One-key login network optimization method and device, electronic equipment and storage medium

Abstract

The invention discloses a one-key login network state optimization method and device, electronic equipment and a storage medium, and belongs to the technical field of communication. The method comprises the following steps: acquiring one-key login multi-modal data including base station information, terminal equipment network state data, operator core network node information and a one-key login operation log; constructing a multivariate heterogeneous graph according to the base station, the terminal equipment and the operator core network node; inputting the multivariate heterogeneous graph into a graph neural network for data fusion to obtain network state fusion data; and generating an optimal network switching strategy based on the network state fusion data. The invention aims to solve the problems of low success rate and poor experience of one-key login in a weak network environment due to the fact that the network state cannot be comprehensively sensed in the prior art, and the technical effect of remarkably improving the stability and success rate of one-key login service is achieved.

Description

Technical Field

[0001] The present invention relates to the field of network communication technology, and in particular to a one-key login network status optimization method, device, electronic device and storage medium. Background Art

[0002] In mobile internet applications, the "one-click login" feature, due to its convenience, has become a mainstream solution for improving user registration and login experience. This feature relies heavily on cellular networks to obtain user phone numbers, so the quality of the network directly determines the stability and success rate of the service.

[0003] The current mobile network environment is extremely complex. Base station density, signal coverage strength, and network load vary significantly across regions. For example, in remote areas, base stations are sparse, resulting in severe signal attenuation; in urban cores, user density is high, leading to network congestion; and in special scenarios like subways, signal shielding is significant. These factors can lead to response timeouts and number retrieval failures in the "one-click login" service, which not only degrades the user experience but can even lead to user churn.

[0004] To address the issue of network connection optimization, several solutions have been proposed in the prior art. One prior art method, based on a graph neural network, determines the target base station. This method constructs a heterogeneous graph based on base station and terminal information and trains a model to determine the optimal handover target. However, the inventors discovered during their research that these prior art solutions still suffer from the following technical drawbacks:

[0005] 1. The data dimension is single and cannot fully reflect the network status: This solution only considers the information on the wireless access network side composed of terminals and base stations, ignoring the load and congestion status of the operator's core network (such as the serving gateway SGW and packet data network gateway PGW). The performance bottleneck of the core network is also a major cause of service timeouts.

[0006] 2. Lack of performance feedback for specific services: This solution optimizes for universal "handoff smoothness" but does not address specific services such as "one-click login," which require extremely high instantaneous network quality within 5 seconds. Model training is not dependent on the ultimate success of the service, resulting in a disconnect between optimization and actual service needs.

[0007] 3. The model lacks generalization capabilities and is difficult to adapt to dynamic changes in the network: The solution does not clearly specify how to deal with dynamic changes in network topology. For example, when a new base station comes online or an old base station goes offline due to a failure, its graph neural network model that relies on fixed meta-paths may be unable to process unseen nodes, resulting in a decline in prediction performance.

[0008] In view of this, how to comprehensively consider multi-dimensional information such as terminals, base stations, core networks, and combine performance feedback of specific services to build an intelligent optimization system that can dynamically adapt to changes in network topology to improve the success rate of one-click login in weak network environments is a technical problem that technical personnel in this field urgently need to solve. Summary of the Invention

[0009] The main purpose of the present invention is to provide a one-click login network status optimization method, device, electronic device and storage medium, aiming to overcome the technical problems in the prior art such as low one-click login success rate caused by single data dimension, lack of business feedback and poor model generalization ability.

[0010] To achieve the above objectives, the present invention provides a one-key login network status optimization method, which includes: Obtaining one-key login multimodal data, the one-key login multimodal data including: base station information, network status data of a terminal device connected to the base station, operator core network node information, and one-key login operation log; Preprocessing the one-key login multimodal data to obtain preprocessed multimodal data; Constructing a multi-heterogeneous graph based on the base station, the terminal device, and the operator core network node, wherein the base station, the terminal device, and the operator core network node serve as nodes, and connection relationships between the nodes serve as edges; Inputting the multivariate heterogeneous graph into a preset graph neural network, fusing the preprocessed multimodal data through the graph neural network to obtain network state fusion data; Based on the network status fusion data, an optimal network switching strategy is generated, where the optimal network switching strategy is used to instruct the terminal device to perform network switching when executing a one-key login service.

