Model lifecycle management method and apparatus, communication device, and storage medium

Abstract

The application relates to a model lifecycle management method and device, a communication device, a storage medium and a computer program product. The method is applied to a first network device, and the method comprises the following steps: receiving signaling sent by a second network device; wherein the signaling at least contains model configuration information; performing an artificial intelligence model training and / or data collection process according to the model configuration information; and sending a model configuration response message to the second network device based on the execution result of the artificial intelligence model training and / or data collection process. By using the method, the artificial intelligence model can be deployed on the wireless access network side, the spectrum efficiency, user experience rate and positioning accuracy can be improved, the time delay can be reduced, the security and privacy protection can be improved, and the AI-related indicators can be optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication, and in particular to a model lifecycle management method and device, a communication device, a storage medium and a computer program product. BACKGROUND

[0002] With the development of wireless communication technology, artificial intelligence native (AI native) design has emerged. The fusion of artificial intelligence (AI) and communication technology will affect related capability indicators in the Key Performance Indicator (KPI) capability indicator system. For example, AI native can improve spectrum efficiency, user experience rate and positioning accuracy, reduce latency, improve security and privacy protection, and optimize AI-related indicators. AI native is a necessary and important technical feature in future mobile communication systems to meet and improve various capability indicators.

[0003] Therefore, there is an urgent need for a model lifecycle management method for deploying an artificial intelligence model on the Radio Access Network (RAN) side. SUMMARY

[0004] The embodiments of the present application provide a model lifecycle management method and device, a communication device, a storage medium and a computer program product, which can deploy an artificial intelligence model on the Radio Access Network (RAN) side.

[0005] A model lifecycle management method, the method is applied to a first network device, and the method comprises:

[0006] receiving signaling sent by a second network device; wherein the signaling contains at least model configuration information;

[0007] According to the model configuration information, performing an artificial intelligence model training and / or data collection process;

[0008] Based on the execution result of the artificial intelligence model training and / or data collection process, sending a model configuration response message to the second network device.

[0009] In one embodiment, the model configuration information contains at least one of indication information and training configuration information; the indication information is used to indicate whether to perform artificial intelligence model training; and the training configuration information includes at least one of model construction geographic range, storage space information, computing power information, model structure configuration information, training parameter information and data collection configuration information.

[0010] In one of the embodiments, the storage space information represents the storage space required by the artificial intelligence model training and / or data collection process; and the computing power information represents the computing power required by the artificial intelligence model training and / or data collection process.

[0011] In one of the embodiments, the model structure configuration information includes at least one of pre-training indication information, model type, and model structure parameters; and the pre-training indication information represents whether it is a pre-trained model.

[0012] In one of the embodiments, the training parameter information includes at least one of dataset size, loss function type, training period, deployment condition, maximum model training times, and training interval.

[0013] In one of the embodiments, the data collection configuration information includes at least one of dataset number, dataset type, dataset function, data collection pilot signal configuration, data collection period, data collection sample type, data collection precision, and data collection amount.

[0014] In one of the embodiments, when the model configuration information includes data collection configuration information, the executing the artificial intelligence model training and / or data collection process according to the model configuration information includes at least one of:

[0015] executing a data collection process based on the data collection configuration information; and,

[0016] obtaining training sample data, and training an artificial intelligence model based on the training sample data and the model configuration information; the training sample data includes at least one of collected sample data and / or pre-stored sample data.

[0017] In one of the embodiments, the obtaining training sample data includes at least one of:

[0018] If the data amount of the pre-stored sample data is less than a preset data amount threshold, the collected sample data is obtained, and the step of training an artificial intelligence model based on the training sample data and the model configuration information is executed when the total data amount of the collected sample data and the pre-stored sample data reaches the data amount threshold.

[0019] If the data amount of the pre-stored sample data is greater than or equal to the preset data amount threshold, the pre-stored sample data is used as the training sample data.

[0020] In one of the embodiments, the obtaining training sample data includes:

[0021] After the data collection is completed, the collected sample data is taken as training sample data.

[0022] In one of the embodiments, based on the execution result of the artificial intelligence model training and / or data collection process, a model configuration response message is sent to the second network device, including at least one of the following:

[0023] If the artificial intelligence model training or the data collection process fails to start, a start failure response message is sent to the second network device;

[0024] If the artificial intelligence model training starts successfully, a model training start response message is sent to the second network device;

[0025] If the artificial intelligence model training is completed, a model training completion message is sent to the second network device.

[0026] In one of the embodiments, after the model training completion message is sent to the second network device, the method further includes at least one of the following:

[0027] A model deployment process is performed on the trained artificial intelligence model;

[0028] A model deployment message sent by the second network device is received, and a model deployment process is performed on the trained artificial intelligence model.

[0029] In one of the embodiments, the method further includes:

[0030] When the artificial intelligence model training fails, a cumulative model training number of the artificial intelligence model is determined.

[0031] In one of the embodiments, the method further includes at least one of the following:

[0032] If the cumulative model training number is less than a maximum model training number, the artificial intelligence model is retrained, and the cumulative model training number is updated;

[0033] If the cumulative model training number is greater than or equal to the maximum model training number, the model training of the artificial intelligence model is stopped, a model training timer is started, and when the timing duration of the model training timer is greater than or equal to a training interval, the artificial intelligence model is retrained.

[0034] In one of the embodiments, before the artificial intelligence model is retrained, the method further includes:

[0035] According to a preset model update strategy, the artificial intelligence model is updated.

[0036] In one of the embodiments, the updating the artificial intelligence model according to the preset model updating strategy comprises at least one of the following:

[0037] changing a model structure of the artificial intelligence model according to the preset model updating strategy;

[0038] changing an impulse function or a loss function of the artificial intelligence model according to the preset model updating strategy;

[0039] transforming a model type of the artificial intelligence model according to the preset model updating strategy.

[0040] In one of the embodiments, before the retraining the artificial intelligence model, the method further comprises:

[0041] changing the training sample data according to a preset training sample data updating strategy.

[0042] In one of the embodiments, the first network device comprises at least one of a terminal, a base station, a master board card, an intelligent computing board card, a computing power board card, a baseband board and a radio access network network element.

[0043] In one of the embodiments, the signaling comprises at least one of user plane signaling, intelligent plane signaling, data plane signaling, computing plane signaling, radio resource control signaling, medium access control-control element, downlink control information, management plane signaling, control plane signaling, system message, non-access layer signaling, access layer signaling and dedicated configuration signaling.

[0044] A model lifecycle management method, the method is applied to a second network device, and the method comprises:

[0045] sending signaling to a first network device; wherein the signaling at least contains model configuration information, the model configuration information instructs the first network device to execute an artificial intelligence model training and / or data collection process;

[0046] receiving a model configuration response message sent by the first network device; the model configuration response message is generated based on an execution result of the artificial intelligence model training and / or data collection process.

[0047] In one of the embodiments, the receiving the model configuration response message sent by the first network device comprises at least one of the following:

[0048] receiving a start failure response message sent by the first network device;

[0049] receiving a model training start response message sent by the first network device;

[0050] receive a model training completion message sent by the first network device.

[0051] In one of the embodiments, after the receiving the model training completion message sent by the first network device, the method further comprises:

[0052] sending a model deployment message to the first network device; the model deployment message is used to instruct the first network device to perform a model deployment process on the trained artificial intelligence model.

[0053] In one of the embodiments, the second network device comprises at least one of a base station, a core network element, and a network management operation and maintenance management.

[0054] A model lifecycle management apparatus, the apparatus is applied to a first network device, the apparatus comprises:

[0055] a first receiving module, configured to receive signaling sent by a second network device; wherein the signaling at least contains model configuration information;

[0056] a first executing module, configured to execute an artificial intelligence model training and / or data collection process according to the model configuration information;

[0057] a first sending module, configured to send a model configuration response message to the second network device based on the execution result of the artificial intelligence model training and / or data collection process.

