Decision-making method and system for dynamic network switching of eSIM card

Abstract

The invention relates to a decision-making method and system for dynamic network switching of an eSIM (Embedded Subscriber Identity Module) card. The method comprises the following steps: S1, acquiring multi-modal data in real time, and performing preprocessing scoring; s2, dynamic weight distribution is carried out; s3, network switching is carried out; and S4, federal learning self-optimization is carried out. Through multi-modal data real-time acquisition, dynamic weight distribution and federated learning self-optimization mechanisms, the problems of frequent switching, neglect of user economic requirements and lack of dynamic optimization capability caused by signal fluctuation in the traditional network switching technology are solved; the method has the advantages of dynamically adapting to network environment change, multi-dimensionally optimizing a network selection strategy and improving long-term performance through autonomous evolution, a'perception-decision-execution-evolution 'closed loop is finally formed, user experience and economical cost are dynamically balanced, and limitation of traditional static decision is broken through.

Description

Technical Field

[0001] The present invention relates to the field of mobile communication technologies, and in particular to a decision method and system for dynamic network switching of an eSIM card. Background Art

[0002] In the field of mobile communications, "Handover" (HO) refers to the technical process of transferring the current network connection of a terminal device (such as a mobile phone, tablet, etc.) from one base station (or cell) to another base station (or cell) during movement.

[0003] Currently, traditional network handover technologies rely primarily on hard handover strategies with fixed thresholds. For example, the trigger condition relies solely on a single signal strength indicator, and the handover target is a pre-configured fixed backup network (such as the highest-priority partner operator network). The implementation is based on a "break-before-make" approach, resulting in brief service interruptions. This hard handover strategy has the following significant limitations: (1) Poor adaptability of static thresholds: When the signal strength fluctuates around the threshold, unnecessary switching will be frequently triggered, resulting in a decline in user experience.

[0004] (2) Single decision-making dimension: Existing technologies only focus on the single indicator of signal strength, completely ignoring key factors such as user package charges (such as traffic excess charges, operator priority) and service types (such as video, voice, and low latency requirements), resulting in a mismatch between economic efficiency and service needs.

[0005] (3) Lack of self-evolution capability: The model is rigid for a long time and cannot be globally and dynamically optimized according to the environment or user local operations, resulting in the switching strategy being out of touch with the real-time environment.

[0006] In view of the above problems, the existing technology is in urgent need of improvement. Summary of the Invention

[0007] In view of the above-mentioned defects in the prior art, a first aspect of the present invention provides a decision method for dynamic network switching of an eSIM card, which includes the following steps: Step S1: Real-time collection and pre-processing of multimodal data; the server continuously collects signal quality parameters and package information of user terminal devices and identifies the service type and its QoS (Quality of Service) requirements, and performs a comprehensive score based on the collected data information to obtain a comprehensive signal quality score S signal , tariff cost score S cost And business matching score S qos ; Step S2: Dynamic weight allocation: define the weight factors, determine the comprehensive utility value U of each available network and determine the network priority; Among them: comprehensive utility value U=αS signal +βS cost +γS qos ; α, β, and γ are the weight factors of signal quality, tariff cost, and service matching, respectively; Step S3: Network switching: triggering a network interface switching instruction to switch to the optimal network; Step S4: Federated learning self-optimization; based on user feedback and changes in the network environment, the weight factor is fine-tuned locally, and only the updated weight factor is encrypted and uploaded to the cloud center. The cloud center generates a new global weight factor and sends it to all terminal devices, that is, the weight distribution is dynamically optimized for all user devices at the same time through the federated learning mechanism.