[0011] Furthermore, the preprocessing includes: data cleaning, data integration, feature extraction and normalization.

[0012] Furthermore, the characteristics of the edges and / or nodes of the multivariate heterogeneous graph include: signal strength, network delay, packet loss rate, transmission rate, one-key login success rate, one-key login response time, transmission bandwidth, signaling traffic, congestion index, switching success rate and at least one of link reliability.

[0013] Furthermore, the step of inputting the multivariate heterogeneous graph into the preset graph neural network and fusing the preprocessed multimodal data through the graph neural network includes: Processing the multivariate heterogeneous graph through a heterogeneous graph pre-training model to obtain a first fusion feature; Learning the multivariate heterogeneous graph through a heterogeneous graph meta-learner, dynamically generating and selecting a meta-path suitable for the current one-click login optimization task; Based on the selected meta-path, information aggregation is performed on the multi-heterogeneous graph to obtain a second fusion feature; The first fusion feature and the second fusion feature are fused in parallel using an attention mechanism.

[0014] Furthermore, the method also includes: training the graph neural network using a hybrid training mode, and the hybrid training mode includes: at least one of supervised learning, incremental learning, transfer learning and online learning.

[0015] Furthermore, the use of online learning to train the graph neural network includes: pushing the dynamic network switching plan generated by the optimal network switching strategy to the operator core network node for execution by the terminal device; collecting the login results after the terminal device executes the switch as feedback data; and optimizing the graph neural network in real time based on the feedback data.

[0016] Another aspect of the present invention further provides a one-key login network status optimization device, the device comprising: A data acquisition and preprocessing module is configured to acquire one-key login multimodal data, the one-key login multimodal data including: base station information, network status data of terminal devices connected to the base station, operator core network node information, and one-key login operation log; and preprocess the one-key login multimodal data to obtain preprocessed multimodal data; a heterogeneous graph construction module, configured to construct a multi-heterogeneous graph based on the base station, the terminal device, and the operator core network node, wherein the base station, the terminal device, and the operator core network node serve as nodes, and the connection relationships between the nodes serve as edges; A data fusion module, configured to input the multivariate heterogeneous graph into a preset graph neural network, and fuse the preprocessed multimodal data through the graph neural network to obtain network state fusion data; A policy generation module is used to generate an optimal network switching policy based on the network status fusion data, and the optimal network switching policy is used to instruct the terminal device to perform network switching when executing the one-key login service.

[0017] Furthermore, the data fusion module is specifically used to: Processing the multivariate heterogeneous graph through a heterogeneous graph pre-training model to obtain a first fusion feature; Learning the multivariate heterogeneous graph through a heterogeneous graph meta-learner, dynamically generating and selecting a meta-path suitable for the current one-click login optimization task; Based on the selected meta-path, information aggregation is performed on the multi-meta heterogeneous graph to obtain a second fusion feature; The first fusion feature and the second fusion feature are fused in parallel by using an attention mechanism.

[0018] Another aspect of the present application also provides an electronic device, comprising a processor and a memory; the memory is used to store computer execution instructions, and the processor is connected with the memory through a bus; when the electronic device is running, the processor executes the computer execution instructions stored in the memory, so that the electronic device executes the method of any one of the preceding aspects.

[0019] Still another aspect of the present application also provides a computer readable storage medium, which stores instructions, when the instructions are run on a computer, the computer executes the method of any one of the preceding aspects.

[0020] Compared with the prior art, the technical scheme provided by the present application has at least the following beneficial effects: 1. The comprehensiveness and accuracy of network state awareness are improved: by innovatively including the "operator core network node" in the heterogeneous graph and fusing the "one-key login operation log", the present application can build an end-to-end network state view from the access network, core network to the upper layer service, solving the problem that the existing technology cannot fully evaluate the network quality relying only on wireless side data, thereby improving the accuracy of subsequent decision-making.

[0021] 2. The adaptability and generalization ability of the model to dynamic network environment are enhanced: the present application introduces a "task-oriented heterogeneous graph meta-learner", realizes the dynamic generation and selection of meta-path, and enables the model to process nodes that have not been seen during training (such as newly added or offline base stations), and has inductive learning ability. This solves the fundamental problem that the model relying on fixed meta-path in the prior art has poor generalization ability in real dynamic networks.

