AI service processing method, related device, storage medium and computer program product

Abstract

Provided are an artificial intelligence (AI) service processing method, related device, storage medium and computer program product, the method comprising: a first network device receiving first information sent by a second network device, the first information being used by the first network device to determine an AI service; determining a to-be-processed AI service based on the first information; performing service arrangement processing on the AI service to obtain a solution set corresponding to processing the AI service; and sending the solution set to the third network device, wherein the solution set is used for the third network device to select an optimal solution from the solution set.

Description

Technical Field

[0001] The present application relates to the field of wireless communication technology, and in particular to an artificial intelligence (AI) service processing method, related equipment, storage medium, and computer program product. Background Art

[0002] Among the related technologies, the technical means required for wireless networks to provide AI services may include fifth-generation mobile communication technology (5G) networks, edge intelligent terminal devices, distributed computing, and AI model compression and optimization. The combined effect of these technical means promotes wireless networks to achieve efficient, real-time, and reliable AI service support.

[0003] However, in related technologies, on the one hand, devices in wireless networks generally have low computing power and storage capacity, making it difficult to handle complex AI tasks, which may lead to problems such as slow processing speed and delayed response time. On the other hand, although 5G networks provide higher bandwidth, network congestion and bandwidth limitations may still occur in certain scenarios. Large-scale data transmission and real-time AI tasks require greater bandwidth support, otherwise the performance and response speed of AI services may be affected. On the other hand, the update and deployment of AI models may require larger bandwidth and computing resources. Therefore, in related technologies, it is difficult to ensure that sufficient network resources can be provided for AI services when performing AI service processing, making it difficult to meet the network resource requirements of AI services. Summary of the Invention

[0004] In view of this, the embodiments of the present application hope to provide an AI service processing method, related equipment, storage medium and computer program product, which can provide more sufficient network resources for AI services and meet the network resource requirements of AI services.

[0005] The technical solution of the embodiment of the present application is implemented as follows:

[0006] In a first aspect, an embodiment of the present application provides an AI service processing method, which is applied to a first network device, and the method includes:

[0007] receiving first information sent by the second network device, where the first information is used by the first network device to determine the AI ​​service;

[0008] Determining an AI service to be processed based on the first information;

[0009] Perform service orchestration on AI services to obtain a set of solutions corresponding to AI services;

[0010] The solution set is sent to a third network device, and the solution set is used for the third network device to select an optimal solution from the solution set.

[0011] In a second aspect, an AI service processing method is provided in the embodiments of the present application, and the method is applied to a third network device, and the method comprises the following steps:

[0012] receiving a solution set sent by a first network device, wherein the solution set is a set of all solutions corresponding to processing of an AI service;

[0013] selecting an optimal solution from the solution set based on the solution set.

[0014] In a third aspect, an AI service processing method is provided in the embodiments of the present application, and the method is applied to a first device, and the method comprises the following steps:

[0015] receiving an AI task deployment scheme sent by a third network device;

[0016] deploying a corresponding AI task or AI subtask in the first device based on the AI task deployment scheme.

[0017] In a fourth aspect, a first network device is provided in the embodiments of the present application, and the first network device comprises the following units:

[0018] a first receiving unit, configured to receive first information sent by a second network device, wherein the first information is used for the first network device to determine an AI service;

[0019] a determining unit, configured to determine an AI service to be processed based on the first information;

[0020] a processing unit, configured to perform service orchestration processing on the AI service to obtain a solution set corresponding to the AI service to be processed;

[0021] a sending unit, configured to send the solution set to a third network device, and the solution set is used for the third network device to select an optimal solution from the solution set.

[0022] In a fifth aspect, a third network device is provided in the embodiments of the present application, and the third network device comprises the following units:

[0023] a second receiving unit, configured to receive a solution set sent by a first network device, wherein the solution set is a set of all solutions corresponding to processing of an AI service;

[0024] a selecting unit, configured to select an optimal solution from the solution set based on the solution set.

[0025] In a sixth aspect, an embodiment of the present application provides a first device, the first device comprising:

[0026] a third receiving unit configured to receive an AI task deployment scheme sent by a third network device;

[0027] a deployment unit configured to perform corresponding AI task or AI subtask deployment on the first device based on the AI task deployment scheme.

[0028] In a seventh aspect, an embodiment of the present application provides a first network device, the first network device comprising a first processor and a first memory; the first processor implements the AI service processing method on the first network device side when executing a running program stored in the first memory.

[0029] In an eighth aspect, an embodiment of the present application provides a third network device, the third network device comprising a second processor and a second memory; the second processor implements the AI service processing method on the third network device side when executing a running program stored in the second memory.

[0030] In a ninth aspect, an embodiment of the present application provides a first device, the first device comprising a third processor and a third memory; the third processor implements the AI service processing method on the first device side when executing a running program stored in the third memory.

[0031] In a tenth aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the AI service processing method on the first network device side, or the computer program being executed by a processor to implement the AI service processing method on the third network device side, or the computer program being executed by a processor to implement the AI service processing method on the first device side.

[0032] In an eleventh aspect, an embodiment of the present application provides a computer program product comprising a computer program, the computer program being executed by a processor to implement the AI service processing method on the first network device side, or the computer program being executed by a processor to implement the AI service processing method on the third network device side, or the computer program being executed by a processor to implement the AI service processing method on the first device side.

[0033] The embodiment of the application provides an AI service processing method, related equipment, a storage medium and a computer program product. The method comprises the following steps: a first network device receives first information sent by a second network device, the first information being used for the first network device to determine an AI service; based on the first information, an AI service to be processed is determined; service arrangement processing is performed on the AI service, so that a solution set corresponding to the AI service is obtained; the solution set is sent to a third network device, and the solution set is used for the third network device to select an optimal solution from the solution set; the third network device receives the solution set sent by the first network device; wherein the solution set is a set of all solutions corresponding to the AI service; based on the solution set, an optimal solution is selected from the solution set; and the first device receives an AI task deployment scheme sent by the third network device; and based on the AI task deployment scheme, corresponding AI tasks or AI subtasks are deployed on the first device. By using the implementation scheme, after the first network device determines the AI service to be processed according to the first information sent by the second network device, all solutions capable of processing the AI service are further determined, the third network device further selects an optimal solution for processing the AI service from all the solutions, and finally, the AI task deployment scheme for processing the AI service is generated according to the optimal solution. Since the optimal solution selected can meet the network resources required when the AI service is processed, the network resources on the first device can meet the demand of the AI tasks for network resources after the AI task deployment scheme generated according to the optimal solution is used and the AI tasks are deployed through the first device, so that more sufficient network resources are ensured for processing the AI service, and the demand of the AI service for network resources is met. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 An AI service processing flow provided by the embodiment of the application Figure 1 ;

[0035] Figure 2 An Internet of Vehicles-intelligent collision prediction interaction flow provided by the embodiment of the application

[0036] Figure 3 An AI task interaction flow provided by the embodiment of the application Figure 1 ;

[0037] Figure 4 An AI task interaction flow provided by the embodiment of the application Figure 2 ;

[0038] Figure 5 An AI service processing flow provided by the embodiment of the application Figure 2 ;

[0039] Figure 6 An AI service processing flow provided for an embodiment of the present application Figure 3 ;

[0040] Figure 7 An AI service processing flow provided for an embodiment of the present application Figure 4 ;

[0041] Figure 8 An AI service processing flow provided for an embodiment of the present application

[0042] Figure 9 A component structure of a first network device provided for an embodiment of the present application Figure 1 ;

[0043] Figure 10 A component structure of a first network device provided for an embodiment of the present application Figure 2 ;

[0044] Figure 11 A component structure of a third network device provided for an embodiment of the present application Figure 1 ;

[0045] Figure 12 A component structure of a third network device provided for an embodiment of the present application Figure 2 ;

[0046] Figure 13 A component structure of a first device provided for an embodiment of the present application Figure 1 ;

[0047] Figure 14 A component structure of a first device provided for an embodiment of the present application Figure 2 . DETAILED DESCRIPTION

[0048] In order to enable a person skilled in the art to better understand the features and technical contents of the embodiments of the present application, the technical solutions of the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments of the present application. The accompanying drawings are only used for reference and are not intended to limit the embodiments of the present application.

[0049] Unless otherwise defined, all technical and scientific terms used in the embodiments of the present application have the same meanings as those commonly understood by a person skilled in the art to which the present application belongs. The terms used in the embodiments of the present application are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0050] In the following description, "some embodiments" are referred to, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. It should also be noted that the terms "first / second / third" referred to in the embodiments of the present application are only used to distinguish similar objects, and do not represent a specific order of the objects. Understandably, "first / second / third" can be interchanged with a specific order or sequence as allowed, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0051] In related technologies, the method of wireless network providing AI service can adopt the following key technologies:

[0052] (1) 5G network: 5G network can provide higher bandwidth, lower latency and larger capacity, which provides a better foundation for wireless network providing AI service. Through the support of 5G network, more efficient data transmission can be realized, meeting the requirements of AI service for processing large-scale data and real-time performance.

