Optimization Method, Device, Communication Equipment and Storage Medium for Radio Access Network

Abstract

The present application relates to an optimization method, device, communication equipment and storage medium for a radio access network, and relates to the field of wireless communication technologies. The method includes: receiving network performance data sent by a radio access network node; sending the network performance data to a central cloud; receiving a network optimization task sent by the central cloud; the network optimization task is obtained based on the network performance data and a first model deployed in the central cloud; decomposing the network optimization task to obtain at least one corresponding execution instruction; the execution instruction carries a node identifier and a model identifier of a second model, the second model is deployed on the radio access network node and the model size is smaller than the size of the first model; sending the execution instruction to an optimization target node corresponding to the node identifier, triggering the optimization target node to call the second model corresponding to the model identifier to execute the execution instruction, so as to optimize the network performance of the optimization target node. Using this method can improve the optimization efficiency in the radio access network.

Description

Technical Field

[0001] The present application relates to the field of wireless communication technologies, and in particular, to an optimization method, apparatus, communication device, storage medium, and computer program product for a radio access network. Background Art

[0002] With the development of artificial intelligence technologies, large models have become an important means for network intelligent optimization. Through large models, network optimization can be performed on each radio access network (RAN) in the entire network. Since large models have high requirements for computing resources, traditional computing architectures cannot support the efficient operation of large models.

[0003] In related technologies, the execution of large model tasks can be supported through a central cloud computing mode and an edge computing mode. In the central cloud computing mode, large models related to network optimization can be deployed in a cloud high-performance computing cluster, and global analysis and decision-making can be performed through powerful computing capabilities; in the edge computing mode, computing resources can be deployed at network edge nodes, and corresponding edge nodes can be used to process some network optimization tasks, thereby sharing the computing pressure of the central cloud.

[0004] However, the central cloud computing mode has a process of uploading data from the base station to the cloud and then feedback to the base station, which has high latency and cannot meet the real-time response requirements. In the edge computing mode, due to the limited computing resources and uneven resource distribution of edge nodes, the computing power of the edge nodes does not support large models to process complex tasks. Therefore, due to the high latency characteristics of the central cloud and the insufficient computing power of edge nodes in related technologies, there is a problem of insufficient optimization efficiency for the application scenario of optimizing each RAN in the entire network through large models. Summary of the Invention

[0005] Based on this, it is necessary to provide an optimization method, apparatus, communication device, computer-readable storage medium, and computer program product for a radio access network that can improve the optimization efficiency in view of the above technical problems.

[0006] In a first aspect, the present application provides an optimization method for a radio access network, which is applied to the AI layer of the radio access network. The method includes:

[0007] Receiving network performance data sent by a radio access network node;

[0008] Sending the network performance data to the central cloud;

[0009] Receiving a network optimization task sent by the central cloud; the network optimization task is obtained based on the network performance data and a first model deployed on the central cloud;

[0010] Decompose the network optimization task to obtain at least one corresponding execution instruction; the execution instruction carries a node identifier and a model identifier of a second model, and the second model is deployed on a radio access network node and has a model size smaller than that of the first model;

[0011] Send the execution instruction to the node to be optimized corresponding to the node identifier, and trigger the node to be optimized to call the second model corresponding to the model identifier to execute the execution instruction, so as to optimize the network performance of the node to be optimized; the node to be optimized includes at least the radio access network node.

[0012] In one embodiment, the sending the network performance data to the central cloud includes:

[0013] Send the network performance data to the central cloud when a preset reporting condition is met;

[0014] The preset reporting condition includes at least one of a preset reporting period, a preset data volume, and receiving a preset user feedback message.

[0015] In one embodiment, the network optimization task includes a node identifier and a parameter to be optimized of the node to be optimized;

[0016] The decomposing the network optimization task to obtain at least one corresponding execution instruction includes:

[0017] Based on the node identifier and the parameter to be optimized included in the network optimization task, determine at least one second model for the node to be optimized corresponding to the node identifier, and generate a model call instruction and a parameter configuration instruction;

[0018] The model call instruction carries the model identifier of the determined second model;

[0019] The model call instruction is used to call the determined second model, and predict a predicted adjustment value of each parameter to be optimized through the second model; the parameter configuration instruction is used to adjust the parameter to be optimized to the predicted adjustment value.

[0020] In one embodiment, after sending the execution instruction to the node to be optimized corresponding to the node identifier, the method further includes:

[0021] After the node to be optimized executes the execution instruction, receive new network performance data sent by the radio access network node, and send the new network performance data to the central cloud;

[0022] In the case where the verification result of the central cloud for the new network performance data is a verification failure, receive the regenerated network optimization task sent by the central cloud, and return to execute the steps of decomposing the network optimization task until the verification result is a verification success.

[0023] In a second aspect, the present application provides an optimization method for a radio access network, which is applied to a central cloud. The method includes:

[0024] Receive the network performance data of the radio access network nodes sent by the AI layer of the radio access network;

[0025] Based on the network performance data and a first model deployed in the central cloud, determine a network optimization task;

[0026] Send the network optimization task to the AI layer of the radio access network to trigger the AI layer of the radio access network to decompose the network optimization task, obtain at least one corresponding execution instruction carrying the node identifier and the model identifier of the second model, and send the execution instruction to the to-be-optimized node corresponding to the node identifier, triggering the to-be-optimized node to call the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the to-be-optimized node;

[0027] Wherein, the second model is deployed on the radio access network nodes and the model size is smaller than the size of the first model; the to-be-optimized nodes at least include the radio access network nodes.

[0028] In one embodiment, the determining the network optimization task based on the network performance data and the first model deployed in the central cloud includes:

[0029] Obtain the historical performance data of the radio access network nodes and the existing configuration parameters of the radio access network nodes;

[0030] Input the historical performance data, the existing configuration parameters, and the network performance data into the first model to obtain a network optimization task for the network performance data; the network optimization task includes the node identifier of the to-be-optimized node and the to-be-optimized parameters.

[0031] In one embodiment, the method further includes:

[0032] After the to-be-optimized node executes the execution instruction, receive the new network performance data of the radio access network nodes sent by the AI layer of the radio access network;

[0033] Determine the performance change of the new network performance data;

[0034] If the performance change meets the expected goal, determine that the verification result of the new network performance data is successful verification;

[0035] If the performance change does not meet the expected goal, determine that the verification result of the new network performance data is failed verification.

[0036] In one embodiment, the method further includes: if the verification result of the new network performance data is failed verification, return to execute the step of determining a network optimization task based on the network performance data and a first model deployed in the central cloud, and sending the network optimization task to the AI layer of the radio access network until the verification result is successful verification.

[0037] In a third aspect, the present application provides an optimization method for a radio access network, which is applied to a radio access network node. The method includes:

[0038] Send network performance data to the AI layer of the radio access network, so that the AI layer of the radio access network sends the network performance data to the central cloud;

[0039] Receive at least one execution instruction returned by the AI layer of the radio access network; the execution instruction is obtained by the AI layer of the radio access network decomposing a network optimization task sent by the central cloud; the network optimization task is obtained based on the network performance data and a first model deployed in the central cloud; the execution instruction carries a node identifier and a model identifier of a second model, and the second model is deployed on the radio access network node and the model size is smaller than the size of the first model;

[0040] Call the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the radio access network node.

[0041] In one embodiment, the sending the network performance data to the AI layer of the radio access network includes:

[0042] Obtain network performance data in real time;

[0043] If it is determined based on the network performance data that a network monitoring index is abnormal, send the network performance data to the AI layer of the radio access network; or,

[0044] If a preset user feedback message is received, send the network performance data and the preset user feedback message to the AI layer of the radio access network.

[0045] In one embodiment, the execution instruction includes a model call instruction and a parameter configuration instruction;

[0046] Receiving at least one execution instruction returned by the wireless access network AI layer and invoking a second model corresponding to the model identifier to execute the execution instruction includes:

[0047] Receiving at least one model invocation instruction and at least one parameter configuration instruction returned by the wireless access network AI layer;

[0048] Based on the model identifier carried in the model invocation instruction, invoking a second model corresponding to the model identifier to predict a predicted adjustment value for the parameter to be optimized;

[0049] Adjusting the parameter to be optimized to the predicted adjustment value based on the parameter configuration instruction.

[0050] In one embodiment, after invoking the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the wireless access network node, the method further includes:

[0051] After executing the execution instruction, re-acquiring the current network performance data;

[0052] Sending the current network performance data to the wireless access network AI layer to send the network performance data to the central cloud through the wireless access network AI layer for verification.

[0053] In a fourth aspect, the present application further provides an optimization device for a wireless access network. The device includes:

[0054] A first receiving module, configured to receive network performance data sent by a wireless access network node;

[0055] A first sending module, configured to send the network performance data to the central cloud;

[0056] A second receiving module, configured to receive a network optimization task sent by the central cloud; the network optimization task is obtained based on the network performance data and a first model deployed on the central cloud;

[0057] A decomposition module, configured to decompose the network optimization task to obtain at least one corresponding execution instruction; the execution instruction carries a node identifier and a model identifier of a second model, and the second model is deployed on the wireless access network node and the model size is smaller than the size of the first model;

[0058] A second sending module, configured to send the execution instruction to a node to be optimized corresponding to the node identifier, triggering the node to be optimized to invoke the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the node to be optimized; the node to be optimized includes at least the wireless access network node.

[0059] Fifth aspect, the present application further provides an optimization device for a radio access network. The device includes:

[0060] a receiving module, configured to receive network performance data of radio access network nodes sent by the AI layer of the radio access network;

[0061] a task determination module, configured to determine a network optimization task based on the network performance data and a first model deployed in the central cloud;

[0062] a sending module, configured to send the network optimization task to the AI layer of the radio access network, so as to trigger the AI layer of the radio access network to decompose the network optimization task, obtain at least one corresponding execution instruction carrying a node identifier and a model identifier of a second model, and send the execution instruction to a to-be-optimized node corresponding to the node identifier, trigger the to-be-optimized node to call the second model corresponding to the model identifier to execute the execution instruction, so as to optimize the network performance of the to-be-optimized node; wherein, the second model is deployed in a radio access network node and the model size is smaller than the size of the first model; the to-be-optimized node at least includes the radio access network node.