[0022] 3. The closed-loop self-adaptation of network optimization is realized, and the service success rate is significantly improved: the present application adopts a hybrid training mode, especially through online learning, using the real one-key login success / failure result as feedback to continuously and real-timely optimize the model. This closed-loop optimization mechanism guided by service results ensures that the network strategy always evolves towards improving the success rate of a specific service, significantly improving the stability and success rate of the "one-key login" service in complex network environments. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. 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 any creative work.

[0024] Figure 1 This is a flowchart of a one-click login network status optimization method based on a heterogeneous graph neural network provided according to an embodiment of the present invention.

[0025] Figure 2 Schematic diagram of a multivariate heterogeneous graph constructed according to an embodiment of the present invention.

[0026] Figure 3 Schematic diagram of a feature fusion architecture proposed according to an embodiment of the present invention.

[0027] Figure 4 It is a structural block diagram of an optimization device provided according to an embodiment of the present invention.

[0028] Figure 5 The figure is a schematic diagram of the hardware structure of an electronic device provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] To make the technical problems, technical solutions and advantages to be solved by the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention.

[0030] An embodiment of the present invention provides a one-click login network status optimization system based on a heterogeneous graph neural network. The core concept of this system is to comprehensively collect multi-dimensional data from terminals, base stations, operator core networks, and the one-click login service itself. Through an advanced graph neural network model that can adapt to dynamic changes in network topology, it performs deep data fusion and analysis, accurately predicts the one-click login success rate under different network paths, and generates the optimal network switching strategy, thereby significantly improving the one-click login success rate and user experience in complex environments such as weak networks.

[0031] The solution of the present invention can be applied to 4G / 5G mobile communication networks and deployed by operators or third-party service providers. In a typical application scenario, when a user clicks the "one-click login" button on a mobile app, the optimization system described in the present invention is triggered. This system replaces the traditional fixed network connection process and dynamically selects the optimal network link for this login request within a very short time (e.g., within 5 seconds). This may involve switching to a neighboring base station with better signal strength, and in the future, operator network sharing will be implemented, i.e., China Unicom accounts will use telecom operator nodes), to maximize the success of the login.

[0032] See also Figure 1 , which shows a flow chart of a one-key login network status optimization method based on a heterogeneous graph neural network according to an embodiment of the present invention. Figure 1 As shown, the method may specifically include the following steps:

[0033] S101: Obtain one-key login multimodal data, where the one-key login multimodal data includes: base station information, network status data of a terminal device connected to the base station, operator core network node information, and one-key login operation log; preprocess the one-key login multimodal data to obtain preprocessed multimodal data.

[0034] In this step, the optimization system serves as the data collection hub, acquiring all data related to the one-click login service from multiple heterogeneous data sources in real time or near real time. Multimodal data is necessary because a single data source cannot fully capture the complex reasons for login failures.

[0035] Specifically, the one-click login multimodal data includes but is not limited to: One-click login operation log: This data directly reflects the final results of the service. It may include: the user's mobile phone number (de-identified identifier), public IP address, private IP address, login initiation time, total login time, login result (success / failure / timeout), etc.

[0036] Base station information: reflects the physical status of the wireless access network. This information may include: the base station's global cell identifier (LAC+CELLID), signal strength (RSRP), signal quality (RSRQ), signal-to-noise ratio (SINR), historical handover success rate, link reliability, etc.

[0037] Operator core network node information: This is a key data dimension that distinguishes this invention from existing technologies, reflecting the status of the data transmission backbone link. Operator core network nodes broadly refer to network elements such as the MME (Mobility Management Entity), SGW (Serving Gateway), and PGW (Packet Data Network Gateway) in the 4G (LTE) core network, and the AMF (Access and Mobility Management Function), SMF (Session Management Function), and UPF (User Plane Function) in the 5G core network. Information may include the IP address of the connected core network node, network type (4G / 5G), transmission bandwidth, real-time signaling traffic, and congestion index.

[0038] Terminal device network status data: reflects the user's network experience, which may include: terminal geographic location information, network connection delay, packet loss rate, instantaneous transmission rate, etc.

[0039] Historical data: One-click login to the server will record the response request information. The required information can be queried from different data logs using the known information, thus completing pre-training, that is, the basic model construction phase.