[0053] (2) Edge intelligent terminal device: Edge intelligent terminal device (such as smart phone, smart speaker, etc.) integrates processing capability and AI algorithm, which can process part of AI tasks locally. By using edge intelligent terminal device, part of the computing tasks of AI service can be moved from the network center to the edge device, reducing network load and improving response speed.

[0054] (3) Distributed computing: Using distributed computing technology, AI services are dispersed to multiple nodes for parallel processing, thereby improving computing efficiency and speed. Through reasonable resource scheduling and task allocation, distributed computing can realize efficient AI service support.

[0055] (4) AI model compression and optimization: For wireless network providing AI service, it is very important to compress and optimize AI model. By reducing the size of the model and simplifying the calculation process, the demand for computing resources can be reduced, and the performance of AI service in limited resource environment can be improved.

[0056] In summary, the related technical solutions of wireless network providing AI service include 5G network, edge intelligent terminal device, distributed computing and AI model compression and optimization, etc. These technical solutions can jointly promote wireless network to realize efficient, real-time and reliable AI service support.

[0057] However, the following problems still exist through the above technical solutions:

[0058] 1. Computing resource limitation: Devices in wireless networks (e.g., terminals, base stations, and management devices at the upper layer of base stations, etc.) usually have low computing power and storage capacity, making it difficult to handle complex AI tasks. This can result in slow processing speed, delayed response time, and other issues.

[0059] 2. Network bandwidth limitation: Although 5G networks provide higher bandwidth, network congestion and bandwidth limitations can still occur in certain scenarios. Large-scale data transmission and real-time AI tasks require greater bandwidth support, otherwise the performance and response speed of AI services can be affected.

[0060] 3. Difficulty in model updating and deployment: AI model updating and deployment can require significant bandwidth and computing resources, which can pose certain challenges in wireless networks. In application scenarios with high real-time requirements, the ability to quickly deploy and update models becomes particularly important.

[0061] To address the above issues, embodiments of the present application provide an AI service processing method, as shown in Figure 1 The method can be applied to a first network device and can include the following steps:

[0062] S101, receiving first information sent by a second network device.

[0063] The first information is used by the first network device to determine an AI service.

[0064] In embodiments of the present application, the first network device can include an AI service orchestration management function (SMO).

[0065] In embodiments of the present application, the second network device can include an AI service platform.

[0066] In embodiments of the present application, the first information can be understood as AI service information corresponding to a user-subscribed AI service, and the first information includes one or more of the following: a user identifier (ID), a subscribed AI service case number, and a corresponding AI service template.

[0067] It should be noted that one or more can also be understood as at least one.

[0068] It should be noted that AI service templates are not limited to standardized formulation, network operator pre-configuration, and other methods.

[0069] In embodiments of the present application, the AI service template includes at least second information identifying the AI service and first network resource information describing the AI service requirements.

[0070] The second information includes one or more of the following: an AI service case number, an AI service scenario identifier, and an AI service sub-scene identifier. The first network resource information includes at least one or more of the following: an AI service triggering mode, an AI service network range, an AI task identifier, AI service required input data, an AI model, an AI model task, and AI model output data.

[0071] In the embodiments of the present application, the second information identifying the AI service in the AI service template includes an AI service case number, an AI service scenario ID, and an AI service sub-scene ID, as shown in Table 1 below.

[0072] Table 1

[0073] AI service use case number AI service scenario ID AI service sub-scenario ID

[0074] In the embodiments of the present application, the first network resource information describing the AI service requirement in the AI service template is as shown in Table 2:

[0075] Table 2

[0076]

[0077] In the embodiments of the present application, the contents involved in the AI service template in Table 1 and Table 2 are further described in detail, wherein:

[0078] 1) AI service case number: the AI service case number can uniquely represent the AI service that a user can subscribe to, and can also correspond to the AI service requirement for the first network resource, thereby translating the AI service that a user can subscribe to into a requirement for network resources. For example, the AI service case number corresponds to a specific AI service, and through the AI service case number, the network resource requirement for the specific AI service can also be determined. Each AI service case number corresponds to a first network resource requirement.

[0079] 2) AI service scenario ID: the network (for example, an AI service platform) pre-numbers according to the AI service application scenarios that can be provided.

[0080] 3) AI service sub-scene ID: the network (for example, an AI service platform) can manage and number according to scene layering, which is conducive to expanding more scenarios of AI services that the network can provide.

[0081] 4) AI service triggering mode: it is the triggering mode of the AI service to start the business process, that is, the AI service related business process is started when the triggering condition is met.

[0082] The triggering mode can be event triggering, which includes but is not limited to: positioning detection, weather detection, accident detection, etc.; the triggering mode can also be switch triggering, for example, the UE triggers the start of the service process by clicking the AI service interface.

[0083] 5) AI service network range: the network range providing AI services for the user after triggering the AI service process.

[0084] 6) AI task ID / description: the functional module required to complete the AI service.

[0085] 7) AI service required input data: application data used by the AI service.

[0086] 8) AI model: selection of AI model and source of AI model.

[0087] 9) AI model task: selection of AI model, model training, model verification optimization, model inference, etc. for executing AI service tasks.

[0088] 10) AI model output data: conclusions / decisions, etc. output by the AI model.

[0089] In the embodiments of the present application, the AI service template includes, but is not limited to, the contents in Table 1 and Table 2 below. Specifically, it can be selected according to the actual situation, and the embodiments of the present application do not make specific limitations.

[0090] Exemplarily, taking the vehicle networking-intelligent collision prediction as an example, the vehicle networking-intelligent collision prediction interaction flowchart is shown in Figure 2 The corresponding AI service template for the vehicle networking-intelligent collision prediction in Figure 2 is shown in Table 3 and Table 4 below. Table 3 describes the AI service use case number, AI service scenario and AI service sub-scenario of the vehicle networking-intelligent collision prediction; Table 4 describes the network resource requirements of the vehicle networking-intelligent collision prediction.

[0091] Table 3 is shown as follows:

[0092] Table 3

[0093]

[0094] It should be noted that the corresponding relationship in Table 3 can be understood as 1.1-vehicle networking-intelligent collision detection; 1.2-vehicle networking-intersection visual fusion; 1.3-vehicle networking-automatic driving simulation system; 1.4-vehicle networking-satellite assisted driving in unpopulated areas. That is, under the AI service scenario of vehicle networking, AI service sub-scenarios such as intelligent collision prediction, intersection visual fusion, automatic driving simulation system, satellite assisted driving in unpopulated areas, etc. can be included, and each AI service sub-scenario corresponds to an AI service use case number.

[0095] In the embodiment of the present application, after determining the AI service use case No. 1.1, the network resources required for the AI service corresponding to 1.1 can be determined, as shown in the following Table 4:

[0096] Table 4

[0097]

[0098] In the embodiment of the present application, AI service template can be formulated on the AI service platform, after the formulation is completed, the AI service user sends an AI service subscription request to the AI service platform of the network side, and the AI service platform can translate the AI service template according to the AI service subscription request to obtain the AI service use case number corresponding to the AI service subscription request. Further, the AI service platform completes the AI service subscription and sends a subscription success request to the subscribed user. Wherein, the subscribed user corresponds to the user sending the AI service subscription request.

[0099] In the embodiment of the present application, the AI service platform sends the first information containing the AI service template, the user ID and the AI service use case number subscribed by the user to the SMO, and the SMO can further determine the AI service to be processed according to the content contained in the first information after receiving the first information.

[0100] S102, determining the AI service to be processed based on the first information.

[0101] In the embodiment of the present application, after the AI service platform sends the first information to the SMO, the SMO can determine the AI service to be processed according to the first information.

[0102] In the embodiment of the present application, the first information contains the user ID, the AI service use case number subscribed by the user and the corresponding AI service template, so that the SMO can determine the AI service to be processed corresponding to the user ID when receiving the first information.

[0103] For example, if the AI service number subscribed by the user A is 1.1, the AI service to be processed corresponding to 1.1 can be determined in the corresponding AI service template. If Table 3 is taken as an example, the SMO can determine that the AI service to be processed is intelligent collision prediction.

[0104] S103, performing service arrangement processing on the AI service to obtain a solution set corresponding to the processing AI service.

[0105] In the embodiment of the present application, the AI service can be completed by a plurality of AI tasks, wherein the AI task is a functional module required to complete the AI service.