[0063] Sixth aspect, the present application further provides an optimization device for a radio access network. The device includes:

[0064] a sending module, configured to send network performance data to the AI layer of the radio access network, so that the AI layer of the radio access network sends the network performance data to the central cloud;

[0065] a receiving module, configured to receive at least one execution instruction returned by the AI layer of the radio access network; the execution instruction is obtained by the AI layer of the radio access network decomposing a network optimization task sent by the central cloud; the network optimization task is obtained based on the network performance data and a first model deployed in the central cloud; the execution instruction carries a node identifier and a model identifier of a second model, the second model is deployed in a radio access network node and the model size is smaller than the size of the first model;

[0066] an instruction execution module, configured to call the second model corresponding to the model identifier to execute the execution instruction, so as to optimize the network performance of the radio access network node.

[0067] Seventh aspect, the present application further provides a communication device. The communication device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps of the method described in the first aspect to the third aspect are implemented.

[0068] In an eighth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, and when the computer program is executed by a processor, the steps of the method described in the first aspect to the third aspect are implemented.

[0069] In a ninth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the method described in the first aspect to the third aspect are implemented.

[0070] For the above-mentioned method, device, communication device, storage medium, and computer program product for optimizing a radio access network, the AI layer of the radio access network can receive network performance data sent by radio access network nodes, and send the network performance data to the central cloud, and receive a network optimization task sent by the central cloud based on the network performance data and a first model deployed in the central cloud; decompose the network optimization task to obtain at least one execution instruction corresponding to the node identifier and the model identifier of the second model, and the second model is deployed in the radio access network node and the model size is smaller than that of the first model; send the execution instruction to the to-be-optimized node corresponding to the node identifier, and trigger the to-be-optimized node to call the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the to-be-optimized node; and the to-be-optimized node at least includes the radio access network node. Through the above method, the radio access network node can send network performance data to the central cloud through the AI layer, generate a network optimization task through a large model deployed in the central cloud, decompose the network optimization task in the AI layer of the radio access network, and send the decomposed execution instruction to the to-be-optimized node, so that the to-be-optimized node can call a locally deployed small model to execute the execution instruction, which can reduce the network resource consumption between the central cloud and the radio access network node, improve the real-time performance of the central cloud, and share the optimization task of the large model deployed in the central cloud through the small model deployed in the radio access network node, improving the flexibility of resource allocation and resource utilization rate, thereby improving the optimization efficiency of the network performance of the radio access network node. Description of the Drawings

[0071] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0072] Figure 1 It is an application environment diagram of the method for optimizing a radio access network in an embodiment;

[0073] Figure 2Schematic flowchart of an optimization method for a radio access network in an embodiment;

[0074] Figure 3 Schematic flowchart of an optimization method for a radio access network in another embodiment;

[0075] Figure 4 Schematic flowchart of an optimization method for a radio access network in yet another embodiment;

[0076] Figure 5 Schematic flowchart of an optimization method for a radio access network in still another embodiment;

[0077] Figure 6 Schematic diagram of wireless intelligent management and orchestration in an embodiment;

[0078] Figure 7 Structural block diagram of an optimization device for a radio access network in an embodiment;

[0079] Figure 8 Structural block diagram of an optimization device for a radio access network in another embodiment;

[0080] Figure 9 Structural block diagram of an optimization device for a radio access network in yet another embodiment;

[0081] Figure 10 Internal structure diagram of a communication device in an embodiment. Detailed implementation manners

[0082] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0083] The optimization method for a radio access network provided by the embodiments of the present application can be applied to an application environment as Figure 1 shown. Among them, the radio access network AI layer 102 can communicate with the central cloud 106. The radio access network AI layer 102 manages multiple affiliated radio access networks, and the radio access network AI layer 102 can communicate with and monitor radio access network nodes 104 in the radio access network. The user terminal 108 can establish a communication connection with each radio access network node 104 in the radio access network, and determine the network performance data of the radio access network node 104 based on the network state of the user terminal 108 in the radio access network node 104.

[0084] In one example, the user terminal 108 may send user feedback information to the radio access network to feedback abnormal network status. The radio access network node 104 may monitor its own network performance data. The radio access network node 104 may send the network performance data and the user feedback information to the radio access network AI layer 102, and the radio access network AI layer 102 may uniformly send the above data information to the central cloud 106. After determining the above data information, the central cloud 106 may determine a network optimization task for the radio access network node 104 based on the first model deployed in the central cloud 106. The central cloud 106 returns the network optimization task to the radio access network AI layer 102. The radio access network AI layer 102 may decompose the network optimization task to obtain at least one execution instruction carrying the node identifier of the node to be optimized and the model identifier of the second model, and send the above execution instruction to the radio access network node 104 corresponding to the node identifier. The node to be optimized may call the second model through the model identifier, so as to execute the execution instruction to optimize the network performance of the node to be optimized. For example, the node to be optimized at least includes the radio access network node that sends the network performance data.

[0085] In one example, each radio access network node 104 may send the network performance data monitored in real time to the radio access network AI layer 102, and the radio access network AI layer 102 may periodically summarize and send the network performance data of each node to the central cloud 106. The central cloud 106 may analyze the network performance data through the first model deployed in the central cloud 106 to determine whether to generate a network optimization task, and in the case of generating a network optimization task, determine the network optimization task of the radio access network node 104 corresponding to the network performance data. The central cloud 106 sends each network optimization task to the radio access network AI layer 102. The radio access network AI layer 102 may decompose the network optimization task to obtain multiple execution instructions. The execution instruction may carry the node identifier corresponding to the node to be optimized for the radio access network node, and the model identifier of the second model to be called. The radio access network AI layer may send the corresponding execution instruction to the corresponding node to be optimized according to the node identifier, so that the node to be optimized calls its own computing power resources to call the second model and optimize the network performance of the node.

[0086] Among them, the radio access network node 104 in the radio access network can be a base station, an antenna, a radio frequency unit, a baseband unit, a transmission device, etc. (also referred to as a radio access network node), and a CPU and a GPU device are deployed in each node. The radio access network AI layer 102 can be an edge node in the radio access network, or a server or a server cluster that communicates with the radio access network independently. The central cloud 106 can be implemented by an independent cloud server or a server cluster composed of multiple cloud servers. The central cloud 106 can be deployed in the radio access network or in a cloud server that is communicatively connected to the radio access network, that is, the central cloud can be deployed inside or outside the radio access network, and no specific limitation is made in the embodiments of the present application. The user terminal 108 can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc.

[0087] In an exemplary embodiment, as Figure 2 shown, an optimization method for a radio access network is provided. Taking the method applied to Figure 1 the radio access network AI layer 102 in

[0088] as an example, the following steps S202 to step S210 are included. Among them:

[0089] The radio access network RAN (Radio Access Network) is a key component in a communication system, responsible for connecting user equipment (such as mobile phones, tablet computers, Internet of Things terminals, etc.) to the core network through wireless signals, so as to realize communication with the Internet or other communication networks. The radio access network node is the core device unit that constitutes the radio access network, responsible for realizing the wireless communication connection between the user equipment and the core network. The node can be composed of a base station, a distributed base station, a relay node, and an access point. The network performance data is a key indicator for measuring the network operation status and user experience. For example, the network performance data can include, but is not limited to, traffic load, latency, packet loss rate, etc. The radio access network AI layer is the logical control plane for radio access network nodes with AI computing power resources. The radio access network AI layer can be deployed in the data center to which the radio access network belongs or in edge devices in the radio access network. The radio access network AI layer is used to control each radio access network node.

[0090] Specifically, the AI layer of the radio access network can receive network performance data sent by multiple radio access network nodes in the radio access network managed by the AI layer of the radio access network. In one example, each radio access network node can monitor its own network performance data in real time and send the network performance data to the AI layer of the radio access network in real time or at regular intervals. In another example, the radio access network node can also receive user feedback information sent by the user terminal. If the user feedback information is preset user feedback information, the radio access network node can obtain the current network performance data and send the network performance data and the user feedback information to the AI layer of the radio access network.

[0091] Step S204: Send the network performance data to the central cloud.

[0092] Among them, the central cloud is the core infrastructure in the cloud computing architecture and can be a large-scale cloud data center deployed centrally. The central cloud is used to process, store, and manage large-scale data and computing resources. In one example, the central cloud is composed of a high-performance computing cluster, and a global AI large model is deployed in the central cloud. Through the AI large model, functions such as global problem analysis and prediction, solution generation, whole-network status monitoring, and historical data management can be performed. In one example, the first model can be the AI large model deployed in the central cloud.

[0093] Specifically, the AI layer of the radio access network can send the network performance data of each radio access network node to the central cloud for the central cloud to monitor and perform subsequent processing. In one example, the AI layer of the radio access network can transmit the network performance data to the central cloud at regular intervals according to the pre-configured time. For example, the AI layer of the radio access network can collect the network performance data sent by each radio access network node and send the above network performance data to the central cloud at a fixed time every month; or send the above network performance data when the amount of collected data reaches the set data volume. In one example, the AI layer of the radio access network can receive the user feedback information corresponding to each radio access network node. When the user feedback information reaches a certain number or receives the preset user feedback information, the AI layer of the radio access network can send the user feedback information and the network performance data of the corresponding radio access network node to the central cloud together. In this embodiment, the network performance data can also be sent to the central cloud even if the fixed time has not been reached to avoid the accumulation of user feedback information and the excessive waiting time of users.

[0094] Step S206: Receive the network optimization task sent by the central cloud.

