Network switching method, network congestion prediction method, device and electronic equipment

Abstract

The application discloses a network switching method, a network congestion prediction method, a device and an electronic device. The network switching method comprises the following steps: a first terminal acquires a first prediction result; the first prediction result is a value representing a network congestion state in a future first time period, which is predicted based on a historical performance index sequence of the first terminal; a second prediction result and a corresponding confidence level are acquired; the second prediction result is a regional prediction curve representing a change of a network congestion probability in a future second time period with time, which is predicted based on terminal group data; the confidence level is used for representing the reliability of the regional prediction curve; the length of the second time period is longer than the length of the first time period; the first prediction result and the second prediction result are fused according to the confidence level to obtain a fusion curve; based on the fusion curve, a comprehensive utility value of switching from a current service network to at least one candidate network is evaluated; and a network switching decision is made according to the evaluation result of the comprehensive utility value of the at least one candidate network.

Description

Technical Field

[0001] This application belongs to the field of mobile communication technology, specifically relating to a network handover method, a network congestion prediction method, an apparatus, and an electronic device. Background Technology

[0002] With the large-scale deployment of 5G networks and the evolution of future mobile communication technologies such as 6G, mobile communication networks are developing towards ultra-high bandwidth, ultra-low latency, and ultra-large-scale connectivity. However, in typical scenarios such as high-speed movement (e.g., high-speed rail, autonomous driving) and ultra-dense user environments (e.g., sporting events, large gatherings), network traffic exhibits high dynamism and suddenness, easily triggering localized cell congestion. This congestion directly leads to a sharp deterioration in the quality of user experience (QoE), manifesting as video stuttering, voice interruptions, and a sharp increase in interaction latency, directly impacting the user experience in video, voice calls, and online interactive applications.

[0003] In related technologies, network optimization mainly relies on centralized processing mechanisms at the core network or base station level. This mechanism involves periodically collecting network performance metrics for offline analysis or simple threshold alarms, and then adjusting parameters manually or through automated scripts based on the analysis results. While this method can achieve a certain level of resource balancing at a macro level, in scenarios with rapid dynamic changes such as high-speed mobility and high user density, its inherent base station statistical lag and cloud command delivery delays will significantly degrade the terminal experience, making it difficult to achieve timely, accurate, and efficient network switching. Summary of the Invention

[0004] The purpose of this application is to provide a network handover method, a network congestion prediction method, a device, and an electronic device that can ensure timely, accurate, and efficient network handover for terminals in scenarios with drastic dynamic changes, such as high-speed movement and high user density.

[0005] In a first aspect, embodiments of this application provide a network switching method, executed by a first terminal, the method comprising: Obtain a first prediction result; wherein the first prediction result is a value that represents the network congestion status in the first time period, predicted based on the historical performance index sequence of the first terminal. Obtain a second prediction result and its corresponding confidence level; wherein, the second prediction result is a regional prediction curve representing the change of network congestion probability over time in the second time period, obtained based on terminal group data; the confidence level is used to characterize the reliability of the regional prediction curve; the duration of the second time period is longer than the duration of the first time period; Based on the confidence level, the first prediction result and the second prediction result are fused to obtain a fusion curve; Based on the fusion curve, evaluate the overall utility value of switching from the current service network to at least one candidate network; A network switching decision is made based on the evaluation results of the comprehensive utility value of the at least one candidate network.

[0006] Secondly, embodiments of this application provide a network congestion prediction method, executed by a prediction device, the method comprising: Receive network status data reported from multiple terminals; The received network status data is spatiotemporally aggregated; Based on the aggregated data, a second prediction result is generated; wherein, the second prediction result is a regional prediction curve characterizing the change of network congestion probability over time in the second future time period; The confidence level of the second prediction result is calculated based on the number of terminals participating in the aggregation and the consistency of the network status data reported by each terminal. The second prediction result and the confidence level are sent to the requesting terminal.

[0007] Thirdly, embodiments of this application provide a network switching device applied to a first terminal, the device comprising: The first acquisition module is used to acquire a first prediction result; wherein the first prediction result is a value that represents the network congestion status in the future first time period, which is predicted based on the historical performance index sequence of the first terminal. The second acquisition module is used to acquire a second prediction result and the corresponding confidence level; wherein, the second prediction result is a regional prediction curve that represents the change of network congestion probability over time in the second time period, obtained based on terminal group data; the confidence level is used to characterize the reliability of the regional prediction curve; the duration of the second time period is longer than the duration of the first time period; The fusion module is used to fuse the first prediction result and the second prediction result according to the confidence level to obtain a fusion curve; An evaluation module is used to evaluate the overall utility value of switching from the current service network to at least one candidate network based on the fusion curve. The decision module is used to perform network switching decisions based on the evaluation results of the comprehensive utility values ​​of the at least one candidate network.

[0008] Fourthly, embodiments of this application provide a network congestion prediction device, applied to a prediction device, the device comprising: The receiving module is used to receive network status data reported from multiple terminals; The aggregation module is used to perform spatiotemporal aggregation on the received network status data; The generation module is used to generate a second prediction result based on the aggregated data; wherein, the second prediction result is a regional prediction curve characterizing the change of network congestion probability over time in the second future time period; The calculation module is used to calculate the confidence level of the second prediction result based on the number of terminals participating in the aggregation and the consistency of the network status data reported by each terminal. The first sending module is used to send the second prediction result and the confidence level to the requesting terminal.

[0009] Fifthly, embodiments of this application provide an electronic device, which includes a processor and a memory. The memory stores programs or instructions that can run on the processor. When the program or instructions are executed by the processor, they implement the steps of the network switching method as described in the first aspect, or the steps of the network congestion prediction method as described in the second aspect.

[0010] In a sixth aspect, embodiments of this application provide a computer-readable storage medium storing a program or instructions that, when executed by a processor, implement the steps of the network switching method as described in the first aspect, or the steps of the network congestion prediction method as described in the second aspect.

[0011] In a seventh aspect, embodiments of this application provide a chip, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the network switching method as described in the first aspect, or to implement the steps of the network congestion prediction method as described in the second aspect.

[0012] Eighthly, embodiments of this application provide a computer program product stored in a storage medium, which is executed by at least one processor to implement the network switching method as described in the first aspect, or to implement the steps of the network congestion prediction method as described in the second aspect.

[0013] In this embodiment, a first terminal obtains a first prediction result; wherein the first prediction result is a value representing the network congestion state in the future first time period, predicted based on the historical performance index sequence of the first terminal; a second prediction result and its corresponding confidence level are obtained; wherein the second prediction result is a regional prediction curve representing the change of network congestion probability over time in the future second time period, predicted based on terminal group data; the confidence level is used to characterize the reliability of the regional prediction curve; the duration of the second time period is longer than the duration of the first time period; the first prediction result and the second prediction result are fused according to the confidence level to obtain a fused curve; based on the fused curve, the comprehensive utility value of switching from the current serving network to at least one candidate network is evaluated; and a network switching decision is executed according to the evaluation result of the comprehensive utility value of at least one candidate network.

[0014] As can be seen, in this embodiment, by fusing millisecond / second-level short-term predictions on the terminal side and minute-level regional predictions on the network side, a fused prediction result with both rapid response capability and macro-trend judgment can be provided to the terminal in scenarios with drastic dynamic changes such as high-speed movement and high user density. This provides a basis for timely and accurate network handover decisions. Based on this fused prediction result, the handover utility of each candidate network is evaluated, enabling the terminal to execute predictive handover decisions. This avoids frequent handovers caused by instantaneous signal fluctuations while proactively mitigating network congestion, achieving timely, accurate, and efficient network handover in such dynamic scenarios. Attached Figure Description

[0015] Figure 1 This is a flowchart of a network switching method provided by some embodiments of this application; Figure 2 This is a flowchart of a network congestion prediction method provided by some embodiments of this application; Figure 3 These are example diagrams of a network handover system provided in some embodiments of this application; Figure 4 This is an example diagram comparing the number of network handovers for different schemes provided in some embodiments of this application; Figure 5 These are example graphs showing the variation of network congestion prediction accuracy over time for different schemes provided in some embodiments of this application; Figure 6 This is a structural block diagram of a network switching device provided in some embodiments of this application; Figure 7 This is a structural block diagram of a network congestion prediction device provided in some embodiments of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided in some embodiments of this application; Figure 9 This is a schematic diagram of the hardware structure of an electronic device that implements the various embodiments of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0017] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0018] As mobile communication systems evolve from 5G to 6G, they continue to improve in terms of bandwidth, connection density, and service diversity. However, in high-speed mobility scenarios (such as high-speed rail and urban expressways) and ultra-high user density scenarios (such as concerts and sporting events), network traffic exhibits bursty and tidal characteristics, which can easily lead to local cell congestion. This manifests as decreased throughput, increased latency jitter, higher retransmission rates, and service interruptions, severely impacting the user experience of video, voice, and real-time interactive services.

[0019] In related technologies, network optimization mechanisms mainly rely on background statistics and periodic parameter adjustments at the base station level. While this approach is effective in macro-level resource allocation, its response time is long (minutes to hours), making it difficult to cope with sudden traffic surges that occur within seconds. Furthermore, its decisions are based on network-side aggregated metrics, failing to detect real-time degradation in the quality of experience (QoE) for individual users.

[0020] On the terminal side, traditional handover decisions are mostly based on threshold comparisons of instantaneous wireless signal quality (such as RSRP and RSRQ), which is a delayed triggering method: handover is only executed when the signal has deteriorated to the point of affecting services. This method lacks the ability to predict network congestion trends and does not fully consider the power consumption, signaling load, and service interruption risks brought about by the handover itself, which may lead to unnecessary frequent handovers or suboptimal handover decisions.

[0021] In summary, the relevant technologies have the following technical problems in network congestion control and handover judgment: delayed response: the passive triggering mechanism based on post-event detection cannot avoid congestion in advance before users perceive any damage; insufficient predictive ability: the lack of a two-layer predictive framework that integrates real-time terminal perception and regional load trends makes it difficult to balance the real-time nature and accuracy of decision-making; single decision-making dimension: the handover decision does not systematically quantify and evaluate the costs of energy consumption, signaling, and service continuity, and cannot achieve overall optimization of experience, energy efficiency, and network load.

[0022] Therefore, there is an urgent need to propose a network switching method that can achieve millisecond / second-level real-time perception at the terminal side, minute-level trend prediction at the network side or group side, and integrate the information from both sides for multi-dimensional cost assessment, in order to meet the service quality assurance requirements in future high-dynamic and high-density network scenarios.

[0023] The network switching method provided in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0024] It should be noted that the network switching method provided in this application embodiment is applicable to terminals. In practical applications, the terminals include, but are not limited to, mobile terminals such as smartphones, tablets, smartwatches, and smart bracelets, in-vehicle terminals, and Internet of Things devices.

[0025] Figure 1 This is a flowchart illustrating a network handover method provided in some embodiments of this application. The method is executed by a first terminal (such as a smartphone, vehicle-mounted terminal, IoT device, etc.) to achieve proactive and intelligent network handover. Figure 1 As shown, the method may include the following steps: step 101, step 102, step 103, step 104 and step 105.

[0026] In step 101, a first prediction result is obtained; wherein, the first prediction result is a value that represents the network congestion status in the first time period, which is predicted based on the historical performance index sequence of the first terminal.

[0027] In this embodiment of the application, the terminal performs short-term, high-frequency local network congestion prediction based on its own collected historical performance data.

[0028] In this embodiment of the application, the historical performance index sequence refers to a multi-dimensional data sequence that is continuously collected by the terminal over a period of time, reflecting the network status and user experience, and is usually organized in the form of a time sliding window.

[0029] In this embodiment of the application, the first future period is a relatively short prediction time window, such as 1 to 10 seconds in the future.

[0030] In this embodiment, the first prediction result is a quantified numerical value, such as congestion probability (0~1), comprehensive congestion index score, or predicted QoE (such as lag rate). Generally, the higher the value, the greater the probability of network congestion occurring in the first time period. However, in subsequent fusion processing, the first prediction result can be expanded into a short-term prediction component according to the algorithm.

[0031] For example, the terminal collects data such as video stuttering rate, application layer round-trip time (RTT), and wireless signal strength (RSRP / RSRQ) once per second. This data is used to form a sliding window of the past 5 seconds, which is then input into a lightweight local LSTM model. This model outputs a probability value, for example, 0.65, of network congestion within the next 3 seconds.

[0032] In this embodiment of the application, by obtaining the first prediction result, the millisecond / second level rapid perception and prediction of the network congestion status of the terminal is realized, ensuring the real-time response.

[0033] In step 102, a second prediction result and its corresponding confidence level are obtained; wherein, the second prediction result is a regional prediction curve that is predicted based on terminal group data and represents the change of network congestion probability over time in the second time period; the confidence level is used to characterize the reliability of the regional prediction curve; the duration of the second time period is longer than the duration of the first time period.

[0034] In this embodiment, the terminal obtains longer-term regional network trend prediction information based on a wider range of terminal group data from the prediction device (network side or neighboring terminal cluster), and attaches reliability assessment information of the prediction information, namely, the second prediction result and its confidence level.

