Model switching method, device and equipment

Abstract

The invention relates to the technical field of communication, and relates to a model switching method, device and equipment. In the application, a terminal device determines that a model switching condition is satisfied, and sends an analysis subscription request to a core network device through an accessed target access network device; receiving at least one piece of AI model related information fed back by the core network equipment through the target access network equipment; the AI model related information is related information of an AI model which is determined from a model list by the core network equipment based on UE analysis subscription request information and UE capability information included in the analysis subscription request and is matched with the terminal equipment; obtaining model switching configuration information through the target access network equipment based on the at least one piece of AI model related information; and determining the switched target AI model based on the model switching configuration information, and executing a model switching process. Through an interaction process among the access network, the terminal and the core network, model switching is realized, different wireless scenes are adapted, network intelligence is realized, and overhead is reduced.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of communication technology, and in particular to a model switching method, device and equipment. Background Art

[0002] With the development of science and technology, the application of artificial intelligence (AI) models is becoming more and more widespread. In related technologies, it is proposed that 6G networks need to integrate AI technology to enable more intelligent application scenarios. However, a major problem in the future application of AI in wireless scenarios is the limitation of generalization performance, that is, how AI models can demonstrate strong learning capabilities in a variety of complex and unknown wireless scenarios is currently a problem.

[0003] At present, the concept of Life Cycle Management (LCM) has been proposed in 3GPP standardization research, and life cycle management covers model switching, which refers to deactivating the currently activated AI model and activating another AI model. Model switching is switching between models, where each model is suitable for a specific scenario / configuration / site, and the models may have different model structures or share a common model structure. A group of models may have different input / output formats, or different preprocessing methods. AI models can be trained for different scenarios, and model switching based on applicable scenarios can effectively improve generalization performance.

[0004] Therefore, as a method to improve the generalization ability of AI models, model switching has broad application prospects in the 6th Generation Mobile Communication Technology (6G) network. It can achieve significant accuracy gains with lower communication and computing overhead, thereby flexibly adapting to different wireless scenarios. However, the relevant technology does not propose a specific implementation method for 6G network application model switching.

[0005] In summary, how to realize model switching in 6G networks to flexibly adapt to different wireless scenarios, effectively realize network intelligence, and reduce signaling overhead is a problem that needs to be solved at present. Summary of the invention

[0006] The embodiments of the present application provide a model switching method, apparatus and device for implementing model switching in a 6G network, thereby flexibly adapting to different wireless scenarios, effectively realizing network intelligence and reducing signaling overhead.

[0007] In a first aspect, an embodiment of the present application provides a model switching method, which is applied to a terminal device (User Equipment, UE), and the method includes:

[0008] Determine that the model switching condition is met, and send an analysis subscription request to the core network device through the accessed target access network device; the analysis subscription request includes: UE analysis subscription request information and UE capability information;

[0009] Receiving, through the target access network device, at least one AI model related information fed back by the core network device; the AI ​​model related information is related information of the AI ​​model adapted to the terminal device, determined from the model list by the core network device based on the UE analyzing the subscription request information and the UE capability information;

[0010] Based on at least one AI model related information, obtain corresponding model switching configuration information through the target access network device;

[0011] Based on the model switching configuration information, determine the target AI model to be switched in, and execute the model switching process.

[0012] In a second aspect, an embodiment of the present application provides a model switching method, which is applied to a target access network device (Radio Access Network, RAN), and the method includes:

[0013] Receiving an analysis subscription request sent by a terminal device after satisfying a model switching condition; the analysis subscription request includes: UE analysis subscription request information and UE capability information;

[0014] Sending the analysis subscription request to the core network device, and receiving at least one AI model related information fed back by the core network device; the AI ​​model related information is the related information of the AI ​​model adapted to the terminal device determined by the core network device from the model list based on the UE analysis subscription request information and the UE capability information;

[0015] Send at least one AI model related information to the terminal device, so that the terminal device obtains corresponding model switching configuration information through the target access network device based on the at least one AI model related information; determines the target AI model to be switched based on the model switching configuration information, and executes the model switching process.

[0016] In a third aspect, an embodiment of the present application provides a model switching method, which is applied to a core network device, and the method includes:

[0017] Receiving an analysis subscription request sent by a target access network device; wherein the analysis subscription request is sent to the target access network device after the terminal device determines that the model switching condition is met, and the analysis subscription request includes: UE analysis subscription request information and UE capability information, and the target access network device is accessed by the terminal device;

[0018] Based on the UE analyzing the subscription request information and the UE capability information, determine at least one AI model adapted to the terminal device from the model list;

[0019] The AI ​​model related information of at least one AI model is fed back to the terminal device through the target access network device, so that the terminal device obtains the corresponding model switching configuration information through the target access network device based on the at least one AI model related information; based on the model switching configuration information, determines the target AI model to be switched in, and executes the model switching process.

[0020] In a fourth aspect, an embodiment of the present application provides a model switching device, which is applied to a terminal device UE, and the device includes:

[0021] The first sending unit is used to determine that the model switching condition is met, and send an analysis subscription request to the core network device through the accessed target access network device; the analysis subscription request includes: UE analysis subscription request information and UE capability information;

[0022] A first receiving unit is configured to receive, through a target access network device, at least one AI model related information fed back by a core network device; the AI ​​model related information is related information of an AI model adapted to the terminal device, determined from a model list by the core network device based on UE analysis of subscription request information and UE capability information;

[0023] An acquisition unit, configured to acquire corresponding model switching configuration information through a target access network device based on at least one AI model related information;

[0024] The execution unit is used to determine the target AI model to be switched based on the model switching configuration information and to execute the model switching process.

[0025] In a fifth aspect, an embodiment of the present application provides a model switching device, which is applied to a target access network device, and the device includes:

[0026] The second receiving unit is used to receive an analysis subscription request sent by the terminal device after the model switching condition is met; the analysis subscription request includes: UE analysis subscription request information and UE capability information;

[0027] A forwarding unit, configured to send the analysis subscription request to the core network device, and receive at least one AI model related information fed back by the core network device; the AI ​​model related information is related information of the AI ​​model adapted to the terminal device, determined by the core network device from the model list based on the UE analysis subscription request information and the UE capability information;

[0028] The second sending unit is used to send at least one AI model related information to the terminal device, so that the terminal device obtains corresponding model switching configuration information through the target access network device based on the at least one AI model related information; determines the target AI model to be switched based on the model switching configuration information, and executes the model switching process.

[0029] In a sixth aspect, an embodiment of the present application provides a model switching device, which is applied to a core network device, and the device includes:

[0030] The third receiving unit is used to receive an analysis subscription request sent by a target access network device; wherein the analysis subscription request is sent to the target access network device after the terminal device determines that the model switching condition is met, and the analysis subscription request includes: UE analysis subscription request information and UE capability information, and the target access network device is accessed by the terminal device;

[0031] A determination unit, configured to determine at least one AI model adapted to the terminal device from a model list based on UE analysis of subscription request information and UE capability information;

[0032] A feedback unit is used to feed back AI model related information of at least one AI model to a terminal device through a target access network device, so that the terminal device obtains corresponding model switching configuration information through the target access network device based on at least one AI model related information; determines the target AI model to be switched in based on the model switching configuration information, and executes the model switching process.

[0033] In the seventh aspect, an embodiment of the present application provides a terminal device, which includes: a processor and a memory; wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor implements the steps of the terminal device execution model switching method provided in the embodiment of the present application.

[0034] In the eighth aspect, an embodiment of the present application provides a target access network device, which includes: a processor and a memory; wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor implements the steps of the target access network device execution model switching method provided in the embodiment of the present application.

[0035] In the ninth aspect, an embodiment of the present application provides a core network device, which includes: a processor and a memory; wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor implements the steps of the core network device execution model switching method provided in the embodiment of the present application.

[0036] In the tenth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the model switching method provided in the embodiment of the present application are implemented.

[0037] In the eleventh aspect, an embodiment of the present application provides a computer program product, which includes a computer program, and the computer program is stored in a computer-readable storage medium; when the processor of an electronic device reads the computer program from the computer-readable storage medium, the processor executes the computer program, so that the electronic device executes the steps of the model switching method provided in the embodiment of the present application.

[0038] The above solution of the present application has at least the following beneficial effects:

[0039] The embodiment of the present application provides a model switching method, device and equipment, which relates to the field of communication technology. In the embodiment of the present application, the terminal device determines that the model switching condition is met, and sends an analysis subscription request to the core network device through the target access network device to which it is connected, and the analysis subscription request includes: UE analysis subscription request information and UE capability information; and receives at least one AI model related information fed back by the core network device through the target access network device; the AI ​​model related information is the relevant information of the AI ​​model adapted to the terminal device determined from the model list by the core network device based on the UE analysis subscription request information and UE capability information; and then based on at least one AI model related information, obtain the corresponding model switching configuration information through the target access network device; based on the model switching configuration information, determine the target AI model to be cut in, and execute the model switching process. It can be seen that in the embodiment of the present application, through the signaling interaction process between the terminal side, the access network side and the core network side, the terminal side can obtain the AI ​​model related information, and determine the target AI model to be cut in based on the AI ​​model related information, and then execute the model switching; therefore, the model switching is realized in the 6G network through the signaling interaction method, so as to flexibly adapt to different wireless scenarios, effectively realize network intelligence, and reduce signaling overhead.

[0040] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or be understood by practicing the present application. The purpose and other advantages of the present application can be realized and obtained by the structures specifically pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0042] Figure 1 A schematic diagram of a model switching system architecture provided in an embodiment of the present application;

[0043] Figure 2 An interactive flow chart of a model switching method provided in an embodiment of the present application;

[0044] Figure 3 An interactive flow chart of a model switching process based on AI model related information provided in an embodiment of the present application;

[0045] Figure 4 A method flow chart of a method for executing a model switching method of a terminal device provided in an embodiment of the present application;

[0046] Figure 5 A structural diagram of a terminal device provided in an embodiment of the present application;

[0047] Figure 6 A structural diagram of a first model switching device provided in an embodiment of the present application;

[0048] Figure 7 A method flow chart of a method for switching a target access network device execution model provided in an embodiment of the present application;

[0049] Figure 8 A structural diagram of a target access network device provided in an embodiment of the present application;

[0050] Fig. 9 A structural diagram of a second model switching device provided in an embodiment of the present application;

[0051] Fig.10 A method flow chart of a core network device execution model switching method provided in an embodiment of the present application;

[0052] Fig.11 A structural diagram of a core network device provided in an embodiment of the present application;

[0053] Fig.12 This is a structural diagram of the third model switching device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0055] Some of the terms used in the embodiments of the present application are explained below to facilitate understanding by those skilled in the art.