[0095] Among them, the network optimization task is obtained based on network performance data and a first model deployed in the central cloud. The network optimization task is used to improve the network performance of radio access network nodes by adjusting the network parameters and network configurations of the radio access network nodes. The network optimization task can determine the radio access network nodes to be adjusted and their corresponding configuration parameters. The first model can be a large AI model deployed in the central cloud. The large AI model can handle network optimization problems, resource allocation problems, etc. for the global radio access network, generate corresponding solutions for the above problems, and obtain network optimization tasks or resource scheduling tasks for each radio access network node based on the solutions. In addition, the large AI model can also monitor the status of the global radio access network, and store and manage the historical data of radio access network nodes.

[0096] Specifically, the central cloud can determine whether the radio access network nodes corresponding to the network performance data need network optimization or resource scheduling based on the first model deployed in the central cloud and the network performance data sent by the radio access network. When it is determined that the radio access network nodes corresponding to the network performance data need network optimization, a network optimization task for the nodes to be optimized can be generated through the first model of the central cloud. In one example, the network optimization task can carry information about the nodes to be optimized, the strategies for optimizing the nodes, the indicators that the optimized network performance should reach, etc. The central cloud can send the generated network optimization task to the AI layer of the radio access network.

[0097] Step S208, decompose the network optimization task to obtain at least one corresponding execution instruction.

[0098] Among them, the execution instruction is an instruction for being received and executed by the radio access network nodes. The execution instruction is used to optimize the configuration parameters in the radio access network nodes. In one example, the execution instruction can optimize the configuration parameters related to each user terminal in the radio access network nodes; the execution instruction carries a node identifier and a model identifier of a second model; the node identifier is used to locate the node to be optimized, and the model identifier is used to locate the second model. The second model is an AI model deployed in any radio access network node and the model size (or complexity) is smaller than the size (or complexity) of the first model. The second model can perform prediction and analysis on configuration parameters in specific dimensions. In one example, the second model can predict the configuration parameters to be optimized. For example, the second model can be a load balancing model and can predict the network resources that should be allocated to each user terminal in the radio access network node.

[0099] Specifically, after receiving the network optimization task sent by the central cloud, the AI layer of the radio access network can decompose the network optimization task to determine the nodes to be optimized, the parameters to be optimized in the nodes to be optimized, and the second model for predicting the parameters to be optimized required to complete the network optimization task. Based on this, the AI layer of the radio access network can generate at least one execution instruction for the steps to be executed in the network optimization task, and store the node identifier of the node to be optimized and the model identifier of the second model in each execution instruction. In one example, the execution instruction can be used to determine and call at least one second model required for the network optimization task, and the execution instruction can be used to adjust the parameters to be optimized.

[0100] Step S210: Send the execution instruction to the node to be optimized corresponding to the node identifier, and trigger the node to be optimized to call the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the node to be optimized.

[0101] Among them, the nodes to be optimized include at least radio access network nodes. Since there are many factors affecting each radio access network node, a radio access network node can be affected by multiple other radio access network nodes. Therefore, the nodes to be optimized can be at least one radio access network node, and the nodes to be optimized include at least the radio access network nodes that send network performance data.

[0102] Specifically, the AI layer of the radio access network can send the execution instructions to the nodes to be optimized corresponding to each node identifier according to the node identifiers in each execution instruction. In one example, the AI layer of the radio access network can package multiple execution instructions to obtain a set of execution instructions. A set of execution instructions can contain multiple execution instructions with an execution sequence. A set of execution instructions can optimize a node to be optimized, and the AI layer of the radio access network can send a set of execution instructions to the node to be optimized corresponding to the node identifier.

[0103] In the above optimization method for a radio access network, the AI layer of the radio access network can receive network performance data sent by radio access network nodes, and send the network performance data to the central cloud, and receive a network optimization task obtained by the central cloud based on the network performance data and a first model deployed in the central cloud; decompose the network optimization task to obtain at least one execution instruction carrying a node identifier and a model identifier of a second model, where the second model is deployed on the radio access network node and the model size is smaller than that of the first model; send the execution instruction to the to-be-optimized node corresponding to the node identifier, and trigger the to-be-optimized node to call the second model corresponding to the model identifier to execute the execution instruction, so as to optimize the network performance of the to-be-optimized node; and the to-be-optimized node at least includes a radio access network node. Through the above method, the radio access network node sends network performance data to the central cloud through the AI layer, can generate a network optimization task through a large model deployed in the central cloud, decompose the network optimization task in the radio access network AI layer, and send the decomposed execution instruction to the to-be-optimized node, so that the to-be-optimized node can call a small model deployed locally to execute the execution instruction, which can reduce the network resource consumption between the central cloud and the radio access network node, improve the real-time performance of the central cloud, and share the optimization task of the large model deployed in the central cloud through the small model deployed on the radio access network node, improving the flexibility of resource allocation and resource utilization rate, thereby improving the optimization efficiency of the network performance of the radio access network node.

[0104] In an exemplary embodiment, the specific implementation process of the step "sending the network performance data to the central cloud" includes:

[0105] When a preset reporting condition is satisfied, send the network performance data to the central cloud. Among them, the preset reporting condition includes at least one of a preset reporting period, a preset data volume, and receiving a preset user feedback message.

[0106] In an example, if the interval between the current time and the time of the last upload of network performance data satisfies the preset reporting period, it is determined that the preset reporting condition corresponding to the preset reporting period is satisfied; if the data volume of the network performance data stored in the radio access network AI layer is greater than the preset data volume threshold, it is determined that the preset reporting condition corresponding to the preset data volume is satisfied; if the user feedback message received by the radio access network AI layer is a preset user feedback message, it is determined that the preset reporting condition corresponding to receiving the preset user feedback message is satisfied.

[0107] For example, the preset reporting period of the AI layer of the radio access network is the 1st of each month. If the current time is March 1st, it is determined that the preset reporting condition corresponding to the preset reporting period is met. The AI layer of the radio access network can send the network performance data of the radio access network nodes to the central cloud on March 1st. For example, if the data volume of the network performance data stored in the AI layer of the radio access network is greater than the preset data volume threshold, such as the data volume of the network performance data is greater than 1 GB, it is determined that the preset reporting condition corresponding to the preset data volume is met. The AI layer of the radio access network can send the network performance data of the radio access network nodes to the central cloud. For example, if the AI layer of the radio access network receives at least one piece of user feedback information for the radio access network nodes and the user feedback information is a user complaint, it is determined that the user feedback information is the preset user feedback information, and it is determined that the preset reporting condition corresponding to receiving the preset user feedback information is met. The AI layer of the radio access network can send the user feedback information and the network performance data to the central cloud.

[0108] In this embodiment, through multiple configured preset reporting conditions, it is possible to control that in the case corresponding to the preset reporting conditions, by uploading the network performance data from the AI layer of the radio access network to the central cloud, the rationality of uploading the network performance data is ensured, and the network resource consumption of the central cloud is reduced.

[0109] In an exemplary embodiment, the network optimization task includes the node identifier and the parameter to be optimized of the node to be optimized. The specific implementation process of the step "decompose the network optimization task to obtain at least one corresponding execution instruction" includes:

[0110] Based on the node identifier and the parameter to be optimized included in the network optimization task, at least one second model is determined for the node to be optimized corresponding to the node identifier, and a model call instruction and a parameter configuration instruction are generated.

[0111] Among them, the network optimization task may carry the node identifier and the parameter to be optimized of the node to be optimized. It can be understood that the network optimization task may also carry other task-related information, such as the global optimization goal or the global optimization scheme, etc. The model call instruction carries the model identifier of the determined second model; the model call instruction is used to call the determined second model and predict the predicted adjustment value of each parameter to be optimized through the second model; the parameter configuration instruction is used to adjust the parameter to be optimized to the predicted adjustment value. The predicted adjustment value is the prediction result obtained by the node to be optimized by calling the second model deployed in the radio access network node and predicting through the second model.

[0112] In one example, the AI layer of the radio access network can decompose network optimization tasks, generate a model call instruction corresponding to a second model for predicting an optimization parameter to be optimized based on the node identifier and the parameter to be optimized carried in the network optimization task, and generate a parameter configuration instruction for the parameter to be optimized. For example, the AI layer of the radio access network can parse each field in the network optimization task to determine the node identifier of the node to be optimized and the parameter to be optimized. The AI layer of the radio access network can determine at least one second model required for the node to be optimized configured in the network optimization task. Based on the required at least one second model, the AI layer of the radio access network determines the model identifier defined in the AI layer of the radio access network for the second model. The AI layer of the radio access network can integrate the node identifier and the model identifier to obtain a model call instruction. After determining the model call instruction, the AI layer of the radio access network can generate a parameter configuration instruction corresponding to the parameter to be optimized based on the model call instruction and the node to be optimized.

[0113] In this embodiment, by determining at least one second model for the node to be optimized corresponding to the node identifier based on the node identifier and the parameter to be optimized included in the network optimization task, and generating a model call instruction and a parameter configuration instruction, the complex network optimization task can be decomposed into multiple executable instructions. The second model to be called by the node to be optimized is determined through the model call instruction in the execution instruction, and the parameter to be optimized is adjusted through the parameter configuration instruction, and the parameter to be optimized is adjusted to the predicted adjustment value predicted by the second model. It is possible to perform the specific optimization process without the participation of the first model, and the AI layer of the radio access network takes over the actual optimization process, and directly controls the radio access network node with the decomposed model call instruction and parameter configuration instruction, so as to use the computing power resource of the radio access network node to call the second model, cooperate the first model and the second model, optimize the computing power resources of the central cloud and the radio access network point, improve the rationality of resource allocation, and improve the efficiency of network optimization.

[0114] In an exemplary embodiment, after the step of "sending the execution instruction to the node to be optimized corresponding to the node identifier", the optimization method of the radio access network further includes:

[0115] After the node to be optimized executes the execution instruction, receive the new network performance data sent by the radio access network node, and send the new network performance data to the central cloud; in the case where the verification result of the central cloud for the new network performance data fails, receive the regenerated network optimization task sent by the central cloud, and return to the step of decomposing the network optimization task until the verification result is successful.