[0035] In this embodiment of the application, terminal group data refers to data aggregated from network status data reported anonymously by multiple terminals within a certain geographical area.

[0036] In this embodiment of the application, the regional prediction curve is a data format that represents the trend line of the network congestion probability in the region changing over a future period of time (such as 1-5 minutes).

[0037] In this embodiment, confidence level is a quantitative indicator used to evaluate the reliability of the regional prediction curve. Its calculation can be based on factors such as the number of terminals participating in data aggregation and data consistency (variance).

[0038] In this embodiment of the application, the future second period is a relatively long prediction time window, which is significantly longer than the first period, for example, the future 1 to 5 minutes.

[0039] For example, when a terminal detects an impending risk of network congestion, it sends a query request to a nearby mobile edge computing (MEC) server. The MEC server aggregates data from hundreds of terminals in the carriage or venue over the past few minutes, generates a regional congestion probability curve for the next two minutes using a spatiotemporal prediction model, calculates the confidence level of the current prediction to be 0.8 (out of 1.0), and sends it to the terminal.

[0040] In this embodiment, by acquiring the second prediction result and its confidence level, minute-level macro-trend prediction is achieved, overcoming the limitations of a single terminal perspective in regional load perception, thereby improving the accuracy and stability of the prediction. Simultaneously, the confidence level, as a quantitative assessment of the reliability of the regional prediction result, provides a crucial input parameter for subsequent fusion decision-making.

[0041] In step 103, the first prediction result and the second prediction result are fused according to the confidence level to obtain the fusion curve.

[0042] In this embodiment, the terminal does not simply select one of the two, but intelligently weights and fuses the short-term numerical prediction and the long-term curve prediction to form a more comprehensive and reliable final prediction curve, namely the fused curve.

[0043] In this embodiment of the application, the fusion process is a data processing procedure that dynamically determines the weights of the first prediction result and the second prediction result in the fusion based on the confidence level of the second prediction result.

[0044] In this embodiment of the application, the fusion curve refers to a single curve generated after fusion processing that characterizes the change of future network congestion probability over time. It integrates instantaneous fluctuations and long-term trend information.

[0045] For example, the terminal receives a regional prediction curve (confidence level 0.8). According to preset rules (e.g., when the confidence level is higher than 0.7, the regional prediction weight is set to 0.7 and the local prediction weight is set to 0.3), the local prediction value is expanded into a decaying short-term component, and then weighted and superimposed with the time-aligned regional prediction curve to generate a fusion curve from the current moment to the next 2 minutes.

[0046] In this embodiment, the advantages of a two-layer prediction mechanism are complemented by fusing the first and second prediction results. This fusion process dynamically adjusts information weights based on confidence levels: when group data is reliable (high confidence), the reliance on regional macro trends is increased to improve decision stability; when data is sparse or noisy (low confidence), the reliance on the terminal's local real-time perception is strengthened to ensure basic responsiveness. Through this adaptive weight adjustment mechanism, the terminal achieves a balance between robustness and adaptability in a dynamic network environment.

[0047] In step 104, based on the fusion curve, the overall utility value of switching from the current service network to at least one candidate network is evaluated.

[0048] In this embodiment, the terminal uses a fusion prediction curve to prospectively evaluate the cost-effectiveness of switching to various available candidate networks (such as neighboring 5G and Wi-Fi 6), i.e., the overall utility value.

[0049] In this embodiment, the overall utility value is a quantitative indicator used to compare the advantages and disadvantages of different handover options. Its calculation is typically: Overall Utility Value = Handover Benefits - Handover Costs. The handover benefits are generally positively correlated with the congestion severity predicted by the convergence curve as avoidable through handover. Handover costs include at least power consumption costs (scanning, signaling interaction, new link maintenance), signaling overhead costs (consumption of network resources), and service interruption risk costs (the impact of handover failure or delay on current services).

[0050] For example, the terminal extracts the predicted probability 1.5 seconds later (considering the handover execution time) from the fusion curve as the basis for revenue. For each candidate network, the power consumption and number of standard signaling messages required for its handover process are estimated, and a higher interruption risk weight is assigned in combination with the current video call service. Finally, the comprehensive utility value of each candidate network is calculated through a preset utility function. For example, the comprehensive utility value of network A is +0.5, and the comprehensive utility value of network B is +0.3.

[0051] In this embodiment, based on the fusion curve, the comprehensive utility value of switching from the current serving network to at least one candidate network is evaluated, transforming the handover decision from a simple "strongest signal" problem into a multi-objective optimization problem. While pursuing a better user experience (avoiding congestion), it also considers terminal power consumption and network efficiency.

[0052] In step 105, a network switching decision is made based on the evaluation results of the comprehensive utility value of at least one candidate network.

[0053] In this embodiment, the terminal compares the overall utility values ​​of all candidate networks and makes a final network switching decision and execution based on the comparison results.

[0054] In this embodiment, the switching decision may include: determining whether to switch and which target network to switch to. For example, the candidate network with the highest overall utility value that exceeds a certain threshold is selected as the target network; if no network has a positive overall utility value, the current network connection is maintained.

[0055] For example, the terminal comparison found that candidate network A had the highest overall utility value (+0.5) and exceeded the preset handover trigger threshold (+0.2). Therefore, it was determined to switch to network A and the underlying protocol stack was triggered to execute the standard handover process.

[0056] In this embodiment, by combining the fusion prediction results with multi-dimensional cost assessment, a complete decision-making loop from "perception-prediction-assessment" to "action" is completed. This decision-making process is based on forward-looking quantitative analysis of the fusion curve, enabling the terminal to proactively switch to the optimal candidate network before the user experience is actually impaired. This achieves precise avoidance of network congestion and improves the connection stability and service continuity of the terminal in high-speed mobile and high-density scenarios.

[0057] As can be seen from the above embodiments, in this embodiment, the first terminal obtains a first prediction result; wherein, the first prediction result is a value representing the network congestion state in the future first time period, predicted based on the historical performance index sequence of the first terminal; a second prediction result and the corresponding confidence level are obtained; wherein, the second prediction result is a regional prediction curve representing the change of network congestion probability over time in the future second time period, predicted based on terminal group data; the confidence level is used to characterize the reliability of the regional prediction curve; the duration of the second time period is longer than the duration of the first time period; the first prediction result and the second prediction result are fused according to the confidence level to obtain a fused curve; based on the fused curve, the comprehensive utility value of switching from the current serving network to at least one candidate network is evaluated; and a network switching decision is executed according to the evaluation result of the comprehensive utility value of at least one candidate network.

[0058] As can be seen, in this embodiment, by fusing millisecond / second-level short-term predictions on the terminal side and minute-level regional predictions on the network side, a fused prediction result with both rapid response capability and macro-trend judgment can be provided to the terminal in scenarios with drastic dynamic changes such as high-speed movement and high user density. This provides a basis for timely and accurate network handover decisions. Based on this fused prediction result, the handover utility of each candidate network is evaluated, enabling the terminal to execute predictive handover decisions. This avoids frequent handovers caused by instantaneous signal fluctuations while proactively mitigating network congestion, achieving timely, accurate, and efficient network handover in such dynamic scenarios.

[0059] In some embodiments provided in this application, step 101 may specifically include the following steps: step 1011 and step 1012; In step 1011, the historical performance index sequence is organized into a sliding window containing data from the most recent N consecutive sampling time points; where N is a positive integer.

[0060] In this embodiment of the application, a structured, fixed-length recent historical input is constructed for the terminal's local short-term prediction model by organizing the historical performance index sequence into a sliding window containing data from the most recent N consecutive sampling time points.

[0061] In this embodiment, the sliding window is a data processing technique that maintains a fixed-size buffer (of length N) that always retains the data of the latest N consecutive sampling points. Each time new sampling data arrives, the oldest data is removed, the new data is moved in, and the window slides forward.

[0062] In this embodiment of the application, the sampling time point refers to the moment when the terminal collects performance indicators at a fixed time interval (such as once per second).

[0063] For example, the sampling period is set to 1 second, and the sliding window size is N=10. The terminal continuously collects data. The data contained in the sliding window at the current moment (t=10 seconds) is a sequence of historical performance indicators for the 10 moments t=1 second, 2 seconds, ..., 10 seconds. In the next second (t=11 seconds), new data enters, the window slides, and its content becomes the data for t=2 seconds, 3 seconds, ..., 11 seconds.

[0064] In this embodiment, by organizing discrete sampling points into continuous short time segments of fixed length, the terminal's local short-term prediction model can capture the changing trends and dynamic patterns of network status in the recent period (e.g., the previous 10 seconds), thereby overcoming the limitations of judgment based on single-point instantaneous values. This sliding window processing method, on the one hand, provides a fixed dimension and unified format for standardized model input, simplifying model design and ensuring the stability and repeatability of the inference process; on the other hand, its sliding update mechanism ensures that the model is always calculated based on the latest and most relevant historical data, thus meeting the real-time processing requirements of network switching decisions.

[0065] In step 1012, the data in the sliding window is input into the terminal's local short-term prediction model, and the terminal's local short-term prediction model outputs the first prediction result.

[0066] In this embodiment of the application, the preprocessed time series data is fed into a dedicated lightweight model deployed locally on the terminal, namely the terminal local short-term prediction model. This model is used to learn the mapping relationship between historical patterns and future congestion states and output prediction values.

[0067] In this embodiment, the terminal-local short-term prediction model is an optimized (e.g., pruned, quantized) machine learning model deployed and running on a terminal device, specifically designed to process local time-series data and make second-level future predictions. Examples include Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), or other lightweight temporal neural networks.

[0068] In this embodiment of the application, the terminal local short-term prediction model calculates based on the input sliding window data and the parameters learned internally, and finally outputs a quantitative prediction result, such as the congestion probability value in the next few seconds, i.e., the first prediction result.

[0069] For example, the sliding window vector containing data from 10 time points is input into a pre-trained and optimized LSTM model. The hidden layers of this LSTM model process the sequence information and finally output a scalar value between 0 and 1 through a fully connected layer, such as 0.72, which means that the probability of congestion occurring within the next 3 seconds is predicted to be 72%.

[0070] In this embodiment, distributed intelligent processing is achieved by deploying a lightweight time-series prediction model on the terminal side. This approach localizes prediction capabilities, avoiding the privacy risks and interaction delays associated with transmitting data back to the network, and ensuring extremely low response times down to the millisecond level. Simultaneously, the model undergoes lightweight optimizations such as pruning and quantization, enabling its inference process to adapt to the limited computing power and power consumption budget of the terminal, thus ensuring the practicality and deployability of the solution. Furthermore, such models (e.g., Long Short-Term Memory networks, LSTM) can automatically learn and extract complex nonlinear temporal dependencies from well-organized historical sequence data, achieving more accurate short-term predictions of network congestion trends compared to traditional methods based on simple rules or statistics.

[0071] In some embodiments, the first prediction result described above can be a comprehensive congestion index C, used to quantify the degree of future network congestion. Specifically, the model calculates this index by weighted summation of its predicted key network parameters.

[0072] For example, the following formula can be used: C=α×RTT pred +β×(1-SINR pred / SINR ref )+γ×Loss pred Among them, RTT pred SINR pred and Loss pred These represent the round-trip time, signal-to-interference-plus-noise ratio (SINR), and packet loss rate at a specific future time (e.g., Δt seconds later) predicted by the model, respectively; SINR refThis is a preset reference signal-to-noise ratio baseline value; α, β, and γ are the weighting coefficients for each term, and satisfy α + β + γ = 1. For example, a preferred set of weighting coefficients is: α = 0.35, β = 0.25, γ = 0.5. The higher the value of this index C, the greater the predicted risk of future congestion.

[0073] As can be seen, in this embodiment, a highly efficient and standardized short-term prediction process is constructed by using a sliding window mechanism to organize and preprocess discrete sampled data in a time series manner, and by utilizing a lightweight time series model deployed locally on the terminal for inference calculation. This process ensures prediction accuracy while taking into account millisecond-level real-time response capabilities and the limited computing and power consumption resources of the terminal, providing crucial prediction information input for subsequent rapid switching decisions.

[0074] In some embodiments provided in this application, the above-mentioned historical performance index sequence may include at least two types of indices: application layer experience quality indices, wireless link layer measurement indices, and terminal motion state context indices.

[0075] In this embodiment, a multi-dimensional indicator fusion strategy is adopted, rather than a single type of data, to comprehensively and accurately describe the network conditions faced by the terminal. These three types of indicators provide information from three different levels: user perception, physical link, and environmental context.

[0076] In this embodiment, the application layer quality of experience (QoE) metric is an objective parameter that directly reflects the end user's subjective experience with the service they are currently using and can be quantified at the application layer. It resides at the upper layer of the protocol stack and is the ultimate manifestation of network performance. For video services, the application layer QoE metric includes, but is not limited to: stuttering frequency, stuttering duration, initial buffering time, and video frame rate drop rate. For voice / real-time communication services, the application layer QoE metric includes, but is not limited to: voice intermittency perception, end-to-end latency, and voice quality score (such as MOS score). For web browsing / interactive application services, the application layer QoE metric includes, but is not limited to: page load time and application response latency.

