Model training method, reasoning method, electronic equipment and storage medium

Abstract

The invention relates to a wireless communication technology, and discloses a model training method, a reasoning method, electronic equipment and a storage medium. A request is received through a data service management network element, a data processing network element, a data storage network element and a data acquisition network element for executing tasks are determined according to a data service type identifier and a data service capability identifier in the request, and corresponding tasks are issued to the determined network elements, so that the network elements respectively execute specific tasks. Through interaction cooperation of the data service management network element, the data processing network element, the data storage network element and the data acquisition network element, an AI model training function and an AI reasoning function of the data service system are realized.

Description

Technical Field

[0001] The present application relates to the field of wireless communication technology, and in particular to a model training method, an inference method, an electronic device and a storage medium. Background Art

[0002] ITU's "Recommendation on the Framework and Overall Objectives of IMT for 2030 and Beyond" proposes six application scenarios for 6G: immersive communication, ultra-large-scale connection, extremely high reliability and low latency, integration of artificial intelligence (AI) and communication, integration of perception and communication, and ubiquitous connection. Among them, the first three scenarios are the evolution and enhancement of the three major scenarios of 5G, and the last three scenarios are emerging scenarios born for integrated innovation. It is clear that perception and AI are new capabilities of 6G, which brings new requirements and challenges to the existing network architecture.

[0003] Perception capabilities can generate a huge amount of data, which is used by AI technology to make autonomous judgments and decisions, thereby improving network performance or directly supporting upper-layer applications. The intelligent characteristics of 6G are reflected in the application of AI algorithms, which will greatly enhance the overall performance of the network, improve user experience and optimize efficiency. At the same time, ubiquitous computing capabilities and rich data resources provide solid support for AI to meet the diverse needs of different industries and applications. With the advancement of 6G, more diverse data will emerge, which needs to be efficiently circulated and shared between different network architectures and network element nodes. With the acceleration of commercialization of the AI ​​industry, the demand for forward-looking data set products and highly customized data services will gradually dominate the market.

[0004] In future mobile communication networks, data processing based on data service system architecture has become a trend. Summary of the invention

[0005] In order to support the data service system to implement AI data services, the present application provides a model training method, an inference method, an electronic device and a storage medium.

[0006] According to one aspect of an embodiment of the present application, a model training method is disclosed, which is executed by a data service management network element. The model training method includes: receiving a model training request, the model training request at least including a data service type identifier and a data service capability identifier; determining a model training execution network element according to the data service type identifier and the data service capability identifier, the model training execution network element at least including a data processing network element and a data storage network element; sending a data processing task to the determined data processing network element so that the data processing network element completes the model training according to the data processing task and training data and sends the training result to the data storage network element, and sending a data storage task to the data storage network element so that the data storage network element stores the training result.

[0007] In an exemplary embodiment, the model training execution network element includes a data acquisition network element. After determining the model training execution network element according to the data service type identifier and the data service capability identifier, the model training method further includes: sending a data acquisition task to the determined data acquisition network element so that the data acquisition network element sends the training data to the data processing network element.

[0008] According to one aspect of an embodiment of the present application, a reasoning method is disclosed, which is executed by a data service management network element, and the data service management network element is obtained by training using the aforementioned model training method. The reasoning method includes: receiving an reasoning request, and the reasoning request includes at least a data service type identifier and a data service capability identifier; determining an reasoning execution network element according to the data service type identifier and the data service capability identifier, and the reasoning execution network element includes at least a data processing network element and a data storage network element; sending a data collection task to the determined data storage network element so that the data storage network element sends the stored model data to the data processing network element; sending a data processing task to the determined data processing network element so that the data processing network element inputs the data to be inferred into the model data sent by the data storage network element to obtain an reasoning result.

[0009] According to one aspect of an embodiment of the present application, an electronic device is disclosed, which includes one or more processors and a memory, wherein the memory is used to store one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device implements the model training method or reasoning method as described above.

[0010] According to one aspect of an embodiment of the present application, a computer-readable storage medium is disclosed, which stores computer-readable instructions. When the computer-readable instructions are executed by a processor of a computer, the computer executes the model training method or reasoning method as described above.

[0011] According to one aspect of an embodiment of the present application, a model training method is disclosed, which is executed by a data service system, wherein the data service system includes at least a data service management network element, a data processing network element and a data storage network element, and the model training method includes: the data service management network element receives a model training request, and the model training request includes at least a data service type identifier and a data service capability identifier; the data service management network element determines a model training execution network element according to the data service type identifier and the data service capability identifier, and the model training execution network element includes at least the data processing network element and the data storage network element; the data service management network element sends a data processing task to the determined data processing network element, and sends a data storage task to the determined data storage network element; the data processing network element completes the model training according to the data processing task and the training data, and sends the training result to the data storage network element; the data storage network element stores the training result.

[0012] In an exemplary embodiment, completing model training according to the data processing task and the training data includes: establishing a processing task table with the processing task identifier of the data processing task and the data service capability identifier as an index to store the data processing task; determining a training algorithm according to the processing task identifier, the data service capability identifier and the training data; and completing model training based on the determined training algorithm.

[0013] In an exemplary embodiment, when the data storage network element receives a data storage task sent by the data service management network element, it establishes a storage task table using the storage task identifier of the data storage task and the data service capability identifier as an index to store the data storage task, and sends the training data to the data processing network element; when the data storage network element receives a training result sent by the data processing network element, it stores the training result using the storage task identifier as an index.

