Multi-agent 5G base station intelligent energy-saving method and device based on large and small model fusion

Abstract

The invention discloses a multi-agent 5G base station intelligent energy-saving method and device based on large and small model fusion, and relates to the field of wireless communication network energy conservation, and the method can automatically recognize a network deployment scene and configure an initial energy-saving strategy corresponding to the scene through the technology of fusion of a large language data model and a generative AI model. And screening out an energy-saving cell based on the initial energy-saving strategy corresponding to each cell or region, obtaining an energy-saving strategy and an energy-saving interaction instruction corresponding to the energy-saving cell, and controlling network equipment corresponding to the energy-saving cell to enter an energy-saving state. And further iteratively optimizing a corresponding energy-saving strategy on line for the cell entering the energy-saving state according to the energy-saving effect and the change area of the load performance index, and searching a balance point between the energy-saving threshold parameter and the network performance, thereby realizing maximum energy saving.

Description

Technical Field

[0001] This application relates to the technical field of energy saving in wireless communication networks, and particularly to a multi-agent 5G base station intelligent energy saving method and device based on the fusion of large and small models. Background Art

[0002] With the continuous expansion of the scale of communication networks, the energy consumption of network devices is also increasing. Network energy saving has become one of the important topics in the communication field. How to use energy-saving and consumption-reducing technologies to ensure that the power consumption of the communication network is as small as possible, while ensuring high-performance use of the network and reducing the power consumption of the communication network as much as possible, and improving network energy efficiency.

[0003] The main means to improve network energy efficiency include: improving air interface radio technology, optimizing the energy efficiency of device hardware, and how to better match the load and network capacity. Among them, how to better match the load and network capacity has become the current main technical route. However, the threshold parameters for the current network energy saving scheme to enter and exit energy saving are defined by expert experience, and maximum energy saving cannot be achieved. Summary of the Invention

[0004] In view of the above problems, this application provides a multi-agent 5G base station intelligent energy saving method and device based on the fusion of large and small models to solve at least some of the above technical problems. The specific solutions are as follows: The first aspect of this application provides a multi-agent 5G base station intelligent energy saving method based on the fusion of large and small models, including: Identifying the network deployment scenario information corresponding to the current wireless network, and configuring an initial energy saving strategy that matches the network deployment scenario information; Screening the energy saving areas in the current wireless network based on the initial energy saving strategy, and determining the current energy saving strategy corresponding to the energy saving areas based on the in-network data corresponding to the energy saving areas; Generating an energy saving interaction instruction based on the current energy saving strategy and sending it to the wireless network device corresponding to the energy saving area, where the energy saving interaction instruction is used to control the energy saving state of the wireless network device; Online iteratively optimizing the current energy saving strategy according to the energy saving effect corresponding to the energy saving area and the change trend of the key network performance indicators.

[0005] In a possible implementation, the network deployment scenario information includes network coverage information and network configuration information, and the initial energy saving strategy includes an energy saving type, an energy saving time period, and an energy saving threshold parameter; The configuring of the initial energy saving strategy that matches the network deployment scenario information includes: Configure the energy-saving type of the current wireless network based on the network configuration information of the current wireless network, where the energy-saving type includes symbol shutdown, channel shutdown, carrier shutdown, or deep sleep of the radio frequency module; Configure the energy-saving time period and energy-saving threshold parameters corresponding to each area of the current wireless network in combination with the historical traffic load performance indicators and network coverage information corresponding to the current wireless network.

[0006] In a possible implementation, the combining the historical traffic load performance indicators and network coverage information corresponding to the current wireless network to configure the energy-saving time period and energy-saving threshold parameters corresponding to each area of the current wireless network includes: For any area of the current wireless network, determine the idle traffic load level based on the historical traffic load of the any area; Determine the energy-saving threshold parameters corresponding to the any area based on the idle traffic load level of the any area; Determine the energy-saving time period corresponding to the any area according to the traffic load prediction result corresponding to the any area and the energy-saving threshold parameters.

[0007] In a possible implementation, the initial energy-saving policy includes energy-saving threshold parameters, where the energy-saving threshold parameters include the traffic load threshold value for entering the energy-saving state or the energy-saving time period; among them, screening the energy-saving areas from the current wireless network based on the initial energy-saving policy includes: Select the areas with the predicted traffic load lower than the corresponding traffic load threshold value for entering the energy-saving state as the energy-saving areas; Or, Select the areas that are in the corresponding energy-saving time period at the current time as the energy-saving areas.

[0008] In a possible implementation, the determining the current energy-saving policy corresponding to the energy-saving area based on the in-network data corresponding to the energy-saving area includes: Obtain the internal in-network data and external in-network data corresponding to the energy-saving area, where the internal in-network data includes engineering parameter information and terminal measurement report data, and the external in-network data includes external system data that affects the traffic load performance indicators; Based on the internal in-network data and external in-network data, use the load prediction model to obtain the predicted value of the traffic load performance indicators corresponding to the energy-saving area; Based on the predicted value of the traffic load performance indicators and the energy-saving threshold parameters corresponding to the energy-saving area, determine the optimal energy-saving time period corresponding to the energy-saving area; Obtain the current energy-saving policy based on the energy-saving type and the optimal energy-saving time period corresponding to the energy-saving area.

[0009] In a possible implementation, online iteratively optimizing the current energy-saving policy according to the energy-saving effect corresponding to the energy-saving area and the change trend of the key network performance indicators includes: Increasing the current energy-saving threshold parameter corresponding to the energy-saving area by a target adjustment step; Performing energy-saving control on the energy-saving area based on the energy-saving threshold parameter after increasing the target adjustment step, and obtaining the corresponding key network performance indicators; If the key network performance indicators do not decrease or do not exceed the allowable threshold after decreasing, continue to execute the step of increasing the current energy-saving threshold parameter corresponding to the energy-saving area by the target adjustment step; If the key network performance indicators decrease and exceed the allowable threshold, restore the energy-saving threshold parameter to the value before increasing the target adjustment step to obtain the optimal energy-saving threshold parameter.

[0010] In a possible implementation, the method further includes: When the energy-saving area does not enter the energy-saving state, evaluating the network performance KPIs of the energy-saving area and the co-covered area. If the evaluation results of the network performance KPIs of the energy-saving area and the co-covered area are poor, prevent the energy-saving area from entering the energy-saving state; if the evaluation results of the network performance KPIs of the energy-saving area and the co-covered area are excellent or there is no result, allow the energy-saving area to enter the energy-saving state; When the energy-saving area has entered the energy-saving state, evaluating the network performance KPIs of the co-covered area of the energy-saving area. If the evaluation results of the network performance KPIs of the co-covered area are poor, wake up the energy-saving area; if the evaluation results of the network performance KPIs of the co-covered area are excellent or there is no result, the energy-saving area remains in the energy-saving state; Wherein, the co-covered area is an area whose coverage range is the same as or the overlapping range with the energy-saving area exceeds a preset threshold.

[0011] In a possible implementation, the method further includes: Detecting whether there is a conflict in the current energy-saving policy. If there is a conflict, modify the current energy-saving policy. If there is no conflict, generate an energy-saving interaction instruction based on the current energy-saving policy; Detecting whether there is an omitted conflict in the current energy-saving policy based on the execution result of the energy-saving interaction instruction. If there is an omitted conflict, feedback the conflict detection result of the current energy-saving policy to the policy management platform.

[0012] In a possible implementation, detecting whether there is a conflict in the current energy-saving policy includes: Detect whether there is a semantic conflict in the current energy-saving policy. If so, determine that the energy-saving area does not need to save energy. Otherwise, execute the step of generating an energy-saving interaction instruction based on the current energy-saving policy; Detect whether the energy-saving interaction instruction is correct. If it is incorrect, modify the current energy-saving policy; If the energy-saving interaction instruction is correct, detect whether there is an implicit conflict in the current energy-saving policy. If there is an implicit conflict, modify the current energy-saving policy; If there is no implicit conflict in the current energy-saving policy, detect whether there is an explicit conflict in the current energy-saving policy. If there is no explicit conflict, execute the step of generating an energy-saving interaction instruction based on the current energy-saving policy. If there is an explicit conflict, modify the current energy-saving policy.