[0056] Lifecycle management: It is an important concept proposed in the standardization research of the 3rd Generation Partnership Project (3GPP), which covers the processes of AI model data collection, model training, model reasoning, model management, model activation / deactivation, model switching, model rollback, model monitoring, model update, etc. The importance of AI model lifecycle management lies in that it ensures the quality and efficiency of AI models throughout their life cycle and reduces potential usage costs.

[0057] Data subscription mechanism: It is a message paradigm in which the message publisher does not send messages directly to specific message receivers (or message subscribers), but instead divides the published messages into different categories, without knowing which message subscribers will consume this information, that is, without knowing which message subscribers may exist. Similarly, message subscribers can express interest in one or more categories and register or subscribe to the categories of interest to receive messages of interest without knowing which message publishers exist. The data subscription mechanism is used to decouple the relationship between message senders and message subscribers, and each part can be developed, deployed, and expanded independently, enhancing the flexibility and maintainability of the system. When a new message is published, the message subscriber will receive the notification and perform the corresponding operation.

[0058] The word “exemplary” is used hereinafter to mean “serving as an example, example, or illustration.” Any embodiment described as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0059] The terms "first" and "second" in this document are used for descriptive purposes only and should not be understood as explicitly or implicitly indicating relative importance or implicitly indicating the number of technical features indicated. Therefore, features defined as "first" and "second" may explicitly or implicitly include one or more of the features, and in the description of the embodiments of this application, unless otherwise specified, "plurality" means two or more.

[0060] It should be noted that the terms "including" and "having" and their variations involved in the documents of this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0061] The following is a brief introduction to the design concept of the embodiment of the present application:

[0062] With the development of science and technology, AI models are being used more and more widely. In white paper reports released by organizations such as 3GPP and the International Telecommunication Union-Telecommunication Standardization Sector (ITU-T), it is proposed that 6G networks need to integrate AI technology to enable more intelligent application scenarios. However, a major problem in the future application of AI in wireless scenarios is the limitation of generalization performance, that is, how AI models can demonstrate strong learning capabilities in a variety of complex and unknown wireless scenarios is currently a problem.

[0063] Currently, the concept of lifecycle management has been proposed in 3GPP standardization research, and lifecycle management covers model switching, which means deactivating the currently activated AI model and activating another AI model. Model switching is switching between models, where each model is suitable for a specific scenario / configuration / site, and the models may have different model structures or share a common model structure. A group of models may have different input / output formats, or different preprocessing methods. AI models can be trained for different scenarios, and model switching based on applicable scenarios can effectively improve generalization performance.

[0064] Therefore, as a method to improve the generalization ability of AI models, model switching has broad application prospects in 6G networks. It can achieve significant accuracy gains with lower communication and computing overheads, thereby flexibly adapting to different wireless scenarios. However, the relevant technology does not propose a specific implementation method for 6G network application model switching.

[0065] In view of this, the embodiments of the present application provide a model switching method, device and equipment involving the field of communication technology; used to implement model switching in a 6G network, so as to flexibly adapt to different wireless scenarios, effectively realize network intelligence, and reduce signaling overhead.

[0066] In an embodiment of the present application, the terminal device determines that the model switching conditions are met, and sends an analysis subscription request to the core network device through the target access network device to which it is connected, where the analysis subscription request includes: UE analysis subscription request information and UE capability information; and receives at least one AI model-related information fed back by the core network device through the target access network device; the AI ​​model-related information is the relevant information of the AI ​​model determined by the core network device from the model list based on the UE analysis subscription request information and the UE capability information, and adapted to the terminal device; and then based on at least one AI model-related information, determines the target AI model to be switched in, and executes the model switching process.

[0067] It can be seen that in the embodiment of the present application, through the signaling interaction process between the terminal side, the access network side and the core network side, the terminal side can obtain AI model related information, and determine the target AI model to be switched based on the AI ​​model related information, and then perform model switching; therefore, in the 6G network, model switching is realized through signaling interaction, so as to flexibly adapt to different wireless scenarios, effectively realize network intelligence, and reduce signaling overhead.

[0068] Among them, the terminal device is a device with wireless communication function, which can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; it can also be deployed on the water surface (such as ships, etc.); it can also be deployed in the air (such as airplanes, balloons and satellites, etc.). The terminal device can be a mobile phone (Mobile Phone), a tablet computer (pad), a computer with wireless transceiver function, a virtual reality (Virtual Reality, VR) terminal device, an augmented reality (Augmented Reality, AR) terminal device, a wireless terminal device in industrial control (Industrial Control), a wireless terminal device in self-driving (Self Driving), a wireless terminal device in remote medical (Remote Medical), a wireless terminal device in smart grid (Smart Grid), a wireless terminal device in transportation safety (Transportation Safety), a wireless terminal device in smart city (Smart City), a wireless terminal device in smart home (Smart Home), etc.; it can also be various forms of UE, mobile station (Mobile Station, MS).

[0069] The target access network device may be a base station, which is a device that provides wireless communication functions for terminals, including but not limited to: gNB, Radio Network Controller (RNC), Node B (NB), Base Station Controller (BSC), Base Transceiver Station (BTS), Home Base Station (for example, Home evolved NodeB, or Home Node B, HNB), Base Band Unit (BBU), Transmitting and Receiving Point (TRP), Transmitting Point (TP), Centralized Unit (CU), Distributed Unit (DU), Mobile Switching Center, etc. The target access network side device in the present application may also be a device that provides wireless communication functions for terminals in other communication systems that may appear in the future.

[0070] The core network equipment is a service-oriented architecture that includes multiple core network elements, including the Access and Mobility Management Function (AMF) network element on the control plane, the User Plane Function (UPF) network element on the user plane, the Network Data Analytics Function (NWDAF) network element, and the Over-The-Top Server (OTT Server) network element.

[0071] In order to better understand the technical solutions provided in the embodiments of the present application, the following briefly introduces the application scenarios to which the technical solutions provided in the embodiments of the present application are applicable. It should be noted that the application scenarios introduced below are only used to illustrate the embodiments of the present application and are not limited thereto. In specific implementation, the technical solutions provided in the embodiments of the present application can be flexibly applied according to actual needs.

[0072] The preferred embodiments of the present application are described below in conjunction with the drawings in the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application. In addition, the embodiments and features in the embodiments of the present application may be combined with each other if there is no conflict.

[0073] See also Figure 1 , Figure 1A schematic diagram of a model switching system architecture is provided for an embodiment of the present application, and the model switching system includes at least: a terminal device 100, a target access network device 101, a management access network device 102 for managing the target access network device, and a core network device 103; wherein the target access network device 101 is an access network device accessed by the terminal device 100 in the covered area, including at least one of CU and DU; the management access network device 102 can be used to manage multiple target access network devices 101, and the management access network device can be a network orchestration and management system (Orchestration and Management, OAM); the core network device includes multiple core network elements such as AMF network elements, UPF network elements, NWDAF network elements, and OTT Server network elements.

[0074] exist Figure 1 In the model switching system architecture shown:

[0075] Terminal device 100: used for model subscription, performance monitoring, model switching, etc.

[0076] Target access network device 101: used for model training data collection, model training, signaling configuration for terminal device 100, and signaling interaction with core network device 103, etc.

[0077] In one possible implementation, training data used in the model training process is collected from the terminal device 100, and a terminal-level AI model is trained based on the training data obtained from a single terminal device 100, and a cell-level AI model is trained based on the training data obtained from multiple terminal devices 100.

[0078] In one possible implementation, the target access network device 101 interacts with the core network device 103 through the NG interface; specifically, the acquired training data can be reported to the core network device 103, and the model feature information of the terminal-level AI model and the cell-level AI model obtained through training can be reported to the core network device 103.

[0079] For example, see Figure 1 , take the target access network device 101 including gNB-DU and gNB-CU as an example: gNB-DU collects training data from the terminal device through the wireless air interface, and transmits the training data to gNB-CU through the F1 interface, and gNB-CU performs model training based on the training data; similarly, when performing signaling configuration for the terminal device 100, gNB-CU sends signaling configuration to gNB-DU through the FI interface, and gNB-DU performs signaling configuration to the terminal device 100 through the wireless air interface. It should be noted that in the model switching system architecture, the target access network devices can perform signaling interaction through the Xn interface, such as Figure 1As shown in the figure, gNB-CU performs signaling interaction through the Xn interface.

[0080] Management access network equipment 102: used for network operation, management, maintenance, model training data collection, model training, etc.

[0081] In one possible implementation, the management access network device 102 may obtain training data from multiple target access network devices 101 managed by it, and train a corresponding cell-level AI model based on the training data obtained from a single target access network device 101, and train a corresponding regional-level AI model based on the training data obtained from multiple target access network devices 101.

[0082] In one possible implementation, the management access network device 102 may also report the acquired training data to the core network device 103 , and report the model feature information of the cell-level AI model and the regional-level AI model obtained through training to the core network device 103 .

[0083] Core network equipment 103: used for model list establishment, update maintenance, release, model training data collection, model training, etc.

[0084] In a possible implementation, the core network device 103 trains a corresponding regional AI model based on the training data obtained from the target access network device 101 and the management access network device 102; and establishes a model list based on the trained regional AI model, the model feature information reported by the target access network device 101 and the management access network device 102, and determines the AI ​​model adapted to the terminal device 100 based on the model list. Exemplarily, the core network device 103 interacts with the target access network device 101 through the NG interface.

