, the model analysis subscription request is used to obtain the first model from the OAM;
  • the first model includes a first number of model partitions.
  • the device further includes: a receiving module
  • the receiving module is configured to receive the model reasoning data request sent by the control radio access network device, the model reasoning data request is used to obtain the model reasoning data; obtain the model reasoning data from the terminal, and send it to the control radio access network device.
  • the receiving module is also used for:
  • the distributed radio access network device In response to the distributed radio access network device being the first distributed radio access network device, receiving a first inference result sent by the first control radio access network device; and sending the first inference result to the terminal.
  • the receiving module is also used for:
  • the distributed radio access network device is the first distributed radio access network device, receiving performance data sent by the terminal; the performance data is real performance data after the terminal has adjusted an execution strategy based on the first model; A device for controlling the radio access network to send the performance data.
  • the receiving module is also used for:
  • the wireless access network device is the second distributed wireless access network device
  • the access network device sends a model analysis subscription request; wherein, the second distributed wireless access network device is a distributed wireless access network device re-accessed after the terminal switches the distributed wireless access network device.
  • a model reasoning device including:
  • a processor a memory for storing processor-executable instructions; wherein, the processor is configured to: execute the first aspect or the model reasoning method described in any implementation manner of the first aspect, or execute the second aspect Or the model reasoning method described in any implementation manner of the second aspect, or execute the third aspect or the model reasoning method described in any implementation manner of the third aspect.
  • a non-transitory computer-readable storage medium When the instructions in the storage medium are executed by the processor of the mobile terminal, the mobile terminal can execute the first aspect or the first The model inference method described in any implementation manner of the second aspect, or enabling the mobile terminal to execute the model inference method described in the second aspect or any implementation manner of the second aspect, or enabling the mobile terminal to execute the third aspect or The model reasoning method described in any one of the implementation manners of the third aspect.
  • the OAM of the present disclosure divides the model and distributes the model segmentation blocks to different control radio access network devices, so that the control of radio access network devices can be better developed.
  • the AI processing capability solves the problem of insufficient AI processing capability of wireless access network equipment, and is conducive to network load balancing. Fully utilizing local AI processing capabilities can effectively improve model inference efficiency, reduce inference delays, and help balance network loads, providing users with efficient and convenient AI analysis services. .
  • Fig. 1 shows a schematic diagram of a basic functional framework structure according to an exemplary embodiment.
  • Fig. 2 shows a schematic diagram of a network architecture according to an exemplary embodiment.
  • Fig. 3 is a flow chart showing a model reasoning method according to an exemplary embodiment.
  • Fig. 4 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 5 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 6 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 7 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 8 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 9 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 10 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 11 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 12 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 13 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 14 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 15 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 16 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 17 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 18 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 19 is a flow chart showing another model reasoning method according to an exemplary embodiment.
  • Fig. 20 is a flow chart of terminal switching in a model reasoning method according to an exemplary embodiment.
  • Fig. 21 is a flow chart of terminal switching in a model reasoning method according to an exemplary embodiment.
  • Fig. 22 is a schematic diagram of a protocol and interface of a model reasoning method according to an exemplary embodiment.
  • Fig. 23 is a schematic diagram of a protocol and an interface for AI analysis task delivery when a terminal switches under the same gNB-CU in a model-free reasoning method according to an exemplary embodiment.
  • Fig. 24 is a schematic diagram of a protocol and an interface for AI analysis task delivery when a terminal switches between gNB-CUs in a model-free reasoning method according to an exemplary embodiment.
  • Fig. 25 is a block diagram of a model reasoning device according to an exemplary embodiment.
  • Fig. 26 is a block diagram of another model reasoning device according to an exemplary embodiment.
  • Fig. 27 is a block diagram of another model reasoning device according to an exemplary embodiment.
  • Fig. 28 is a block diagram of a model reasoning device according to an exemplary embodiment.
  • Fig. 29 is a block diagram of another model reasoning device according to an exemplary embodiment.
  • FIG. 1 shows a schematic diagram of a basic functional framework structure according to an exemplary embodiment. As shown in FIG. 1 , as an initial architecture, a potential wireless network architecture supporting artificial intelligence is used.
  • the data collection & preparation unit includes data collection and data preprocessing functions, data collection can be performed in multiple network elements, and the provided data includes measurement data, feedback performance data and model performance data, etc.
  • Model Training Iterates the machine learning model through calculation and processing to obtain a better model for reasoning.
  • the input includes training data and model performance feedback.
  • Model inference unit Use the trained machine learning model to generate prediction results or decision results.
  • Execution unit Use the model reasoning results to formulate and execute strategies, and feed back relevant performance results after execution to Data collection.
  • Fig. 2 shows a schematic diagram of a network architecture according to an exemplary embodiment.
  • the system includes a terminal, gNB-DU, gNB-CU and OAM.
  • the terminal accesses the gNB-DU through a wireless channel, and multiple gNB-DUs access the gNB-CU through the F1 interface. Between gNB-CUs Connect via Xn interface.
  • OAM is mainly responsible for the work of the model training functional unit in the wireless network architecture supporting AI, responsible for model training and model segmentation;
  • gNB-CU is responsible for the work of the model reasoning function unit, responsible for completing model reasoning;
  • gNB-DU is mainly responsible for data
  • the work of the collection function unit is responsible for the collection of real-time inference data and the collection of terminal performance feedback data;
  • the terminal is responsible for the work of the action execution function unit and is responsible for making corresponding policy adjustments based on the model reasoning results.
  • the terminal is responsible for the work of the execution (Action) functional unit; the next Generation Node B Distributed Unit (gNB-DU) is responsible for forwarding the analysis request and reasoning results of the terminal, and executing the data collection (Data collection) functional unit. Work.
  • the base station control unit (next Generation Node B Control Unit, gNB-CU) is responsible for forwarding the analysis request and reasoning results of the terminal, and performing the work of the data collection functional unit.
  • OAM is responsible for performing the work of model training and inference (Model training and Model inference) functional units.
  • the execution process includes: the terminal initiates an analysis subscription request to the gNB-DU, the gNB-DU sends the terminal analysis subscription request to the gNB-CU, and the gNB-CU reports the terminal analysis subscription request to the OAM.
  • OAM selects the appropriate model according to the analysis subscription request of the terminal, and starts the model reasoning work.
  • OAM initiates a model inference data request to gNB-CU, and network elements at all levels (gNB-CU, gNB-DU, terminal) collect model inference data according to the inference data request information, process the data and send it to OAM.
  • OAM uses model inference data to perform model inference to obtain inference results, and sends the inference results to gNB-CU, gNB-CU sends the inference results to gNB-DU, gNB-DU sends the inference results to the terminal, and the terminal can use the inference results to make Make corresponding policy adjustments.
  • model inference work is all completed by the OAM network management, and all model inference data needs to be sent to the OAM.
  • This solution uploads real-time model inference data from the wireless side to the network management system, which poses challenges to data security, especially in scenarios where the model inference data includes terminal business data, this solution will be limited.
  • Model reasoning delay includes the transmission delay caused by uploading model reasoning data to OAM, the calculation delay of model reasoning, and the transmission delay caused by OAM sending reasoning results to the terminal.
  • the first part of the delay is relatively large and will As a result, the feedback of inference results is not timely, which affects the terminal service experience.
  • the present disclosure provides a model reasoning method, which assigns model reasoning tasks to different gNB-CUs (that is, control radio access network devices in the embodiments of the present disclosure). Further, on the basis of the artificial intelligence architecture of the wireless network, the model is divided according to the AI processing capability of each network element, and multiple network elements with AI processing capability are selected to assist the model reasoning network elements to which the terminal belongs to jointly complete the model reasoning work. The reasoning results are fed back to the terminal, and the terminal implements corresponding policy adjustments based on the reasoning results, and performs performance feedback to achieve continuous optimization of the model.
  • the specific process is as follows: first, the terminal initiates a model analysis subscription request, and the gNB-CU connected to the terminal generates model subscription request information according to its own AI processing capability and the analysis subscription request information of the terminal, and reports it to OAM. OAM performs model selection and model segmentation, allocation and distribution of model segmentation blocks according to the model subscription request, and sends model segmentation block allocation information to all gNB-CUs participating in joint reasoning.
  • the gNB-CU accessed by the terminal initiates a model reasoning data request, and the relevant network elements collect and process the data and send it to the gNB-CU.
  • the gNB-CU accessed by the terminal uses the model inference data to complete the inference of the first model segmentation block, and sends the inference intermediate results to the gNB-CU where the next model segmentation block is located according to the model segmentation block allocation information, until it is responsible for the last model
  • the gNB-CU of the segmentation block inference task obtains the inference result, it sends the inference result to the gNB-CU accessed by the terminal according to the allocation information of the model segmentation block.
  • the gNB-CU connected to the terminal sends the inference result to the terminal, and the terminal uses the inference result to make corresponding policy adjustments.
  • the gNB-CU collects model performance data and terminal performance feedback data and reports them to the OAM.
  • the OAM trains and optimizes the model, and sends the updated model parameters to the gNB-CU.
  • the re-delivery of wireless network AI analysis tasks is carried out.
  • it can be divided into the following two scenarios:
  • the terminal When the terminal switches to another gNB-DU under the same gNB-CU (that is, the distributed radio access network device in the embodiment of the present disclosure), the terminal re-initiates the analysis subscription request, and the gNB-CU and OAM update the terminal Analytics subscription request information for . If the current reasoning task is not completed when the terminal is handed over, the gNB-CU will continue to complete the reasoning task. After obtaining the reasoning result, it will send the reasoning result to the gNB-DU currently connected to the terminal according to the access location in the update analysis subscription request message. -DU sends the inference result to the terminal. After the terminal switching is completed, the newly connected gNB-DU is responsible for completing relevant data collection and data forwarding tasks.
  • the terminal When the terminal switches to another gNB-CU, the terminal resends a model analysis subscription request, and the newly accessed gNB-CU of the terminal sends a model subscription request to the OAM.
  • the OAM updates the analysis subscription request of the terminal, and sends the updated analysis subscription request information to the source gNB-CU of the terminal. If the current inference task is not completed when the terminal switches, the source gNB-CU completes the inference task, and after obtaining the inference result, sends the inference result to the gNB-CU newly accessed by the terminal according to the access location in the update analysis subscription request message.
  • the source gNB-CU updates the terminal model subscription analysis request information, and is no longer responsible for the tasks related to the terminal model subscription analysis request.
  • the gNB-CU newly accessed by the terminal sends the inference result to the gNB-DU newly accessed by the terminal, and the gNB-DU sends the inference result to the terminal.
  • OAM performs model selection and segmentation again according to the model subscription analysis request sent by the gNB-CU newly accessed by the terminal, and sends the model segmentation block allocation information to the gNB-CU participating in the joint reasoning.
  • It also provides a delivery method for wireless network AI analysis tasks in the scenario of high-speed terminal mobility, which solves the problem of discontinuous AI analysis services caused by terminal switching, ensures the efficiency and continuity of wireless network AI analysis services, and improves Terminal service experience is also conducive to improving the efficiency of wireless network operation.
  • the wireless communication system in the embodiment of the present disclosure is a network that provides a wireless communication function.
  • Wireless communication systems can use different communication technologies, such as code division multiple access (CDMA), wideband code division multiple access (WCDMA), time division multiple access (TDMA) , frequency division multiple access (FDMA), orthogonal frequency-division multiple access (OFDMA), single carrier frequency-division multiple access (single Carrier FDMA, SC-FDMA), carrier sense Multiple Access/Conflict Avoidance (Carrier Sense Multiple Access with Collision Avoidance).
  • CDMA code division multiple access
  • WCDMA wideband code division multiple access
  • TDMA time division multiple access
  • FDMA frequency division multiple access
  • OFDMA orthogonal frequency-division multiple access
  • single Carrier FDMA single Carrier FDMA
  • SC-FDMA carrier sense Multiple Access/Conflict Avoidance
  • Carrier Sense Multiple Access with Collision Avoidance Carrier Sense Multiple Access with Collision Avoidance
  • the network can be divided into 2G (English: generation) network, 3G network, 4G network or future evolution network, such as 5G network, 5G network can also be called a new wireless network ( New Radio, NR).
  • 2G International: generation
  • 3G network 4G network or future evolution network, such as 5G network
  • 5G network can also be called a new wireless network ( New Radio, NR).
  • New Radio New Radio
  • the present disclosure sometimes simply refers to a wireless communication network as a network.
  • the wireless access network device may be: a base station, an evolved base station (evolved node B, base station), a home base station, an access point (access point, AP) in a wireless fidelity (wireless fidelity, WIFI) system, a wireless relay Node, wireless backhaul node, transmission point (transmission point, TP) or transmission and reception point (transmission and reception point, TRP), etc., can also be gNB in the NR system, or it can also be a component or a part of equipment that constitutes a base station Wait.
  • the network device When it is a vehicle-to-everything (V2X) communication system, the network device may also be a vehicle-mounted device. It should be understood that in the embodiments of the present disclosure, there is no limitation on the specific technology and specific equipment form adopted by the network equipment.
  • V2X vehicle-to-everything
  • terminals involved in this disclosure can also be referred to as terminal equipment, user equipment (User Equipment, UE), mobile station (Mobile Station, MS), mobile terminal (Mobile Terminal, MT), etc.
  • a device providing voice and/or data connectivity for example, a terminal may be a handheld device with a wireless connection function, a vehicle-mounted device, and the like.
  • examples of some terminals are: smart phones (Mobile Phone), pocket computers (Pocket Personal Computer, PPC), handheld computers, personal digital assistants (Personal Digital Assistant, PDA), notebook computers, tablet computers, wearable devices, or Vehicle equipment, etc.
  • V2X vehicle-to-everything
  • the terminal device may also be a vehicle-mounted device. It should be understood that the embodiment of the present disclosure does not limit the specific technology and specific device form adopted by the terminal.
  • Fig. 3 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in Figure 3, the model reasoning method is used in the OAM entity, including the following steps.
  • step S11 in response to receiving the model subscription request information sent by the control radio access network device, the first model corresponding to the model subscription request information is determined.
  • the model subscription request information includes the AI processing capability information for controlling the wireless access network device itself, and the terminal model analysis subscription request information.
  • the AI processing capability information includes the computing speed of the base station server and the current surplus computing power.
  • the OAM selects a model that conforms to the terminal model analysis subscription request information, and further determines a model with an appropriate scale among the models that meet the requirements, that is, the first model, according to the AI processing capability of the control radio access network device.
  • the model that meets the requirements and has a suitable scale is referred to as the first model.
  • step S12 the first model is divided to obtain a first number of model segmentation blocks, and the first number of model segmentation blocks are distributed to the first number of control radio access network devices.
  • the OAM divides the first model into a first number of model segmentation blocks according to the AI processing capability information of the control radio access network device, and determines the same number of control radio access network devices according to the first number. Distributing the first number of model segmentation blocks to the first number of control radio access network devices.
  • the first number of control radio access network devices is determined by OAM based on the control radio access network devices adjacent to the control radio access network device that sends the model subscription request information, and the basis for determination may be that the control radio access network device The choice of computing power occupancy and load conditions of network equipment is relatively free.
  • the computing power can be balanced to multiple different control wireless access network devices through the cooperative reasoning method of multiple control wireless access network devices, and the control wireless access network devices can be fully utilized.
  • the local AI processing capability can effectively improve the efficiency of model reasoning.
  • each model segmentation block in the first number of model segmentation blocks corresponds to allocation information.
  • the allocation information includes the inference sequence of the first number of model segmentation blocks, and the control radio access network device corresponding to each model segmentation block.
  • the control radio access network device corresponding to each model segmentation block is included in the allocation information in a manner of corresponding identification.
  • the first number of control radio access network devices includes a first control radio access network device, where the first control radio access network device is a control radio access network device accessed by a terminal.
  • Fig. 4 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in Figure 4, the model reasoning method is used in the OAM entity, including the following steps.
  • step S21 among the control radio access network devices adjacent to the first control radio access network device, a plurality of auxiliary control radio access network devices are determined.
  • the OAM selects an auxiliary control radio access network device that can assist in model reasoning among control radio access network devices adjacent to the first control radio access network device.
  • step S22 among the plurality of auxiliary control radio access network devices, a second quantity of control radio access network devices is determined based on the computing power idle state of each auxiliary control radio access network device.
  • the OAM determines the second number of control radio access network devices that can participate in this model reasoning according to the computing power occupation state and load of each control radio access network device.
  • the second number of controlling radio access network devices is other controlling radio access network devices in the first number except the first controlling radio access network device.
  • step S23 based on the inference order of the first number of model segmentation blocks, the first model segmentation block is sent to the first control radio access network device, and the remaining number of model segmentation blocks are distributed to the second number of control Wireless access network equipment.
  • the OAM sends the first model segmentation block to the first control radio access network device (for example, gNB-CU1), and sends the rest of the model segmentation blocks to other control radio access network devices participating in joint reasoning.
  • the network device and send the allocation information corresponding to the model segmentation block to all control radio access network devices participating in the joint reasoning.
  • Fig. 5 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in Figure 5, the model reasoning method is used in the OAM entity, including the following steps.
  • step S31 the model performance update data sent by the first control radio access network device is received.
  • the first controlling radio access network device compares the received performance data with the first reasoning result of the first model, determines model performance update data, and sends the model performance update data to the OAM.
  • the model performance update data may be model accuracy.
  • the OAM may also receive performance data sent by the first control wireless access device.
  • step S32 the first model is updated based on the model performance update data, the updated model parameters of the first model are determined, and the updated model parameters of the first model are sent to the first control radio access network device.
  • the OAM trains and optimizes the first model according to the performance data and model performance update data, obtains the updated model parameters of the first model, and sends the updated model parameters of the first model to the first control radio access network equipment.
  • the terminal in response to OAM receiving the first model analysis subscription update request, and the information included in the first model analysis subscription update request is the terminal's model analysis subscription information, the terminal is updated based on the first model analysis subscription update request Model analysis request information for .
  • the first model analysis subscription update request is received in response to the OAM, and the information included in the first model analysis subscription update request is the model analysis subscription information of the terminal and the AI of the second control radio access network device Processing capability information, based on model analysis subscription information and AI processing capability information of the second control radio access network device, re-segment the first model, and send the first model segmentation block to the second control radio access network device, The remaining model segmentation blocks are sent to other control radio access network devices participating in inference.
  • Fig. 6 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in Fig. 6, the model reasoning method is used in controlling radio access network equipment, including the following steps.