[0013] The second aspect of the present application provides a multi-agent 5G base station intelligent energy-saving device based on the fusion of large and small models, including: An initial policy configuration module, configured to identify network deployment scenario information corresponding to the current wireless network and configure an initial energy-saving policy matching the network deployment scenario information; A current policy determination module, configured to screen energy-saving areas in the current wireless network based on the initial energy-saving policy and determine the current energy-saving policy corresponding to the energy-saving areas based on the in-network data corresponding to the energy-saving areas; An energy-saving control module, configured to generate an energy-saving interaction instruction based on the current energy-saving policy and send it to a wireless network device corresponding to the energy-saving area, where the energy-saving interaction instruction is used to control the energy-saving state of the wireless network device; An energy-saving policy update module, configured to iteratively optimize the current energy-saving policy online according to the energy-saving effect corresponding to the energy-saving area and the change trend of key network performance indicators.

[0014] The third aspect of the present application provides an electronic device, including at least one processor and a memory connected to the processor, where: The memory is used to store a computer program; The processor is used to execute the computer program so that the electronic device can implement the multi-agent 5G base station intelligent energy-saving method based on the fusion of large and small models in the first aspect or any implementation manner of the first aspect.

[0015] The fourth aspect of the present application provides a computer storage medium, where the storage medium carries one or more computer programs, and when the one or more computer programs are executed by an electronic device, the electronic device can implement the multi-agent 5G base station intelligent energy-saving method in the first aspect or any implementation manner of the first aspect.

[0016] With the above technical solution, the multi-agent 5G base station intelligent energy-saving method based on the fusion of large and small models provided by this application can automatically identify the network deployment scenario and configure the initial energy-saving strategy corresponding to this scenario through the technology of fusing the large language data model and the generative AI model. Based on the initial energy-saving strategies corresponding to each cell or area, the energy-saving cells are screened out, the energy-saving strategies and energy-saving interaction instructions corresponding to the energy-saving cells are obtained, and the network devices corresponding to the energy-saving cells are controlled to enter the energy-saving state. Further, for the cells that enter the energy-saving state, the corresponding energy-saving strategies are iteratively optimized online according to the change areas of the energy-saving effect and the load performance index, and the balance point between the energy-saving threshold parameter and the network performance is found, so as to achieve maximum energy saving. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In combination with the accompanying drawings and with reference to the following specific embodiments, the above and other features, advantages and aspects of the various embodiments of the present disclosure will become more apparent. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic and the original parts and elements are not necessarily drawn to scale.

[0018] Figure 1 It is a schematic structural diagram of a wireless energy-saving system architecture provided by this application; Figure 2 It is a flowchart of a multi-agent 5G base station intelligent energy-saving method based on the fusion of large and small models provided by this application; Figure 3 It is a schematic flow diagram of another multi-agent 5G base station intelligent energy-saving method based on the fusion of large and small models provided by an embodiment of this application; Figure 4 It is a schematic diagram of the process of establishing and using a knowledge base provided by an embodiment of this application; Figure 5 It is a schematic diagram of online iterative optimization of energy-saving threshold parameters provided by an embodiment of this application; Figure 6 It is a schematic flow diagram of an energy-saving strategy conflict management process provided by an embodiment of this application; Figure 7 It is a schematic structural diagram of a multi-agent 5G base station intelligent energy-saving device based on the fusion of large and small models provided by an embodiment of this application; Figure 8 It is a schematic structural diagram of an electronic device provided by an embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] The following describes the embodiments of this application in combination with the drawings in the embodiments of this application. The terms used in the embodiments part of this application are only used to explain the specific embodiments of this application, and are not intended to limit this application.

[0020] The embodiments of the present application will be described below in conjunction with the accompanying drawings. As is known to those of ordinary skill in the art, with the development of technology and the emergence of new scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0021] Terms such as "first" and "second" in the specification, claims, and the above-mentioned accompanying drawings of the present application are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that such terms can be interchanged under appropriate circumstances, which is only a way of distinguishing objects with the same attributes when describing the embodiments of the present application. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion, so that a process, method, system, product, or device comprising a series of units does not have to be limited to those units, but may include other units that are not clearly listed or are inherent to these processes, methods, products, or devices.

[0022] First, the professional terms involved in the present application will be introduced: 5G NR (New Radio) is a core component of the fifth-generation mobile communication technology, which can provide a wireless access technology with higher speed, lower latency, greater connection density, and higher energy efficiency.

[0023] OFDM (Orthogonal Frequency Division Multiplexing) is a multi-carrier transmission technology. By dividing the channel into several orthogonal sub-channels, a high-speed data signal is converted into parallel low-speed sub-data streams and modulated onto each sub-channel for transmission.

[0024] RRC (Radio Resource Control) is the message configuration center and control center of the access layer of the entire wireless communication protocol stack.

[0025] PRB (Physical Resource Block) is the core resource unit in a wireless communication system and the smallest resource unit allocated to users. A PRB consists of continuous time-frequency resources and contains a specific number of sub-carriers and time slots. PRB utilization rate = the number of PRBs allocated to users / the number of available PRBs. The higher the PRB utilization rate, the more effectively the physical resources are utilized, indicating that the system can better meet the needs of users.

[0026] AAU (Active Antenna Unit) is a device that integrates radio frequency and digital processing functions. Its function is to amplify and process wireless signals and then transmit and receive wireless signals through an antenna.

[0027] The BBU (Baseband Unit) is a component in a wireless base station system and is a device integrating digital signal processing and control functions. The function of the BBU is to process the digital signals received from the AAU and control the operation of the wireless base station system.

[0028] The RRU (Remote Radio Unit) is a component of the wireless base station system and is a device integrating radio frequency functions. Its function is to convert digital signals into radio frequency signals and transmit and receive wireless signals through antennas.

[0029] In the process of researching the solution of this application, the inventor found that: the existing multi-agent 5G base station intelligent energy-saving method based on the fusion of large and small models sets the key index thresholds for entering and exiting the energy-saving state by expert experience. When it is monitored that the load index at the cell level meets the threshold of the energy-saving strategy type, the energy-saving strategy corresponding to the enter / exit energy-saving instructions is sent to the specific energy-saving network element through the network element management system to achieve energy saving of the wireless network. In this way, the key index thresholds cannot be adaptively adjusted after being determined, and moreover, the set key index thresholds are relatively conservative and cannot achieve maximum energy saving.

[0030] To solve the above problems, this application provides a multi-agent 5G base station intelligent energy-saving method based on the fusion of large and small models. This solution continuously iteratively optimizes the energy-saving strategy online according to the big data analysis of the full-scenario traffic model, energy-saving effect, and key network performance indicators (which can be called network performance KPIs), searches for the balance point between the key index thresholds and network performance, and achieves maximum energy saving.

[0031] Please refer to Figure 1 , which shows a schematic structural diagram of an energy-saving system architecture of a wireless communication network provided by an embodiment of this application. This system is an energy-saving system architecture based on the Open Radio Access Network (O-RAN) architecture. As Figure 1 shown, this system mainly completes the interaction with other subsystems in the wireless communication system based on the service management and service orchestration (SMO) framework. Here, other subsystems may include the network functions of O-RAN, the operation platform O-Cloud of O-RAN, the core network, external systems, etc.

[0032] Among them, there are significant differences between the service-based and open SMO framework and the traditional operator network management subsystem: The SMO framework provides various services for the network operation and management of network devices, rather than the actual network management after integration. The specific network operation and management work is completed by various services.

[0033] The SMO framework is an open operation and management platform that can integrate network device products from any manufacturer that complies with the O-RAN specification and can apply network operation and management software from any manufacturer that complies with the O-RAN specification.

[0034] In some embodiments of this application, the SMO framework includes the following functions: Cloud Infrastructure OAM, RAN OAM, and Non-Real Time RAN Intelligent Controller (Non-RTRIC).

[0035] Cloud Infrastructure OAM is responsible for the operation, maintenance, and management of the cloud infrastructure. RAN-OAM is responsible for the operation, maintenance, and management of radio access network elements.