[0085] Therefore, through the switching model system provided in the embodiment of the present application, through signaling interaction between different devices, processes such as model generation, model list establishment, and model switching can be implemented in the 6G network.

[0086] The following describes in detail the model switching, the establishment of the model list involved in the model switching process, and the model generation through specific embodiments.

[0087] 1. Model switching:

[0088] See also Figure 2 , Figure 2 An interactive flow chart of a model switching method provided in an embodiment of the present application includes the following steps:

[0089] Step S200: The terminal device determines that the model switching condition is met and sends a model switching request to the core network device.

[0090] In one possible implementation, the terminal device determines that the model switching conditions are met, triggers the model switching process through non-access stratum (NAS) signaling, and reports the model switching request to the core network device from the control plane through AMF.

[0091] The model switching condition includes at least one of the following:

[0092] Condition 1: Event trigger: The model monitoring results reach the model performance threshold.

[0093] Among them, the model monitoring results are associated with the AI ​​model used, as well as the configuration, scenario, site, conditions, etc.; the model monitoring results are at least used to indicate the performance of the AI ​​model currently used by the UE; among them, the performance of the AI ​​model includes: accurate performance indicators (such as: accuracy, convergence, etc.), overhead performance indicators (such as: model parameter quantity, computational complexity, etc.), and at least one of generalization performance indicators; the model monitoring results are also used to indicate: system performance indicators (such as: throughput, bit error rate, block error rate, latency, etc.).

[0094] The model performance threshold is configured by at least one of the terminal device, the target access network device to which the terminal device is connected, other access network devices to which non-terminal devices are connected, the management access network device for managing the target access network device, and the core network device.

[0095] In one possible implementation, the terminal device monitors the performance of the AI ​​model currently in use, obtains the corresponding model monitoring results, and compares the model monitoring results with the set model performance threshold. When it is determined that the model monitoring results reach the model performance threshold, a model switching request is sent to the core network device. It should be noted that the terminal device can monitor the performance of the AI ​​model currently in use in real time, periodically, or at a specified time, and the specific monitoring timing is set according to the actual situation.

[0096] In specific applications, model performance thresholds can be set for the above-mentioned performance indicators respectively, and when the model monitoring results associated with each of N performance indicators reach the corresponding model performance threshold, a model switching request is sent to the core network device, where N is a positive integer greater than or equal to 1; or a comprehensive model performance threshold is set, and the model monitoring results associated with each of the monitored performance indicators are weighted to obtain a comprehensive monitoring result, and the comprehensive monitoring result is compared with the comprehensive model performance threshold; when it is determined that the comprehensive monitoring result reaches the comprehensive model performance threshold, a model switching request is sent to the core network device.

[0097] Condition 2: Non-event trigger: reaching the specified model switching time.

[0098] Among them, the specified model switching time is configured by at least one of the terminal device, the target access network device to which the terminal device is connected, other access network devices to which non-terminal devices are connected, the management access network device for managing the target access network device, and the core network device.

[0099] In specific applications, the specified model switching time can be determined according to a trigger period, that is, the model switching process is periodically started; wherein the trigger period is associated with a specific configuration, scenario, site, condition, etc.

[0100] Step S201: After receiving the model switching request sent by the terminal device, the core network device feeds back a corresponding switching response message to the terminal device in response to the model switching request.

[0101] Step S202, after receiving the switching response message fed back by the core network device to the model switching request, the terminal device sends an analysis subscription request to the connected target access network device; the analysis subscription request includes: at least one of UE analysis subscription request information and UE capability information.

[0102] Among them, the UE analysis subscription request information includes: UE identification, analysis subscription service type, access location information, and analysis of at least one of subscription service quality; the analysis subscription service quality includes: service quality (Quality of Service, QoS) requirements and AI model performance requirements.

[0103] Exemplarily, the UE identifier is a globally unique temporary identifier (Globally Unique Temporary Identifier, GUTI); the analysis of the subscription service type is represented by a request service ID, including: service scenarios and service functions; the access location information mainly includes: the gNB-CU and gNB-DU information currently accessed by the terminal device; the QoS requirement is the service quality requirement of the terminal device in the current state; the AI ​​model performance requirement is the requirements of the terminal device for the AI ​​model in terms of accuracy, convergence, coverage, generalization capability, etc.

[0104] UE capability information includes: at least one of computing capability, storage capability, and AI feature-related information.

[0105] Exemplarily, computing capability includes: the processing speed of the central processing unit (CPU) / graphics processing unit (GPU) of the terminal device, number of threads, number of cores, bit width, etc.; storage capability includes: the memory capacity, reading speed, storage format, etc. of the terminal device; AI feature-related information includes: AI features of the local data of the terminal device, such as the number of data set features, number of samples, dimension, data type, learning rate, data batches and other hyperparameter settings.

[0106] Step S203, the target access network device responds to the analysis subscription request sent by the terminal device and sends the analysis subscription request to the core network device; the analysis subscription request includes: at least one of UE analysis subscription request information and UE capability information.

[0107] Exemplarily, the analysis subscription request is reported to the core network device via control plane signaling.

[0108] Step S204: After receiving the analysis subscription request, the core network device determines at least one AI model adapted to the terminal device from the model list based on at least one of the UE analysis subscription request information and UE capability information contained in the analysis subscription request.

[0109] In a possible implementation, the model list records: the association between AI model categories, AI model identifiers, and AI model feature information; AI model categories include: terminal-level AI models, community-level AI models, and regional-level AI models; AI model feature information includes: model service type, model service quality, and model capability information. For details, please refer to the relevant content of the model list establishment below, which will not be repeated here.

[0110] Since multiple AI models are stored in the model list, when determining the AI ​​model that is compatible with the terminal device from the model list, it is necessary to determine the comprehensive compatibility between the AI ​​model and the terminal device for each AI model based on at least one of the UE analysis subscription request information and UE capability information contained in the analysis subscription request, and then select at least one AI model that meets the compatibility conditions based on the comprehensive compatibility.

[0111] Exemplarily, one or more AI models with the highest comprehensive adaptability are selected as at least one AI model adapted to the terminal device; or one or more AI models whose comprehensive adaptability reaches the adaptability threshold are selected as at least one AI model adapted to the terminal device; or the top K AI models are selected according to the comprehensive adaptability sorting as at least one AI model adapted to the terminal device, where K is a positive integer greater than or equal to 1.

[0112] Since the implementation method for determining the comprehensive adaptability between each AI model and the terminal device is consistent, the implementation method for determining the comprehensive adaptability between AI model I in the model list and the terminal device is described in detail below, taking the determination of the comprehensive adaptability between the AI ​​model I in the model list and the terminal device as an example.

[0113] If the analysis subscription request only includes UE analysis subscription request information, and the UE analysis subscription request information includes: analysis subscription service type, access location information, and analysis subscription service quality, the comprehensive adaptability of AI model I and the terminal device is determined by formula 1:

[0114] Formula 1:

[0115] S I =l1(first fitness of AI model I)+l2(second fitness of AI model I)+k3(network load condition)

[0116] Among them, s I The comprehensive compatibility between AI model I and terminal equipment;

[0117] l1, l2, and l3 are quantization functions, which are used to convert the degree of compatibility between the AI ​​model I and the terminal device into quantifiable values;

[0118] The first adaptability is obtained by matching the analysis subscription service type with the model service type of AI model 1, indicating that the UE analyzes the requested service ID in the subscription request information and the adaptability of the service scenario and service function of AI model 1 recorded in the model list;

[0119] The second adaptability is obtained by matching the analyzed subscription service quality with the model service quality of AI model 1, indicating the adaptability of the UE to the QoS requirements and AI model performance requirements in the analysis subscription request information and the QoS guarantee level, AI model performance, coverage, and generalization capability of AI model 1 recorded in the model list;

[0120] The network load condition is based on the UE analyzing the access location information in the subscription request information and obtaining the network load condition of the terminal device from the management access network device and the target access network device.

[0121] If the analysis subscription request only contains UE capability information, the comprehensive adaptability of AI model I and the terminal device is determined by formula 2:

[0122] Formula 2:

[0123] s I = l4 (the third fitness of AI model I)

[0124] Among them, s I The comprehensive compatibility between AI model I and terminal equipment;

[0125] l4 is a quantization function, which is used to convert the degree of adaptation between the AI ​​model I and the terminal device into a quantifiable value;

[0126] The third adaptability is obtained by matching the UE capability information with the model capability information of AI model I, which indicates the adaptability of the computing capability, storage capability and AI-related feature information in the UE capability information with the model parameter quantity, computational complexity, deployment cost and resource utilization of the AI ​​model I recorded in the model list.

[0127] If the analysis subscription request includes UE analysis subscription request information and UE capability information, and the UE analysis subscription request information includes: analysis subscription service type, access location information, and analysis subscription service quality, then the comprehensive adaptability of AI model I and the terminal device is determined by formula 3:

[0128] Formula 3:

[0129] s I =k1 (first fitness of AI model I) + k2 (second fitness of AI model I) + k3 (network load) + k4 (third fitness of AI model I)

[0130] It should be noted that the meaning of Formula 3 can be found in the above Formula 1 and Formula 2, which will not be repeated here.

[0131] In the present application, based on UE analysis of subscription request information and UE capability information, the compatibility between the AI ​​model and the terminal device can be accurately determined, and then the AI ​​model adapted to the terminal device can be selected according to the compatibility, thereby preventing frequent switching of the AI ​​model, reducing the frequency of interaction between the terminal device, the target access network device, and the core network device, and saving resources.

[0132] In step S205, the core network device sends AI model related information of at least one AI model adapted to the terminal device to the target access network device; wherein the AI ​​model related information includes: an AI model identifier, and corresponding AI model storage location information.

[0133] In one possible implementation, the core network device sends at least one AI model-related information from the NG-C interface to the target access network device to which the terminal device is connected through control plane signaling.

[0134] Step S206: After receiving at least one AI model related information, the target access network device sends at least one AI model related information to the terminal device.

[0135] In one possible implementation, after the target access network device receives at least one AI model related information sent by the core network device, it sends at least one AI model related information to the terminal device through Radio Resource Control (RRC) signaling.