  • step S41 in response to receiving the model analysis subscription request sent by the distributed radio access network device, process the model analysis subscription request to obtain model subscription request information, and send the model subscription request information to the OAM.
  • the model analyzes the subscription request, including the identification of the terminal, analysis request type, and access location information.
  • the terminal accesses the first distributed radio access network device (for example, gNB-DU1), and gNB-DU1 and gNB-DU2 access gNB-CU1.
  • the terminal identifier is GUTI
  • the analysis request type is represented by analysis ID, such as analysis ID1: location prediction analysis service, analysis ID2: load prediction analysis service.
  • the access location mainly includes the control radio access network equipment and distributed radio access network equipment information currently accessed by the terminal.
  • model analysis subscription request sent by the distributed radio access network device
  • model subscription request information In response to the control radio access network device receiving the model analysis subscription request sent by the distributed radio access network device, generate model subscription request information according to its own AI processing capability and model analysis subscription request, and send the model subscription request information to OAM .
  • step S42 the model segmentation block sent by OAM is received.
  • the model segmentation block is a model segmentation block determined by dividing the first model.
  • the first model is determined by the OAM based on the model subscription request information.
  • the computing power can be balanced to multiple different control wireless access network devices through the cooperative reasoning method of multiple control wireless access network devices, and the control wireless access network devices can be fully utilized.
  • the local AI processing capability can effectively improve the efficiency of model reasoning.
  • Fig. 7 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in Fig. 7, the model reasoning method is used in controlling radio access network equipment, including the following steps.
  • step S51 a model reasoning data request is sent to the distributed radio access network device.
  • the model reasoning data request is used to acquire model reasoning data.
  • the radio access network device is controlled to send a model reasoning data request to the distributed radio access network device.
  • the control radio access network device can send a model reasoning data request to the distributed radio access network device accessed by the terminal, or send a model reasoning data request to other distributed radio access network devices participating in reasoning within the range of the control radio access network device.
  • the radio access network device sends a model inference data request.
  • step S52 reasoning is performed on the model segmentation block based on the model inference data to obtain inference intermediate information of the model segmentation block.
  • the model reasoning data includes the model reasoning data collected by the distributed radio access network device and the model reasoning data reported by the terminal.
  • the radio access network device is controlled to perform inference on the model segmentation blocks based on the received model inference data, and determine inference intermediate information of each model segmentation block.
  • each model segmentation block corresponds to allocation information.
  • the radio access network device is controlled to receive the model segmentation block and receive allocation information.
  • the assignment information includes the reasoning order of the first number of model division blocks, and the control radio access network device corresponding to each model division block.
  • the control radio access network device corresponding to each model segmentation block is included in the allocation information in a manner of corresponding identification.
  • Fig. 8 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in FIG. 8, the model reasoning method is used in controlling radio access network equipment, including the following steps.
  • step S61 in response to the fact that the controlling radio access network device is not the last controlling radio access network device, based on the reasoning order, the reasoning intermediate information is sent to the next controlling radio access network device.
  • the current control radio access network device of the current inference model in response to the fact that the control radio access network device of the current inference model is not the control radio access network device where the last model segmentation block is located, the current control radio access network device infers the model segmentation block according to the allocation information The inference sequence of the inference, and the inference intermediate information is sent to the control radio access network device where the next model segmentation block is located.
  • step S62 in response to the fact that the controlling radio access network device is the last controlling radio access network device, after the model reasoning is completed, determine the first reasoning result corresponding to the first model, and send the first reasoning result to the first controller
  • the radio access network device the first control radio access network device is the control radio access network device accessed by the terminal.
  • the control radio access network device of the current inference model in response to the fact that the control radio access network device of the current inference model is the control radio access network device where the last model segmentation block is located, the current control radio access network device infers the model segmentation block according to the allocation information
  • the reasoning sequence of the first model is completed, and the first reasoning result corresponding to the first model is determined, and the first reasoning result is sent to the control radio access network device where the first model segmentation block is located, that is, the first Control wireless access network equipment.
  • the first control radio access network device is a control radio access network device accessed by a terminal.
  • the model reasoning method disclosed in the present disclosure it is guaranteed that the first model segmentation block is reasoned on the control wireless access network device currently accessed by the terminal, and the original reasoning data required by the terminal reasoning is only provided to the currently connected wireless access network device . Only the inference intermediate information is transmitted between other control wireless access network devices participating in the joint inference.
  • the inference intermediate information has been processed by features, has a small amount of data and is difficult to reversely infer terminal information. This mechanism ensures that the wireless network is sensitive. Data security also saves data transmission overhead.
  • Fig. 9 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in FIG. 9 , the model reasoning method is used in controlling radio access network equipment, and includes the following steps.
  • step S71 in response to the controlling radio access network device being the first controlling radio access network device, a first reasoning result is received.
  • the last model segmentation block corresponds to the control radio access network device performing inference to determine the model inference result, that is, the first inference result.
  • the first inference result is sent to the first control radio access network device, and the first control radio access network device obtains the first model inference result corresponding to the first model.
  • step S72 the first reasoning result is sent to the first distributed radio access network device.
  • the first controlling radio access network device determines the first reasoning result of the first model according to the received reasoning result. Send the first reasoning result to the distributed radio access network device accessed by the terminal.
  • Fig. 10 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in FIG. 10 , the model reasoning method is used in controlling radio access network equipment, and includes the following steps.
  • step S81 performance data sent by the first distributed radio access network device is received.
  • the performance data is real performance data after the terminal adjusts the execution strategy based on the first model. After the terminal adjusts the execution strategy based on the first model, it reports the obtained real performance data to the access distributed wireless access network device.
  • the distributed radio access network is sent to the control radio access network device.
  • the distributed radio access network device accessed by the terminal is gNB-DU1
  • the control radio access network device corresponding to gNB-DU1 is gNB-CU1.
  • the terminal determines the real performance data, it sends it to gNB-DU1.
  • gNB-DU1 reports to gNB-CU1.
  • the performance data can be the quantification of the performance improvement brought by the AI analysis service. For example, after the terminal subscribes to a certain analysis and executes corresponding policy adjustments based on the analysis results, it can save 5% of power.
  • step S82 the performance data is processed to obtain model performance update data, and the model performance update data is sent to the OAM.
  • the performance data includes model performance data and performance feedback data fed back by the terminal.
  • the first control radio access network device processes the model performance data fed back by the terminal and the performance feedback data to obtain model performance update data, and sends the model performance update data to the OAM.
  • Fig. 11 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in FIG. 11 , the model reasoning method is used in controlling radio access network equipment, and includes the following steps.
  • step S91 in response to the control radio access network device being the first control radio access network device, send model subscription request information to the OAM.
  • the first control radio access network device is the control radio access network device corresponding to the first distributed wireless network device accessed by the terminal.
  • the first control radio access network device receives the model analysis subscription request sent by the distributed radio access network device, analyzes the subscription request according to its own AI capability and model, generates model subscription request information, and sends the model subscription request information to OAM .
  • the terminal due to the mobility of the terminal, handover of the distributed radio access network equipment accessed may occur.
  • the terminal switches the distributed wireless access network equipment that it accesses, and does not switch the control wireless access network equipment.
  • Fig. 12 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in FIG. 12 , the model reasoning method is used in controlling radio access network equipment, and includes the following steps.
  • step S101 in response to the control radio access network device being the first control radio access network device, if the model analysis subscription request is received again, determine the second distributed radio access network device that resends the model analysis subscription request.
  • the second distributed radio access network device is the distributed radio access network device re-accessed by the terminal after switching the distributed radio access network device.
  • the first control radio access network device In response to the first control radio access network device receiving the model analysis subscription request again, it is determined that the distributed radio access network device accessed by the terminal is switched, the analysis subscription information of the terminal is updated, and the analysis subscription information is reported to the OAM.
  • the first control radio access network device has not completed the reasoning task of the first model, and the first control radio access network device (for example, gNB-CU1) completes the current model reasoning task and obtains the first reasoning result based on The access location in the update analysis request message sends the first reasoning result to the second distributed radio access network device (for example, gNB-DU2), and the gNB-DU2 forwards it to the terminal.
  • the first control radio access network device for example, gNB-CU1
  • step S102 the first reasoning result is sent to the second distributed radio access network device, and a model subscription update request is sent to the OAM.
  • the first control radio access network device sends the first reasoning result to the second distributed radio access network device, and sends a model subscription update request to the OAM, requesting the OAM to update the analysis request information of the terminal.
  • gNB-CU1 updates the analysis request information of the terminal.
  • the gNB-CU1 reports the terminal analysis request message to the OAM, and the OAM updates the terminal analysis request information.
  • gNB-CU1 completes the current model reasoning task, and after obtaining the first reasoning result, sends the first reasoning result to gNB-DU2 according to the access location in the update analysis request message, and gNB-DU2 forwarded to the terminal.
  • the terminal switches the access distributed wireless access network equipment, and switches and controls the wireless access network equipment. Its implementation includes:
  • Fig. 13 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in Figure 13, the model reasoning method is used to control radio access network equipment, including the following steps.
  • step S111 in response to the control radio access network device being the second control radio access network device, if the model analysis subscription request is received again, determine the second distributed radio access network device that resends the model analysis subscription request, And the second control radio access network device that receives the model analysis subscription request again.
  • the second control radio access network device is a control radio access network device corresponding to the second distributed radio access network device.
  • the second distributed radio access network device is the distributed radio access network device that the terminal re-accesses after switching the distributed radio access network device.
  • control radio access network device In response to the control radio access network device re-receiving the model analysis subscription request sent by the distributed radio access network device, and the re-received control radio access network device is the second control radio access network device, determining that the terminal re-accesses The second distributed wireless access network device.
  • step S112 the first reasoning result is sent to the second control radio access network device, and a model subscription update request is sent to the OAM.
  • the first controlling radio access network device sends the first reasoning result to the second controlling radio access network device, and is no longer responsible for the terminal's analysis request. And send a model subscription update request to OAM.
  • gNB-CU2 sends a model subscription request to OAM, including its own AI processing capability information and terminal analysis subscription request information.
  • OAM updates the analysis subscription request information of the current terminal according to the model subscription request; and sends the updated analysis subscription request information to the source base station gNB-CU1.
  • gNB-CU1 completes the reasoning task, obtains the first reasoning result, and then sends the first reasoning result to gNB-CU2 according to the access location in the update analysis request message.
  • gNB-CU1 updates the terminal analysis request information, and is no longer responsible for the tasks related to the terminal analysis request.
  • gNB-CU2 sends the first reasoning result to gNB-DU3, and gNB-DU3 forwards it to the terminal.
  • OAM re-segments the model according to the AI processing capability information in the model subscription request of gNB-CU2, sends the first block to the gNB-CU2 that initiates the request, sends the remaining blocks to other gNB-CUs, and distributes the information of the model segmentation blocks Send to the gNB-CU participating in the joint reasoning.
  • model reasoning method provided in this disclosure, it is ensured that the current first reasoning result of the source distributed wireless access network device is smoothly fed back to the terminal that has been switched, and the newly connected distributed wireless access network device quickly takes over the model reasoning task, thereby It avoids the interruption of the AI analysis service required by the terminal during the switching process, and ensures the continuity and accuracy of the mobile terminal AI analysis service.
  • Fig. 14 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in Fig. 14, the model reasoning method is used in distributed wireless access network equipment, including the following steps.
  • step S121 in response to receiving the model analysis subscription request sent by the terminal, send a model analysis subscription request to the control radio access network device.
  • the model analysis subscription request is used to obtain the first model from the OAM.
  • the first model includes a first number of model partitions.
  • the terminal initiates a model analysis subscription request to the connected distributed wireless access network device, wherein the model analysis subscription request includes a terminal identifier, analysis request type, and access location information.
  • the distributed radio access network device After receiving the model analysis subscription request, the distributed radio access network device sends the model analysis subscription request to the control radio access network device.
  • Fig. 15 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in Fig. 15, the model reasoning method is used in distributed radio access network equipment, including the following steps.
  • step S131 the model inference data request sent by the control radio access network device is received.
  • step S132 the model reasoning data is obtained from the terminal and sent to the control radio access network device.
  • the model reasoning data request is used to acquire model reasoning data.
  • the distributed radio access network device After receiving the model reasoning data request, the distributed radio access network device obtains the model reasoning data from the terminal, and sends the model reasoning data to the controlling radio access network device.
  • Fig. 16 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in FIG. 16, the model reasoning method is used in a distributed radio access network device, and includes the following steps.
  • step S141 in response to the fact that the distributed radio access network device is the first distributed radio access network device, a first reasoning result sent by the first control radio access network device is received.
  • step S142 the first reasoning result is sent to the terminal.
  • the first distributed radio access network device accessed by the terminal receives the first reasoning result sent by the first controlling radio access network device, and sends the first reasoning result to the terminal, so that the terminal Adjust execution strategy.
  • Fig. 17 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in FIG. 17 , the model reasoning method is used in a distributed radio access network device, and includes the following steps.
  • step S151 in response to the fact that the distributed radio access network device is the first distributed radio access network device, performance data sent by the terminal is received.
  • the performance data is real performance data after the terminal adjusts the execution strategy based on the first model
  • step S152 the performance data is sent to the first control radio access network device.
  • the terminal sends real performance data (also referred to as performance feedback data) obtained after adjusting the execution policy to the accessed first distributed radio access network device. And send the performance data to the first control radio access network device corresponding to the first distributed radio access network device.
  • real performance data also referred to as performance feedback data
  • Fig. 18 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in FIG. 18, the model reasoning method is used in a distributed radio access network device, and includes the following steps.
  • step S161 in response to the fact that the wireless access network device is the second distributed wireless access network device, if a model analysis subscription request resent by the terminal is received, determine to send the control information corresponding to the second distributed wireless access network device The radio access network device sends a model analysis subscription request.
  • the second distributed radio access network device is the distributed radio access network device re-accessed by the terminal after switching the distributed radio access network device.
  • the terminal After the terminal switches the accessed distributed wireless access network device, it re-sends a model analysis subscription request to the accessed distributed wireless access network device.
  • the model analysis is sent to its corresponding controlling radio access network device Subscribe request.
  • the OAM, the interaction process between the control radio access network device and the distributed radio access network device is taken as an example for further description.
  • the control radio access network device may be gNB-CU, and the distributed radio access network device may be gNB-DU.
  • Fig. 19 is a flow chart showing a model reasoning method according to an exemplary embodiment. As shown in Figure 19, the following steps are included:
  • step 1 the terminal initiates a model analysis subscription request.
  • step 2 the gNB-CU initiates a model subscription request to the OAM.
  • OAM performs model selection and model segmentation, allocation and distribution of model segmentation blocks according to the model subscription request information, and sends model segmentation block allocation information to gNB-CUs participating in joint reasoning.
  • step 4 the gNB-CU accessed by the terminal initiates a model reasoning data request, and the relevant network elements collect and process the data, and send the data to the gNB-CU.
  • step 5 the gNB-CU accessed by the terminal uses the model inference data to complete the inference of the first model segmentation block, and sends the inference intermediate result to the gNB-CU where the next model segmentation block is located according to the model segmentation block allocation information.
  • step 6 the gNB-CU responsible for the inference task of the last model segmentation block obtains the first inference result and then sends the first inference result to the gNB-CU accessed by the terminal according to the model segmentation block allocation information.
  • step 7 the gNB-CU accessed by the terminal sends the first reasoning result to the terminal, and the terminal uses the first reasoning result to perform corresponding policy adjustment.
  • step 8 the gNB-CU collects model performance data and terminal performance feedback data and reports them to the OAM.
  • the OAM trains and optimizes the model, and sends the updated model to the gNB-CU.
  • Fig. 20 is a flow chart of terminal switching in a model reasoning method according to an exemplary embodiment. As shown in Figure 20, the following steps are included:
  • step 1 the terminal re-initiates an analysis subscription request.
  • step 2 the gNB-CU and the OAM update the model analysis subscription request information of the terminal.
  • step 3 if the current reasoning task is not completed when the terminal is handed over, the gNB-CU completes the reasoning task and sends the first reasoning result to the gNB-DU currently connected to the terminal, and the gNB-DU sends the first reasoning result sent to the terminal.
  • step 4 after the terminal handover is completed, the newly accessed gNB-DU is responsible for completing relevant data collection and data forwarding tasks.
  • Fig. 21 is a flow chart of terminal switching in a model reasoning method according to an exemplary embodiment. As shown in Figure 21, the following steps are included:
  • step 1 the terminal re-initiates the model analysis subscription request.
  • step 2 the gNB-CU newly accessed by the terminal sends model subscription request information to the OAM.
  • step 3 the OAM updates the terminal's analysis subscription request, and sends the updated model analysis subscription request information to the terminal's source gNB-CU.
  • step 4 if the current inference task is not completed when the terminal switches, the source gNB-CU completes the inference task, and sends the first inference result to the gNB-CU newly accessed by the terminal.
  • step 5 the source gNB-CU updates the terminal analysis request information, and is no longer responsible for the analysis request related tasks of the terminal.
  • step 6 the gNB-CU newly accessed by the terminal sends the first inference result to the gNB-DU newly accessed by the terminal, and the gNB-DU sends the first inference result to the terminal.
  • OAM performs model selection and segmentation again according to the model subscription request sent by the gNB-CU newly accessed by the terminal, and sends the model segmentation block allocation information to the gNB-CU participating in the joint reasoning.
  • Fig. 22 is a schematic diagram of a protocol and interface of a model reasoning method according to an exemplary embodiment. As shown in Figure 22, it mainly involves the terminal provided by the embodiment of the present invention, the gNB-DU accessed by the terminal, the gNB-CU accessed by the terminal, and other gNB-CUs participating in joint reasoning (gNB-CU(1) ⁇ gNB- CU(N)) and OAM. details as follows:
  • the terminal sends the model analysis subscription request signaling to the gNB-DU, indicating to initiate a model analysis subscription request to the receiver.
  • the gNB-DU sends analysis subscription request signaling to the gNB-CU.
  • the gNB-CU generates model subscription request information according to its own AI processing capability and analysis subscription request information.