[0008] In the aforementioned decision method for dynamic network switching of an eSIM card, optionally, step S1 specifically includes the following steps: Step S1.1: Acquire signal quality parameters according to a preset frequency, where the signal quality parameters include at least one of the following: Reference Signal Received Power (RSRP), Reference Signal Received Quality (RSRQ), and Signal-to-Interference-plus-Noise Ratio (SINR). The reference signal received power RSRP, reference signal received quality RSRQ, and signal to interference plus noise ratio SINR are scored, and the comprehensive signal quality score S is obtained by combining the three scores and adding weights. signal ; Step S1.2: Parse the user's eSIM card package information, which includes at least one of the following: data limit, roaming charges, and priority rules, and score the package cost using the following formula: Tariff cost score S cost =(total package traffic / remaining traffic×100)×w1+(1 / unit price×100)×w2; Among them, w1 is the weight of the remaining traffic; w2 is the weight of the tariff unit price; the tariff unit price is the tariff within the package; the higher the proportion of remaining traffic, the higher the score, and the lower the tariff unit price, the higher the score; Step S1.3: Determine the service type of the user's current eSIM card based on the port number and data packet size distribution, and output the QoS indicator requirement level; The business matching score S qos The calculation of the key QoS indicators of the business type is combined to set the weight w i , using the following formula: The ratio of negative indicators = demand threshold / actual value i ; Ratio of positive indicators = actual value i / the ratio of the required threshold. If the actual value of the positive indicator exceeds the required threshold, the ratio result will be directly calculated as 1.0 to avoid excessive rewards; Then the weighted sum of the ratio of negative indicators to positive indicators is calculated. Business matching score S qos =(ratio 带宽 ×w 带宽 +Ratio 时延 ×w 时延 +Ratio 抖动 ×w 抖动 )×100, the maximum score is 100 points; Among them, w represents the weight coefficient of each QoS indicator. At the same time, these weight coefficients satisfy w 带宽 +w 时延 +w 抖动 =1.

[0009] In the aforementioned decision method for dynamic network switching of an eSIM card, optionally, in step S1.1, the scoring is based on the following logic and formula: When the RSRP value is less than or equal to -118 dBm, the signal is determined to be unavailable and the RSRP score is set to 0. When the SINR value is less than or equal to -5dB, the SINR score is 0; When the reference signal reception quality RSRQ is less than or equal to -19dB, the reference signal reception quality RSRQ score is 0; The specific formula is as follows: RSRP score = max(0, min(100, (RSRP + 118) × 2.083)) Signal-to-interference-and-noise ratio (SINR) score = max(0, min(100, (SINR+5)×4)) Reference signal reception quality RSRQ score = max(0, min(100, (RSRQ+19)×7.14)).

[0010] In the decision method for dynamic network switching of an eSIM card as described above, optionally, in step S1.3, the service type includes at least one of the following: video, voice, and instant messaging; and the QoS indicator includes at least one of the following: bandwidth, latency, and jitter.

[0011] In the aforementioned decision method for dynamic network switching of an eSIM card, optionally, in step S4, the three weight factors α, β, and γ are fine-tuned based on each user's own historical usage records; in, If user A often experiences video freezes on the high-speed train, the value of α will be automatically increased; If user B's package traffic is frequently overspent, the value of β will be automatically increased; If user C experiences video freezes due to insufficient network bandwidth, the value of γ is automatically increased.

[0012] In the decision-making method for dynamic network switching of an eSIM card as described above, optionally, the user feedback includes at least one of the following: manual network switching and complaints about network freezes; privacy protection is performed during step S4, and the user's privacy data is not uploaded to the cloud center. The privacy data includes at least one of the following: location and package usage, and only the updated amount of the weight factor modification is uploaded.

[0013] In the decision method for dynamic network switching of the eSIM card as described above, optionally, the cloud center classifies the updated weight factors by scenario and performs weighted average, and the terminal device with a larger data volume has a higher weight ratio.

[0014] In order to achieve the above-mentioned purpose, the second aspect of the present invention provides a decision system for dynamic network switching of an eSIM card, wherein a decision method as described in any one of the first aspects is used, including a server, a terminal device and a cloud center that are mutually communicated with each other.

[0015] In order to achieve the above-mentioned purpose, the third aspect of the present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor runs the program, the decision-making method as described in any one of the first aspects above is implemented.

[0016] In order to achieve the above-mentioned purpose, the fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions or a computer program, and when the computer-executable instructions or the computer program are processed and executed, the decision-making method as described in any one of the above-mentioned first aspects is implemented.