[0106] In the embodiments of the present application, the solution set is all solutions contained in the AI service. For example, there are three solutions to complete a specific AI service, and the three solutions constitute the solution set of the specific AI service.

[0107] In the embodiments of the present application, after the SMO determines the AI service to be processed, the solution set corresponding to the AI service is obtained by performing service orchestration processing on the AI service to be processed. Specifically, the AI service subscribed by the user can be completed by multiple AI task IDs / descriptions. The AI task is a functional module required to complete the AI service. First, the AI tasks contained in the AI service are split to obtain different AI subtasks, for example, the AI tasks are split into data collection, model training, model inference, model updating, and other AI subtasks. Then, the corresponding solutions are determined for different AI subtasks, so that all possible solutions to complete the AI service can be formed, and finally the solution set for processing the AI service is constituted.

[0108] It should be noted that the AI service solution set provided by the embodiments of the present application for the AI service can achieve that the AI service can be completed by multiple AI tasks, and the AI task can be a module for implementing different functions, which can be further split into data collection, model training, model inference, model updating, and other AI subtasks. Different split methods correspond to different network resource configurations, so that the AI service solution set is finally obtained.

[0109] In S104, the solution set is sent to a third network device. The solution set is used for the third network device to select an optimal solution from the solution set.

[0110] In the embodiments of the present application, the third network device can include a Near-realtime RAN Intelligent Controller (Near-RT RIC).

[0111] In the embodiments of the present application, after the SMO generates the solution set, the generated solution set is sent to the Near-RT RIC. Since the solution set contains multiple solutions that can solve the AI service, but not all of them are the optimal solutions for solving the AI service, the third network device is required to further select an optimal solution from the solution set.

[0112] It can be understood that in the AI service processing method provided in the embodiments of the present application, after the first network device determines the AI service to be processed according to the first information sent by the second network device, all solutions capable of processing the AI service are further determined, the optimal solution for processing the AI service is further selected from all the solutions through the third network device, and finally the AI task deployment scheme for processing the AI service is generated according to the optimal solution. Since the optimal solution selected by the first network device can meet the network resources required when processing the AI service, the network resources on the first device can meet the demand of each AI task for network resources after the AI task deployment scheme generated according to the optimal solution is deployed by the third network device through the first device, thereby ensuring that more sufficient network resources are provided for processing the AI service, and the demand of the AI service for network resources is met.

[0113] In an embodiment of the present application, the AI service includes a plurality of AI tasks. When the AI service is processed by service orchestration, one or more AI tasks in the plurality of AI tasks can be processed by splitting to obtain a plurality of combined AI subtasks, and a set composed of the plurality of combined AI subtasks and the plurality of AI tasks is determined as a first set. Based on an AI subtask template, a processing mode corresponding to the AI subtask and / or the AI task in the first set is determined, and the processing mode corresponding to the AI subtask and / or the AI task is determined as a second set. The first set and the second set are used to obtain a third set, and the third set is determined as a solution set corresponding to the AI service.

[0114] In an embodiment of the present application, the SMO can process the AI service by service orchestration to obtain a solution set for processing the AI service.

[0115] In an embodiment of the present application, the first set is all possible combination schemes obtained by splitting the AI task.

[0116] In an embodiment of the present application, the second set is a processing mode corresponding to the AI task or the AI subtask.

[0117] In an embodiment of the present application, when the AI task is processed by splitting to obtain different AI subtasks, the splitting processing mode includes but is not limited to the following splitting processing modes:

[0118] 1) Splitting between AI function modules, such as data acquisition, model inference, model training, model selection, model updating, etc.

[0119] 2) Splitting within an AI function module, such as splitting the same AI model to be executed by different network elements.

[0120] In the embodiments of the present application, the SMO splits one or more AI tasks in the plurality of AI tasks in any splitting manner to obtain a plurality of combined AI subtasks. For example, AI task 1 is split to obtain AI task 1 AI subtask 1 and AI task 1 AI subtask 2, and AI task 2 is not split. AI task 1 AI subtask 1, AI task 1 AI subtask 2, and AI task 2 are referred to as one combination.

[0121] In the embodiments of the present application, the first set contains a plurality of combined AI subtasks and a plurality of AI tasks, as shown in Table 5 below.

[0122] Table 5 shows the first set composed of different splits of AI tasks, wherein the first set only shows three combination manners, but the manner of splitting AI tasks is not limited to the following three.

[0123] Table 5

[0124]

[0125] In the embodiments of the present application, AI subtask templates of various AI tasks are designed. For example, in the data collection subtask template of an AI task, the following information is included: data source, data consumer, data type, data task, data granularity, data volume, data processing method, and specific description, which is used for the Near-RT RIC to extract effective information from the data collection subtask template for deployment according to the data collection task, and the Near-RT RIC to perform communication connection resource allocation and guarantee according to the data type, data task (purpose), and other related information.

[0126] In the embodiments of the present application, a data collection subtask template is provided, as shown in Table 6 below:

[0127] Table 6

[0128]

[0129] In the embodiments of the present application, the AI subtask templates of various AI tasks also include an AI model task template, which includes one or more of the following: AI task type, AI model, and AI model management related configuration information.

[0130] In the embodiments of the present application, the AI model task template includes one or more of the following: AI task type, AI model, and AI model management related configuration information, which is used to extract effective information from the AI model task template for deployment and life cycle management of AI tasks, and the Near-RT RIC to perform communication resource and computing resource allocation and guarantee according to the AI task type, AI model, and other related information.

[0131] In the embodiments of the present application, the AI model task template is as shown in Table 7 below:

[0132] Table 7

[0133]

[0134] In the embodiments of the present application, each AI task or AI subtask corresponds to the relevant information in Table 6 and / or Table 7 above, and the third set obtained by combining Table 5 to Table 7 above is determined as a solution set for solving the AI service.

[0135] Exemplarily, the solution set corresponding to the AI service of vehicle networking-intelligent collision prediction can be:

[0136] The AI service of vehicle networking-intelligent collision prediction is completed by AI task 1-service vehicle trajectory prediction and AI task 2-vehicle collision warning decision algorithm, and each task includes data collection, model inference and other processes.

[0137] The SMO can generate an AI service solution set by performing service orchestration processing on the AI service. The solution set can include multiple solutions, two of which can be:

[0138] Solution 1: In solution 1, AI task 1 is divided into two subtasks, and the AI service is completed by AI task 1-subtask 1, AI task 1-subtask 2 and AI task 2.

[0139] Among them, AI task 1-subtask 1 is part of the service vehicle trajectory prediction: for example, in a CNN with 3 hidden layers (expressed in English as hidden layer), the inference tasks of the first 2 layers belong to AI task 1-subtask 1.

[0140] AI task 1-subtask 2 is another part of the service vehicle trajectory prediction: for example, in a CNN with 3 hidden layers, the inference tasks of the last 1 layer belong to AI task 1-subtask 2.

[0141] AI task 2 is the collision warning decision algorithm prediction of the vehicle.

[0142] The process of interaction between AI task 1-subtask 1, AI task 1-subtask 2 and AI task 2 is as shown in Figure 3 .

[0143] Solution 1 also includes the corresponding AI model task template, as shown in Table 8 below:

[0144] Table 8

[0145]

[0146] It should be noted that if the data collection is involved in solution 1, the data collection sub-task template in the foregoing Table 6 can be referred to, and examples are not given here.

[0147] Solution 3: In solution 3, the AI service is completed by AI task 1 and AI task 2 together. Specifically, as shown in Figure 4

[0148] Solution 3 also includes a corresponding AI model task template, as shown in the following Table 9:

[0149] Table 9

[0150]

[0151]

[0152] It should be noted that if the data collection is involved in solution 3, the data collection sub-task template in the foregoing Table 6 can be referred to, and examples are not given here.

[0153] The embodiments of the present application also provide an AI service processing method, as shown in Figure 5 The method can be applied to a third network device, and can include the following steps.

[0154] S501, receiving a solution set sent by a first network device.

[0155] The solution set is a set of all solutions corresponding to processing the AI service.

[0156] In the embodiments of the present application, the third network device includes Near-RT RIC.

[0157] In the embodiments of the present application, the solution set sent by the first network device can refer to the above-mentioned embodiments, and details are not given here.

[0158] S502, selecting an optimal solution from the solution set based on the solution set.

[0159] In the embodiments of the present application, the SMO sends the generated solution set for solving the AI service to the Near-RT RIC to realize the management and arrangement of network resources.