[0036] Non-RTRIC is responsible for the intelligent energy-saving control of network devices. Specifically, an energy-saving application (Energy Saving RAN Application, ESrApp), energy-saving model deployment, and energy-saving model management are deployed inside the Non-RTRIC framework. The energy-saving model training extends outside the Non-RTRIC framework, that is, the energy-saving model training is deployed within the SMO framework, and the energy-saving model is managed within the Non-RTRIC framework. The energy-saving model supports the output of cell-level optimal energy-saving time periods, optimal energy-saving threshold parameters, and energy-saving interaction instructions related to energy saving.

[0037] Among them, ESrApp runs in the Non-Real Time RAN Intelligent Controller (Non-RTRIC) and is responsible for tasks such as network-level energy-saving strategy optimization and long-term data analysis (time span from seconds to hours).

[0038] The energy-saving model of this application is an intelligent energy-saving model based on the integration of a large language model (Large Language Model, ML) and an artificial intelligence (Artificial Intelligence, AI) model.

[0039] Based on the communication between each subsystem in the O-RAN wireless energy-saving system through corresponding interfaces, the following interfaces may be included: O1 interface: A newly added interface between the SMO framework and the internal network elements of O-RAN, used for the SMO to intelligently manage and operate the logical network elements inside O-RAN.

[0040] O2 Interface: It is the interface between the SMO and the O-RAN operation platform (O-Cloud), used for the SMO to intelligently manage and operate each O-RAN network service node running on the O-Cloud.

[0041] The following briefly introduces the overall working process of the Figure 1 energy-saving system shown: ESrApp selects energy-saving cells according to the load prediction results (the load prediction results are obtained by the load prediction model predicting the traffic load of the cell or area in a future period of time, where the traffic load refers to the communication traffic carried per unit time in the communication system and is usually used to measure the utilization efficiency of communication network resources), and sends a request for energy-saving related data of the energy-saving cells to the database. The database responds to this data request and pushes the in-network data (including the data of the current period of the internal network and the external network) to the AI / ML model, enabling the AI / ML model to identify the network deployment scenario based on the in-network data pushed by the database and customize an energy-saving strategy matching this scenario, that is, determine the corresponding optimal energy-saving threshold parameters (i.e., the load thresholds for entering and exiting the energy-saving state) and the optimal energy-saving time period, and generate an energy-saving interaction instruction and provide it to ESrApp. ESrApp performs energy-saving instruction interaction with the O-RAN through the interface between the AI / ML model and the O-RAN via RAN OAM (i.e., the O1 interface in the figure), such as sending the energy-saving strategy and energy-saving instructions to the O-RAN and receiving the energy-saving instruction response information returned by the O-RAN. At the same time, through AI / ML and ESrApp, a real-time quick evaluation is carried out on the cells entering the energy-saving state to continuously iterate and update the energy-saving strategy.

[0042] It can be seen that in the energy-saving system of the wireless communication network provided by this application, the energy-saving application is deployed on the wireless network device nodes. The application deployment does not require a large amount of server hardware and network resources, and at the same time does not require docking with the standard interfaces of different manufacturers, greatly reducing the development and operation costs of the energy-saving device.

[0043] Please refer to Figure 2 , which shows a flowchart of a multi-agent 5G base station intelligent energy-saving method based on the fusion of large and small models provided by an embodiment of this application. This method is applied to the Figure 1 SMO of the system shown, and this method may include the following steps: S101, identify the network deployment scenario information of each cell or area in the current wireless network.

[0044] The energy-saving model can automatically identify the network deployment scenario information of the wireless network where the current cell or area is located, such as network coverage, networking mode, traffic load conditions, etc.

[0045] (1) Network coverage identification The identification of network coverage mainly relies on engineering parameter information and terminal measurement report (MR) data. Through the big data analysis of integrating these two types of information, multi-frequency and multi-mode cell coverage identification is carried out to find the relationship of co-covered cells in the area (including mostly overlapping coverage).

[0046] Engineering parameter information is usually provided by the operator and the data is refined and updated periodically, mainly including information such as site name, longitude and latitude, antenna hanging height and azimuth angle, etc.

[0047] MR is based on the measurement results of wireless terminal devices (such as mobile phones), such as signal strength, signal-to-noise ratio, modem parameters and other information. The statistics of MR data mainly rely on the cell identification, signal strength, and tracking area (TA) carried by the terminal measurement information, and calculate the coverage range of the cell, whether there are scenarios of weak uplink and downlink coverage, etc.

[0048] (2)Network configuration identification With the continuous development and evolution of the network, there are various different networking types in the network: including standalone (SA) / non-standalone (NSA), co-construction and sharing, different hardware device models (including single-mode / mixed-mode for the same hardware), different network networking topologies (such as 4G / 5G sharing the same BBU frame, 4G / 5G using separate BBU frames, etc.).

[0049] The big data model in the energy-saving model distinguishes device types through the configuration identification of base stations and cells, and obtains serving cell information and neighbor cell relationship information from the OMC background data configuration table, which is used to analyze how energy-saving types such as symbol shutdown, channel shutdown, carrier shutdown, and deep sleep of radio frequency modules match with the corresponding configuration information. Among them, OMC, that is, the operation and maintenance center, is a network management system dedicated to monitoring and controlling the operation and performance of communication networks.

[0050] Among them, the serving cell information includes: Resource data: system type, bandwidth, network element name, network element identifier, etc.; Performance data: cell uplink PRB utilization rate, cell downlink PRB utilization rate, number of users in the cell's RRC connected state, etc.; Alarm data: alarm name, alarm impact, alarm time, etc.; Configuration data: power and other related configuration data.

[0051] The neighbor cell relationship information is the neighbor cell list data that has a neighbor cell relationship with this serving cell, and may include resource data such as network system type, bandwidth, network element name, network element identifier, etc.

[0052] The energy-saving model analyzes by combining historical traffic load performance indicators (such as performance indicators related to traffic load like the uplink PRB utilization rate of the cell, the downlink PRB utilization rate of the cell, the number of users in the RRC connected state of the cell, etc.), the analysis results of network configuration, and the network coverage identification results to identify the network deployment scenarios where the current network devices are located, such as including coverage scenarios, networking modes, network traffic load, etc.

[0053] S102, respectively configure initial energy-saving strategies that match the network deployment scenarios of each cell or area.

[0054] The initial energy-saving strategies may include energy-saving types (such as symbol shutdown, channel shutdown, carrier shutdown, or deep sleep of the radio frequency module), energy-saving threshold parameters (traffic load performance indicator thresholds for entering and exiting the energy-saving state), energy-saving time periods, etc.

[0055] Among them, the energy-saving type can be configured according to the network configuration information of the current wireless network (such as serving cell information and neighbor cell relationship information).

[0056] Among them, symbol shutdown is to turn off the transmission of single or multiple symbol periods in the time dimension. A symbol is the basic time unit in the radio frame structure, such as an OFDM symbol in 5G NR. Taking the 5G NR low-frequency system as an example, a radio frame is 10 ms, each radio frame consists of 10 radio sub-frames, each radio sub-frame consists of 2 time slots, each radio sub-frame is 1 ms, and each time slot is 0.5 ms. Each time slot usually consists of 14 symbols. In the actual communication process, the base station is not in the state of maximum traffic at all times, so for the symbols in the sub-frame, not all moments are filled with valid information. At the moment of the symbol period predicted without data transmission, turn off the power amplifier (PA) power supply and the transceiver unit (TRX) switch to reduce the system power consumption, and turn on the PA power supply and the TRX switch in advance at the moment of the symbol period predicted with data transmission to ensure that the service is not affected.

[0057] Channel shutdown is to turn off specific logical or physical channels, such as control channels, traffic channels, or frequency band resource blocks.

[0058] Carrier shutdown is to turn off the entire carrier, that is, a specific frequency bandwidth. For example, in the carrier aggregation scenario, when the load is insufficient, turn off the secondary carrier to reduce energy consumption.

[0059] Deep sleep of the radio frequency module means that when the base station service is in a long-term idle state, the base station radio frequency unit device (AAU) can turn off the power supply of most active devices and enter the sleep state, so as to achieve the purpose of saving energy. This function is especially suitable for 5G base stations because the radio frequency unit devices of 5G base stations contain a large number of active devices and baseband processing units, resulting in a large increase in their no-load power consumption.