[0136] Exemplarily, the target access network device encapsulates the AI ​​model-related information separately in an RRC reconfiguration signaling (RRC-Reconfiguration) or concatenates it in an RRC message and sends it to the terminal device.

[0137] Step S207: After the terminal device receives at least one AI model related information sent by the target access network device, it determines the target AI model to be switched to based on the at least one AI model related information and executes the model switching process.

[0138] In one possible implementation, after the terminal device receives at least one AI model related information sent by the target access network device, it sends an RRC reconfiguration completion signaling (RRC-ReconfigurationComplete) to the target access network device to complete the RRC reconfiguration process.

[0139] In the present application, the terminal device performs signaling interaction with the core network device through the target access network device to obtain AI model related information of at least one AI model adapted to the terminal device, and then can determine the target AI model to be switched in according to the AI ​​model related information, and execute the switching process for the target AI model to ensure the accuracy of model switching in the 6G network, so as to flexibly adapt to different wireless scenarios, effectively realize network intelligence, and reduce signaling overhead. Moreover, the AI ​​model related information in the embodiment of the present application is the related information of the AI ​​model determined by the core network device from the model list, and the model list realizes the coexistence of multiple levels of AI models, that is, the model list stores terminal-level AI models, cell-level AI models, and regional-level AI models, so the model switching process can realize the switching of specific models to meet diverse user needs and service QoS.

[0140] In an embodiment of the present application, when the terminal device determines the target AI model to be switched to based on at least one AI model-related information and executes the model switching process, the terminal device obtains the AI ​​model file and model switching configuration information associated with the AI ​​model identifier from the corresponding model storage node based on the AI ​​model storage location information in each AI model-related information, and can determine the corresponding target AI model based on the model file, and can determine the model switching method based on the model switching configuration information, and execute the model switching process according to the model switching method.

[0141] Since the implementation method of determining the target AI model based on the relevant information of each AI model and executing the model switching process is consistent, a detailed description is given for the terminal device to determine the target AI model based on the relevant information of an AI model and execute the model switching process.

[0142] See also Figure 3 , Figure 3 An interactive flowchart of a model switching process based on AI model related information provided in an embodiment of the present application includes the following steps:

[0143] Step S300: The terminal device sends a model subscription request to the target access network device based on the AI ​​model storage location information; the model subscription request includes: an AI model identifier.

[0144] Step S301: The target access network device receives a model subscription request and determines a model storage node based on the model subscription request.

[0145] Since the model subscription request is sent based on the AI ​​model storage location information, the model subscription request implicitly includes the model storage node.

[0146] In one possible implementation, the model subscription request may also include: AI model storage location information; to explicitly determine the model storage node.

[0147] Step S302: The target access network device sends a model subscription request to the model storage node; the model subscription request includes: an AI model identifier.

[0148] Step S303: The model storage node obtains the model switching configuration information associated with the AI ​​model identifier, as well as the model file.

[0149] Step S304: the model storage node sends the model switching configuration information and the model file to the target access network device.

[0150] In one possible implementation, the model storage node may be at least one of a target access network device accessed by a terminal device, other access network devices at the same level as the target access network device accessed by a non-terminal device, a management access network device for managing the target access network device, or a core network device.

[0151] If the model storage node is the target access network device, there is no need to continue to execute step S302, but directly generate the model switching configuration information associated with the AI ​​model identifier and the model file;

[0152] If the model storage node is another access network device at the same level as the target access network device, the target access network device communicates with the other access network device through the Xn interface, exchanges user plane and control plane information with the other access network device, and obtains the model switching configuration information associated with the AI ​​model identifier, as well as the model file;

[0153] If the model storage node is a management access network device for managing the target access network device, the target access network device obtains the model file associated with the AI ​​model identifier from the storage location specified by the management access network device through a data subscription mechanism, and obtains the model switching configuration information associated with the AI ​​model identifier through a specific interface;

[0154] If the model storage node is a core network device, the target access network device obtains the model switching configuration information associated with the AI ​​model identifier from the control plane through the access and mobility management function AMF, and obtains the model file associated with the AI ​​model identifier from the user plane through the user plane function UPF.

[0155] In one possible implementation, the model switching configuration information includes at least UE identification, model ID to be switched, frequency domain resource allocation, time domain resource allocation, triggering mode indication, reserved UE capability indication, model compression and quantization indication, switching mode indication, usage period indication, model monitoring indicator identification, feedback period indication and fallback mode indication.

[0156] Model files are generally in executable file format, or in an open standard format that can be executed after compilation, supporting migration, sharing, and deployment between multiple deep learning frameworks, such as pt format, onnx format, pb format, etc.

[0157] Step S305: The target access network device feeds back to the terminal device the DCI carrying the model switching configuration information matching the AI ​​model identifier, and the model file matching the AI ​​model identifier.

[0158] In one possible implementation, the target access network device receives and encapsulates the model switching configuration information, then uses DCI to carry the encapsulated model switching configuration signaling, and sends the DCI together with the corresponding model file to the terminal device; wherein the model file is sent via a data radio bearer (DRB).

[0159] In one possible implementation, the DCI for model switching uses a cyclic redundancy check (CRC) scrambled by a cell-radio network temporary identifier (C-RNTI) or a random access radio network temporary identifier (RA-RNTI).

[0160] In one possible implementation, if the switching configuration information indicates to perform model compression and quantization, the target access network device will compress and quantize the model file based on the model compression and quantization indication contained in the model switching configuration information; and send the compressed and quantized model file to the terminal device via DRB.

[0161] In a possible implementation, in order to follow the principle of saving communication overhead and adapt to the model switching process of the terminal device, a DCI configuration method for model switching is defined in the implementation of this application; wherein the meaning of each field of the DCI for model switching and the number of bits are designed as shown in Table 1:

[0162] Table 1

[0163]

[0164]

[0165] Step S306: After receiving the DCI, the terminal device demodulates the DCI and obtains the model switching configuration information.

[0166] Step S307: The terminal device receives the model file at a designated time-frequency resource position of a physical downlink shared channel (PDSCH) based on the indicative scheduling information included in the model switching configuration information.

[0167] The scheduling indication information is the frequency domain resource allocation and time domain resource allocation in the model switching configuration information.

[0168] Step S308: The terminal device determines the target AI model to be switched based on the model file, and executes the model switching process based on the model switching configuration information.

[0169] In a possible implementation, when the model switching process is executed based on the model switching configuration information:

[0170] If the model switching mode included in the model switching configuration information is soft switching, the original AI model will be stopped after the target AI model and the original AI model work together for a set period of time; the set period of time for working together can be configured by the core network device, the target access network device, other access network devices, or the management access network device. That is, after the target AI model and the original AI model work together for a set period of time, if it is monitored that the target AI model works normally, the original AI model will be stopped.

[0171] If the model switching mode included in the model switching configuration information is hard switching, the original AI model is directly stopped and the target AI model is started; that is, after receiving the target AI model, the terminal device immediately stops using the original AI model and uses the target AI model.

[0172] In this application, based on the AI ​​model-related information, the terminal device can accurately obtain the model switching configuration information and model file of the adapted AI model from the model storage node, and then determine the target AI model according to the model file, and enable the AI ​​model according to the model switching configuration information, thereby successfully completing the model switching process.

[0173] In an embodiment of the present application, the terminal device completes the model switching process and feeds back a model switching completion message to the target access network device.

[0174] In a possible implementation, the model switching completion message fed back by the terminal device to the target access network device may be carried by a medium access control layer control element (MAC CE) or RRC signaling.

[0175] In an embodiment of the present application, the terminal device completes the model switching process and, based on the indication feedback information included in the model switching configuration information, feeds back the model monitoring results to the core network device; correspondingly, the core network device maintains and updates the model list based on the model monitoring results.

[0176] In this application, the core network device maintains and updates the model list based on the model monitoring results in order to provide better AI services for terminal devices within the coverage area of ​​the core network device.

[0177] 2. Model list establishment:

[0178] In an embodiment of the present application, in order to maintain synchronization at the core network device, all terminal-level AI models, cell-level AI models, and regional-level AI models within the coverage area of ​​the core network device need to store the corresponding AI model feature information in the core network device for cross-domain model management. The core network device forms a model list for the core network device coverage area based on the AI ​​model feature information.

[0179] In one possible implementation, AI model feature information includes: model service type, model service quality, and model capability information.

[0180] Among them, model service types include: service scenarios and service functions; service scenarios, such as: urban macro cells, rural scenarios, indoor dense coverage scenarios, etc.; service functions, such as: load prediction and analysis services, traffic prediction services, congestion management services, etc.;

[0181] Model service quality includes: QoS guarantee level, that is, the expected service quality; AI model performance, such as convergence performance and accuracy performance during model training; coverage of cell-level AI models and regional AI models; generalization capabilities of cell-level AI models and regional AI models;

[0182] Model capability information includes: the number of model parameters, that is, the number of model parameters contained before model compression; the computational complexity expressed as floating point operations per second (FLOPs); deployment cost, such as expected carbon emissions, etc.; and resource utilization associated with hardware devices.

[0183] Therefore, the model list formed based on the AI ​​model feature information can be shown in Table 2:

[0184] Table 2

[0185]

[0186]

[0187] It should be clarified that the storage locations of the terminal-level AI model, cell-level AI model and regional-level AI model in the model list are different, and are usually stored in the device that generates the AI ​​model, wherein the implementation method of AI model generation is described in detail below. For example, if the terminal-level AI model is generated by the target access network device, then the terminal-level AI model is stored in the corresponding target access network device; if the cell-level AI model is generated by the management access network device, then the cell-level AI model is stored in the corresponding management access network device, and so on. Therefore, the AI ​​model feature information in the model list is specific model feature information uploaded by the device that stores the AI ​​model, and is uniformly decided and deployed by the core network device. Moreover, since the model files are generally large and may generate large transmission overhead, whether to upload the model files to the core network device is an optional option.