[0160] ​In the embodiments of the present application, different AI service solutions have different requirements for network resources. For example, when AI task 1-subtask 1 and AI task 1-subtask 2 are not in the same network element, the model intermediate layer data and model parameters need to be transmitted between network elements. When AI task 1 and AI task 2 are not in the same network element, the output data of vehicle trajectory prediction needs to be transmitted between network elements. Therefore, different AI service solutions correspond to different network resource requirements (such as computing resource, connection resource, etc.), and the network resource requirements can be computing resource, connection resource, etc.

[0161] Therefore, after the SMO sends the solution set to the Near-RT RIC, the Near-RT RIC can determine the network resource requirements corresponding to the AI service solution according to different AI service solutions, so as to further determine the optimal solution for solving the AI service from the solution set according to the network resource requirements corresponding to each AI service solution.

[0162] It can be understood that in the AI service processing method provided in the embodiments of the present application, after the first network device determines the AI service to be processed according to the first information sent by the second network device, it further determines all solutions capable of processing the AI service, and the third network device can further select the optimal solution for processing the AI service from the solution set sent by the first network device. Since the selected optimal solution can meet the network resources required for processing the AI service, it can provide a basis for AI task deployment for the first device, and since the AI task deployment scheme corresponds to the optimal solution, it can ensure that the network resources on the first device can meet the network resource requirements of each AI task, provide more sufficient network resources for AI service processing, and meet the network resource requirements of the AI service.

[0163] In an embodiment of the present application, when selecting the optimal solution from the solution set based on the solution set, for each solution in the solution set, the second network resource required by each solution is estimated according to the information in the AI model task template contained in each solution; the third network resource at the current time is obtained; from the solution set, the solution with the second network resource most similar to the third network resource is determined, and the solution with the second network resource most similar to the third network resource is determined as the optimal solution.

[0164] In the embodiments of the present application, the second network resource and the third network resource include one or more of the following: computing resource, connection resource.

[0165] It should be noted that in addition to computing resource and connection resource, data resource and AI model resource can also be included.

[0166] In the embodiments of the present application, the connection resource can be understood as the connection resource corresponding to the coverage range of the AI service, including one or more of the following: cell physical resource block (Physical Resource Block, PRB) utilization, cell radio channel, cell connection load, etc.

[0167] In the embodiments of the present application, the computing resource can be understood as the computing resource corresponding to the coverage range of the AI service, including one or more of the following: hardware structure (such as graphics processing unit (Graphics Processing Unit, GPU), central processing unit (Central Processing Unit, CPU), etc.), computing power size, data storage space, etc.

[0168] In the embodiments of the present application, the current time can be understood as the time when the AI service is to be processed.

[0169] In the embodiments of the present application, the Near-RT RIC can utilize the solution set sent by the SMO, for each solution in the solution set, by the AI type, AI model, etc. in each solution set, the network resource demand required by different solutions is estimated in advance, including the estimation of the demand for computing resource, connection resource, etc.

[0170] In the embodiments of the present application, after the second network resource corresponding to each solution is estimated, the third network resource, i.e. the real-time perceived network resource, is obtained, including the real-time perceived computing resource, connection resource, etc. The second network resource corresponding to each solution is compared with the third network resource or through the optimal algorithm, and the solution closest to the third network resource is determined as the optimal solution.

[0171] It should be noted that closest can also be understood as most similar, and can also be understood as the same.

[0172] In the embodiments of the present application, since it is possible that the third network resource does not satisfy all solutions, for example, in the case of extreme resource shortage, a solution closest / suitable / matched / similar is selected.

[0173] In the embodiments of the present application, the third network device is used to perceive and manage network resources, and a solution corresponding to an AI service is selected, that is, according to the real-time third network resources perceived by the Near-RT RIC, an optimal algorithm is used to select a solution that best matches the current real-time network resource situation, and a deployment scheme of the AI task is further generated. The solution that best matches the current real-time network resource situation is based on the cross-domain collaborative management and arrangement of multi-objective optimization algorithms such as data, computing power, connection, AI model, etc. For each solution, the requirements of different solutions for network resources can be obtained in advance through information such as AI task type, AI model, etc., including requirements for limited computing power and connection resources of the network, and the complex AI service arrangement can be mapped to the matching and selection of network resource requirements and real-time perceived network resources.

[0174] In an embodiment of the present application, based on the solution set, an optimal solution is selected from the solution set, and based on the optimal solution, an AI task deployment scheme corresponding to the optimal solution is determined; the AI task deployment scheme is sent to the first device, and the first device deploys the corresponding AI task or AI subtask on the first device based on the AI task deployment scheme.

[0175] In the embodiments of the present application, the first device can include one or more of the following base stations, terminals, edge computing nodes. For example, the first device can be a base station, such as a NodeB, a Radio Access Network (RAN).

[0176] In the embodiments of the present application, the AI task deployment scheme refers to the allocation of data collection, AI model tasks, etc. to specific deployed network elements.

[0177] In the embodiments of the present application, when the corresponding AI task deployment scheme is determined based on the optimal solution, the deployment network element of each AI task or AI subtask in the optimal solution is determined according to the computing power resources and connection resources that each network element can carry.

[0178] In the embodiments of the present application, after the deployment network element is determined, the AI task deployment scheme corresponding to the optimal solution is formed, and the deployment scheme is sent to the first device, where the first device can be a device that executes the AI task deployment scheme. The first device deploys the corresponding AI task or AI subtask based on the AI task deployment scheme.

[0179] It should be noted that the determination of the AI task deployment scheme is also affected by the AI service triggering mode, AI service network coverage range, etc. contained in the AI service template, and the data source, data consumer IP, and AI model management related parameters in the solution set can also provide sufficient information for the AI task deployment scheme to realize the full-automatic AI service process.

[0180] Exemplarily, taking the AI service solution selection and AI task deployment scheme generation of the vehicle networking-intelligent collision prediction as an example, the Near-RT RIC determines, according to the perceived network resources, that the service vehicle has the inference data and computing power of the AI task 1, and that the service vehicle and the main service cell transmission channel quality can meet the real-time requirement of AI task 1 inference result transmission and task 2 decision algorithm issuing. The main service cell has sufficient computing power, storage, and other environmental support for AI task 2. Therefore, the AI service solution 3 in Table 5 is selected, and the AI task deployment scheme is generated, as shown in the following Table 10:

[0181] Table 10

[0182]

[0183]

[0184] In the above Table 10, it can be seen that the AI task 1 can be deployed on the terminal side, such as the vehicle side, and the AI task 2 can be deployed on the base station side, such as the NodeB where the vehicle is located. When the terminal side receives the AI task deployment scheme, the AI task 1 is deployed, and when the base station side receives the deployment scheme, the AI task 2 is deployed.

[0185] In an embodiment of the present application, the third network device can also perform AI service lifecycle management and provide quality of service (QoS) guarantee.

[0186] In an embodiment of the present application, when the Near-RT RIC formulates the AI task deployment scheme, the AI service lifecycle process can be executed, and the AI task lifecycle in the AI service solution can also be executed.

[0187] In an embodiment of the present application, the third network device can also receive the QoS information of the AI service and update the AI task deployment scheme. Specifically, the third network device receives the QoS information of the AI service sent by the first device; when the QoS information does not meet the AI service requirement, the optimal solution corresponding to the third network resource is reselected from the solution set, and the updated AI task deployment scheme is determined based on the optimal solution corresponding to the third network resource.

[0188] In the embodiments of the present application, when the AI task is executed by the AI task or AI subtask deployed on different network elements, the third network device receives the QoS information of the AI service, which can be obtained by reporting to the third network device through the first device, or can be obtained by other manners, and the embodiments of the present application do not make specific limitations.

[0189] In the embodiments of the present application, when it is monitored that the QoS information (for example, the prediction accuracy) of the AI service does not meet the AI service requirement, then according to the real-time perceived third network resource, one AI service solution closest to the third network resource or corresponding to the third network resource is selected from the solution set again, and an updated AI task deployment scheme is generated according to the reselected solution, and the deployment is performed again based on the updated AI task deployment scheme and the monitoring of the QoS information is continued.

[0190] It should be noted that the method of reselecting the solution closest to the third network resource according to the third network resource can refer to the foregoing embodiments, which will not be repeated here.

[0191] It should be noted that the process of determining the updated AI task deployment scheme according to the reselected solution can also refer to the foregoing embodiments, which will not be repeated here.

[0192] The embodiments of the present application also provide an AI service processing method, as shown in Figure 6 The method can include the following steps:

[0193] S601, receiving the AI task deployment scheme sent by the third network device.

[0194] In the embodiments of the present application, the first device includes one or more of the following: a base station, a terminal, and an edge computing node.

[0195] It should be noted that the first device is not limited to the devices contained above, and the specific selection can be made according to the actual situation, and the embodiments of the present application do not make specific limitations.