[0017] The present invention provides an eSIM card dynamic network switching decision-making method and system, which solve the problems of frequent switching caused by signal fluctuations, neglect of user economic needs, and lack of dynamic optimization capabilities in traditional network switching technologies through real-time multimodal data collection, dynamic weight allocation, and federated learning self-optimization mechanism. It has the advantages of dynamically adapting to changes in the network environment, optimizing network selection strategies in multiple dimensions, and autonomously evolving to improve long-term performance, ultimately forming a "perception-decision-execution-evolution" closed loop, dynamically balancing user experience and economic costs, and breaking through the limitations of traditional static strategies.

[0018] In summary, the present invention achieves dual optimization of economy and experience. Through the dynamic weight adjustment mechanism, it solves the environmental adaptability problem of the existing fixed-weight model; further improves the stability of the network, and reduces the service interruption rate by 60% in scenarios with fluctuating signal quality; enables the model to self-evolve to adapt to changes in the network environment (such as new base station deployment, tariff package updates, etc.); and protects the model through local deployment, enabling continuous model learning while ensuring privacy.

[0019] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of an embodiment of a decision method for dynamic network switching of an eSIM card provided by the present invention; Figure 2 yes Figure 1 Specific flow diagram of step S1; Figure 3 yes Figure 1 Specific flow diagram of step S2; Figure 4 yes Figure 1 Specific flow diagram of step S3; Figure 5 yes Figure 1 Specific flow diagram of step S4; Figure 6 This is a logical diagram of the self-optimization of the federated learning of the client-side model in step S4. DETAILED DESCRIPTION

[0021] In order to make the technical means, creative features, objectives and effects of the invention easier to understand, the invention is further described below with reference to specific diagrams. However, the invention is not limited to the following implementation cases.

[0022] It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings in this specification are only used to match the contents disclosed in the specification so that people familiar with this technology can understand and read them. They are not used to limit the conditions under which the present invention can be implemented. Therefore, they have no substantive technical significance. Any modification of the structure, change in the proportion relationship or adjustment of the size should still fall within the scope of the technical content disclosed in the present invention without affecting the efficacy and purpose that can be achieved by the present invention.

[0023] Terms such as “comprise” and “include” indicate that in addition to the components directly and explicitly stated in the description and claims, the technical solution of the present invention does not exclude the situation where it has other components that are not directly or explicitly stated.

[0024] In addition, the term "and / or" in this document simply describes an association between related objects, indicating that three possible relationships exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Furthermore, the character " / " in this document generally indicates that the related objects are in an "or" relationship.

[0025] During the research and development process, the inventors of this application discovered that the following situations would lead to serious problems if there were no better solutions: First, it cannot adapt to the dynamically changing network environment. When the signal strength fluctuates near the threshold, it will cause frequent and unnecessary network switching, seriously affecting the user experience.

[0026] Second, existing technologies focus solely on signal strength, completely ignoring key factors such as user package pricing (such as data overage charges and operator priority) and service type (such as video, voice, and low-latency requirements). This single-dimensional decision-making approach makes it difficult to meet users' diverse demands for cost-effectiveness and service experience.

[0027] When user terminals are located in areas with overlapping coverage from multiple operators, network selection based solely on physical layer metrics like RSRP and SINR fails to balance the economic demands of rate-sensitive services with the QoS requirements of real-time interactive services. This single-dimensional decision-making mechanism struggles to adapt to dynamically changing network environments, leading to a mismatch between network switching decisions and actual user needs. This manifests as systemic flaws such as mishandling to high-cost networks and the failure of QoS guarantees for critical services.

[0028] For example, in a high-speed rail mobile scenario, user terminals simultaneously access 5G networks from three carriers. When a train crosses the edge of base station coverage, traditional algorithms trigger network switching based solely on signal strength. This could potentially switch video conferencing services to a carrier network with a higher RSRP value but insufficient remaining data traffic, resulting in excessive data charges. Meanwhile, another carrier network, while still meeting video transmission requirements despite its RSRP dropping to -110dBm, is excluded from the candidate list because the algorithm fails to consider the remaining data traffic weight of the plan. Such misjudgments directly lead to increased service interruption rates and unmanageable user costs, exposing the adaptability limitations of static threshold mechanisms in complex scenarios.