[0040] Raw data often contains noise and is in different formats, so preprocessing is required to improve the efficiency and effectiveness of model training. In one embodiment, this step may include:

[0041] Data cleaning: Process base station information to remove duplicate, erroneous or incomplete base station data records.

[0042] Data standardization / normalization: Normalize the numerical features in the terminal device data, such as signal strength and network latency (for example, map them to the [0, 1] interval) to eliminate the impact of different feature dimensions.

[0043] Feature Engineering: Process login logs to extract useful derived features. For example, convert login timestamps into categorical features such as "whether it is during peak hours" and perform one-hot encoding on operators (China Mobile, China Unicom, and China Telecom) and login results.

[0044] Integration and dimensionality reduction: Integrate preprocessed data from different sources and use methods such as principal component analysis (PCA) to reduce dimensionality and remove redundant information.

[0045] S102: Construct a multi-heterogeneous graph based on the base station, the terminal device and the operator core network node, wherein the base station, the terminal device and the operator core network node serve as nodes and the connection relationship between the nodes serves as an edge.

[0046] The graph structure is the basis of the GNN model. This paper constructs a multi-heterogeneous graph G=(V, E) to describe the various entities in the network and their complex relationships. Figure 2, which is a schematic diagram of a heterogeneous graph.

[0047] Node (V): The diagram contains three core types of nodes: terminal device nodes (U), base station nodes (B), and operator core network nodes (C).

[0048] Edge (E): An edge represents a connection between nodes and includes at least: Terminal-base station (UB) connection: indicates the base station where the terminal is currently stationed or accessible.

[0049] Base station-core network node (BC) connection: indicates the core network element to which the base station belongs or is connected.

[0050] Base station-to-base station (BB) connection: refers to a pair of base stations that are geographically adjacent or have a handover relationship.

[0051] Node and edge features: Use the preprocessed multimodal data as the attribute features of nodes and edges. For example:

[0052] The characteristics of the UB edge can include: signal strength, network latency, historical one-key login success rate, etc.

[0053] The characteristics of BC edges can be: transmission bandwidth, congestion index, etc.

[0054] The characteristics of the BB edge may include: historical handover success rate, link reliability, etc.

[0055] S103: Inputting the multivariate heterogeneous graph into a preset graph neural network, fusing the preprocessed multimodal data through the graph neural network to obtain network state fusion data; This is the core processing step of the present invention. Through the powerful relational learning ability of the graph neural network, discrete multi-dimensional data is fused into a vector representation that can represent the overall state of the network. In a preferred embodiment, the internal implementation architecture of this step is as follows: Figure 3 As shown in Figure 1, it includes two parallel paths: heterogeneous graph pre-training and task-oriented meta-learner. The heterogeneous graph pre-training model implements feature extraction by adopting node content conversion, intra-meta-path aggregation, and inter-meta-path aggregation. The meta-path is defined as follows:

[0056]

[0057] Abbreviated as ( ) form, which describes the node type and The compound relationship between: , where ∘ represents a compound operator on a relation.

[0058] Node content conversion: The feature vector dimensions of different nodes are different. This method maps the original feature vectors to the same space, that is, the dimensions of the feature vectors after mapping are the same.

[0059] Meta-path aggregation: Extract semantic information from each meta-path and perform a weighted aggregation. The specific process involves using an encoder to compress all node features on a meta-path instance into a single vector. Using a general attention mechanism (GAT), different weights are assigned to different meta-path instances, representing different contributions to the target node. The weighted summation yields the node representation for that meta-path.

[0060] Meta-path aggregation: The information from different meta-paths is combined into a final node representation. A global summary (average + linear transformation) is performed on the output of each meta-path. The attention mechanism is again used to determine which meta-path is more important. The results of all meta-paths are weighted and combined. Finally, a linear layer and activation function are applied to output the final embedding.

[0061] The heterogeneous graph meta-learner generates a series of meta-path sequences through a meta-generation module. It then selects the most suitable meta-path for feature extraction through parameter learning. Finally, the features obtained from the two paths are weighted using an attention mechanism to further optimize the features. This architecture aims to address the performance degradation of traditional GNN models when nodes change dynamically. The fusion process is as follows:

[0062] 1. Heterogeneous graph pre-training (such as Figure 3 Upper part): This path is responsible for learning universal, high-quality node and relationship embeddings from large-scale, relatively stable historical heterogeneous graphs. It first transforms the original feature vectors (possibly of different dimensions) of different types of nodes (U, B, C) into the same feature space through node content transformation.