  • the gNB-CU sends the model subscription request signaling to the OAM, instructing to initiate a model subscription request to the receiver.
  • OAM selects the first model that meets the analysis request according to the model subscription information, and divides the model into several blocks according to the AI processing capability information.
  • 5a The OAM sends the first model segmentation block and the allocation information of the model segmentation block to the gNB-CU accessed by the terminal.
  • OAM sends the rest of the model segmentation blocks and the allocation information of the model segmentation blocks to other gNB-CUs participating in the joint inference, and instructs to send the allocation information of the model segmentation blocks.
  • the gNB-CU sends the model reasoning data collection request signaling to the gNB-DU, instructing to initiate a model reasoning data collection request to the receiver.
  • the gNB-DU, the terminal, and the gNB-CU collect data respectively according to the model reasoning data collection request, and send them to the gNB-CU.
  • the gNB-CU uses the collected inference data to perform partial model inference on the first model segmentation block to obtain inference intermediate information.
  • the gNB-CU sends the inference intermediate information to the gNB-CU (1) where the next model segmentation block is located. 10.
  • NB-CU (1) performs partial model inference on the model segmentation block, and obtains inference intermediate information results.
  • the gNB-CU (1) sends the inference intermediate information to the gNB-CU where the next model segmentation block is located, until the gNB-CU (N) where the last model segmentation block is located receives all the inference intermediate information.
  • the gNB-CU(N) performs partial model inference on the last model segmentation block to obtain the first inference result. 13a.
  • the gNB-CU(N) sends the first reasoning result to the gNB-CU accessed by the terminal. 13b.
  • the gNB-CU sends the first reasoning result to the gNB-DU accessed by the terminal. 13c.
  • the gNB-DU sends the first reasoning result to the terminal. 14.
  • the terminal makes corresponding policy adjustments according to the first reasoning result.
  • 15a. The terminal sends the performance feedback data to the gNB-DU.
  • 15b. The gNB-DU sends the performance feedback data to the gNB-CU.
  • the gNB-CU compares the first inference result with the real data to obtain model performance data.
  • the gNB-CU processes model performance data and terminal performance feedback data.
  • the gNB-CU sends the model performance data and terminal performance feedback data to the OAM. 19.
  • OAM trains and optimizes the model based on model performance data and performance feedback data.
  • 20. The OAM sends the updated model parameters to the gNB-CU.
  • Fig. 23 is a schematic diagram of the protocol and interface of AI analysis task delivery when the terminal switches under the same gNB-CU in a model-free reasoning method according to an exemplary embodiment. As shown in Fig. 23 , it mainly relates to embodiments of the present disclosure Provided terminal, terminal source gNB-DU (gNB-DU 1), terminal newly accessed gNB-DU (gNB-DU 2), terminal accessed gNB-CU and OAM. details as follows:
  • the terminal sends the model analysis subscription request signaling to the gNB-DU 2.
  • the gNB-DU 2 sends the model analysis subscription request signaling to the gNB-CU. 2.
  • the gNB-CU updates the analysis request information of the terminal. 3.
  • the gNB-CU sends the analysis subscription update request to the OAM. 4.
  • the OAM updates the analysis request information of the terminal. 5. If the current inference task is not completed when the terminal is switched, the gNB-CU continues to complete the current inference task to obtain the first inference result. 6a.
  • the gNB-CU sends the first reasoning result to the gNB-DU 2 according to the access location in the analysis subscription update request. 6b.
  • the gNB-DU 2 sends the first reasoning result to the terminal. 7. After the terminal switching is completed, gNB-DU 2 is responsible for the data collection and data forwarding tasks related to the terminal analysis request.
  • Fig. 24 is a schematic diagram of the protocol and interface of the AI analysis task delivery when the terminal is switched across gNB-CUs in a model-free reasoning method according to an exemplary embodiment.
  • the terminal the source gNB-DU of the terminal (gNB-DU 1), the gNB-DU newly accessed by the terminal (gNB-DU 3), the source gNB-CU of the terminal (gNB-CU 1), and the gNB-DU newly accessed by the terminal CU (gNB-CU 2), other gNB-CUs (gNB-CU (1) ⁇ gNB-CU (N)) participating in joint reasoning, and OAM. details as follows:
  • the terminal sends the model analysis subscription request signaling to gNB-DU 3. 1b.
  • gNB-DU 3 sends the model analysis subscription request signaling to gNB-CU 2.
  • the gNB-CU 2 uses its own AI processing capability and analysis Subscribe request information, generate model subscription request information.
  • the gNB-CU 2 sends the model subscription request signaling to the OAM.
  • the OAM updates the analysis subscription request information of the current terminal according to the model subscription request.
  • the OAM sends the analysis subscription update request to gNB-CU 1, indicating to initiate an analysis subscription update request to the receiver. 6. If the current inference task is not completed when the terminal is switched, gNB-CU 1 will continue to complete the current inference task and obtain the first inference result. 7.
  • gNB-CU 1 updates the terminal analysis request information, and is no longer responsible for the analysis request related tasks of the terminal.
  • 8a.gNB-CU 1 sends the first reasoning result to gNB-CU 2 according to the access location in the analysis subscription update request.
  • 8b.gNB-CU 2 sends the first reasoning result to gNB-DU 3.
  • 8c.gNB-DU 3. Send the first inference result to the terminal.
  • OAM re-selects and splits models based on the AI processing capability information in the model subscription request.
  • 10a. OAM sends the first model segmentation block and model segmentation block allocation information to gNB-CU 2.
  • 10b. OAM sends the remaining model segmentation blocks and model segmentation block allocation information to gNB-CU for auxiliary reasoning.
  • the embodiment of the present disclosure also provides a model reasoning device.
  • the model reasoning apparatus provided by the embodiments of the present disclosure includes corresponding hardware structures and/or software modules for performing various functions.
  • the embodiments of the present disclosure can be implemented in the form of hardware or a combination of hardware and computer software. Whether a certain function is executed by hardware or computer software drives hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art may use different methods to implement the described functions for each specific application, but such implementation should not be regarded as exceeding the scope of the technical solutions of the embodiments of the present disclosure.
  • Fig. 25 is a block diagram of a model reasoning device according to an exemplary embodiment.
  • the model reasoning apparatus 100 is applied to an OAM entity, and includes a determination module 101 and a sending module 102 .
  • the determining module 101 is configured to determine a first model corresponding to the model subscription request information in response to receiving the model subscription request information sent by the control radio access network device.
  • the sending module 102 is configured to divide the first model to obtain a first number of model segmentation blocks, and distribute the first number of model segmentation blocks to the first number of control radio access network devices.
  • each model segmentation block in the first number of model segmentation blocks corresponds to allocation information.
  • the assignment information includes an inference order of the first number of model partition blocks, and a control radio access network device corresponding to each model partition block.
  • the first number of control radio access network devices includes a first control radio access network device, and the first control radio access network device is a control radio access network device accessed by a terminal.
  • the sending module 102 is configured to, among the control radio access network devices adjacent to the first control radio access network device, determine a plurality of auxiliary control radio access network devices; among the plurality of auxiliary control radio access network devices, Based on the computing power occupancy state and load of each auxiliary control wireless access network device, determine a second number of control wireless access network devices; the second number of control wireless access network devices are other than the first control wireless access device Other control radio access network devices in the first number; based on the reasoning sequence of the first number of model segmentation blocks, sending the first model segmentation block to the first control radio access network device, and dividing the remaining number of model segments The blocks are distributed to the second number of controlling radio access network devices.
  • the model reasoning apparatus further includes: a receiving module 103 .
  • the receiving module 103 shown is configured to receive model performance update data sent by the first control radio access network device.
  • the first model is updated based on the model performance update data, the updated model parameters of the first model are determined, and the updated model parameters of the first model are sent to the first control radio access network device.
  • the receiving module 103 is further configured to update the distributed radio access network device accessed by the terminal in response to receiving the first model analysis subscription update request.
  • the first model analysis subscription update request instructs the terminal to switch the distributed radio access network equipment, and not to switch the control radio access network equipment.
  • the second model analysis subscription update request update the distributed radio access network device accessed by the terminal, and re-segment the first model.
  • the second model analyzes the subscription update request and instructs the terminal to switch the distributed radio access network equipment, and switch and control the radio access network equipment.
  • Fig. 26 is a block diagram of a model reasoning device according to an exemplary embodiment.
  • the model reasoning apparatus 200 is applied to control radio access network equipment, and includes a sending module 201 and a receiving module 202 .
  • the sending module 201 is configured to process the model analysis subscription request to obtain model subscription request information in response to receiving the model analysis subscription request sent by the distributed wireless access network device, and send the model subscription request information to the OAM.
  • the receiving module 202 is configured to receive the model segmentation blocks sent by the OAM.
  • the model segmentation block is a model segmentation block determined by segmenting the first model.
  • the first model is determined by the OAM based on the model subscription request information.
  • the sending module 201 is further configured to send a model reasoning data request to the distributed radio access network device, where the model reasoning data request is used to acquire model reasoning data. Based on the model inference data, reasoning is performed on the model segmentation block, and the inference intermediate information of the model segmentation block is obtained.
  • the model segmentation blocks correspond to allocation information.
  • the assignment information includes an inference order of the first number of model partition blocks, and a control radio access network device corresponding to each model partition block.
  • the sending module 201 is further configured to send the reasoning intermediate information to the next controlling radio access network device based on the reasoning order in response to that the controlling radio access network device is not the last controlling radio access network device.
  • the model reasoning is completed, determine a first reasoning result corresponding to the first model, and send the first reasoning result to the first controlling radio access network device.
  • the first control radio access network equipment is the control radio access network equipment accessed by the terminal.
  • the sending module 201 is further configured to receive the first reasoning result in response to the controlling radio access network device being the first controlling radio access network device.
  • the first reasoning result is sent to the first distributed radio access network device, where the first distributed radio access network device is a distributed radio access network device accessed by the terminal.
  • the receiving module 202 is further configured to receive the performance data sent by the first distributed radio access network device, and the performance data is the terminal Actual performance data after adjusting the execution strategy based on the first model. Process the performance data to obtain model performance update data, and send the model performance update data to the OAM.
  • the sending module 201 is further configured to send model subscription request information to the OAM in response to the controlling radio access network device being the first controlling radio access network device.
  • the first control radio access network device is a control radio access network device corresponding to the first distributed wireless network device accessed by the terminal.
  • the sending module 201 is further configured to determine to resend the second request of the model analysis subscription request if the control radio access network device is the first control radio access network device.
  • the second distributed radio access network device is the distributed radio access network device that the terminal re-accesses after switching the distributed radio access network device. Send the first reasoning result to the second distributed radio access network device, and send a model subscription update request to the OAM.
  • the sending module 201 is further configured to determine to resend the second model analysis subscription request if the control radio access network device is the second control radio access network device, if the model analysis subscription request is re-received.
  • the distributed wireless access network device, and the second control wireless access network device that has received the model analysis subscription request again, the second control wireless access network device is the control wireless access network device corresponding to the second distributed wireless access network device network equipment.
  • the second distributed radio access network device is the distributed radio access network device that the terminal re-accesses after switching the distributed radio access network device. Send the first reasoning result to the second control radio access network device, and send a model subscription update request to the OAM.
  • Fig. 27 is a block diagram of a model reasoning device according to an exemplary embodiment.
  • the model reasoning apparatus 300 is applied to a distributed radio access network device, and includes a sending module 301 .
  • the sending module 301 is configured to send the model analysis subscription request to the control radio access network device in response to receiving the model analysis subscription request sent by the terminal.
  • the model analysis subscription request is used to obtain the first model from the OAM.
  • the first model includes a first number of model partitions.
  • the device further includes: a receiving module 302 .
  • the receiving module 302 is configured to receive a model reasoning data request sent by the control radio access network device, where the model reasoning data request is used to acquire model reasoning data. Obtain model reasoning data from the terminal and send it to the control radio access network device.
  • the receiving module 302 is further configured to receive the first reasoning result sent by the first control radio access network device in response to the distributed radio access network device being the first distributed radio access network device. Send the first reasoning result to the terminal.
  • the receiving module 302 is further configured to receive the performance data sent by the terminal in response to the fact that the distributed radio access network device is the first distributed radio access network device.
  • the performance data is real performance data after the terminal adjusts the execution strategy based on the first model. Send the performance data to the first control radio access network device.
  • the receiving module 302 is further configured to determine whether the wireless access network device is the second distributed wireless access network device, if a model analysis subscription request resent by the terminal is received, to determine the The control radio access network device corresponding to the type radio access network device sends a model analysis subscription request.
  • the second distributed radio access network device is a distributed radio access network device re-accessed by the terminal after switching the distributed radio access network device.
  • Fig. 28 is a block diagram of an apparatus 400 for model reasoning according to an exemplary embodiment.
  • the apparatus 400 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, and the like.
  • device 400 may include one or more of the following components: processing component 402, memory 404, power component 406, multimedia component 408, audio component 410, input/output (I/O) interface 412, sensor component 414, and communication component 416 .
  • the processing component 402 generally controls the overall operations of the device 400, such as those associated with display, telephone calls, data communications, camera operations, and recording operations.
  • the processing component 402 may include one or more processors 420 to execute instructions to complete all or part of the steps of the above method. Additionally, processing component 402 may include one or more modules that facilitate interaction between processing component 402 and other components. For example, processing component 402 may include a multimedia module to facilitate interaction between multimedia component 408 and processing component 402 .
  • the memory 404 is configured to store various types of data to support operations at the device 400 . Examples of such data include instructions for any application or method operating on device 400, contact data, phonebook data, messages, pictures, videos, and the like.
  • the memory 404 can be implemented by any type of volatile or non-volatile storage device or their combination, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), Magnetic Memory, Flash Memory, Magnetic or Optical Disk.
  • SRAM static random access memory
  • EEPROM electrically erasable programmable read-only memory
  • EPROM erasable Programmable Read Only Memory
  • PROM Programmable Read Only Memory
  • ROM Read Only Memory
  • Magnetic Memory Flash Memory
  • Magnetic or Optical Disk Magnetic Disk
  • Power component 406 provides power to various components of device 400 .
  • Power components 406 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for device 400 .
  • the multimedia component 408 includes a screen that provides an output interface between the device 400 and the user.
  • the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from a user.
  • the touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensor may not only sense a boundary of a touch or swipe action, but also detect duration and pressure associated with the touch or swipe action.
  • the multimedia component 408 includes a front camera and/or a rear camera. When the device 400 is in an operation mode, such as a shooting mode or a video mode, the front camera and/or the rear camera can receive external multimedia data. Each front camera and rear camera can be a fixed optical lens system or have focal length and optical zoom capability.
  • the audio component 410 is configured to output and/or input audio signals.
  • the audio component 410 includes a microphone (MIC), which is configured to receive external audio signals when the device 400 is in operation modes, such as call mode, recording mode and voice recognition mode. Received audio signals may be further stored in memory 404 or sent via communication component 416 .
  • the audio component 410 also includes a speaker for outputting audio signals.
  • the I/O interface 412 provides an interface between the processing component 402 and a peripheral interface module.
  • the peripheral interface module may be a keyboard, a click wheel, a button, and the like. These buttons may include, but are not limited to: a home button, volume buttons, start button, and lock button.
  • Sensor assembly 414 includes one or more sensors for providing status assessments of various aspects of device 400 .
  • the sensor component 414 can detect the open/closed state of the device 400, the relative positioning of components, such as the display and keypad of the device 400, and the sensor component 414 can also detect a change in the position of the device 400 or a component of the device 400 , the presence or absence of user contact with the device 400 , the device 400 orientation or acceleration/deceleration and the temperature change of the device 400 .
  • the sensor assembly 414 may include a proximity sensor configured to detect the presence of nearby objects in the absence of any physical contact.
  • Sensor assembly 414 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications.
  • the sensor component 414 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor or a temperature sensor.
  • the communication component 416 is configured to facilitate wired or wireless communication between the apparatus 400 and other devices.
  • the device 400 can access wireless networks based on communication standards, such as WiFi, 2G or 3G, or a combination thereof.
  • the communication component 416 receives broadcast signals or broadcast related information from an external broadcast management system via a broadcast channel.
  • the communication component 416 also includes a near field communication (NFC) module to facilitate short-range communication.
  • NFC near field communication
  • the NFC module may be implemented based on Radio Frequency Identification (RFID) technology, Infrared Data Association (IrDA) technology, Ultra Wide Band (UWB) technology, Bluetooth (BT) technology and other technologies.
  • RFID Radio Frequency Identification
  • IrDA Infrared Data Association
  • UWB Ultra Wide Band
  • Bluetooth Bluetooth
  • apparatus 400 may be programmed by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable A gate array (FPGA), controller, microcontroller, microprocessor or other electronic component implementation for performing the methods described above.
  • ASICs application specific integrated circuits
  • DSPs digital signal processors
  • DSPDs digital signal processing devices
  • PLDs programmable logic devices
  • FPGA field programmable A gate array
  • controller microcontroller, microprocessor or other electronic component implementation for performing the methods described above.
  • non-transitory computer-readable storage medium including instructions, such as the memory 404 including instructions, which can be executed by the processor 420 of the device 400 to implement the above method.
  • the non-transitory computer readable storage medium may be ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, and the like.
  • Fig. 29 is a block diagram of an apparatus 500 for model reasoning according to an exemplary embodiment.
  • the apparatus 500 may be provided as a server. 29, apparatus 500 includes processing component 522, which further includes one or more processors, and memory resources represented by memory 532 for storing instructions executable by processing component 522, such as application programs.
  • the application program stored in memory 532 may include one or more modules each corresponding to a set of instructions.
  • the processing component 522 is configured to execute instructions to perform the above method.
  • Device 500 may also include a power component 526 configured to perform power management of device 500 , a wired or wireless network interface 550 configured to connect device 500 to a network, and an input-output (I/O) interface 558 .
  • the device 500 can operate based on an operating system stored in the memory 532, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or the like.
  • “plurality” in the present disclosure refers to two or more, and other quantifiers are similar thereto.
  • “And/or” describes the association relationship of associated objects, indicating that there may be three types of relationships, for example, A and/or B may indicate: A exists alone, A and B exist simultaneously, and B exists independently.
  • the character “/” generally indicates that the contextual objects are an “or” relationship.
  • the singular forms “a”, “said” and “the” are also intended to include the plural unless the context clearly dictates otherwise.
  • first, second, etc. are used to describe various information, but the information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another, and do not imply a specific order or degree of importance. In fact, expressions such as “first” and “second” can be used interchangeably.
  • first information may also be called second information, and similarly, second information may also be called first information.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Software Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Artificial Intelligence (AREA)
  • Microelectronics & Electronic Packaging (AREA)
  • Physics & Mathematics (AREA)
  • Condensed Matter Physics & Semiconductors (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Hardware Design (AREA)
  • Power Engineering (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本公开是关于一种模型推理方法、模型推理装置及存储介质。其中，模型推理方法，应用于操作维护管理OAM实体，所述方法包括：响应于接收到控制无线接入网设备发送的模型订阅请求信息，确定与所述模型订阅请求信息对应的第一模型；将所述第一模型进行分割，得到第一数量的模型分割块，并将所述第一数量的模型分割块分发至第一数量的控制无线接入网设备。通过本公开可以能够有效提高模型推理效率，减小推理时延，同时有助于均衡网络负载。