[0029] Finally, existing network switching algorithms lack the ability to evolve autonomously and are unable to dynamically optimize based on user behavior feedback and network environment changes (such as the addition of new 5G base stations and adjustments to operator pricing policies). This static model exhibits significant lack of adaptability in rapidly changing network environments, severely restricting long-term performance improvements.

[0030] In summary, if the above problems are not resolved, the correlation between multimodal network parameters will not be effectively converted into a basis for handover decisions, resulting in a continuous deterioration in network resource utilization and user experience. The coupling effect of signal quality fluctuations and tariff policy changes will lead to an abnormal increase in handover frequency, thereby triggering signaling storms and core network load imbalance. The lack of service type identification will exacerbate the mismatch between network service capabilities and service needs, especially in URLLC service scenarios, which may cause millisecond-level delays to exceed the standard. The long-term lack of self-optimization mechanisms will cause generational differences in network handover strategies and dynamic environments, ultimately leading to an irreversible decline in mobile network service quality.

[0031] like Figures 1 to 5 As shown, the present invention provides a decision method for dynamic network switching of an eSIM card, which may include the following steps: Step S1: Real-time multimodal data collection and preprocessing scoring. This involves simultaneously collecting signal quality parameters, user package information, and service type data, and converting these data into comparable quantitative scores using a pre-set scoring algorithm.

[0032] In step S1, the server continuously collects the signal quality parameters and package information of the user terminal device and identifies the service type and its QoS requirements, and performs a comprehensive score based on the collected data information to obtain a comprehensive signal quality score S signal , tariff cost score S cost And business matching score S qos By integrating three indicators: signal strength, economic cost, and service quality, the one-sidedness of decision-making based on a single signal indicator can be solved.

[0033] In an optional embodiment, if Figure 2 As shown, step S1 may further include the following steps: Step S1.1: Obtain signal quality parameters according to the preset frequency. The signal quality parameters may include reference signal received power RSRP, reference signal received quality RSRQ, signal to interference and noise ratio SINR, etc. Then, score the reference signal received power RSRP, reference signal received quality RSRQ, and signal to interference and noise ratio SINR, and comprehensively weight the three scores to obtain a comprehensive signal quality score S signal It should be noted that the reference signal refers to multiple signal samples of the terminal device collected by the local server, and the local server in this embodiment can be a lightweight AI model integrated in the terminal device.

[0034] Optionally, the acquisition frequency of the signal quality parameters may be set to once per second or dynamically adjusted according to the network environment, for example, automatically increased to twice per second when the moving speed exceeds 60 kilometers per hour.

[0035] In this embodiment, the scoring in step S1.1 is based on the following logic and formula: When the RSRP value is less than or equal to -118 dBm, the signal is determined to be unavailable and the RSRP score is set to 0. When the SINR value is less than or equal to -5dB, the SINR score is 0; When the reference signal reception quality RSRQ is less than or equal to -19dB, the reference signal reception quality RSRQ score is 0; The specific formula is as follows: RSRP score = max(0, min(100, (RSRP + 118) × 2.083)) Signal-to-interference-and-noise ratio (SINR) score = max(0, min(100, (SINR+5)×4)) Reference signal reception quality RSRQ score = max(0, min(100, (RSRQ+19)×7.14)).

[0036] In this embodiment, the setting of a critical threshold is used to determine whether the signal is available. For example, -118dBm of RSRP is used as the demarcation point for signal availability. When the signal parameter is lower than the threshold, it is directly assigned a score of 0 to exclude invalid signals. For valid signal parameters, they are mapped to a standardized scoring interval of 0-100 through a linear conversion formula. For example, the conversion coefficient of RSRP of 2.083 is designed to linearly map the original value interval of -118dBm to -73dBm to 0-100 points. Similarly, the conversion coefficient of SINR of 4 corresponds to the original value interval of -5dB to 20dB, and the conversion coefficient of RSRQ of 7.14 corresponds to the original value interval of -19dB to -5dB. The calculation results are bilaterally limited by the max and min functions to avoid exceeding the standard scoring range. The scoring formulas for different signal parameters are designed with differentiated coefficients so that each parameter is superimposable within the same scoring interval, providing a normalized basis for subsequent comprehensive calculations.