[0196] In the embodiments of the present application, the first device receives the AI task deployment scheme sent by the Near-RT RIC.

[0197] In the embodiments of the present application, if the AI service needs to be deployed on one network element, for example, the base station side, the AI task deployment scheme can be sent to the base station side through the Near-RT RIC; and if the AI service needs to be deployed on two network elements, for example, the base station side and the terminal side, the AI task deployment scheme can be sent to the base station side through the Near-RT RIC, and the base station side sends the AI task deployment scheme to the terminal side.

[0198] It should be noted that because the number of first devices completing the AI task deployment scheme can be one or more, if it is one first device, the first device does not need to perform the AI task deployment scheme forwarding process, and if it is multiple first devices, the AI task deployment scheme needs to be sent to each first device performing an AI task or an AI subtask.

[0199] S602, deploying the corresponding AI task or AI subtask on the first device based on the AI task deployment scheme.

[0200] In the embodiments of the present application, after receiving the AI task deployment scheme, the first device performs AI task or AI subtask deployment corresponding to the AI task deployment scheme on the corresponding first device.

[0201] In the embodiments of the present application, when deploying the AI task or the AI subtask, taking an example that an AI service is completed by AI task 1 and AI task 2, AI task 1 is completed on the vehicle side, and AI task 2 is completed on the base station side, when the AI task deployment scheme is received, AI task 1 is deployed on the vehicle side, and AI task 2 is deployed on the base station side. In another example, the AI service is completed by AI task 1 and AI task 2, and both AI task 1 and AI task 2 are completed on the base station side, then when the AI task deployment scheme is received, AI task 1 and AI task 2 are deployed on the base station side. In another example, if the AI service is completed by AI task 1_subtask 1, AI task 2_subtask 2, and AI task 2, AI task 1_subtask 1 is completed on the vehicle side, and AI task 2_subtask 2 and AI task 2 are completed on the base station side, then AI task 1_subtask 1 is deployed on the vehicle side, and AI task 2_subtask 2 and AI task 2 are deployed on the base station side.

[0202] It can be understood that in the AI service processing method provided in the embodiments of the present application, because the selected optimal solution can meet the network resources required when processing the AI service, after the AI task deployment scheme generated according to the optimal solution and the AI task deployment performed by the first device, the network resources on the first device can meet the demand of each AI task for network resources, thereby ensuring that more sufficient network resources are provided for processing the AI service, and meeting the demand of the AI service for network resources.

[0203] In an embodiment of the present application, if the number of first devices is multiple, after receiving the AI task deployment scheme sent by the third network device, when the AI task deployment scheme is deployed on the multiple first devices, on the multiple first devices, the life cycle of the corresponding AI task and AI subtask is executed based on the AI deployment scheme, and the intermediate execution result obtained by executing the life cycle of the AI task and / or AI subtask is interacted between the multiple first devices; based on the intermediate execution result, a target execution strategy is determined, and the target execution strategy is executed by using a target device in the multiple first devices.

[0204] In an embodiment of the present application, the target device is a device that executes the target execution strategy of the final output, which can be a base station included in the first device, can also be a terminal included in the first device, or can be other devices included in the first device, which is not specifically limited here.

[0205] In an embodiment of the present application, when multiple first devices are required to complete the AI deployment scheme, the corresponding AI task or AI subtask in the AI deployment scheme is deployed on each first device. Because all AI tasks or AI subtasks are executed to complete a complete AI service, each first device needs to start executing the corresponding AI task or AI subtask when the trigger condition of the corresponding AI task or AI subtask is triggered, and output the result after execution.

[0206] In an embodiment of the present application, there is a corresponding execution sequence for executing the corresponding AI task or AI subtask on the first device. Only after the previous AI task or AI subtask is executed, the life cycle execution of the next AI task or AI subtask can be triggered.

[0207] In an embodiment of the present application, the intermediate execution result of the AI task or subtask is interacted between the first devices. The intermediate execution result is obtained in turn according to the execution sequence of the AI task or AI subtask, and the target execution strategy is determined according to the last obtained intermediate execution result.

[0208] It should be noted that the AI task or AI subtask can also be executed in parallel. For example, the first device includes a base station 1, a base station 2 and a terminal, AI task 1 is executed on the base station 1, AI task 2 is executed on the base station 2, and AI task 3 is executed on the terminal. Then, the base station 1 and the base station 2 can execute AI task 1 and AI task 2 at the same time, and then the execution results are sent to the terminal at the same time, and the terminal further executes AI task 3 according to the execution results sent by the base station 1 and the base station 2.

[0209] It should be noted that the target execution strategy represents a solution strategy for completing the AI service.

[0210] In the embodiment of the present application, after obtaining the target execution strategy, the target execution strategy is transmitted to the first device in the plurality of first devices that executes the target execution strategy, and the target execution strategy is executed through the first device to complete the AI service.

[0211] Exemplarily, taking the vehicle-to-everything-intelligent collision prediction as an example, the AI service needs to be completed by AI task 1 and AI task 2 together, AI task 1 is deployed on the vehicle side, and AI task 2 is deployed on the base station side. After the corresponding AI task deployment is completed on the vehicle side and the base station side, the vehicle side detects the AI service generation cycle process trigger condition, executes the AI task 1 life cycle when the trigger condition is met, and sends the model prediction value (intermediate execution result) output by the AI task 1 to the base station side after the execution ends. The base station side performs the AI task 2 life cycle execution, and sends the anti-collision strategy output by the AI task 2 to the vehicle side for finally executing the strategy, and the vehicle side executes the anti-collision strategy.

[0212] In an embodiment of the present application, after the first device executes the target execution strategy, the first device further obtains the QoS information of the AI service; and the QoS information is sent to the third network device, and the QoS information is used for the third network device to reselect an optimal solution corresponding to the third network resource from the solution set when the QoS information does not meet the AI service demand, and determine an updated AI task deployment scheme based on the optimal solution corresponding to the third network resource.

[0213] In the embodiment of the present application, the QoS information of the AI service can be a prediction accuracy rate and the like.

[0214] In the embodiment of the present application, the first device is one, and the first device is an execution device of the target execution strategy. Then, the first device can determine the QoS information of the AI service based on the execution result of the target execution strategy, and send the QoS information to the Near-RT RIC, and the Near-RT RIC further processes based on the QoS information of the AI service.

[0215] In the embodiment of the present application, the first device is one, and the first device is an execution device of the target execution strategy. Then, the first device can determine the QoS information of the AI service based on the execution result of the target execution strategy, and send the QoS information to the Near-RT RIC, and the Near-RT RIC further processes based on the QoS information of the AI service.

[0216] Based on the above embodiment, the present application further provides a flowchart of AI service processing, as shown in Figure 7As shown, mainly includes the following steps:

[0217] 1. AI service subscription and translation, including the following:

[0218] AI service template formulation

[0219] User subscription AI service

[0220] Query AI service template

[0221] Transmission AI service template-AI service with number

[0222] 2. AI service orchestration processing, including the following:

[0223] AI task decomposition solution set

[0224] AI subtask template formulation

[0225] Transmission AI service corresponding solution set

[0226] 3. Optimal solution selection and AI task deployment scheme generation, including the following:

[0227] Select the optimal solution

[0228] Generate AI task deployment scheme

[0229] Transmission AI task deployment scheme

[0230] 4. AI task execution control, including the following:

[0231] AI task deployment

[0232] AI task life cycle management

[0233] QoS guarantee

[0234] 5. QoS information monitoring and dynamic adjustment of the optimal solution.

[0235] Based on the above embodiment, the embodiment of the application also provides a specific flowchart of AI service processing, as shown in Figure 8 , mainly includes the following implementation steps:

[0236] Step 1, AI service subscription and translation, including:

[0237] 1. The AI service platform formulates an AI service template;

[0238] 2. The AI service platform executes user subscription AI service;

[0239] 3. The AI service platform sends AI service information to the SMO: user ID, subscribed AI service case number, and corresponding AI service template.

[0240] It should be noted that in the process of AI service subscription and translation, an AI service template is designed, which includes AI service case number, AI service scene ID, AI service sub-scene ID, AI service trigger mode, AI service network range, AI task identifier, AI service required input data, AI model, AI model task, and AI model output data, which are used to generate AI service cases that users can subscribe to and describe network resource demand information of AI services.

[0241] Step 2, AI service orchestration processing, obtaining an AI service solution set, including:

[0242] 4. The SMO determines the first set after AI task decomposition. Determine the AI service to be processed, and combine the AI tasks contained in the AI service after decomposition to obtain the first set.

[0243] 5. The SMO formulates an AI sub-task template, and determines a second set based on the AI sub-task template. The AI sub-task template includes a data collection sub-task template and / or an AI model task template.