[0188] In an embodiment of the present application, after completing the establishment of the model list, the core network device publishes information about available AI models to all terminal devices within the coverage area of ​​the core network device, so that the terminal devices can subscribe to the model according to their own needs. Exemplarily, the core network device sends the model list to all target access network devices within the coverage area of ​​the core network device through the NG interface, and the target access network device of each cell broadcasts the model list information to all terminal devices in the cell by broadcasting; for example, the target access network device broadcasts the model list information to all terminal devices in the cell through System Information Broadcast (SIB) signaling.

[0189] Since the model list may involve privacy information, the model list published by the core network device may be a subset of the model list it maintains, such as publishing model scenarios and functions, service quality and other information.

[0190] The core network equipment can update the maintenance model list in real time and notify all terminal devices in the network.

[0191] 3. Model generation:

[0192] In the embodiment of the present application, based on the different computing power and storage capabilities of terminal devices and customized AI service requirements, there may be multiple levels of AI models in the network, including at least the following three types:

[0193] Terminal-level AI model: The target access network device collects model training data from a single terminal device, and the target access network device completes AI model training based on the collected training data to obtain a terminal-level AI model specific to the terminal device, and the terminal-level AI model is stored in the target access network device. It should be noted that the terminal-level AI model generally has fewer model parameters and a simpler model structure, and is mainly generated to provide customized services to specific terminal devices.

[0194] Cell-level AI model: The target access network device corresponding to a cell in the network collects model training data from multiple terminal devices in the cell, and the target access network device or the management access network device of the corresponding cell completes AI model training based on the collected training data to obtain a cell-level AI model, which is stored in the target access network device or the management access network device. Exemplarily, if the target access network device obtains a cell-level AI model through training, it is stored in the target access network device; if the management access network device obtains a cell-level AI model through training, it is stored in the management access network device.

[0195] In a possible implementation, the target access network device confirms the terminal devices participating in the model training in the cell through SIB1, and collects training data from the terminal devices participating in the model training. Furthermore, the collected training data may be transmitted to the management access network device.

[0196] It should be noted that the community AI-level model needs to fit data samples from multiple terminal devices. Compared with the terminal-level model, its model parameters are larger and the model structure is more complex. It is used to provide comprehensive AI services for terminal devices in the community.

[0197] Regional-level AI model: In a specific area in the network, which is generally a coverage area covering multiple cells, model training data is collected from multiple terminal devices in multiple cells in the region, and the core network equipment completes AI model training based on the collected training data, or the management access network equipment and the core network equipment collaborate to complete the model training, to obtain a regional-level AI model, which is stored in the core network equipment or the management access network equipment of each cell.

[0198] In one possible implementation, the target access network device of each cell uploads the collected training data to the UPF network element through the user plane, so that the core network device performs model training based on the training data; wherein the training data is collected by the target access network device of each cell through SIB1 to confirm the terminal devices participating in the model training of this cell.

[0199] It should be noted that the regional-level AI model is used to provide AI services with better generalization performance for each terminal device in this area. Compared with the terminal-level AI model and the community-level AI model, the model generally has a larger number of parameters, a complex model structure, and is non-customized.

[0200] In this application, a multi-level model generation method is proposed. Models at different levels are suitable for different scenarios. Based on the applicable scenarios and combined with the determined AI model, model switching can effectively improve the generalization performance.

[0201] Based on the same inventive concept, the embodiment of the present application also provides a model switching method. Since the device corresponding to the method is the terminal device in the model switching system of the embodiment of the present application, and the principle of solving the problem by the method is similar to that of the terminal device in the model switching system, the implementation of the method can refer to the implementation of the terminal device in the model switching system, and the repeated parts will not be repeated.

[0202] See also Figure 4 , Figure 4 A method flow chart of a method for executing a model switching method of a terminal device provided in an embodiment of the present application includes the following steps:

[0203] Step S400, determining that the model switching condition is met, sending an analysis subscription request to the core network device through the accessed target access network device; the analysis subscription request includes: UE analysis subscription request information and UE capability information;

[0204] Step S410, receiving at least one AI model related information fed back by the core network device through the target access network device; the AI ​​model related information is related information of the AI ​​model adapted to the terminal device, determined from the model list by the core network device based on the UE analyzing the subscription request information and the UE capability information;

[0205] Step S420, based on at least one AI model related information, obtaining corresponding model switching configuration information through the target access network device;

[0206] Step S430, based on the model switching configuration information, determine the target AI model to be switched in, and execute the model switching process.

[0207] In a possible implementation, the model switching condition includes at least one of the following:

[0208] The model monitoring result reaches the model performance threshold; the model monitoring result is at least used to indicate the performance of the AI ​​model currently used by the UE;

[0209] The specified model switching time is reached.

[0210] In a possible implementation, after determining that the model switching condition is met, before sending the analysis subscription request to the core network device through the accessed target access network device, the method further includes:

[0211] A model switching request is sent to a core network device, and a switching response message fed back by the core network device in response to the model switching request is received.

[0212] In a possible implementation, the AI ​​model related information includes: an AI model identifier, and corresponding AI model storage location information;

[0213] Based on at least one AI model related information, determine the target AI model to be switched to, and execute the model switching process, including:

[0214] Based on at least one AI model related information, corresponding model switching configuration information is obtained through the target access network device, including:

[0215] Based on each AI model related information in the at least one AI model related information, respectively perform:

[0216] Based on the AI ​​model storage location information, a model subscription request is sent to the corresponding model storage node through the target access network device; the model subscription request includes: an AI model identifier;

[0217] Through the target access network device, a subscription success message fed back by the model storage node is received; the subscription success message includes: model switching configuration information matching the AI ​​model identifier.

[0218] In a possible implementation, the model storage node is at least one of: a target access network device, other access network devices at the same level as the target access network device, a management access network device for managing the target access network device, and a core network device.

[0219] In a possible implementation, obtaining corresponding model switching configuration information through a target access network device includes:

[0220] Receiving downlink control information DCI of bearer model switching configuration information sent by the target access network device;

[0221] The model switching configuration information is fed back to the target access network device by the model storage node after receiving the model subscription request.

[0222] In a possible implementation, based on the model switching configuration information, determining the target AI model to be switched in includes:

[0223] Based on the indication scheduling information included in the model switching configuration information, the model file is received at the designated time-frequency resource position of the physical downlink shared channel PDSCH; the model file is obtained by the target access network from the corresponding model storage node based on the AI ​​model storage location information in the AI ​​model related information, and sent through the data radio bearer DRB;

[0224] Based on the model file, determine the target AI model to be used.

[0225] In a possible implementation, based on the model switching configuration information, a model switching process is performed, including:

[0226] If the model switching mode included in the model switching configuration information is soft switching, the original AI model is stopped after the target AI model and the original AI model work together for a set period of time;

[0227] If the model switching mode included in the model switching configuration information is hard switching, the original AI model is stopped directly and the target AI model is started.

[0228] In a possible implementation, after executing the model switching process based on the model switching configuration information, the method further includes:

[0229] Based on the indication feedback information included in the model switching configuration information, the model monitoring result is fed back to the core network device, so that the core network device maintains and updates the model list based on the model monitoring result.

[0230] correspond Figure 4 The present application also provides a terminal device, see Figure 5 , is a structural diagram of a terminal device provided in an embodiment of the present application, the terminal device includes: a processor 500, a memory 501 and a transceiver 502.

[0231] The processor 500 is responsible for managing the bus architecture and general processing, the memory 501 can store data used by the processor 500 when performing operations, and the transceiver 502 is used to receive and send data under the control of the processor 500.

[0232] The bus architecture may include any number of interconnected buses and bridges, specifically one or more processors 500 represented by a processor and various circuits of a memory 501 represented by a memory 501 are linked together. The bus architecture may also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface. The processor 500 is responsible for managing the bus architecture and general processing, and the memory 501 may store data used by the processor 500 when performing operations.

[0233] The process disclosed in the embodiment of the present application can be applied to the processor 500, or implemented by the processor 500. In the implementation process, each step of the signal processing process can be completed by the hardware integrated logic circuit or software instructions in the processor 500. The processor 500 can be a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, and can implement or execute the methods, steps and logic block diagrams disclosed in the embodiment of the present application. The general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware processor to be executed, or the hardware and software modules in the processor 500 can be combined and executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 501, and the processor 500 reads the information in the memory 501 and completes the steps of the signal processing process in combination with its hardware.

[0234] The processor 500 is used to read the program in the memory 501 and execute Figure 4 The model switching method shown.

[0235] correspond Figure 4 The embodiment of the present application also provides a model switching device, which is applied to a terminal device. The principle of solving the problem by the device is similar to the method of executing the model switching embodiment of the above-mentioned terminal device. Therefore, the implementation of the device can refer to the implementation of the above-mentioned method, and the repeated parts will not be repeated.

[0236] See also Figure 6 , is a structural diagram of a first model switching device provided in an embodiment of the present application, the model switching device 600 comprises:

[0237] The first sending unit 601 is used to determine that the model switching condition is met, and send an analysis subscription request to the core network device through the accessed target access network device; the analysis subscription request includes: UE analysis subscription request information and UE capability information;

[0238] The first receiving unit 602 is configured to receive, through the target access network device, at least one AI model related information fed back by the core network device; the AI ​​model related information is related information of the AI ​​model adapted to the terminal device, determined from the model list by the core network device based on the UE analyzing the subscription request information and the UE capability information;

[0239] An acquiring unit 603 is configured to acquire corresponding model switching configuration information through a target access network device based on at least one AI model related information;

[0240] The execution unit 604 is used to determine the target AI model to be switched based on the model switching configuration information, and to execute the model switching process.

[0241] In a possible implementation, the model switching condition includes at least one of the following:

[0242] The model monitoring result reaches the model performance threshold; the model monitoring result is at least used to indicate the performance of the AI ​​model currently used by the UE;

[0243] The specified model switching time is reached.

[0244] In a possible implementation manner, the first sending unit 601 is further configured to:

[0245] After determining that the model switching conditions are met, before sending an analysis subscription request to the core network device through the connected target access network device, a model switching request is sent to the core network device, and a switching response message fed back by the core network device to the model switching request is received.

