5G network intelligent switching management method based on artificial intelligence

Abstract

The invention provides a 5G network intelligent switching management method based on artificial intelligence, and the method comprises the following steps: collecting and processing network related data, and obtaining a network data set; constructing a switching demand prediction model used for predicting network switching time and a target and a switching effect prediction model used for predicting network performance after switching; optimizing the initial switching strategy in combination with reinforcement learning to obtain an optimal switching strategy; deploying a network switching engine model on the edge computing node, and adjusting an optimal switching strategy in real time according to the network environment change; a virtual simulation digital environment is constructed for network simulation, an optimal switching strategy is evaluated according to the simulated virtual network environment, and then a network switching engine model is optimized in real time according to an evaluation result. According to the invention, the intelligent switching decision is realized, the precision and efficiency of the switching decision can be improved, the network adaptability and diversified requirements are enhanced, and the user experience is improved.

Description

Technical Field

[0001] The present invention relates to the field of network management technology, and in particular to a 5G network intelligent switching management method based on artificial intelligence. Background Art

[0002] In 5G networks, handover management is a key mechanism to ensure that user equipment (UE) maintains continuous connection when moving between different base stations. Existing handover management methods include but are not limited to: 1) When the user equipment detects that the signal strength of the neighboring base station exceeds the signal strength of the current base station, the handover is triggered. But this may lead to a ping-pong effect, that is, the device frequently switches between two base stations, affecting the stability of the connection. 2) Switching is performed based on the distance between the user equipment and the base station, usually combined with the signal strength. However, considering only the distance may not be accurate because the signal quality is also affected by other factors (such as interference and obstacles). 3) Suitable for high-speed moving user equipment, the switching timing is optimized by predicting its moving path and speed. The process requires accurate speed and path prediction, and errors will cause unnecessary switching or delays. Summary of the invention

[0003] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a 5G network intelligent switching management method based on artificial intelligence, which not only realizes intelligent switching decisions, but also improves the accuracy and efficiency of switching decisions, enhances network adaptability and diversified needs, and improves user experience.

[0004] To achieve the above object, the present invention provides the following solution: a 5G network intelligent switching management method based on artificial intelligence, comprising the following steps:

[0005] Collecting network-related data and preprocessing the network-related data to obtain a network data set;

[0006] Using the network data set to construct a switching demand prediction model for predicting network switching time and target and a switching effect prediction model for predicting network performance after switching;

[0007] Formulate an initial switching strategy based on the constructed model, and optimize the initial switching strategy in combination with reinforcement learning to obtain an optimal switching strategy;

[0008] Deploy a network switching engine model on the edge computing node and adjust the optimal switching strategy in real time according to changes in the network environment;

[0009] A virtual simulation digital environment is constructed to perform network simulation, the optimal switching strategy is evaluated according to the simulated virtual network environment, and then the network switching engine model is optimized in real time according to the evaluation result.

[0010] Optionally, the network-related data includes channel quality indicators, mobility data, network topology data, historical switching data, user behavior data and environmental data;

[0011] Among them, the channel quality indicator is obtained from the channel quality information reported by the user equipment, the mobile data includes user location, speed, acceleration and moving direction, the network topology data includes base station location, coverage, adjacency and base station load, the historical switching data includes switching events, switching success rate, switching failure reasons and switching delay, the user behavior data includes user application type, traffic pattern, service quality requirements, and the environmental data includes weather conditions, geographical obstacle information and road traffic information.

[0012] Optionally, collecting network-related data and preprocessing the network-related data to obtain a network data set includes:

[0013] Collect network-related data, perform denoising, anomaly detection and missing data filling on the network data to obtain an initial data set;

[0014] Dividing the initial data set into two types: continuous values ​​and discrete values, selecting different data standardization methods for different data types, and completing the standardization operation of the initial data set;

[0015] Based on the initial data set after the normalization operation, the time series characteristics of the signal strength are calculated, the periodic pattern of the time series characteristics is identified, and the time series analysis results are obtained. Based on the time series analysis results, the interaction characteristics and statistical characteristics of the user equipment and the network base station are constructed, and then the feature selection algorithm is used to extract the features to obtain relevant feature data;

[0016] The switching events in the initial data set are identified in combination with the relevant feature data, and the identified switching events are classified and labeled to complete the preprocessing of the network-related data.