[0077] In this embodiment, the wireless link layer measurement index is a physical layer parameter that reflects the quality of the wireless connection between the terminal and the current serving base station / access point, obtained by measurement at the terminal's physical layer or protocol stack layer. The wireless link layer measurement index includes, but is not limited to: signal strength and quality such as Reference Signal Received Power (RSRP), Reference Signal Received Quality (RSRQ), and Signal-to-Interference-plus-Noise Ratio (SINR); link performance such as Uplink / Downlink Block Error Rate (BLER), number of Physical Resource Blocks (PRBs) allocated by the scheduler, and physical layer throughput; and connection reliability such as Wireless Link Failure (RLF) event count and handover success rate history.

[0078] In this embodiment, the terminal motion state context indicator is auxiliary information that describes the terminal's own state and external environment, and may have a systematic impact on network connectivity. The terminal motion state context indicator includes, but is not limited to: instantaneous speed (from GPS / inertial sensors), acceleration, motion direction (whether facing or away from the base station), serving cell ID, geographical location (latitude and longitude), whether it is at the cell edge, and current time period (such as peak business hours), terminal screen status (screen on / off), etc.

[0079] As can be seen, in this embodiment, a multi-dimensional network state awareness system is constructed by integrating application-layer experience quality indicators, wireless link-layer measurement indicators, and terminal motion state context indicators. This system overcomes the potential user experience misjudgments that may result from relying solely on wireless signal strength, and provides richer, deeper clues for the prediction model through comprehensive analysis of multi-indicator change patterns, supporting earlier and more accurate predictions. Simultaneously, the introduction of contextual information enables the terminal to identify specific scenarios such as high-speed rail and densely populated venues, achieving adaptive adjustment of prediction sensitivity and decision threshold.

[0080] In some embodiments provided in this application, the above-mentioned 102 may specifically include the following steps: step 1021 and step 1022; In step 1021, if the first prediction result meets the short-term congestion risk condition, a prediction query request is sent to the prediction device.

[0081] In this application embodiment, a mechanism for obtaining the second prediction result and its confidence level on demand is defined, instead of periodically or indiscriminately requesting the second prediction result and its confidence level, in order to save signaling overhead and terminal power consumption.

[0082] In this embodiment, the short-term congestion risk condition is a judgment logic based on the first prediction result, used to confirm whether there is an urgent local network congestion risk that needs further verification. It can be a threshold judgment (e.g., the predicted congestion probability > 0.6), a continuous triggering mechanism (e.g., two consecutive predictions exceeding the threshold), or a trend judgment (the upward slope of the predicted value exceeds the threshold).

[0083] In this embodiment, the prediction device is a computing entity capable of providing regional-level network state prediction. It can be a network-side device (such as a mobile edge computing (MEC) server or a computing unit built into a base station) or a second terminal (such as a cluster head terminal elected in an ad hoc network).

[0084] In this embodiment, the prediction query request is a signaling or data message initiated by a terminal to request future network congestion prediction information for its current location or serving cell area. This request typically includes the terminal's location identifier (such as cell ID, geographic location) and a possible timestamp.

[0085] For example, the terminal's local short-term prediction model outputs a congestion probability of 0.7 for the next 3 seconds, and the preset short-term risk threshold is 0.65. Since 0.7 > 0.65, the short-term congestion risk condition is determined to be met. The terminal then generates a prediction query request, which includes its currently connected cell ID (such as Cell ID 12345), and sends it to the MEC server serving the area via the uplink channel.

[0086] In this embodiment, a local risk triggering mechanism is used to activate the region prediction query only when a short-term congestion risk is detected, thus avoiding unnecessary signaling overhead and power consumption introduced by continuous high-frequency interaction between the terminal and the network. Simultaneously, this on-demand acquisition mechanism ensures that the requested region prediction information is highly targeted, directly responding to potential network problems currently faced by the terminal, thereby improving information utilization efficiency and the accuracy of the final handover decision.

[0087] In step 1022, the second prediction result and the corresponding confidence level returned by the prediction device in response to the prediction query request are received.

[0088] In this embodiment, the prediction device responds to the received prediction query request, performs corresponding calculations or query operations, and returns a second prediction result and its confidence level to the terminal, namely, the regional prediction curve and its reliability quantification value. This process completes a full interactive closed loop from the terminal initiating the request to obtaining prediction information.

[0089] For example, after receiving a prediction query request from a terminal (Cell ID 12345), the MEC server retrieves from its regional status database the latest calculated congestion probability curve for the next 2 minutes for that cell region (i.e., the second prediction result), along with a confidence level (e.g., 0.85) derived from the number of terminal samples and data variance used to calculate the curve. The MEC server then sends the curve and confidence level together to the requesting terminal via the downlink channel.

[0090] In this embodiment, a request-response mechanism enables the terminal to obtain the regional prediction information and its confidence level required for intelligent fusion decision-making, completing the decision-making closed loop from local early warning to external information acquisition. This mechanism ensures that the prediction results received by the terminal are generated in real time based on the latest data, possessing high timeliness and relevance. Simultaneously, the asynchronous interaction mode allows the terminal to execute other processing tasks in parallel while waiting for a response, helping to optimize decision-making timing and reduce end-to-end latency.

[0091] As can be seen, in this embodiment of the application, the regional prediction information is activated on demand by using the local risk warning of the terminal as the trigger condition. This ensures that the terminal can obtain high-quality prediction information in a timely manner to support accurate decision-making, while optimizing the consumption of signaling resources and the efficiency of end-to-end response.

[0092] In some embodiments provided in this application, the prediction device is a network-side device or a second terminal; wherein, when the prediction device is a second terminal, the second terminal is the cluster head terminal elected by the temporary cluster formed with the first terminal.

[0093] This application's embodiments clarify two possible forms of the prediction device, demonstrating the flexibility of the network switching system architecture. This is not a simple choice between two options, but rather an adaptive design for different deployment environments and resource conditions. It covers everything from centralized cloud / edge computing models with infrastructure support to distributed terminal collaboration models without infrastructure dependence.

[0094] In this embodiment, network-side equipment refers to a network functional entity deployed on the mobile communication network infrastructure side, possessing certain computing, storage, and communication capabilities. Examples include mobile edge computing (MEC) servers, base stations with computing capabilities (such as centralized units with CU / DU separation architecture), or network elements in the core network responsible for regional data analysis.

[0095] In this embodiment, the second terminal, also known as the cluster head terminal, refers to other user terminal devices (such as smartphones or vehicle terminals) located in the same geographical area as the first terminal that initiated the request. When multiple such terminals spontaneously organize into a temporary cluster to complete a common task (such as regional prediction), a terminal is elected as the cluster head terminal through a certain election algorithm (usually based on indicators such as remaining battery power, computing power, and signal stability). The cluster head terminal is used to coordinate communication within the cluster, aggregate member data, perform prediction calculations, and interact with the first terminal as an external representative.

[0096] In this embodiment, the temporary cluster is a dynamically formed terminal self-organizing network based on device-to-device (D2D) communication or sidelink, whose members, topology, and lifecycle change with the scenario and requirements.

[0097] For example, taking network-side devices as an example, in a high-speed rail scenario, MEC servers deployed in areas such as carriages or base stations along the line continuously receive and process anonymous network status data periodically reported by all passenger terminals in that area. When a passenger terminal (the first terminal) sends a prediction query request, the MEC server generates a regional prediction curve based on its maintained regional data model and returns it.

[0098] For example, taking the second terminal (cluster head terminal) as an example, at a large open-air concert venue, there is no fixed MEC coverage. Multiple audience terminals (including the first terminal) automatically discover each other via Wi-Fi Direct or 5G Sidelink, forming a temporary cluster. By comparing battery level (>50%) and signal strength (RSRP>-90 dBm), a terminal with sufficient battery and good signal is elected as the cluster head terminal. The cluster head terminal collects summary information such as recent throughput and latency from several surrounding terminals, runs a lightweight crowd prediction model, and returns the result to the first terminal that made the request.

[0099] As can be seen, this application embodiment enhances deployment flexibility and environmental robustness by supporting prediction devices in both network-side and terminal cluster forms. In scenarios with stable network infrastructure (such as cities and high-speed rail), mobile edge computing nodes or enhanced base stations can be used to provide powerful centralized computing and a global data perspective. In scenarios with missing, damaged, or overloaded infrastructure (such as disaster sites and large gatherings), decentralized group prediction can be achieved through self-organizing terminal clusters, ensuring service availability and survivability. Furthermore, in terms of resource utilization, the network-side mode offloads the computing load to the infrastructure, suitable for scenarios with limited terminal resources or large-scale terminals; the terminal cluster mode utilizes idle terminal resources to achieve distributed intelligence, and its prediction capabilities can be enhanced as the cluster size increases, demonstrating good scalability. Simultaneously, the terminal cluster mode avoids uploading sensitive user data to the remote network side through local data exchange and processing between terminals, effectively reducing the risk of privacy leaks.

[0100] In some embodiments provided in this application, the network switching method provided is... Figure 1 Based on the illustrated embodiment, the following step can be added: Step 106; In step 106, network status data is sent to the prediction device according to the first cycle; wherein, the network status data is obtained by de-sensitizing the historical performance index sequence.

[0101] In this embodiment, a mechanism is introduced in which the terminal periodically reports data to the prediction device, providing a continuous data input source for the prediction device to generate high-quality regional prediction curves and their confidence levels. This mechanism is independent of prediction query requests triggered by local risks at the terminal, constituting the data supply side.

[0102] In this embodiment, the first period is a preset time interval used to control the frequency of data reporting, such as every 10 seconds or every 30 seconds. The setting of the first period needs to balance data freshness and uplink signaling overhead.

[0103] In this embodiment of the application, network status data refers to a processed subset of network-related information selected and reported by the terminal to support regional prediction.

[0104] In this embodiment, de-identification is a data preprocessing technique used to remove or obfuscate personally identifiable information from data, thereby protecting user privacy. De-identification includes, but is not limited to: anonymization (removing the terminal's unique identifier), generalization (blurring precise locations into a grid of regions), adding noise (differential privacy techniques), and data aggregation (reporting only statistical features such as mean and variance).

[0105] For example, the terminal performs an operation every 30 seconds (first cycle): collecting the average RTT, 95th percentile latency, average downlink throughput, and primary serving cell ID over the past 30 seconds. Identification information such as IMEI is removed, and the GPS coordinates are obfuscated into a 100-meter square grid code, which is then sent to the MEC server via a lightweight uplink message.

[0106] In this embodiment, a periodic network status data reporting mechanism provides a continuous and stable regional network status data stream to the prediction device, enabling it to build and maintain a dynamically updated regional network situational view, thus providing a data foundation for generating high-precision regional prediction curves. This mechanism achieves an effective balance between data sharing and privacy protection: by de-identifying the reported data (e.g., anonymizing and generalizing), terminals can contribute local data to serve the collective interest of improving the overall prediction quality of the region, while strictly protecting user privacy. Furthermore, compared to real-time reporting of raw, massive data streams, the periodic reporting of summary data reduces the pressure on terminal uplink bandwidth and network-side data storage and processing resources, optimizing overall overhead.

[0107] Accordingly, the second prediction result and the corresponding confidence level are generated by the prediction device in the following way: aggregating network status data reported from multiple terminals; inputting the aggregated data into the regional group prediction model of the prediction device, and the regional group prediction model outputting the second prediction result; calculating the confidence level based on the number of terminals participating in the aggregation and the consistency of the network status data reported by each terminal.

[0108] In this embodiment, the prediction device collects and integrates terminal data reported periodically from all terminals within its coverage area. For example, the MEC server receives summary data (such as latency and throughput) reported every 30 seconds from hundreds of terminals in the same cell, and categorizes and stores it according to geographical location or cell sector to form a multi-terminal, multi-time point regional dataset.

[0109] In this embodiment of the application, network status data reported from multiple terminals is aggregated to form a group data view, which gathers the local experience of a single terminal into a macro view that can reflect the overall network load and performance trend of the region.

[0110] In this embodiment, the regional population prediction model is a machine learning model (such as a spatiotemporal graph neural network ST-GNN or a sequence model that considers inter-cell correlations) deployed on the prediction device for analyzing the spatiotemporal characteristics of regional data. For example, the MEC server inputs the regional aggregated data (time series + spatial topology) of the past 5 minutes into the model, and the model outputs the congestion probability change curve of the region in the next 2 minutes, which is the second prediction result.

[0111] In this embodiment, a regional group prediction model is used to achieve minute-level trend prediction. This model can capture the spatiotemporal patterns of congestion propagation between different cells within a region, providing more stable and forward-looking prediction information than any single-terminal perspective.

[0112] In this embodiment, the number of terminals participating in the aggregation refers to the sample size. Generally, the larger the number, the more reliable the statistical results. For example, a prediction based on data from 100 terminals has a higher confidence level than a prediction based on data from 5 terminals.

[0113] In this embodiment, data consistency refers to the degree of dispersion of similar indicators (such as latency) reported by different terminals. Small variance indicates a uniform regional state and high prediction reliability; large variance indicates a complex regional state or the presence of edge users, increasing prediction uncertainty.