[0244] 6. The SMO determines the solution set of the AI service. The third set obtained by combining the first set and the second set is determined as the solution set of the AI service.

[0245] Step 3, optimal solution selection and AI task deployment scheme generation, including:

[0246] 7. The Near-RT RIC selects an optimal solution corresponding to the AI service.

[0247] 8. The Near-RT RIC determines the AI task deployment scheme.

[0248] 9. The Near-RT RIC issues the AI task deployment scheme to the first device 1, i.e., the RAN.

[0249] 10. The RAN sends the AI task deployment scheme to the first device 2, i.e., the UE.

[0250] It should be noted that if the first device is only one, it only needs to be issued once, and if the first device includes multiple devices, it needs to be forwarded internally.

[0251] Step 4, taking scheme 3 in Table 5 as an example: AI task execution control, including:

[0252] 11. The RAN performs AI task deployment.

[0253] 12. UE performs AI task deployment, detects AI service life cycle process trigger condition;

[0254] 13. UE performs AI task 1 life cycle execution;

[0255] 14. UE sends AI task 1 output model prediction value to RAN;

[0256] 15. RAN performs AI task 2 life cycle execution;

[0257] 16. RAN transmits AI task 2 output anti-collision decision to UE;

[0258] 17. UE executes anti-collision decision;

[0259] Step 5, QoS information monitoring of AI service and dynamic adjustment of optimal solution, including:

[0260] 18. UE reports AI service QoS information to RAN;

[0261] 19. RAN reports AI service QoS information to Near-RT RIC;

[0262] 20. Near-RT RIC reselects the optimal solution of AI service according to the real-time perceived third network resource and AI service QoS information, and regenerates the updated AI task deployment scheme.

[0263] Based on the above steps, in the process of providing AI service in wireless network, the following points are mainly embodied in 3GPP protocol or ORAN protocol:

[0264] 1) SMO->Near-RT RIC: issue AI service solution set (corresponding to ORAN protocol);

[0265] 2) Near-RT RIC->gNB: issue AI task deployment scheme (corresponding to ORAN protocol);

[0266] 3) gNB->UE (or intelligent terminal): issue AI task deployment scheme (corresponding to 3GPP RRM process);

[0267] 4) UE (or intelligent terminal)->gNB: report AI model output model prediction value (corresponding to 3GPP enhanced user plane / signaling plane transmission IE);

[0268] 5) gNB->UE (or intelligent terminal): issue anti-collision decision (corresponding to 3GPP enhanced user plane / signaling plane transmission IE);

[0269] 6) Intelligent terminal -> gNB: Report the QoS information of AI service, such as prediction accuracy (corresponding to 3GPP enhanced user plane / signaling plane transmission IE);

[0270] 7) gNB -> Near-RT RIC: Report the QoS information of AI service, such as prediction accuracy (corresponding to the IE in ORAN E2 subscription and reporting information).

[0271] The AI service processing method proposed in the embodiments of the present application has the following technical advantages compared with related technologies:

[0272] The embodiments of the present application propose a method and process for providing the computing power, data, AI algorithm, and connection resources required by AI services in a wireless network, mainly embodied in the following aspects: signing AI services, describing network resource requirements through AI service corresponding AI sub-task templates, providing network resource requirement pre-configuration through AI service solution sets, generating AI task deployment schemes based on real-time network resources, executing the life cycle of AI services by the first device and providing corresponding QoS guarantee, and monitoring the QoS information of AI services by the first device and further updating the AI task deployment scheme through the third network device. Since this method can describe the requirements of computing power, data, AI algorithm, and connection resources of AI services on network resources through AI service corresponding AI sub-task templates, AI services can be effectively utilized by distributed and fragmented wireless network computing power and connection resources by decomposing AI services into AI tasks and AI sub-tasks, and formulating network resource pre-configuration of AI service solutions according to different decomposition methods. Through AI QoS detection / prediction, AI task deployment schemes can be generated according to the dynamic changes of real-time network resources. Different decomposition methods and deployment methods will correspond to different network resource requirements, and the requirements of different solutions on network resources, including limited computing power and connection resources, can be obtained in advance through information such as AI task type and AI model. Complex AI services can be mapped to network resource requirements. Finally, the problem of limited computing power and connection resources caused by the distributed, fragmented, and real-time fluctuation characteristics of wireless network resources, which makes it difficult to deploy and real-time guarantee model performance, is solved to meet the demand of low service response time and high service accuracy of network edge AI services.

[0273] Based on the above embodiments, in another embodiment of the present application, a first network device 1 is provided, as shown in the figure, the first network device 1 comprises: Figure 9

[0274] A first receiving unit 10 is configured to receive first information sent by a second network device, the first information being used for the first network device to determine an AI service.

[0275] ​The determining unit 11 is configured to determine an AI service to be processed based on first information.

[0276] The processing unit 12 is configured to perform service orchestration processing on the AI service to obtain a solution set corresponding to the AI service.

[0277] The sending unit 13 is configured to send the solution set to a third network device, and the solution set is used for the third network device to select an optimal solution from the solution set.

[0278] In an embodiment, the AI service includes a plurality of AI tasks.

[0279] In an embodiment, the first network device 1 further includes a splitting unit.

[0280] The splitting unit is configured to perform splitting processing on one or more AI tasks in the plurality of AI tasks to obtain a plurality of combined AI subtasks, and determine a set composed of the plurality of combined AI subtasks and the plurality of AI tasks as a first set.

[0281] The determining unit 11 is further configured to determine a processing mode corresponding to the AI subtask and / or the AI task in the first set based on an AI subtask template, and determine the processing mode corresponding to the AI subtask and / or the AI task as a second set.

[0282] The determining unit 11 is further configured to obtain a third set by using the first set and the second set, and determine the third set as the solution set corresponding to the AI service.

[0283] In an embodiment, the AI subtask template includes an AI model task template, and the AI model task template includes one or more of the following: an AI task type, an AI model, and configuration information related to AI model management.

[0284] In an embodiment, the first information includes an AI service template, and the AI service template includes at least second information identifying the AI service and first network resource information describing a requirement of the AI service; the second information includes one or more of the following: an AI service use case number, an AI service scenario identifier, and an AI service sub-scenario identifier; and the first network resource information includes one or more of the following: an AI service triggering mode, an AI service network range, an AI task identifier, AI service required input data, an AI model, an AI model task, and AI model output data.

[0285] The embodiment of the application provides a first network device, the first network device receives first information sent by a second network device, the first information is used for the first network device to determine an AI service; based on the first information, an AI service to be processed is determined; the AI service is subjected to service arrangement processing, and a solution set corresponding to the AI service is obtained; the solution set is sent to a third network device, and the solution set is used for the third network device to select an optimal solution from the solution set, so it can be seen that the first network device provided by the embodiment of the application determines the AI service to be processed according to the first information sent by the second network device, further determines all solutions capable of processing the AI service, further selects the optimal solution for processing the AI service from all the solutions through the third network device, and finally generates an AI task deployment scheme for processing the AI service according to the optimal solution. It can be seen that the optimal solution selected through the first network device can meet the network resources required when the AI service is processed, and then the network resources on the first device can meet the demand of each AI task for network resources after the third network device generates the AI task deployment scheme according to the optimal solution and deploys the AI task through the first device, so that more sufficient network resources are ensured to process the AI service, and the demand of the AI service for network resources is met.

[0286] Figure 10 A schematic diagram of the composition structure of the first network device 1 provided by the embodiment of the application is shown in the actual application, based on the same disclosure concept of the above embodiment, as shown in the figure, Figure 10 The first network device 1 of the embodiment of the application includes a first processor 14, a first memory 15 and a first communication bus 16.

[0287] In the specific embodiment process, the first receiving unit 10, the determining unit 11, the processing unit 12, the sending unit 13, and the splitting unit described above can be implemented by a first processor 14 located on the first network device 1. The first processor 14 can be at least one of an application specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a CPU, a controller, a microcontroller, or a microprocessor. It can be understood that, for different devices, the electronic device used to implement the processor function can also be other devices, and the embodiments of the present application do not make specific limitations.

[0288] In the embodiments of the present application, the first communication bus 16 is used to realize the connection and communication between the first processor 14 and the first memory 15. When the first processor 14 executes the running program stored in the first memory 15, the following AI service processing method is realized:

[0289] receiving first information sent by a second network device, the first information being used for the first network device to determine an AI service; determining an AI service to be processed based on the first information; performing service orchestration processing on the AI service to obtain a solution set corresponding to the AI service; and sending the solution set to a third network device, the solution set being used for the third network device to select an optimal solution from the solution set.

[0290] In an embodiment, the AI service includes a plurality of AI tasks.