[0017] Optionally, using the network data set to construct a switching demand prediction model for predicting network switching time and target includes:

[0018] Extracting current and historical signal quality, user equipment mobility mode, base station status and environmental information from the relevant feature data, and arranging the extracted data into a time series data structure to obtain a first training data set;

[0019] Using an LSTM network to capture the time dependency of channel quality, user mobility, and environmental characteristics to obtain a time series model, and then using a sliding window technique to input the first training data set into the time series model for training, learn and predict the future state of network use, and determine the timing of network switching;

[0020] Building a graph neural network model based on the network topology structure, inputting the first training data set into the graph neural network model, and obtaining the status and relationship changes of neighboring base stations; wherein each node in the graph neural network model represents a base station, and each edge represents a connection relationship between base stations;

[0021] The time series model and the graph neural network model are deployed and integrated to obtain a switching demand prediction model, and the switching demand prediction model is used to predict real-time data to obtain the network switching time and the switching target base station.

[0022] Optionally, using the network data set to construct a handover effect prediction model for predicting post-handover network performance includes:

[0023] Extracting base station features, connection device requirements, and overall network status from the relevant feature data, and integrating the extracted data into a second training data set;

[0024] Constructing a multi-layer perceptron, introducing residual connection and batch normalization into the multi-layer perceptron to obtain a deep neural network, training the deep neural network using the second training data set, and integrating a random forest and a gradient boosting tree into the deep learning network to obtain a deep learning model;

[0025] The deep learning model is trained using the network performance data after switching to predict and evaluate the performance indicator trend of the network after switching.

[0026] Optionally, an initial switching strategy is formulated based on the constructed prediction model, and the initial switching strategy is optimized in combination with reinforcement learning to obtain an optimal switching strategy, including:

[0027] Acquire and integrate the current network state, user state and environment state to obtain state information, encode the state information into a feature vector of fixed length, obtain state representation, and complete state definition;

[0028] Define the discrete action space and continuous action space after network switching, and define the immediate reward and long-term reward after network switching;

[0029] Integrate the defined states, actions, and rewards to get a reinforcement learning framework;

[0030] Continuously training the reinforcement learning framework using the initial data set to obtain a strategy model, and optimizing the initial switching strategy output by the prediction model using the strategy model to obtain an optimal switching strategy;

[0031] The policy model is deployed in access network and core network nodes to perform real-time online decision making.

[0032] Optionally, a network switching engine model is deployed on the edge computing node, and the optimal switching strategy is adjusted in real time according to changes in the network environment, including:

[0033] Selecting a lightweight model, and performing model compression and containerization operations on the lightweight model to obtain a network switching engine model;

[0034] The initial data set is input into the network switching engine model, and the network switching operation priority and target base station recommendation are output to perform real-time adjustment of the optimal switching strategy.

[0035] Optionally, constructing a virtual simulated digital environment to perform network simulation, evaluating the optimal switching strategy according to the simulated virtual network environment, and then optimizing the network switching engine model in real time according to the evaluation result, including:

[0036] Use physical simulation tools to simulate network infrastructure, signal interference, and user mobility, and then use simulation tools to integrate real user behavior patterns to perform network simulation;

[0037] Selecting a digital twin as a strategy testing platform, and introducing a multi-dimensional simulation matrix into the strategy testing platform to evaluate the optimal switching strategy;

[0038] The network switching engine model is calibrated and the optimal switching strategy is adjusted according to the evaluation result of the optimal switching strategy.