[0063] Then, based on a set of artificially set meta-paths that can capture important semantics (for example, UBC means that the user is connected to a core network via a base station), intra-meta-path aggregation (such as using the attention mechanism to aggregate path instances) and inter-meta-path aggregation (such as using the attention mechanism to aggregate the outputs of different meta-paths) are performed to generate the first fusion feature.

[0064] Known information: Nodes are classified into three categories: U, B, and C, representing terminals, base stations, and operator core network nodes. V represents the union of these three categories of nodes: U, B, and C.

[0065] Node content conversion: All original node features have been projected into a unified latent space to obtain the initial node embedding .

[0066] The aggregation process within the meta-path is as follows: For a predefined set of common meta paths: , the model aggregates each meta-path instance through the attention mechanism to obtain the nodes under the meta-path For example, for the meta-path :

[0067]

[0068] in, is a node Along the metapath The neighbor set of is a meta-path instance encoder (such as a mean, linear, or relational rotary encoder), is the normalized weight calculated by the attention mechanism.

[0069] The aggregation process between meta-paths is as follows: The node Representation under all predefined meta paths Aggregate again through the attention mechanism to obtain its universal pre-trained embedding:

[0070] in, is achieved through an attention mechanism (e.g. computation With learnable attention vectors The weight obtained by the similarity and Softmax normalization represents the meta-path right importance.

[0071] The β parameter calculation process is as follows: For nodes Each meta-path of , the model will calculate it and The higher the similarity, the more important the meta-path is for generating a common node representation.

[0072] The specific calculation method is as follows: First, After a linear transformation (e.g. ,in is a learnable weight matrix), and we get a Then, we calculate the transformed representation and , or a scalar value is calculated through a small neural network (such as a single-layer perceptron) to represent the unnormalized attention score.

[0073] Finally, the scores of all meta-paths are normalized by the Softmax function to obtain the final attention weight of each meta-path .

[0074] For example:

[0075]

[0076] in is the activation function, is a learnable weight matrix.

[0077] 2. Task-oriented heterogeneous graph meta-learner ( Figure 3 Lower part): This path is the key innovation of this invention, responsible for dynamically adapting to network changes. It is an "intelligent meta-path discovery algorithm." When the network changes (such as the launch of a new base station) or when optimization for a specific task is required, the meta-generation module in the meta-learner dynamically generates or selects from a large set of candidates the meta-path sequence that best suits the current network state and task.

[0078] Subsequently, information aggregation is performed based on these dynamically selected meta-paths to generate the second fusion feature This feature captures specific, instantaneous information about the current "snapshot" of the network.

[0079] 3. Parallel attention fusion: Finally, the system uses a top-level attention mechanism to perform a weighted fusion of the universal features output by the pre-trained model (the first fused features) and the immediate features output by the meta-learner (the second fused features). This allows the model to leverage stable knowledge learned over time while quickly responding to immediate network changes, resulting in a final, highly accurate fusion of network state data.

[0080] The two-path feature aggregation process is as follows:

[0081]

[0082]

[0083]

[0084] in, and They are the pre-trained features of heterogeneous graphs and the features extracted by the heterogeneous graph element path learner.

[0085] is a learnable attention vector with the same size as and same.

[0086] and are two scores, indicating and The importance of.

[0087] Softmax Guarantee and According to the probability distribution, .

[0088] S104: Generate an optimal network switching strategy based on the network status fusion data, where the optimal network switching strategy is used to instruct the terminal device to perform network switching when executing the one-key login service.

[0089] After obtaining the fused data, the system generates decisions through a downstream prediction module.

[0090] Predicting login success rate: The network status fusion data is input into a prediction model (such as a simple multi-layer perceptron (MLP) or a module with an attention mechanism). The model predicts the success rate of one-click login if the terminal uses different potential paths (for example, connecting to neighboring base station A or neighboring base station B) under the current network status.