[0058] A model lifecycle management apparatus, the apparatus is applied to a second network device, the apparatus comprises:

[0059] a second sending module, configured to send signaling to a first network device; wherein the signaling at least contains model configuration information, the model configuration information instructs the first network device to execute an artificial intelligence model training and / or data collection process;

[0060] a second receiving module, configured to receive a model configuration response message sent by the first network device; the model configuration response message is generated based on the execution result of the artificial intelligence model training and / or data collection process.

[0061] A communication device, comprising a receiver, a processor and a transmitter;

[0062] the receiver is configured to receive signaling sent by a second network device; wherein the signaling at least contains model configuration information;

[0063] the processor is configured to execute an artificial intelligence model training and / or data collection process according to the model configuration information;

[0064] The transmitter is configured to send a model configuration response message to the second network device based on the execution result of the artificial intelligence model training and / or data collection process.

[0065] A communication device comprises a transmitter and a receiver.

[0066] The transmitter is configured to send signaling to a first network device, wherein the signaling contains at least model configuration information indicating that the first network device performs an artificial intelligence model training and / or data collection process.

[0067] The receiver is configured to receive a model configuration response message sent by the first network device, wherein the model configuration response message is generated based on the execution result of the artificial intelligence model training and / or data collection process.

[0068] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above method.

[0069] A chip comprising programmable logic circuitry and / or program instructions, which, when the chip is running, can perform the steps of the above method.

[0070] A computer program product comprising a computer program, which, when executed by a processor, implements the steps of the above method.

[0071] The above model lifecycle management method, device, communication device, storage medium, and computer program product, the method is applied to a first network device, and the method comprises: receiving signaling sent by a second network device, wherein the signaling contains at least model configuration information; performing an artificial intelligence model training and / or data collection process according to the model configuration information; and sending a model configuration response message to the second network device based on the execution result of the artificial intelligence model training and / or data collection process. In this way, the first network device performs an artificial intelligence model training and / or data collection process according to the model configuration information in the signaling sent by the second network device, and sends a model configuration response message to the second network device based on the execution result, thereby realizing the deployment of an artificial intelligence model on the side of a wireless access network, i.e., the first network device, which can improve spectral efficiency, user experience rate, and positioning accuracy, reduce time delay, improve security and privacy protection, and optimize AI-related indicators. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 An application environment diagram of the model lifecycle management method in one embodiment;

[0073] Figure 2 A flowchart of the model lifecycle management method in one embodiment;

[0074] Figure 3 For an embodiment, a flowchart of a process of performing the artificial intelligence model training and / or data collection process step according to the model configuration information in the case that the model configuration information contains data collection configuration information;

[0075] Figure 4 For an embodiment, a flowchart of a process of the step of obtaining the training sample data;

[0076] Figure 5 For an embodiment, a flowchart of a process of the step of sending the model configuration response message to the second network device based on the execution result of the artificial intelligence model training and / or data collection process;

[0077] Figure 6 For an embodiment, a flowchart of a process of the step that the model lifecycle management method can further include after sending the model training completion message to the second network device;

[0078] Figure 7 For an embodiment, a flowchart of a process of the step that the model lifecycle management method can further include;

[0079] Figure 8 For an embodiment, a flowchart of a process of the step of updating the artificial intelligence model according to the preset model update strategy;

[0080] Figure 9 For another embodiment, a flowchart of a process of the model lifecycle management method;

[0081] Figure 10 For an embodiment, a flowchart of a process of the step of receiving the model configuration response message sent by the first network device;

[0082] Figure 11 For an embodiment, a signaling interaction flowchart of the model lifecycle management method;

[0083] Figure 12 For an embodiment, a structural block diagram of the model lifecycle management apparatus;

[0084] Figure 13 For another embodiment, a structural block diagram of the model lifecycle management apparatus;

[0085] Figure 14 For an embodiment, an internal structural diagram of the communication device;

[0086] Figure 15 For another embodiment, an internal structural diagram of the communication device. DETAILED DESCRIPTION

[0087] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0088] Figure 1 An application scenario of a model lifecycle management method provided by an embodiment of the present application is shown in FIG. 1. As shown in FIG. 1, the scenario includes a first network device 100 and a second network device 200. The first network device 100 and the second network device 200 perform data transmission through a network. Figure 1

[0089] The first network device 100 can include at least one of a terminal, a base station, a master board card, an intelligent computing board card, a computing power board card, a baseband board and a radio access network element. The second network device 200 can include at least one of a base station, a core network element and an operation administration and maintenance (OAM).

[0090] The base station can be a base transceiver station (BTS) in a global system of mobile communication (GSM) or a code division multiple access (CDMA), can be a nodeB (NB) in a wideband code division multiple access (WCDMA), can be an evolutional nodeB (eNB or eNodeB) in LTE, or can be a relay station or an access point, or can be a base station in a 5G network, etc., and is not limited herein.

[0091] ​A terminal can be a wireless terminal, which can refer to a device that provides voice and / or other data connectivity to a user, or a hand-held device having a wireless connection capability, or other processing device connected to a wireless modem. The wireless terminal can communicate with one or more core networks via a Radio Access Network (RAN), and can be a mobile terminal, such as a mobile telephone (or "cellular" telephone) and a computer with a mobile termination that can be, for example, portable, pocket, hand-held, computer-included, or car-mounted, and can exchange language and / or data with a radio access network. The wireless terminal can also be referred to as a system, a subscriber unit, a subscriber station, a mobile station, a mobile, a remote station, a remote terminal, an access terminal, a user terminal, a user agent, a user device, or user equipment, without limitation.

[0092] The technical solutions of the present application and how the technical solutions solve the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described again in some embodiments. The embodiments of the present application will be described below with reference to the drawings.

[0093] In one embodiment, as shown in Figure 2 , a model lifecycle management method is provided, which is applied to the first network device 100 in Figure 1 for example, and includes the following steps:

[0094] Step 201, receiving signaling sent by a second network device.

[0095] The signaling contains at least model configuration information.

[0096] In the embodiments of the present application, the lifecycle of the model refers to the entire process from the creation of the model to the abandonment / invalidation of the model. The model lifecycle management can include at least one of model generation, model construction, model training, and model deployment. It can be understood that the model training can represent the model construction or the model generation. The first network device is a network device to which the artificial intelligence model is to be deployed. The artificial intelligence model can be an artificial intelligence / machine learning (AI / ML) model. The artificial intelligence model can be a resource scheduling model and / or a decision-making model. The resource can be a time domain resource, a frequency domain resource, and a space domain resource. The artificial intelligence model can be used in a multi-carrier coordination scenario, a multi-frequency coordination scenario, an Industrial Internet of Things (IIoT) scenario, a telecommunications customized network scenario, and a millimeter wave deployment scenario. The model configuration information indicates the first network device to perform an artificial intelligence model training and / or a data collection process.

[0097] Optionally, the signaling can include at least one of user plane signaling, intelligent plane signaling, data plane signaling, computing plane signaling, Radio Resource Control (RCC) signaling, Medium Access Control- Control Element (MAC CE), Downlink Control Information (DCI), management plane signaling, control plane signaling, system message, Non-Access-stratum (NAS) signaling, Access-stratum (AS) signaling, and dedicated configuration signaling.

[0098] Specifically, the second network device 200 sends signaling containing at least model configuration information to the first network device 100. The first network device 100 receives the signaling sent by the second network device 200.

[0099] In step 202, according to the model configuration information, an artificial intelligence model training and / or a data collection process is performed.

[0100] Optionally, the model configuration information can contain at least one of indication information and training configuration information. The indication information is used to indicate whether to perform the artificial intelligence model training. The training configuration information can include at least one of model construction geographic range, storage space information, computing power information, model structure configuration information, training parameter information, and data collection configuration information.

[0101] Optionally, the model construction geographic range can be determined in one or more of the following manners: according to a neighbor relation or a tracking area (TA), indicated by an array of cell identity numbers (IDs), indicated by a base station identifier (including a plurality of carriers supported by the base station), determined according to a preset neighbor relation group or a cell group (the preset neighbor relation group or the cell group is determined by the second network device 200), and determined according to a configured geographic range base station or cell number.