[0039] The present invention discloses the following technical effects by providing a 5G network intelligent switching management method based on artificial intelligence:

[0040] 1. High accuracy of handover decision: By building a handover demand prediction model, it is possible to predict when handover is needed and which target base station to switch to. By building a handover effect prediction model, it is possible to predict the network performance after handover, such as throughput, latency, signal strength, etc. At the same time, in the process of building the prediction model, by combining a variety of methods such as time series models, graph neural networks, and deep neural networks, the prediction model can effectively capture the time and space characteristics and the complex relationship between them, and improve the decision quality of intelligent handover of 5G networks in dynamic environments. This intelligent and innovative data processing and prediction modeling method ensures the continuous optimization of network performance and user experience.

[0041] 2. Strong switching strategy stability: By combining reinforcement learning, it is possible to learn the optimal switching strategy, balancing immediate benefits (such as network performance) and long-term benefits (such as overall network health and user experience). It also enables 5G network switching management to dynamically adapt to complex and changing network environments, and implement adaptive, intelligently optimized switching strategies. It significantly improves the accuracy and efficiency of 5G network switching decisions, and ultimately improves user experience and network resource utilization.

[0042] 3. High switching decision efficiency: By deploying a lightweight network switching engine model, the optimal switching strategy can be adjusted in real time. At the same time, model compression technology (such as knowledge distillation and quantization) is used in model design to ensure efficient operation of the model. Docker or Kubernetes is used for containerization to facilitate management and expansion, which greatly improves the decision efficiency and response speed of switching.

[0043] 4. Continuous optimization: By building a virtual environment, it is possible to evaluate strategies and estimate the impact on throughput, latency, user feedback, etc., to achieve continuous optimization of switching strategies.

[0044] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0046] Figure 1 A schematic diagram of a method flow chart provided by an embodiment of the present invention;

[0047] Figure 2 A schematic diagram of a process for constructing a prediction model provided by an embodiment of the present invention;

[0048] Figure 3 A schematic diagram of a flow chart of network switching management provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0051] like Figure 1-3 As shown, the present invention provides a 5G network intelligent switching management method based on artificial intelligence, comprising the following steps:

[0052] 1. Collect network-related data and preprocess the network-related data to obtain a network data set.

[0053] 1.1 Network-related data collection

[0054] Channel Quality Indicator (CQI): Collects channel quality information reported by user equipment. User equipment (UE) periodically reports channel quality indicators (CQI) to the base station, which reflect the current status of the wireless channel. The base station monitors CQI in real time through physical layer signal measurement to ensure the timeliness of the data. The base station aggregates CQI data from multiple UEs to form an overall view of regional channel quality.

[0055] Mobile data: including user location, speed, acceleration, direction of movement, etc. Use GPS, Wi-Fi positioning, Bluetooth beacon and other technologies to obtain the location information of the user's device. Use the device's built-in accelerometer, gyroscope and other sensors to obtain speed, acceleration and direction of movement. Infer the user's movement path and speed by analyzing changes in signal strength.

[0056] Network topology data: base station location, coverage, adjacency, base station load, etc. The base station automatically discovers neighboring base stations through the SON function and updates adjacency. The base station location, coverage, and load information are regularly obtained from the network management system (NMS). The load of the base station is monitored in real time to identify high-load and low-load areas.

[0057] Historical switching data: past switching events, success rates, failure reasons, switching delays, etc. Store historical switching data in the data warehouse to facilitate subsequent analysis and model training.

[0058] User behavior data: user application type, traffic pattern, service quality requirements, etc. Use DPI (deep packet inspection) technology to identify user application type and traffic pattern. Monitor user service quality requirements in real time, such as bandwidth, latency, jitter, etc. Build user profiles based on user historical behavior data and predict future needs.