Description

一种模型推理方法、模型推理装置及存储介质 技术领域
本公开涉及无线通信技术领域，尤其涉及一种模型推理方法、模型推理装置及存储介质。
背景技术
在新一代通信技术中，网络智能化合自动化相关行为的决策，需要采用人工智能和机器学习获取大量可用数据，包括终端和网络侧设备采集的数据。基于所述数据利用机器学习算法进行推理、训练，提取出不同级别的相关模型。
在相关技术中，终端向通过无线接入设备向操作维护管理(Operation Administration and Maintenance，OAM)网元请求订阅模型，其模型的推理工作全部由操作维护管理(Operation Administration and Maintenance，OAM)网元来完成，模型推理工作时需要将所有模型推理数据上传至OAM，同时，OAM还需要根据模型推理数据对模型进行训练。因此，当OAM同时接收到多个订阅模型请求时，OAM无法满足多个订阅请求并提供模型推理结果，会导致模型推理结果反馈时延增大，并且降低系统工作效率。
发明内容
为克服相关技术中存在的问题，本公开提供一种模型推理方法、模型推理装置及存储介质。
根据本公开实施例的第一方面，提供一种模型推理方法，应用于操作维护管理OAM实体，所述方法包括：
响应于接收到控制无线接入网设备发送的模型订阅请求信息，确定与所述模型订阅请求信息对应的第一模型；将所述第一模型进行分割，得到第一数量的模型分割块，并将所述第一数量的模型分割块分发至第一数量的控制无线接入网设备。
一种实施方式中，所述第一数量的模型分割块中每个模型分割块对应有分配信息；
所述分配信息包括第一数量的模型分割块的推理顺序，以及与每个模型分割块对应的所述控制无线接入网设备。
一种实施方式中，所述第一数量的控制无线接入网设备包括第一控制无线接入网设备，所述第一控制无线接入网设备为终端接入的控制无线接入网设备；
所述将所述第一数量的模型分割块分发至第一数量的控制无线接入网设备，包括：
在与所述第一控制无线接入网设备相邻的控制无线接入网设备中，确定多个辅助控制无线接入网设备；
在所述多个辅助控制无线接入网设备中，基于每个所述辅助控制无线接入网设备的算力占用状态和负载，确定第二数量的控制无线接入网设备；所述第二数量的控制无线接入网设备为除第一控制无线接入设备以外第一数量中其他的控制无线接入网设备；
基于第一数量的模型分割块的推理顺序，将第一个模型分割块发送至所述第一控制无线接入网设备，并将剩余数量的模型分割块分发至所述第二数量的控制无线接入网设备。
一种实施方式中，所述模型推理方法还包括：
接收第一控制无线接入网设备发送的模型性能更新数据；基于所述模型性能更新数据更新所述第一模型，确定所述第一模型更新后的模型参数，并向所述第一控制无线接入网设备发送所述第一模型更新后的模型参数。
一种实施方式中，所述模型推理方法还包括：
响应于接收到第一模型分析订阅更新请求，更新终端接入的分布式无线接入网设备；其中，所述第一模型分析订阅更新请求指示终端切换分布式无线接入网设备，且不切换控制无线接入网设备；
或
响应于接收到第二模型分析订阅更新请求，更新终端接入的分布式无线接入网设备，并重新对所述第一模型进行分割；其中，所述第二模型分析订阅更新请求指示终端切换分布式无线接入网设备，并切换控制无线接入网设备。
根据本公开实施例的第二方面，提供一种模型推理方法，应用于控制无线接入网设备，所述方法包括：
响应于接收到分布式无线接入网设备发送的模型分析订阅请求，对所述模型分析订阅请求进行处理得到模型订阅请求信息，并向OAM发送所述模型订阅请求信息；接收OAM发送的模型分割块；所述模型分割块为分割第一模型确定的模型分割块；所述第一模型为OAM基于所述模型订阅请求信息确定的。
一种实施方式中，所述向OAM发送模型订阅请求信息之后，所述方法还包括：
向分布式无线接入网设备发送模型推理数据请求，所述模型推理数据请求用于获取模型推理数据；基于所述模型推理数据对所述模型分割块进行推理，得到模型分割块的推理中间信息。
一种实施方式中，所述模型分割块对应有分配信息；所述分配信息包括第一数量的模型分割块的推理顺序，以及与每个模型分割块对应的所述控制无线接入网设备；
所述模型推理方法还包括：
响应于所述控制无线接入网设备不是最后一个控制无线接入网设备，基于所述推理顺 序，将推理中间信息发送至下一个控制无线接入网设备；响应于所述控制无线接入网设备为最后一个控制无线接入网设备，模型推理完成后，确定与第一模型对应的第一推理结果，将所述第一推理结果发送至第一控制无线接入网设备，所述第一控制无线接入网设备为终端接入的控制无线接入网设备。
一种实施方式中，所述方法还包括：
响应于所述控制无线接入网设备为第一控制无线接入网设备，接收所述第一推理结果；向第一分布式无线接入网设备发送所述第一推理结果，所述第一分布式无线接入网设备为终端接入的分布式无线接入网设备。
一种实施方式中，所述将所述第一推理结果发送至第一分布式无线接入网设备之后，所述模型推理方法还包括：
接收第一分布式无线接入网设备发送的性能数据，所述性能数据为终端基于第一模型调整执行策略后的真实性能数据；对所述性能数据进行处理，得到模型性能更新数据，并向OAM发送所述模型性能更新数据。
一种实施方式中，向OAM发送模型订阅请求信息，包括：
响应于所述控制无线接入网设备为第一控制无线接入网设备，向OAM发送模型订阅请求信息；其中，所述第一控制无线接入网设备为终端接入的第一分布式无线网络设备对应的控制无线接入网设备。
一种实施方式中，所述模型推理方法还包括：
响应于所述控制无线接入网设备为第一控制无线接入网设备，若重新接收到模型分析订阅请求，确定重新发送所述模型分析订阅请求的第二分布式无线接入网设备；所述第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备；将第一推理结果发送至第二分布式无线接入网设备，并向OAM发送模型订阅更新请求。
一种实施方式中，所述模型推理方法还包括：
响应于所述控制无线接入网设备为第二控制无线接入网设备，若重新接收到模型分析订阅请求，确定重新发送所述模型分析订阅请求的第二分布式无线接入网设备，以及重新接收到模型分析订阅请求的第二控制无线接入网设备，所述第二控制无线接入网设备为第二分布式无线接入网设备对应的控制无线接入网设备；所述第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备；将第一推理结果发送至第二控制无线接入网设备，并向OAM发送模型订阅更新请求。
根据本公开实施例的第三方面，提供一种模型推理方法，应用于分布式无线接入网设 备，所述方法包括：
响应于接收到终端发送的模型分析订阅请求，向控制无线接入网设备发送所述模型分析订阅请求；其中，所述模型分析订阅请求用于向OAM获取第一模型；所述第一模型包括第一数量的模型分割块。
一种实施方式中，所述方法还包括：
接收控制无线接入网设备发送的模型推理数据请求，所述模型推理数据请求用于获取模型推理数据；向终端获取模型推理数据，并发送至控制无线接入网设备。
一种实施方式中，所述方法还包括：
响应于所述分布式无线接入网设备为第一分布式无线接入网设备，接收第一控制无线接入网设备发送的第一推理结果；将所述第一推理结果发送至终端。
一种实施方式中，所述将所述第一推理结果发送至终端之后，所述方法还包括：
响应于所述分布式无线接入网设备为第一分布式无线接入网设备，接收终端发送的性能数据；所述性能数据为终端基于第一模型调整执行策略后的真实性能数据；向第一控制无线接入网设备发送所述性能数据。
一种实施方式中，所述方法还包括：
响应于所述无线接入网设备为第二分布式无线接入网设备，若接收到终端重新发送的模型分析订阅请求，确定向与所述第二分布式无线接入网络设备对应的控制无线接入网设备发送模型分析订阅请求；其中，所述第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备。
根据本公开实施例的第四方面，提供一种模型推理装置，应用于操作维护管理OAM实体，所述装置包括：
确定模块，用于响应于接收到控制无线接入网设备发送的模型订阅请求信息，确定与所述模型订阅请求信息对应的第一模型；发送模块，用于将所述第一模型进行分割，得到第一数量的模型分割块，并将所述第一数量的模型分割块分发至第一数量的控制无线接入网设备。
一种实施方式中，所述第一数量的模型分割块中每个模型分割块对应有分配信息；
所述分配信息包括第一数量的模型分割块的推理顺序，以及与每个模型分割块对应的所述控制无线接入网设备。
一种实施方式中，所述第一数量的控制无线接入网设备包括第一控制无线接入网设备，所述第一控制无线接入网设备为终端接入的控制无线接入网设备；
发送模块，用于：
在与所述第一控制无线接入网设备相邻的控制无线接入网设备中，确定多个辅助控制无线接入网设备；
在所述多个辅助控制无线接入网设备中，基于每个所述辅助控制无线接入网设备的算力占用状态和负载，确定第二数量的控制无线接入网设备；所述第二数量的控制无线接入网设备为除第一控制无线接入设备以外第一数量中其他的控制无线接入网设备；
基于第一数量的模型分割块的推理顺序，将第一个模型分割块发送至所述第一控制无线接入网设备，并将剩余数量的模型分割块分发至所述第二数量的控制无线接入网设备。
一种实施方式中，所述模型推理装置还包括：接收模块；
所示接收模块用于，接收第一控制无线接入网设备发送的模型性能更新数据；基于所述模型性能更新数据更新所述第一模型，确定所述第一模型更新后的模型参数，并向所述第一控制无线接入网设备发送所述第一模型更新后的模型参数。
一种实施方式中，所述接收模块还用于：
响应于接收到第一模型分析订阅更新请求，更新终端接入的分布式无线接入网设备；其中，所述第一模型分析订阅更新请求指示终端切换分布式无线接入网设备，且不切换控制无线接入网设备；
或
响应于接收到第二模型分析订阅更新请求，更新终端接入的分布式无线接入网设备，并重新对所述第一模型进行分割；其中，所述第二模型分析订阅更新请求指示终端切换分布式无线接入网设备，并切换控制无线接入网设备。
根据本公开实施例的第五方面，提供一种模型推理装置，应用于控制无线接入网设备，所述装置包括：
发送模块，用于响应于接收到分布式无线接入网设备发送的模型分析订阅请求，对所述模型分析订阅请求进行处理得到模型订阅请求信息，并向OAM发送所述模型订阅请求信息；接收模块，用于接收OAM发送的模型分割块；所述模型分割块为分割第一模型确定的模型分割块；所述第一模型为OAM基于所述模型订阅请求信息确定的。
一种实施方式中，所述发送模块还用于：
向分布式无线接入网设备发送模型推理数据请求，所述模型推理数据请求用于获取模型推理数据；基于所述模型推理数据对所述模型分割块进行推理，得到模型分割块的推理中间信息。
一种实施方式中，所述模型分割块对应有分配信息；所述分配信息包括第一数量的模型分割块的推理顺序，以及与每个模型分割块对应的所述控制无线接入网设备；
所述发送模块还用于：
响应于所述控制无线接入网设备不是最后一个控制无线接入网设备，基于所述推理顺序，将推理中间信息发送至下一个控制无线接入网设备；响应于所述控制无线接入网设备为最后一个控制无线接入网设备，模型推理完成后，确定与第一模型对应的第一推理结果，将所述第一推理结果，发送至第一控制无线接入网设备，所述第一控制无线接入网设备为终端接入的控制无线接入网设备。
一种实施方式中，所述发送模块还用于：
响应于所述控制无线接入网设备为第一控制无线接入网设备，接收所述第一推理结果；向第一分布式无线接入网设备发送所述第一推理结果，所述第一分布式无线接入网设备为终端接入的分布式无线接入网设备。
一种实施方式中，所述将所述第一推理结果发送至第一分布式无线接入网设备之后，所述接收模块还用于：
接收第一分布式无线接入网设备发送的性能数据，所述性能数据为终端基于第一模型调整执行策略后的真实性能数据；对所述性能数据进行处理，得到模型性能更新数据，并向OAM发送所述模型性能更新数据。
一种实施方式中，所述发送模块还用于：
响应于所述控制无线接入网设备为第一控制无线接入网设备，向OAM发送模型订阅请求信息；其中，所述第一控制无线接入网设备为终端接入的第一分布式无线网络设备对应的控制无线接入网设备。
一种实施方式中，所述发送模块还用于：
响应于所述控制无线接入网设备为第一控制无线接入网设备，若重新接收到模型分析订阅请求，确定重新发送所述模型分析订阅请求的第二分布式无线接入网设备；所述第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备；将第一推理结果发送至第二分布式无线接入网设备，并向OAM发送模型订阅更新请求。
一种实施方式中，所述发送模块还用于：
响应于所述控制无线接入网设备为第二控制无线接入网设备，若重新接收到模型分析订阅请求，确定重新发送所述模型分析订阅请求的第二分布式无线接入网设备，以及重新接收到模型分析订阅请求的第二控制无线接入网设备，所述第二控制无线接入网设备为第二分布式无线接入网设备对应的控制无线接入网设备；所述第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备；将第一推理结果发 送至第二控制无线接入网设备，并向OAM发送模型订阅更新请求。
根据本公开实施例的第六方面，提供一种模型推理装置，应用于分布式无线接入网设备，所述装置包括：
发送模块，用于响应于接收到终端发送的模型分析订阅请求，向控制无线接入网设备发送所述模型分析订阅请求；其中，所述模型分析订阅请求用于向OAM获取第一模型；所述第一模型包括第一数量的模型分割块。
一种实施方式中，所述装置还包括：接收模块；
接收模块，用于接收控制无线接入网设备发送的模型推理数据请求，所述模型推理数据请求用于获取模型推理数据；向终端获取模型推理数据，并发送至控制无线接入网设备。
一种实施方式中，所述接收模块，还用于：
响应于所述分布式无线接入网设备为第一分布式无线接入网设备，接收第一控制无线接入网设备发送的第一推理结果；将所述第一推理结果发送至终端。
一种实施方式中，所述接收模块，还用于：
响应于所述分布式无线接入网设备为第一分布式无线接入网设备，接收终端发送的性能数据；所述性能数据为终端基于第一模型调整执行策略后的真实性能数据；向第一控制无线接入网设备发送所述性能数据。
一种实施方式中，所述接收模块，还用于：
响应于所述无线接入网设备为第二分布式无线接入网设备，若接收到终端重新发送的模型分析订阅请求，确定向与所述第二分布式无线接入网络设备对应的控制无线接入网设备发送模型分析订阅请求；其中，所述第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备。
根据本公开实施例的第七方面，提供一种模型推理装置，包括：
处理器；用于存储处理器可执行指令的存储器；其中，所述处理器被配置为：执行第一方面或第一方面任意一种实施方式中所述的模型推理方法，或执行第二方面或第二方面任意一种实施方式中所述的模型推理方法，或执行第三方面或第三方面任意一种实施方式中所述的模型推理方法。
根据本公开实施例的第八方面，提供一种非临时性计算机可读存储介质，当所述存储介质中的指令由移动终端的处理器执行时，使得移动终端能够执行第一方面或第一方面任意一种实施方式中所述的模型推理方法，或使得移动终端能够执行第二方面或第二方面任意一种实施方式中所述的模型推理方法，或使得移动终端能够执行第三方面或第三方面任意一种实施方式中所述的模型推理方法。
本公开的实施例提供的技术方案可以包括以下有益效果：通过本公开OAM将模型分割，并将模型分割块分发至不同的控制无线接入网设备，可以更好地开发无线接入网设备的AI处理能力，解决了无线接入网设备AI处理能力不足的问题，并有利于网络负载均衡。分利用本地AI处理能力，能够有效提高模型推理效率，减小推理时延，同时有助于均衡网络负载，为用户提供高效、便捷的AI分析服务。。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性和解释性的，并不能限制本公开。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，示出了符合本公开的实施例，并与说明书一起用于解释本公开的原理。
图1根据一示例性实施例示出的一种基本功能性框架结构示意图。
图2根据一示例性实施例示出的一种网络架构示意图。
图3是根据一示例性实施例示出的一种模型推理方法的流程图。
图4是根据一示例性实施例示出的又一种模型推理方法的流程图。
图5是根据一示例性实施例示出的又一种模型推理方法的流程图。
图6是根据一示例性实施例示出的又一种模型推理方法的流程图。
图7是根据一示例性实施例示出的又一种模型推理方法的流程图。
图8是根据一示例性实施例示出的又一种模型推理方法的流程图。
图9是根据一示例性实施例示出的又一种模型推理方法的流程图。
图10是根据一示例性实施例示出的又一种模型推理方法的流程图。
图11是根据一示例性实施例示出的又一种模型推理方法的流程图。
图12是根据一示例性实施例示出的又一种模型推理方法的流程图。
图13是根据一示例性实施例示出的又一种模型推理方法的流程图。
图14是根据一示例性实施例示出的又一种模型推理方法的流程图。
图15是根据一示例性实施例示出的又一种模型推理方法的流程图。
图16是根据一示例性实施例示出的又一种模型推理方法的流程图。
图17是根据一示例性实施例示出的又一种模型推理方法的流程图。
图18是根据一示例性实施例示出的又一种模型推理方法的流程图。
图19是根据一示例性实施例示出的又一种模型推理方法的流程图。
图20是根据一示例性实施例示出的一种模型推理方法中终端切换的流程图。
图21是根据一示例性实施例示出的一种模型推理方法中终端切换的流程图。
图22是根据一示例性实施例示出的一种模型推理方法的协议和接口原理图。
图23是根据一示例性实施例示出的一种无模型推理方法中终端在同一gNB-CU下切换时AI分析任务交付的协议和接口原理图。