[0114] In this embodiment of the application, a function for calculating the confidence level can be designed, for example, confidence level = f(sample size, 1 / data variance), which is normalized to the interval [0,1].

[0115] In this embodiment, a key quality metric for the second prediction result is provided by calculating and feeding back a confidence score. This confidence score serves as the core input for dynamic weight fusion on the terminal side, enabling the terminal to intelligently assess and quantify its reliance on the prediction information for that region. Simultaneously, the introduction of this "quality label" enhances the transparency and interpretability of the entire decision-making process. It not only allows the terminal to understand the source of the prediction result's reliability but also provides a clear quantitative basis for automatically adopting more conservative decision-making strategies (such as reducing the weighting of the prediction) in situations with insufficient data samples or poor consistency, thereby improving the overall robustness and reliability of decision-making in various complex scenarios.

[0116] As can be seen, in this embodiment of the application, a complete "data reporting" system is constructed through periodic data reporting and background processing mechanisms. Aggregate analysis Predictive generation The "quality assessment" technology loop not only provides reliable backend data and computing support for terminals to obtain high-precision regional predictions on demand, but also achieves collaborative mining and intelligent transformation of the value of group data at the overall level under the premise of strict privacy protection.

[0117] In some embodiments provided in this application, step 103 may specifically include the following steps: step 1031 and step 1032; In step 1031, based on the confidence level, a first fusion weight corresponding to the first prediction result and a second fusion weight corresponding to the second prediction result are determined; wherein, the value of the second fusion weight is positively correlated with the confidence level; and / or, the value of the first fusion weight is negatively correlated with the confidence level.

[0118] In this embodiment of the application, instead of using fixed weights during fusion, the influence of the two information sources in the final decision is dynamically adjusted based on the reliability of the regional prediction curve.

[0119] In this embodiment of the application, the first fusion weight (denoted as W) local The first fusion weight W refers to the weighting coefficient assigned to the terminal's own first prediction result (value). local It decreases as confidence level C increases, for example, W local =1-C.

[0120] In this embodiment of the application, the second fusion weight (denoted as W) region The second fusion weight (W) refers to the weighting coefficient assigned to the second prediction result (regional prediction curve) from the external source. region It increases with increasing confidence level C, for example, W region =C.

[0121] For example, the terminal receives a region prediction curve with a confidence level C = 0.8. A linear weighting rule is used: W region =C=0.8, W local =1-C=0.2. This means that in this fusion, 80% of the trust is given to regional trends, and 20% of the trust is given to instantaneous perception.

[0122] In this embodiment, an adaptive information trust strategy is implemented through a dynamic weight allocation mechanism based on confidence level. This strategy enables the terminal to intelligently assess the reliability of external regional prediction data and dynamically adjust the weight of information sources accordingly: when the regional prediction confidence is high, it tends to adopt more stable macroeconomic trends to avoid misjudgments that may be caused by instantaneous network fluctuations within the terminal itself; when the regional prediction confidence is low, it shifts its focus to real-time sensing data from the terminal's local environment to ensure that basic responsiveness is not compromised by fluctuations in the quality of external information. This mechanism enhances the robustness of fusion decision-making, giving it inherent fault tolerance to problems such as noise, latency, or insufficient samples that may exist in regional prediction data. Even if the quality of external predictions temporarily declines, the terminal can effectively isolate its negative impact on the final decision by automatically reducing its weight. Ultimately, this dynamic weighting mechanism achieves an adaptive optimization balance between decision stability (dependent on group trends) and sensitivity (dependent on local perception), thus providing a key guarantee for executing accurate predictive network switching.

[0123] In step 1032, the first prediction result and the second prediction result are weighted and fused according to the first fusion weight and the second fusion weight to obtain the fusion curve.

[0124] In this embodiment of the application, weighted fusion is a data fusion method that multiplies each input data by its corresponding weight coefficient and then combines (usually adds) them together.

[0125] In this embodiment of the application, the fusion curve is the result of weighted fusion, which is a new and comprehensive time series curve that represents the terminal's final judgment on future network congestion.

[0126] For example, continuing from the previous example, the weight has been determined to be W. local =0.2, W region =0.8. Assume the first prediction result is a numerical value P. local =0.7 (probability of congestion at some future moment), the second prediction result is a curve P region(t) First, the value P needs to be... local It can be expanded into a short-term curve component f(P) through a certain method (such as exponential decay). local Then, a weighted calculation is performed point-by-time: P fused(t) =0.2×f(P local , t)+0.8×P region(t) P obtained fused(t) This is the fusion curve.

[0127] In this embodiment, weighted fusion synthesizes prediction information from different time scales and data formats into a single fusion curve, providing a unified and authoritative predictive basis for subsequent switching decisions. This fusion process, through dynamic weight allocation, selectively strengthens the contribution of high-quality information sources while weakening the influence of low-quality or highly uncertain information sources, thereby ensuring that the final prediction curve retains and highlights the most valuable key features (whether long-term macro trends or short-term key transients). Furthermore, because the weight values ​​change continuously with confidence levels, the fusion curve achieves a smooth transition, avoiding system behavior oscillations caused by abrupt changes in the decision-making basis, thus making the overall switching process more stable and predictable.

[0128] As can be seen, in this embodiment of the application, a dynamic weighted fusion algorithm based on confidence is constructed. The algorithm first establishes an intelligent trust evaluation model, and then executes the model, thereby seamlessly and optimally integrating heterogeneous, multi-source prediction information into a high-quality fusion prediction curve.

[0129] In some embodiments provided in this application, the above-mentioned 1032 may specifically include the following steps: step 10321 and step 10322; In step 10321, the second prediction result is time-axis aligned to compensate for the time delay introduced from the generation of the second prediction result to its reception by the first terminal.

[0130] In this embodiment, considering that the regional prediction curve is generated based on a future timeline of its "generation time," while a certain period of time has passed by the time the terminal receives and prepares to use it, if alignment is not performed, the fusion may be based on an incorrect time reference, leading to prediction failure.

[0131] In this embodiment of the application, time axis alignment is a data transformation operation that shifts a time series data (curve) forward (in the future direction) along its time axis by a specific time interval.

[0132] In this embodiment of the application, the time delay refers to the time difference (Δt=t) between the moment when the prediction device generates the second prediction result and the moment when the first terminal actually receives the result. receive -t generate This delay mainly includes network transmission delay and possible processing queuing delay. In practical applications, the delay Δt can be calculated based on the timestamp carried in the second prediction result and the terminal's current system time, or it can be provided by the network side when sending data.

[0133] For example, the MEC server generates a regional prediction curve for the next 2 minutes (i.e., the interval [0 seconds, 120 seconds]) at time T=0. This curve is transmitted over the air interface and processed by the terminal, and is received by the terminal at time T=0.5. At this point, if used directly, the state predicted by the curve at time T=1 actually corresponds to the state at time T=1.5 in the real world (because the "future 1 second" assumed during generation has already passed 0.5 seconds). The alignment process involves shifting this curve backward by Δt=0.5 seconds on the time axis, making its horizontal coordinate [0.5 seconds, 120.5 seconds], so that each point on the curve correctly corresponds to the actual future time.

[0134] In this embodiment, by performing time axis alignment processing on the second prediction result, the consistency of the time base is first ensured. This is an absolute prerequisite for achieving accurate information fusion. It ensures that the local prediction based on the current time of the terminal and the regional prediction generated based on historical time can be compared and synthesized under a unified real-world time base, thus giving the fusion result a clear physical meaning. Secondly, this processing directly improves the timeliness and accuracy of the prediction, enabling the regional prediction curve to correctly reflect the future network trend starting from the current time rather than a certain time in the past. This ensures the timeliness of the fused prediction curve and provides reliable input for core forward-looking decision-making. Especially in high-dynamic network scenarios such as high-speed mobile networks, the network state may change significantly within hundreds of milliseconds. Time alignment processing becomes a necessary technical means to effectively compensate for transmission and processing delays and prevent prediction inaccuracies caused by such delays.

[0135] In step 10322, a weighted fusion is performed based on the first fusion weight, the second fusion weight, the first prediction result, and the aligned second prediction result to obtain the fusion curve.

[0136] In this embodiment, weighted fusion refers to linearly combining the expanded form of the first prediction result with the time-aligned second prediction result according to their respective weights. Since the first prediction result is usually a value for a specific future moment, while the second prediction result is a curve, the former needs to be expanded into a proximal prediction component that varies with time (denoted as f(P1, t)), for example, through constant expansion or exponential decay models.

[0137] For example, let the aligned region prediction curve be P2. aligned(t) The first prediction result, P1 = 0.7, is expanded exponentially into component f(P1, t). If the weights are fixed at W1 = 0.2 and W2 = 0.8, the fusion curve is: P fused(t) =0.2×f(P1, t)+0.8×P2 aligned(t) .

[0138] In some embodiments, the first fusion weight and the second fusion weight are time-varying weights that change over time. Accordingly, step 10322 above may specifically include the following steps: expanding the first prediction result into a proximal prediction component; weighting the proximal prediction component using a first time-varying weight function corresponding to the first fusion weight, and weighting the time-axis aligned second prediction result using a second time-varying weight function corresponding to the second fusion weight; wherein the value of the first time-varying weight function monotonically decreases with time, and the value of the second time-varying weight function monotonically increases with time; summing the weighted two results to obtain the fusion curve.

[0139] In this embodiment, considering that the reliability of information at different time scales should evolve over time, the terminal's own instantaneous perception is more important and accurate for the nearer future; while for the more distant future, macro-trends based on historical group data are more valuable for reference. The time-varying weight function is precisely designed to characterize this dynamic shift in reliability.

[0140] In this embodiment, the time-varying weights refer to weighting coefficients that are functions W1(t) and W2(t) with respect to future time t. The first time-varying weighting function W1(t) represents the confidence level of the local prediction component, and its value monotonically decreases (e.g., W1(t) = exp(-α × (t - t)). now The second time-varying weighting function W2(t) represents the confidence level of the regional prediction curve, and its value is monotonically increasing (e.g., W2(t) = 1 - exp(-β×(t - t)). now The typical fusion formula is: P fused(t) =W1(t)×f(P1,t)+W2(t)×P2 aligned(t) And it often satisfies W1(t)+W2(t)=1.

[0141] For example, let t now =0, P1=0.8, f(P1, t)=0.8×exp(-0.3×t), W1(t)=exp(-0.5×t), W2(t)=1-exp(-0.5×t). For t=0.5 seconds (short term): W1(0.5)≈0.78, W2(0.5)≈0.22, local forecasting dominates. For t=2 seconds (long term): W1(2)≈0.37, W2(2)≈0.63, regional forecasting dominates.

[0142] In this embodiment, the fusion process employs a time-varying weight function. Firstly, it adheres to the fundamental predictive principle that "short-term predictions rely on instantaneous signals, while long-term predictions rely on statistical trends," thus making the fusion result theoretically closer to the optimal estimate. Secondly, it provides more accurate and finer-grained input information for subsequent decision-making steps (e.g., extracting predicted values ​​for specific future moments from the fusion curve). Simultaneously, because the weight function changes continuously over time, the generated fusion curve exhibits excellent smoothness, avoiding abrupt changes in decision-making criteria due to weight jumps, thereby improving the stability of trend analysis and other operations based on this curve. Furthermore, the specific form of the time-varying weight function (e.g., exponential, linear, sigmoid) and its parameters are highly configurable, enabling offline optimization or online adaptive learning based on different application scenarios (e.g., high-speed rail, densely populated urban areas), thereby enhancing the environmental adaptability of the core algorithm.

[0143] As can be seen, this application embodiment constructs a fusion execution hierarchy from basic to advanced. The fixed-weight fusion method provides a clear and stable foundational framework, while the time-varying weight fusion method, by establishing a dynamic model of "information trust evolving over time," achieves a more cognitively consistent and refined utilization of predicted information. Both methods aim to efficiently and accurately transform the intelligently allocated trust into a high-quality fusion prediction curve, thereby jointly ensuring the performance of the fusion algorithm in both logical intelligence and engineering reliability dimensions. This provides crucial technical support for ultimately achieving accurate, timely, and efficient intelligent network switching.

[0144] In some embodiments provided in this application, step 104 may specifically include the following steps: step 1041, step 1042, step 1043 and step 1044; In step 1041, the predicted value of network congestion state corresponding to the future target time is extracted from the fusion curve; wherein the future target time is determined based on the estimated time required to perform network handover.

[0145] In this embodiment of the application, the switching decision is not based on the "present" or "arbitrary future", but on a future moment with clear significance that takes into account the time consumption of the switching execution process itself.

[0146] In this embodiment, the future target time is a specific future point in time, and its selection logic is: current time + estimated time required to perform network handover. The estimated time refers to the time expected to be required for the entire process from when the terminal makes a handover decision and triggers the handover process, to when a stable connection is actually established on the new network and service flow is restored. It includes signaling interaction latency, radio link establishment latency, etc.

[0147] In this embodiment of the application, the network congestion state prediction value refers to the value derived from the fusion curve P. fused(t) The value read at t = the future target time usually represents the probability or degree of network congestion at that time.