[0060] In an exemplary embodiment, the AI algorithm model of the energy-saving model automatically configures an initial energy-saving strategy matching the scenario, which may include the following processes: A1. Determine the idle traffic load level of the area or cell based on the historical traffic load data of the area or cell; A2. Determine the energy-saving threshold parameter based on the idle traffic load level of the area or cell.

[0061] The energy-saving threshold parameter includes the thresholds of the traffic load performance indicators for entering and exiting the energy-saving state; for example, the energy-saving threshold parameters corresponding to different energy-saving types are also different. For example, the energy-saving threshold parameter for symbol shutdown is higher than that for channel shutdown, and the threshold for channel shutdown is higher than that for carrier shutdown.

[0062] A3. Determine the applicable energy-saving time periods for each area or cell respectively according to the energy-saving threshold parameters and traffic load prediction results corresponding to each cell or area.

[0063] In order to make the energy-saving strategy adapt to the traffic load, the load prediction kernel is introduced into the AI algorithm model in the energy-saving model. Based on the statistics of the performance indicators related to the traffic load (such as the utilization rates of uplink and downlink PRBs, the number of RRC users, etc.), and modeling by distinguishing different network deployment scenarios, the time characteristics of weekdays / holidays, etc., the load prediction is completed by using the time series prediction algorithm, driving the timely application of the energy-saving time periods and energy-saving threshold parameters, and optimizing the energy-saving effect while ensuring the network performance.

[0064] The energy-saving time period can be determined according to the time period when the predicted load meets the performance indicator threshold of the energy-saving condition. For example, if it is predicted that the traffic load-related performance indicators of a certain cell meet the threshold value for entering the energy-saving state during the period from t1 to t2, then the energy-saving time period of this cell is determined to be the period from t1 to t2.

[0065] S103. Screen out the energy-saving cells that meet the energy-saving conditions from each cell or area, request the internal and external current network data corresponding to the energy-saving cells to be provided to the energy-saving model, and obtain the current energy-saving strategy corresponding to the energy-saving cells output by the energy-saving model.

[0066] The current energy-saving strategy includes the optimal energy-saving time period, the optimal energy-saving threshold parameter, and the energy-saving interaction instruction.

[0067] In an exemplary embodiment, for any cell or area in the current wireless network, it is judged whether the current time period is included in the range of the energy-saving time period corresponding to this cell or area (the energy-saving time period in the initial energy-saving strategy corresponding to this cell or area). If so, this cell is determined to be an energy-saving cell.

[0068] In another exemplary embodiment, for any cell or area in the current wireless network, the actual measured value of the traffic load of the cell or area is obtained, and it is determined whether the measured value of the traffic load timing is lower than the energy-saving threshold parameter corresponding to the cell or area (e.g., the traffic load threshold value). If it is lower than or equal to, it is determined that the energy-saving condition is met, and access to the energy-saving state is allowed. If it is higher, it is determined that the energy-saving condition is not met and entry into the energy-saving state is not allowed.

[0069] If the current cell or area meets the corresponding energy-saving conditions, the database is requested to provide the internal and external existing network data of the cell or area to the energy-saving model. For example, the internal existing network data may include engineering parameter information and MR data, and the external existing network data may include data from external systems, such as complaint data, large-scale personnel flow information, Internet ticketing and major event information, and network heavy protection information. Thus, the energy-saving model outputs the corresponding energy-saving strategy for the current time period based on the internal and external existing network data corresponding to the cell or area, such as the optimal energy-saving time period, the optimal energy-saving threshold parameter, and the energy-saving interaction instruction, etc. The energy-saving interaction instruction is automatically generated by the energy-saving model according to the northbound command line development requirements of the wireless network device corresponding to the cell or area.

[0070] S104, perform energy-saving instruction interaction with the wireless network device corresponding to the energy-saving cell.

[0071] See Figure 1 , the ESrApp application can obtain the energy-saving strategy and energy-saving interaction instruction output by the energy-saving model, and perform energy-saving instruction interaction through the interface O1 between RAN OAM and O-RAN. That is, the network device corresponding to the energy-saving cell is controlled to enter the energy-saving state through the energy-saving instruction.

[0072] S105, for the cell or area that enters the energy-saving state, online iteratively optimize the energy-saving strategy corresponding to the cell or area according to the change trend of the energy-saving effect and the traffic load performance index.

[0073] The energy-saving strategy can be dynamically optimized according to the network performance KPIs for a period of time (e.g., one day). Specifically, the energy-saving duration can be dynamically optimized according to the KPIs to achieve a balance between the energy-saving duration and the network performance. The energy-saving threshold parameter (the traffic load performance index threshold for entering and exiting the energy-saving state) can also be continuously optimized according to the network performance KPIs.

[0074] The network performance KPIs mainly include key performance indicators related to network quality. For example, they include KPIs such as establishment type, call drop type, handover type, and user experience type, such as network connection rate, network disconnection rate, network handover success rate, network traffic, etc.

[0075] Meanwhile, real-time quick evaluation and guarantee can also be carried out on the communities or regions that enter the energy-saving state, and the entry and exit of the energy-saving state can be updated in real time according to the quick evaluation results.

[0076] The multi-agent 5G base station intelligent energy-saving method based on the fusion of large and small models provided in this embodiment can automatically identify the network deployment scenario and configure the corresponding initial energy-saving strategy through the technology of fusing the large language data model and the generative AI model. Based on the initial energy-saving strategies corresponding to each community or region, energy-saving communities are screened out, the energy-saving strategies and energy-saving interaction instructions corresponding to the energy-saving communities are obtained, and the network devices corresponding to the energy-saving communities are controlled to enter the energy-saving state. Further, for the communities that enter the energy-saving state, the corresponding energy-saving strategies are iteratively optimized online according to the change areas of the energy-saving effect and the load performance index, and the balance point between the energy-saving threshold parameter and the network performance is found, so as to achieve maximum energy saving.

[0077] Please refer to Figure 3 , which shows a schematic flow chart of another multi-agent 5G base station intelligent energy-saving method based on the fusion of large and small models provided in the embodiment of the present application. This method adds the detailed processes of the energy-saving model training process and the energy-saving control process on the basis of the embodiment shown in Figure 2 .

[0078] As Figure 3 shown, this method may include the following steps: S201, construct a knowledge base based on external data.

[0079] In an exemplary embodiment, the process of constructing the knowledge base may include the following process: (1) Collect external data For example, the external data may include documents on the wireless network manufacturer or operator side and data retrieved by the external system API plug-in.

[0080] The documents on the wireless network manufacturer or operator side may include: the description document related to the wireless network element model map (mainly providing knowledge of the wireless network hardware), the interface technical requirements document of the wireless network capacity scheduling subsystem (mainly providing knowledge of the resources, alarms, and performance of the wireless network), the wireless MR description document (mainly providing the network coverage of the wireless network at the cell level), and the northbound command line development guide document (mainly providing operation control instructions for the wireless network element).

[0081] Retrieving plugin data from external system APIs may include: a centralized complaint system (mainly providing information such as the location where complaints occur and the associated radio cells), an Internet traffic analysis system (mainly providing large-scale personnel flow information), Internet ticketing, and major event information (mainly providing information on the attendance of personnel at entertainment, sports, and other related events), and network heavy protection information (heavy protection scenario information and detailed information on base stations or cells included in heavy protection).

[0082] (2)Data Processing Since the model has certain limitations on the size of the input data, after completing the above data collection, it is necessary to split the large document into smaller text chunks, and these text chunks should maintain semantic integrity as much as possible. For example, it can be split according to chapters. The purpose of splitting is to ensure that the size of each text chunk is suitable for model processing while minimizing the loss of context information.

[0083] Furthermore, convert the split text chunks into numerical vectors. This step can be completed by calling a vector conversion server. Specifically, the text data of the text chunks can be sent to the server for vector conversion, and the server returns the vectors corresponding to the text data. These vectors can capture the semantic information of the text, and similar texts are close to each other in the vector space, that is, the vector data obtained by converting semantically similar texts is also similar.

[0084] (3)Result Storage Store the vectors, construct and store a knowledge graph based on the text vectors, and store the original document.