[0014] In an exemplary embodiment, the data service management network element also sends a data collection task to the determined data storage network element; the data storage network element establishes a collection task table using the collection task identifier of the data collection task and the data service capability identifier as an index to store the data collection task; the data storage network element sends the training data carrying the collection task identifier to the data processing network element.

[0015] In an exemplary embodiment, the model training execution network element includes a data acquisition network element. After the model training execution network element is determined according to the data service type identifier and the data service capability identifier, the model training method further includes: the data service management network element sends a data acquisition task to the determined data acquisition network element; the data acquisition network element establishes a collection task table to store the data collection task with the collection task identifier of the data collection task and the data service capability identifier as indexes; the data acquisition network element collects training data and sends the training data carrying the collection task identifier to the data processing network element.

[0016] According to one aspect of an embodiment of the present application, a reasoning method is disclosed, which is executed by a data service system, and the data service system is trained by the aforementioned model training method. The reasoning method includes: the data service management network element receives an reasoning request, and the reasoning request includes at least a data service type identifier and a data service capability identifier; the data service management network element determines an reasoning execution network element based on the data service type identifier and the data service capability identifier, and the reasoning execution network element includes at least a data processing network element and a data storage network element; the data service management network element sends a data processing task to the determined data processing network element, and sends a data collection task to the determined data storage network element; the data storage network element sends the stored model data to the data processing network element; the data processing network element inputs the data to be inferred into the model data to obtain an reasoning result.

[0017] In an exemplary embodiment, after the data service management network element determines the inference execution network element based on the data service type identifier and the data service capability identifier, and before the data processing network element inputs the data to be inferred into the model data, the inference method further includes: the data service management network element feeds back the determined address information of the data processing network element to the inference requester, so that the inference requester sends the data to be inferred to the data processing network element; and the data processing network element receives the data to be inferred sent by the inference requester.

[0018] The technical solution provided by the embodiments of the present application includes at least the following beneficial effects:

[0019] The technical solution provided in the present application receives model training requests through a data service management network element, determines the data processing network element and data storage network element that execute model training according to the data service type identifier and the data service capability identifier, sends a data processing task to the determined data processing network element so that the data processing network element completes the model training according to the data processing task and the training data and sends the training results to the data storage network element, and sends a data storage task to the data storage network element so that the data storage network element stores the training results, thereby realizing the AI ​​model training function of the data service system.

[0020] The technical solution provided in the present application receives inference requests through a data service management network element, determines the data processing network element and the data storage network element that execute inference according to the data service type identifier and the data service capability identifier, and sends a data collection task to the determined data storage network element so that the data storage network element sends the stored model data to the data processing network element, and sends a data processing task to the determined data processing network element so that the data processing network element inputs the data to be inferred into the model data, thereby obtaining the inference result, thereby realizing the AI ​​inference function of the data service system.

[0021] It should be understood that the foregoing general description and the following detailed description are exemplary only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0023] Figure 1 is a schematic diagram of a 6G network system supporting data services according to some embodiments.

[0024] Figure 2 is a schematic diagram of a 6G network system supporting data services according to some other embodiments.

[0025] Figure 3 is a flowchart of a model training method according to a first exemplary embodiment.

[0026] Figure 4 is a flowchart of a model training method according to a second exemplary embodiment.

[0027] Figure 5 is a block diagram of a model training device according to some embodiments.

[0028] Figure 6 is a flowchart of a model training method according to a third exemplary embodiment.

[0029] Figure 7 is an interaction diagram of a model training service according to some embodiments.

[0030] Figure 8 This is an interaction diagram of a model training service in which a data storage network element does not store training data in an embodiment.

[0031] Fig. 9 This is an interaction diagram of a model training service when a data storage network element stores training data in an embodiment.

[0032] Fig.10is a flowchart showing an inference method according to the first exemplary embodiment.

[0033] Fig.11 is a block diagram of an inference device according to some embodiments.

[0034] Fig.12 is a flowchart showing an inference method according to the second exemplary embodiment.

[0035] Fig.13 is a flowchart showing an inference method according to the third exemplary embodiment.

[0036] Fig.14 is an interaction diagram of an inference service shown according to some embodiments.

[0037] Fig.15 is an interaction diagram of an inference service according to some other embodiments.

[0038] Fig.16 is an interaction diagram of selecting an execution network element according to some embodiments.

[0039] Fig.17 It is a block diagram of an electronic device according to an exemplary embodiment.

[0040] Fig.18 It is a block diagram of a computer system according to an exemplary embodiment.

[0041] The following are the descriptions of the reference numerals:

[0042] 500, model training device; 501, receiving module; 502, processing module; 503, sending module; 1100, reasoning device, 1101, receiving module; 1102, processing module; 1103, sending module; 1700, electronic device; 1701, processor; 1702, memory; 1800, computer system; 1801, CPU; 1802, ROM; 1803, storage part; 1804, RAM; 1805, bus; 1806, I / O interface; 1807, input part; 1808, output part; 1809, communication part; 1810, drive; 1811, removable medium. DETAILED DESCRIPTION

[0043] Although the present application can be easily embodied in different forms of embodiments, only some of the specific embodiments are shown in the drawings and described in detail in this specification. It should be understood that this description should be regarded as an exemplary illustration of the principles of the present application and is not intended to limit the present application to that described herein.

[0044] In addition, the terms "include", "comprising", "having" and any variations thereof mentioned in the description of this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or modules is not limited to the listed steps or modules, but may optionally include other steps or modules not listed, or may optionally include other steps or modules inherent to these processes, methods, products or devices.

[0045] It should be noted that, in the embodiments of the present application, words such as "exemplary" or "by way of example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "by way of example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "by way of example" is intended to present related concepts in a specific way.