[0116] Among them, the new network performance data refers to the network performance data monitored by the radio access network node after a period of time after the node to be optimized executes the execution instruction and completes the network performance optimization. The verification result is the verification result of whether the new network performance data meets the pre-estimated optimization effect by the central cloud. The verification result can be verification success or verification failure. Verification success means that the new network performance data has met the pre-estimated optimization effect. Verification failure means that the new network performance data has not reached the pre-estimated optimization effect.

[0117] For example, the new network performance data can be latency. The new latency value can be 50 ms, while the pre-estimated latency value is 100 ms. Since the first latency value is less than the pre-estimated latency value, it is determined that the new network performance data has reached the pre-estimated optimization effect. The pre-estimated optimization effect can be the improvement rate. For example, the latency value before optimization is 100 ms, and the latency value after optimization is 50 ms. It is determined that the improvement rate is 100%. Since the threshold of the improvement rate in the pre-estimated optimization effect is 50%, it is determined that the improvement rate is greater than the improvement rate threshold, and the verification result is determined to be verification success.

[0118] Specifically, after the node to be optimized executes the execution instruction, the AI layer of the radio access network can receive the new network performance data sent by the radio access network node and send the new network performance data to the central cloud. If the verification result of the central cloud for the new network performance data is verification failure, receive the regenerated network optimization task returned by the central cloud. The AI layer of the radio access network can decompose the network optimization task to generate at least one corresponding execution instruction. The execution instruction carries the node identifier and the model identifier of the second model, and sends the execution instruction to the node to be optimized corresponding to the node identifier. The node to be optimized can perform optimization again according to the execution instruction, and after the node to be optimized executes the execution instruction, receive the new network performance data. Repeat the above process until the verification result of the central cloud for the new network performance data is verification success.

[0119] In this embodiment, by receiving the new network performance data sent by the radio access network node, sending the new network performance data to the central cloud; in the case where the verification result of the central cloud for the new network performance data is verification failure, receiving the regenerated network optimization task sent by the central cloud and iteratively performing the above decomposition process, so as to perform network optimization again, which can improve the network stability of the radio access network node after optimizing the network performance.

[0120] In an exemplary embodiment, as Figure 3 shown, an optimization method for a radio access network is provided. Taking the method applied to Figure 1 the central cloud 106 therein as an example, it includes the following steps S302 to S306. Among them:

[0121] Step S302: Receive the network performance data of the radio access network nodes sent by the radio access network AI layer.

[0122] Specifically, the central cloud can receive the network performance data of each radio access network node sent by the radio access network AI layer. In one example, the radio access network AI layer can transmit the network performance data to the central cloud at regular intervals according to pre-configured time. In one example, the central cloud can receive the user feedback information corresponding to each radio access network node sent by the radio access network AI layer.

[0123] Step S304: Determine the network optimization task based on the network performance data and the first model deployed in the central cloud.

[0124] In one example, the central cloud can input the network performance data into the first model deployed in the central cloud, and analyze the network performance data through pre-constructed prompt words and the first model to determine the network optimization task corresponding to the network performance data. The pre-constructed prompt words can configure the data structure of the network optimization task. For example, the network optimization task can include information such as the node identifier of the node to be optimized and the identifier of the second model to be called.

[0125] For example, when analyzing the network performance data, the first model can combine the historical performance data of each radio access network node stored in the central cloud and the historical configuration parameters of each radio access network node to analyze the current network performance parameters, thereby improving the accuracy of the analysis. The first model can first determine whether the network performance data is abnormal. If it is determined that the network performance data is abnormal, the first model can further determine the reason for the network performance degradation and determine a solution based on the reason. Finally, the first model generates a network optimization task corresponding to the solution.

[0126] Step S306: Send the network optimization task to the radio access network AI layer to trigger the radio access network AI layer to decompose the network optimization task, obtain at least one corresponding execution instruction carrying the node identifier and the model identifier of the second model, and send the execution instruction to the to-be-optimized node corresponding to the node identifier, triggering the to-be-optimized node to call the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the to-be-optimized node.

[0127] Among them, the second model is deployed on the radio access network node and the model size is smaller than that of the first model; the to-be-optimized nodes include at least radio access network nodes.

[0128] Specifically, after generating a network optimization task, the central cloud can send the network optimization task to the AI layer of the radio access network. After receiving the network optimization task sent by the central cloud, the AI layer of the radio access network can decompose the network optimization task to determine the nodes to be optimized, the parameters to be optimized in the nodes to be optimized, and the second model for predicting the parameters to be optimized. Based on this, the AI layer of the radio access network can generate at least one execution instruction for the steps to be performed in the network optimization task, and store the node identifier of the node to be optimized and the model identifier of the second model in each execution instruction. The AI layer of the radio access network can send the execution instructions to the multiple nodes to be optimized corresponding to the node identifiers according to the node identifiers in each execution instruction. The nodes to be optimized can call the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the nodes to be optimized.

[0129] In the above optimization method of the radio access network, the central cloud receives the network performance data of the radio access network nodes sent by the AI layer of the radio access network, determines a network optimization task based on the network performance data and the first model deployed in the central cloud, and sends the network optimization task to the AI layer of the radio access network to trigger the AI layer of the radio access network to decompose the network optimization task, obtain at least one corresponding execution instruction carrying the node identifier and the model identifier of the second model, and send the execution instruction to the node to be optimized corresponding to the node identifier, triggering the node to be optimized to call the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the node to be optimized. It can obtain network performance data from the AI layer of the radio access network, generate a network optimization task through the large model deployed in the central cloud, decompose the network optimization task in the AI layer of the radio access network, and send the decomposed execution instruction to the node to be optimized, so that the node to be optimized can call the small model deployed in the radio access network node, thereby executing the execution instruction, which can reduce the network resource consumption between the central cloud and the radio access network nodes, improve the real-time performance of the central cloud, and share the optimization task of the large model deployed in the central cloud through the small model deployed in the radio access network node, improving the flexibility of resource allocation and resource utilization, and thus improving the optimization efficiency of the network performance of the radio access network nodes.

[0130] In an exemplary embodiment, the specific implementation process of the step "determine a network optimization task based on the network performance data and the first model deployed in the central cloud" includes:

[0131] Obtain the historical performance data of the radio access network nodes and the existing configuration parameters of the radio access network nodes; input the historical performance data, the existing configuration parameters, and the network performance data into the first model to obtain a network optimization task for the network performance data.

[0132] Among them, the historical performance data is the network performance data of the radio access network nodes in the past period of time. The existing configuration parameters are the current configuration parameters of each radio access network node for network performance. The network optimization task includes the node identifier of the node to be optimized and the parameter to be optimized.

[0133] Specifically, the central cloud can control the first model to process according to the pre-configured prompt words, obtain the historical performance data of the radio access network nodes, and the existing configuration parameters of the radio access network nodes. The central cloud can analyze the current network performance data based on the historical performance data and the existing configuration parameters through the pre-configured prompt words based on the first model, determine whether there is a network performance anomaly, and generate an optimal solution for the network performance anomaly in the case of a performance anomaly. The central cloud can convert the solution into a network optimization task with a pre-configured data structure through the pre-configured prompt words. In an example, the node identifier of the node to be optimized and the parameter to be optimized can be saved in the data structure.

[0134] In an example, the central cloud can input the pre-configured data acquisition prompt words into the first model, and through the pre-configured data acquisition prompt words, determine the historical performance data related to the radio access network node corresponding to the received network performance data, and the existing configuration parameters corresponding to the radio access network node. For example, the central cloud can obtain the historical performance data of the radio access network node in the past year, and obtain the historical configuration parameters of the radio access network node in the past year, as well as the current existing configuration parameters. The central cloud can input the pre-configured data acquisition prompt words into the first model. The central cloud can analyze the network performance data according to the pre-configured anomaly analysis prompt words and the historical performance data to determine whether the network performance data is abnormal. In the case of abnormal network performance data, based on the historical configuration parameters and the existing configuration parameters, determine how to adjust the parameters corresponding to the radio access network node, so as to obtain an optimal solution. The central cloud can input the pre-configured task generation prompt words into the first model. Through the task generation prompt words and the optimal solution, the first model can convert the optimal solution into structured data according to the pre-configured data structure and determine that the converted structured data is the network optimization task.

[0135] In this embodiment, through the first model, the network performance data can be analyzed based on the historical performance data and the existing configuration parameters, so as to obtain the optimal solution in the case of performance anomaly of the network performance data, and generate a network optimization task corresponding to the solution. It can analyze based on historical data and a large model, and be limited by a pre-configured data structure to generate a network optimization task, which can improve the accuracy of generating the network optimization task and the stability of the generated network optimization task.

[0136] In an exemplary embodiment, the method for optimizing a radio access network further includes: after an execution instruction is executed by a node to be optimized, receiving new network performance data of a radio access network node sent by the AI layer of the radio access network; determining a performance change of the new network performance data; if the performance change meets the expected target, determining that the verification result of the new network performance data is verification successful; if the performance change does not meet the expected target, determining that the verification result of the new network performance data is verification failed.

[0137] Wherein, the performance change is the change range of the network performance data of the radio access network node before and after network performance optimization of the radio access network node. The expected target may be the lowest threshold corresponding to the change range of a certain network performance data. For example, when the network performance data is delay, the expected target corresponding to the delay may be a promotion rate with a performance change reaching 50%.

[0138] Specifically, after the execution instruction is executed by the node to be optimized, the central cloud can receive the new network performance data sent by the AI layer of the radio access network. The central cloud can determine the performance change between the new network performance data and the network performance data before optimization based on the new network performance data and the network performance data before optimization. In one example, the central cloud can subtract the new network performance data from the network performance data before optimization to obtain a difference value, and determine it as the performance change. The central cloud can calculate the proportion corresponding to the difference value in the network performance data based on the difference value to obtain a change rate, and determine it as the performance change.

[0139] The central cloud can configure the expected target corresponding to each network performance data, and determine whether the performance change of each network performance data meets the expected target. For example, for each new network performance data, when the performance change meets the expected target, determining that the verification result of the new network performance data is verification successful. When the performance change does not meet the expected target, determining that the verification result of the new network performance data is verification failed.