[0085] The knowledge base constructed based on the large language model is capable of efficiently storing a large number of text vectors related to energy conservation, quickly retrieving the text vectors most similar to a given vector, and supporting complex query operations.

[0086] S202, train the energy-saving model using the historical data of the wireless network and the knowledge base.

[0087] The historical data of the wireless network may include: performance data (such as uplink and downlink PRB utilization rates of cells, the number of users in the RRC connected state, etc.), resource data (such as network mode, bandwidth, network element name, network element identifier, etc.), alarm data (such as alarm name, alarm impact, alarm time, etc.), parameter configuration data (such as power and other related configuration data), and MR data (such as signal strength, signal-to-noise ratio, etc.). It is necessary to manually label the policy execution effect labels in the historical data, such as energy-saving efficiency, network quality change indicators, etc., and further construct the mapping relationship between the input (such as network status, such as MR data, external system data, etc.) and output (policy execution effect) of the energy-saving model.

[0088] The energy-saving model is built based on the Retrieval Augmented Generation (RAG) model. Among them, RAG is a model that combines retrieval and generation technologies. It generates answers or content by referring to information in an external knowledge base, and has strong interpretability and customization capabilities. RAG retrieves relevant information from a large-scale document collection and uses this information to guide text generation, thereby improving the quality and accuracy of predictions. The retrieval function obtains relevant information from the external knowledge base according to the user's query content. Enhancement means embedding the user's query content and the retrieved relevant knowledge together into a preset prompt template. Generation means inputting the retrieved and enhanced prompt content into a large language model to generate the required output.

[0089] The knowledge base stores domain knowledge and practical experience in a structured manner, enabling the energy-saving model to break through the limitations of the data-driven paradigm. Specifically, the knowledge base provides professional rules and experience for the energy-saving model and enhances the interpretability of decisions. During the training process of the energy-saving model, the knowledge base can enhance the data quality of the energy-saving model and expand dynamic features; during the iterative optimization process of the energy-saving model, the knowledge base provides corresponding support for the continuous learning of the energy-saving model. In addition, in terms of ensuring energy-saving effects, the knowledge base provides corresponding support for the energy-saving model to achieve policy conflict detection, exception recovery, and effect verification.

[0090] In addition, it is necessary to annotate semantic labels for text paragraphs in the knowledge base, such as power adjustment rules, traffic load prediction methods, etc., so as to improve the prediction accuracy of RAG.

[0091] In an exemplary embodiment, the basic energy-saving model can be trained in an offline pre-training manner. For example, the historical network state data corresponding to the current base station node (such as within the past 3 months) is used to train the basic energy-saving model. Specifically, the historical network state data can be input into the initial model to obtain the corresponding energy-saving strategy and the corresponding strategy execution effect, and the loss value between the predicted strategy execution effect and the actual effect execution effect is calculated according to the loss function. If the loss value is greater than the preset threshold, the model parameters in the basic model are adjusted according to the loss value, and the above process is repeated until the loss value is less than or equal to the preset threshold to obtain the basic prediction model corresponding to each base station node.

[0092] Furthermore, contrastive learning is used to optimize the vector representation of the knowledge base. This process deeply integrates domain knowledge and representation learning, improving the retrieval accuracy of the knowledge base. In the scenario of generating energy-saving strategies, it provides intelligent energy-saving technologies such as load prediction, energy-saving period optimization, energy-saving threshold parameter optimization, and fast evaluation guarantee. Through multi-dimensional intelligent scheduling of energy-saving time, energy-saving threshold parameters, etc., and mutual coordination of multiple frequency layers, the energy-saving capabilities of the base station are fully exploited to maximize the energy-saving effective duration while ensuring user perception.

[0093] Among them, contrastive learning is a self-supervised learning method that learns the surface vector representation of text by comparing sample pairs. The core idea of contrastive learning is that similar samples should be close to each other in the representation space, while dissimilar samples should be far from each other. The process of optimizing the vector representation by contrastive learning can be as follows: (1) Constructing generated samples The samples involved in this embodiment include three types of samples: anchor, positive, and negative. Anchor represents the original text paragraph (such as "AAU deep sleep trigger condition"); Positive represents the manually rewritten synonymous description (such as "radio frequency unit sleep activation rule"); Negative represents similar but irrelevant text (such as "base station heat dissipation system start-stop logic"). Further, a loss function is constructed.

[0094] (2) Constructing the loss function Calculate the loss value between the historical vector representation and the current vector representation corresponding to the same sample according to the loss function. Among them, the historical vector representation refers to the vector representation of the sample obtained through the historical vector conversion model, and the current vector representation is the vector representation of the sample obtained based on the updated vector conversion model. Update the model parameters of the vector representation model according to the loss value.

[0095] (3) HyDE enhanced representation HyDE (Hypothetical Document Embeddings) is a technique for enhancing retrieval performance by generating hypothetical documents. The core idea of HyDE is to generate a hypothetical document that is closer to the embedding space of the target document than the original query, thereby improving retrieval accuracy. Specifically, HyDE first generates a hypothetical document that better captures the intent of the query, and then vectorizes the hypothetical document to generate an embedding representation. Use the vectorized hypothetical document to retrieve similar documents in the database and find the real document closer to the hypothetical document.

[0096] Furthermore, each base station node updates its local model parameters through personalized federated instruction tuning (PFIT), and dynamic weighted averaging is adopted during global model aggregation to adapt to different network load scenarios.

[0097] (1) Model aggregation under the federated learning framework: By designing a dynamic weighted averaging algorithm model, the weight calculation is based on the number of network users, network traffic, resource utilization entropy value, and user service perception quality score of the base station node to construct a dynamic weighted averaging algorithm model. For heterogeneous base station devices, the feature space mapping technology is used to unify the model input uniquely, and the knowledge distillation is used to compress the model parameter differences, and the core energy-saving decision logic is retained to achieve parameter alignment.

[0098] (2)Network energy-saving adaptation solution Build a real-time adaptation engine through a two-layer decision-making mechanism and a load prediction module. The first-layer decision-making mechanism: a fast energy-saving response layer based on a rule engine. The second-layer decision-making mechanism: generate a refined policy through model inference. The load prediction module uses an LSTM+Attention model to achieve load prediction for the next hour. Output a network energy-saving adaptation solution through the real-time adaptation engine combined with the scenario feature library.

[0099] Among them, LSTM is a long short-term memory network. Its core idea is to introduce a long-term memory unit (CellState) to store long-term information, and selectively update or delete information through a set of carefully designed gating mechanisms (including input gate, forget gate, and output gate). These gating mechanisms can control the inflow and outflow of information from the cell state, thereby achieving effective learning of long-term dependencies. The Attention model is a mechanism that enables a neural network to focus on key information when processing sequential data by dynamically allocating weights.

[0100] Such as Figure 1 shown, the trained energy-saving model is deployed in the Non-RTRIC framework of the SMO framework. In addition, an energy-saving model management module is also deployed in the Non-RTRIC framework to manage the energy-saving model. For example, update the energy-saving policy of the energy-saving model.

[0101] Please refer to Figure 4 , which shows a schematic diagram of the process of establishing and using a knowledge base provided by an embodiment of the present application. The collected external data is preprocessed to obtain a document object, the document object is segmented to obtain the segmented text paragraphs (i.e., text blocks), and further the text paragraphs are converted into corresponding vectors, and corresponding indexes are constructed for the text vectors to generate a professional knowledge base.

[0102] When receiving a wireless energy-saving problem, convert the wireless energy-saving problem into a corresponding energy-saving problem vector, retrieve the top k relevant text vectors matching the energy-saving problem vector from the knowledge base, and combine the problem vector and the relevant text vectors in the knowledge base into a context. Call the large language model in the energy-saving model to return results related to the problem. Among them, the returned results may be operations without instruction interaction, or operations with instruction interaction.

[0103] S203, ESrApp screens energy-saving communities and requests the database to push the required internal and external in-network data to the energy-saving model.

[0104] After the ESrApp selects the community that needs energy conservation, it notifies the database to push the in-network and out-of-network data of this community to the energy conservation model, such as MR data and data from external systems, such as complaint data, large-scale personnel flow information, Internet ticketing and major event information, and network heavy protection information.