[0046] The exemplary embodiments will be described in detail below. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Instead, they are only examples of devices and methods consistent with some aspects of the present application as detailed in the Summary of the Invention.

[0047] In a distributed network, AI model training and reasoning processes involve a variety of data types, which are collectively referred to as AI data, typically including models, gradients, confidence levels, and vectors. In order to adapt to the development of the 6G era, AI data services are gradually focusing on comprehensive data management, requiring the construction of a comprehensive framework that includes data collection, preprocessing, distribution, analysis, and storage, and then providing data to various users in the form of services.

[0048] In the 6G data service system, how to implement model training and reasoning of AI data services is still a problem that needs to be solved. To this end, this application proposes a model training method, reasoning method, electronic device and storage medium applied in the 6G data service system to support the data service system to implement AI data services.

[0049] First, some nouns or terms that appear in the process of explaining the embodiments of the present application are subject to the following explanations.

[0050] Data Service (DS): refers to providing data as a service product after operations such as data collection, preprocessing, and analysis.

[0051] User Equipment (UE): various terminal devices with wireless communication functions that support users to access network services.

[0052] Radio Access Network (RAN): provides wireless access control, data transmission and other services for user equipment in a specific area, and supports wireless perception of specific areas or targets and acquisition of perception measurement data.

[0053] Data Plane Function (DPF): implements functions such as data collection, transmission, preprocessing, and analysis.

[0054] Data Storage Function (DSF): stores collected data and data service results, etc.

[0055] Service Exposure Function (SEF): provides core network services to the outside world.

[0056] The Data Service Management Function (DSMF) translates and decomposes data service requirements into data service tasks, selects specific NFs with different data capabilities such as DPF and DSF, forms an executable UE, RAN, DPF, and DSF logical topology, and controls the work chain to implement specific data service functions. DSMF also needs to support the data service capability reporting function of DPF.

[0057] Network Function (NF): In a communication network, a device or software entity with specific functions and independent operation capabilities.

[0058] Connection Service Management Function (CSMF): Receives connection service tasks from SOF and controls the establishment, modification and release of the connection from UE to the data network DN.

[0059] Service Orchestration Function (SOF): A network element that supports the orchestration of connection services and data services across and within domains in 6G distributed networks.

[0060] Data collection (DC): Obtaining data from data sources, including user data, network data, AI data, perception data, etc.

[0061] Data Storage (DS): stores collected data and data service results, etc.

[0062] Data Processing (DP): data preprocessing, analysis, etc.

[0063] Evolved Access and Mobility Management Function (eAMF).

[0064] Evolved Network Repository Function (eNRF).

[0065] Application Function (AF) network element.

[0066] User plane function (UPF)

[0067] Data Network (DN): such as operator services, Internet access or third-party services.

[0068] Quality of Service (QoS): In network services, the quality of service can be improved by ensuring transmission bandwidth, reducing transmission delay, reducing data packet loss rate and delay jitter, etc.

[0069] Internet Protocol Address (IP Address): An identifier assigned to each device connected to a network.

[0070] Figure 1 is a schematic diagram of a 6G network system supporting data services according to some embodiments.

[0071] like Figure 1 As shown, the 6G network system supporting data services can be in the form of service-oriented interfaces. The system architecture adds DSMF, DPF and DSF network elements related to data services to control and implement functions such as data collection, transmission, processing and storage. At the same time, SOF can also be added to support cross-domain and intra-domain connection services and data service orchestration in 6G distributed networks. Accordingly, the system architecture enhances some other network elements to support the converged control of connection and data services.

[0072] The data service management network element is connected to the registration center, service exposure function network element, application function network element, access and mobility management function network element, connection service management function network element, data storage network element, and data plane function network element through a data bus, and data is transmitted through the data bus.

[0073] A data bus is introduced into the 6G network system architecture for data services to transmit collected data, intermediate processing data, or final data service results related to AI and synaesthesia. The introduced data bus separates data transmission aggregation from business logic and can be used to transmit large amounts of data to improve data exchange efficiency. At the same time, the data bus also supports the transmission of new types of data such as perception and AI, such as structured data with topological information, gradient data of AI models, and inference data.

[0074] Through a unified data service architecture, the data in the system can be managed throughout its life cycle. At the same time, the unified data service architecture and process can support 6G's new perception, AI and other data services, meet the needs of future 6G networks for perception of the external environment of the network and AI model training and reasoning, and greatly expand the network's ability to go beyond connectivity.

[0075] Figure 2 FIG. 1 is a schematic diagram of a 6G network system supporting data services according to some other embodiments. Figure 2 As shown, the 6G network system supporting data services can also be in the form of a point-to-point interface. The 6G network system in this interface form will not be described here.

[0076] Next, the model training process and reasoning process of this application are explained.

[0077] Figure 3 is a flowchart of a model training method according to a first exemplary embodiment.

[0078] The model training method is executed by the data service management network element, such as Figure 3 As shown, the model training method includes the following steps S301 to S303.

[0079] In step S301, a model training request is received.

[0080] The model training request includes at least a data service type identifier and a data service capability identifier. The data service type identifier can be used to identify whether the request is model training or reasoning. The data service type identifier can also further identify what kind of model is being trained, such as the training of a traffic prediction model, and further identify what kind of reasoning is, such as traffic prediction.

[0081] The model training request may be initiated by the service demander, which may be UE, RAN, AF, etc.

[0082] In step S302, a model training execution network element is determined according to the data service type identifier and the data service capability identifier.