[0140] In this embodiment, by receiving the new network performance data of the radio access network node sent by the AI layer of the radio access network after the execution instruction is executed by the node to be optimized; determining the performance change of the new network performance data; and determining the verification result of the new network performance data according to the performance change, so as to determine whether the network performance optimization is successful, the verification accuracy of the central cloud can be improved.

[0141] In an exemplary embodiment, the method for optimizing a radio access network further includes: if the verification result of the new network performance data is verification failed, then returning to the step of determining a network optimization task based on the network performance data and a first model deployed on the central cloud, and sending the network optimization task to the AI layer of the radio access network until the verification result is verification successful.

[0142] Specifically, when the verification result is successful, the central cloud does not need to perform other actions. When the verification result is a failure, the central cloud can analyze the new network performance data for which the verification fails by invoking the first model, and regenerate the corresponding network optimization task. The central cloud can send the regenerated network optimization task to the AI layer of the radio access network for decomposition to obtain at least one corresponding execution instruction, and send it to the radio access network node so that the radio access network node executes the execution instruction. The above steps can be continuously repeated until the verification result of the central cloud for the new network performance data is successful.

[0143] In this embodiment, by repeatedly executing the steps of determining the network optimization task based on the network performance data and the first model deployed in the central cloud, and sending the network optimization task to the AI layer of the radio access network until the verification result is successful, the network stability of the radio access network nodes after optimizing the network performance can be improved.

[0144] In an exemplary embodiment, as Figure 4 shown, an optimization method for a radio access network is provided. Taking the radio access network node 104 in Figure 1 as an example, the following steps S402 to S406 are included. Among them:

[0145] Step S402: Send the network performance data to the AI layer of the radio access network so that the AI layer of the radio access network sends the network performance data to the central cloud.

[0146] Specifically, the radio access network node can monitor multiple user terminals with communication connections, and determine the network performance data of the radio access network node based on the network performance data corresponding to each user terminal. In one example, the radio access network node can monitor the network performance data of each user terminal, for example, traffic load, latency, packet loss rate. And statistically analyze the network performance data of each user terminal to obtain the network performance data of the radio access network node. For example, the process of statistical analysis can be to determine the corresponding median, average value, etc. based on the network performance data of the user terminal.

[0147] In one example, the radio access network node can send the network performance data to the AI layer of the radio access network. For example, the radio access network node can send the network performance data to the AI layer of the radio access network in real time, and the radio access network node can send the network performance data corresponding to each preset period to the AI layer of the radio access network at preset period intervals.

[0148] In one example, a radio access network node may receive user feedback information sent by a user terminal. The user feedback information may be a complaint information from the user regarding insufficient network performance of the radio access network node, or may be feedback information indicating good network performance provided by the user. The radio access network node may send the user feedback information to the radio access network AI layer and send corresponding network performance data. The radio access network node may send the corresponding user feedback information and network performance data within each preset period to the radio access network AI layer at intervals of the preset period.

[0149] Step S404: Receive at least one execution instruction returned by the radio access network AI layer.

[0150] Among them, the execution instruction is obtained by the radio access network AI layer decomposing the network optimization task sent by the central cloud; the network optimization task is based on the network performance data and the first model deployed in the central cloud; the execution instruction carries the node identifier and the model identifier of the second model, and the second model is deployed on the radio access network node and the model size is smaller than that of the first model.

[0151] Specifically, if at least one execution instruction obtained by the radio access network AI layer decomposing the network optimization task carries the node identifier corresponding to the radio access network node, it is determined that the radio access network node is a node to be optimized. The radio access network node may receive at least one execution instruction sent by the radio access network AI layer based on the node identifier. In one example, the radio access network AI layer may determine the node to be optimized according to the node identifier carried in the execution instruction and send the corresponding execution instruction to the node to be optimized. The radio access network node may be the node to be optimized, and the radio access network node may receive the execution instruction carrying its own node identifier.

[0152] Step S406: Invoke the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the radio access network node.

[0153] Specifically, after the radio access network node receives the execution instruction carrying the node identifier, it may invoke the required second model based on the execution instruction and optimize the radio access network node according to the invocation result of the second model, thereby improving the network performance of the radio access network node. In one example, the radio access network node may receive a set of execution instructions obtained by packing multiple execution instructions, and execute the multiple execution instructions in the order of the set of execution instructions, thereby invoking the second model through the execution instruction, obtaining the inference result of the second model, and optimizing the radio access network node according to the inference result and the execution instruction.

[0154] In the above optimization method for a radio access network, network performance data can be sent to the radio access network AI layer through a radio access network node, so that the radio access network AI layer sends the network performance data to the central cloud; receive at least one execution instruction returned by the radio access network AI layer; call the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the radio access network node, where the execution instruction is obtained by the radio access network AI layer decomposing the network optimization task sent by the central cloud; the network optimization task is obtained based on the network performance data and the first model deployed in the central cloud; the execution instruction carries the node identifier and the model identifier of the second model, and the second model is deployed on the radio access network node and the model size is smaller than the size of the first model. Based on the above optimization method, network performance data can be sent to the central cloud through the radio access network AI layer, a network optimization task can be generated by the first model deployed in the central cloud, the network optimization task can be decomposed at the radio access network AI layer, the decomposed execution instruction can be returned to the radio access network node through the radio access network AI layer, and the radio access network node can call the second model deployed in each radio access network node to execute the execution instruction, which can reduce the network resource consumption between the central cloud and the radio access network node, improve the real-time performance of the central cloud, and share the optimization task of the first model deployed in the central cloud through the second model deployed on the radio access network node, improving the flexibility of resource allocation and resource utilization rate, thereby improving the optimization efficiency of the network performance of the radio access network node.

[0155] In an exemplary embodiment, the specific implementation process of the step "sending network performance data to the radio access network AI layer" includes:

[0156] Obtain network performance data in real time; if it is determined that the network monitoring index is abnormal based on the network performance data, send the network performance data to the radio access network AI layer; or, if a preset user feedback message is received, send the network performance data and the preset user feedback message to the radio access network AI layer.

[0157] Among them, an abnormal network monitoring index means that the index of the network performance data obtained in real time does not match the pre-configured index range; the preset user feedback message can be a pre-configured user feedback message. For example, the complaint information of the user about the insufficient network performance can be configured as the preset user feedback message.

[0158] Specifically, the radio access network node can obtain the network performance data between each user terminal and the radio access network node in real time. The radio access network node can merge the network performance data corresponding to each user terminal to obtain the network performance data corresponding to the radio access network node. For each item of network performance data, the radio access network can determine whether the network performance data is within the pre-configured index range. If the network performance data is within the pre-configured index range, it is determined that the network monitoring index is normal. If the network performance data is outside the pre-configured index range, it is determined that the network monitoring index is abnormal. In the case where the network monitoring index is abnormal, the radio access network node can send the network performance data to the radio access network AI layer.

[0159] In one example, after the radio access network node determines the network performance data, if the radio access network node receives user feedback information sent by the user terminal, such as a complaint information regarding insufficient network performance, it can send the network performance data and the user feedback information to the radio access network AI layer. For example, if the user terminal sends multiple pieces of user feedback information to the radio access network node, and the number of pieces of information that are preset user feedback information in the user feedback information reaches the feedback threshold, the radio access network node can obtain the current network performance data and package and send the obtained preset user feedback information and the current network performance data to the radio access network AI layer.

[0160] In this embodiment, by sending the network performance data to the radio access network AI layer respectively through the abnormal network monitoring index and the actual scenario corresponding to the preset user feedback information, the rationality of reporting the network performance data can be improved, and the network resources between the radio access network node and the radio access network AI layer can be optimized.

[0161] In an exemplary embodiment, the execution instructions include a model call instruction and a parameter configuration instruction. The specific implementation process of the step "receiving at least one execution instruction returned by the radio access network AI layer and calling the second model corresponding to the model identifier to execute the execution instruction" includes:

[0162] Receiving at least one model call instruction and at least one parameter configuration instruction returned by the radio access network AI layer; based on the model identifier carried in the model call instruction, calling the second model corresponding to the model identifier to predict the predicted adjustment value corresponding to the parameter to be optimized; adjusting the parameter to be optimized to the predicted adjustment value based on the parameter configuration instruction.

[0163] Among them, the model call instruction is used to call the determined second model and predict the predicted adjustment value of each parameter to be optimized through the second model; the parameter configuration instruction is used to adjust the parameter to be optimized to the predicted adjustment value. The predicted adjustment value is the prediction result obtained by the node to be optimized by calling the second model deployed in the radio access network node and through the second model.

[0164] Specifically, after decomposing the network optimization tasks at the AI layer of the radio access network, multiple execution instructions composed of at least one model invocation instruction and at least one parameter configuration instruction can be obtained. The AI layer of the radio access network can send the corresponding execution instructions to the radio access network nodes according to the node identifiers carried in the execution instructions. Based on this, the radio access network nodes can receive the model invocation instructions and parameter configuration instructions carrying their own node identifiers.

[0165] The radio access network nodes can execute the model invocation instructions, and according to the model identifiers carried in the model invocation instructions, invoke the second model corresponding to the model identifiers, and predict the predicted adjustment value corresponding to the parameter to be optimized through data such as the network performance parameters of the radio access network nodes themselves, the historical parameters corresponding to the parameters to be optimized, etc.

[0166] In one example, the radio access network nodes can, according to the model identifiers carried in the model invocation instructions, determine the radio access network nodes in the radio access network where the second model corresponding to the model identifiers is deployed, and initiate a call request to the radio access network nodes where the second model is deployed, so that the second model can be invoked by the radio access network nodes. The radio access network nodes can send the input data related to the second model, such as the network performance parameters of the radio access network nodes themselves, the historical parameters corresponding to the parameters to be optimized, etc., to the radio access network nodes where the second model is deployed, for the radio access network nodes where the second model is deployed to invoke the second model and predict the predicted adjustment value corresponding to the parameter to be optimized. The second model can be a load balancing model for network load or a power control model for power, which is not specifically limited in the embodiments of the present application.