[0059] Environmental data: such as weather conditions, geographic obstacles, road traffic, etc. Obtain real-time weather conditions through interfaces with weather service providers. Use GIS data to identify geographic obstacles and road traffic conditions. Use drones for environmental monitoring in specific areas to obtain real-time geographic and traffic information.

[0060] 1.2 Preprocess the network-related data to obtain a network data set

[0061] 1.21 Collect network-related data, perform denoising, anomaly detection and missing data filling on the network data to obtain an initial data set.

[0062] 1) Denoising method: Use a low-pass filter to filter out high-frequency noise in signal transmission and retain useful low-frequency information.

[0063] 2) Statistical methods: Use statistical indicators such as mean and median to identify and remove abnormal data points.

[0064] 3) Anomaly Detection:

[0065] IQR (interquartile range) method: By calculating the interquartile range of the data, significantly deviated large and small values ​​can be identified and eliminated.

[0066] Machine learning methods: such as the Isolation Forest model, which automatically detects abnormal data.

[0067] 4) Missing data filling:

[0068] Interpolation method: Use linear interpolation and Lagrange interpolation methods to fill in missing data.

[0069] Regression imputation: Predict missing values ​​using a simple regression model.

[0070] Multiple imputation: Fill in the gaps by training multiple models fusion results.

[0071] 1.22 The initial data set is divided into two types: continuous values ​​and discrete values, and different data standardization methods are selected for different data types to complete the standardization operation of the initial data set. For example, continuous values ​​are normalized and standardized, and discrete values ​​are one-hot encoded or binarized.

[0072] 1.23 Based on the initial data set after the normalization operation, the time series characteristics of the signal strength are calculated, the periodic patterns of the time series characteristics are identified, and the timing analysis results are obtained. Based on the timing analysis results, the interaction characteristics and statistical characteristics of the user equipment and the network base station are constructed, and then the feature selection algorithm is used to extract the features to obtain relevant feature data.

[0073] Time series analysis: Calculate the trend of signal strength, volatility (such as standard deviation) and other time series characteristics. Identify periodic patterns (such as short-term load peaks) to predict switching needs.

[0074] Interaction feature construction: Generate interaction features such as signal-to-noise ratio (SNR) and coverage overlap rate based on the associated data of the device and base station.

[0075] Statistical features: Features such as maximum value, minimum value, mean value, variance, kurtosis, and skewness can be used to describe signal quality.

[0076] 1.24 Identify the switching events in the initial data set in combination with the relevant feature data, and classify and label the identified switching events to complete the preprocessing of the network-related data.

[0077] Event classification: For example, use historical switching logs to identify and label each switching event (success, failure, delay, etc.). The classification process corresponds to the model training label one by one.

[0078] Annotation: For example, automatically annotate common switching events using predefined rules (e.g., signal loss time exceeding a certain threshold is considered a failure).

[0079] 2. Using the network data set, a switching demand prediction model for predicting network switching time and target and a switching effect prediction model for predicting network performance after switching are constructed.

[0080] 2.1 Switching demand forecasting model

[0081] Current and historical signal quality, user equipment mobility mode (position, speed and direction), base station status (load, available resources) and environmental information (weather, traffic) are extracted from the relevant feature data, and the extracted data are organized into a time series data structure to obtain a first training data set.

[0082] The LSTM network is used to capture the time dependency of channel quality, user mobility and environmental characteristics to obtain a time series model. Then, the sliding window technology is used to input the first training data set into the time series model for training, learn and predict the future state of network usage, and determine the timing of network switching.

[0083] A graph neural network model is constructed based on the network topology structure, and the first training data set is input into the graph neural network model to obtain the status and relationship changes of neighboring base stations; wherein each node in the graph neural network model represents a base station, and each edge represents the connection relationship between base stations; information propagation and aggregation are performed on the network graph so that the model captures the status and relationship changes of neighboring base stations.

[0084] Use historical switching records for supervised learning, implement cross-validation and hyperparameter optimization techniques (such as random search, Bayesian optimization) to improve model performance.