[0291] In an embodiment, the first processor 14 is further configured to split one or more AI tasks in the plurality of AI tasks to obtain a plurality of combined AI subtasks, determine a set composed of the plurality of combined AI subtasks and the plurality of AI tasks as a first set, determine a processing mode corresponding to the AI subtasks and / or the AI tasks in the first set based on an AI subtask template, determine the processing mode corresponding to the AI subtasks and / or the AI tasks as a second set, and obtain a third set by using the first set and the second set, and determine the third set as the solution set corresponding to the AI service.

[0292] In an embodiment, the AI sub-task template includes an AI model task template, and the AI model task template includes one or more of the following: an AI task type, an AI model, and configuration information related to AI model management.

[0293] In an embodiment, the first information includes an AI service template, and the AI service template includes at least second information identifying an AI service and first network resource information describing a requirement of the AI service; the second information includes one or more of the following: an AI service use case number, an AI service scenario identifier, and an AI service sub-scenario identifier; and the first network resource information includes one or more of the following: an AI service triggering mode, an AI service network range, an AI task identifier, AI service required input data, an AI model, an AI model task, and AI model output data.

[0294] Based on the above embodiments, in another embodiment of the present application, a third network device 2 is provided, as shown in Figure 11 The third network device 2 includes:

[0295] A second receiving unit 20 is configured to receive a solution set sent by the first network device; the solution set is a set of all solutions corresponding to the AI service.

[0296] A selection unit 21 is configured to select an optimal solution from the solution set based on the solution set.

[0297] In an embodiment, the third network device 2 can further include a prediction unit, an acquisition unit, and a determination unit.

[0298] The prediction unit is configured to, for each solution in the solution set, predict a second network resource required by each solution according to information in an AI model task template included in each solution.

[0299] The acquisition unit is configured to acquire a third network resource at a current time.

[0300] The determination unit is configured to determine, from the solution set, a solution in which the second network resource is most similar to the third network resource, and determine the solution in which the second network resource is most similar to the third network resource as the optimal solution.

[0301] In an embodiment, the second network resource and the third network resource include one or more of the following: computing power resources and connection resources.

[0302] In an embodiment, the third network device 2 can further include a sending unit.

[0303] The determining unit is further configured to determine an AI task deployment scheme corresponding to the optimal one solution based on the optimal one solution.

[0304] The sending unit is configured to send the AI task deployment scheme to the first device, and the first device deploys a corresponding AI task or AI subtask on the first device based on the AI task deployment scheme.

[0305] The second receiving unit 20 is further configured to receive quality of service (QoS) information of the AI service sent by the first device.

[0306] The selecting unit 21 is further configured to, when the QoS information does not meet the requirement of the AI service, reselect an optimal solution corresponding to third network resources from the solution set, and determine an updated AI task deployment scheme based on the optimal solution corresponding to the third network resources.

[0307] The third network device provided in the embodiments of the present application receives a solution set sent by a first network device, wherein the solution set is a set of all solutions corresponding to processing of an AI service; and an optimal one solution is selected from the solution set based on the solution set. As can be seen, the third network device provided in the embodiments of the present application, after the first network device determines the AI service to be processed based on the first information sent by the second network device, further determines all solutions capable of processing the AI service, the third network device can further select an optimal solution for processing the AI service from the solution set sent by the first network device, the selected optimal solution can meet the network resources required when processing the AI service, and therefore can provide a basis for AI task deployment of the first device. Since the AI task deployment scheme corresponds to the optimal solution, the network resources on the first device can meet the requirement of the AI tasks for the network resources, more sufficient network resources are provided for processing of the AI service, and the requirement of the AI service for the network resources is met.

[0308] Figure 12 A schematic diagram of a composition structure of the third network device 2 provided in the embodiments of the present application is shown in FIG. 2. In actual application, based on the same disclosure concept of the above embodiments, as shown in FIG. 2, the third network device 2 of the embodiments of the present application includes a second processor 22, a second memory 23 and a second communication bus 24. Figure 12

[0309] ​In the specific embodiment process, the second receiving unit 20, the selection unit 21, the estimation unit, the acquisition unit, the determination unit, and the sending unit described above can be implemented by a second processor 22 located on the third network device 2. The second processor 22 can be at least one of an ASIC, a DSP, a DSPD, a PLD, an FPGA, a CPU, a controller, a microcontroller, and a microprocessor. It can be understood that, for different devices, the electronic device used to implement the processor function can also be other devices, and the embodiments of the present application do not make specific limitations.

[0310] In the embodiments of the present application, the second communication bus 24 is used to realize the connection and communication between the second processor 22 and the second memory 23. When the second processor 22 executes the running program stored in the second memory 23, the following AI service processing method is realized:

[0311] receiving a solution set sent by the first network device; wherein the solution set is a set of all solutions corresponding to the processing of the AI service; and selecting an optimal solution from the solution set based on the solution set.

[0312] In an embodiment, the second processor 22 is further configured to, for each solution in the solution set, estimate the second network resource required by each solution according to the information in the AI model task template contained in each solution; acquire the third network resource at the current time; and determine, from the solution set, a solution that is most similar to the third network resource in terms of the second network resource, and determine the solution that is most similar to the third network resource in terms of the second network resource as the optimal solution.

[0313] In an embodiment, the second network resource and the third network resource include one or more of the following: computing power resource, connection resource.

[0314] In an embodiment, the second processor 22 is further configured to determine an AI task deployment scheme corresponding to the optimal solution based on the optimal solution; and send the AI task deployment scheme to the first device, so that the first device deploys the corresponding AI task or AI subtask on the first device based on the AI task deployment scheme.

[0315] In an embodiment, the second processor 22 is further configured to receive the QoS information of the AI service sent by the first device; and when the QoS information does not meet the demand of the AI service, reselect an optimal solution corresponding to the third network resource from the solution set, and determine an updated AI task deployment scheme based on the optimal solution corresponding to the third network resource.

[0316] Based on the above embodiments, in another embodiment of the present application, a first device 3 is provided, which comprises a first processor 31, a first memory 32, a first communication bus 33, and a first receiving unit 34, a first selection unit 35, a first estimation unit, a first acquisition unit, a first determination unit, and a first sending unit.Figure 13 The first device 3 includes:

[0317] The third receiving unit 30 is configured to receive an AI task deployment scheme sent by a third network device.

[0318] The deployment unit 31 is configured to deploy corresponding AI tasks or AI subtasks on the first device based on the AI task deployment scheme.

[0319] In an embodiment, the first device includes one or more of the following: a base station, a terminal, an edge computing node.

[0320] In an embodiment, the first device 3 can further include an execution unit.

[0321] The execution unit is configured to, when the AI task deployment scheme is deployed on a plurality of first devices, execute the life cycle of the corresponding AI tasks and AI subtasks on the plurality of first devices based on the AI deployment scheme, and interact the intermediate execution results obtained by executing the life cycle of the AI tasks and / or AI subtasks among the plurality of first devices.

[0322] The execution unit is further configured to determine a target execution strategy based on the intermediate execution results, and execute the target execution strategy by using a target device in the plurality of first devices.

[0323] In an embodiment, the first device 3 can further include an obtaining unit and a sending unit.

[0324] The obtaining unit is configured to obtain QoS information of the AI service.

[0325] The sending unit is configured to send the QoS information to the third network device, and the QoS information is used for the third network device to reselect an optimal solution corresponding to the third network resource from the solution set when the QoS information does not meet the demand of the AI service, and determine an updated AI task deployment scheme based on the optimal solution corresponding to the third network resource.

[0326] The embodiment of the present application provides a first device, which receives an AI task deployment scheme sent by a third network device, and deploys corresponding AI tasks or AI subtasks on the first device based on the AI task deployment scheme. It can be seen that the first device provided in the embodiment of the present application can meet the network resource required for processing the AI service due to the selected optimal solution, so that the network resource on the first device can meet the demand of each AI task for the network resource after the AI task deployment scheme generated according to the optimal solution is deployed on the first device, thereby guaranteeing that more sufficient network resource is provided for processing the AI service, and meeting the demand of the AI service for the network resource.

[0327] Figure 14A schematic diagram of a constituent structure of a first device 3 provided in an embodiment of the present application is shown in FIG. 3. In actual application, based on the same disclosure concept of the above embodiment, the first device 3 of the embodiment of the present application includes a third processor 32, a third memory 33 and a third communication bus 34. Figure 14

[0328] In a specific embodiment process, the third receiving unit 30, the deployment unit 31, the execution unit, the obtaining unit and the sending unit described above can be implemented by the third processor 32 located on the first device 3. The third processor 32 can be at least one of an ASIC, a DSP, a DSPD, a PLD, a FPGA, a CPU, a controller, a microcontroller, a microprocessor. It can be understood that for different devices, the electronic device used to implement the above processor function can also be other, and the embodiment of the present application does not make specific limitation.