[0037] Specifically, the scoring process of signal quality parameters is decomposed into three steps: threshold determination, linear conversion, and range limiting. First, invalid signals are filtered out through a preset critical threshold. For example, when the RSRP is lower than -118dBm, 0 points are directly output. For valid signal values, a linear conversion formula is used for quantification. For example, in the RSRP conversion formula, every 1dBm increase in the original value can increase the score by 2.083 points until the upper limit of 100 points is reached. In this process, the conversion coefficient is accurately calculated as 100 / (highest value-lowest value). For example, the difference between the highest RSRP value of -73dBm and the lowest value of -118dBm is 45dBm, so the coefficient is 100 / 45≈2.083. The min function is then used to limit the conversion result to within 100 points, and the max function is used to ensure that the score is not lower than 0 points. This step-by-step approach allows for precise quantification of weak signal differences. For example, when RSRP increases from -118dBm to -117dBm, the score increases from 0 to 2.083 points. Traditional methods are unable to distinguish signal quality differences in this critical region. By unifying the scores of the three signal parameters into the same dimensional range, the subsequent weighted calculation of the comprehensive signal quality score is comparable and operational, avoiding calculation errors caused by directly combining parameters of different dimensionalities.

[0038] The comprehensive signal quality score is calculated by comprehensively analyzing the three key indicators of RSRP, SINR and RSRQ, and a certain weight can be set. For example, in this embodiment, the weight of RSRP is 0.6, the weight of RSRQ is 0.2, and the weight of SINR is 0.2. signal =RSRP score × 0.6 + SINR score × 0.2 + RSRQ score × 0.2.

[0039] Through the above-mentioned technical solution, the present invention achieves accurate quantitative assessment of signal quality parameters. By setting a critical threshold and a linear conversion formula, subtle differences in signal quality can be effectively distinguished, avoiding the misjudgment problem near the critical value caused by traditional fixed threshold strategies. The linear mapping method enables the scoring results to continuously reflect changes in signal quality, improving the accuracy of network status assessment. In addition, the unified 0-100 point scoring standard makes different signal quality parameters comparable, providing a reliable data foundation for subsequent comprehensive signal quality assessment, thereby improving the reliability and adaptability of network switching decisions.

[0040] Step S1.2: Parse the user's eSIM card package information, which may include traffic limits, roaming charges, priority rules, etc., and score the package cost using the following formula: Tariff cost score S cost =(Total package traffic / Remaining traffic × 100) × w1 + (1 / Unit price × 100) × w2. w1 is the weight of remaining traffic; w2 is the weight of unit price; Unit price is the package fee. A higher weight for remaining traffic indicates a higher score, while a lower unit price indicates a higher score.

[0041] Package information can be obtained through the operator's API or local database. During the package information parsing process, the remaining traffic weight w1 is set to a value range of 0.6 to 0.8, and the tariff unit price weight w2 is set to a value range of 0.2 to 0.4, ensuring that traffic usage status plays a dominant role in the score.

[0042] Preferably, the remaining traffic weight w1 is set to 0.6, and the tariff unit price weight w2 is set to 0.4.

[0043] Step S1.3: Determine the service type of the user's current eSIM card by port number and data packet size distribution, and output the QoS indicator requirement level. Service matching score S qos It is used to quantify the degree to which the candidate network meets the current business needs. Its calculation is combined with the key QoS indicators of the business type to set the weight w i , using the following formula: The ratio of negative indicators = demand threshold / actual value i ; Ratio of positive indicators = actual value i / the ratio of the required threshold. If the actual value of the positive indicator exceeds the required threshold, the ratio result will be directly calculated as 1.0 to avoid excessive rewards; Then the weighted sum of the ratio of negative indicators to positive indicators is calculated. Business matching score S qos =(ratio 带宽 ×w 带宽 +Ratio时延 ×w 时延 +Ratio 抖动 ×w 抖动 )×100, the maximum score is 100 points; Among them, w represents the weight coefficient of each QoS indicator. At the same time, these weight coefficients satisfy w 带宽 +w 时延 +w 抖动 =1.