[0083] The model training execution network element includes at least a data processing network element and a data storage network element. The data processing network element can realize functions such as data collection, transmission, preprocessing, and analysis. The data storage network element can store collected data and data service results.

[0084] In step S303, a data processing task is sent to the determined data processing network element so that the data processing network element completes the model training according to the data processing task and the training data and sends the training result to the data storage network element, and a data storage task is sent to the data storage network element so that the data storage network element stores the training result.

[0085] In some embodiments, the data storage network element does not store the training data required for model training, and the model training execution network element further includes a data acquisition network element, which collects training data, and the data processing network element performs model training based on the training data collected by the data acquisition network element.

[0086] Figure 4 is a flowchart of a model training method according to a second exemplary embodiment.

[0087] The model training method is executed by the data service management network element, such as Figure 4 As shown, the model training method includes the following steps S401 to S403.

[0088] In step S401, a model training request is received.

[0089] The model training request includes at least a data service type identifier and a data service capability identifier.

[0090] In step S402, a model training execution network element is determined according to the data service type identifier and the data service capability identifier.

[0091] The model training execution network element includes at least a data processing network element, a data storage network element, and a data acquisition network element. The data processing network element is used to realize functions such as data acquisition, transmission, preprocessing, and analysis. The data storage network element is used to store the collected data and data service results. The data acquisition network element is used to collect training data.

[0092] In step S403, a data collection task is sent to the determined data collection network element so that the data collection network element sends the training data to the data processing network element, a data processing task is sent to the determined data processing network element so that the data processing network element completes the model training according to the data processing task and the training data and sends the training result to the data storage network element, and a data storage task is sent to the data storage network element so that the data storage network element stores the training result.

[0093] In other embodiments, the data storage network element pre-stores the training data required for model training, and the model training execution network element may only include a data processing network element and a data storage network element. After determining the model training execution network element based on the data service type identifier and the data service capability identifier, a data collection task is also sent to the data storage network element, and the training data is sent to the data processing network element through the data storage network element.

[0094] See next Figure 5 As shown, this embodiment provides a model training device 500, which is used for an electronic device, and includes a receiving module 501, a processing module 502 and a sending module 503. Among them, the receiving module 501 is used to receive a model training request, and the model training request includes at least a data service type identifier and a data service capability identifier. The processing module 502 is used to determine a model training execution network element according to the data service type identifier and the data service capability identifier, and the model training execution network element includes at least a data processing network element and a data storage network element. The sending module 503 is used to send a data processing task to the determined data processing network element so that the data processing network element completes the model training according to the data processing task and the training data and sends the training result to the data storage network element, and sends a data storage task to the data storage network element so that the data storage network element stores the training result.

[0095] In some embodiments, the data storage network element does not store the training data required for model training, and the model training execution network element further includes a data acquisition network element, through which the training data is collected, and the data processing network element performs model training based on the training data collected by the data acquisition network element. The sending module 503 is also used to send a data acquisition task to the data acquisition network element, so that the data acquisition network element sends the training data to the data processing network element.

[0096] In other embodiments, the data storage network element pre-stores the training data required for model training, and the sending module 503 is also used to send a data collection task to the data storage network element so that the data storage network element sends the training data to the data processing network element.

[0097] The implementation process of the functions and effects of each module in the above-mentioned model training device 500 is specifically described in the implementation process of the corresponding steps in the above-mentioned model training method, which will not be repeated here.

[0098] Figure 6 is a flowchart of a model training method according to a third exemplary embodiment.

[0099] The model training method is executed by a data service system, which includes at least a data service management network element, a data processing network element, and a data storage network element. Figure 6As shown, the model training method includes the following steps S601 to S605.

[0100] In step S601, the data service management network element receives a model training request.

[0101] The model training request includes at least a data service type identifier and a data service capability identifier.

[0102] In step S602, the data service management network element determines the model training execution network element according to the data service type identifier and the data service capability identifier.

[0103] Among them, the model training execution network element includes at least a data processing network element and a data storage network element.

[0104] In step S603, the data service management network element sends a data processing task to the determined data processing network element, and sends a data storage task to the determined data storage network element.

[0105] In step S604, the data processing network element completes model training according to the data processing task and the training data, and sends the training result to the data storage network element.

[0106] In some embodiments, the data processing network element establishes a processing task table to store data processing tasks using the processing task identifier and data service capability identifier of the data processing task as an index; determines a training algorithm based on the processing task identifier, the data service capability identifier and the training data; and completes model training based on the determined training algorithm.

[0107] In step S605, the data storage network element stores the training result.

[0108] In some embodiments, when the data storage network element receives a data storage task sent by the data service management network element, it establishes a storage task table using the storage task identifier and data service capability identifier of the data storage task as an index to store the data storage task, and sends the training data to the data processing network element; when receiving the training results sent by the data processing network element, it stores the training results using the storage task identifier as an index.

[0109] In some embodiments, the data storage network element does not store the training data required for model training, and the model training execution network element further includes a data acquisition network element, which collects training data, and the data processing network element performs model training based on the training data collected by the data acquisition network element.

[0110] In detail, after the data service management network element determines the model training execution network element based on the data service type identifier and the data service capability identifier, it sends the data collection task to the determined data collection network element. The data collection network element establishes a collection task table with the collection task identifier and the data service capability identifier of the data collection task as index to store the data collection tasks, collects training data, and sends the training data carrying the collection task identifier to the data processing network element.

[0111] In other embodiments, the data storage network element pre-stores the training data required for model training, and after the data service management network element determines the model training execution network element according to the data service type identifier and the data service capability identifier, it also sends the data collection task to the determined data storage network element. The data storage network element uses the collection task identifier and the data service capability identifier of the data collection task as an index to establish a collection task table to store the data collection tasks, and sends the training data carrying the collection task identifier to the data processing network element.