[0148] For example, based on historical data analysis or standard protocols, it is known that a complete handover in this network environment takes an average of τ = 2 seconds. The current time is T. now Then the future target time is set as T. target =T now +2 seconds. The terminal is merging from the Pf curve. used(t) Read t=T target The predicted value at time P value =0.85.

[0149] In this embodiment, forward-looking decision-making is achieved by determining the future target time based on the estimated handover execution time and extracting the predicted value of that time from the fusion curve. By aligning the decision point with the network state after the handover, the goal is to enter a better network environment rather than blindly responding to instantaneous signal fluctuations. This improves the accuracy and effectiveness of the decision-making process and avoids time mismatches and errors that may arise from premature prediction or decisions based on the current instantaneous state. It ensures precise alignment between the experience improvements brought about by the handover action and the changes in network state on the timeline. Simultaneously, it provides a unified time benchmark for subsequent cost and benefit evaluation, ensuring consistency in comparison among candidate networks under the same future scenario.

[0150] In step 1042, if the predicted network congestion state is greater than a first threshold, the switching cost of switching to a candidate network is evaluated for each of the at least one candidate network, in order to calculate the overall utility value.

[0151] In this embodiment, a switch is deemed necessary only when there is a significant risk of congestion (exceeding a first threshold) at the predicted target time, thereby initiating a cost assessment of each alternative scheme and avoiding unnecessary computational overhead.

[0152] In this embodiment of the application, the first threshold is a preset threshold value used to determine whether the predicted congestion risk is high enough to justify paying the switching cost to avoid it.

[0153] In this embodiment of the application, the switching cost refers to the sum of the multidimensional costs required for a terminal to perform an operation of switching from the current network to the target candidate network.

[0154] In some embodiments, handover costs may include at least one of the following: power consumption costs, signaling overhead costs, and service interruption risk costs. Power consumption costs include the additional energy consumed for scanning the target network, performing signaling interactions, and power calibration on the new link. Signaling overhead costs include the load on air interface resources and core network processing capabilities caused by control plane signaling generated during the handover process. Service interruption risk costs quantify the damage to the current service experience caused by potential brief service interruptions, data retransmissions, or even handover failures during the handover process.

[0155] For example, the first threshold is set to 0.7. The predicted value P extracted in step 1041... value =0.85>0.7, meeting the condition. The terminal identified three candidate networks: neighboring 5G cell A, Wi-Fi network B, and neighboring 5G cell C. For each candidate network, the terminal estimated: switching to A requires high-power scanning (high power consumption), but the signaling process is standard (medium signaling overhead), and the historical handover success rate is high (low risk); switching to Wi-Fi B is fast scanning (low power consumption), but requires new authentication (high signaling overhead), and the signal is unstable (high risk).

[0156] In this embodiment, by introducing a comprehensive evaluation of switching costs across multiple dimensions, including power consumption, signaling overhead, and service interruption risk, network switching decisions are elevated from the traditional "finding the strongest signal" to a resource optimization and allocation problem, incorporating terminal energy consumption and network signaling load into core considerations. This evaluation mechanism can be deeply integrated with the terminal's real-time status (such as remaining battery power and service type) to achieve personalized decision-making strategies. For example, when battery power is insufficient, the power consumption cost weight can be dynamically increased to guide energy saving. Furthermore, this mechanism has inherent stability guarantees: even if congestion risk is predicted, if the overall cost of all candidate networks is too high (expected net benefit is negative), the terminal can choose to maintain the current connection, thereby suppressing negative-benefit switching and enhancing connection stability and user experience consistency.

[0157] In step 1043, the switching benefits are determined based on the fusion curve.

[0158] In this embodiment, the extent to which switching to a candidate network can avoid the congestion predicted in the original network is quantified, thereby improving the user experience.

[0159] In this embodiment, the switching benefit is a quantifiable value representing the experience loss avoided or the performance improvement gained through switching. It is typically positively correlated with the predicted congestion severity at a future target time in the current network (serving network), and may also be related to the candidate network's predicted load or signal quality (if available).

[0160] For example, assuming that switching to the candidate network can completely avoid the congestion of the original network at the target time, the profit G can be directly equal to the predicted value P extracted from the fusion curve. value (e.g., 0.85), or its monotonically increasing functional relationship G=g(P) value A more refined evaluation can incorporate information from the candidate network side (if available), such as the candidate network's own load predictions.

[0161] In this embodiment of the application, the abstract goal of "improving user experience" is transformed into a specific and calculable numerical indicator by quantifying the switching benefits based on the fusion curve. This provides an operable quantitative basis for the subsequent comprehensive consideration of switching costs across multiple dimensions, ensuring that the core decision-making always focuses on solving the fundamental problem of network congestion and that the benefit assessment directly and accurately reflects the degree of congestion that can be avoided through switching.

[0162] In step 1044, the overall utility value for the candidate network is calculated based on the switching cost and switching benefit.

[0163] In this embodiment of the application, benefits and costs are measured on the same dimension, and a score representing the net value of each candidate network, namely the comprehensive utility value, is calculated.

[0164] In this embodiment, the comprehensive utility value is a scalar value used to comprehensively evaluate the advantages and disadvantages of switching to the candidate network. It is typically calculated using a utility function, which defines a mathematical model that combines benefits and multi-dimensional costs into a single indicator, for example: U=G-(λ1×C power +λ2×C signaling +λ3×C risk Where G is the switching benefit, and C is the switching benefit. power C signaling and C risk These are the switching costs, and λ1, λ2, and λ3 are the weighting coefficients for each cost.

[0165] For example, for candidate network A, its return G is estimated. A =0.85, cost C powerA =0.1, C signalingA =0.05,C riskA =0.02. Preset weights λ1=1.0, λ2=0.5, λ3=2.0 (higher weight for business interruption risk), then U A =0.85-(1.0×0.1+0.5×0.05+2.0×0.02)=0.85-0.21=0.64. Similarly, calculate the utility values ​​of other candidate networks.

[0166] In this embodiment, a comprehensive utility function is designed to integrate multiple optimization objectives, such as improving user experience, reducing terminal power consumption, reducing network signaling overhead, and ensuring service continuity, into a unified quantitative framework for collaborative optimization. This mechanism transforms each candidate network into a rankingable comprehensive utility value, simplifying network selection into a clear numerical comparison problem with direct and unambiguous decision logic. Furthermore, by dynamically adjusting the weighting coefficients of various costs within the utility function, the system can adapt to different strategy orientations, such as "prioritizing ultimate user experience" and "prioritizing terminal energy saving."

[0167] As can be seen, in this embodiment, the predicted state at the decision moment is extracted from the fusion curve in a forward-looking manner, and the costs and benefits of each candidate network are evaluated in parallel after confirming the necessity of switching. Finally, the comprehensive utility score is synthesized through a multi-objective utility function, which realizes the transformation of traditional experience-based switching decisions into a computable, optimizable, and interpretable scientific decision problem, which is the key to realizing intelligent network switching.

[0168] In some embodiments provided in this application, the network switching method may further include the following step: Step 107; In step 107, the trend of the fusion curve near the future target time is obtained.

[0169] In this embodiment of the application, a secondary verification or enhancement judgment mechanism is introduced before evaluating the comprehensive utility value. This mechanism not only cares about the static predicted value at the future target time, but also pays attention to the dynamic trend of the prediction curve near that time, that is, whether the future state tends to improve or deteriorate, providing additional information about the changing momentum for decision-making.

[0170] In this embodiment, the trend of change refers to the direction and rate of change of the fusion curve within a time window near the future target time. It is usually obtained by calculating the first derivative (slope) of the curve at that point or by analyzing the numerical changes in a small interval before and after that point.

[0171] In this embodiment of the application, the worsening network congestion state means that the trend indicates that the predicted probability or degree of network congestion will continue or accelerate for a short period of time after the target time in the future.

[0172] For example, the terminal extracts the future target time T from the fusion curve. target =Predicted value P after 2 seconds value =0.75 (greater than the first threshold of 0.7). Meanwhile, calculating the slope of the curve near t=2 seconds yields slope = +0.1 per second. This indicates that after the handover, network congestion not only remains high but continues to rise rapidly.

[0173] In this embodiment, the context-awareness of decision-making is enhanced by acquiring and analyzing the changing trend of the fusion curve near a future target time. This method introduces dynamic information in the time dimension, allowing the decision-making basis to simultaneously include the static predicted value at a future time and its evolution trend, thereby forming a more comprehensive judgment on the future behavior of the network. This trend information provides a stronger handover trigger signal: when a high predicted value and a deteriorating trend occur simultaneously, the terminal can more accurately determine the urgency and necessity of the handover, equivalent to adding a "trend confirmation" step. Simultaneously, the trend information supports fine-grained control over the timing of decisions, and can be used to determine whether to "immediately execute the handover" or "postpone observation." For example, when the predicted value is high but the trend shows that the network status is rapidly improving, the terminal can choose to delay the handover to expect the network to recover autonomously, thereby avoiding unnecessary handover operations and their associated costs.

[0174] Accordingly, step 1042 may specifically include the following steps: Step 10421; In step 10421, when the predicted value of network congestion is greater than a first threshold and the trend of change indicates that the network congestion is worsening, the switching cost of switching to a candidate network is evaluated for each of the at least one candidate network.

[0175] In this embodiment, the threshold for triggering cost assessment is increased from a single predicted value threshold to an AND condition of the "predicted value threshold" and the "deterioration trend". This is a more stringent and intelligent triggering logic.

[0176] In this application embodiment, two conditions are defined for performing subsequent complex evaluations (cost-benefit calculations): Static risk condition: The predicted value is greater than the first threshold, which confirms that there is a high risk of congestion at the target time.

[0177] Dynamic risk conditions: The changing trend indicates "deterioration", which confirms that the risk not only exists but is also intensifying, thus ruling out the short-lived peak situation of "high risk that is about to pass".

[0178] For example, continuing from the previous example, P value =0.75 (>0.7) and slope=+0.1>0, both conditions are met. Therefore, the terminal initiates a handover cost assessment for all candidate networks. Conversely, if P value =0.75 but slope=-0.2 (rapid improvement), then even if the predicted value exceeds the threshold, cost assessment will not be triggered, and the terminal may choose to continue monitoring.

[0179] In this embodiment, a dual-verification triggering mechanism based on state values ​​and changing trends effectively suppresses frequent and excessive handovers. This mechanism intelligently identifies and filters false alarms caused by brief, instantaneous congestion peaks. When the terminal determines that the network status is about to improve on its own, it maintains the current connection without triggering a handover process, thereby avoiding a large number of unnecessary and high-overhead handover operations and improving the stability of network connections and the continuity of user experience. Simultaneously, it optimizes the accuracy of decision-making and the efficiency of system resource utilization: only when network risks are simultaneously confirmed as "high-level" and "continuously deteriorating" will the subsequent complex candidate network evaluation and utility calculation process be initiated, ensuring that computing resources and decision-making logic are concentrated on the most critical scenarios, thus improving the efficiency of the overall decision-making process.

[0180] As can be seen, in this embodiment of the application, by incorporating dynamic trend analysis in the time dimension, the terminal has the advanced ability to accurately distinguish between "persistent network threats" and "transient state fluctuations". This enables the entire handover decision not only to initiate action at the right time based on prediction, but also to choose to remain silent at the right time, thereby avoiding invalid operations.

[0181] In some embodiments provided in this application, step 105 may specifically include the following steps: step 1051 and step 1052; In step 1051, the candidate network with the highest comprehensive utility value is selected as the target network from all candidate networks.

[0182] In this embodiment of the application, after completing the comprehensive utility evaluation of all candidate networks, a unique optimal execution target is determined from multiple alternatives through a simple and definite comparison rule.

[0183] In this embodiment, the target network is the network that the terminal ultimately selects and prepares to switch to. For example, there are three candidate networks, and their calculated comprehensive utility values ​​are as follows: U A =0.64, U B =0.50, U C =-0.10. The terminal selects the network with the highest overall utility value, i.e., network A (U A =0.64) as the target network.

[0184] In this embodiment of the application, the optimal selection is achieved globally by selecting the network with the highest comprehensive utility value from all candidate networks as the target, ensuring that the finally selected network is the only option with the best comprehensive benefits after quantitative evaluation, rather than simply being better than the current connection or being randomly selected.

[0185] In step 1052, if the overall utility value of the target network exceeds the second threshold, a switching operation to the target network is performed.

[0186] In this embodiment of the application, even if the optimal network is selected, the switch is not unconditional. An "absolute return" threshold is introduced to ensure that the switch is worthwhile only if the expected net return of the selected optimal solution is significant enough, given the inherent uncertainty and small cost of the switch.

[0187] In this embodiment, the second threshold is a preset absolute threshold value used to determine whether a handover is worthwhile. It represents the minimum positive utility that the terminal considers acceptable for a handover action. Only when the utility value of the target network exceeds this threshold is the handover finally approved for execution.

[0188] In this embodiment of the application, performing a handover operation means triggering the underlying Radio Access Technology (RAT) protocol stack of the terminal to initiate and complete the connection migration to the target network according to the standard process (such as the handover process defined by 3GPP).