[0167] In one example, the radio access network nodes can adjust the parameter to be optimized according to the parameter configuration instructions. The radio access network nodes can, based on the predicted adjustment value for the parameter to be optimized predicted by the second model, adjust the corresponding parameter to be optimized value to the predicted adjustment value through the parameter configuration instructions. For example, the power value of the radio access network node is A, and the power value predicted by the second model is B, and the parameter configuration instructions can control the power value of the radio access network to be adjusted from A to B.

[0168] In this embodiment, through the model invocation instructions, the second model can be invoked, and the predicted adjustment value for the parameter to be optimized can be predicted, and through the parameter configuration instructions, the parameter to be optimized can be adjusted to the predicted adjustment value, so that the parameter to be optimized can be accurately adjusted, and the accuracy and efficiency of optimizing the network performance can be improved.

[0169] In an exemplary embodiment, after the step "invoking the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the radio access network node", the optimization method of the radio access network further includes:

[0170] After executing the execution instruction, re-obtain the current network performance data; send the current network performance data to the wireless access network AI layer, so as to send the network performance data to the central cloud through the wireless access network AI layer for verification.

[0171] In one example, after the wireless access network node executes the execution instruction, after a period of time, the wireless access network node can re-obtain the current network performance data. The wireless access network node can send the current network performance data to the wireless access network AI layer, and send the network performance data to the central cloud through the wireless access network AI layer. The central cloud can verify the network performance data. If the verification result of the central cloud for the network performance data is verification failure, receive at least one re-generated execution instruction returned by the wireless access network AI layer. The wireless access network node can perform optimization again according to the execution instruction, and after the wireless access network node executes the execution instruction, repeat the above process until the verification result of the central cloud for the network performance data is verification success.

[0172] In this embodiment, by re-obtaining the current network performance data, sending the current network performance data to the wireless access network AI layer, so as to send the network performance data to the central cloud through the wireless access network AI layer for verification, receiving the re-generated execution instruction sent by the wireless access network AI layer, and executing the instruction, so as to perform network optimization again, the network stability of the wireless access network node after optimizing the network performance can be improved.

[0173] As Figure 5 shown, the following combines a specific embodiment to describe in detail the specific execution process of the above optimization method for the wireless access network. The embodiment of the present application proposes a technical solution based on the collaboration of the central cloud large model and the AI RAN small model, and realizes the intelligent and automatic optimization of the network through the RAN AI Layer, especially for the dynamic adjustment and optimization problems of network configuration. The embodiment of the present application is mainly applied to the dynamic optimization scenario in a complex network environment. For example: for network performance optimization, when the user experience drops due to load change, interference enhancement or improper configuration in a certain base station area, the large model of the central cloud can analyze the root cause of the problem and command the small model of the AI RAN node to perform optimization operations through the RAN AI Layer. For intelligent complaint handling: after the user feedbacks network problems through the customer service or mobile application, the large model predicts the optimal solution according to historical data and complaint content, and collaborates with the small model for optimization. For resource scheduling and configuration, in a multi-base station scenario, the RAN AI Layer is used to overall schedule fragmented computing power resources to ensure the efficient collaborative work of multiple small models.

[0174] In one example, the network element devices and their functions required for the technical solution based on the cooperation between the central cloud large model and the AI RAN small model include: the central cloud, the RAN AI Layer, and the AI RAN node.

[0175] (1) The central cloud. Form: A high-performance computing cluster that deploys a global AI large model and has powerful computing and storage capabilities. Functions: Global problem analysis and prediction, solution generation, full-network status monitoring, and historical data management.

[0176] (2) The RAN AI Layer. Form: The AI RAN logical control plane, which can be deployed in a data center or edge device to control the AI RAN node. Functions include: model scheduling, data processing, service orchestration, and general computing resource allocation. For example, model scheduling can be to select and assign small models to the AI RAN node according to the central cloud instructions. Data processing can be to aggregate and preprocess performance data and transfer it to the central cloud or the AI RAN node. Service orchestration can be to translate the solutions of the large model into execution instructions for the AIRAN. General computing resource allocation can be to coordinate fragmented GPU computing power to support the efficient operation of small models.

[0177] (3) The AI RAN node. Form: An intelligent base station with computing capabilities, built with a CPU and a GPU, and can directly run small models on the RAN side. Functions: Local inference and optimization, including real-time data processing, execution configuration adjustment, and uploading of optimized effect data.

[0178] In an exemplary embodiment, the specific implementation process of the technical solution based on the cooperation between the central cloud large model and the AI RAN small model includes:

[0179] Step 1: Collection and upload of performance data. Execution entity: The AI RAN node; Trigger condition: User feedback on network performance degradation or abnormal network monitoring metrics. Processing actions: The AI RAN node monitors its own performance data in real time (such as traffic load, latency, packet loss rate, etc.). After preprocessing the data, it uploads it to the data management function in the RAN AI Layer. Result: Generate a local performance data report and send it to the data management function in the RAN AI Layer. Role: Provide data support for the subsequent analysis of the large model in the central cloud.

[0180] Step 2: Network-wide Performance Analysis and Optimization Requirement Generation. Execution Entity: Central Cloud; Trigger Condition: The performance data uploaded by the data management function in the RAN AI Layer to the Central Cloud reaches the analysis threshold or a user complaint triggers an analysis request. Processing Action: The Central Cloud large model receives the network-wide data aggregated by the RAN AI Layer. Analyze the root cause of the network performance degradation (such as increased interference, uneven load) based on historical data and existing configurations. Predict the optimal solution and generate an optimization task (including target nodes, adjustment parameters, target effects, etc.). Result: An optimization task instruction. Function: Provide a basis for the RAN AI Layer to issue a specific execution plan.

[0181] Step 3: Task Decomposition and Model Scheduling of Optimization Tasks; Execution Entity: RAN AI Layer; Trigger Condition: Receive the optimization task instruction issued by the Central Cloud. Processing Action: Decompose the optimization task into specific operation steps (such as which nodes need to be adjusted and which parameters to adjust). Select an appropriate small model (such as a load balancing model, power control model) according to the task requirements and node characteristics. The model management function in the RAN AI Layer issues specific execution instructions to relevant AI RAN nodes. Result: A set of task allocation instructions, including the call information of the small model and configuration adjustment information. Function: Coordinate fragmented computing power resources and small model execution to achieve efficient task allocation.

[0182] Step 4: Localized Optimization and Configuration Adjustment; Execution Entity: AI RAN Node; Trigger Condition: Receive the task instruction issued by the RAN AI Layer. Processing Action: Call the specified small model and complete the inference in combination with local real-time data. Execute specific configuration adjustments (such as adjusting power, optimizing antenna direction, switching load, etc.). Result: The network configuration parameters are optimized and the user experience is improved. Function: Quickly respond to optimization requirements through localized small model execution and improve network performance.

[0183] Step 5: Optimization Effect Verification and Feedback; Execution Entity: AI RAN Node and RAN AI Layer; Trigger Condition: Within a certain period of time after the configuration adjustment is completed, the change in performance data reaches the verification standard. Processing Action: The AI RAN node monitors the adjusted performance data and generates feedback data. Upload the feedback data to the data management function in the RAN AI Layer, and the data management function in the RAN AI Layer further aggregates and uploads it to the Central Cloud. Result: The Central Cloud verifies whether the optimization effect achieves the expected goal. Function: Close-loop verification of the effectiveness of the optimization task and provide a basis for subsequent optimization.

[0184] Step 6: Further optimization or task completion; Execution entity: Central cloud and RAN AI Layer; Trigger condition: The verification result of the optimization effect fails to meet the expected goal or the user continues to complain. Processing actions: If the verification effect is not good, the central cloud re-analyzes the root cause of the problem, adjusts the optimization plan, and re-issues the task. If the verification effect is good, the task ends. Result: The optimization task enters a new round of cycle or is terminated. Function: Ensure the continuity and closed-loop control of the optimization task.

[0185] In an exemplary embodiment, as Figure 6 shown, it represents the relationship between different radio access network nodes and the Radio Access Network Intelligence Management and Orchestration Function (RAN AI Layer). It can be seen that the Radio Access Network Intelligence Management and Orchestration Function (RAN AI Layer) can include four modules: model management function, data management service, service management orchestration, and computing and resource scheduling. And the Radio Access Network Intelligence Management and Orchestration Function can schedule and control, for example, 3GPP RAN, cloud-native RAN, and other RAN solutions.

[0186] Through the above technical solutions, this embodiment realizes the collaborative optimization between the large model and the small model between the central cloud and AI RAN, makes full use of the fragmented computing power resources of AI RAN, and demonstrates significant technical advantages in the network optimization scenario. Specifically, in this embodiment, the central cloud large model is used for global analysis and decision-making, combined with the refined decomposition of the optimization task by the RAN AI Layer and the small model scheduling, to improve the accuracy and efficiency of task allocation; at the same time, the local deployment of the small model on the AI RAN side can quickly respond to dynamic network requirements, effectively reduce latency and improve the user experience. Through the closed-loop optimization verification mechanism, this embodiment ensures the effectiveness and reliability of network configuration adjustment, significantly reduces the cost of manual intervention, and improves the intelligence and automation level of the network. This technical solution solves the problems of low resource utilization efficiency and high optimization latency in the prior art.

[0187] Specifically, the beneficial effects of this embodiment are as follows:

[0188] (1) Collaborative optimization between the central cloud large model and the AI RAN small model. In this embodiment, the large model is deployed in the central cloud for global analysis and decision-making, and the small model is deployed in the AI RAN to be responsible for local inference and execution, realizing end-to-end task collaborative optimization. This collaborative mechanism significantly improves the computing efficiency of the model and the network response speed. Compared with the prior art, it breaks through the limitations of the single deployment mode and achieves the goal of precise and dynamic resource allocation.