[0091] Generate an optimal strategy: The system selects the network path with the highest predicted success rate as the optimal path and generates a corresponding network handover strategy. For example, a strategy could be "recommend that the terminal handover from the current base station X to the neighboring base station Y" or "recommend that the operator's core network node assign a public network IP address of xxx.xxx.xxx.xxx to the current terminal."

[0092] S105: Train the model using a hybrid training mode.

[0093] In order to ensure that the model always maintains high performance in a complex and changing environment, this paper adopts a hybrid training mode: Supervised learning: Use a large amount of historical labeled data (including network status and corresponding successful login results) to train an initial, well-performing baseline model.

[0094] Incremental learning: For newly arriving streaming data, there is no need to completely retrain the model. Instead, parameters are fine-tuned based on the baseline model to adapt to recent changes in network characteristics and save computing resources.

[0095] Transfer learning: When the system needs to be deployed to a new city or region, the backbone network parameters of the model can be frozen, and only the policy modules related to the new region can be fine-tuned to achieve rapid knowledge transfer and cross-domain adaptation.

[0096] Online learning: This is key to achieving closed-loop optimization. After the system executes the generated handover policy, it immediately receives the actual login result (success or failure). This result serves as a new training sample. Using online learning algorithms (such as Q-learning or policy gradients in reinforcement learning), the policy generation module is dynamically adjusted in real time to quickly respond to sudden network issues.

[0097] Generate and execute dynamic network switching plans, and provide feedback for continuous optimization.

[0098] This is a closed-loop execution and optimization process.

[0099] 1. The system forms a specific and executable dynamic network switching solution based on the optimal network switching strategy generated in step S104.

[0100] 2. The solution is pushed to the operator's core network node through a standardized interface.

[0101] 3. The core network node sends instructions to the target terminal to perform network switching operations (such as switching cells).

[0102] 4. After the terminal switches, it continues the one-click login process. The final login result is collected by the feedback module and transmitted back to the training system as input for online learning and the next round of iterative optimization of the model.

[0103] Through the above steps, the present invention constructs a network status optimization system capable of comprehensive perception, intelligent decision-making, dynamic adaptation, and closed-loop optimization, thereby effectively solving the success rate problem of "one-click login" in a weak network environment.

[0104] See also Figure 4 , which shows a structural block diagram of an optimization device 400 provided according to an embodiment of the present invention. The device 400 can implement any of the aforementioned method embodiments. Specifically, the device 400 may include:

[0105] Data Acquisition and Preprocessing Module 401: This module is responsible for executing S101 and comprehensively collects all multimodal data related to the one-click login service from multiple heterogeneous data sources, including base station information, terminal network status data, operator core network node information, and one-click login operation logs. The module also performs operations such as cleaning, standardization, and feature engineering on the collected raw data.

[0106] Heterogeneous graph construction module 402: responsible for executing S102, constructing a multi-dimensional heterogeneous graph including terminals, base stations, and operator core network nodes based on the preprocessed data.

[0107] Data Fusion Module 403: Responsible for executing S103, it is the core of the device. It integrates a pre-set graph neural network model. In a preferred embodiment, this unit is further divided into a heterogeneous graph pre-training sub-module and a task-oriented meta-learner sub-module, and includes an attention fusion unit for fusing the features output by these two sub-modules to generate the final fused network state data.

[0108] Strategy generation module 404: responsible for executing S104, receiving the network status fusion data output by the data fusion module 404, and predicting the login success rate of different network paths through the internal prediction model, and generating the optimal network switching strategy accordingly.

[0109] Training module 405: responsible for executing S105, using a hybrid training mode to train and continuously optimize the models in the data fusion unit 404 and the strategy generation unit 404.

[0110] See also Figure 5 , which shows a hardware structure diagram of an electronic device 500 according to an embodiment of the present invention. The electronic device 500 may be a server, a workstation or a dedicated network device. Figure 5 As shown, the electronic device 500 includes a processor 501, a memory 502, and a bus 503 connecting them.

[0111] The memory 502 stores a computer program (ie, computer executable instructions). When the computer program is executed by the processor 501 , the steps in any of the aforementioned method embodiments can be implemented.

[0112] The processor 501 may be a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated AI chip (such as an ASIC or FPGA). The memory 502 may include random access memory (RAM) and / or non-volatile memory (NVM) such as a hard disk or a solid-state drive (SSD).