[0085] The time series model and the graph neural network model are deployed and integrated to obtain a switching demand prediction model, and the switching demand prediction model is used to predict real-time data to obtain the network switching time and the switching target base station.

[0086] 2.2 Switching effect prediction model:

[0087] Base station characteristics (signal strength, load, historical performance indicators), connection device requirements (bandwidth, delay sensitivity) and overall network status (congestion level, interference situation) are extracted from the relevant feature data, and the extracted data are integrated into a second training data set.

[0088] A multilayer perceptron (MLP) is constructed to process nonlinear feature combinations. Residual connections and batch normalization are introduced into the multilayer perceptron to optimize the training stability and speed of the deep network and obtain a deep neural network.

[0089] The deep neural network is trained using the second training data set, and a random forest (automatically selecting the optimal switching target base station through multiple decision tree voting outputs) and a gradient boosting tree (gradually adjusting the decision error and combining multiple weak learners to improve the prediction ability of the overall model) are integrated in the deep learning network to obtain a deep learning model.

[0090] The deep learning model is trained using the network performance data after switching to predict and evaluate the performance indicator trend of the network after switching.

[0091] 3. Formulate an initial switching strategy based on the constructed model, and optimize the initial switching strategy by combining reinforcement learning to obtain the optimal switching strategy. Including:

[0092] 3.1 State definition:

[0093] Current network status: including signal quality, base station load, throughput, latency, etc.

[0094] User status: location information, speed, service requirements, etc.

[0095] Environmental conditions: information such as weather and road traffic.

[0096] The current network state, user state and environment state are acquired and integrated to obtain state information, and the state information is encoded into a feature vector of a fixed length to obtain a state representation, thereby completing the state definition.

[0097] 3.2 Action Definition

[0098] Discrete action space: such as switching to a specific base station, enabling or disabling a specific function.

[0099] Continuous action space: Adjust switching parameters (such as switching thresholds, power control parameters).

[0100] 3.3 Reward Definition

[0101] Instant rewards: Instant rewards are given based on network performance indicators (such as throughput, signal strength) and user experience (such as latency, packet loss rate) after switching.

[0102] Long-term rewards: Consider the long-term optimization of network resources, such as reducing network jitter and improving system stability.

[0103] 3..4 Integrate the defined states, actions and rewards to obtain a reinforcement learning framework, use the initial data set to continuously train the reinforcement learning framework to obtain a policy model, use the policy model to optimize the initial switching policy output by the prediction model, and obtain the optimal switching policy. Deploy the policy model in the access network and core network nodes to make real-time online decisions.

[0104] During the training of the reinforcement learning framework, a deep Q-network (DQN) can be selected to handle situations with limited switching choices (discrete actions). A deep neural network (DNN) is used to estimate the Q-value function, which evaluates the value of each possible action in each state.

[0105] 4. Deploy the network switching engine model on the edge computing node and adjust the optimal switching strategy in real time according to changes in the network environment. Including:

[0106] Edge computing nodes (MEC) are close to user devices. By deploying models on MEC servers, data transmission delays can be significantly reduced, and the real-time and response speed of switching decisions can be improved. In edge computing nodes (MEC), protocols can be established to ensure that neighboring edge nodes can efficiently share necessary status information, and distributed ledger technology can be used to ensure data consistency and integrity during information sharing.

[0107] Choose lightweight models (such as MobileNet and TinyML, which are suitable for running in the limited resources of terminal devices) to reduce the network burden. Integrate and use hardware resources such as device CPU, GPU, and NPU to optimize model inference speed and power consumption.

[0108] The lightweight model is subjected to model compression (such as knowledge distillation and quantization to ensure efficient operation) and containerization operation (such as containerization using Docker or Kubernetes to facilitate management and expansion) to obtain a network switching engine model.