[0102] Optionally, the storage space information represents a storage space required by the artificial intelligence model training and / or data collection process. The computing power information represents a computing power required by the artificial intelligence model training and / or data collection process. The computing power information can include at least one of computing power requirement information (also referred to as central processing unit (CPU) capability requirement information) and intelligent computing power requirement information (also referred to as graphic processing unit (GPU) capability requirement information).

[0103] Optionally, the model structure configuration information includes at least one of pre-training indication information, a model type, and model structure parameters. The pre-training indication information represents whether it is a pre-trained model. The model type includes at least one of a transformer, a deep neural network (DNN), a convolution neural network (CNN), and a multilayer perceptron (MLP). The model structure parameters represent a model structure configuration. For example, the model structure parameters are in a standardized representation format or in a mainstream model representation format.

[0104] Optionally, the training parameter information can include at least one of a data set size, a loss function type, a training period, a deployment condition (also referred to as a deployment threshold), a maximum model training number, and a training interval (also referred to as a training duration or a training waiting window). The data set size can be a batch size of the data set.

[0105] Optionally, the data collection configuration information can include at least one of a data set number, a data set type, a data set function, a data collection pilot signal configuration, a data collection period, a data collection sample type, a data collection precision, and a data collection amount.

[0106] Step 203, based on the execution result of the artificial intelligence model training and / or data collection process, a model configuration response message is sent to the second network device.

[0107] In an example embodiment, a base station receives signaling sent by a network management operation and maintenance management. The signaling contains at least model configuration information. The model configuration information can include at least one of an on indication of a time domain artificial intelligence model, an off indication of the time domain artificial intelligence model, an on indication of a frequency domain artificial intelligence model, an off indication of the frequency domain artificial intelligence model, an on indication of a space domain artificial intelligence model, an off indication of the space domain artificial intelligence model, a model enabling threshold, a performance detection threshold, a model training period, a data collection time length, a serving cell frequency point, and a neighbor cell frequency point. Then, the base station performs an artificial intelligence model training and / or data collection process according to the model configuration information. Specifically, the base station collects resource scheduling management data of a radio access network according to the model configuration information. The artificial intelligence model is a resource scheduling management model of the radio access network. The resource scheduling management data can include at least one of a reference signal receiving power (RSRP) of a time domain serving cell frequency point and a neighbor cell frequency point, a reference signal receiving power of a frequency domain serving cell frequency point and a neighbor cell frequency point, a reference signal receiving power of a space domain serving cell frequency point and a neighbor cell frequency point, a user transmission rate, a cell spectrum efficiency, a user transmission signal to noise ratio (SINR), a layer 1 (L1) or layer 3 (L3) reference signal receiving power. Then, the base station sends a model configuration response message to the network management operation and maintenance management based on an execution result of the artificial intelligence model training and / or data collection process.

[0108] In another example embodiment, the base station receives signaling sent by the network management operation and maintenance management. The signaling contains at least model configuration information. The model configuration information can include at least one of a measurement quantity of a model input, a base station identifier, a transmission and receiving point (TRP) identifier, and a beam identifier. The measurement quantity of the model input includes at least one of a channel impulse response (CIR), a reference signal received power, a time of arrival (TOA), and a time difference of arrival (TDOA). Then, the base station performs an artificial intelligence model training and / or data collection process according to the model configuration information. Specifically, the base station collects AI positioning enhancement information or AI-based beam management information according to the model configuration information. The artificial intelligence model is an AI positioning enhancement model or an AI-based beam management model. The AI positioning enhancement information can include at least one of different cell reference signal received powers, times of arrival, time differences of arrival, channel impulse responses, beam identifiers, cell identifiers, and frequency point identifiers. The AI-based beam management information can include at least one of different cell reference signal received powers, times of arrival, time differences of arrival, channel impulse responses, beam identifiers, cell identifiers, and frequency point identifiers. Then, the base station sends a model configuration response message to the network management operation and maintenance management based on the execution result of the artificial intelligence model training and / or data collection process.

[0109] Through the scheme, the first network device performs an artificial intelligence model training and / or data collection process according to model configuration information in signaling sent by the second network device, and sends a model configuration response message to the second network device based on the execution result, so that the artificial intelligence model is deployed on the first network device, i.e., the wireless access network side, which can improve the spectrum efficiency, user experience rate, and positioning accuracy, reduce the time delay, improve the security and privacy protection, and optimize the AI-related indicators. Moreover, the scheme can realize signaling configuration about AI / ML model deployment between the wireless network and the base station on the RAN side, enable the network to control the AI model generation and training, and enable AI to improve the performance of the wireless air interface system. Furthermore, the scheme defines a specific configuration scheme for model construction and model training of the wireless access network, defines a specific design scheme for related signaling interaction and configuration information for configuring model construction for model training by the base station, gateway, or core network element, and enables the base station, gateway, or core network element to perform model construction on the base station, intelligent computing board card, main control board, or new network element on the RAN side through the model configuration information, which has high flexibility.

[0110] In one embodiment, as Figure 3As shown, in a case where the model configuration information comprises the data collection configuration information, performing the artificial intelligence model training and / or the data collection process according to the model configuration information can comprise at least one of the following:

[0111] At step 301, the data collection process is performed based on the data collection configuration information.

[0112] Optionally, in a case where the data collection configuration information comprises a data collection period, the first network device performs the data collection process according to the data collection period.

[0113] Optionally, if the artificial intelligence model satisfies a preset model deployment condition, the data collection process is stopped.

[0114] At step 302, the training sample data is obtained, and the artificial intelligence model is trained based on the training sample data and the model configuration information.

[0115] The training sample data comprises at least one of the collected sample data and / or the pre-stored sample data.

[0116] Optionally, in a case where the model configuration information comprises storage space information and / or computing power information, the first network device determines whether to start the artificial intelligence model training based on the storage space information and / or the computing power information. If it is determined to start the artificial intelligence model training, the first network device trains the artificial intelligence model based on the training sample data and the model configuration information.

[0117] Through the scheme, the model configuration information comprises the data collection configuration information, the data collection process is performed based on the data collection configuration information, the training sample data is obtained, and the artificial intelligence model is trained based on the training sample data and the model configuration information, thereby realizing that the first network device performs the artificial intelligence model training and / or the data collection process, and thereby realizing that the artificial intelligence model is deployed at the first network device, i.e., the wireless access network side.

[0118] In one embodiment, as shown, Figure 4 The training sample data can comprise at least one of the following:

[0119] At step 401, if the data amount of the pre-stored sample data is less than a preset data amount threshold, the collected sample data is obtained, and the step of training the artificial intelligence model based on the training sample data and the model configuration information is performed when the total data amount of the collected sample data and the pre-stored sample data reaches the data amount threshold.

[0120] In the embodiments of the present application, different artificial intelligence models correspond to different data amount thresholds. The data amount of the pre-stored sample data can be 0.

[0121] Specifically, if the data amount of the pre-stored sample data is less than the preset data amount threshold, the first network device acquires the collected sample data. When the total data amount of the collected sample data and the pre-stored sample data reaches the data amount threshold, the first network device constructs the collected sample data and the pre-stored sample data into training sample data, and performs the step of training the artificial intelligence model based on the training sample data and the model configuration information.

[0122] In step 402, if the data amount of the pre-stored sample data is greater than or equal to the preset data amount threshold, the pre-stored sample data is taken as the training sample data.

[0123] Through the scheme, the data amount of the pre-stored sample data is compared with the preset data amount threshold. In the case that the data amount of the pre-stored sample data is insufficient, the collected sample data and the pre-stored sample data are constructed into training sample data. In the case that the data amount of the pre-stored sample data is sufficient, the pre-stored sample data is directly taken as the training sample data. The data amount used for training the artificial intelligence model can be ensured to be sufficient, and the first network device can perform the artificial intelligence model training and / or data collection process, and the artificial intelligence model can be deployed at the first network device, i.e., the wireless access network side. Moreover, the scheme can realize decoupling of the training sample data and the artificial intelligence model.