图24是根据一示例性实施例示出的一种无模型推理方法中终端跨gNB-CU切换时AI分析任务交付的协议和接口原理图。
图25是根据一示例性实施例示出的一种模型推理装置框图。
图26是根据一示例性实施例示出的又一种模型推理装置框图。
图27是根据一示例性实施例示出的又一种模型推理装置框图。
图28是根据一示例性实施例示出的一种模型推理装置的框图。
图29是根据一示例性实施例示出的又一种模型推理装置的框图。
具体实施方式
这里将详细地对示例性实施例进行说明，其示例表示在附图中。下面的描述涉及附图时，除非另有表示，不同附图中的相同数字表示相同或相似的要素。以下示例性实施例中所描述的实施方式并不代表与本公开相一致的所有实施方式。相反，它们仅是与如所附权利要求书中所详述的、本公开的一些方面相一致的装置和方法的例子。
在新一代智能化和自动化相关行为的决策，需要采用人工智能和机器学习获取大量可用数据，包括终端和网络侧采集的数据，并依靠机器学习算法对输入数据进行挖掘，并提取不同级别的相关模型，利用所获得的模型来驱动。为了实现大数据使能的人工智能无线网络，支持人工智能的无线网络框架，AI模块/平台的功能，输入和输出以及与无线网元的关系等关键技术是亟待研究的问题。
因此针对无线接入网设备(RAN)侧智能性优化的研究项目(Study Item)：NR和ENDC数据采集的增强研究。并对其设计准则、基本概念、适用案例、标准影响进行讨论。图1根据一示例性实施例示出的一种基本功能性框架结构示意图，如图1所示，作为初始架构，潜在的支持人工智能的无线网络架构。
其中，数据收集&准备单元(Data collection&preparation)：包含数据采集和数据预处理功能，数据采集可以在多个网元执行，提供的数据包括测量数据、反馈的性能数据和模型的性能数据等。
模型训练单元(Model Training)：通过运算和处理来迭代机器学习模型以得到更好的用于进行推理的模型，输入包括训练数据以及模型性能反馈等。
模型推理单元(Model inference)：使用训练好的机器学习模型来生成预测结果或者决策结果。
执行单元(Action)：利用模型推理结果制定并执行策略，并将执行后相关性能结果反馈给Data collection。
通过以上无线网络人工智能的架构，为提升无线网络终端业务体验注入了智能化动力。为了保持无线网络人工智能分析服务的连续性和准确性，提高无线网络人工智能的运行效率，需要对各个AI功能单元之间的交互进行进一步的规范与优化，使得无线网络人工智能架构具有更强的适应性和可扩展性。
在相关技术中，终端发起模型分析订阅请求后，均由OAM进行推理和训练。图2根据一示例性实施例示出的一种网络架构示意图。如图2所示，该系统包括终端，gNB-DU，gNB-CU和OAM，终端通过无线信道接入gNB-DU，多个gNB-DU通过F1接口接入gNB-CU，gNB-CU之间通过Xn接口连接。OAM主要负责承担支持AI的无线网络架构中模型训练功能单元的工作，负责模型训练和模型分割等工作；gNB-CU承担模型推理功能单元的工作，负责完成模型推理；gNB-DU则主要承担数据收集功能单元的工作，负责实时推理数据的收集，终端性能反馈数据收集等工作；终端承担动作执行功能单元的工作，负责依据模型推理结果做出相应的策略调整。
其中，终端负责执行(Action)功能单元的工作；基站分布式单元(next Generation Node B Distributed Unit，gNB-DU)负责转发终端的分析请求和推理结果，并执行数据收集(Data collection)功能单元的工作。基站控制单元(next Generation Node B Control Unit，gNB-CU)负责转发终端的分析请求和推理结果，并执行数据收集功能单元的工作。OAM负责执行模型训练和推理(Model training和Model inference)功能单元的工作。
其执行流程包括：终端向gNB-DU发起分析订阅请求，gNB-DU将该终端的分析订阅请求发送给gNB-CU，gNB-CU向OAM上报终端的分析订阅请求。OAM依据终端的分析订阅请求选择合适的模型，开启模型推理工作。OAM向gNB-CU发起模型推理数据请求，各级网元(gNB-CU、gNB-DU、终端)根据推理数据请求信息收集模型推理数据，进行数据处理后发送给OAM。OAM采用模型推理数据进行模型推理得到推理结果，将推理结果发送给gNB-CU，gNB-CU将推理结果发送给gNB-DU，gNB-DU将推理结果发送给终端，终端即可采用推理结果做出相应策略调整。
但是相关技术中，存在以下问题：
(1)模型推理工作全部由OAM网管来完成，需要将模型推理数据全部发送给OAM。这种方案将实时的模型推理数据从无线侧上传到网管，对数据的安全性造成了挑战，尤其是在模型推理数据包含终端业务数据的场景中，这种方案将会受到限制。
(2)完成模型推理工作时需要将所有模型推理数据上传至OAM，需要进行实时的数 据传输，在无线通信资源受限的情况下，这种方案会增大网络负载。
(3)模型推理时延包括模型推理数据上传到OAM带来的传输时延、模型推理的计算时延、OAM发送推理结果给终端带来的传输时延，其中第一部分时延较大，会造成推理结果反馈不及时，影响终端业务体验。
(4)将所有模型推理任务均卸到OAM，同时OAM还需要完成模型训练工作，当终端的分析订阅请求密集时，会产生OAM算力不够的情况，将会降低系统工作效率。
基于此本公开提供一种模型推理方法，将模型推理任务分配给不同的gNB-CU(即本公开实施例中的控制无线接入网设备)。进一步地，在无线网络人工智能架构的基础上，依据各网元的AI处理能力对模型进行分割，选择多个具有AI处理能力的网元辅助终端所属的模型推理网元共同完成模型推理工作，并将推理结果反馈给终端，终端依据推理结果执行相应的策略调整，并进行性能反馈，实现模型的持续优化。
具体流程为：首先终端发起模型分析订阅请求，终端接入的gNB-CU根据自身AI处理能力和终端的分析订阅请求信息，生成模型订阅请求信息并上报至OAM。OAM根据模型订阅请求进行模型选择和模型分割、模型分割块的分配与分发，并将模型分割块分配信息发送给所有参与联合推理的gNB-CU。终端接入的gNB-CU发起模型推理数据请求，相关网元进行数据收集和处理并发送给该gNB-CU。终端接入的gNB-CU采用模型推理数据完成第一个模型分割块的推理，并依据模型分割块分配信息将推理中间结果发送给下一个模型分割块所在的gNB-CU，直至负责最后一个模型分割块推理任务的gNB-CU得到推理结果后依据模型分割块分配信息将推理结果发送给终端接入的gNB-CU。终端接入的gNB-CU将推理结果发送给终端，终端采用推理结果进行相应的策略调整。gNB-CU收集模型性能数据和终端性能反馈数据并上报给OAM，OAM对模型进行训练优化，并将更新后的模型参数发送给gNB-CU。
在终端具有高速移动性的场景下，开展无线网络AI分析任务的重新交付工作。在分析请求信息中加上终端接入位置信息，在终端接入位置发生改变后通过重新发起分析请求来全局维护分析请求信息中的接入位置信息，并加入推理结果转发、重新进行模型选择与分割等流程来实现终端切换后模型推理任务的顺利交付，保证AI分析服务的连续性和准确性。具体可分为以下两种场景：
1)当终端切换时同一gNB-CU下的另一gNB-DU(即，本公开实施例中的分布式无线接入网设备)时，终端重新发起分析订阅请求，gNB-CU、OAM更新终端的分析订阅请求信息。若终端发生切换时当前推理任务未完成，则gNB-CU继续完成推理任务，得到推理结果后依据更新分析订阅请求消息中接入位置将推理结果发送给终端当前接入的 gNB-DU，该gNB-DU将推理结果发送给终端。终端切换完成后，由新接入的gNB-DU负责完成相关数据收集和数据转发任务。
2)当终端切换至另一gNB-CU时，终端重新发送模型分析订阅请求，终端新接入的gNB-CU向OAM发送模型订阅请求。OAM更新终端的分析订阅请求，并将更新的分析订阅请求信息发送给终端的源gNB-CU。若终端切换时当前推理任务未完成，则源gNB-CU完成推理任务，得到推理结果后依据更新分析订阅请求消息中接入位置将推理结果发送给终端新接入的gNB-CU。源gNB-CU更新终端模型订阅分析请求信息，不再负责该终端的模型订阅分析请求相关任务。终端新接入的gNB-CU将推理结果发送给终端新接入的gNB-DU，该gNB-DU将推理结果发送给终端。OAM依据终端新接入的gNB-CU发送的模型订阅分析请求重新进行模型选择和分割，并将模型分割块分配信息发送给参与联合推理的gNB-CU。通过该实施方式，可以更好地开发基站的AI处理能力，解决了基站AI处理能力不足的问题，并有利于网络负载均衡。并提供了在终端高速移动性场景下无线网络AI分析任务的交付方法，解决了终端切换带来的AI分析服务不连续的问题，保障了无线网络AI分析服务的高效性和连续性，提升了终端业务体验，同时也有利于提高无线网络运行效率。
进一步可以理解的是，本公开实施例的无线通信系统，是一种提供无线通信功能的网络。无线通信系统可以采用不同的通信技术，例如码分多址(code division multiple access,CDMA)、宽带码分多址(wideband code division multiple access，WCDMA)、时分多址(time division multiple access，TDMA)、频分多址(frequency division multiple access，FDMA)、正交频分多址(orthogonal frequency-division multiple access，OFDMA)、单载波频分多址(single Carrier FDMA，SC-FDMA)、载波侦听多路访问/冲突避免(Carrier Sense Multiple Access with Collision Avoidance)。根据不同网络的容量、速率、时延等因素可以将网络分为2G(英文：generation)网络、3G网络、4G网络或者未来演进网络，如5G网络，5G网络也可称为是新无线网络(New Radio，NR)。为了方便描述，本公开有时会将无线通信网络简称为网络。
进一步的，本公开中涉及的网络设备也可以称为无线接入网设备。该无线接入网设备可以是：基站、演进型基站(evolved node B，基站)、家庭基站、无线保真(wireless fidelity，WIFI)系统中的接入点(access point，AP)、无线中继节点、无线回传节点、传输点(transmission point，TP)或者发送接收点(transmission and reception point，TRP)等，还可以为NR系统中的gNB，或者，还可以是构成基站的组件或一部分设备等。当为车联网(V2X)通信系统时，网络设备还可以是车载设备。应理解，本公开的实施例中，对网络 设备所采用的具体技术和具体设备形态不做限定。
进一步的，本公开中涉及的终端，也可以称为终端设备、用户设备(User Equipment，UE)、移动台(Mobile Station，MS)、移动终端(Mobile Terminal，MT)等，是一种向用户提供语音和/或数据连通性的设备，例如，终端可以是具有无线连接功能的手持式设备、车载设备等。目前，一些终端的举例为：智能手机(Mobile Phone)、口袋计算机(Pocket Personal Computer，PPC)、掌上电脑、个人数字助理(Personal Digital Assistant，PDA)、笔记本电脑、平板电脑、可穿戴设备、或者车载设备等。此外，当为车联网(V2X)通信系统时，终端设备还可以是车载设备。应理解，本公开实施例对终端所采用的具体技术和具体设备形态不做限定。
图3是根据一示例性实施例示出的一种模型推理方法的流程图。如图3所示，模型推理方法用于OAM实体中，包括以下步骤。
在步骤S11中，响应于接收到控制无线接入网设备发送的模型订阅请求信息，确定与模型订阅请求信息对应的第一模型。
在本公开实施例中，模型订阅请求信息包括控制无线接入网设备自身AI处理能力信息，以及终端模型分析订阅请求信息。其中，AI处理能力信息包括基站服务器计算速度和当前富余算力。OAM根据模型订阅信息选择符合终端模型分析订阅请求信息的模型，进一步根据控制无线接入网设备的AI处理能力，在符合要求的模型中确定合适规模的模型，即第一模型。本公开为便描述，将符合要求且合适规模的模型称为第一模型。
在步骤S12中，将第一模型进行分割，得到第一数量的模型分割块，并将第一数量的模型分割块分发至第一数量的控制无线接入网设备。
在本公开实施例中，OAM依据控制无线接入网设备的AI处理能力信息将第一模型分割为第一数量的模型分割块，根据第一数量确定相同数量的控制无线接入网设备。将第一数量的模型分割块分发至第一数量的控制无线接入网设备。其中第一数量的控制无线接入网设备为，OAM基于与发送模型订阅请求信息的控制无线接入网设备相邻的控制无线接入网设备中确定的，确定的依据可以是控制无线接入网设备的算力占用情况和负载情况等选择较为空闲。
通过本公开实施例提供的模型推理方法，可以通过多个控制无线接入网设备协同推理的方法，将算力均衡到多个不同的控制无线接入网设备，充分利用控制无线接入网设备本地的AI处理能力，有效提高模型推理效率。
在本公开一些实施例中，第一数量的模型分割块中每个模型分割块对应有分配信息。其中分配信息包括第一数量的模型分割块的推理顺序，以及与每个模型分割块对应的控制 无线接入网设备。其中，与每个模型分割块对应的控制无线接入网设备以相应标识的方式包括在分配信息中。
在本公开一些实施例中，第一数量的控制无线接入网设备包括第一控制无线接入网设备，其中，第一控制无线接入网设备为终端接入的控制无线接入网设备。
图4是根据一示例性实施例示出的一种模型推理方法的流程图。如图4所示，模型推理方法用于OAM实体中，包括以下步骤。
在步骤S21中，在与第一控制无线接入网设备相邻的控制无线接入网设备中，确定多个辅助控制无线接入网设备。
在本公开实施例中，OAM在第一控制无线接入网设备相邻的控制无线接入网设备中，选择可以辅助进行模型推理的辅助控制无线接入网设备。
在步骤S22中，在多个辅助控制无线接入网设备中，基于每个辅助控制无线接入网设备的算力空闲状态，确定第二数量的控制无线接入网设备。
在本公开实施例中，OAM根据每个控制无线接入网设备算力占用状态和负载，确定可以参该次模型推理的第二数量的控制无线接入网设备。其中第二数量的控制无线接入网设备为除第一控制无线接入设备以外第一数量中其他的控制无线接入网设备。
在步骤S23中，基于第一数量的模型分割块的推理顺序，将第一个模型分割块发送至第一控制无线接入网设备，并将剩余数量的模型分割块分发至第二数量的控制无线接入网设备。
在本公开实施例中，OAM将第一个模型分割块发送给第一控制无线接入网设备(例如，gNB-CU1)，将其余的模型分割块发送给其他参与联合推理的控制无线接入网设备，并将与模型分割块对应的分配信息发送给所有参与联合推理的控制无线接入网设备。
图5是根据一示例性实施例示出的一种模型推理方法的流程图。如图5所示，模型推理方法用于OAM实体中，包括以下步骤。
在步骤S31中，接收第一控制无线接入网设备发送的模型性能更新数据。
在本公开实施例中，第一控制无线接入网设备根据接收的性能数据，与第一模型的第一推理结果进行对比，确定模型性能更新数据，并将模型性能更新数据发送至OAM。其中模型性能更新数据可以是模型精度。OAM还可以接收第一控制无线接入设备发送的性能数据。
在步骤S32中，基于模型性能更新数据更新第一模型，确定第一模型更新后的模型参数，并向第一控制无线接入网设备发送所述第一模型更新后的模型参数。
在本公开实施例中，OAM根据性能数据和模型性能更新数据对第一模型进行训练优 化，得到第一模型更新后的模型参数，并将第一模型更新后的模型参数发送至第一控制无线接入网设备。
在本公开一些实施例中，响应于OAM接收到第一模型分析订阅更新请求，且第一模型分析订阅更新请求包括的信息为终端的模型分析订阅信息，基于第一模型分析订阅更新请求更新终端的模型分析请求信息。
在本公开一些实施例中，响应于OAM接收到第一模型分析订阅更新请求，且第一模型分析订阅更新请求包括的信息为终端的模型分析订阅信息和第二控制无线接入网设备的AI处理能力信息，基于模型分析订阅信息和第二控制无线接入网设备的AI处理能力信息，重新对第一模型进行分割，将第一个模型分割块发送给第二控制无线接入网设备，剩余的模型分割块发送至参与推理的其他控制无线接入网设备。
图6是根据一示例性实施例示出的一种模型推理方法的流程图。如图6所示，模型推理方法用于控制无线接入网设备中，包括以下步骤。
在步骤S41中，响应于接收到分布式无线接入网设备发送的模型分析订阅请求，对模型分析订阅请求进行处理得到模型订阅请求信息，并向OAM发送模型订阅请求信息。
在本公开实施例中，模型分析订阅请求，包括终端的标识，分析请求类型、接入位置信息。示例性的，终端接入第一分布式无线接入网设备(例如，gNB-DU1)，gNB-DU1和gNB-DU2接入gNB-CU1。终端标识为GUTI，分析请求类型以分析ID来表示，如分析ID1：位置预测分析服务，分析ID2：负载预测分析服务。接入位置主要包含终端当前接入的控制无线接入网设备和分布式无线接入网设备信息。
响应于控制无线接入网设备接收到分布式无线接入网设备发送的模型分析订阅请求，根据自身AI处理能力和模型分析订阅请求，生成模型订阅请求信息，并向OAM发送该模型订阅请求信息。
在步骤S42中，接收OAM发送的模型分割块。
在本公开实施例中，模型分割块为分割第一模型确定的模型分割块。第一模型为OAM基于模型订阅请求信息确定的。
通过本公开实施例提供的模型推理方法，可以通过多个控制无线接入网设备协同推理的方法，将算力均衡到多个不同的控制无线接入网设备，充分利用控制无线接入网设备本地的AI处理能力，有效提高模型推理效率。