[0109] The initial data set is input into the network switching engine model, and the network switching operation priority and target base station recommendation are output. The trust level of available base stations is evaluated using a deep learning or reinforcement learning model, and a switching recommendation is generated to make real-time adjustments to the optimal switching strategy.

[0110] 5. Construct a virtual simulation digital environment to perform network simulation, evaluate the optimal switching strategy based on the simulated virtual network environment, and then optimize the network switching engine model in real time based on the evaluation results. Including:

[0111] Use physical simulation tools (such as NS-3 or Omnet++) to simulate network infrastructure, signal interference, and user mobility. Then use simulation tools to integrate real user behavior patterns, consider the interaction of different user devices and applications, and perform network simulation.

[0112] A digital twin is selected as a strategy testing platform, and a multi-dimensional simulation matrix is ​​introduced into the strategy testing platform to evaluate the optimal switching strategy and its impact on throughput, latency, user feedback, etc.

[0113] Among them, 1) Digital Twin is a conceptual system of interaction between the physical world and the digital space. It is both a new technology and a new paradigm. Digital twins make full use of physical models, sensor updates, operation history and other data, integrate simulation technology, and map in three-dimensional space to reflect the entire life cycle of the corresponding physical equipment. Simply put, a digital twin creates a digital version of a "clone". 2) A multidimensional simulation matrix refers to a matrix with multiple dimensions (such as two-dimensional, three-dimensional, etc.), whose elements can be numerical, symbolic, or other data types. This matrix is ​​very useful in simulation systems, especially when it is necessary to deal with complex systems or perform multidimensional data analysis.

[0114] The network switching engine model is calibrated and the optimal switching strategy is adjusted according to the evaluation result of the optimal switching strategy to achieve continuous optimization.

[0115] Therefore, the present invention provides a 5G network intelligent switching management method based on artificial intelligence, which not only realizes intelligent switching decisions, but also improves the accuracy and efficiency of switching decisions, enhances network adaptability and diversified needs, and improves user experience.

[0116] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0117] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A 5G network intelligent switching management method based on artificial intelligence, characterized in that: The following steps are involved: Collecting network-related data and preprocessing the network-related data to obtain a network data set; Using the network data set to construct a switching demand prediction model for predicting network switching time and target and a switching effect prediction model for predicting network performance after switching; Formulate an initial switching strategy based on the constructed model, and optimize the initial switching strategy in combination with reinforcement learning to obtain an optimal switching strategy; Deploy a network switching engine model on the edge computing node and adjust the optimal switching strategy in real time according to changes in the network environment; A virtual simulation digital environment is constructed to perform network simulation, the optimal switching strategy is evaluated according to the simulated virtual network environment, and then the network switching engine model is optimized in real time according to the evaluation result.

2. According to the artificial intelligence-based 5G network intelligent switching management method according to claim 1, it is characterized in that: The network-related data includes channel quality indicators, mobility data, network topology data, historical switching data, user behavior data and environmental data; Among them, the channel quality indicator is obtained from the channel quality information reported by the user equipment, the mobile data includes user location, speed, acceleration and moving direction, the network topology data includes base station location, coverage, adjacency and base station load, the historical switching data includes switching events, switching success rate, switching failure reasons and switching delay, the user behavior data includes user application type, traffic pattern, service quality requirements, and the environmental data includes weather conditions, geographical obstacle information and road traffic information.

3. According to the artificial intelligence-based 5G network intelligent switching management method according to claim 2, it is characterized in that: Collect network-related data and pre-process the network-related data to obtain a network data set, including: Collect network-related data, perform denoising, anomaly detection and missing data filling on the network data to obtain an initial data set; Dividing the initial data set into two types: continuous value and discrete value, selecting different data standardization methods for different data types, and completing the standardization operation of the initial data set; Based on the initial data set after the normalization operation, the time series characteristics of the signal strength are calculated, the periodic pattern of the time series characteristics is identified, and the time series analysis results are obtained. Based on the time series analysis results, the interaction characteristics and statistical characteristics of the user equipment and the network base station are constructed, and then the feature selection algorithm is used to extract the features to obtain relevant feature data; The switching events in the initial data set are identified in combination with the relevant feature data, and the identified switching events are classified and labeled to complete the preprocessing of the network-related data.