[0044] Specifically, service types are identified by analyzing packet size distribution. For example, video packets typically exceed 1500 bytes and are distributed continuously, while instant messaging packets exhibit intermittent distribution characteristics of less than 500 bytes. Service types can include video, voice, and instant messaging. QoS can include bandwidth, latency, and jitter. For example, the required QoS level for video streaming requires bandwidth > 5Mbps and latency < 100ms.

[0045] This embodiment can calculate the matching score according to different business types. Taking video streaming as an example, Assume that the actual network bandwidth is 8Mbps and the latency is 80ms. Bandwidth is a positive indicator (the larger the value, the better), latency is a negative indicator (the smaller the value, the better), and similarly, jitter is also a negative indicator.

[0046] ratio 带宽 =Actual value / demand threshold= 8 / 5 = 1.6. If the actual value exceeds the demand, the ratio of the two will be directly calculated as 1.0 to avoid excessive rewards.

[0047] ratio 时延 =Demand threshold / actual value = 100 / 80 = 1.25.

[0048] Business matching score S qos =(1.0×0.6+1.25×0.4)×100=110. Since the final score exceeds 100 points, it is also calculated as 100 points.

[0049] Step S2: Dynamic weight allocation.

[0050] like Figure 3 As shown, this step can dynamically adjust the multi-dimensional decision weights according to user preferences and real-time scenarios, define weight factors, determine the comprehensive utility value U of each available network, determine the network priority, and sort the available networks to generate a priority list.

[0051] Among them: comprehensive utility value U=αS signal +βS cost +γSqos ; α, β, and γ are the weight factors of signal quality, tariff cost, and service matching degree, respectively.

[0052] Optionally, in the weight distribution phase, for the video stream, α=0.4, β=0.2, γ=0.4 can be initially set. For each available network, use the formula U=0.4S signal +0.2S cost +0.4S qos Calculate the comprehensive utility value U. Then sort the networks from high to low according to the U value to generate a priority list.

[0053] Assume that a user device has multiple different eSIM profile network configurations. Each profile (the digital carrier of the operator's network configuration, that is, the software configuration file of the eSIM card) corresponds to a different operator or a different network (4G, 5G). Calculate the comprehensive utility values U1, U2, and U3 of each profile network respectively, and select the optimal profile network.

[0054] Step S3: Network switching.

[0055] like Figure 4 As shown, triggering network interface switching instructions (such as disableProfile, enableProfile) to seamlessly switch to the optimal network.

[0056] Furthermore, during the network switching phase, if the current network is not the optimal network in the list, and the U value of the optimal network is higher than the current network by a certain threshold (which can be set by the user, such as 10%), a switching instruction is triggered. During the switching process, the new network connection is first established, and then the old network connection is disconnected, to achieve seamless switching.

[0057] Step S4: Federated learning self-optimization.

[0058] like Figures 5 and 6 As shown in Figure 2, this step continuously optimizes model parameters based on user feedback and environmental changes to improve long-term decision-making performance.

[0059] Specifically, step S4 fine-tunes the weight factor locally based on user feedback and changes in the network environment, and only encrypts and uploads the updated weight factor to the cloud center. The cloud center generates a new global weight factor and sends it to all terminal devices, that is, dynamically optimizes the weight distribution of all user devices at the same time through the federated learning mechanism.

[0060] In step S4, the three weight factors α, β, and γ are fine-tuned based on each user's historical usage records. in, If user A often experiences video freezes on the high-speed train, the α value will be automatically increased, giving more attention to signal stability. If user B frequently exceeds the data usage limit, the value of β will be automatically increased, and the user will pay more attention to the price. If user C experiences video freezes due to insufficient network bandwidth, the value of γ is automatically increased, and more emphasis is placed on business matching.

[0061] Federated learning uses the method of "decentralized training and centralized aggregation" to allow all user devices to jointly optimize the weight factor parameters (α, β, γ) of the local model while protecting user privacy.