[0189] For example, continuing the previous example, the utility value U of the selected target network A A =0.64. The preset second threshold is U. threshold =0.2. Since 0.64 > 0.2, the handover condition is met, and the terminal immediately triggers the standard handover procedure to network A. If U A The value is 0.15 (although it is still the highest among the candidate networks), but since 0.15 < 0.2, the terminal will not perform a network switch and will maintain the current network connection.

[0190] In this embodiment, a second threshold is set to avoid network switching with only marginal benefits. While such a switch may be relatively optimal among candidate networks, its net gain is minimal, and its actual value may be negative after factoring in the switching risk. The second threshold mechanism filters out such scenarios, allowing only switching operations where the expected benefit significantly outweighs the risk cost. Simultaneously, this threshold provides a safety buffer for the terminal, enhancing stability: even if the evaluation model has errors or network state fluctuations, as long as the optimal utility value does not significantly exceed the threshold, the terminal maintains its current connection, thereby suppressing frequent oscillations at the decision boundary. Furthermore, the second threshold, as a configurable parameter, supports diverse decision-making strategies: a high threshold (e.g., 0.5) corresponds to a "conservative / stability-first" mode, a low threshold (e.g., 0.05) corresponds to an "active / experience-first" mode, and a zero threshold degenerates into a "relatively optimal, switch" strategy.

[0191] As can be seen, in this embodiment of the application, the relative correctness of the selected direction is ensured by "optimal comparison"; and the absolute value and stability of the executed action are guaranteed by "absolute threshold adjudication".

[0192] Figure 2This is a flowchart illustrating a network congestion prediction method provided in some embodiments of this application. The method is executed by a prediction device (such as a network-side device like a mobile edge computing node or a cluster head terminal of a terminal cluster) to generate regional network trend predictions and assess their reliability. Figure 2 As shown, the method may include the following steps: step 201, step 202, step 203, step 204 and step 205.

[0193] In step 201, network status data reported from multiple terminals is received.

[0194] In this embodiment of the application, the prediction device can continuously or periodically collect network status summary information of multiple terminals within its service area.

[0195] In this embodiment of the application, network status data refers to the processed network performance and status summary information reported by the terminal, which is usually desensitized (e.g., anonymized, data generalized) to protect privacy.

[0196] For example, an MEC server deployed in a high-speed train carriage receives summary data reported by hundreds of passengers' mobile phones every 30 seconds, including average RTT, throughput quantiles, and fuzzy location grid codes.

[0197] In this embodiment of the application, by receiving network status data reported from multiple terminals, collective perception information is aggregated, providing a data foundation for constructing a regional network view.

[0198] In step 202, spatiotemporal aggregation is performed on the received network status data.

[0199] In this embodiment of the application, the original reported data can be processed to integrate discrete point data from different terminals and times into formatted data that can reflect the overall spatiotemporal characteristics of the region.

[0200] In this embodiment of the application, spatiotemporal aggregation is a data processing operation, including: 1) spatial aggregation: integrating multiple terminal data belonging to the same geographical area (such as the same cell or grid) within the same time period (such as calculating the average, median, and variance); 2) temporal aggregation: organizing regional aggregated data within multiple consecutive periods into a sequence according to time order.

[0201] For example, the MEC server arranges the spatial average (such as average latency) of all terminal data in carriage A area, which is taken every 30 seconds over the past 5 minutes, in chronological order to form a time series containing 10 time points that reflects the latency evolution of the area.

[0202] In this embodiment, by spatiotemporally aggregating network status data from multiple terminals, key statistical features characterizing the overall network behavior pattern of the region are extracted, reducing the dimensionality and complexity of the original data and providing refined input for the prediction model. Simultaneously, this aggregation constructs a standardized data tensor containing time-series and spatial correlation information, providing necessary data format support for directly applying spatiotemporal prediction models (such as spatiotemporal graph neural networks), which is a prerequisite for implementing regional trend prediction.

[0203] In step 203, a second prediction result is generated based on the aggregated data; wherein, the second prediction result is a regional prediction curve characterizing the change of network congestion probability over time in the second future time period.

[0204] In this embodiment of the application, aggregated spatiotemporal data is used to infer the future network state evolution trend of the region through a dedicated prediction model.

[0205] In this embodiment, the regional prediction curve is a curve with the horizontal axis representing future time (a second time period, such as 1-5 minutes in the future) and the vertical axis representing the predicted network congestion probability or degree index. It characterizes the network state trend of the entire region (rather than a single terminal).

[0206] For example, the latency time series and other features (such as throughput series and changes in the number of terminals) formed above are input into a pre-trained spatiotemporal graph neural network (ST-GNN). This model captures the mutual influence and temporal evolution patterns between cells within a region, and outputs a continuous prediction curve of the network congestion probability in the region within the next 3 minutes.

[0207] In this embodiment, by generating regional prediction curves based on aggregated data, minute-level network trend prediction is achieved, providing the terminal with a more stable long-term network state outlook than second-level short-term predictions, supporting the terminal in forward-looking connection planning. Simultaneously, by employing a prediction model capable of modeling spatial dependencies (such as a spatiotemporal graph neural network), this prediction not only reveals the temporal evolution trend of the network state but also captures the potential propagation patterns of congestion between adjacent regions, imbuing the prediction results with spatial dimension insights and enhancing the comprehensive value of regional prediction information.

[0208] In step 204, the confidence level of the second prediction result is calculated based on the number of terminals participating in the aggregation and the consistency of the network status data reported by each terminal.

[0209] In this embodiment of the application, the quality of the prediction results is evaluated to generate a quantitative index that reflects the reliability of the prediction, namely the confidence level.

[0210] In this embodiment of the application, confidence level is a quantitative indicator used to evaluate the reliability of the current regional prediction curve.

[0211] In this embodiment, the number of terminals participating in the aggregation is the sample size. Generally, the larger the sample size, the more reliable the statistical estimate and the higher the confidence level.

[0212] In this embodiment, data consistency refers to the degree of dispersion (e.g., variance) of data reported by different terminals within the same area. The more consistent the data, the more uniform the regional state, and the more reliable the prediction; the larger the variance, the more complex the regional state or the existence of edge anomalies, and the increased uncertainty of the prediction.

[0213] For example, this prediction is based on data from 50 terminals over the past 5 minutes (medium sample size). The variance of the "latency" metric reported by these terminals was calculated, revealing a small variance (indicating that most users have similar experiences). Considering both the small sample size and low variance, the confidence level of this prediction was calculated to be 0.82 (range 0-1) using a pre-defined formula.

[0214] In this embodiment, a quality assessment label is provided for the regional prediction curve by calculating a confidence level, thus attaching clear reliability information to the prediction output and supporting the terminal in subsequent decision-making. This confidence level serves as the core input parameter for dynamic weight fusion, enabling the terminal to intelligently adjust its level of trust in the regional prediction based on quantitative indicators. Furthermore, when the aggregated data sample is insufficient or has poor consistency, the confidence level reflects a decline in prediction quality, prompting the terminal to adopt a conservative strategy, thereby enhancing its adaptability and stability in scenarios with fluctuating data quality.

[0215] In step 205, the second prediction result and confidence level are sent to the requesting terminal.

[0216] In this embodiment of the application, the requesting terminal refers to the terminal that previously sent the prediction query request.

[0217] For example, the MEC server packages the generated congestion probability curve for the next 3 minutes, along with its confidence level of 0.82, and sends it to the terminal that previously initiated the request via the downlink channel.

[0218] In this embodiment of the application, by sending the second prediction result and its confidence level to the requesting terminal, not only is the prediction content (prediction curve) of the regional network status provided, but also a quantitative assessment of the reliability of the prediction result (confidence level) is provided simultaneously, providing complete and interpretable input information for the terminal to perform fusion decision-making.

[0219] As can be seen from the above embodiments, in this embodiment, by receiving data from a group of terminals, performing spatiotemporal aggregation, model prediction, and quality assessment, a prediction curve with confidence labels is finally generated and sent to the terminals. This process is the core backend service supporting the "terminal-led, network / group-assisted" intelligent handover architecture, and the high-quality prediction information it produces provides key input for terminals to make accurate and forward-looking network handover decisions.

[0220] In some embodiments provided in this application, the prediction device is a network-side device; accordingly, step 201 above may specifically include the following steps: periodically receiving network status data reported by each terminal in an anonymized manner.

[0221] As can be seen, in this embodiment, the periodic mechanism ensures the continuity and stability of the prediction model's input data, and the anonymization process protects user privacy from a technical perspective. Furthermore, compared to event-triggered reporting, the periodic reporting mode facilitates uplink resource pre-scheduling and traffic shaping on the network side, reducing signaling impact and resource contention. Simultaneously, network-side devices can leverage their infrastructure advantages to aggregate terminal data from a wide area, generating regional network prediction information with a macroscopic perspective and global representativeness.

[0222] In some embodiments provided in this application, the prediction device is a cluster head terminal, which is a terminal that forms a temporary cluster together with the first terminal and is elected by the temporary cluster; accordingly, the above step 201 may specifically include the following steps: receiving network status data sent by other terminals in the temporary cluster.

[0223] In this embodiment, under the terminal self-organizing mode, when the local short-term prediction result of the first terminal meets the preset risk conditions, it can initiate a cluster formation request to neighboring terminals via broadcast. In response to this request, terminals in close proximity participate in collaboration based on a preset admission policy (such as when the number of responses reaches a threshold), forming a temporary terminal cluster. After the cluster is established, each member terminal dynamically elects a cluster head terminal based on preset election rules, comprehensively evaluating indicators such as remaining battery power, signal quality, and computing power. This head terminal is responsible for subsequent data aggregation, prediction calculation, and information interaction with other terminals within the cluster.

[0224] As can be seen, in this embodiment, a self-organizing temporary cluster and cluster head election mechanism by terminals achieves group prediction capabilities independent of infrastructure, effectively ensuring system resilience and service availability in scenarios with missing network coverage or damaged infrastructure. Terminals directly exchange network status data using short-range communication (such as D2D and Sidelink), reducing transmission latency while avoiding data uploads to the operator's network, thus achieving privacy protection. Furthermore, this mechanism supports dynamic changes in cluster members and topology, and its prediction capability naturally enhances as the cluster size increases and the data sample size grows, exhibiting good dynamic adaptability and scalability.

[0225] In some embodiments provided in this application, step 203 may specifically include the following steps: generating a second prediction result based on the aggregated data according to a second period. The second period is longer than the first period.

[0226] As can be seen, in this embodiment, the second period (e.g., 30 seconds) is significantly longer than the first period for data reporting (e.g., 10 seconds). This setting allows the prediction device to accumulate data over a longer time window, thereby enabling more robust spatiotemporal aggregation and avoiding frequent high-load model inference. Predicting based on data over a longer period helps smooth out instantaneous fluctuations, generating more stable regional trend prediction curves, which are more consistent with predicting trends rather than instantaneous states. Simultaneously, the extended prediction period directly reduces the computational frequency and energy consumption burden on network-side devices or cluster head terminals, contributing to improved long-term operational stability.

[0227] To facilitate a comprehensive understanding of the technical solutions provided in the embodiments of this application, the following is combined with... Figure 3 The example diagram shown is used to illustrate this point.

[0228] like Figure 3 As shown, the network switching system consists of a first terminal, a prediction device, and other terminals; The first terminal executes steps 301-306 and 310-312; the prediction device executes steps 308 and 309; and the other terminals execute step 307. In step 301, the first terminal collects a local performance indicator sequence; for example, the performance indicator sequence includes at least two of the following types of indicators: application layer experience quality indicators, wireless link layer measurement indicators, and terminal motion state context indicators.

[0229] In step 302, the first terminal aggregates the collected performance index sequence.

[0230] In step 303, the first terminal uses a local short-term prediction model to process the aggregated performance index sequence and predict the local network congestion risk value.

[0231] In step 304, the first terminal performs desensitization processing on the aggregated performance index sequence and sends the desensitized network status data to the prediction device.

[0232] In step 305, the first terminal determines whether the network congestion risk value is greater than the risk threshold. If so, it proceeds to step 306; otherwise, it proceeds to step 301.

[0233] In step 306, the first terminal sends a prediction query request to the prediction device.

[0234] In step 307, other terminals send network status data to the prediction device.

[0235] In step 308, the prediction device aggregates network status data reported from multiple terminals; it uses a regional group prediction model to process the aggregated data to obtain a regional prediction curve; and it calculates the confidence level based on the number of terminals participating in the aggregation and the consistency of the network status data reported by each terminal.

[0236] In step 309, the prediction device responds to the prediction query request by returning the regional prediction curve and its confidence level to the first terminal.

[0237] In step 310, the first terminal performs a fusion process on the network congestion risk value and the regional prediction curve based on the confidence level to obtain a fused curve.

[0238] In step 311, the first terminal evaluates the overall utility value of switching from the current serving network to each candidate network based on the fusion curve, considering both the switching benefits and switching costs.

[0239] In step 312, the first terminal selects the target network and switches to the target network based on the evaluation results of the comprehensive utility value of each candidate network.