[0112] Figure 7 is an interaction diagram of a model training service according to some embodiments.

[0113] like Figure 7 As shown, the model training service includes events 1-8. The AI ​​service demander is, for example, AF, the data service management network element is, for example, DSMF, the data processing network element is, for example, DPF, the data storage network element is, for example, DSF, and the data collection network element can be DSF or other network elements, such as eAMF, RAN, etc.

[0114] In event 1, the AI ​​service demand direction sends an AI model training request to the data service management network element. The model training request, for example, is for traffic identification, classification, prediction, etc. The model training request contains relevant parameters such as AI training task configuration.

[0115] In event 2, the data service management network element orchestrates and selects appropriate AI data service execution network elements, namely the data collection network element, data processing network element, and data storage network element, based on parameters such as the data service type identifier and data service capability identifier in the AI ​​model training request.

[0116] In event 3, the data service management network element dispatches tasks such as data collection / data analysis / data processing / data storage to each execution network element, and informs it of the specific tasks and the objects of output reporting.

[0117] In event 4, the data service execution network elements cooperate to perform the AI ​​model training task.

[0118] In event 5, the data processing network element sends a request to store the training result to the data storage network element.

[0119] In event 6, the data storage network element stores the AI ​​model training results based on the training result request.

[0120] In event 7, the data storage network element responds with the storage result to the data processing network element.

[0121] In event 8, the data processing network element feeds back the completion of the AI ​​model training task to the data service management network element, so as to feed back the completion of the AI ​​model training task to the AI ​​service demander through the data service management network element.

[0122] Figure 8 This is an interaction diagram of a model training service in which a data storage network element does not store training data in an embodiment.

[0123] like Figure 8 As shown, the model training service includes events 1 to 12. eAMF / RAN is used for data collection.

[0124] In event 1, AF sends an AI model training request to SEF.

[0125] The request includes parameters such as request type identifier, request node location information, training task identifier, AI data service type identifier, and data service capability identifier.

[0126] Among them, the request type is used to identify whether it is an AI type request or a perception type request, or other types of requests. The request node location information is the location information of the AF. The training task identifier is used to identify the specific type of training task. The AI ​​data service type identifier is used to identify whether it is an AI model training service type or an AI reasoning service type, or other AI services. The data service capability identifier is used to determine the specific data service capability type, such as data collection capability, data analysis capability, data processing capability, data storage capability, etc.

[0127] In event 2, SEF initiates an AI model training request to DSMF.

[0128] In event 3, DSMF selects appropriate RAN, DPF, and DSF network elements according to the task configuration arrangement.

[0129] In event 4, DSMF dispatches data storage / data processing / data collection tasks to each execution network element:

[0130] 4a.DSMF sends data processing tasks to the selected DPF.

[0131] 4b.DPF uses the processing task identifier (task_id) and the data service capability identifier (task_cap) as indexes to establish a processing task table to store data processing tasks.

[0132] 4c.DSMF sends the data storage task to the selected DSF.

[0133] 4d.DSF uses the storage task ID and data service capability ID as indexes to create a storage task table to store data storage tasks.

[0134] 4e.DSMF sends data collection tasks to eAMF / RAN.

[0135] 4f.eAMF / RAN establishes a collection task table to store data collection tasks using the collection task identifier and data service capability identifier as indexes.

[0136] In event 5, eAMF / RAN performs data collection.

[0137] In event 6, the eAMF / RAN sends the training data carrying the collection task identifier to the DPF.

[0138] In event 7, DPF searches the processing task table according to the processing task identifier and the data service capability identifier, parses the training data, determines the algorithm according to the processing task identifier, data type, data service capability type, and combines the training data and data result content to complete the model training.

[0139] In event 8, after completing the model training, DPF sends a model training result storage request to DSF.

[0140] In event 9, DSF stores the model training results using the storage task identifier as an index.

[0141] In event 10, the DSF responds with the storage result to the DPF.

[0142] In event 11, the DPF responds with the result to the DSMF.

[0143] In event 12, DSMF feeds back to AF that the model training task is completed.

[0144] Fig. 9 This is an interaction diagram of a model training service when a data storage network element stores training data in an embodiment.

[0145] like Fig. 9 As shown, the model training service includes events 1-11.

[0146] In event 1, AF sends an AI model training request to SEF.

[0147] In event 2, SEF initiates an AI model training request to DSMF.

[0148] In event 3, DSMF selects appropriate RAN, DPF, and DSF network elements according to the task configuration arrangement.

[0149] In event 4, DSMF dispatches data storage / data processing / data collection tasks to each execution network element:

[0150] 4a.DSMF sends data processing tasks to the selected DPF.

[0151] 4b.DPF uses the processing task ID and data service capability ID as indexes to establish a processing task table to store data processing tasks.

[0152] 4c.DSMF sends the data storage task to the selected DSF.

[0153] 4d.DSF uses the storage task ID and data service capability ID as indexes to create a storage task table to store data storage tasks.

[0154] 4e.DSMF sends the data collection task to the selected DSF.

[0155] 4f.DSF uses the collection task ID and data service capability ID as indexes to create a collection task table to store data collection tasks.

[0156] In event 5, the DSF sends the training data carrying the collection task identifier to the DPF.

[0157] In event 6, DPF searches the processing task table according to the processing task identifier and the data service capability identifier, parses the training data, determines the algorithm according to the processing task identifier, data type, data service capability type, and combines the training data and data result content to complete the model training.