4. According to the artificial intelligence-based 5G network intelligent switching management method of claim 3, it is characterized in that: The network data set is used to construct a switching demand prediction model for predicting network switching time and target, including: Extracting current and historical signal quality, user equipment mobility mode, base station status and environmental information from the relevant feature data, and arranging the extracted data into a time series data structure to obtain a first training data set; Using an LSTM network to capture the time dependency of channel quality, user mobility, and environmental characteristics to obtain a time series model, and then using a sliding window technique to input the first training data set into the time series model for training, learn and predict the future state of network use, and determine the timing of network switching; Building a graph neural network model based on the network topology structure, inputting the first training data set into the graph neural network model, and obtaining the status and relationship changes of neighboring base stations; wherein each node in the graph neural network model represents a base station, and each edge represents a connection relationship between base stations; The time series model and the graph neural network model are deployed and integrated to obtain a switching demand prediction model, and the switching demand prediction model is used to predict real-time data to obtain the network switching time and the switching target base station.

5. According to the artificial intelligence-based 5G network intelligent switching management method of claim 4, it is characterized in that: The network data set is used to construct a switching effect prediction model for predicting network performance after switching, including: Extracting base station features, connection device requirements, and overall network status from the relevant feature data, and integrating the extracted data into a second training data set; Constructing a multi-layer perceptron, introducing residual connection and batch normalization into the multi-layer perceptron to obtain a deep neural network, training the deep neural network using the second training data set, and integrating a random forest and a gradient boosting tree into the deep learning network to obtain a deep learning model; The deep learning model is trained using the network performance data after switching to predict and evaluate the performance indicator trend of the network after switching.

6. According to the artificial intelligence-based 5G network intelligent switching management method of claim 5, it is characterized in that: An initial switching strategy is formulated based on the constructed prediction model, and the initial switching strategy is optimized in combination with reinforcement learning to obtain an optimal switching strategy, including: Acquire and integrate the current network state, user state and environment state to obtain state information, encode the state information into a feature vector of fixed length, obtain state representation, and complete state definition; Define the discrete action space and continuous action space after network switching, and define the immediate reward and long-term reward after network switching; Integrate the defined states, actions, and rewards to get a reinforcement learning framework; Continuously training the reinforcement learning framework using the initial data set to obtain a strategy model, and optimizing the initial switching strategy output by the prediction model using the strategy model to obtain an optimal switching strategy; The policy model is deployed in access network and core network nodes to perform real-time online decision making.

7. The 5G network intelligent switching management method based on artificial intelligence according to claim 6 is characterized in that: The network switching engine model is deployed on the edge computing node, and the optimal switching strategy is adjusted in real time according to changes in the network environment, including: Selecting a lightweight model, and performing model compression and containerization operations on the lightweight model to obtain a network switching engine model; The initial data set is input into the network switching engine model, and the network switching operation priority and target base station recommendation are output to perform real-time adjustment of the optimal switching strategy.

8. The 5G network intelligent switching management method based on artificial intelligence according to claim 7 is characterized in that: Constructing a virtual simulation digital environment to perform network simulation, evaluating the optimal switching strategy according to the simulated virtual network environment, and then optimizing the network switching engine model in real time according to the evaluation result, including: Use physical simulation tools to simulate network infrastructure, signal interference, and user mobility, and then use simulation tools to integrate real user behavior patterns to perform network simulation; Selecting a digital twin as a strategy testing platform, and introducing a multi-dimensional simulation matrix into the strategy testing platform to evaluate the optimal switching strategy; The network switching engine model is calibrated and the optimal switching strategy is adjusted according to the evaluation result of the optimal switching strategy.

Images