[0113] The present invention also provides a computer-readable storage medium (such as a USB flash drive, an optical disk, or a hard disk) on which computer instructions are stored. When the instructions are executed on a computer, the computer executes any of the aforementioned method embodiments.

[0114] The above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention.

Claims

1. A one-key login network status optimization method, characterized in that: include: Obtaining one-key login multimodal data, the one-key login multimodal data including: base station information, network status data of a terminal device connected to the base station, operator core network node information, and one-key login operation log; preprocessing the one-key login multimodal data to obtain preprocessed multimodal data; Constructing a multi-heterogeneous graph based on the base station, the terminal device, and the operator core network node, wherein the base station, the terminal device, and the operator core network node serve as nodes, and connection relationships between the nodes serve as edges; Inputting the multivariate heterogeneous graph into a preset graph neural network, fusing the preprocessed multimodal data through the graph neural network to obtain network state fusion data; Based on the network status fusion data, an optimal network switching strategy is generated, where the optimal network switching strategy is used to instruct the terminal device to perform network switching when executing a one-key login service.

2. The method according to claim 1, characterized in that The preprocessing includes: data cleaning, data integration, feature extraction and normalization.

3. The method according to claim 1, characterized in that The characteristics of the edges and / or nodes of the multivariate heterogeneous graph include: at least one of: signal strength, network delay, packet loss rate, transmission rate, one-key login success rate, one-key login response time, transmission bandwidth, signaling traffic, congestion index, switching success rate and link reliability.

4. The method according to claim 1, wherein The step of inputting the multivariate heterogeneous graph into the preset graph neural network and fusing the pre-processed multimodal data through the graph neural network includes: Processing the multivariate heterogeneous graph through a heterogeneous graph pre-training model to obtain a first fusion feature; Learning the multivariate heterogeneous graph through a heterogeneous graph meta-learner, dynamically generating and selecting a meta-path suitable for the current one-click login optimization task; Based on the selected meta-path, information aggregation is performed on the multi-heterogeneous graph to obtain a second fusion feature; The first fusion feature and the second fusion feature are fused in parallel using an attention mechanism.

5. The method according to claim 1, wherein The method further comprises: The graph neural network is trained using a hybrid training mode, where the hybrid training mode includes at least one of supervised learning, incremental learning, transfer learning, and online learning.

6. The method according to claim 5, characterized in that The adopting online learning to train the graph neural network includes: Pushing the dynamic network switching solution generated by the optimal network switching strategy to the operator's core network node for execution by the terminal device; Collecting a login result after the terminal device performs the handover as feedback data; Based on the feedback data, the graph neural network is optimized in real time.

7. A one-key login network status optimization device, characterized in that: include: A data acquisition and preprocessing module is used to acquire one-key login multimodal data, the one-key login multimodal data including: base station information, network status data of terminal devices connected to the base station, operator core network node information, and one-key login operation log; preprocess the one-key login multimodal data to obtain preprocessed multimodal data; a heterogeneous graph construction module, configured to construct a multi-heterogeneous graph based on the base station, the terminal device, and the operator core network node, wherein the base station, the terminal device, and the operator core network node serve as nodes, and the connection relationships between the nodes serve as edges; A data fusion module, configured to input the multivariate heterogeneous graph into a preset graph neural network, and fuse the preprocessed multimodal data through the graph neural network to obtain network state fusion data; A policy generation module is used to generate an optimal network switching policy based on the network status fusion data, and the optimal network switching policy is used to instruct the terminal device to perform network switching when executing the one-key login service.

8. The device according to claim 7, characterized in that The data fusion module is specifically used for: Processing the multivariate heterogeneous graph through a heterogeneous graph pre-training model to obtain a first fusion feature; Learning the multivariate heterogeneous graph through a heterogeneous graph meta-learner, dynamically generating and selecting a meta-path suitable for the current one-click login optimization task; Based on the selected meta-path, information aggregation is performed on the multi-heterogeneous graph to obtain a second fusion feature; The first fusion feature and the second fusion feature are fused in parallel using an attention mechanism.

9. An electronic device, characterized in that: include: processor, and memory; The memory is used to store computer-executable instructions, and the processor is connected to the memory via a bus; When the electronic device is running, the processor executes the computer-executable instructions stored in the memory, so that the electronic device performs the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the method according to any one of claims 1 to 6.

Images