[0105] S204, the ESrApp obtains the energy conservation strategy of the energy conservation community output by the energy conservation model.

[0106] Such as Figure 1 As shown, the ESrApp subscribes to the energy conservation strategy of the energy conservation model through the R1 interface. The energy conservation strategy may include the optimal energy conservation period, the optimal energy conservation threshold parameter, and the energy conservation interaction instruction.

[0107] (1) Optimal energy conservation period In an exemplary embodiment, the optimal energy conservation period can be determined according to the predicted load corresponding to each prediction time granularity (for example, 15 minutes) and the energy conservation threshold parameter corresponding to this energy conservation community. Taking the energy conservation type of carrier shutdown as an example, the process of determining the optimal time period for carrier shutdown in the energy conservation community is as follows: The predicted average number of RRC users of the carrier ≤ the carrier-level UE quantity threshold for carrier shutdown; The predicted usage rate of uplink / downlink PRBs on the carrier ≤ the uplink / downlink PRB usage rate load threshold for carrier shutdown; The predicted usage rate of uplink / downlink PRBs in the energy conservation community group ≤ the uplink / downlink PRB usage rate wake-up threshold of all basic coverage communities; where the energy conservation community group includes the currently selected energy conservation communities and co-coverage communities (communities with the same or mostly overlapping coverage areas as the energy conservation communities). All basic coverage communities are the energy conservation communities and all their co-coverage communities.

[0108] If all the above indicators corresponding to the same time granularity simultaneously meet the corresponding conditions, it is considered that the predicted load of this granularity meets the carrier shutdown condition. Further, filter out the time periods of at least 2 consecutive time granularities or more from the original energy conservation available time periods as the final carrier shutdown available energy conservation time. At the same time, select at least 4 consecutive time granularities or more as the final available energy conservation time for deep sleep.

[0109] In the case where deep sleep cannot meet the operator's requirement for extreme energy conservation, in order to achieve a more extreme energy conservation effect, when the traffic load is low, the AAU / RRU can be directly powered off. This energy conservation type is called automatic start-stop energy conservation. The automatic start-stop of the AAU / RRU will power off the entire AAU / RRU (except the power software module). The energy conservation entry condition is the same as that of carrier shutdown / deep sleep in the 5GNR system, and the energy conservation wake-up only supports timed wake-up.

[0110] For automatic start-stop, if the traffic load performance index after conversion needs to be considered, the conversion method is as follows: when the energy-saving threshold parameter corresponding to carrier shutdown is lower than the preset value, the original energy-saving threshold parameter remains unchanged without conversion; when the energy-saving threshold parameter corresponding to carrier shutdown is higher than the preset threshold, the original energy-saving threshold parameter is reduced by a certain value as the automatic start-stop threshold. The period with at least 9 consecutive event granularities of traffic load lower than the automatic start-stop threshold is the final energy-saving period for automatic start-stop.

[0111] For the channel shutdown method, only the energy-saving threshold parameter of the primary serving cell is considered, and the load conditions of any neighboring cells do not need to be concerned. Filter out at least two consecutive time-granularity time periods, and then take the intersection of the energy-saving periods of the carrier channels in all co-antenna groups as the final energy-saving period for channel shutdown.

[0112] (2)Optimal energy-saving threshold parameter For energy-saving, the higher the threshold value for entering the shutdown state, the better the energy-saving effect, that is, it is easier to enter the energy-saving state. However, in traditional energy-saving schemes, in order to take into account the differences in various scenarios, the set threshold value is relatively conservative, that is, the threshold value is low, which means it is more difficult to enter the energy-saving state, resulting in a poor energy-saving effect.

[0113] In an exemplary embodiment, a traffic load performance index rollback self-optimization strategy is adopted to obtain the optimal energy-saving threshold parameter, that is, according to the full-scenario traffic load prediction model, the energy-saving effect and the change trend of network performance KPIs are used to continuously iteratively optimize the energy-saving threshold parameter online. Specifically, it can include the following processes: Extract the corresponding performance index data for each cell periodically (for example, once a day), use the clustering algorithm to find the optimal adjustment step for different threshold parameters, and adjust the corresponding energy-saving threshold parameter according to the optimal adjustment step. After the energy-saving threshold parameter is optimized and refreshed, monitor the network performance KPIs of the wireless network, and continuously iterate the energy-saving threshold parameter within the allowable floating range until the best balance point between energy-saving and network performance is reached, that is, the optimal energy-saving threshold parameter.

[0114] Please refer to Figure 5 , after iteratively optimizing the threshold parameter, determine whether the network performance KPI has declined. If so, continue to determine whether the network performance KPI exceeds the allowable threshold. If it does not exceed the allowable threshold, continue to iteratively optimize the threshold parameter; if it exceeds the allowable threshold, roll back the threshold parameter to the value before iterative optimization, and continue to determine whether the network performance KPI has declined. If the network performance KPI has not declined, continue to iteratively optimize the threshold parameter until the best balance point between energy-saving and performance is reached.

[0115] In addition, the above predicted load is obtained through the load prediction algorithm in the energy-saving model.

[0116] In an exemplary embodiment, when modeling the load prediction algorithm, positive-effect cells, negative-effect cells, and non-effect cells are distinguished based on historical load data. The subsequence splitting prediction method for the same day within a cycle is adopted, and the influence of holiday factors on prediction indicators is combined. Using the second-order exponential smoothing prediction algorithm, a prediction model with the optimal calculation performance and the best optimization effect is obtained.

[0117] Cell scenario division: A positive-effect cell refers to a cell where the performance index value increases at the start of a holiday and decreases at the end of the holiday; a non-effect cell refers to a cell where there is no obvious change in the performance index during the holiday; a negative-effect cell refers to a cell where the performance index value decreases at the start of the holiday and increases at the end of the holiday.

[0118] By comprehensively modeling holiday factors, load development trends, and periodicity, the accuracy of cell load prediction can be significantly improved. Holiday effects and periodic changes are very key features. Incorporating trend modeling can help capture long-term change trends, and combining machine learning or deep learning methods can more comprehensively predict cell loads.

[0119] At the same time, for some cells with strong load burst characteristics, the prediction results are probabilistically relatively poor. In this case, consider screening energy-saving cells from the aspects of load stability and load volatility, and perform smoothing processing on the data.

[0120] According to the performance indicators related to the load, cell scenarios are divided. Except for the scenarios where the load indicators soar / drop sharply, the accuracy of cell-level traffic load prediction is significantly improved, reaching over 94.5%.

[0121] S205, ESrApp performs interaction of energy-saving instructions through the O1 interface between RAN OAM and O-RAN.

[0122] S206, ESrApp performs real-time quick evaluation and guarantee for energy-saving cells, and updates the energy-saving status according to the quick evaluation results.

[0123] In an exemplary embodiment, the real-time quick evaluation and guarantee for updating the energy-saving status may include the following situations: In the pre-energy-saving stage, trigger the real-time network performance KPI quick evaluation of energy-saving cells and co-covered cells (cells whose coverage ranges are the same as or mostly overlap with that of the energy-saving cells), and use the network performance KPI data in the historical time period (such as the previous 2 hours) for evaluation. If the evaluation result of the network performance KPI of the energy-saving cell or the co-covered cell is poor, prevent the energy-saving cell from entering the energy-saving state; if the evaluation results of the energy-saving cell and the co-covered cell are excellent or there is no result, allow the energy-saving cell to enter the energy-saving state.

[0124] During the energy-saving preparation stage, trigger the quick evaluation of the network performance KPIs of the energy-saving cell and the co-covered cell, and use the network performance KPI data in the energy-saving preparation stage for evaluation. If the evaluation result of the energy-saving cell or the co-covered cell is poor, prevent the energy-saving cell from entering the energy-saving state; if the evaluation results of the energy-saving cell and the co-covered cell are excellent or there is no result, allow the energy-saving cell to enter the energy-saving state.

[0125] During the energy-saving stage, trigger the periodic quick evaluation of the co-covered cell. The co-covered cell performs the network performance KPI evaluation for a specified period (such as 1 minute). If the network performance KPI evaluation result of the co-covered cell is poor, wake up the energy-saving cell; if the evaluation result of the co-covered cell is excellent or there is no result, do not wake up the energy-saving cell.