图7是根据一示例性实施例示出的一种模型推理方法的流程图。如图7所示，模型推理方法用于控制无线接入网设备中，包括以下步骤。
在步骤S51中，向分布式无线接入网设备发送模型推理数据请求。
在本公开实施例中，模型推理数据请求用于获取模型推理数据。控制无线接入网设备向分布式无线接入网设备发送模型推理数据请求。其中，需要说明的是，控制无线接入网设备可以向终端接入的分布式无线接入网设备发送模型推理数据请求，也可以向该控制无线接入网设备范围内其他参与推理的分布式无线接入网设备发送模型推理数据请求。
在步骤S52中，基于模型推理数据对模型分割块进行推理，得到模型分割块的推理中间信息。
在本公开实施例中，模型推理数据包括分布式无线接入网设备收集的模型推理数据和终端上报的模型推理数据。控制无线接入网设备基于接收的模型推理数据对模型分割块进行推理，确定每个模型分割块的推理中间信息。
在本公开一些实施例中，每个模型分割块对应有分配信息。控制无线接入网设备接收模型分割块，并接收分配信息。其中，分配信息包括第一数量的模型分割块的推理顺序，以及与每个模型分割块对应的所述控制无线接入网设备。其中，与每个模型分割块对应的控制无线接入网设备以相应标识的方式包括在分配信息中。
图8是根据一示例性实施例示出的一种模型推理方法的流程图。如图8所示，模型推理方法用于控制无线接入网设备中，包括以下步骤。
在步骤S61中，响应于控制无线接入网设备不是最后一个控制无线接入网设备，基于推理顺序，将推理中间信息发送至下一个控制无线接入网设备。
在本公开实施例中，响应于当前推理模型的控制无线接入网设备不是最后一个模型分割块所在的控制无线接入网设备，当前控制无线接入网设备根据分配信息中，推理模型分割块的推理顺序，将推理中间信息发送至下一个模型分割块所在的控制无线接入网设备。
在步骤S62中，响应于控制无线接入网设备为最后一个控制无线接入网设备，模型推理完成后，确定与第一模型对应的第一推理结果，将第一推理结果发送至第一控制无线接入网设备，第一控制无线接入网设备为终端接入的控制无线接入网设备。
在本公开实施例中，响应于当前推理模型的控制无线接入网设备是最后一个模型分割块所在的控制无线接入网设备，当前控制无线接入网设备根据分配信息中，推理模型分割块的推理顺序，完成第一模型的推理，并确定与第一模型相对应的第一推理结果，将第一推理结果发送至第一个模型分割块所在的控制无线接入网设备，即第一控制无线接入网设备。其中，第一控制无线接入网设备为终端接入的控制无线接入网设备。
通过本公开的模型推理方法，保证第一个模型分割块在终端当前接入的控制无线接入网设备进行推理，终端推理所需的原始推理数据只提供给当前接入的无线接入网设备。在其他参与联合推理的控制无线接入网设备间只传输推理中间信息，推理中间信息经过了特 征处理，具有较小的数据量且难以反向推理出终端信息，这种机制保证了无线网络敏感数据的安全性同时也节省了数据传输开销。
图9是根据一示例性实施例示出的一种模型推理方法的流程图。如图9所示，模型推理方法用于控制无线接入网设备中，包括以下步骤。
在步骤S71中，响应于控制无线接入网设备为第一控制无线接入网设备，接收第一推理结果。
在本公开实施例中，最后一个模型分割块对应得控制无线接入网设备将推理完成确定模型推理结果，即第一推理结果。将该第一推理结果发送至第一控制无线接入网设备，第一控制无线接入网设备得到与第一模型相对应的第一模型推理结果。
在步骤S72中，向第一分布式无线接入网设备发送第一推理结果。
在本公开实施例中，第一控制无线接入网设备根据接收得推理结果确定第一模型的第一推理结果。将第一推理结果发送至终端接入的分布式无线接入网设备。
图10是根据一示例性实施例示出的一种模型推理方法的流程图。如图10所示，模型推理方法用于控制无线接入网设备中，包括以下步骤。
在步骤S81中，接收第一分布式无线接入网设备发送的性能数据。
在本公开实施例中，性能数据为终端基于第一模型调整执行策略后的真实性能数据。终端基于该第一模型调整执行策略后，将得到的真实性能数据上报至接入的分布式无线接入网络设备。该分布式无线接入网络发送至控制无线接入网设备中。示例性的，终端接入的分布式无线接入网设备为gNB-DU1，与gNB-DU1对应的控制无线接入网设备为gNB-CU1，终端确定真实性能数据后，发送给gNB-DU1，gNB-DU1上报给gNB-CU1。
其中，性能数据可以是AI分析服务带来的性能提升的量化，如终端订阅某种分析并依据分析结果执行相应的策略调整后，实现省电5％。
在步骤S82中，对性能数据进行处理，得到模型性能更新数据，并向OAM发送模型性能更新数据。
在本公开实施例中，性能数据包括模型性能数据和终端反馈的性能反馈数据。第一控制无线接入网设备根据终端反馈的模型性能数据和性能反馈数据进行处理，得到模型性能更新数据，并将模型性能更新数据发送至OAM。
图11是根据一示例性实施例示出的一种模型推理方法的流程图。如图11所示，模型推理方法用于控制无线接入网设备中，包括以下步骤。
在步骤S91中，响应于控制无线接入网设备为第一控制无线接入网设备，向OAM发送模型订阅请求信息。
在本公开实施例中，第一控制无线接入网设备为终端接入的第一分布式无线网络设备对应的控制无线接入网设备。第一控制无线接入网设备接收到分布式无线接入网设备发送的模型分析订阅请求，根据自身的AI能力和模型分析订阅请求，生成模型订阅请求信息，并将模型订阅请求信息发送至OAM。
在本公开一些实施例中，由于终端的移动性，可能会发生切换接入的分布式无线接入网设备。一种实施方式中，终端切换接入的分布式无线接入网设备，不切换控制无线接入网设备。
图12是根据一示例性实施例示出的一种模型推理方法的流程图。如图12所示，模型推理方法用于控制无线接入网设备中，包括以下步骤。
在步骤S101中，响应于控制无线接入网设备为第一控制无线接入网设备，若重新接收到模型分析订阅请求，确定重新发送模型分析订阅请求的第二分布式无线接入网设备。
在本公开实施例中，第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备。
响应于第一控制无线接入网设备重新接收到模型分析订阅请求，确定终端接入的分布式无线接入网设备发生切换，更新终端的分析订阅信息，并将分析订阅信息上报至OAM。一种方式中，第一控制无线接入网设备未完成第一模型的推理任务，第一控制无线接入网设备(例如，gNB-CU1)完成当前模型推理任务、得到第一推理结果后依据更新分析请求消息中接入位置将第一推理结果发送给第二分布式无线接入网设备(例如，gNB-DU2)，由gNB-DU2转发给终端。
在步骤S102中，将第一推理结果发送至第二分布式无线接入网设备，并向OAM发送模型订阅更新请求。
在本公开实施例中，第一控制无线接入网设备将第一推理结果发送至第二分布式无线接入网设备，并向OAM发送模型订阅更新请求，请求OAM更新终端的分析请求信息。
示例性的，以第一控制无线接入网设备为gNB-CU1，第二分布式无线接入网设备为gNB-DU2为例。gNB-CU1更新终端的分析请求信息。gNB-CU1将终端分析请求消息上报给OAM，OAM更新终端的分析请求信息。响应于终端切换时当前推理任务未完成，gNB-CU1完成当前模型推理任务、得到第一推理结果后依据更新分析请求消息中接入位置将第一推理结果发送给gNB-DU2，由gNB-DU2转发给终端。
一种实施方式中，终端切换接入的分布式无线接入网设备，并切换控制无线接入网设备。其实施方式包括：
图13是根据一示例性实施例示出的一种模型推理方法的流程图。如图13所示，模型 推理方法用于控制无线接入网设备中，包括以下步骤。
在步骤S111中，响应于控制无线接入网设备为第二控制无线接入网设备，若重新接收到模型分析订阅请求，确定重新发送模型分析订阅请求的第二分布式无线接入网设备，以及重新接收到模型分析订阅请求的第二控制无线接入网设备。
在本公开实施例中，第二控制无线接入网设备为第二分布式无线接入网设备对应的控制无线接入网设备。第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备。
响应于控制无线接入网设备重新接收到分布式无线接入网设备发送的模型分析订阅请求，且重新接收的控制无线接入网设备为第二控制无线接入网设备，确定终端重新接入的第二分布式无线接入网设备。
在步骤S112中，将第一推理结果发送至第二控制无线接入网设备，并向OAM发送模型订阅更新请求。
在本公开实施例中，第一控制无线接入网设备将第一推理结果发送至第二控制无线接入网设备，不再负责该终端的分析请求。并向OAM发送模型订阅更新请求。
示例性的，以第二控制无线接入网设备为gNB-CU2，第二分布式无线接入网设备为gNB-DU3为例。gNB-CU2向OAM发送模型订阅请求，包括自身AI处理能力信息和终端分析订阅请求信息。OAM依据模型订阅请求更新当前终端的分析订阅请求信息；并将更新的分析订阅请求信息发送给源基站gNB-CU1。响应于终端切换时当前推理任务未完成，gNB-CU1完成推理任务、得到第一推理结果后依据更新分析请求消息中接入位置将第一推理结果发送给gNB-CU2。gNB-CU1更新终端分析请求信息，不再负责该终端的分析请求相关任务。gNB-CU2将第一推理结果发送给gNB-DU3，gNB-DU3转发给终端。OAM依据gNB-CU2的模型订阅请求中的AI处理能力信息对模型进行重新分割，将第一块发送给发起请求的gNB-CU2，其余块发送给其他gNB-CU，并将模型分割块分配信息发送给参与联合推理的给gNB-CU。
通过本公开提供的模型推理方法，保证源分布式无线接入网设备当前第一推理结果顺利反馈给已经发生切换的终端，新接入的分布式无线接入网设备迅速接管模型推理任务，从而避免切换过程中终端所需的AI分析服务中断，保障了移动终端AI分析服务的连续性和准确性。
图14是根据一示例性实施例示出的一种模型推理方法的流程图。如图14所示，模型推理方法用于分布式无线接入网设备中，包括以下步骤。
在步骤S121中，响应于接收到终端发送的模型分析订阅请求，向控制无线接入网设 备发送模型分析订阅请求。
在本公开实施例中，模型分析订阅请求用于向OAM获取第一模型。第一模型包括第一数量的模型分割块。
终端向接入的分布式无线接入网设备发起模型分析订阅请求，其中，模型分析订阅请求包括终端标识，分析请求类型、接入位置信息。分布式无线接入网设备接收到模型分析订阅请求后，向控制无线接入网设备发送该模型分析订阅请求。
图15是根据一示例性实施例示出的一种模型推理方法的流程图。如图15所示，模型推理方法用于分布式无线接入网设备中，包括以下步骤。
在步骤S131中，接收控制无线接入网设备发送的模型推理数据请求。
在步骤S132中，向终端获取模型推理数据，并发送至控制无线接入网设备。
在本公开实施例中，模型推理数据请求用于获取模型推理数据。分布式无线接入网设备接收到模型推理数据请求后，向终端获取模型推理数据，并向控制无线接入网设备发送该模型推理数据。
图16是根据一示例性实施例示出的一种模型推理方法的流程图。如图16所示，模型推理方法用于分布式无线接入网设备中，包括以下步骤。
在步骤S141中，响应于分布式无线接入网设备为第一分布式无线接入网设备，接收第一控制无线接入网设备发送的第一推理结果。
在步骤S142中，将第一推理结果发送至终端。
在本公开实施例中，终端接入的第一分布式无线接入网设备，接收第一控制无线接入网设备发送的第一推理结果，并将第一推理结果发送至终端，以使终端调整执行策略。
图17是根据一示例性实施例示出的一种模型推理方法的流程图。如图17所示，模型推理方法用于分布式无线接入网设备中，包括以下步骤。
在步骤S151中，响应于分布式无线接入网设备为第一分布式无线接入网设备，接收终端发送的性能数据。
在本公开实施例中，性能数据为终端基于第一模型调整执行策略后的真实性能数据；
在步骤S152中，向第一控制无线接入网设备发送性能数据。
在本公开实施例中，终端向接入的第一分布式无线接入网设备，发送调整执行策略后得到真实性能数据，(也可以称为性能反馈数据)。并向与第一分布式无线接入网设备对应的第一控制无线接入网设备发送该性能数据。
图18是根据一示例性实施例示出的一种模型推理方法的流程图。如图18所示，模型推理方法用于分布式无线接入网设备中，包括以下步骤。
在步骤S161中，响应于无线接入网设备为第二分布式无线接入网设备，若接收到终端重新发送的模型分析订阅请求，确定向与第二分布式无线接入网络设备对应的控制无线接入网设备发送模型分析订阅请求。
在本公开实施例中，第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备。
终端切换接入的分布式无线接入网设备后，重新向接入的分布式无线接入网设备发送模型分析订阅请求。响应于接收的分布式无线接入网络为第二分布式无线接入网设备，即终端重新接入的分布式无线接入网设备，则向其对应的控制无线接入网设备发送该模型分析订阅请求。
在本公开一些实施例中，以OAM，控制无线接入网设备和分布式无线接入网设备之间的交互过程为例进一步进行说明。其中控制无线接入网设备可以是gNB-CU，分布式无线接入网设备可以是gNB-DU。图19是根据一示例性实施例示出的一种模型推理方法的流程图。如图19所示，包括以下步骤：
在步骤1中，终端发起模型分析订阅请求。
在步骤2中，gNB-CU向OAM发起模型订阅请求。
在步骤3中，OAM根据模型订阅请求信息进行模型选择和模型分割、模型分割块的分配与分发，并将模型分割块分配信息发送给参与联合推理的gNB-CU。
在步骤4中，终端接入的gNB-CU发起模型推理数据请求，相关网元进行数据收集和处理，并将数据发送给该gNB-CU。
在步骤5中，终端接入的gNB-CU采用模型推理数据完成第一个模型分割块的推理，并依据模型分割块分配信息将推理中间结果发送给下一个模型分割块所在的gNB-CU。
在步骤6中，负责最后一个模型分割块推理任务的gNB-CU得到第一推理结果之后依据模型分割块分配信息将第一推理结果发送给终端接入的gNB-CU。
在步骤7中，终端接入的gNB-CU将第一推理结果发送给终端，终端采用第一推理结果进行相应的策略调整。
在步骤8中，gNB-CU收集模型性能数据和终端性能反馈数据并上报给OAM，OAM对模型进行训练优化，并将更新后的模型发送给gNB-CU。
在本公开一些实施例中，由于终端的移动性，可能会发生切换接入的分布式无线接入网设备。一种实施方式中，终端切换接入的分布式无线接入网设备，不切换控制无线接入网设备。图20是根据一示例性实施例示出的一种模型推理方法中终端切换的流程图。如图20所示，包括以下步骤：
在步骤1中，终端重新发起分析订阅请求。
在步骤2中，gNB-CU、OAM更新终端的模型分析订阅请求信息。
在步骤3中，若终端发生切换时当前推理任务未完成，则gNB-CU完成推理任务，并将第一推理结果发送给终端当前接入的gNB-DU，该gNB-DU将第一推理结果发送给终端。
在步骤4中，终端切换完成后，由新接入的gNB-DU负责完成相关数据收集和数据转发任务。
在本公开一些实施例中，由于终端的移动性，可能会发生切换接入的分布式无线接入网设备。一种实施方式中，终端切换接入的分布式无线接入网设备，且切换控制无线接入网设备。图21是根据一示例性实施例示出的一种模型推理方法中终端切换的流程图。如图21所示，包括以下步骤：
在步骤1中，终端重新发起模型分析订阅请求。
在步骤2中，终端新接入的gNB-CU向OAM发送模型订阅请求信息。
在步骤3中，OAM更新终端的分析订阅请求，并将更新的模型分析订阅请求信息发送给终端的源gNB-CU。
在步骤4中，若终端切换时当前推理任务未完成，则源gNB-CU完成推理任务，将第一推理结果发送给终端新接入的gNB-CU。
在步骤5中，源gNB-CU更新终端分析请求信息，不再负责该终端的分析请求相关任务。
在步骤6中，终端新接入的gNB-CU将第一推理结果发送给终端新接入的gNB-DU，该gNB-DU将第一推理结果发送给终端。
在步骤7中，OAM依据终端新接入的gNB-CU发送的模型订阅请求重新进行模型选择和分割，并将模型分割块分配信息发送给参与联合推理的gNB-CU。
图22是根据一示例性实施例示出的一种模型推理方法的协议和接口原理图。如图22所示，主要涉及本发明实施例提供的终端、终端接入的gNB-DU、终端接入的gNB-CU、参与联合推理的其他gNB-CU(gNB-CU(1)～gNB-CU(N))以及OAM。具体如下：
1a.终端将模型分析订阅请求信令发送给gNB-DU，指示向接收方发起模型分析订阅请求。1b.gNB-DU将分析订阅请求信令发送给gNB-CU。2.gNB-CU根据自身AI处理能力和分析订阅请求信息，生成模型订阅请求信息。3.gNB-CU将模型订阅请求信令发送给OAM，指示向接收方发起模型订阅请求。4.OAM根据模型订阅信息选择符合分析请求的第一模型，并依据AI处理能力信息将模型分割为若干块。5a.OAM将第一个模型分割块及模型分割块分配信息发送给终端接入的gNB-CU。5b.OAM将其余模型分割块及模型分割 块分配信息发送给参与联合推理的其他gNB-CU，指示发送模型分割块分配信息。6.gNB-CU将模型推理数据收集请求信令发送给gNB-DU，指示向接收方发起模型推理数据收集请求。7.gNB-DU、终端、gNB-CU依据模型推理数据收集请求分别收集数据，并发送给gNB-CU。8.gNB-CU采用收集的推理数据对第一个模型分割块进行部分模型推理，得到推理中间信息。9.gNB-CU将推理中间信息发送给下一个模型分割块所在的gNB-CU(1)。10.NB-CU(1)对模型分割块进行部分模型推理，得到推理中间信息结果。11.gNB-CU(1)将推理中间信息发送给下一个模型分割块所在的gNB-CU，直至最后一个模型分割块所在的gNB-CU(N)接收到所有的推理中间信息。12.gNB-CU(N)对最后一个模型分割块进行部分模型推理，得到第一推理结果。13a.gNB-CU(N)将第一推理结果发送给终端接入的gNB-CU。13b.gNB-CU将第一推理结果发送给终端接入的gNB-DU。13c.gNB-DU将第一推理结果发送给终端。14.终端根据第一推理结果做出相应策略调整。15a.终端将性能反馈数据发送给gNB-DU。15b.gNB-DU将性能反馈数据发送给gNB-CU。16.gNB-CU将第一推理结果与真实数据进行对比，得到模型性能数据。17.gNB-CU对模型性能数据和终端性能反馈数据进行处理。18.gNB-CU将模型性能数据和终端性能反馈数据发送给OAM。19.OAM基于模型性能数据和性能反馈数据对模型进行训练优化。20.OAM将更新后的模型参数发送给gNB-CU。