[0124] In one embodiment, acquiring the training sample data can include: after the data collection is completed, taking the collected sample data as the training sample data.

[0125] Through the scheme, the sample data collected in this data collection process is directly taken as the training sample data. The data amount used for training the artificial intelligence model can be ensured to be sufficient, the first network device can perform the artificial intelligence model training and / or data collection process, the artificial intelligence model can be deployed at the first network device, i.e., the wireless access network side, the real-time performance of the sample training data can be improved, and the accuracy of the artificial intelligence model can be improved.

[0126] In one embodiment, as shown in FIG. 5, based on the execution result of the artificial intelligence model training and / or data collection process, the model configuration response message can be sent to the second network device, which can include at least one of the following: Figure 5

[0127] In step 501, if the artificial intelligence model training or the data collection process fails to start, a start failure response message is sent to the second network device.

[0128] Optionally, the start failure response message can include failure cause information. The failure cause information can be an error code. The failure cause information can include at least one of insufficient computing power, insufficient storage space, busy base station, and invalid configuration.​

[0129] Optionally, the first network device can perform the process of handling the failure of starting the artificial intelligence model training or the data collection process according to a preset starting failure handling rule.

[0130] In the embodiment of the present application, the model training starting response message indicates that the starting of the artificial intelligence model training is successful.

[0131] In the embodiment of the present application, the model training completion message indicates that the artificial intelligence model training is completed.

[0132] In the embodiment of the present application, the model training completion message indicates that the artificial intelligence model training is completed.

[0133] In the embodiment of the present application, the model training completion message indicates that the artificial intelligence model training is completed.

[0134] Through the scheme, the starting failure response message is sent to the second network device, the process of handling the failure of starting the artificial intelligence model training or the data collection process can be performed in time, the deployment of the artificial intelligence model at the first network device, i.e., the wireless access network side, is guaranteed, and the efficiency of deploying the artificial intelligence model at the first network device, i.e., the wireless access network side, is improved; the model training starting response message and the model training completion message are sent to the second network device, the stable model training is guaranteed, and the deployment of the artificial intelligence model at the first network device, i.e., the wireless access network side, is guaranteed.

[0135] In one embodiment, as shown in FIG. 6, after the model training completion message is sent to the second network device, the method can further include at least one of the following: Figure 6

[0136] Step 601: performing a model deployment process on the trained artificial intelligence model.

[0137] In the embodiment of the present application, after the model training completion message is sent to the second network device, the first network device performs the model deployment process on the trained artificial intelligence model.

[0138] Step 602: receiving a model deployment message sent by the second network device, and performing the model deployment process on the trained artificial intelligence model.

[0139] In the embodiment of the present application, the model deployment message (also referred to as the model online message) instructs the first network device to perform the model deployment process on the trained artificial intelligence model.

[0140] ​Specifically, after sending the model training completion message to the second network device, the first network device waits for the model deployment message sent by the second network device. Then, the first network device receives the model deployment message sent by the second network device, and performs the model deployment process on the trained artificial intelligence model.

[0141] By the scheme, after sending the model training completion message to the second network device, the model deployment process is directly performed on the trained artificial intelligence model, which can improve the efficiency of deploying the artificial intelligence model at the first network device, i.e., the wireless access network side; and after receiving the model deployment message sent by the second network device, the model deployment process is performed on the trained artificial intelligence model again, which can improve the flexibility of deploying the artificial intelligence model at the first network device, i.e., the wireless access network side.

[0142] In one embodiment, the method can further include: when the artificial intelligence model training fails, determining a cumulative model training number of the artificial intelligence model.

[0143] Optionally, if the artificial intelligence model does not converge after the model training of the preset training round, the first network device determines that the artificial intelligence model training fails.

[0144] Optionally, when the artificial intelligence model training fails, the first network device obtains a cumulative model training number of the artificial intelligence model.

[0145] In one embodiment, as shown in FIG. 7, Figure 7 the method can further include at least one of the following:

[0146] Step 701: If the cumulative model training number is less than the maximum model training number, the artificial intelligence model is retrained, and the cumulative model training number is updated.

[0147] In the embodiment of the present application, the first network device compares the cumulative model training number with the maximum model training number contained in the model configuration information. If the cumulative model training number is less than the maximum model training number, the first network device re-trains the artificial intelligence model, and adds 1 to the cumulative model training number.

[0148] Step 702: If the cumulative model training number is greater than or equal to the maximum model training number, the model training of the artificial intelligence model is stopped, a model training timer is started, and when the timing duration of the model training timer is greater than or equal to the training interval, the artificial intelligence model is retrained.

[0149] In the embodiment of the present application, if the accumulated model training number is greater than or equal to the maximum model training number, the first network device stops the model training of the artificial intelligence model, starts a model training timer, or resets the model training timer. When the timing duration of the model training timer is greater than or equal to the training interval, the first network device re-trains the artificial intelligence model.

[0150] Through the present solution, when the artificial intelligence model training fails, the accumulated model training number of the artificial intelligence model is determined, and when the accumulated model training number is greater than or equal to the maximum model training number, the model training of the artificial intelligence model is stopped, and after the training interval, the artificial intelligence model is re-trained, which can avoid the influence of the current network condition on the artificial intelligence model training, save computing power, and reduce energy consumption. Moreover, the present solution proposes a variety of response mechanisms when the model training succeeds or fails, supports the sending of model training related configurations, supports attempting multiple times of model training in the case of timer configuration and / or maximum training number configuration when the model training fails, improves the success probability of model construction, and can be implemented in a live network.

[0151] In one embodiment, before re-training the artificial intelligence model, the method can further include: updating the artificial intelligence model according to a preset model update strategy.

[0152] In the embodiment of the present application, the model update strategy can include at least one of a model structure update strategy, an impulse function update strategy, a loss function update strategy, and a model type update strategy.

[0153] Through the present solution, before re-training the artificial intelligence model, the artificial intelligence model is updated, which improves the success probability of model training and improves the success probability of deploying the artificial intelligence model on the first network device, i.e., the wireless access network side, and can be implemented in a live network.

[0154] In one embodiment, as shown in Figure 8 According to the preset model update strategy, updating the artificial intelligence model can include at least one of the following:

[0155] Step 801: According to a preset model update strategy, the model structure of the artificial intelligence model is changed.

[0156] In the embodiment of the present application, changing the model structure of the artificial intelligence model can include at least one of adding a convolution layer, reducing a convolution layer, adding a self-attention module, reducing a self-attention module, adding a residual structure, reducing a residual structure, adding a fully connected layer, and reducing a fully connected layer.

[0157] Specifically, the first network device determines a target model structure and a target number of layers corresponding to the target model structure according to a preset model updating strategy. Then, the first network device updates the number of layers of the target model structure to the target number of layers. The target model structure can include at least one of a convolutional layer, a self-attention module, a residual structure, and a fully connected layer.

[0158] Optionally, in a case where the model updating strategy includes a model structure change sequence and a preset mapping relationship between a model structure update number and a number of layers, the first network device determines a target model structure changed this time according to the model structure change sequence. Then, the first network device queries, in the preset mapping relationship between the model structure update number and the number of layers, a target number of layers corresponding to the target model structure changed this time. The model structure change sequence represents an order of changing model structures.

[0159] Optionally, the first network device randomly determines, in the preset model structure, a target model structure changed this time. Then, the first network device randomly determines, in the preset number of layers, a target number of layers of the target model structure changed this time.

[0160] Step 802, according to a preset model updating strategy, changing an impulse function or a loss function of an artificial intelligence model.

[0161] In an embodiment of the present application, the first network device determines a target impulse function or a target loss function of the artificial intelligence model according to a preset model updating strategy. Then, the first network device changes the impulse function or the loss function of the artificial intelligence model to the target impulse function or the target loss function.

[0162] Optionally, the first network device determines a target impulse function or a target loss function changed this time according to a preset impulse function or loss function complexity sequence. The impulse function or loss function complexity sequence represents a complexity ranking of each impulse function or loss function.