[0126] The multi-agent 5G base station intelligent energy-saving method based on the fusion of large and small models provided in this embodiment uploads relevant documents such as the model map of wireless network devices to the knowledge base through the capabilities of the large language model. The large language model autonomously learns the energy-saving strategies supported by the models and automatically generates corresponding energy-saving instructions for entering and exiting energy-saving, greatly shortening the development and operation and maintenance costs of energy-saving products, and avoiding the problem of energy-saving accidents affecting user perception due to untimely update of instructions with the iterative upgrade of wireless devices. Moreover, the energy-saving model based on the large language model can effectively combine sudden abnormal events that affect user perception, such as complaints and sudden high-load events, in a timely manner, and adjust the energy-saving strategy in a timely manner to achieve the balance between energy-saving effect and wireless perception.

[0127] In addition, considering that the mobile communication network is a highly coupled complex system, the users of the energy-saving system may know little about the internal details of the system, and at the same time, the requirements for energy-saving strategies may come from different network departments. Therefore, conflicts between energy-saving strategies are inevitable. This application also provides an energy-saving strategy conflict management based on cognitive enhancement, which manages energy-saving strategies by identifying semantic conflicts, explicit conflicts, implicit conflicts, and omission conflicts through the large language model. As Figure 6 shown, this management method further includes the following steps on the basis of Figure 4 : S301, detect whether there is a semantic conflict in the energy-saving strategy; if not, execute S302; if so, determine that the cell does not need energy-saving.

[0128] If the objects of multiple energy-saving strategies have an intersection, and there are mutual exclusions, repetitions, or inclusions in the energy-saving intention semantics, it is determined that there is a semantic conflict. Semantic conflicts can be detected through historical energy-saving strategies and rules defined based on expert experience. When a semantic conflict in the energy-saving strategy is detected, the system will not execute the energy-saving strategy and feedback the failure of creating the energy-saving strategy to the initiator of the energy-saving strategy.

[0129] For example, when multiple sites need to enable the energy-saving function in combination and the energy-saving threshold parameters need to be automatically optimized, and there is a mutual exclusion between the energy-saving duration and the performance indicators, it is determined that there is a semantic conflict.

[0130] S302. Determine whether the energy-saving policy is correct; if it is correct, execute S303; if it is incorrect, modify the energy-saving policy.

[0131] S303. Detect whether there is an implicit conflict in the energy-saving policy; if not, execute S304; if so, modify the energy-saving policy.

[0132] An implicit conflict in the energy-saving policy means that although the energy-saving parameter values required by multiple energy-saving policies are not mutually exclusive, it is highly likely that they cannot be achieved simultaneously in a specific scenario. For the implicit conflict of the energy-saving policy, a machine learning algorithm can be used for identification, or the digital twin platform can be used to issue the intentions of these energy-saving policies simultaneously in the twin target network to identify the conflict. If an implicit conflict is detected, the system will not execute the energy-saving policy and will feedback to the initiator of the energy-saving policy that the intention creation / activation / modification fails.

[0133] For example, when multiple sites need to enable the energy-saving function in combination and the energy-saving threshold parameters need to be automatically optimized, but the same coverage cell list is included in other scenarios such as important road scenarios, key guarantee scenarios (important meetings), etc.

[0134] S304. Detect whether there is an explicit conflict in the energy-saving policy; if not, execute S305; if so, modify the energy-saving policy.

[0135] An explicit conflict in the energy-saving policy means that there is an intersection among multiple energy-saving policy objects and the affected parameters, and the required value ranges of the intersecting parameters are mutually exclusive. The key to detecting the explicit conflict of the energy-saving policy lies in the system's ability to accurately judge the parameter value ranges required for each energy-saving policy to be achieved. After detecting an explicit conflict in the energy-saving policy, the system will not execute the energy-saving policy and will feedback to the initiator of the energy-saving policy that the creation / activation / modification of the energy-saving policy fails.

[0136] For example, when multiple sites need to enable the energy-saving function in combination and the energy-saving threshold parameters need to be automatically optimized, the influence factors of neighboring cells need to be considered. If there are cells that need to judge the carrier load of the neighboring cell relationship, and the same cell is a neighboring cell of different primary serving cells, and the A primary serving cell judges that it can save energy while the B cell judges that it cannot save energy, there is a mutual exclusion in the required value ranges of the parameters at this time.

[0137] S305. Generate an energy-saving interaction instruction.

[0138] S306. Execute the energy-saving interaction instruction.

[0139] Send the energy-saving interaction instruction to the network device corresponding to the energy-saving cell for execution.

[0140] S307. Detect whether there are any missing conflicts in the energy-saving policy based on the execution result of the energy-saving interaction instruction; if so, escalate to manual assistance; if not, continue to detect whether there are any missing conflicts.

[0141] A missing conflict means that ESrApp discovers conflicts among multiple executed energy-saving policies through real-time monitoring. Missing conflicts are mainly due to immature semantic, explicit, and implicit conflict technologies, or incomplete energy-saving policies caused by unexpected events. After discovering a missing conflict, it is escalated to the energy-saving policy management that coordinates the large and small models or escalated to manual assistance for resolution. The energy-saving policy can be updated and adjusted in real time, or a certain energy-saving intention can be prioritized to eliminate the conflict.

[0142] For example, when multiple sites need to combine to enable the energy-saving function and require automatic optimization of the energy-saving threshold parameters, and there are sudden dense human activities around a community that has entered the energy-saving state, such as: temporary aggregation of live broadcast users, etc.

[0143] Corresponding to the above-described embodiment of the multi-agent 5G base station intelligent energy-saving method based on the integration of large and small models, the present application also provides an embodiment of a multi-agent 5G base station intelligent energy-saving device based on the integration of large and small models.

[0144] As Figure 7 shown, the multi-agent 5G base station intelligent energy-saving device provided in this embodiment based on the integration of large and small models includes: An initial policy configuration module 101, configured to identify network deployment scenario information corresponding to the current wireless network and configure an initial energy-saving policy that matches the network deployment scenario information.

[0145] A current policy determination module 102, configured to screen energy-saving areas in the current wireless network based on the initial energy-saving policy and determine the current energy-saving policy corresponding to the energy-saving areas based on the in-network data corresponding to the energy-saving areas.

[0146] An energy-saving control module 103, configured to generate an energy-saving interaction instruction based on the current energy-saving policy and send it to the wireless network device corresponding to the energy-saving area, where the energy-saving interaction instruction is used to control the energy-saving state of the wireless network device.

[0147] An energy-saving policy update module 104, configured to iteratively optimize the current energy-saving policy online according to the energy-saving effect corresponding to the energy-saving area and the change trend of the key network performance indicators.

[0148] Please refer to Figure 8 , the present application also provides an embodiment of an electronic device. The electronic device includes a bus 201, a processor 202, a communication interface 203, and a memory 204. The processor 202, the memory 204, and the communication interface 203 communicate with each other through the bus 201.

[0149] The bus 201 can be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience in representation, only a thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0150] The processor 202 can be any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP), etc.

[0151] The memory 204 can include volatile memory, such as random access memory (RAM). The memory 204 can also include non-volatile memory, such as read-only memory (ROM), flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0152] Among them, the memory 204 can be used to store software codes related to the multi-agent 5G base station intelligent energy-saving method based on size model fusion provided by this application. The processor 202 can execute the steps of the method in the memory, or can also schedule other units to implement corresponding functions.

[0153] An embodiment of this application also provides a computer program product, including computer-readable instructions. When the computer-readable instructions run on an electronic device, the electronic device is enabled to implement any one of the multi-agent 5G base station intelligent energy-saving methods based on size model fusion provided by the embodiments of this application.

[0154] An embodiment of this application also provides a computer storage medium. The storage medium carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can be enabled to implement any one of the multi-agent 5G base station intelligent energy-saving methods based on size model fusion provided by the embodiments of this application.

[0155] It should be further noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the attached drawings of the device embodiments provided in this application, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines.