[0240] like Figure 4 and Figure 5 As shown, tests were conducted using a simulation platform / experimental network: In high-speed mobile scenarios, compared with traditional network-side driven handover schemes, the proposed solution significantly improves the reduction of network handover frequency and the congestion prediction accuracy. These results demonstrate that the proposed solution can balance energy consumption and system stability while ensuring user experience.

[0241] The network handover method provided in this application can be executed by a network handover device. This application uses the example of a network handover device executing the network handover method to illustrate the network handover device provided in this application.

[0242] The network congestion prediction method provided in this application can be executed by a network congestion prediction device. This application uses the example of a network congestion prediction device executing the network congestion prediction method to illustrate the network congestion prediction device provided in this application.

[0243] Figure 6 This is a structural block diagram of a network switching device provided in some embodiments of this application, applied to a first terminal, such as... Figure 6 As shown, the network switching device 600 may include: a first acquisition module 601, a second acquisition module 602, a fusion module 603, an evaluation module 604, and a decision module 605; The first acquisition module 601 is used to acquire a first prediction result; wherein, the first prediction result is a value that represents the network congestion status in the future first time period, which is predicted based on the historical performance index sequence of the first terminal. The second acquisition module 602 is used to acquire a second prediction result and the corresponding confidence level; wherein, the second prediction result is a regional prediction curve based on terminal group data, which characterizes the change of network congestion probability over time in the second future time period; the confidence level is used to characterize the reliability of the regional prediction curve; the duration of the second time period is longer than the duration of the first time period; The fusion module 603 is used to fuse the first prediction result and the second prediction result according to the confidence level to obtain a fusion curve; The evaluation module 604 is used to evaluate the overall utility value of switching from the current service network to at least one candidate network based on the fusion curve. The decision module 605 is used to perform a network switching decision based on the evaluation result of the comprehensive utility value of the at least one candidate network.

[0244] As can be seen from the above embodiments, by fusing millisecond / second-level short-term predictions on the terminal side and minute-level regional predictions on the network side, this embodiment can provide the terminal with fused prediction results that combine rapid response capabilities with macro-trend judgment in scenarios with drastic dynamic changes such as high-speed movement and high user density. This provides a basis for timely and accurate network handover decisions. Based on these fused prediction results, the handover utility of each candidate network is evaluated, enabling the terminal to execute predictive handover decisions. This avoids frequent handovers caused by instantaneous signal fluctuations while proactively mitigating network congestion, achieving timely, accurate, and efficient network handover in such dynamic scenarios.

[0245] Optionally, as an embodiment, the first acquisition module 601 is specifically used to organize the historical performance index sequence into a sliding window containing data from the most recent N consecutive sampling time points; where N is a positive integer; input the data in the sliding window into the terminal local short-term prediction model, and output the first prediction result by the terminal local short-term prediction model.

[0246] Optionally, as an embodiment, the historical performance index sequence includes at least two of the following categories of indices: application layer experience quality indices, wireless link layer measurement indices, and terminal motion state context indices.

[0247] Optionally, as an embodiment, the second acquisition module 602 is specifically used to send a prediction query request to the prediction device when the first prediction result meets the short-term congestion risk condition; and to receive the second prediction result and the corresponding confidence level returned by the prediction device in response to the prediction query request.

[0248] Optionally, as an embodiment, the prediction device is a network-side device or a second terminal; Wherein, if the prediction device is the second terminal, the second terminal is the cluster head terminal elected by the temporary cluster formed with the first terminal.

[0249] Optionally, as an embodiment, the network switching device 600 may further include: The second sending module is used to send network status data to the prediction device according to the first period; wherein the network status data is obtained by de-identifying the historical performance index sequence; The second prediction result and the corresponding confidence level are generated by the prediction device in the following manner: aggregating the network status data reported from multiple terminals; inputting the aggregated data into the regional group prediction model of the prediction device, and outputting the second prediction result by the regional group prediction model; calculating the confidence level based on the number of terminals participating in the aggregation and the consistency of the network status data reported by each terminal.

[0250] Optionally, as an embodiment, the fusion module 603 is specifically used to determine, based on the confidence level, a first fusion weight corresponding to the first prediction result and a second fusion weight corresponding to the second prediction result; and to perform weighted fusion on the first prediction result and the second prediction result based on the first fusion weight and the second fusion weight to obtain the fusion curve; wherein the value of the second fusion weight is positively correlated with the confidence level; and / or, the value of the first fusion weight is negatively correlated with the confidence level.

[0251] Optionally, as an embodiment, the fusion module 603 is specifically used to perform time axis alignment processing on the second prediction result to compensate for the time delay introduced by the second prediction result from generation to reception by the first terminal; and to perform weighted fusion based on the first fusion weight, the second fusion weight, the first prediction result and the aligned second prediction result to obtain the fusion curve.

[0252] Optionally, as an embodiment, the first fusion weight and the second fusion weight are time-varying weights that change over time; The fusion module 603 is specifically used to expand the first prediction result into a near-end prediction component; to weight the near-end prediction component using a first time-varying weight function corresponding to the first fusion weight, and to weight the second prediction result after time axis alignment using a second time-varying weight function corresponding to the second fusion weight; wherein the value of the first time-varying weight function decreases monotonically with time, and the value of the second time-varying weight function increases monotonically with time; and to sum the two weighted components to obtain the fusion curve.

[0253] Optionally, as an embodiment, the evaluation module 604 is specifically used to extract the predicted network congestion state value corresponding to a future target time from the fusion curve; wherein the future target time is determined based on the estimated time required to perform network handover; if the predicted network congestion state value is greater than a first threshold, for each of the at least one candidate network, the handover cost of switching to the candidate network is evaluated; the handover benefit is determined based on the fusion curve; and the comprehensive utility value for the candidate network is calculated based on the handover cost and the handover benefit.

[0254] Optionally, as an embodiment, the network switching device 600 may further include: The third acquisition module is used to acquire the change trend of the fusion curve near the future target time; The evaluation module 604 is specifically used to evaluate the switching cost of switching to the candidate network for each of the at least one candidate network when the predicted value of the network congestion state is greater than the first threshold and the trend of change indicates that the network congestion state is deteriorating.

[0255] Figure 7 This is a structural block diagram of a network congestion prediction device provided in some embodiments of this application, applied to prediction devices, such as... Figure 7 As shown, the network congestion prediction device 700 may include: a receiving module 701, an aggregation module 702, a generation module 703, a calculation module 704, and a first sending module 705; The receiving module 701 is used to receive network status data reported from multiple terminals; The aggregation module 702 is used to perform spatiotemporal aggregation on the received network status data; The generation module 703 is used to generate a second prediction result based on the aggregated data; wherein, the second prediction result is a regional prediction curve characterizing the change of network congestion probability over time in the second future time period; The calculation module 704 is used to calculate the confidence level of the second prediction result based on the number of terminals participating in the aggregation and the consistency of the network status data reported by each terminal. The first sending module 705 is used to send the second prediction result and the confidence level to the requesting terminal.

[0256] As can be seen from the above embodiments, this embodiment defines a standard, efficient, and self-consistent regional network congestion prediction service process. This process begins with the reception of data from a group of terminals, undergoes spatiotemporal aggregation, model-based prediction, and intrinsic quality assessment, and finally produces a prediction curve with a clear confidence level label, which is then distributed to the terminals. This process is the core backend service supporting the efficient operation of the entire "terminal-led, network / group-assisted" intelligent handover architecture. The high-quality, interpretable regional prediction information it produces constitutes essential key information for terminals to achieve accurate, forward-looking, and intelligent network handover decisions.

[0257] Optionally, as an embodiment, the prediction device is a network-side device; The receiving module 701 is specifically used to periodically receive network status data reported by each terminal in an anonymized manner.

[0258] Optionally, as an embodiment, the prediction device is a cluster head terminal, which is a terminal that forms a temporary cluster together with the first terminal and is elected by the temporary cluster; The receiving module 701 is specifically used to receive network status data sent by other terminals in the temporary cluster.

[0259] Optionally, as an embodiment, the generation module 703 is specifically used to generate a second prediction result based on the aggregated data according to a second cycle.

[0260] The network switching device or network congestion prediction device in the embodiments of this application can be an electronic device or a component in an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal or other devices besides a terminal. For example, the electronic device can be a mobile phone, tablet computer, laptop computer, handheld computer, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, ultra-mobile personal computer (UMPC), netbook or personal digital assistant (PDA), etc. It can also be a server, network attached storage (NAS), personal computer (PC), television (TV), ATM or self-service machine, etc. The embodiments of this application do not specifically limit it.

[0261] The network switching device or network congestion prediction device in the embodiments of this application can be a device with an operating system. The operating system can be Android, iOS, or other possible operating systems, and this application does not specifically limit it.

[0262] The network switching device provided in this application embodiment can achieve the above-mentioned functions. Figure 1 To avoid repetition, the various processes implemented in the method embodiment shown will not be described again here.

[0263] The network congestion prediction device provided in this application embodiment can achieve the above-mentioned... Figure 2 To avoid repetition, the various processes implemented in the method embodiment shown will not be described again here.

[0264] Optionally, such as Figure 8 As shown, this application embodiment also provides an electronic device 800, including a processor 801 and a memory 802. The memory 802 stores a program or instructions that can run on the processor 801. When the program or instructions are executed by the processor 801, they implement the various steps of the above-described network switching method embodiment or the various steps of the above-described network congestion prediction method embodiment, and can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0265] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0266] Figure 9 This is a schematic diagram of the hardware structure of an electronic device that implements the various embodiments of this application. The electronic device 900 includes, but is not limited to, components such as: a radio frequency unit 901, a network module 902, an audio output unit 903, an input unit 904, a sensor 905, a display unit 906, a user input unit 907, an interface unit 908, a memory 909, and a processor 910.

[0267] Those skilled in the art will understand that the electronic device 900 may also include a power supply (such as a battery) for supplying power to various components. The power supply may be logically connected to the processor 910 through a power management system, thereby enabling functions such as managing charging, discharging, and power consumption through the power management system. Figure 9 The electronic device structure shown does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements, which will not be elaborated here.

[0268] In some embodiments, when the electronic device is a first terminal, the processor 910 is configured to: acquire a first prediction result; wherein the first prediction result is a value representing the network congestion state in a future first time period, predicted based on the historical performance index sequence of the first terminal; acquire a second prediction result and a corresponding confidence level; wherein the second prediction result is a regional prediction curve representing the change of network congestion probability over time in a future second time period, predicted based on terminal group data; the confidence level is used to characterize the reliability of the regional prediction curve; the duration of the second time period is longer than the duration of the first time period; perform fusion processing on the first prediction result and the second prediction result according to the confidence level to obtain a fusion curve; evaluate the comprehensive utility value of switching from the current serving network to at least one candidate network based on the fusion curve; and execute a network switching decision based on the evaluation result of the comprehensive utility value of the at least one candidate network.

[0269] As can be seen, in this embodiment, by fusing millisecond / second-level short-term predictions on the terminal side and minute-level regional predictions on the network side, a fused prediction result with both rapid response capability and macro-trend judgment can be provided to the terminal in scenarios with drastic dynamic changes such as high-speed movement and high user density. This provides a basis for timely and accurate network handover decisions. Based on this fused prediction result, the handover utility of each candidate network is evaluated, enabling the terminal to execute predictive handover decisions. This avoids frequent handovers caused by instantaneous signal fluctuations while proactively mitigating network congestion, achieving timely, accurate, and efficient network handover in such dynamic scenarios.

[0270] Optionally, as an embodiment, the processor 910 is specifically configured to organize the historical performance index sequence into a sliding window containing data from the most recent N consecutive sampling time points; where N is a positive integer; input the data in the sliding window into the terminal local short-term prediction model, and output the first prediction result by the terminal local short-term prediction model.

[0271] Optionally, as an embodiment, the historical performance index sequence includes at least two of the following categories of indices: application layer experience quality indices, wireless link layer measurement indices, and terminal motion state context indices.

[0272] Optionally, as an embodiment, the processor 910 is specifically configured to send a prediction query request to the prediction device when the first prediction result meets the short-term congestion risk condition; and receive the second prediction result and corresponding confidence level returned by the prediction device in response to the prediction query request.

[0273] Optionally, as an embodiment, the prediction device is a network-side device or a second terminal; wherein, when the prediction device is the second terminal, the second terminal is the cluster head terminal elected by the temporary cluster formed with the first terminal.

[0274] Optionally, as an embodiment, the processor 910 is further configured to send network status data to the prediction device according to a first cycle; wherein the network status data is obtained by de-identifying the historical performance index sequence; The second prediction result and the corresponding confidence level are generated by the prediction device in the following manner: aggregating the network status data reported from multiple terminals; inputting the aggregated data into the regional group prediction model of the prediction device, and outputting the second prediction result by the regional group prediction model; calculating the confidence level based on the number of terminals participating in the aggregation and the consistency of the network status data reported by each terminal.

[0275] Optionally, as an embodiment, the processor 910 is specifically configured to determine, based on the confidence level, a first fusion weight corresponding to the first prediction result and a second fusion weight corresponding to the second prediction result; and to perform weighted fusion on the first prediction result and the second prediction result based on the first fusion weight and the second fusion weight to obtain the fusion curve; wherein the value of the second fusion weight is positively correlated with the confidence level; and / or, the value of the first fusion weight is negatively correlated with the confidence level.