[0246] In a possible implementation, the AI ​​model related information includes: an AI model identifier, and corresponding AI model storage location information; the acquisition unit 603 is specifically used to:

[0247] Based on each AI model related information in the at least one AI model related information, respectively perform:

[0248] Based on the AI ​​model storage location information, a model subscription request is sent to the corresponding model storage node through the target access network device; the model subscription request includes: an AI model identifier;

[0249] Through the target access network device, a subscription success message fed back by the model storage node is received; the subscription success message includes: model switching configuration information matching the AI ​​model identifier.

[0250] In a possible implementation, the model storage node is at least one of: a target access network device, other access network devices at the same level as the target access network device, a management access network device for managing the target access network device, and a core network device.

[0251] In a possible implementation, the acquisition unit 603 acquires corresponding model switching configuration information through the target access network device, specifically for:

[0252] Receiving downlink control information DCI of bearer model switching configuration information sent by the target access network device;

[0253] The model switching configuration information is fed back to the target access network device by the model storage node after receiving the model subscription request.

[0254] In a possible implementation, the execution unit 604 determines the target AI model to be switched based on the model switching configuration information, specifically for:

[0255] Based on the indication scheduling information included in the model switching configuration information, the model file is received at the designated time-frequency resource position of the physical downlink shared channel PDSCH; the model file is obtained by the target access network from the corresponding model storage node based on the AI ​​model storage location information in the AI ​​model related information, and sent through the data radio bearer DRB;

[0256] Based on the model file, determine the target AI model to be used.

[0257] In a possible implementation, the execution unit 604 executes the model switching process based on the model switching configuration information, specifically for:

[0258] If the model switching mode included in the model switching configuration information is soft switching, the original AI model is stopped after the target AI model and the original AI model work together for a set period of time;

[0259] If the model switching mode included in the model switching configuration information is hard switching, the original AI model is stopped directly and the target AI model is started.

[0260] In a possible implementation, after executing the model switching process based on the model switching configuration information, the execution unit 604 is further configured to:

[0261] Based on the indication feedback information included in the model switching configuration information, the model monitoring result is fed back to the core network device, so that the core network device maintains and updates the model list based on the model monitoring result.

[0262] Based on the same inventive concept, an embodiment of the present application also provides a model switching method. Since the device corresponding to the method is the target access network device in the model switching system of the embodiment of the present application, and the principle of solving the problem by the method is similar to that of the target access network device in the model switching system, the implementation of the method can refer to the implementation of the target access network device in the model switching system, and the repeated parts will not be repeated.

[0263] See also Figure 7 , Figure 7 A method flow chart of a method for switching a target access network device execution model provided in an embodiment of the present application includes the following steps:

[0264] Step S700, receiving an analysis subscription request sent by a terminal device after satisfying a model switching condition; the analysis subscription request includes: UE analysis subscription request information and UE capability information;

[0265] Step S710, sending the analysis subscription request to the core network device, and receiving at least one AI model related information fed back by the core network device; the AI ​​model related information is related information of the AI ​​model adapted to the terminal device, determined from the model list by the core network device based on the UE analysis subscription request information and the UE capability information;

[0266] Step S720, sending at least one AI model related information to the terminal device, so that the terminal device obtains corresponding model switching configuration information through the target access network device based on at least one AI model related information; based on the model switching configuration information, determines the target AI model to be switched in, and executes the model switching process.

[0267] In a possible implementation, the AI ​​model related information includes: an AI model identifier, and corresponding AI model storage location information;

[0268] After sending at least one AI model related information to the terminal device, the method further includes:

[0269] Receiving a model subscription request sent by a terminal device; the model subscription request is sent based on the AI ​​model storage location information, and the model subscription request includes: an AI model identifier;

[0270] Feedback to the terminal device is the DCI that carries the model switching configuration information that matches the AI ​​model identifier, as well as the model file that matches the AI ​​model identifier.

[0271] In a possible implementation, after receiving the model subscription request sent by the terminal device, feeding back to the terminal device the DCI carrying the model switching configuration information matching the AI ​​model identifier, and before the model file matching the AI ​​model identifier, further comprising:

[0272] Based on the model subscription request, the model storage node is determined, and the model switching configuration information matching the AI ​​model identifier and the model file matching the AI ​​model identifier are obtained from the model storage node.

[0273] In a possible implementation, if the model storage node is a target access network device, the model switching configuration information and the model file are directly generated;

[0274] If the model storage node is another access network device at the same level as the target access network device, the user plane and control plane information is exchanged with the other access network device through the Xn interface to obtain the model switching configuration information and the model file;

[0275] If the model storage node is a management access network device for managing the target access network device, the model file is obtained from the storage location specified by the management access network device through a data subscription mechanism, and the model switching configuration information is obtained through a specific interface;

[0276] If the model storage node is a core network device, the model switching configuration information is obtained from the control plane through the access and mobility management function AMF, and the model file is obtained from the user plane through the user plane function UPF.

[0277] In a possible implementation, feeding back a model file matching the AI ​​model identifier to the terminal device includes:

[0278] Based on the model compression and quantization indication included in the model switching configuration information, the model file is compressed and quantized;

[0279] The compressed and quantized model file is fed back to the terminal device; the model file is sent via DRB.

[0280] correspond Figure 7 The present application also provides a target access network device, see Figure 8, is a structural diagram of a target access network device provided in an embodiment of the present application, and the target access network device includes: a processor 800, a memory 801, and a transceiver 802.

[0281] The processor 800 is responsible for managing the bus architecture and general processing, the memory 801 can store data used by the processor 800 when performing operations, and the transceiver 802 is used to receive and send data under the control of the processor 800.

[0282] The bus architecture may include any number of interconnected buses and bridges, specifically one or more processors 800 represented by a processor and various circuits of a memory 801 represented by a memory 801 are linked together. The bus architecture may also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface. The processor 800 is responsible for managing the bus architecture and general processing, and the memory 801 may store data used by the processor 800 when performing operations.

[0283] The process disclosed in the embodiment of the present application can be applied to the processor 800, or implemented by the processor 800. In the implementation process, each step of the signal processing process can be completed by the hardware integrated logic circuit or software instructions in the processor 800. The processor 800 can be a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, and can implement or execute the methods, steps and logic block diagrams disclosed in the embodiment of the present application. The general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware processor to be executed, or the hardware and software modules in the processor 800 can be combined and executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 801, and the processor 800 reads the information in the memory 801 and completes the steps of the signal processing process in combination with its hardware.

[0284] The processor 800 is used to read the program in the memory 801 and execute Figure 7 The model switching method shown.

[0285] correspond Figure 7 The embodiment of the present application also provides a model switching device, which is applied to the target access network equipment. The principle of solving the problem by the device is similar to the method of executing the model switching embodiment of the above-mentioned target access network equipment. Therefore, the implementation of the device can refer to the implementation of the above-mentioned method, and the repeated parts will not be repeated.

[0286] See also Fig. 9 , is a structural diagram of a second model switching device provided in an embodiment of the present application, the model switching device 900 includes:

[0287] The second receiving unit 901 is used to receive an analysis subscription request sent by a terminal device after the model switching condition is met; the analysis subscription request includes: UE analysis subscription request information and UE capability information;

[0288] The forwarding unit 902 is configured to send the analysis subscription request to the core network device, and receive at least one AI model related information fed back by the core network device; the AI ​​model related information is the related information of the AI ​​model adapted to the terminal device, determined by the core network device from the model list based on the UE analysis subscription request information and the UE capability information;

[0289] The second sending unit 903 is used to send at least one AI model related information to the terminal device, so that the terminal device obtains corresponding model switching configuration information through the target access network device based on the at least one AI model related information; determines the target AI model to be switched based on the model switching configuration information, and executes the model switching process.

[0290] In a possible implementation, the AI ​​model related information includes: an AI model identifier, and corresponding AI model storage location information;

[0291] After the second sending unit 903 sends at least one AI model related information to the terminal device, it is also used to:

[0292] Receiving a model subscription request sent by a terminal device; the model subscription request is sent based on the AI ​​model storage location information, and the model subscription request includes: an AI model identifier;

[0293] Feedback to the terminal device is the DCI that carries the model switching configuration information that matches the AI ​​model identifier, as well as the model file that matches the AI ​​model identifier.

[0294] In a possible implementation manner, the second sending unit 903 is further configured to:

[0295] After receiving the model subscription request sent by the terminal device, before feeding back to the terminal device the DCI carrying the model switching configuration information that matches the AI ​​model identifier and the model file that matches the AI ​​model identifier, based on the model subscription request, determine the model storage node, and obtain the model switching configuration information that matches the AI ​​model identifier and the model file that matches the AI ​​model identifier from the model storage node.

[0296] In a possible implementation, if the model storage node is a target access network device, the model switching configuration information and the model file are directly generated;

[0297] If the model storage node is another access network device at the same level as the target access network device, the user plane and control plane information is exchanged with the other access network device through the Xn interface to obtain the model switching configuration information and the model file;

[0298] If the model storage node is a management access network device for managing the target access network device, the model file is obtained from the storage location specified by the management access network device through a data subscription mechanism, and the model switching configuration information is obtained through a specific interface;

[0299] If the model storage node is a core network device, the model switching configuration information is obtained from the control plane through access and AMF, and the model file is obtained from the user plane through UPF.

[0300] In a possible implementation, the second sending unit 903 feeds back a model file matching the AI ​​model identifier to the terminal device, including:

[0301] Based on the model compression and quantization indication included in the model switching configuration information, the model file is compressed and quantized;

[0302] The compressed and quantized model file is fed back to the terminal device; the model file is sent via DRB.

[0303] Based on the same inventive concept, the embodiment of the present application also provides a model switching method. Since the device corresponding to the method is the core network device in the model switching system of the embodiment of the present application, and the principle of solving the problem by the method is similar to that of the core network device in the model switching system, the implementation of the method can refer to the implementation of the core network device in the model switching system, and the repeated parts will not be repeated.