[0158] In event 7, after completing the model training, DPF sends a model training result storage request to DSF.

[0159] In event 8, DSF stores the model training results using the storage task identifier as an index.

[0160] In event 9, the DSF responds with the storage result to the DPF.

[0161] In event 10, the DPF responds with the result to the DSMF.

[0162] In event 11, DSMF feeds back to AF that the model training task is completed.

[0163] Fig.10 is a flowchart showing an inference method according to the first exemplary embodiment.

[0164] The reasoning method is executed by a data service management network element, and the data service management network element can be Figure 3 or Figure 4 The model training method shown is obtained by training, such as Fig.10As shown, the reasoning method includes the following steps S1001 to S1003.

[0165] In step S1001, an inference request is received.

[0166] The inference request includes at least a data service type identifier and a data service capability identifier. The data service type identifier can be used to identify whether the request is model training or inference. The data service type identifier can also further identify what kind of model is being trained, such as the training of a traffic prediction model, and further identify what kind of inference is being performed, such as traffic prediction.

[0167] The inference request may be initiated by a service demander, which may be a UE, RAN, AF, etc.

[0168] In step S1002, the inference execution network element is determined according to the data service type identifier and the data service capability identifier.

[0169] The reasoning execution network element at least includes a data processing network element and a data storage network element. The data processing network element can realize functions such as data collection, transmission, preprocessing, and analysis. The data storage network element can store collected data and data service results.

[0170] In step S1003, a data acquisition task is sent to the determined data storage network element so that the data storage network element sends the stored model data to the data processing network element; a data processing task is sent to the determined data processing network element so that the data processing network element inputs the data to be inferred into the model data sent by the data storage network element to obtain an inference result.

[0171] See next Fig.11 As shown, this embodiment provides an inference device 1100, which is used in an electronic device, and includes a receiving module 1101, a processing module 1102 and a sending module 1103. Among them, the receiving module 1101 is used to receive an inference request, and the inference request includes at least a data service type identifier and a data service capability identifier. The processing module 1102 is used to determine the inference execution network element according to the data service type identifier and the data service capability identifier, and the inference execution network element includes at least a data processing network element and a data storage network element. The sending module 1103 is used to send a data acquisition task to the determined data storage network element, so that the data storage network element sends the stored model data to the data processing network element; send a data processing task to the determined data processing network element, so that the data processing network element inputs the data to be inferred into the model data sent by the data storage network element to obtain an inference result.

[0172] The implementation process of the functions and effects of each module in the above-mentioned reasoning device 1100 is specifically described in the implementation process of the corresponding steps in the above-mentioned reasoning method, which will not be repeated here.

[0173] Fig.12 is a flowchart showing an inference method according to the second exemplary embodiment.

[0174] The reasoning method is executed by a data service system, and the data service system can be Figure 6 The model training method shown is obtained by training, which at least includes a data service management network element, a data processing network element and a data storage network element. Fig.12 As shown, the model training method includes the following steps S1201 to S1205.

[0175] In step S1201, the data service management network element receives an inference request.

[0176] The inference request includes at least a data service type identifier and a data service capability identifier.

[0177] In step S1202, the data service management network element determines the inference execution network element according to the data service type identifier and the data service capability identifier.

[0178] Among them, the inference execution network element at least includes a data processing network element and a data storage network element.

[0179] In step S1203, the data service management network element sends a data processing task to the determined data processing network element, and sends a data collection task to the determined data storage network element.

[0180] In step S1204, the data storage network element sends the stored model data to the data processing network element.

[0181] In step S1205, the data processing network element inputs the data to be inferred into the model data to obtain the inference result.

[0182] In some embodiments, the data processing network element directly receives the data to be inferred sent by the service demander. After the data service management network element determines the inference execution network element according to the data service type identifier and the data service capability identifier, before the data processing network element inputs the data to be inferred into the model data, the data service management network element feeds back the address information of the determined data processing network element to the inference requester, so that the inference requester sends the data to be inferred to the data processing network element; the data processing network element receives the data to be inferred sent by the inference requester.

[0183] like Fig.13 As shown, the model training method includes the following steps S1301 to S1306.

[0184] In step S1301, the data service management network element receives an inference request.

[0185] The inference request includes at least a data service type identifier and a data service capability identifier.

[0186] In step S1302, the data service management network element determines the inference execution network element according to the data service type identifier and the data service capability identifier.

[0187] Among them, the inference execution network element at least includes a data processing network element and a data storage network element.

[0188] In step S1303, the data service management network element sends a data processing task to the determined data processing network element, and sends a data collection task to the determined data storage network element.

[0189] In step S1304, the data storage network element sends the stored model data to the data processing network element.

[0190] In step S1305, the data service management network element feeds back the determined address information of the data processing network element to the inference requesting party, so that the inference requesting party sends the data to be inferred to the data processing network element.

[0191] In step S1306, the data processing network element receives the data to be inferred sent by the inference requester, and inputs the data to be inferred into the model data to obtain the inference result.

[0192] Fig.14 is an interaction diagram of an inference service shown according to some embodiments.

[0193] like Fig.14 As shown, the reasoning service includes events 1 to 8. The AI ​​service demander is, for example, AF, the data service management network element is, for example, DSMF, the data processing network element is, for example, DPF, and the data storage network element is, for example, DSF.

[0194] In event 1, the AI ​​service demander initiates an AI reasoning request to the data service management network element.

[0195] In event 2, the data service management network element orchestration selects appropriate data processing network elements and data storage network elements for scheduling.