[0276] Optionally, as an embodiment, the processor 910 is specifically used to perform time axis alignment processing on the second prediction result to compensate for the time delay introduced by the second prediction result from generation to reception by the first terminal; and to perform weighted fusion based on the first fusion weight, the second fusion weight, the first prediction result and the aligned second prediction result to obtain the fusion curve.

[0277] Optionally, as an embodiment, the first fusion weight and the second fusion weight are time-varying weights that change over time; The processor 910 is specifically used to expand the first prediction result into a proximal prediction component; to weight the proximal prediction component using a first time-varying weight function corresponding to the first fusion weight, and to weight the second prediction result after time axis alignment using a second time-varying weight function corresponding to the second fusion weight; wherein the value of the first time-varying weight function monotonically decreases with time, and the value of the second time-varying weight function monotonically increases with time; and to sum the two weighted components to obtain the fusion curve.

[0278] Optionally, as an embodiment, the processor 910 is specifically configured to extract a predicted network congestion state value corresponding to a future target time from the fusion curve; wherein the future target time is determined based on the estimated time required to perform network handover; if the predicted network congestion state value is greater than a first threshold, for each of the at least one candidate network, evaluate the handover cost of switching to the candidate network; determine the handover benefit based on the fusion curve; and calculate the comprehensive utility value for the candidate network based on the handover cost and the handover benefit.

[0279] Optionally, as an embodiment, the processor 910 is specifically configured to acquire the changing trend of the fusion curve near the future target time; and, when the predicted value of the network congestion state is greater than the first threshold and the changing trend indicates that the network congestion state is deteriorating, evaluate the switching cost of switching to the candidate network for each of the at least one candidate network.

[0280] In some embodiments, when the electronic device is a prediction device, the processor 910 is configured to receive network status data reported from multiple terminals; perform spatiotemporal aggregation on the received network status data; generate a second prediction result based on the aggregated data; wherein the second prediction result is a regional prediction curve characterizing the change of network congestion probability over time in a future second time period; calculate the confidence level of the second prediction result based on the number of terminals participating in the aggregation and the consistency of the network status data reported by each terminal; and send the second prediction result and the confidence level to the requesting terminal.

[0281] As can be seen, this application defines a standard, efficient, and self-consistent regional network congestion prediction service process. This process begins with the reception of data from a group of terminals, undergoes spatiotemporal aggregation, model-based prediction, and intrinsic quality assessment, and ultimately produces a prediction curve with a clear confidence level label, which is then distributed to the terminals. This process is the core backend service supporting the efficient operation of the entire "terminal-led, network / group-assisted" intelligent handover architecture. The high-quality, interpretable regional prediction information it produces constitutes crucial information indispensable for terminals to make accurate, forward-looking, and intelligent network handover decisions.

[0282] Optionally, as an embodiment, the prediction device is a network-side device; the processor 910 is specifically used to periodically receive network status data reported by each terminal in an anonymized manner.

[0283] Optionally, as an embodiment, the prediction device is a cluster head terminal, which is a terminal that forms a temporary cluster together with the first terminal and is elected by the temporary cluster; the processor 910 is specifically used to receive network status data sent by other terminals in the temporary cluster.

[0284] Optionally, as one embodiment, the processor 910 is specifically configured to generate a second prediction result based on the aggregated data according to a second cycle.

[0285] It should be understood that, in this embodiment, the input unit 904 may include a graphics processing unit (GPU) 9041 and a microphone 9042. The GPU 9041 processes image data of still images or videos obtained by an image capture device (such as a camera) in video capture mode or image capture mode. The display unit 906 may include a display panel 9061, which may be configured in the form of a liquid crystal display, an organic light-emitting diode, or the like. The user input unit 907 includes at least one of a touch panel 9071 and other input devices 9072. The touch panel 9071 is also called a touch screen. The touch panel 9071 may include a touch detection device and a touch controller. Other input devices 9072 may include, but are not limited to, physical keyboards, function keys (such as volume control buttons, power buttons, etc.), trackballs, mice, and joysticks, which will not be described in detail here.

[0286] The memory 909 can be used to store software programs and various data. The memory 909 may primarily include a first storage area for storing programs or instructions and a second storage area for storing data. The first storage area may store the operating system, application programs or instructions required for at least one function (such as sound playback, image playback, etc.). Furthermore, the memory 909 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM). The memory 909 in the embodiments of this application includes, but is not limited to, these and any other suitable types of memory.

[0287] Processor 910 may include one or more processing units; optionally, processor 910 integrates an application processor and a modem processor, wherein the application processor mainly handles operations involving the operating system, user interface, and applications, and the modem processor mainly handles wireless communication signals, such as a baseband processor. It is understood that the aforementioned modem processor may also not be integrated into processor 910.

[0288] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described network handover method embodiments or the various steps of the above-described network congestion prediction method embodiments, and can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0289] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0290] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above network switching method embodiment, or to implement the various steps of the above network congestion prediction method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0291] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0292] This application provides a computer program product, which is stored in a storage medium and executed by at least one processor to implement the various processes of the network switching method embodiment described above, or to implement the various steps of the network congestion prediction method embodiment described above, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0293] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0294] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this 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 computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0295] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A network handover method, characterized in that, The method, executed by a first terminal, includes: Obtain a first prediction result; wherein the first prediction result is a value that represents the network congestion status in the first time period, predicted based on the historical performance index sequence of the first terminal. Obtain a second prediction result and its corresponding confidence level; wherein, the second prediction result is a regional prediction curve representing the change of network congestion probability over time in the second time period, obtained based on terminal group data; the confidence level is used to characterize the reliability of the regional prediction curve; the duration of the second time period is longer than the duration of the first time period; Based on the confidence level, the first prediction result and the second prediction result are fused to obtain a fusion curve; Based on the fusion curve, evaluate the overall utility value of switching from the current service network to at least one candidate network; A network switching decision is made based on the evaluation results of the comprehensive utility value of the at least one candidate network.

2. The method according to claim 1, characterized in that, Obtaining the first prediction result includes: The historical performance index sequence is organized into a sliding window containing data from the most recent N consecutive sampling time points; where N is a positive integer; The data within the sliding window is input into the terminal's local short-term prediction model, and the terminal's local short-term prediction model outputs the first prediction result.

3. The method according to claim 1 or 2, characterized in that, The historical performance index sequence includes at least two of the following categories: application layer experience quality indexes, wireless link layer measurement indexes, and terminal motion state context indexes.

4. The method according to claim 1, characterized in that, The process of obtaining the second prediction result and the corresponding confidence level includes: If the first prediction result meets the short-term congestion risk condition, a prediction query request is sent to the prediction device. Receive the second prediction result and the corresponding confidence level returned by the prediction device in response to the prediction query request.

5. The method according to claim 4, characterized in that, The prediction device is a network-side device or a second terminal; Wherein, if the prediction device is the second terminal, the second terminal is the cluster head terminal elected by the temporary cluster formed with the first terminal.

6. The method according to claim 4 or 5, characterized in that, The method further includes: According to the first cycle, network status data is sent to the prediction device; wherein, the network status data is obtained by de-identifying the historical performance index sequence; The second prediction result and its corresponding confidence level are generated by the prediction device in the following manner: Aggregate the network status data reported from multiple terminals; The aggregated data is input into the regional population prediction model of the prediction device, and the regional population prediction model outputs the second prediction result. The confidence level is calculated based on the number of terminals participating in the aggregation and the consistency of the network status data reported by each terminal.

7. The method according to claim 1, characterized in that, The step of fusing the first prediction result and the second prediction result based on the confidence level to obtain a fusion curve includes: Based on the confidence level, determine the first fusion weight corresponding to the first prediction result and the second fusion weight corresponding to the second prediction result; Based on the first fusion weight and the second fusion weight, the first prediction result and the second prediction result are weighted and fused to obtain the fusion curve; Wherein, the value of the second fusion weight is positively correlated with the confidence level; and / or, the value of the first fusion weight is negatively correlated with the confidence level.

8. The method according to claim 7, characterized in that, The step of weighting and fusing the first prediction result and the second prediction result according to the first fusion weight and the second fusion weight to obtain the fusion curve includes: The second prediction result is time-axis aligned to compensate for the time delay introduced from the generation of the second prediction result to its reception by the first terminal; Based on the first fusion weight, the second fusion weight, the first prediction result, and the aligned second prediction result, a weighted fusion is performed to obtain the fusion curve.

9. The method according to claim 8, characterized in that, The first fusion weight and the second fusion weight are time-varying weights that change over time; The weighted fusion based on the first fusion weight, the second fusion weight, the first prediction result, and the aligned second prediction result to obtain the fusion curve includes: The first prediction result is expanded into a near-end prediction component; The near-end prediction component is weighted using a first time-varying weight function corresponding to the first fusion weight, and the second prediction result after time axis alignment is weighted using a second time-varying weight function corresponding to the second fusion weight; wherein, the value of the first time-varying weight function decreases monotonically with time, and the value of the second time-varying weight function increases monotonically with time; The weighted sums are then summed to obtain the fusion curve.

10. The method according to claim 1, 8, or 9, characterized in that, The evaluation of the overall utility value of switching from the current service network to at least one candidate network based on the fusion curve includes: From the fusion curve, the predicted value of network congestion state corresponding to the future target time is extracted; wherein, the future target time is determined based on the estimated time required to perform network handover; If the predicted network congestion state value is greater than a first threshold, the switching cost of switching to the candidate network is evaluated for each of the at least one candidate network. The switching benefits are determined based on the fusion curve. Calculate the overall utility value for the candidate network based on the switching cost and the switching benefit.

11. The method according to claim 10, characterized in that, The method further includes: Obtain the trend of the fusion curve near the future target time; When the predicted network congestion state value is greater than a first threshold, the step of evaluating the switching cost to each of the at least one candidate network includes: If the predicted network congestion state is greater than the first threshold and the trend of change indicates that the network congestion state is deteriorating, the switching cost of switching to the candidate network is evaluated for each of the at least one candidate network.

12. A network congestion prediction method, characterized in that, Performed by a prediction device, the method includes: Receive network status data reported from multiple terminals; The received network status data is spatiotemporally aggregated; Based on the aggregated data, a second prediction result is generated; wherein, the second prediction result is a regional prediction curve characterizing the change of network congestion probability over time in the second future time period; The confidence level of the second prediction result is calculated based on the number of terminals participating in the aggregation and the consistency of the network status data reported by each terminal. The second prediction result and the confidence level are sent to the requesting terminal.

13. The method according to claim 12, characterized in that, The prediction device is a network-side device; The receiving of network status data reported from multiple terminals includes: It periodically receives network status data reported by each terminal in an anonymized manner.

14. The method according to claim 12, characterized in that, The prediction device is a cluster head terminal, which is a terminal that forms a temporary cluster with the first terminal and is elected by the temporary cluster. The receiving of network status data reported from multiple terminals includes: Receive network status data sent by other terminals within the temporary cluster.

15. The method according to claim 12, characterized in that, The generation of a second prediction result based on the aggregated data includes: According to the second cycle, a second prediction result is generated based on the aggregated data.

16. A network switching device, characterized in that, Applied to a first terminal, the device includes: The first acquisition module is used to acquire a first prediction result; wherein the first prediction result is a value that represents the network congestion status in the future first time period, which is predicted based on the historical performance index sequence of the first terminal. The second acquisition module is used to acquire a second prediction result and the corresponding confidence level; wherein, the second prediction result is a regional prediction curve that represents the change of network congestion probability over time in the second time period, obtained based on terminal group data; the confidence level is used to characterize the reliability of the regional prediction curve; the duration of the second time period is longer than the duration of the first time period; The fusion module is used to fuse the first prediction result and the second prediction result according to the confidence level to obtain a fusion curve; An evaluation module is used to evaluate the overall utility value of switching from the current service network to at least one candidate network based on the fusion curve. The decision module is used to perform network switching decisions based on the evaluation results of the comprehensive utility values ​​of the at least one candidate network.

17. A network congestion prediction device, characterized in that, Applied to a prediction device, the device includes: The receiving module is used to receive network status data reported from multiple terminals; The aggregation module is used to perform spatiotemporal aggregation on the received network status data; The generation module is used to generate a second prediction result based on the aggregated data; wherein, the second prediction result is a regional prediction curve characterizing the change of network congestion probability over time in the second future time period; The calculation module is used to calculate the confidence level of the second prediction result based on the number of terminals participating in the aggregation and the consistency of the network status data reported by each terminal. The first sending module is used to send the second prediction result and the confidence level to the requesting terminal.

18. An electronic device, characterized in that, The electronic device includes a processor and a memory, the memory storing programs or instructions that can run on the processor, the programs or instructions being executed by the processor to implement the steps of the network switching method as described in any one of claims 1 to 11, or to implement the steps of the network congestion prediction method as described in any one of claims 12 to 15.

19. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the network switching method as described in any one of claims 1 to 11, or the steps of the network congestion prediction method as described in any one of claims 12 to 15.

Images