[0304] See also Fig.10 , Fig.10 A method flow chart of a core network device execution model switching method provided in an embodiment of the present application includes the following steps:

[0305] Step S1000, receiving an analysis subscription request sent by a target access network device; wherein the analysis subscription request is sent to the target access network device by the terminal device after determining that a model switching condition is met, and the analysis subscription request includes: UE analysis subscription request information and UE capability information, and the target access network device is accessed by the terminal device;

[0306] Step S1010, based on the UE analyzing the subscription request information and the UE capability information, determining at least one AI model adapted to the terminal device from the model list;

[0307] Step S1020, the AI ​​model related information of at least one AI model is fed back to the terminal device through the target access network device, so that the terminal device obtains the corresponding model switching configuration information through the target access network device based on the at least one AI model related information; based on the model switching configuration information, determines the target AI model to be switched in, and executes the model switching process.

[0308] In a possible implementation, the model list records: an association relationship between an AI model category, an AI model identifier, and AI model feature information;

[0309] Among them, AI model categories include: terminal-level AI model, community-level AI model, and regional-level AI model;

[0310] AI model feature information includes: model service type, model service quality, and model capability information.

[0311] In a possible implementation, if the UE analyzes the subscription request information including: analyzing the subscription service type, the access location information, and analyzing the subscription service quality, then based on the UE analyzing the subscription request information and the UE capability information, determining at least one AI model adapted to the terminal device from the model list includes:

[0312] For each AI model in the model list, execute:

[0313] Matching the analysis subscription service type with the model service type of the AI ​​model to obtain a first degree of fitness; matching the analysis subscription service quality with the model service quality of the AI ​​model to obtain a second degree of fitness; matching the UE capability information with the model capability information of the AI ​​model to obtain a third degree of fitness;

[0314] and, determining a network load of the terminal device based on the access location information;

[0315] Determine the comprehensive fitness of the AI ​​model based on the first fitness, the second fitness, the third fitness, and the network load;

[0316] Based on the comprehensive fitness of each AI model in the model list, at least one AI model that meets the fitness condition is selected from the model list.

[0317] In one possible implementation, the terminal-level AI model is a target access network device to which the terminal device is connected, and is obtained through training based on training data collected from the terminal device;

[0318] The cell-level AI model is an access network device corresponding to the target cell, which is trained based on training data collected from multiple terminal devices in the target cell; wherein the access network device is a target access network device accessed by the terminal device in the target cell and a management access network device for managing the target access network device;

[0319] The regional-level AI model is a core network device that is trained based on training data collected from multiple terminal devices in multiple target cells covered by the core network device.

[0320] In a possible implementation, after the model list is established, available model information is sent to all terminal devices in the core network device load area, so that the terminal devices can subscribe to the model based on their own needs.

[0321] correspond Fig.10 The present application also provides a core network device, see Fig.11 , is a structural diagram of a core network device provided in an embodiment of the present application, and the core network device includes: a transceiver 1102 and a memory 1101 and a transceiver 1102.

[0322] The processor 1100 is responsible for managing the bus architecture and general processing, the memory 1101 can store data used by the processor 1100 when performing operations, and the transceiver 1102 is used to receive and send data under the control of the transceiver 1102 .

[0323] The bus architecture may include any number of interconnected buses and bridges, specifically one or more processors 1100 represented by the processor and various circuits of the memory 1101 represented by the memory 1101 are linked together. The bus architecture may also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface. The processor 1100 is responsible for managing the bus architecture and general processing, and the memory 1101 may store data used by the processor 1100 when performing operations.

[0324] The process disclosed in the embodiment of the present application can be applied to the processor 1100, or implemented by the processor 1100. In the implementation process, each step of the signal processing process can be completed by the hardware integrated logic circuit or software instructions in the processor 1100. The processor 1100 can be a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, and can implement or execute the methods, steps and logic block diagrams disclosed in the embodiment of the present application. The general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware processor to execute, or the hardware and software modules in the processor 1100 can be executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 1101, and the processor 1100 reads the information in the memory 1101 and completes the steps of the signal processing process in combination with its hardware.

[0325] The processor 1100 is used to read the program in the memory 1101 and execute Fig.10 The model switching method shown.

[0326] correspond Fig.10 The embodiment of the present application also provides a model switching device, which is applied to the core network equipment. The principle of solving the problem by the device is similar to the method of executing the model switching embodiment of the above-mentioned core network equipment. Therefore, the implementation of the device can refer to the implementation of the above-mentioned method, and the repeated parts will not be repeated.

[0327] See also Fig.12 , is a structural diagram of a third model switching device provided in an embodiment of the present application, the model switching device 1200 includes:

[0328] The third receiving unit 1201 is used to receive an analysis subscription request sent by a target access network device; wherein the analysis subscription request is sent to the target access network device after the terminal device determines that the model switching condition is met, and the analysis subscription request includes: UE analysis subscription request information and UE capability information, and the target access network device is accessed by the terminal device;

[0329] The determining unit 1202 is configured to determine at least one AI model adapted to the terminal device from the model list based on the UE analyzing the subscription request information and the UE capability information;

[0330] The feedback unit 1203 is used to feed back the AI ​​model related information of at least one AI model to the terminal device through the target access network device, so that the terminal device obtains the corresponding model switching configuration information through the target access network device based on the at least one AI model related information; determines the target AI model to be switched based on the model switching configuration information, and executes the model switching process.

[0331] In a possible implementation, the model list records: an association relationship between an AI model category, an AI model identifier, and AI model feature information;

[0332] Among them, AI model categories include: terminal-level AI model, community-level AI model, and regional-level AI model;

[0333] AI model feature information includes: model service type, model service quality, and model capability information.

[0334] In a possible implementation manner, if the UE analyzes the subscription request information including: analyzing the subscription service type, the access location information, and analyzing the subscription service quality, the determining unit 1202 is specifically configured to:

[0335] For each AI model in the model list, execute:

[0336] Matching the analysis subscription service type with the model service type of the AI ​​model to obtain a first degree of fitness; matching the analysis subscription service quality with the model service quality of the AI ​​model to obtain a second degree of fitness; matching the UE capability information with the model capability information of the AI ​​model to obtain a third degree of fitness;

[0337] and, determining a network load of the terminal device based on the access location information;

[0338] Determine the comprehensive fitness of the AI ​​model based on the first fitness, the second fitness, the third fitness, and the network load;

[0339] Based on the comprehensive fitness of each AI model in the model list, at least one AI model that meets the fitness condition is selected from the model list.

[0340] In one possible implementation, the terminal-level AI model is a target access network device to which the terminal device is connected, and is obtained through training based on training data collected from the terminal device;

[0341] The cell-level AI model is an access network device corresponding to the target cell, which is trained based on training data collected from multiple terminal devices in the target cell; wherein the access network device is a target access network device accessed by the terminal device in the target cell and a management access network device for managing the target access network device;

[0342] The regional-level AI model is a core network device that is trained based on training data collected from multiple terminal devices in multiple target cells covered by the core network device.

[0343] In a possible implementation, after the model list is established, available model information is sent to all terminal devices in the core network device load area, so that the terminal devices can subscribe to the model based on their own needs.

[0344] An embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on an electronic device, the computer program is used to enable the electronic device to execute the model switching method of the above embodiment.

[0345] The computer-readable storage medium in the above embodiments can be any available medium or data storage device that can be accessed by the processor in the device, including but not limited to magnetic storage such as floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO), etc., optical storage such as CDs, DVDs, BDs, HVDs, etc., and semiconductor storage such as ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drives (SSDs), etc.

[0346] In some possible implementations, various aspects of the model switching method provided in the embodiments of the present application can also be implemented in the form of a program product, which includes a computer program. When the program product is run on an electronic device, the computer program is used to enable the electronic device to execute the steps of the model switching method according to various exemplary embodiments of the present application described above in this specification.

[0347] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0348] The program product of the embodiment of the present application may adopt a portable compact disk read-only memory (CD-ROM) and include a computer program, and can be run on an electronic device. However, the program product of the present application is not limited thereto, and in this document, a readable storage medium may be any tangible medium containing or storing a program, which can be used by or in combination with a command execution system, apparatus, or device.

[0349] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, wherein a readable computer program is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with a command execution system, apparatus, or device.

[0350] The computer program embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0351] The computer program for performing the operation of the present application can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and also conventional procedural programming languages ​​such as "C" language or similar programming languages. The computer program can be executed entirely on the user electronic device, partially on the user electronic device, as a separate software package, partially on the user electronic device and partially on a remote electronic device, or entirely on the remote electronic device. In the case of a remote electronic device, the remote electronic device can be connected to the user electronic device through any type of network including a local area network (LAN) or a wide area network (WAN), or can be connected to an external electronic device (for example, using an Internet service provider to connect through the Internet).

[0352] It should be noted that, although several units or subunits of the device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided into multiple units to be embodied.

[0353] In addition, although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0354] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain a computer-usable computer program.

[0355] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program commands. These computer program commands can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the commands executed by the processor of the computer or other programmable data processing device generate commands for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0356] These computer program commands may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the commands stored in the computer readable memory produce an article of manufacture comprising a command device, the command device implementing the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0357] These computer program commands can also be loaded onto a computer or other programmable data processing device so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the commands executed on the computer or other programmable device provide the instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0358] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0359] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A model switching method, characterized in that: Applied to a terminal device UE, the method includes: Determine that the model switching condition is met, and send an analysis subscription request to the core network device through the accessed target access network device; the analysis subscription request includes: UE analysis subscription request information and UE capability information; Receiving, through the target access network device, at least one artificial intelligence AI model related information fed back by the core network device; the AI ​​model related information is related information of the AI ​​model adapted to the terminal device, determined from the model list by the core network device based on the UE analyzing the subscription request information and the UE capability information; Based on the at least one AI model related information, obtaining corresponding model switching configuration information through the target access network device; Based on the model switching configuration information, the target AI model to be switched in is determined, and the model switching process is executed.

2. The method according to claim 1, characterized in that The model switching condition includes at least one of the following: The model monitoring result reaches the model performance threshold; the model monitoring result is at least used to indicate the performance of the AI ​​model currently used by the UE; The specified model switching time is reached.