[0196] In event 3, the data service management network element dispatches data processing / data collection tasks to each execution network element.

[0197] In event 4, the data service management network element feeds back the address of the selected data processing network element to the AI ​​service demander.

[0198] In event 5, the AI ​​service demand direction sends the data to be inferred to the data processing network element.

[0199] In event 6, the data processing network element interacts with the data storage network element that previously executed the training process to obtain a training model, and inputs the data to be inferred into the training model to obtain an inference result.

[0200] In event 7, the data processing network element responds with the inference result to the AI ​​service demander.

[0201] In Event 8, the AI ​​service demander performs subsequent operations based on the inference results.

[0202] Fig.15 is an interaction diagram of an inference service according to some other embodiments.

[0203] like Fig.15 As shown, the inference service includes events 1-12.

[0204] In event 1, AF sends an AI inference request to SEF.

[0205] In event 2, SEF initiates an AI reasoning request to DSMF.

[0206] In event 3, DSMF performs orchestration and selects appropriate DSF and DPF network elements according to the inference request.

[0207] In event 4, DSMF dispatches data processing / data collection tasks to DPF and DSF:

[0208] 4a.DSMF sends data processing tasks to the selected DPF.

[0209] 4b.DPF uses the processing task identifier (guess_task_id) and the data service capability identifier (task_cap) as indexes to create a processing task table to store data processing tasks.

[0210] 4c.DSMF sends the data collection task to the selected DSF.

[0211] 4d.DSF determines that the AI ​​data service type is AI reasoning, and uses the collection task ID and data service capability ID as indexes to find the previous AI model training task.

[0212] In event 5, the DSMF feeds back the address of the selected DPF to the AF.

[0213] In event 6, the AF sends the data to be inferred to the DPF.

[0214] In event 7, DPF uses the processing task identifier and the data service capability identifier as indexes to look up the processing task table and stores the data to be inferred.

[0215] In event 8, DPF initiates a request to DSF to call the AI ​​training model.

[0216] In event 9, DSF sends the AI ​​training results obtained by querying the local task to DPF.

[0217] In event 10, DPF searches the processing task table for the data to be inferred based on the processing task identifier and the data service capability identifier as an index, and executes the AI ​​inference task based on the loaded AI model to obtain the inference result.

[0218] In event 11, the DPF outputs the AI ​​reasoning result to the AF.

[0219] In event 12, AF performs subsequent related operations based on the AI ​​reasoning results.

[0220] Next, how the data service management network element selects the execution network element is described.

[0221] After receiving an AI data service request (model training request / inference request), the data service management network element needs to select AI data service executors with different data service capabilities to control the completion of various AI data service tasks. In some embodiments, the data service management network element queries and selects network elements that meet the data service type and data service capabilities required by the AI ​​model training request / inference request through the network registration center as the execution network element. Fig.16 As shown, it specifically includes events 1-5.

[0222] In event 1, the data service management network element sends a data service discovery request message to query the network registration center for data service executor instances with different data capabilities that meet the data service requirements. The request message includes, for example, the following content: data service executor type identifier; data service business type (DC, DS, DP service); data service area; data service execution node information; specific QoS requirements of the data service; data transmission protocol type, etc.

[0223] In event 2, the network registration center authorizes the data service management network element to discover the data service.

[0224] In event 3, in the data service request response message, the network registration center provides the data service management network element with information on multiple groups of data service executor instances that meet the data service requirements and are grouped according to data capabilities (for example, the IP addresses of multiple groups of NFs with different data service capabilities).

[0225] In event 4, the data service management network element selects one or more data service executors (execution network elements) in the data service task work chain from multiple groups of data service executors with different data capabilities, based on the DC, DS, and DP data capabilities of the data service executors obtained through subscription.

[0226] In event 5, the data service management network element sends the data service task to the data service executor.

[0227] See also Fig.17 As shown, this embodiment provides an electronic device 1700, which includes one or more processors 1701 and a memory 1702, and the memory 1702 is used to store one or more programs. When the one or more programs are executed by one or more processors 1701, the electronic device 1700 implements the model training method or reasoning method of the present application.

[0228] Fig.18 The following schematically shows a computer system structure block diagram for implementing some embodiments of the present application. It should be noted that: Fig.18 The computer system shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0229] like Fig.18 As shown, the computer system 1800 includes a central processing unit 1801 (CPU), which can perform various appropriate actions and processes according to the program stored in the read-only memory 1802 (ROM) or the program loaded from the storage part 1803 to the random access memory 1804 (RAM). Various programs and data required for the operation of the device are also stored in the random access memory 1804. The central processing unit 1801, the read-only memory 1802 and the random access memory 1804 are connected to each other through a bus 1805. The input / output interface 1806 (Input / Output interface, i.e., I / O interface) is also connected to the bus 1805.

[0230] The following components are connected to the input / output interface 1806: an input section 1807 including a keyboard, a mouse, etc.; an output section 1808 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker; a storage section 1803 including a hard disk, etc.; and a communication section 1809 including a network interface card such as a LAN card, a modem, etc. The communication section 1809 performs communication processing via a network such as the Internet. A drive 1810 is also connected to the input / output interface 1806 as needed. A removable medium 1811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1810 as needed so that a computer program read therefrom is installed into the storage section 1803 as needed.

[0231] In particular, according to an embodiment of the present application, the processes described in the various method flow charts can be implemented as computer software programs. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flow chart. In such an embodiment, the computer program can be downloaded and installed from a network through the communication part 1809, and / or installed from a removable medium 1811. When the computer program is executed by the central processor 1801, various functions defined in the device of the present application are executed.