[0163] Optionally, the first network device randomly determines, in the preset impulse function or loss function, a target impulse function or a target loss function of the artificial intelligence model.

[0164] Step 803, according to a preset model updating strategy, transforming a model type of an artificial intelligence model.

[0165] In an embodiment of the present application, the first network device determines a target model type of the artificial intelligence model according to a preset model updating strategy. Then, the first network device changes the model type of the artificial intelligence model to the target model type.

[0166] Optionally, the first network device determines the target impulse function or the target model type of the current change according to a preset model type complexity sequence. The model type complexity sequence represents the complexity ranking of each model type.

[0167] Optionally, the first network device randomly determines the target model type of the artificial intelligence model in the preset model type.

[0168] Optionally, the first network device determines the target model update type of the current artificial intelligence model update according to an artificial intelligence model update complexity sequence. Then, the first network device updates the artificial intelligence model according to the target model update type. The artificial intelligence model update complexity sequence represents the complexity ranking of the model update types of various artificial intelligence models from small to large. The model update type can include at least one of changing the model structure of the artificial intelligence model, changing the impulse function or loss function of the artificial intelligence model, and transforming the model type of the artificial intelligence model. For example, the artificial intelligence model update complexity sequence can be, in sequence, changing the model structure of the artificial intelligence model, changing the impulse function or loss function of the artificial intelligence model, and transforming the model type of the artificial intelligence model.

[0169] Through the scheme, before retraining the artificial intelligence model, the artificial intelligence model is updated by at least one of changing the model structure of the artificial intelligence model, changing the impulse function or loss function of the artificial intelligence model, and transforming the model type of the artificial intelligence model, a plurality of model enhancement mechanisms before retraining the model after the model training fails are proposed, the model is optimized, the model training performance is enhanced, the success probability of the model training is further improved, and the success probability of deploying the artificial intelligence model at the first network device, i.e., the radio access network side, is further improved.

[0170] In one embodiment, before retraining the artificial intelligence model, the method can further include: changing the training sample data according to a preset training sample data update strategy.

[0171] Optionally, the first network device uses different numbered data sets as training sample data.

[0172] Optionally, the first network device uses a plurality of numbered data sets to construct the training sample data.

[0173] Through the scheme, before retraining the artificial intelligence model, the training sample data is changed, the success probability of the model training is improved, the success probability of deploying the artificial intelligence model at the first network device, i.e., the radio access network side, is improved, and the scheme can be implemented in the existing network.

[0174] In one embodiment, as Figure 9As shown, a model lifecycle management method is provided, and the method is applied to Figure 1 The second network device 200 in the network is taken as an example for illustration, and the method comprises the following steps:

[0175] Step 901, signaling is sent to the first network device.

[0176] The signaling contains at least model configuration information, and the model configuration information indicates the first network device to execute an artificial intelligence model training and / or data collection process.

[0177] Step 902, a model configuration response message sent by the first network device is received.

[0178] The model configuration response message is generated based on an execution result of the artificial intelligence model training and / or data collection process.

[0179] Through the scheme, the second network device sends signaling containing at least model configuration information to the first network device, indicates the first network device to execute an artificial intelligence model training and / or data collection process, and receives a model configuration response message sent by the first network device, which is generated based on an execution result of the artificial intelligence model training and / or data collection process. The artificial intelligence model can be deployed on the first network device, i.e., the wireless access network side, through the second network device, the spectrum efficiency, user experience rate and positioning accuracy can be improved, the time delay can be reduced, the security and privacy protection can be improved, and the AI-related indicators can be optimized. Moreover, the scheme can realize signaling configuration about AI / ML model deployment between the wireless network and the base station on the RAN side, enable the network to control the generation and training of the AI model, and enable AI to improve the performance of the wireless air interface system. Moreover, the scheme defines a specific configuration scheme for model construction and model training of the wireless access network, defines a specific design scheme for related signaling interaction and configuration information for configuring model construction of the base station and other RAN side network elements for model training by the base station, the gateway or the core network element, and enables the base station, the gateway or the core network element to perform model construction on the base station, the intelligent computing board card, the main control board or the new network element on the RAN side and other network element devices through the model configuration information, thereby achieving high flexibility.

[0180] In one embodiment, as shown in Figure 10 The model configuration response message sent by the first network device can comprise at least one of the following:

[0181] Step 1001, a start failure response message sent by the first network device is received.

[0182] Step 1002, a model training start response message sent by the first network device is received.

[0183] Step 1003, a model training completion message sent by the first network device is received.

[0184] By the scheme, the starting failure response message sent by the first network device is received, the processing of the starting failure of the artificial intelligence model training or data collection process is performed in time, the artificial intelligence model is deployed on the first network device, i.e., the wireless access network side, and the efficiency of deploying the artificial intelligence model on the first network device, i.e., the wireless access network side, is improved; the model training starting response message and the model training completion message sent by the first network device are received, the stable model training is ensured, and the artificial intelligence model is deployed on the first network device, i.e., the wireless access network side, through the second network device.

[0185] In one embodiment, after receiving the model training completion message sent by the first network device, the method can further include: sending a model deployment message to the first network device.

[0186] The model deployment message is used to instruct the first network device to perform a model deployment process on the trained artificial intelligence model.

[0187] By the scheme, after receiving the model training completion message sent by the first network device, the model deployment message is sent to the first network device to instruct the first network device to perform a model deployment process on the trained artificial intelligence model, and the flexibility of deploying the artificial intelligence model on the first network device, i.e., the wireless access network side, is improved.

[0188] In one embodiment, Figure 11 A signaling interaction flowchart of a model life cycle management method is provided. As shown in the figure, the method is applied to a model life cycle management system, and the model life cycle management system at least includes a first network device and a second network device. The method includes the following steps. Figure 11

[0189] Step 1101: The second network device sends signaling to the first network device. The signaling at least includes model configuration information, and the model configuration information instructs the first network device to perform an artificial intelligence model training and / or data collection process.

[0190] Step 1102: The first network device performs an artificial intelligence model training and / or data collection process according to the model configuration information.

[0191] Step 1103: The first network device determines whether the artificial intelligence model training or data collection process is started successfully.

[0192] If the artificial intelligence model training or data collection process is started unsuccessfully, step 1104 is performed; if the artificial intelligence model training is started successfully, step 1105 is performed.

[0193] Step 1104: The first network device sends a starting failure response message to the second network device.​

[0194] In step 1105, the first network device sends a model training start response message to the second network device.

[0195] In step 1106, the first network device determines whether the artificial intelligence model training is completed.

[0196] If the artificial intelligence model training is completed, step 1107 is performed; otherwise, no processing is performed.

[0197] In step 1107, the first network device sends a model training completion message to the second network device.

[0198] In step 1108, the second network device sends a model deployment message to the first network device.

[0199] In step 1109, the first network device performs a model deployment process on the trained artificial intelligence model.

[0200] It should be understood that, although Figure 2-11 the steps in the flowchart of the model lifecycle management method are displayed in sequence according to the arrows, these steps are not necessarily executed in sequence according to the arrows. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, Figure 2-11 at least part of the steps in the model lifecycle management method can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0201] In one embodiment, as shown in Figure 12 a model lifecycle management apparatus 1200 is provided, which is applied to a first network device and includes a first receiving module 1210, a first execution module 1220, and a first sending module 1230, wherein:

[0202] The first receiving module 1210 is configured to receive signaling sent by a second network device, wherein the signaling at least contains model configuration information.

[0203] The first execution module 1220 is configured to perform an artificial intelligence model training and / or data collection process according to the model configuration information.

[0204] The first sending module 1230 is configured to send a model configuration response message to the second network device based on the execution result of the artificial intelligence model training and / or data collection process.

[0205] Optionally, the model configuration information at least includes one of indication information and training configuration information; the indication information is used to indicate whether to perform artificial intelligence model training; and the training configuration information includes at least one of model construction geographic range, storage space information, computing power information, model structure configuration information, training parameter information and data collection configuration information.