[0329] In the embodiment of the present application, the third communication bus 34 is used to realize the connection and communication between the third processor 32 and the third memory 33. When the third processor 32 executes the running program stored in the third memory 33, the following AI service processing method is realized:

[0330] receiving an AI task deployment scheme sent by a third network device; and deploying corresponding AI tasks or AI subtasks on the first device based on the AI task deployment scheme.

[0331] In an embodiment, the first device includes one or more of the following: a base station, a terminal, an edge computing node.

[0332] In an embodiment, the third processor 32 is further configured to, when the AI task deployment scheme is deployed on a plurality of first devices, execute the life cycle of the corresponding AI tasks and AI subtasks on the plurality of first devices based on the AI deployment scheme, and interact between the plurality of first devices on the intermediate execution results obtained by executing the life cycle of the AI tasks and / or AI subtasks; based on the intermediate execution results, determine a target execution strategy, and execute the target execution strategy by using a target device in the plurality of first devices.

[0333] In an embodiment, the third processor 32 is further configured to obtain QoS information of the AI service; and send the QoS information to the third network device, wherein the QoS information is used for the third network device to reselect an optimal solution corresponding to the third network resource from a solution set when the QoS information does not meet the demand of the AI service, and determine an updated AI task deployment scheme based on the optimal solution corresponding to the third network resource.

[0334] ​Based on the above embodiments, the embodiments of the present application provide a storage medium having a computer program stored thereon, the computer readable storage medium stores one or more programs, the one or more programs can be executed by one or more processors, and are applied to the first network device / third network device / first device, and the computer program implements the AI service processing method as described above.

[0335] Based on the above embodiments, the embodiments of the present application provide a computer program product, comprising a computer program, the computer program can be executed by one or more processors, and is applied to the first network device / third network device / first device, and the computer program implements the AI service processing method as described above.

[0336] It should be noted that in the embodiments of the present application, the terms "comprising", "containing" or any other variants thereof are intended to cover non-exclusive containing, so that the processes, methods, articles or devices including a series of elements not only include those elements, but also include other elements not explicitly listed, or include elements inherent to such processes, methods, articles or devices. Without more limitations, the element defined by the sentence "including a…" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0337] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment method can be realized by means of software and necessary general hardware platform, of course, it can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the embodiments of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a plurality of instructions for making an image display device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) execute the methods described in various embodiments of the present application.

[0338] The above is only a specific implementation of the embodiments of the present application, but the protection scope of the present application is not limited thereto. Any skilled person in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An artificial intelligence (AI) service processing method, characterized in that: The method is applied to a first network device, and the method comprises: receiving first information sent by a second network device, the first information being used for the first network device to determine an AI service; determining an AI service to be processed based on the first information; performing service orchestration processing on the AI service to obtain a solution set corresponding to processing the AI service; sending the solution set to a third network device, the solution set being used for the third network device to select an optimal solution from the solution set.

2. The method of claim 1, wherein, The AI service comprises a plurality of AI tasks, and the performing service orchestration processing on the AI service to obtain a solution set corresponding to processing the AI service comprises: performing splitting processing on one or more AI tasks in the plurality of AI tasks to obtain a plurality of combined AI subtasks, and determining a set composed of the plurality of combined AI subtasks and the plurality of AI tasks as a first set; determining a processing mode corresponding to an AI subtask and / or an AI task in the first set based on an AI subtask template, and determining the processing mode corresponding to the AI subtask and / or the AI task as a second set; obtaining a third set by using the first set and the second set, and determining the third set as the solution set corresponding to the AI service.

3. The method of claim 2, wherein, The AI subtask template comprises an AI model task template, and the AI model task template comprises one or more of the following: an AI task type, an AI model, and configuration information related to AI model management.

4. The method according to claim 1, wherein The first information comprises an AI service template, and the AI service template comprises at least second information identifying an AI service and first network resource information describing a requirement of the AI service; wherein the second information comprises one or more of the following: an AI service use case number, an AI service scenario identifier, and an AI service sub-scenario identifier, and the first network resource information comprises at least one or more of the following: an AI service triggering mode, an AI service network range, an AI task identifier, AI service required input data, an AI model, an AI model task, and AI model output data.

5. An AI service processing method, characterized by, The method is applied to a third network device, and the method comprises: receiving a solution set sent by a first network device; wherein the solution set is a set of all solutions corresponding to processing an AI service; selecting an optimal solution from the solution set based on the solution set.

6. The method of claim 5, wherein, The selecting an optimal solution from the solution set based on the solution set comprises: for each solution in the solution set, estimating second network resources required by the each solution according to information in an AI model task template contained in the each solution; obtaining third network resources at a current time; from the solution set, determining a solution in which the second network resources are most similar to the third network resources, and determining the solution in which the second network resources are most similar to the third network resources as the optimal solution.

7. The method of claim 6, wherein, The second network resource and the third network resource include one or more of the following: computing resource, connection resource.

8. The method of claim 5, wherein, After the optimal one solution is selected from the solution set based on the solution set, the method further includes: Based on the optimal one solution, determining an AI task deployment scheme corresponding to the optimal one solution; Sending the AI task deployment scheme to the first device, and the first device deploys the corresponding AI task or AI subtask on the first device based on the AI task deployment scheme.

9. The method of claim 8, wherein, The method further includes: Receiving quality of service (QoS) information of the AI service sent by the first device; When the QoS information does not meet the AI service requirement, reselecting an optimal solution corresponding to the third network resource from the solution set, and determining an updated AI task deployment scheme based on the optimal solution corresponding to the third network resource.

10. An AI service processing method, characterized in that: Applied to the first device, the method includes: Receiving an AI task deployment scheme sent by a third network device; Based on the AI task deployment scheme, deploying corresponding AI tasks or AI subtasks on the first device.

11. The method of claim 10, wherein, The first device includes one or more of the following: base station, terminal, edge computing node.

12. The method of claim 10, wherein, The number of the first devices is multiple, and after receiving the AI task deployment scheme sent by the third network device, the method further includes: When the AI task deployment scheme is deployed on multiple first devices, based on the AI deployment scheme, executing the life cycle of the corresponding AI tasks and AI subtasks on the multiple first devices, and interacting the intermediate execution results obtained by executing the life cycle of the AI tasks and / or AI subtasks among the multiple first devices; Based on the intermediate execution results, determining a target execution strategy, and executing the target execution strategy by using a target device in the multiple first devices.

13. The method of claim 12, wherein, The method further includes: Obtaining QoS information of an AI service; Sending the QoS information to a third network device, and the QoS information is used for the third network device to reselect an optimal solution corresponding to the third network resource from a solution set when the QoS information does not meet the AI service requirement, and determine an updated AI task deployment scheme based on the optimal solution corresponding to the third network resource.

14. A first network device, comprising: The first network device includes: A first receiving unit configured to receive first information sent by a second network device, the first information being used for the first network device to determine an AI service; A determining unit configured to determine an AI service to be processed based on the first information; A processing unit configured to perform service orchestration processing on the AI service to obtain a solution set corresponding to the AI service; A sending unit configured to send the solution set to a third network device, the solution set being used for the third network device to select an optimal one solution from the solution set.

15. A third network device, comprising: The third network device includes: The second receiving unit is configured to receive a solution set sent by the first network device, wherein the solution set is a set of all solutions corresponding to the AI service; The selecting unit is configured to select an optimal solution from the solution set based on the solution set.

16. A first device, comprising: The first device comprises: The third receiving unit is configured to receive an AI task deployment scheme sent by a third network device; The deployment unit is configured to perform corresponding AI task or AI subtask deployment on the first device based on the AI task deployment scheme.

17. A first network device, comprising: The first network device comprises a first processor and a first memory; the first processor implements the method of any one of claims 1 to 4 when executing a running program stored in the first memory.

18. A third network device, comprising: The third network device comprises a second processor and a second memory; the second processor implements the method of any one of claims 5 to 9 when executing a running program stored in the second memory.

19. A first device, comprising: The first device comprises a third processor and a third memory; the third processor implements the method of any one of claims 10 to 13 when executing a running program stored in the third memory.

20. A storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the method of any one of claims 1 to 4, or the computer program is executed by the processor to implement the method of any one of claims 5 to 9, or the computer program is executed by the processor to implement the method of any one of claims 10 to 13.

21. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1 to 4, or the computer program is executed by the processor to implement the method of any one of claims 5 to 9, or the computer program is executed by the processor to implement the method of any one of claims 10 to 13.

Images