图23是根据一示例性实施例示出的一种无模型推理方法中终端在同一gNB-CU下切换时AI分析任务交付的协议和接口原理图，如图23所示，主要涉及本公开实施例提供的终端、终端的源gNB-DU(gNB-DU 1)、终端新接入的gNB-DU(gNB-DU 2)、终端接入的gNB-CU以及OAM。具体如下：
1a.终端将模型分析订阅请求信令发送给gNB-DU 2。1b.gNB-DU 2将，模型分析订阅请求信令发送给gNB-CU。2.gNB-CU更新终端的分析请求信息。3.gNB-CU将分析订阅更新请求发送给OAM。4.OAM更新终端的分析请求信息。5.若终端切换时当前推理任务未完成，则gNB-CU继续完成当前推理任务，得到第一推理结果。6a.gNB-CU依据分析订阅更新请求中接入位置将第一推理结果发送给gNB-DU 2。6b.gNB-DU 2将第一推理结果发送给终端。7.终端切换完成后，gNB-DU 2负责终端分析请求相关数据收集和数据转发任务。
图24是根据一示例性实施例示出的一种无模型推理方法中终端跨gNB-CU切换时AI分析任务交付的协议和接口原理图，如图24所示，主要涉及本公开实施例提供的终端、终端的源gNB-DU(gNB-DU 1)、终端新接入的gNB-DU(gNB-DU 3)、终端的源gNB-CU(gNB-CU 1)、终端新接入的gNB-CU(gNB-CU 2)、参与联合推理的其他gNB-CU(gNB-CU (1)～gNB-CU(N))以及OAM。具体如下：
1a.终端将模型分析订阅请求信令发送给gNB-DU 3。1b.gNB-DU 3将模型分析订阅请求信令发送给gNB-CU 2。2.gNB-CU 2根据自身AI处理能力和分析订阅请求信息，生成模型订阅请求信息。3.gNB-CU 2将模型订阅请求信令发送给OAM。4.OAM根据模型订阅请求更新当前终端的分析订阅请求信息。5.OAM将分析订阅更新请求发送给gNB-CU 1，指示向接收者发起分析订阅更新请求。6.若终端切换时当前推理任务未完成，则gNB-CU 1继续完成当前推理任务，得到第一推理结果。7.gNB-CU 1更新终端分析请求信息，不再负责该终端的分析请求相关任务。8a.gNB-CU 1依据分析订阅更新请求中接入位置将第一推理结果发送给gNB-CU 2。8b.gNB-CU 2将第一推理结果发送给gNB-DU 3。8c.gNB-DU 3将第一推理结果发送给终端。9.OAM依据模型订阅请求中的AI处理能力信息重新进行模型选择和模型分割。10a.OAM将第一个模型分割块及模型分割块分配信息发送给gNB-CU 2。10b.OAM将其余模型分割块及模型分割块分配信息发送给辅助推理的gNB-CU。
基于相同的构思，本公开实施例还提供一种模型推理装置。
可以理解的是，本公开实施例提供的模型推理装置为了实现上述功能，其包含了执行各个功能相应的硬件结构和/或软件模块。结合本公开实施例中所公开的各示例的单元及算法步骤，本公开实施例能够以硬件或硬件和计算机软件的结合形式来实现。某个功能究竟以硬件还是计算机软件驱动硬件的方式来执行，取决于技术方案的特定应用和设计约束条件。本领域技术人员可以对每个特定的应用来使用不同的方法来实现所描述的功能，但是这种实现不应认为超出本公开实施例的技术方案的范围。
图25是根据一示例性实施例示出的一种模型推理装置框图。参照图25，该模型推理装置100应用于操作维护管理OAM实体，包括确定模块101和发送模块102。
确定模块101，用于响应于接收到控制无线接入网设备发送的模型订阅请求信息，确定与模型订阅请求信息对应的第一模型。发送模块102，用于将第一模型进行分割，得到第一数量的模型分割块，并将第一数量的模型分割块分发至第一数量的控制无线接入网设备。
在本公开实施例中，第一数量的模型分割块中每个模型分割块对应有分配信息。
分配信息包括第一数量的模型分割块的推理顺序，以及与每个模型分割块对应的控制无线接入网设备。
在本公开实施例中，第一数量的控制无线接入网设备包括第一控制无线接入网设备，第一控制无线接入网设备为终端接入的控制无线接入网设备。
发送模块102，用于在与第一控制无线接入网设备相邻的控制无线接入网设备中，确 定多个辅助控制无线接入网设备；在多个辅助控制无线接入网设备中，基于每个辅助控制无线接入网设备的算力占用状态和负载，确定第二数量的控制无线接入网设备；第二数量的控制无线接入网设备为除第一控制无线接入设备以外第一数量中其他的控制无线接入网设备；基于第一数量的模型分割块的推理顺序，将第一个模型分割块发送至第一控制无线接入网设备，并将剩余数量的模型分割块分发至第二数量的控制无线接入网设备。
在本公开实施例中，模型推理装置还包括：接收模块103。
所示接收模块103用于，接收第一控制无线接入网设备发送的模型性能更新数据。基于模型性能更新数据更新第一模型，确定第一模型更新后的模型参数，并向第一控制无线接入网设备发送第一模型更新后的模型参数。
在本公开实施例中，接收模块103还用于响应于接收到第一模型分析订阅更新请求，更新终端接入的分布式无线接入网设备。其中，第一模型分析订阅更新请求指示终端切换分布式无线接入网设备，且不切换控制无线接入网设备。或，响应于接收到第二模型分析订阅更新请求，更新终端接入的分布式无线接入网设备，并重新对第一模型进行分割。其中，第二模型分析订阅更新请求指示终端切换分布式无线接入网设备，并切换控制无线接入网设备。
图26是根据一示例性实施例示出的一种模型推理装置框图。参照图26，该模型推理装置200应用于控制无线接入网设备，包括发送模块201和接收模块202。
发送模块201，用于响应于接收到分布式无线接入网设备发送的模型分析订阅请求，对模型分析订阅请求进行处理得到模型订阅请求信息，并向OAM发送模型订阅请求信息。接收模块202，用于接收OAM发送的模型分割块。模型分割块为分割第一模型确定的模型分割块。第一模型为OAM基于模型订阅请求信息确定的。
在本公开实施例中，发送模块201还用于向分布式无线接入网设备发送模型推理数据请求，模型推理数据请求用于获取模型推理数据。基于模型推理数据对模型分割块进行推理，得到模型分割块的推理中间信息。
在本公开实施例中，模型分割块对应有分配信息。分配信息包括第一数量的模型分割块的推理顺序，以及与每个模型分割块对应的控制无线接入网设备。
发送模块201还用于响应于控制无线接入网设备不是最后一个控制无线接入网设备，基于推理顺序，将推理中间信息发送至下一个控制无线接入网设备。响应于控制无线接入网设备为最后一个控制无线接入网设备，模型推理完成后，确定与第一模型对应的第一推理结果，将所述第一推理结果发送至第一控制无线接入网设备，第一控制无线接入网设备为终端接入的控制无线接入网设备。
在本公开实施例中，发送模块201还用于响应于控制无线接入网设备为第一控制无线接入网设备，接收所述第一推理结果。向第一分布式无线接入网设备发送第一推理结果，第一分布式无线接入网设备为终端接入的分布式无线接入网设备。
在本公开实施例中，将第一推理结果发送至第一分布式无线接入网设备之后，接收模块202还用于接收第一分布式无线接入网设备发送的性能数据，性能数据为终端基于第一模型调整执行策略后的真实性能数据。对性能数据进行处理，得到模型性能更新数据，并向OAM发送模型性能更新数据。
在本公开实施例中，发送模块201还用于响应于控制无线接入网设备为第一控制无线接入网设备，向OAM发送模型订阅请求信息。其中，第一控制无线接入网设备为终端接入的第一分布式无线网络设备对应的控制无线接入网设备。
在本公开实施例中，发送模块201还用于响应于控制无线接入网设备为第一控制无线接入网设备，若重新接收到模型分析订阅请求，确定重新发送模型分析订阅请求的第二分布式无线接入网设备。第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备。将第一推理结果发送至第二分布式无线接入网设备，并向OAM发送模型订阅更新请求。
在本公开实施例中，发送模块201还用于响应于控制无线接入网设备为第二控制无线接入网设备，若重新接收到模型分析订阅请求，确定重新发送模型分析订阅请求的第二分布式无线接入网设备，以及重新接收到模型分析订阅请求的第二控制无线接入网设备，第二控制无线接入网设备为第二分布式无线接入网设备对应的控制无线接入网设备。第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备。将第一推理结果发送至第二控制无线接入网设备，并向OAM发送模型订阅更新请求。
图27是根据一示例性实施例示出的一种模型推理装置框图。参照图27，该模型推理装置300应用于分布式无线接入网设备，包括发送模块301。
发送模块301，用于响应于接收到终端发送的模型分析订阅请求，向控制无线接入网设备发送模型分析订阅请求。其中，模型分析订阅请求用于向OAM获取第一模型。第一模型包括第一数量的模型分割块。
在本公开实施例中，装置还包括：接收模块302。
接收模块302，用于接收控制无线接入网设备发送的模型推理数据请求，模型推理数据请求用于获取模型推理数据。向终端获取模型推理数据，并发送至控制无线接入网设备。
在本公开实施例中，接收模块302，还用于响应于分布式无线接入网设备为第一分布 式无线接入网设备，接收第一控制无线接入网设备发送的第一推理结果。将第一推理结果发送至终端。
在本公开实施例中，接收模块302，还用于响应于分布式无线接入网设备为第一分布式无线接入网设备，接收终端发送的性能数据。性能数据为终端基于第一模型调整执行策略后的真实性能数据。向第一控制无线接入网设备发送性能数据。
在本公开实施例中，接收模块302，还用于响应于无线接入网设备为第二分布式无线接入网设备，若接收到终端重新发送的模型分析订阅请求，确定向与第二分布式无线接入网络设备对应的控制无线接入网设备发送模型分析订阅请求。其中，第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备。
关于上述实施例中的装置，其中各个模块执行操作的具体方式已经在有关该方法的实施例中进行了详细描述，此处将不做详细阐述说明。
图28是根据一示例性实施例示出的一种用于模型推理的装置400的框图。例如，装置400可以是移动电话，计算机，数字广播终端，消息收发设备，游戏控制台，平板设备，医疗设备，健身设备，个人数字助理等。
参照图28，装置400可以包括以下一个或多个组件：处理组件402，存储器404，电力组件406，多媒体组件408，音频组件410，输入/输出(I/O)接口412，传感器组件414，以及通信组件416。
处理组件402通常控制装置400的整体操作，诸如与显示，电话呼叫，数据通信，相机操作和记录操作相关联的操作。处理组件402可以包括一个或多个处理器420来执行指令，以完成上述的方法的全部或部分步骤。此外，处理组件402可以包括一个或多个模块，便于处理组件402和其他组件之间的交互。例如，处理组件402可以包括多媒体模块，以方便多媒体组件408和处理组件402之间的交互。
存储器404被配置为存储各种类型的数据以支持在装置400的操作。这些数据的示例包括用于在装置400上操作的任何应用程序或方法的指令，联系人数据，电话簿数据，消息，图片，视频等。存储器404可以由任何类型的易失性或非易失性存储设备或者它们的组合实现，如静态随机存取存储器(SRAM)，电可擦除可编程只读存储器(EEPROM)，可擦除可编程只读存储器(EPROM)，可编程只读存储器(PROM)，只读存储器(ROM)，磁存储器，快闪存储器，磁盘或光盘。
电力组件406为装置400的各种组件提供电力。电力组件406可以包括电源管理系统，一个或多个电源，及其他与为装置400生成、管理和分配电力相关联的组件。
多媒体组件408包括在所述装置400和用户之间的提供一个输出接口的屏幕。在一些 实施例中，屏幕可以包括液晶显示器(LCD)和触摸面板(TP)。如果屏幕包括触摸面板，屏幕可以被实现为触摸屏，以接收来自用户的输入信号。触摸面板包括一个或多个触摸传感器以感测触摸、滑动和触摸面板上的手势。所述触摸传感器可以不仅感测触摸或滑动动作的边界，而且还检测与所述触摸或滑动操作相关的持续时间和压力。在一些实施例中，多媒体组件408包括一个前置摄像头和/或后置摄像头。当装置400处于操作模式，如拍摄模式或视频模式时，前置摄像头和/或后置摄像头可以接收外部的多媒体数据。每个前置摄像头和后置摄像头可以是一个固定的光学透镜系统或具有焦距和光学变焦能力。
音频组件410被配置为输出和/或输入音频信号。例如，音频组件410包括一个麦克风(MIC)，当装置400处于操作模式，如呼叫模式、记录模式和语音识别模式时，麦克风被配置为接收外部音频信号。所接收的音频信号可以被进一步存储在存储器404或经由通信组件416发送。在一些实施例中，音频组件410还包括一个扬声器，用于输出音频信号。
I/O接口412为处理组件402和外围接口模块之间提供接口，上述外围接口模块可以是键盘，点击轮，按钮等。这些按钮可包括但不限于：主页按钮、音量按钮、启动按钮和锁定按钮。
传感器组件414包括一个或多个传感器，用于为装置400提供各个方面的状态评估。例如，传感器组件414可以检测到装置400的打开/关闭状态，组件的相对定位，例如所述组件为装置400的显示器和小键盘，传感器组件414还可以检测装置400或装置400一个组件的位置改变，用户与装置400接触的存在或不存在，装置400方位或加速/减速和装置400的温度变化。传感器组件414可以包括接近传感器，被配置用来在没有任何的物理接触时检测附近物体的存在。传感器组件414还可以包括光传感器，如CMOS或CCD图像传感器，用于在成像应用中使用。在一些实施例中，该传感器组件414还可以包括加速度传感器，陀螺仪传感器，磁传感器，压力传感器或温度传感器。
通信组件416被配置为便于装置400和其他设备之间有线或无线方式的通信。装置400可以接入基于通信标准的无线网络，如WiFi，2G或3G，或它们的组合。在一个示例性实施例中，通信组件416经由广播信道接收来自外部广播管理系统的广播信号或广播相关信息。在一个示例性实施例中，所述通信组件416还包括近场通信(NFC)模块，以促进短程通信。例如，在NFC模块可基于射频识别(RFID)技术，红外数据协会(IrDA)技术，超宽带(UWB)技术，蓝牙(BT)技术和其他技术来实现。
在示例性实施例中，装置400可以被一个或多个应用专用集成电路(ASIC)、数字信号处理器(DSP)、数字信号处理设备(DSPD)、可编程逻辑器件(PLD)、现场可编程门阵列(FPGA)、控制器、微控制器、微处理器或其他电子元件实现，用于执行上述方法。
在示例性实施例中，还提供了一种包括指令的非临时性计算机可读存储介质，例如包括指令的存储器404，上述指令可由装置400的处理器420执行以完成上述方法。例如，所述非临时性计算机可读存储介质可以是ROM、随机存取存储器(RAM)、CD-ROM、磁带、软盘和光数据存储设备等。
图29是根据一示例性实施例示出的一种用于模型推理的装置500的框图。例如，装置500可以被提供为一服务器。参照图29，装置500包括处理组件522，其进一步包括一个或多个处理器，以及由存储器532所代表的存储器资源，用于存储可由处理组件522的执行的指令，例如应用程序。存储器532中存储的应用程序可以包括一个或一个以上的每一个对应于一组指令的模块。此外，处理组件522被配置为执行指令，以执行上述方法。
装置500还可以包括一个电源组件526被配置为执行装置500的电源管理，一个有线或无线网络接口550被配置为将装置500连接到网络，和一个输入输出(I/O)接口558。装置500可以操作基于存储在存储器532的操作系统，例如Windows ServerTM，Mac OS XTM，UnixTM,LinuxTM，FreeBSDTM或类似。
进一步可以理解的是，本公开中“多个”是指两个或两个以上，其它量词与之类似。“和/或”，描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。字符“/”一般表示前后关联对象是一种“或”的关系。单数形式的“一种”、“所述”和“该”也旨在包括多数形式，除非上下文清楚地表示其他含义。
进一步可以理解的是，术语“第一”、“第二”等用于描述各种信息，但这些信息不应限于这些术语。这些术语仅用来将同一类型的信息彼此区分开，并不表示特定的顺序或者重要程度。实际上，“第一”、“第二”等表述完全可以互换使用。例如，在不脱离本公开范围的情况下，第一信息也可以被称为第二信息，类似地，第二信息也可以被称为第一信息。
进一步可以理解的是，本公开实施例中尽管在附图中以特定的顺序描述操作，但是不应将其理解为要求按照所示的特定顺序或是串行顺序来执行这些操作，或是要求执行全部所示的操作以得到期望的结果。在特定环境中，多任务和并行处理可能是有利的。
本领域技术人员在考虑说明书及实践这里公开的发明后，将容易想到本公开的其它实施方案。本申请旨在涵盖本公开的任何变型、用途或者适应性变化，这些变型、用途或者适应性变化遵循本公开的一般性原理并包括本公开未公开的本技术领域中的公知常识或惯用技术手段。说明书和实施例仅被视为示例性的，本公开的真正范围和精神由下面的权利要求指出。
应当理解的是，本公开并不局限于上面已经描述并在附图中示出的精确结构，并且可以在不脱离其范围进行各种修改和改变。本公开的范围仅由所附的权利要求来限制。