[0156] Through the description of the above embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general hardware. Of course, it can also be implemented by dedicated hardware, including application-specific integrated circuits, dedicated CPUs, dedicated memories, dedicated components, etc. Generally, functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structures used to implement the same function can also be various, such as analog circuits, digital circuits, or dedicated circuits. However, for this application, software program implementation is a better implementation method in more cases. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disc of a computer, and includes several instructions to enable a computer device (which can be a personal computer, training device, or network device, etc.) to execute the methods described in various embodiments of this application.

[0157] In the above embodiments, it can be implemented in whole or in part through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product.

[0158] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (such as coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that includes one or more integrated available media. The available medium may be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

Claims

1. A multi-agent 5G base station intelligent energy-saving method based on large and small model fusion, characterized in that: include: Identify network deployment scenario information corresponding to the current wireless network, and configure an initial energy-saving strategy that matches the network deployment scenario information; Filtering the energy-saving area in the current wireless network based on the initial energy-saving strategy, and determining the current energy-saving strategy corresponding to the energy-saving area based on the existing network data corresponding to the energy-saving area; Generate an energy-saving interaction instruction based on the current energy-saving strategy and send it to the wireless network device corresponding to the energy-saving area, wherein the energy-saving interaction instruction is used to control the energy-saving state of the wireless network device; According to the energy-saving effect corresponding to the energy-saving area and the change trend of the key indicators of network performance, the current energy-saving strategy is optimized through online iteration.

2. The method according to claim 1, characterized in that The network deployment scenario information includes network coverage information and network configuration information, and the initial energy-saving strategy includes energy-saving type, energy-saving period and energy-saving threshold parameters; The configuration of the initial energy-saving strategy matching the network deployment scenario information includes: Configuring an energy saving type of the current wireless network based on the network configuration information of the current wireless network, the energy saving type including symbol shutdown, channel shutdown, carrier shutdown or deep sleep of a radio frequency module; In combination with the historical traffic load performance index and network coverage information corresponding to the current wireless network, the energy-saving time period and energy-saving threshold parameters corresponding to each area of ​​the current wireless network are configured.

3. The method according to claim 2, characterized in that The configuring the energy-saving period and energy-saving threshold parameters corresponding to each area of ​​the current wireless network in combination with the historical traffic load performance indicator and network coverage information corresponding to the current wireless network includes: For any area of ​​the current wireless network, determining an idle time traffic load level based on a historical traffic load of the any area; Determine the energy saving threshold parameter corresponding to any one of the areas based on the idle time traffic load level of any one of the areas; The energy-saving time period corresponding to any one of the areas is determined according to the traffic load prediction result corresponding to any one of the areas and the energy-saving threshold parameter.

4. The method according to any one of claims 1 to 3, characterized in that: The initial energy-saving strategy includes an energy-saving threshold parameter, and the energy-saving threshold parameter includes a traffic load threshold value or an energy-saving time period for entering an energy-saving state; wherein, selecting an energy-saving area from the current wireless network based on the initial energy-saving strategy includes: Selecting an area where the predicted traffic load value is lower than the corresponding traffic load threshold value for entering energy saving as an energy saving area; or, The area in which the current period is in the corresponding energy-saving time period is selected as the energy-saving area.

5. The method according to any one of claims 1 to 3, characterized in that: The determining the current energy-saving strategy corresponding to the energy-saving area based on the existing network data corresponding to the energy-saving area includes: Acquire internal existing network data and external existing network data corresponding to the energy-saving area, wherein the internal existing network data includes engineering parameter information and terminal measurement report data, and the external existing network data includes external system data that affects the traffic load performance index; Based on the internal existing network data and the external existing network data, using a load prediction model to obtain a predicted value of a traffic load performance indicator corresponding to the energy-saving area; Determining an optimal energy-saving period corresponding to the energy-saving area based on the predicted value of the traffic load performance indicator and the energy-saving threshold parameter corresponding to the energy-saving area; A current energy-saving strategy is obtained based on the energy-saving type and the optimal energy-saving period corresponding to the energy-saving area.

6. The method according to claim 1, characterized in that The online iterative optimization of the current energy-saving strategy according to the energy-saving effect corresponding to the energy-saving area and the change trend of the key network performance indicators includes: The current energy-saving threshold parameter corresponding to the energy-saving area is increased by a target adjustment step size; Performing energy-saving control on the energy-saving area based on the energy-saving threshold parameter after increasing the target adjustment step, and obtaining corresponding key network performance indicators; If the key network performance indicator does not decrease or does not exceed the allowable threshold after decreasing, continue to execute the step of increasing the target adjustment step size of the current energy-saving threshold parameter corresponding to the energy-saving area; If the key network performance indicator decreases and exceeds the allowable threshold, the energy-saving threshold parameter is restored to the value before increasing the target adjustment step size to obtain the optimal energy-saving threshold parameter.

7. The method according to claim 1 or 6, characterized in that: The method further comprises: When the energy-saving area has not entered the energy-saving state, the network performance KPI of the energy-saving area and the area with the same coverage is evaluated. If the evaluation result of the network performance KPI of the energy-saving area and the area with the same coverage is poor, the energy-saving area is prevented from entering the energy-saving state; if the evaluation result of the network performance KPI of the energy-saving area and the area with the same coverage is excellent or no result, the energy-saving area is allowed to enter the energy-saving state; When the energy-saving area has entered the energy-saving state, the network performance KPI of the same coverage area as the energy-saving area is evaluated, and if the evaluation result of the network performance KPI of the same coverage area is poor, the energy-saving area is awakened; if the evaluation result of the network performance KPI of the same coverage area is excellent or no result, the energy-saving area remains in the energy-saving state; The same coverage area is an area whose coverage is the same as that of the energy-saving area or whose overlapping range exceeds a preset threshold.

8. The method according to claim 1, characterized in that: The method further comprises: Detecting whether there is a conflict in the current energy-saving strategy, if there is a conflict, modifying the current energy-saving strategy, and if there is no conflict, generating an energy-saving interaction instruction based on the current energy-saving strategy; Based on the execution result of the energy-saving interaction instruction, it is detected whether the current energy-saving strategy has an omission conflict, and if there is an omission conflict, the conflict detection result of the current energy-saving strategy is fed back to the strategy management platform.

9. The method according to claim 8, characterized in that Detecting whether the current energy-saving strategy has a conflict includes: Detect whether there is a semantic conflict in the current energy-saving strategy, if so, determine that the energy-saving area does not need to save energy, if not, execute the step of generating an energy-saving interaction instruction based on the current energy-saving strategy; Detecting whether the energy-saving interaction instruction is correct, and if not correct, modifying the current energy-saving strategy; If the energy-saving interaction instruction is correct, detecting whether there is an implicit conflict in the current energy-saving strategy, and modifying the current energy-saving strategy if there is an implicit conflict; If the current energy-saving strategy does not have an implicit conflict, then detect whether the current energy-saving strategy has an explicit conflict. If there is no explicit conflict, execute the step of generating an energy-saving interaction instruction based on the current energy-saving strategy. If there is an explicit conflict, modify the current energy-saving strategy.

10. A multi-agent 5G base station intelligent energy-saving device based on large and small model fusion, characterized in that: include: An initial policy configuration module, used to identify network deployment scenario information corresponding to the current wireless network, and configure an initial energy-saving policy that matches the network deployment scenario information; A current strategy determination module, configured to screen the energy-saving area in the current wireless network based on the initial energy-saving strategy, and determine the current energy-saving strategy corresponding to the energy-saving area based on the existing network data corresponding to the energy-saving area; An energy-saving control module, used for generating an energy-saving interaction instruction based on the current energy-saving strategy and sending the energy-saving interaction instruction to the wireless network device corresponding to the energy-saving area, wherein the energy-saving interaction instruction is used for controlling the energy-saving state of the wireless network device; The energy-saving strategy updating module is used to iteratively optimize the current energy-saving strategy online according to the energy-saving effect corresponding to the energy-saving area and the changing trend of the key indicators of network performance.

11. An electronic device, characterized in that: The method comprises at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program so that the electronic device can implement the multi-agent 5G base station intelligent energy saving method based on large and small model fusion as described in any one of claims 1 to 9.

12. A computer storage medium, characterized in that: The storage medium carries one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the multi-agent 5G base station intelligent energy-saving method based on large and small model fusion as described in any one of claims 1 to 9.

Images