[0232] It should be noted that the computer-readable medium shown in the embodiment of the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, - but not limited to - an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor device, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a 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. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by an instruction execution device, a device or a device or used in combination with it. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable program code 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 above. Computer readable signal media may also be any computer readable medium other than computer readable storage media, which may send, propagate, or transmit programs for use by or in conjunction with instruction execution devices, devices, or devices. The program code contained on the computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0233] Those skilled in the art should be aware that in one or more of the above examples, the functions described in this application can be implemented by hardware, software, firmware, or any combination thereof. When implemented by software, these functions can be stored in a computer-readable storage medium or transmitted as one or more instructions or codes on a computer-readable storage medium.

[0234] Through the description of the above implementation methods, technical personnel in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0235] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the module division is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0236] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be performed without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A model training method, executed by a data service management network element, characterized in that: The model training method comprises: Receiving a model training request, wherein the model training request includes at least a data service type identifier and a data service capability identifier; Determine a model training execution network element according to the data service type identifier and the data service capability identifier, wherein the model training execution network element includes at least a data processing network element and a data storage network element; Send a data processing task to the determined data processing network element so that the data processing network element completes model training according to the data processing task and training data and sends the training result to the data storage network element, and send a data storage task to the data storage network element so that the data storage network element stores the training result.

2. The model training method according to claim 1, characterized in that: The model training execution network element includes a data collection network element. After the model training execution network element is determined according to the data service type identifier and the data service capability identifier, the model training method further includes: A data collection task is sent to the determined data collection network element, so that the data collection network element sends the training data to the data processing network element.

3. A reasoning method, executed by a data service management network element, wherein the data service management network element is trained by the model training method according to claim 1 or 2, characterized in that: The reasoning method includes: receiving an inference request, wherein the inference request includes at least a data service type identifier and a data service capability identifier; Determine an inference execution network element according to the data service type identifier and the data service capability identifier, wherein the inference execution network element at least includes a data processing network element and a data storage network element; Sending a data collection task to the determined data storage network element, so that the data storage network element sends the stored model data to the data processing network element; The data processing task is sent to the determined data processing network element, so that the data processing network element inputs the data to be inferred into the model data sent by the data storage network element to obtain the inference result.

4. An electronic device, characterized in that: The electronic device comprises: one or more processors; A memory for storing one or more programs, which, when executed by the one or more processors, enables the electronic device to implement the method according to any one of claims 1 to 3.

5. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor of a computer, the computer is caused to execute the method according to any one of claims 1 to 3.

6. A model training method, executed by a data service system, wherein the data service system comprises at least a data service management network element, a data processing network element and a data storage network element, characterized in that: The model training method comprises: The data service management network element receives a model training request, where the model training request includes at least a data service type identifier and a data service capability identifier; The data service management network element determines a model training execution network element according to the data service type identifier and the data service capability identifier, and the model training execution network element at least includes the data processing network element and the data storage network element; The data service management network element sends a data processing task to the determined data processing network element, and sends a data storage task to the determined data storage network element; The data processing network element completes model training according to the data processing task and the training data, and sends the training result to the data storage network element; The data storage network element stores the training result.

7. The model training method according to claim 6, characterized in that: The completing the model training according to the data processing task and the training data includes: Using the processing task identifier of the data processing task and the data service capability identifier as indexes, a processing task table is established to store the data processing task; Determine a training algorithm according to the processing task identifier, the data service capability identifier and the training data; Complete model training based on the determined training algorithm; and / or When the data storage network element receives the data storage task sent by the data service management network element, it establishes a storage task table with the storage task identifier of the data storage task and the data service capability identifier as an index to store the data storage task, and sends the training data to the data processing network element; When the data storage network element receives the training result sent by the data processing network element, it stores the training result using the storage task identifier as an index.

8. The model training method according to claim 6 or 7, characterized in that: The data service management network element further sends a data collection task to the determined data storage network element; The data storage network element establishes a collection task table to store the data collection task by using the collection task identifier and the data service capability identifier of the data collection task as indexes; The data storage network element sends the training data carrying the collection task identifier to the data processing network element; or The model training execution network element includes a data collection network element. After the model training execution network element is determined according to the data service type identifier and the data service capability identifier, the model training method further includes: The data service management network element sends a data collection task to the determined data collection network element; The data collection network element establishes a collection task table to store the data collection task by using the collection task identifier and the data service capability identifier of the data collection task as indexes; The data collection network element collects training data and sends the training data carrying a collection task identifier to the data processing network element.

9. An inference method, executed by a data service system, wherein the data service system is trained by the model training method according to any one of claims 6 to 8, characterized in that: The reasoning method includes: The data service management network element receives an inference request, where the inference request includes at least a data service type identifier and a data service capability identifier; The data service management network element determines an inference execution network element according to the data service type identifier and the data service capability identifier, wherein the inference execution network element at least includes a data processing network element and a data storage network element; The data service management network element sends a data processing task to the determined data processing network element, and sends a data collection task to the determined data storage network element; The data storage network element sends the stored model data to the data processing network element; The data processing network element inputs the data to be inferred into the model data to obtain an inference result.

10. The inference method according to claim 9, characterized in that: After the data service management network element determines the inference execution network element according to the data service type identifier and the data service capability identifier, and before the data processing network element inputs the data to be inferred into the model data, the inference method further includes: The data service management network element feeds back the determined address information of the data processing network element to the inference requesting party, so that the inference requesting party sends the data to be inferred to the data processing network element; The data processing network element receives the data to be inferred sent by the inference requester.

Images