3. The method according to claim 1, characterized in that After determining that the model switching condition is met, before sending the analysis subscription request to the core network device through the connected target access network device, it also includes: A model switching request is sent to the core network device, and a switching response message fed back by the core network device in response to the model switching request is received.

4. The method according to claim 1, characterized in that The AI ​​model related information includes: an AI model identifier, and corresponding AI model storage location information; The acquiring corresponding model switching configuration information through the target access network device based on the at least one AI model related information includes: Based on each AI model related information in the at least one AI model related information, respectively perform: Based on the AI ​​model storage location information, a model subscription request is sent to the corresponding model storage node through the target access network device; the model subscription request includes: the AI ​​model identifier; Receiving, through the target access network device, a subscription success message fed back by the model storage node; The subscription success message includes: model switching configuration information matching the AI ​​model identifier.

5. The method according to claim 4, characterized in that The model storage node is: at least one of the target access network device, other access network devices at the same level as the target access network device, a management access network device for managing the target access network device, and the core network device.

6. The method according to any one of claims 1 to 5, characterized in that: The obtaining corresponding model switching configuration information through the target access network device includes: Receiving downlink control information DCI sent by the target access network device and carrying the model switching configuration information; The model switching configuration information is fed back to the target access network device by the model storage node after receiving the model subscription request.

7. The method according to any one of claims 1 to 5, characterized in that: The determining the target AI model to be switched based on the model switching configuration information includes: Based on the indication scheduling information included in the model switching configuration information, a model file is received at a specified time-frequency resource position of a physical downlink shared channel PDSCH; the model file is obtained by the target access network from a corresponding model storage node based on the AI ​​model storage location information in the AI ​​model related information, and sent through a data radio bearer DRB; Based on the model file, determine the target AI model to be switched into.

8. The method according to any one of claims 1 to 5, characterized in that: Based on the model switching configuration information, a model switching process is executed, including: If the model switching mode included in the model switching configuration information is soft switching, after the target AI model and the original AI model work together for a set time, the original AI model is stopped; If the model switching mode included in the model switching configuration information is hard switching, the original AI model is directly stopped and the target AI model is started.

9. The method according to any one of claims 1 to 5, characterized in that: After executing the model switching process based on the model switching configuration information, the method further includes: Based on the indication feedback information included in the model switching configuration information, the model monitoring result is fed back to the core network device, so that the core network device maintains and updates the model list based on the model monitoring result.

10. A model switching method, characterized in that: Applied to a target access network device RAN, the method comprises: Receiving an analysis subscription request sent by a terminal device after satisfying a model switching condition; the analysis subscription request includes: UE analysis subscription request information and UE capability information; Sending the analysis subscription request to a core network device, and receiving at least one AI model related information fed back by the core network device; the AI ​​model related information is related information of the AI ​​model adapted to the terminal device, determined by the core network device from the model list based on the UE analysis subscription request information and the UE capability information; The at least one AI model related information is sent to the terminal device, so that the terminal device obtains corresponding model switching configuration information through the target access network device based on the at least one AI model related information; determines the target AI model to be switched based on the model switching configuration information, and executes the model switching process.

11. The method according to claim 10, characterized in that The AI ​​model related information includes: an AI model identifier, and corresponding AI model storage location information; After sending the at least one AI model related information to the terminal device, the method further includes: receiving a model subscription request sent by the terminal device; the model subscription request is sent based on the AI ​​model storage location information, and the model subscription request includes: the AI ​​model identifier; Feedback to the terminal device is a DCI carrying model switching configuration information that matches the AI ​​model identifier, as well as a model file that matches the AI ​​model identifier.

12. The method according to claim 11, characterized in that After receiving the model subscription request sent by the terminal device, feeding back to the terminal device the DCI carrying the model switching configuration information matching the AI ​​model identifier, and before the model file matching the AI ​​model identifier, further comprising: Based on the model subscription request, a model storage node is determined, and model switching configuration information matching the AI ​​model identifier and a model file matching the AI ​​model identifier are obtained from the model storage node.

13. The method according to claim 12, characterized in that If the model storage node is the target access network device, directly generating the model switching configuration information and the model file; If the model storage node is another access network device at the same level as the target access network device, the user plane and control plane information interaction is performed with the other access network device through the Xn interface to obtain the model switching configuration information and the model file; If the model storage node is a management access network device for managing the target access network device, the model file is obtained from a storage location specified by the management access network device through a data subscription mechanism, and the model switching configuration information is obtained through a specific interface; If the model storage node is the core network device, the model switching configuration information is obtained from the control plane through the access and mobility management function AMF, and the model file is obtained from the user plane through the user plane function UPF.

14. The method according to claim 11, characterized in that Feedback to the terminal device a model file matching the AI ​​model identifier, including: Based on the model compression and quantization indication included in the model switching configuration information, compress and quantize the model file; The compressed and quantized model file is fed back to the terminal device; the model file is sent via the data radio bearer DRB.

15. A model switching method, characterized in that: Applied to a core network device, the method comprises: Receiving an analysis subscription request sent by a target access network device; wherein the analysis subscription request is sent to the target access network device after the terminal device determines that a model switching condition is met, and the analysis subscription request includes: UE analysis subscription request information and UE capability information, and the target access network device is accessed by the terminal device; Determine at least one AI model adapted to the terminal device from a model list based on analyzing the subscription request information and the UE capability information by the UE; The AI ​​model related information of the at least one AI model is fed back to the terminal device through the target access network device, so that the terminal device obtains corresponding model switching configuration information through the target access network device based on the at least one AI model related information; determines the target AI model to be switched in based on the model switching configuration information, and executes the model switching process.

16. The method according to claim 15, characterized in that The model list records: the association between the AI ​​model category, the AI ​​model identifier, and the AI ​​model feature information; The AI ​​model categories include: terminal-level AI model, community-level AI model, and regional-level AI model; The AI ​​model feature information includes: model service type, model service quality, and model capability information.

17. The method according to claim 16, characterized in that If the UE analyzing the subscription request information includes: analyzing the subscription service type, the access location information, and analyzing the subscription service quality, then determining at least one AI model adapted to the terminal device from the model list based on the UE analyzing the subscription request information and the UE capability information includes: For each AI model in the model list, execute: Matching the analysis subscription service type with the model service type of the AI ​​model to obtain a first degree of fitness; matching the analysis subscription service quality with the model service quality of the AI ​​model to obtain a second degree of fitness; matching the UE capability information with the model capability information of the AI ​​model to obtain a third degree of fitness; and determining a network load of the terminal device based on the access location information; Determining a comprehensive fitness of the AI ​​model based on the first fitness, the second fitness, the third fitness, and the network load; Based on the comprehensive fitness of each AI model in the model list, at least one AI model that meets the fitness condition is selected from the model list.

18. The method according to claim 15, characterized in that The terminal-level AI model is a target access network device to which the terminal device is connected, and is obtained through training based on training data collected from the terminal device; The cell-level AI model is an access network device corresponding to the target cell, and is obtained through training based on training data collected from multiple terminal devices in the target cell; wherein the access network device is a target access network device accessed by the terminal device in the target cell, and a management access network device for managing the target access network device; The regional-level AI model is a core network device, which is trained based on training data collected from multiple terminal devices in multiple target cells covered by the core network device.

19. The method according to claim 15, characterized in that The method further comprises: After determining that the model list is established, available model information is sent to all terminal devices in the core network device load area, so that the terminal devices can subscribe to models based on their own needs.

20. A model switching device, characterized in that: Applied to a terminal device UE, the device includes: The first sending unit is used to determine that the model switching condition is met, and send an analysis subscription request to the core network device through the accessed target access network device; the analysis subscription request includes: UE analysis subscription request information and UE capability information; A first receiving unit is configured to receive, through the target access network device, at least one AI model related information fed back by the core network device; the AI ​​model related information is related information of the AI ​​model adapted to the terminal device, determined from the model list by the core network device based on analyzing the subscription request information and the UE capability information of the UE; An acquiring unit, configured to acquire corresponding model switching configuration information through the target access network device based on the at least one AI model related information; The execution unit is used to determine the target AI model to be switched based on the model switching configuration information and to execute the model switching process.

21. A model switching device, characterized in that: Applied to a target access network device, the device comprises: The second receiving unit is used to receive an analysis subscription request sent by the terminal device after the model switching condition is met; the analysis subscription request includes: UE analysis subscription request information and UE capability information; A forwarding unit, configured to send the analysis subscription request to a core network device, and receive at least one AI model related information fed back by the core network device; the AI ​​model related information is related information of an AI model adapted to the terminal device, determined by the core network device from a model list based on the UE analysis subscription request information and the UE capability information; The second sending unit is used to send the at least one AI model related information to the terminal device, so that the terminal device obtains corresponding model switching configuration information through the target access network device based on the at least one AI model related information; determines the target AI model to be switched based on the model switching configuration information, and executes the model switching process.

22. A model switching device, characterized in that: Applied to core network equipment, the device includes: A third receiving unit is configured to receive an analysis subscription request sent by a target access network device; wherein the analysis subscription request is sent to the target access network device after the terminal device determines that the model switching condition is met, and the analysis subscription request includes: UE analysis subscription request information and UE capability information, and the target access network device is accessed by the terminal device; A determining unit, configured to determine at least one AI model adapted to the terminal device from a model list based on analyzing the subscription request information and the UE capability information by the UE; A feedback unit is used to feed back the AI ​​model related information of the at least one AI model to the terminal device through the target access network device, so that the terminal device obtains corresponding model switching configuration information through the target access network device based on the at least one AI model related information; determines the target AI model to be switched based on the model switching configuration information, and executes the model switching process.

23. A terminal device UE, characterized in that: The device comprises: a processor and a memory; wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any one of the methods in claims 1 to 9.

24. A target access network device, characterized in that: The device comprises: a processor and a memory; wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any one of the methods of claims 10 to 14.

25. A core network device, characterized in that: The device comprises: a processor and a memory; wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any one of the methods of claims 15 to 19.

Images