[0206] Optionally, the storage space information represents storage space required by the artificial intelligence model training and / or data collection process; and the computing power information represents computing power required by the artificial intelligence model training and / or data collection process.

[0207] Optionally, the model structure configuration information includes at least one of pre-training indication information, model type and model structure parameter; and the pre-training indication information represents whether it is a pre-trained model.

[0208] Optionally, the training parameter information includes at least one of data set size, loss function type, training period, deployment condition, maximum model training number and training interval.

[0209] Optionally, the data collection configuration information includes at least one of data set number, data set type, data set function, data collection pilot signal configuration, data collection period, data collection sample type, data collection precision and data collection amount.

[0210] Optionally, in the case where the model configuration information includes the data collection configuration information, the first execution module 1220 is specifically configured to:

[0211] perform a data collection process based on the data collection configuration information; and

[0212] obtain training sample data, and train an artificial intelligence model based on the training sample data and the model configuration information; the training sample data includes at least one of collected sample data and / or pre-stored sample data.

[0213] Optionally, the first execution module 1220 is specifically configured to:

[0214] if the data amount of the pre-stored sample data is less than a preset data amount threshold, obtain the collected sample data, and perform the step of training the artificial intelligence model based on the training sample data and the model configuration information when the total data amount of the collected sample data and the pre-stored sample data reaches the data amount threshold;

[0215] if the data amount of the pre-stored sample data is greater than or equal to the preset data amount threshold, use the pre-stored sample data as the training sample data.

[0216] Optionally, the first execution module 1220 is specifically configured to:

[0217] After the data collection is completed, the collected sample data is taken as training sample data.

[0218] Optionally, the first sending module 1230 is specifically configured to:

[0219] If the artificial intelligence model training or the data collection process fails to start, a start failure response message is sent to the second network device;

[0220] If the artificial intelligence model training starts successfully, a model training start response message is sent to the second network device;

[0221] If the artificial intelligence model training is completed, a model training completion message is sent to the second network device.

[0222] Optionally, the apparatus 1200 further includes at least one of the following:

[0223] A second execution module configured to execute a model deployment process on the trained artificial intelligence model;

[0224] A third execution module configured to receive a model deployment message sent by the second network device and execute a model deployment process on the trained artificial intelligence model.

[0225] Optionally, the apparatus 1200 further includes:

[0226] A determination module configured to determine a cumulative model training number of the artificial intelligence model when the artificial intelligence model training fails.

[0227] Optionally, the apparatus 1200 further includes at least one of the following:

[0228] A first retraining module configured to retrain the artificial intelligence model and update the cumulative model training number if the cumulative model training number is less than a maximum model training number;

[0229] A second retraining module configured to stop the model training of the artificial intelligence model, start a model training timer, and retrain the artificial intelligence model when a timing duration of the model training timer is greater than or equal to a training interval if the cumulative model training number is greater than or equal to the maximum model training number.

[0230] Optionally, the apparatus 1200 further includes:

[0231] A first updating module configured to update the artificial intelligence model according to a preset model updating strategy.

[0232] Optionally, the first updating module is specifically configured to:

[0233] change a model structure of the artificial intelligence model according to a preset model updating strategy;

[0234] change an impulse function or a loss function of the artificial intelligence model according to the preset model updating strategy;

[0235] transform a model type of the artificial intelligence model according to the preset model updating strategy.

[0236] Optionally, the apparatus 1200 further includes:

[0237] a second updating module configured to change the training sample data according to a preset training sample data updating strategy.

[0238] Optionally, the first network device includes at least one of a terminal, a base station, a master board card, an intelligent computing board card, a computing power board card, a baseband board, and a radio access network network element.

[0239] Optionally, the signaling includes at least one of user plane signaling, intelligent plane signaling, data plane signaling, computing plane signaling, radio resource control signaling, medium access control-control element, downlink control information, management plane signaling, control plane signaling, system message, non-access stratum signaling, access stratum signaling, and dedicated configuration signaling.

[0240] In one embodiment, as shown in Figure 13 FIG. 13, a model lifecycle management apparatus 1300 is provided, which is applied to a second network device and includes a second sending module 1310 and a second receiving module 1320, wherein:

[0241] The second sending module 1310 is configured to send signaling to a first network device, wherein the signaling at least includes model configuration information, and the model configuration information indicates the first network device to perform an artificial intelligence model training and / or data collection process.

[0242] The second receiving module 1320 is configured to receive a model configuration response message sent by the first network device, and the model configuration response message is generated based on an execution result of the artificial intelligence model training and / or data collection process.

[0243] Optionally, the second receiving module 1320 is specifically configured to:

[0244] receive a start failure response message sent by the first network device;

[0245] receive a model training start response message sent by the first network device;

[0246] receiving the model training completion message sent by the first network device.

[0247] Optionally, the apparatus 1300 further includes:

[0248] a third sending module, configured to send a model deployment message to the first network device; the model deployment message is used to instruct the first network device to perform a model deployment process on the trained artificial intelligence model.

[0249] Optionally, the second network device includes at least one of a base station, a core network element, and network management operation and maintenance management.

[0250] The specific limitations of the model life cycle management apparatus can be referred to the limitations of the model life cycle management method in the foregoing, which will not be repeated here. Each module in the above model life cycle management apparatus can be realized by software, hardware, and combinations thereof, in whole or in part. The above modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to call and execute the operations corresponding to each module by the processor.

[0251] Figure 14 The access network device provided by the embodiment of the present application is shown in the structural schematic diagram. The access network device can include a receiver 1401, a memory 1402, a processor 1403, at least one communication bus 1404, and a transmitter 1405. The communication bus 1404 is used to realize the communication connection between the elements. The memory 1402 can contain a high-speed RAM memory, and can also include a non-volatile storage NVM, such as at least one disk memory. The memory 1402 can store various programs for completing various processing functions and implementing the method steps of the embodiment. In the embodiment, the transmitter 1405 can be a radio frequency processing module or a baseband processing module in the access network device, and the receiver 1401 can also be a radio frequency processing module or a baseband processing module in the access network device. The transmitter 1405 and the receiver 1401 can be integrated together to realize a transceiver. The transmitter 1405 and the receiver 1401 can be coupled to the processor 1403, which can realize the receiving or transmitting action under the indication or control action of the processor 1403.

[0252] In the embodiment, the receiver 1401 is used to receive the signaling sent by the second network device; wherein the signaling at least contains model configuration information;

[0253] The processor 1403 is used to perform an artificial intelligence model training and / or data collection process according to the model configuration information;

[0254] The transmitter 1405 is configured to send a model configuration response message to the second network device based on the execution result of the artificial intelligence model training and / or data collection process.

[0255] In one embodiment, the model configuration information at least includes one of indication information and training configuration information; the indication information is used to indicate whether to perform the artificial intelligence model training; and the training configuration information includes at least one of a model construction geographical range, storage space information, computing power information, model structure configuration information, training parameter information, and data collection configuration information.

[0256] In one embodiment, the storage space information represents storage space required by the artificial intelligence model training and / or data collection process; and the computing power information represents computing power required by the artificial intelligence model training and / or data collection process.

[0257] In one embodiment, the model structure configuration information includes at least one of pre-training indication information, a model type, and model structure parameters; and the pre-training indication information represents whether it is a pre-trained model.

[0258] In one embodiment, the training parameter information includes at least one of a data set size, a loss function type, a training period, a deployment condition, a maximum model training number, and a training interval.

[0259] In one embodiment, the data collection configuration information includes at least one of a data set number, a data set type, a data set function, a data collection pilot signal configuration, a data collection period, a data collection sample type, a data collection precision, and a data collection amount.

[0260] In one embodiment, when the model configuration information includes the data collection configuration information, the processor 1403 is specifically configured to:

[0261] perform a data collection process based on the data collection configuration information; and

[0262] obtain training sample data, and train the artificial intelligence model based on the training sample data and the model configuration information; the training sample data includes at least one of the collected sample data and / or pre-stored sample data.