[0189] (2) Intelligent scheduling capabilities of the RAN AI Layer. In this embodiment, the RAN AI Layer is introduced as the core control plane, with the capabilities of model scheduling, data processing, service orchestration, and communication computing resource allocation. Through the scheduling of small models by the RAN AI Layer, computing power resources can be flexibly allocated according to different scenarios, enabling the efficient utilization of fragmented computing power of AI RAN nodes. Compared with the traditional method that relies on central cloud computing, this solution significantly reduces network latency and improves the response ability to dynamic service requirements.

[0190] (3) Introduction of a closed-loop optimization mechanism. In the network configuration optimization scenario, in this embodiment, the large model analyzes the reasons for problems such as user complaints, generates optimization strategies, and sends them to the RAN AI Layer, where the small models execute the adjustments locally. At the same time, the system continuously monitors user experience feedback (such as changes in the number of work orders) to verify the optimization effect, forming a closed-loop optimization mechanism. Compared with the traditional one-way optimization scheme, this embodiment can correct in a timely manner, improving the reliability and effectiveness of optimization.

[0191] (4) Efficient utilization of fragmented computing power. By using the local GPU resources of AI RAN nodes for small model inference, the present invention effectively integrates the fragmented computing power resources within the RAN and reduces the dependence on central cloud computing resources. Compared with the existing technology that mainly relies on cloud computing power, this solution reduces the overall computing power cost while ensuring computing performance and provides greater flexibility for computing power resource scheduling.

[0192] (5) Intelligent task decomposition for scenarios. This embodiment supports flexibly adjusting optimization strategies according to specific scenarios (such as network problems complained by users), decomposing complex problems into multiple small tasks, and having the small models complete them respectively. This task decomposition and execution method makes the optimization process more efficient and accurate. Compared with the traditional static rule-based optimization method, the present invention has stronger adaptability and intelligent features.

[0193] Through the above technical means, this embodiment shows significant technical advantages in aspects such as multi-level computing power collaboration, utilization of fragmented computing power, closed-loop control of network optimization, and dynamic response ability, providing an innovative solution for AI-driven intelligent network construction.

[0194] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this document, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0195] Based on the same inventive concept, an embodiment of the present application also provides an optimization device for a radio access network for implementing the optimization method for the radio access network described above. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the optimization device for the radio access network provided below can refer to the limitations on the optimization method for the radio access network in the above text, and will not be repeated here.

[0196] In an exemplary embodiment, as Figure 7 shown, an optimization device 700 for a radio access network is provided, including: a first receiving module 701, a first sending module 702, a second receiving module 703, a decomposition module 704, and a second sending module 705, where:

[0197] The first receiving module 701 is configured to receive network performance data sent by a radio access network node;

[0198] The first sending module 702 is configured to send the network performance data to the central cloud;

[0199] The second receiving module 703 is configured to receive a network optimization task sent by the central cloud; the network optimization task is obtained based on the network performance data and a first model deployed in the central cloud;

[0200] The decomposition module 704 is configured to decompose the network optimization task to obtain at least one corresponding execution instruction; the execution instruction carries a node identifier and a model identifier of a second model, and the second model is deployed on the radio access network node and the model size is smaller than the size of the first model;

[0201] The second sending module 705 is configured to send the execution instruction to the to-be-optimized node corresponding to the node identifier, and trigger the to-be-optimized node to call the second model corresponding to the model identifier to execute the execution instruction, so as to optimize the network performance of the to-be-optimized node; the to-be-optimized node at least includes a radio access network node.

[0202] Further, the first sending module 702 is specifically configured to: when a preset reporting condition is met, send network performance data to the central cloud; the preset reporting condition includes at least one of a preset reporting period, a preset data volume, and receiving a preset user feedback message.

[0203] Further, the network optimization task includes a node identifier of a node to be optimized and a parameter to be optimized. The decomposition module 704 is specifically configured to: based on the node identifier and the parameter to be optimized included in the network optimization task, determine at least one second model for the node to be optimized corresponding to the node identifier, and generate a model call instruction and a parameter configuration instruction; the model call instruction carries a model identifier of the determined second model; the model call instruction is used to call the determined second model, and predict a predicted adjustment value of each parameter to be optimized through the second model; the parameter configuration instruction is used to adjust the parameter to be optimized to the predicted adjustment value.

[0204] Further, the optimization device for the radio access network further includes a verification module, which is specifically configured to: after the node to be optimized executes the execution instruction, receive new network performance data sent by the radio access network node, and send the new network performance data to the central cloud; in the case where the verification result of the central cloud for the new network performance data fails the verification, receive the regenerated network optimization task sent by the central cloud, and return to execute the steps of decomposing the network optimization task until the verification result is successful.

[0205] In an exemplary embodiment, as Figure 8 shown, an optimization device 800 for a radio access network is provided, including: a receiving module 801, a task determination module 802, and a sending module 803, where:

[0206] The receiving module 801 is configured to receive network performance data of radio access network nodes sent by the radio access network AI layer;

[0207] The task determination module 802 is configured to determine a network optimization task based on the network performance data and a first model deployed in the central cloud;

[0208] The sending module 803 is configured to send the network optimization task to the radio access network AI layer to trigger the radio access network AI layer to decompose the network optimization task, obtain at least one corresponding execution instruction carrying a node identifier and a model identifier of a second model, and send the execution instruction to the node to be optimized corresponding to the node identifier, triggering the node to be optimized to call the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the node to be optimized; wherein, the second model is deployed on the radio access network node and the model size is smaller than the size of the first model; the node to be optimized includes at least a radio access network node.

[0209] Further, the task determination module 802 is specifically configured to: obtain the historical performance data of the radio access network node and the existing configuration parameters of the radio access network node; input the historical performance data, the existing configuration parameters, and the network performance data into the first model to obtain a network optimization task for the network performance data; the network optimization task includes the node identifier of the node to be optimized and the parameter to be optimized.

[0210] Further, the optimization device 800 of the radio access network further includes a verification module, which is specifically configured to: after the node to be optimized executes the execution instruction, receive the new network performance data of the radio access network node sent by the AI layer of the radio access network; determine the performance change of the new network performance data; if the performance change meets the expected target, determine that the verification result of the new network performance data is verification successful; if the performance change does not meet the expected target, determine that the verification result of the new network performance data is verification failed.

[0211] Further, the verification module is specifically further configured to: if the verification result of the new network performance data is verification failed, return to the step of executing to determine the network optimization task based on the network performance data and the first model deployed in the central cloud, and send the network optimization task to the AI layer of the radio access network until the verification result is verification successful.

[0212] In an exemplary embodiment, as Figure 9 shown, an optimization device 900 of a radio access network is provided, including: a sending module 901, a receiving module 902, and an instruction execution module 903, where:

[0213] The sending module 901 is configured to send the network performance data to the AI layer of the radio access network, so that the AI layer of the radio access network sends the network performance data to the central cloud;

[0214] The receiving module 902 is configured to receive at least one execution instruction returned by the AI layer of the radio access network; the execution instruction is obtained by the AI layer of the radio access network decomposing the network optimization task sent by the central cloud; the network optimization task is obtained based on the network performance data and the first model deployed in the central cloud; the execution instruction carries the node identifier and the model identifier of the second model, and the second model is deployed on the radio access network node and the model size is smaller than the size of the first model;

[0215] The instruction execution module 903 is configured to call the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the radio access network node.

[0216] Further, the sending module 901 is specifically configured to: obtain network performance data in real time; if it is determined based on the network performance data that a network monitoring indicator is abnormal, send the network performance data to the AI layer of the radio access network; or, if a preset user feedback message is received, send the network performance data and the preset user feedback message to the AI layer of the radio access network.

[0217] Further, the execution instruction includes a model call instruction and a parameter configuration instruction; the receiving module 902 and the instruction execution module 903 are specifically configured to: receive at least one model call instruction and at least one parameter configuration instruction returned by the AI layer of the radio access network; based on the model identifier carried in the model call instruction, call a second model corresponding to the model identifier, and predict a predicted adjustment value corresponding to the parameter to be optimized; adjust the parameter to be optimized to the predicted adjustment value based on the parameter configuration instruction.

[0218] Further, the optimization device 900 of the radio access network further includes a verification module, which is specifically configured to: after executing the execution instruction, re-obtain the current network performance data; send the current network performance data to the AI layer of the radio access network, so as to send the network performance data to the central cloud through the AI layer of the radio access network for verification.

[0219] Each module in the above-mentioned optimization device of the radio access network can be implemented in whole or in part by software, hardware, and their combination. The above-mentioned modules can be embedded in the processor in the communication device in hardware form or independent of the processor, or stored in the memory in the communication device in software form, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

[0220] In an exemplary embodiment, a communication device is provided. The communication device can be a server, and its internal structure diagram can be as Figure 10 shown. The communication device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the communication device is used to provide computing and control capabilities. The memory of the communication device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the communication device is used to store the network performance data of the radio access network nodes. The input / output interface of the communication device is used to exchange information between the processor and external devices. The communication interface of the communication device is used to communicate with external terminals through a network connection. The computer program, when executed by the processor, implements an optimization method for a radio access network.

[0221] Those skilled in the art can understand that Figure 10 The structure shown in Figure 10 is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the communication device to which the solution of this application is applied. The specific communication device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0222] In an exemplary embodiment, a communication device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0223] In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0224] In an embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0225] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0226] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in this application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0227] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0228] The above-described embodiments merely represent several implementation manners of this application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of this application. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several modifications and improvements can still be made, and these all belong to the protection scope of this application. Therefore, the protection scope of this application should be subject to the appended claims.