Claims (23)

  1. 一种模型推理方法，其特征在于，应用于操作维护管理OAM实体，所述方法包括：
    响应于接收到控制无线接入网设备发送的模型订阅请求信息，确定与所述模型订阅请求信息对应的第一模型；
    将所述第一模型进行分割，得到第一数量的模型分割块，并将所述第一数量的模型分割块分发至第一数量的控制无线接入网设备。
  2. 根据权利要求1所述的模型推理方法，其特征在于，所述第一数量的模型分割块中每个模型分割块对应有分配信息；
    所述分配信息包括第一数量的模型分割块的推理顺序，以及与每个模型分割块对应的所述控制无线接入网设备。
  3. 根据权利要求1所述的模型推理方法，其特征在于；所述第一数量的控制无线接入网设备包括第一控制无线接入网设备，所述第一控制无线接入网设备为终端接入的控制无线接入网设备；
    所述将所述第一数量的模型分割块分发至第一数量的控制无线接入网设备，包括：
    在与所述第一控制无线接入网设备相邻的控制无线接入网设备中，确定多个辅助控制无线接入网设备；
    在所述多个辅助控制无线接入网设备中，基于每个所述辅助控制无线接入网设备的算力占用状态和负载，确定第二数量的控制无线接入网设备；所述第二数量的控制无线接入网设备为除第一控制无线接入设备以外第一数量中其他的控制无线接入网设备；
    基于第一数量的模型分割块的推理顺序，将第一个模型分割块发送至所述第一控制无线接入网设备，并将剩余数量的模型分割块分发至所述第二数量的控制无线接入网设备。
  4. 根据权利要求3所述的模型推理方法，其特征在于，所述模型推理方法还包括：
    接收第一控制无线接入网设备发送的模型性能更新数据；
    基于所述模型性能更新数据更新所述第一模型，确定所述第一模型更新后的模型参数，并向所述第一控制无线接入网设备发送所述第一模型更新后的模型参数。
  5. 根据权利要求3所述的模型推理方法，其特征在于，所述模型推理方法还包括：
    响应于接收到第一模型分析订阅更新请求，更新终端接入的分布式无线接入网设备；其中，所述第一模型分析订阅更新请求指示终端切换分布式无线接入网设备，且不切换控制无线接入网设备；
    或
    响应于接收到第二模型分析订阅更新请求，更新终端接入的分布式无线接入网设备，并重新对所述第一模型进行分割；其中，所述第二模型分析订阅更新请求指示终端切换分布式无线接入网设备，并切换控制无线接入网设备。
  6. 一种模型推理方法，其特征在于，应用于控制无线接入网设备，所述方法包括：
    响应于接收到分布式无线接入网设备发送的模型分析订阅请求，对所述模型分析订阅请求进行处理得到模型订阅请求信息，并向OAM发送所述模型订阅请求信息；
    接收OAM发送的模型分割块；所述模型分割块为分割第一模型确定的模型分割块；所述第一模型为OAM基于所述模型订阅请求信息确定的。
  7. 根据权利要求6所述的模型推理方法，其特征在于，所述向OAM发送模型订阅请求信息之后，所述方法还包括：
    向分布式无线接入网设备发送模型推理数据请求，所述模型推理数据请求用于获取模型推理数据；
    基于所述模型推理数据对所述模型分割块进行推理，得到模型分割块的推理中间信息。
  8. 根据权利要求7所述的模型推理方法，其特征在于，所述模型分割块对应有分配信息；所述分配信息包括第一数量的模型分割块的推理顺序，以及与每个模型分割块对应的所述控制无线接入网设备；
    所述模型推理方法还包括：
    响应于所述控制无线接入网设备不是最后一个控制无线接入网设备，基于所述推理顺序，将推理中间信息发送至下一个控制无线接入网设备；
    响应于所述控制无线接入网设备为最后一个控制无线接入网设备，模型推理完成后，确定与第一模型对应的第一推理结果，将所述第一推理结果发送至第一控制无线接入网设备，所述第一控制无线接入网设备为终端接入的控制无线接入网设备。
  9. 根据权利要求8所述的模型推理方法，其特征在于，所述方法还包括：
    响应于所述控制无线接入网设备为第一控制无线接入网设备，接收所述第一推理结果；
    向第一分布式无线接入网设备发送所述第一推理结果，所述第一分布式无线接入网设备为终端接入的分布式无线接入网设备。
  10. 根据权利要求9所述的模型推理方法，其特征在于，所述将所述第一推理结果发送至第一分布式无线接入网设备之后，所述模型推理方法还包括：
    接收第一分布式无线接入网设备发送的性能数据，所述性能数据为终端基于第一模型调整执行策略后的真实性能数据；
    对所述性能数据进行处理，得到模型性能更新数据，并向OAM发送所述模型性能更新数据。
  11. 根据权利要求6所述的模型推理方法，其特征在于，向OAM发送模型订阅请求信息，包括：
    响应于所述控制无线接入网设备为第一控制无线接入网设备，向OAM发送模型订阅请求信息；
    其中，所述第一控制无线接入网设备为终端接入的第一分布式无线网络设备对应的控制无线接入网设备。
  12. 根据权利要求11所述的模型推理方法，其特征在于，所述模型推理方法还包括：
    响应于所述控制无线接入网设备为第一控制无线接入网设备，若重新接收到模型分析订阅请求，确定重新发送所述模型分析订阅请求的第二分布式无线接入网设备；所述第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备；
    将第一推理结果发送至第二分布式无线接入网设备，并向OAM发送模型订阅更新请求。
  13. 根据权利要求12所述的模型推理方法，其特征在于，所述模型推理方法还包括：
    响应于所述控制无线接入网设备为第二控制无线接入网设备，若重新接收到模型分析订阅请求，确定重新发送所述模型分析订阅请求的第二分布式无线接入网设备，以及重新接收到模型分析订阅请求的第二控制无线接入网设备，所述第二控制无线接入网设备为第二分布式无线接入网设备对应的控制无线接入网设备；所述第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备；
    将第一推理结果发送至第二控制无线接入网设备，并向OAM发送模型订阅更新请求。
  14. 一种模型推理方法，其特征在于，应用于分布式无线接入网设备，所述方法包括：
    响应于接收到终端发送的模型分析订阅请求，向控制无线接入网设备发送所述模型分析订阅请求；
    其中，所述模型分析订阅请求用于向OAM获取第一模型；所述第一模型包括第一数量的模型分割块。
  15. 根据权利要求14所述的模型推理方法，其特征在于，所述方法还包括：
    接收控制无线接入网设备发送的模型推理数据请求，所述模型推理数据请求用于获取模型推理数据；
    向终端获取模型推理数据，并发送至控制无线接入网设备。
  16. 根据权利要求14所述的模型推理方法，其特征在于，所述方法还包括：
    响应于所述分布式无线接入网设备为第一分布式无线接入网设备，接收第一控制无线接入网设备发送的第一推理结果；
    将所述第一推理结果发送至终端。
  17. 根据权利要求16所述的模型推理方法，其特征在于，所述将所述第一推理结果发送至终端之后，所述方法还包括：
    响应于所述分布式无线接入网设备为第一分布式无线接入网设备，接收终端发送的性能数据；所述性能数据为终端基于第一模型调整执行策略后的真实性能数据；
    向第一控制无线接入网设备发送所述性能数据。
  18. 根据权利要求14所述的模型推理方法，其特征在于，所述方法还包括：
    响应于所述无线接入网设备为第二分布式无线接入网设备，若接收到终端重新发送的模型分析订阅请求，确定向与所述第二分布式无线接入网络设备对应的控制无线接入网设备发送模型分析订阅请求；
    其中，所述第二分布式无线接入网设备为终端切换分布式无线接入网设备后重新接入的分布式无线接入网设备。
  19. 一种模型推理装置，其特征在于，应用于操作维护管理OAM实体，所述装置包括：
    确定模块，用于响应于接收到控制无线接入网设备发送的模型订阅请求信息，确定与所述模型订阅请求信息对应的第一模型；
    发送模块，用于将所述第一模型进行分割，得到第一数量的模型分割块，并将所述第一数量的模型分割块分发至第一数量的控制无线接入网设备。
  20. 一种模型推理装置，其特征在于，应用于控制无线接入网设备，所述装置包括：
    发送模块，用于响应于接收到分布式无线接入网设备发送的模型分析订阅请求，对所述模型分析订阅请求进行处理得到模型订阅请求信息，并向OAM发送所述模型订阅请求信息；
    接收模块，用于接收OAM发送的模型分割块；所述模型分割块为分割第一模型确定的模型分割块；所述第一模型为OAM基于所述模型订阅请求信息确定的。
  21. 一种模型推理装置，其特征在于，应用于分布式无线接入网设备，所述装置包括：
    发送模块，用于响应于接收到终端发送的模型分析订阅请求，向控制无线接入网设备发送所述模型分析订阅请求；
    其中，所述模型分析订阅请求用于向OAM获取第一模型；所述第一模型包括第一数量的模型分割块。
  22. 一种模型推理装置，其特征在于，包括：
    处理器；
    用于存储处理器可执行指令的存储器；
    其中，所述处理器被配置为：执行权利要求1-5中任意一项所述的模型推理方法，或执行权利要求6-13中任意一项所述的模型推理方法，或执行权利要求14-18中任意一项所述的模型推理方法。
  23. 一种非临时性计算机可读存储介质，当所述存储介质中的指令由移动终端的处理器执行时，使得移动终端能够执行权利要求1-5中任意一项所述的模型推理方法，或使得移动终端能够执行权利要求6-13中任意一项所述的模型推理方法，或使得移动终端能够执行权利要求14-18中任意一项所述的模型推理方法。
PCT/CN2021/092900 2021-05-10 2021-05-10 一种模型推理方法、模型推理装置及存储介质 WO2022236638A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
CN202180001523.6A CN115669030A (zh) 2021-05-10 2021-05-10 一种模型推理方法、模型推理装置及存储介质
PCT/CN2021/092900 WO2022236638A1 (zh) 2021-05-10 2021-05-10 一种模型推理方法、模型推理装置及存储介质
EP21941214.5A EP4340300A1 (en) 2021-05-10 2021-05-10 Model inference method, model inference apparatus, and storage medium
US18/290,201 US20240323099A1 (en) 2021-05-10 2021-05-10 Method for model inference

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2021/092900 WO2022236638A1 (zh) 2021-05-10 2021-05-10 一种模型推理方法、模型推理装置及存储介质

Publications (1)

Publication Number Publication Date
WO2022236638A1 true WO2022236638A1 (zh) 2022-11-17

Family

ID=84029061

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/092900 WO2022236638A1 (zh) 2021-05-10 2021-05-10 一种模型推理方法、模型推理装置及存储介质

Country Status (4)

Country Link
US (1) US20240323099A1 (zh)
EP (1) EP4340300A1 (zh)
CN (1) CN115669030A (zh)
WO (1) WO2022236638A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118784495A (zh) * 2023-04-07 2024-10-15 大唐移动通信设备有限公司 模型传输方法及装置

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112311564A (zh) * 2019-07-23 2021-02-02 华为技术有限公司 应用mos模型的训练方法、设备及系统
CN112514441A (zh) * 2020-11-02 2021-03-16 北京小米移动软件有限公司 一种通信方法、通信装置及存储介质
CN112673593A (zh) * 2018-09-11 2021-04-16 三星电子株式会社 在电信网络中的和与电信网络相关的改进

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112673593A (zh) * 2018-09-11 2021-04-16 三星电子株式会社 在电信网络中的和与电信网络相关的改进
CN112311564A (zh) * 2019-07-23 2021-02-02 华为技术有限公司 应用mos模型的训练方法、设备及系统
CN112514441A (zh) * 2020-11-02 2021-03-16 北京小米移动软件有限公司 一种通信方法、通信装置及存储介质

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
CMCC: "Solutions for AI-based load balancing", 3GPP DRAFT; R3-212505, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG3, no. Online; 20210517 - 20210528, 7 May 2021 (2021-05-07), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France , XP052002551 *
LENOVO, MOTOROLA MOBILITY: "Discussion on standard impact to support AI functionality", 3GPP DRAFT; R3-212180, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG3, no. Online; 20210517 - 20210527, 7 May 2021 (2021-05-07), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France , XP052002357 *
NOKIA, NOKIA SHANGHAI BELL: "AI/ML Architecture Discussion", 3GPP DRAFT; R3-212371, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG3, no. E-meeting; 20210517 - 20210527, 6 May 2021 (2021-05-06), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France , XP052001648 *
ZTE, CHINA UNICOM: "Initial analyse on the interface impact with AI-based RAN architecture", 3GPP DRAFT; R3-212034, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG3, no. 20210517 - 20210527, 7 May 2021 (2021-05-07), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France , XP052002280 *

Also Published As

Publication number Publication date
US20240323099A1 (en) 2024-09-26
CN115669030A (zh) 2023-01-31
EP4340300A1 (en) 2024-03-20

Similar Documents

Publication Publication Date Title
US20230217366A1 (en) Access method, access apparatus, and storage medium
WO2021258370A1 (zh) 一种通信处理方法、通信处理装置及存储介质
US20220201650A1 (en) Paging response method and device, paging method and device
WO2021046674A1 (zh) 数据处理方法、装置、电子设备和计算机可读存储介质
WO2022088181A1 (zh) 一种通信方法、通信装置及存储介质
WO2022252162A1 (zh) 一种模型训练方法、模型训练装置及存储介质
CN113498132B (zh) 移动性管理方法、源基站、目标基站及终端设备
WO2022236638A1 (zh) 一种模型推理方法、模型推理装置及存储介质
WO2023123123A1 (zh) 一种频域资源确定方法、装置及存储介质
EP4436259A1 (en) Network task processing method and apparatus based on artificial intelligence, and storage medium
CN110177379B (zh) 基站接入方法及系统
CN116888937A (zh) 一种人工智能通信方法、装置及存储介质
JP2018527765A (ja) サービス接続を確立する方法、装置、プログラム、及び記録媒体
WO2022236831A1 (zh) 一种模型学习方法、模型学习装置及存储介质
WO2022133689A1 (zh) 一种模型传输方法、模型传输装置及存储介质
WO2022257085A1 (zh) 一种模型数据管理方法、模型数据管理装置及存储介质
WO2022021079A1 (zh) 信道参数确定方法和装置、存储介质
WO2022036610A1 (zh) 一种通信方法、通信装置及存储介质
WO2024138375A1 (zh) 一种通信方法、装置、设备及存储介质
WO2019227486A1 (zh) 测量方法、装置、系统及存储介质
WO2024088119A1 (zh) 数据处理方法、装置、终端及网络侧设备
WO2024197957A1 (zh) 一种通信方法、装置及存储介质
WO2024182960A1 (zh) 业务预测方法、装置及存储介质
CN115277687B (zh) 资源数据获取方法、资源数据获取装置及存储介质
WO2024067437A1 (zh) 模型去激活方法、信息发送方法、装置及设备

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 202347083767

Country of ref document: IN

WWE Wipo information: entry into national phase

Ref document number: 2021941214

Country of ref document: EP

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2021941214

Country of ref document: EP

Effective date: 20231211