[0263] In one embodiment, the processor 1403 is specifically configured to:

[0264] if the data amount of the pre-stored sample data is less than a preset data amount threshold, obtain the collected sample data, and perform the step of training the artificial intelligence model based on the training sample data and the model configuration information when the total data amount of the collected sample data and the pre-stored sample data reaches the data amount threshold;

[0265] If the data amount of the pre-stored sample data is greater than or equal to the preset data amount threshold, the pre-stored sample data is used as the training sample data.

[0266] In one embodiment, the processor 1403 is specifically configured to:

[0267] After the data collection is completed, the collected sample data is used as the training sample data.

[0268] In one embodiment, the transmitter 1405 is specifically configured to:

[0269] If the artificial intelligence model training or the data collection process fails to start, a start failure response message is sent to the second network device;

[0270] If the artificial intelligence model training starts successfully, a model training start response message is sent to the second network device;

[0271] If the artificial intelligence model training is completed, a model training completion message is sent to the second network device.

[0272] In one embodiment, the processor 1403 is further configured to:

[0273] performing a model deployment process on the trained artificial intelligence model;

[0274] receiving a model deployment message sent by the second network device, and performing a model deployment process on the trained artificial intelligence model.

[0275] In one embodiment, the processor 1403 is further configured to:

[0276] When the artificial intelligence model training fails, determining a cumulative model training number of the artificial intelligence model.

[0277] In one embodiment, the processor 1403 is further configured to:

[0278] If the cumulative model training number is less than the maximum model training number, the artificial intelligence model is retrained, and the cumulative model training number is updated;

[0279] If the cumulative model training number is greater than or equal to the maximum model training number, the model training of the artificial intelligence model is stopped, a model training timer is started, and when the timing duration of the model training timer is greater than or equal to the training interval, the artificial intelligence model is retrained.

[0280] In one embodiment, the processor 1403 is further configured to:

[0281] According to a preset model update strategy, the artificial intelligence model is updated.

[0282] In an embodiment, the processor 1403 is further configured to:

[0283] change the model structure of the artificial intelligence model according to a preset model updating strategy;

[0284] change the impulse function or loss function of the artificial intelligence model according to a preset model updating strategy;

[0285] transform the model type of the artificial intelligence model according to a preset model updating strategy.

[0286] In an embodiment, the processor 1403 is further configured to:

[0287] change the training sample data according to a preset training sample data updating strategy.

[0288] In an embodiment, the first network device comprises at least one of a terminal, a base station, a master board card, an intelligent computing board card, a computing power board card, a baseband board, and a radio access network network element.

[0289] In an embodiment, the signaling comprises at least one of user plane signaling, intelligent plane signaling, data plane signaling, computing plane signaling, radio resource control signaling, medium access control-control element, downlink control information, management plane signaling, control plane signaling, system message, non-access stratum signaling, access stratum signaling, and dedicated configuration signaling.

[0290] Figure 15 A structural diagram of an access network device is provided in the embodiments of the present application. The access network device can include a receiver 1501, a memory 1502, a processor 1503, at least one communication bus 1504, and a transmitter 1505. The communication bus 1504 is used to realize the communication connection between the elements. The memory 1502 can contain a high-speed RAM memory and can also include a non-volatile storage NVM, such as at least one disk memory. The memory 1502 can store various programs for completing various processing functions and implementing the method steps of the embodiments. In the embodiments, the transmitter 1505 can be a radio frequency processing module or a baseband processing module in the access network device, and the receiver 1501 can also be a radio frequency processing module or a baseband processing module in the access network device. The transmitter 1505 and the receiver 1501 can be integrated together to realize a transceiver. Both the transmitter 1505 and the receiver 1501 can be coupled to the processor 1503, which can realize the receiving or transmitting action under the indication or control action of the processor 1503.

[0291] In the embodiments, the transmitter 1505 is configured to send signaling to the first network device, and the signaling at least contains model configuration information, which indicates the first network device to perform an artificial intelligence model training and / or data acquisition process.

[0292] The receiver 1501 is configured to receive a model configuration response message sent by the first network device, wherein the model configuration response message is generated based on an execution result of the artificial intelligence model training and / or data collection process.

[0293] In an embodiment, the receiver 1501 is specifically configured to:

[0294] receive a start failure response message sent by the first network device;

[0295] receive a model training start response message sent by the first network device;

[0296] receive a model training completion message sent by the first network device.

[0297] In an embodiment, the transmitter 1505 is further configured to:

[0298] send a model deployment message to the first network device, wherein the model deployment message is used to instruct the first network device to perform a model deployment process on the trained artificial intelligence model.

[0299] In an embodiment, the second network device includes at least one of a base station, a core network element, and a network management operation and maintenance management.

[0300] In an embodiment, a computer readable storage medium is provided, and the computer readable storage medium stores a computer program. The computer program is executed by a processor to implement the steps of the above method.

[0301] The embodiments of the present application also provide a computer program product containing instructions, which, when executed on a computer, cause the computer to perform the steps of the above method.

[0302] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, database or other medium used in each embodiment provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0303] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of each technical feature in the above embodiments are not described, but as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0304] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for those skilled in the art, without departing from the concept of the present application, some modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A model lifecycle management method, characterized by, The method is applied to a first network device, and the method comprises: receiving signaling sent by a second network device; wherein the signaling comprises at least model configuration information; performing an artificial intelligence model training and / or data collection process according to the model configuration information; sending a model configuration response message to the second network device based on the execution result of the artificial intelligence model training and / or data collection process; when the artificial intelligence model training fails, determining the cumulative model training number of the artificial intelligence model; if the cumulative model training number is less than the maximum model training number, retraining the artificial intelligence model and updating the cumulative model training number; if the cumulative model training number is greater than or equal to the maximum model training number, stopping the model training of the artificial intelligence model, starting a model training timer, and retraining the artificial intelligence model when the timing duration of the model training timer is greater than or equal to a training interval; before the retraining of the artificial intelligence model, the method further comprises at least one of the following: changing the model structure of the artificial intelligence model according to a preset model update strategy; transforming the model type of the artificial intelligence model according to a preset model update strategy.

2. The method of claim 1, wherein, The model configuration information comprises at least one of indication information and training configuration information; the indication information is used to indicate whether to perform artificial intelligence model training; the training configuration information comprises at least one of model construction geographic range, storage space information, computing power information, model structure configuration information, training parameter information, and data collection configuration information.

3. The method of claim 2, wherein, The storage space information represents the storage space required for the artificial intelligence model training and / or data collection process; the computing power information represents the computing power required for the artificial intelligence model training and / or data collection process.

4. The method of claim 2, wherein, The model structure configuration information comprises at least one of pre-training indication information, model type, and model structure parameter; the pre-training indication information represents whether it is a pre-trained model.

5. The method of claim 2, wherein, The training parameter information comprises at least one of data set size, loss function type, training period, deployment condition, maximum model training number, and training interval.

6. The method of claim 2, wherein, The data collection configuration information comprises at least one of data set number, data set type, data set function, data collection pilot signal configuration, data collection period, data collection sample type, data collection precision, and data collection quantity.

7. The method according to claim 1 or 2, characterized in that, In the case where the model configuration information comprises data collection configuration information, performing an artificial intelligence model training and / or data collection process according to the model configuration information comprises at least one of the following: performing a data collection process based on the data collection configuration information; and, obtaining training sample data, and training an artificial intelligence model based on the training sample data and the model configuration information; the training sample data comprises at least one of collected sample data and / or pre-stored sample data.

8. The method of claim 7, wherein, The obtaining of training sample data comprises at least one of the following: If the data amount of the pre-stored sample data is less than the preset data amount threshold, the collected sample data is acquired, and when the total data amount of the collected sample data and the pre-stored sample data reaches the data amount threshold, the step of training an artificial intelligence model based on the training sample data and the model configuration information is performed. If the data amount of the pre-stored sample data is greater than or equal to the preset data amount threshold, the pre-stored sample data is used as the training sample data.