Claims

1. An optimization method for a radio access network, characterized in that Applied to the AI layer of the radio access network, the method includes: Receiving network performance data sent by a radio access network node; Sending the network performance data to the central cloud; Receiving a network optimization task sent by the central cloud; the network optimization task is obtained based on the network performance data and a first model deployed in the central cloud; the network optimization task includes a node identifier of a node to be optimized and a parameter to be optimized; Decomposing the network optimization task to obtain at least one corresponding execution instruction; the execution instruction carries a node identifier and a model identifier of a second model, and the second model is deployed on a radio access network node and has a model size smaller than that of the first model; Sending the execution instruction to the node to be optimized corresponding to the node identifier, triggering the node to be optimized to call the second model corresponding to the model identifier to execute the execution instruction, so as to optimize the network performance of the node to be optimized; the node to be optimized at least includes the radio access network node that sent the network performance data; Wherein, the execution instruction includes a model call instruction and a parameter configuration instruction; the model call instruction is used to call the second model determined by the radio access network AI layer to predict a predicted adjustment value of each parameter to be optimized through the second model; the parameter configuration instruction is used to adjust the parameter to be optimized to the predicted adjustment value.

2. The method according to claim 1, wherein The sending the network performance data to the central cloud includes: Sending the network performance data to the central cloud when a preset reporting condition is met; The preset reporting condition includes at least one of a preset reporting period, a preset data volume, and receiving a preset user feedback message.

3. The method according to claim 1, characterized in that The decomposing the network optimization task to obtain at least one corresponding execution instruction includes: Based on the node identifier and the parameter to be optimized included in the network optimization task, determining at least one second model for the node to be optimized corresponding to the node identifier, and generating a model call instruction and a parameter configuration instruction; The model call instruction carries the model identifier of the determined second model.

4. The method according to any one of claims 1 to 3, characterized in that After the sending the execution instruction to the node to be optimized corresponding to the node identifier, the method further includes: After the node to be optimized executes the execution instruction, receiving new network performance data sent by the radio access network node, and sending the new network performance data to the central cloud; When the verification result of the central cloud for the new network performance data is verification failure, receiving a regenerated network optimization task sent by the central cloud, and returning to execute the step of decomposing the network optimization task until the verification result is verification success.

5. An optimization method for a radio access network, characterized in that Applied to the central cloud, the method includes: Receiving network performance data of a radio access network node sent by the radio access network AI layer; Based on the network performance data and a first model deployed in the central cloud, determining a network optimization task; the network optimization task includes a node identifier of a node to be optimized and a parameter to be optimized; Send the network optimization task to the AI layer of the radio access network to trigger the AI layer of the radio access network to decompose the network optimization task, obtain at least one corresponding execution instruction carrying the node identifier and the model identifier of the second model, and send the execution instruction to the to-be-optimized node corresponding to the node identifier, triggering the to-be-optimized node to call the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the to-be-optimized node; Wherein, the second model is deployed on a radio access network node and the model size is smaller than the size of the first model; the to-be-optimized node at least includes the radio access network node that sends the network performance data; Wherein, the execution instruction includes a model call instruction and a parameter configuration instruction; the model call instruction is used to call the second model determined by the AI layer of the radio access network to predict the predicted adjustment value of each to-be-optimized parameter through the second model; the parameter configuration instruction is used to adjust the to-be-optimized parameter to the predicted adjustment value.

6. The method according to claim 5, wherein The determining the network optimization task based on the network performance data and the first model deployed on the central cloud includes: Obtain the historical performance data of the radio access network node and the existing configuration parameters of the radio access network node; Input the historical performance data, the existing configuration parameters and the network performance data into the first model to obtain a network optimization task for the network performance data; the network optimization task includes the node identifier of the to-be-optimized node and the to-be-optimized parameter.

7. The method according to claim 5 or 6, characterized in that, The method further includes: After the to-be-optimized node executes the execution instruction, receive the new network performance data of the radio access network node sent by the AI layer of the radio access network; Determine the performance change of the new network performance data; If the performance change meets the expected target, determine that the verification result of the new network performance data is verification successful; If the performance change does not meet the expected target, determine that the verification result of the new network performance data is verification failed.

8. The method according to claim 7, wherein The method further includes: If the verification result of the new network performance data is verification failed, return to execute the steps of determining the network optimization task based on the network performance data and the first model deployed on the central cloud and sending the network optimization task to the AI layer of the radio access network until the verification result is verification successful.

9. An optimization method for a radio access network, characterized in that Applied to a radio access network node, the method includes: Send network performance data to the AI layer of the radio access network so that the AI layer of the radio access network sends the network performance data to the central cloud; Receive at least one execution instruction returned by the wireless access network AI layer; the execution instruction is obtained by the wireless access network AI layer decomposing the network optimization task sent by the central cloud; the network optimization task is obtained based on the network performance data and a first model deployed on the central cloud, and the network optimization task includes the node identifier of the node to be optimized and the parameter to be optimized, and the node to be optimized at least includes the wireless access network node that sends the network performance data; the execution instruction carries the node identifier and the model identifier of the second model, and the second model is deployed on the wireless access network node and the model size is smaller than the size of the first model; Invoke the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the wireless access network node; Wherein, the execution instruction includes a model invocation instruction and a parameter configuration instruction; the model invocation instruction is used to invoke the second model determined by the wireless access network AI layer to predict the predicted adjustment value of each parameter to be optimized through the second model; the parameter configuration instruction is used to adjust the parameter to be optimized to the predicted adjustment value.

10. The method according to claim 9, wherein The sending the network performance data to the wireless access network AI layer includes: Obtain network performance data in real time; If it is determined based on the network performance data that the network monitoring index is abnormal, send the network performance data to the wireless access network AI layer; or, If the preset user feedback information is received, send the network performance data and the preset user feedback information to the wireless access network AI layer.

11. The method according to claim 9, wherein The execution instruction includes a model invocation instruction and a parameter configuration instruction; The receiving at least one execution instruction returned by the wireless access network AI layer and invoking the second model corresponding to the model identifier to execute the execution instruction includes: Receive at least one model invocation instruction and at least one parameter configuration instruction returned by the wireless access network AI layer; Based on the model identifier carried in the model invocation instruction, invoke the second model corresponding to the model identifier to predict the predicted adjustment value corresponding to the parameter to be optimized; Based on the parameter configuration instruction, adjust the parameter to be optimized to the predicted adjustment value.

12. The method according to any one of claims 9-11, characterized in that, After invoking the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the wireless access network node, the method further includes: After executing the execution instruction, obtain the current network performance data again; Send the current network performance data to the wireless access network AI layer, so that the wireless access network AI layer sends the network performance data to the central cloud for verification.

13. An optimization device for a radio access network, characterized in that, The device includes: A first receiving module, configured to receive network performance data sent by a wireless access network node; A first sending module, configured to send the network performance data to the central cloud; A second receiving module, configured to receive the network optimization task sent by the central cloud; the network optimization task is obtained based on the network performance data and a first model deployed on the central cloud; the network optimization task includes the node identifier of the node to be optimized and the parameter to be optimized; A decomposition module, configured to decompose the network optimization task to obtain at least one corresponding execution instruction; the execution instruction carries a node identifier and a model identifier of a second model, and the second model is deployed on a radio access network node and has a model size smaller than that of the first model; A second sending module, configured to send the execution instruction to the node to be optimized corresponding to the node identifier, and trigger the node to be optimized to call the second model corresponding to the model identifier to execute the execution instruction, so as to optimize the network performance of the node to be optimized; the node to be optimized at least includes the radio access network node that sends the network performance data; Wherein, the execution instruction includes a model call instruction and a parameter configuration instruction; the model call instruction is used to call the second model determined by the radio access network AI layer to predict the predicted adjustment value of each to-be-optimized parameter through the second model; the parameter configuration instruction is used to adjust the to-be-optimized parameter to the predicted adjustment value.

14. An optimization device for a radio access network, characterized in that, The apparatus includes: A receiving module, configured to receive network performance data of a radio access network node sent by a radio access network AI layer; A task determination module, configured to determine a network optimization task based on the network performance data and a first model deployed on a central cloud; the network optimization task includes a node identifier of a node to be optimized and a to-be-optimized parameter; A sending module, configured to send the network optimization task to the radio access network AI layer, so as to trigger the radio access network AI layer to decompose the network optimization task to obtain at least one execution instruction carrying a node identifier and a model identifier of a second model, and send the execution instruction to the node to be optimized corresponding to the node identifier, and trigger the node to be optimized to call the second model corresponding to the model identifier to execute the execution instruction, so as to optimize the network performance of the node to be optimized; wherein, the second model is deployed on a radio access network node and has a model size smaller than that of the first model; the node to be optimized at least includes the radio access network node that sends the network performance data; Wherein, the execution instruction includes a model call instruction and a parameter configuration instruction; the model call instruction is used to call the second model determined by the radio access network AI layer to predict the predicted adjustment value of each to-be-optimized parameter through the second model; the parameter configuration instruction is used to adjust the to-be-optimized parameter to the predicted adjustment value.

15. An optimization device for a radio access network, characterized in that, The apparatus includes: A sending module, configured to send network performance data to a radio access network AI layer, so that the radio access network AI layer sends the network performance data to a central cloud; A receiving module, configured to receive at least one execution instruction returned by the AI layer of the radio access network; the execution instruction is obtained by the AI layer of the radio access network decomposing the network optimization task sent by the central cloud; the network optimization task is obtained based on the network performance data and a first model deployed in the central cloud, and the network optimization task includes the node identifier of the node to be optimized and the parameter to be optimized, and the node to be optimized at least includes the radio access network node that sends the network performance data; the execution instruction carries the node identifier and the model identifier of the second model, and the second model is deployed on the radio access network node and the model size is smaller than the size of the first model; An instruction execution module, configured to call the second model corresponding to the model identifier to execute the execution instruction to optimize the network performance of the radio access network node; Wherein, the execution instruction includes a model call instruction and a parameter configuration instruction; the model call instruction is used to call the second model determined by the AI layer of the radio access network to predict the predicted adjustment value of each parameter to be optimized through the second model; the parameter configuration instruction is used to adjust the parameter to be optimized to the predicted adjustment value.

16. A communication device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 12 are implemented.

17. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 12 are implemented.

18. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 12 are implemented.

Images