9. The method of claim 7, wherein, The acquiring training sample data comprises: After the data acquisition is completed, the collected sample data is used as the training sample data.

10. The method of claim 1, wherein, The sending of the model configuration response message to the second network device based on the execution result of the artificial intelligence model training and / or data acquisition process comprises at least one of the following: If the artificial intelligence model training or the data acquisition process fails to start, a start failure response message is sent to the second network device; If the artificial intelligence model training starts successfully, a model training start response message is sent to the second network device; If the artificial intelligence model training is completed, a model training completion message is sent to the second network device.

11. The method of claim 10, wherein, After the sending of the model training completion message to the second network device, the method further comprises at least one of the following: performing a model deployment process on the trained artificial intelligence model; receiving a model deployment message sent by the second network device and performing a model deployment process on the trained artificial intelligence model.

12. The method of claim 1, wherein, The updating of the artificial intelligence model according to the preset model update strategy comprises: According to the preset model update strategy, the impulse function or the loss function of the artificial intelligence model is changed.

13. The method of claim 1, wherein, Before the re-training of the artificial intelligence model, the method further comprises: According to the preset training sample data update strategy, the training sample data is changed.

14. The method of claim 1, wherein, The first network device comprises at least one of a terminal, a base station, a main control board, an intelligent computing board, a computing power board, a baseband board and a radio access network element.

15. The method of claim 1, wherein, The signaling comprises at least one of user plane signaling, intelligent plane signaling, data plane signaling, computing plane signaling, radio resource control signaling, medium access control control element, downlink control information, management plane signaling, control plane signaling, system message, non-access layer signaling, access layer signaling and dedicated configuration signaling.

16. A model lifecycle management method, characterized by, The method is applied to a second network device, and the method comprises: sending signaling to a first network device; wherein the signaling at least contains model configuration information, and the model configuration information instructs the first network device to perform an artificial intelligence model training and / or data acquisition process; receiving a model configuration response message sent by the first network device; the model configuration response message is generated based on the execution result of the artificial intelligence model training and / or data acquisition process; The first network device determines a cumulative model training number of the artificial intelligence model when the artificial intelligence model training fails, and if the cumulative model training number is less than a maximum model training number, re-trains the artificial intelligence model and updates the cumulative model training number. If the cumulative model training number is greater than or equal to the maximum model training number, stops the model training of the artificial intelligence model, starts a model training timer, and when the timing duration of the model training timer is greater than or equal to a training interval, re-trains the artificial intelligence model. Before the re-training of the artificial intelligence model, the method further includes at least one of the following: According to a preset model update strategy, changing a model structure of the artificial intelligence model; According to a preset model update strategy, changing a model type of the artificial intelligence model.

17. The method of claim 16, wherein, The receiving of the model configuration response message sent by the first network device includes at least one of the following: Receiving a start failure response message sent by the first network device; Receiving a model training start response message sent by the first network device; Receiving a model training completion message sent by the first network device.

18. The method of claim 17, wherein, After the receiving of the model training completion message sent by the first network device, the method further includes: Sending a model deployment message to the first network device; the model deployment message is used to instruct the first network device to perform a model deployment process on the trained artificial intelligence model.

19. The method of claim 16, wherein, The second network device includes at least one of a base station, a core network element, and a network management operation and maintenance management.

20. An apparatus for model lifecycle management, the apparatus comprising: The device is applied to a first network device, and the device includes: A first receiving module configured to receive signaling sent by a second network device; wherein the signaling at least contains model configuration information; A first execution module configured to perform an artificial intelligence model training and / or data collection process according to the model configuration information; A first sending module configured to send a model configuration response message to the second network device based on an execution result of the artificial intelligence model training and / or data collection process; A determination module configured to determine a cumulative model training number of the artificial intelligence model when the artificial intelligence model training fails; The device further includes: A first re-training module configured to re-train the artificial intelligence model and update the cumulative model training number if the cumulative model training number is less than a maximum model training number; A second re-training module configured to stop the model training of the artificial intelligence model, start a model training timer, and re-train the artificial intelligence model when the timing duration of the model training timer is greater than or equal to a training interval if the cumulative model training number is greater than or equal to the maximum model training number; Before the re-training of the artificial intelligence model, the device further includes a first update module configured to at least one of the following: According to a preset model update strategy, changing a model structure of the artificial intelligence model; According to a preset model update strategy, changing a model type of the artificial intelligence model.

21. An apparatus for model lifecycle management, the apparatus comprising: The device is applied to a second network device, and the device comprises: a second sending module configured to send signaling to a first network device, wherein the signaling comprises at least model configuration information, and the model configuration information instructs the first network device to perform an artificial intelligence model training and / or data collection process; a second receiving module configured to receive a model configuration response message sent by the first network device, wherein the model configuration response message is generated based on an execution result of the artificial intelligence model training and / or data collection process; when the artificial intelligence model training fails, the first network device determines a cumulative model training number of the artificial intelligence model, and if the cumulative model training number is less than a maximum model training number, the first network device re-trains the artificial intelligence model and updates the cumulative model training number; if the cumulative model training number is greater than or equal to the maximum model training number, the first network device stops the model training of the artificial intelligence model, starts a model training timer, and re-trains the artificial intelligence model when a timing duration of the model training timer is greater than or equal to a training interval; before the re-training of the artificial intelligence model, the first network device is further configured to perform at least one of the following: change a model structure of the artificial intelligence model according to a preset model update strategy; and transform a model type of the artificial intelligence model according to the preset model update strategy.

22. A communications device, comprising: comprise: a receiver, a processor and a transmitter; the receiver is configured to receive signaling sent by a second network device, wherein the signaling comprises at least model configuration information; the processor is configured to perform an artificial intelligence model training and / or data collection process according to the model configuration information; the transmitter is configured to send a model configuration response message to the second network device based on an execution result of the artificial intelligence model training and / or data collection process; the processor is further configured to determine a cumulative model training number of the artificial intelligence model when the artificial intelligence model training fails; the processor is further configured to re-train the artificial intelligence model and update the cumulative model training number if the cumulative model training number is less than a maximum model training number; if the cumulative model training number is greater than or equal to the maximum model training number, the first network device stops the model training of the artificial intelligence model, starts a model training timer, and re-trains the artificial intelligence model when a timing duration of the model training timer is greater than or equal to a training interval; before the re-training of the artificial intelligence model, the processor is further configured to perform at least one of the following: change a model structure of the artificial intelligence model according to a preset model update strategy; and transform a model type of the artificial intelligence model according to the preset model update strategy.

23. A communications device, characterized by comprise: a transmitter and a receiver; the transmitter is configured to send signaling to a first network device, wherein the signaling comprises at least model configuration information, and the model configuration information instructs the first network device to perform an artificial intelligence model training and / or data collection process; The receiver is configured to receive a model configuration response message sent by the first network device, wherein the model configuration response message is generated based on an execution result of the artificial intelligence model training and / or data collection process; When the artificial intelligence model training fails, the first network device determines a cumulative model training number of the artificial intelligence model, and if the cumulative model training number is less than a maximum model training number, the first network device re-trains the artificial intelligence model and updates the cumulative model training number; If the cumulative model training number is greater than or equal to the maximum model training number, the first network device stops the model training of the artificial intelligence model, starts a model training timer, and re-trains the artificial intelligence model when a timing duration of the model training timer is greater than or equal to a training interval; Before the re-training of the artificial intelligence model, the first network device is further configured to perform at least one of the following: According to a preset model update strategy, change a model structure of the artificial intelligence model; According to a preset model update strategy, change a model type of the artificial intelligence model.

24. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, implements the steps of the method of any one of claims 1 to 19.

25. A chip, characterized by The chip includes a programmable logic circuit and / or program instructions, and when the chip is running, the steps of the method of any one of claims 1 to 19 are implemented.

26. A computer program product comprising a computer program, characterised in that, The computer program, when executed by a processor, implements the steps of the method of any one of claims 1 to 19.

Images