[0062] Among them, historical usage records are parsed into problem events in specific scenarios, such as video freezes, traffic overspending, or insufficient bandwidth, and a mapping relationship between event types and weight factors is established. When a user is detected to have video freezes in a high-mobility scenario, the adjustment range of the signal quality weight α can be set based on the freeze frequency. For example, for every 10% increase in the number of freezes, α increases by 0.1; when the user's package traffic consumption rate exceeds the preset threshold, the adjustment amount of the tariff cost weight β can be linearly related to the overspending ratio. For example, for a 20% overspending, β increases by 0.15; when the service bandwidth demand is not met, the adjustment value of the service matching weight γ is calculated based on the actual bandwidth gap. For example, for every 5Mbps increase in the bandwidth gap, γ increases by 0.05. The above adjustment process is completed on the local terminal, and the adjusted weight factor is uploaded to the federated learning framework through an encrypted channel and aggregated and optimized with the global model parameters.

[0063] Specifically, in high-speed rail scenarios, when a user terminal detects three consecutive instances of video playback freezes and positioning data indicates a speed exceeding 200 km / h, an adjustment mechanism for the signal quality weight α is automatically triggered. Based on the historical proportion of high-speed rail freezes to the total number of freezes, for example, if the proportion exceeds 70%, the α value is gradually increased from the initial 0.4 to 0.6. When the adjusted α value is used in the comprehensive utility calculation, the signal quality score contributes 30%, prompting network handover decisions to prioritize base stations with stable signals. Simultaneously, the updated α value is differentially privacy processed and uploaded to the cloud to participate in the global weight aggregation of the federated learning model. This process not only preserves the local privacy of user behavior data but also enables personalized optimization of network selection strategies through dynamic adjustment of weight factors. This results in a 25% improvement in signal stability and a 40% reduction in video freezes for high-speed rail users during subsequent network handovers. For traffic-sensitive users, when the daily traffic consumption rate in the past week is monitored to exceed 15% of the package limit, the β value will be automatically increased from 0.3 to 0.45, prompting the system to give priority to networks with lower rates. Actual tests have shown that the probability of users' monthly traffic overspending can be reduced from 35% to 12%.

[0064] In this embodiment, user feedback can include complaints about manual network switching and lag. For example, if a user manually switches from a cheaper network to one with better signal, α may need to be automatically adjusted higher. Privacy protection is implemented during the upload of the weight factor to the cloud center after local fine-tuning in step S4. The user's private data, which may include location and package usage, is not uploaded to the cloud center. This data is isolated and stored on the local terminal device, and only encrypted weight factor updates (i.e., incremental data) are uploaded to the cloud via a secure transmission channel.

[0065] For example, when a user manually switches to a 5G network, the terminal device records the operation event and triggers the local weight factor adjustment algorithm, and generates a binary update package after hashing the incremental value of the signal quality weight α.

[0066] During the federated learning self-optimization phase, the terminal device first triggers a local weight adjustment mechanism based on user feedback. If a manual network switch is detected, a weight correction is calculated based on the difference in network performance before and after the switch. If a lag complaint is received, the γ value is dynamically adjusted based on the service type's compatibility. The adjusted weight factors are processed in the local encryption module using an asymmetric encryption algorithm. For example, the updates for α, β, and γ are converted into floating-point values in a differentially private format and then encapsulated into ciphertext packets. After receiving the encrypted updates from each terminal, the cloud center generates global weight parameters using an aggregation algorithm. However, the original location trajectory or package consumption data cannot be parsed. This establishes a one-way isolated channel between user behavior data and model parameter updates, preserving the collaborative optimization capabilities of federated learning while preventing the possibility of inferring private information through weights. This approach ensures that the original private data never leaves the user device. After receiving the encrypted updates, the cloud center cannot infer the user's specific location or package usage, thereby protecting user privacy.

[0067] In this embodiment, the cloud center classifies the updated weight factors by scenario and performs weighted averaging, and the terminal devices with larger data volumes have higher weight proportions.

[0068] Optionally, scenarios can be categorized into network usage scenarios based on geographic location, such as defining high-speed rail areas as mobile scenarios and residential areas as fixed scenarios. Alternatively, scenarios can be categorized based on network access type, such as distinguishing 5G NSA and SA networking environments into different scenario categories. During the weighted averaging process, weight percentages are calculated using a data volume threshold to achieve differentiated processing. For example, when the historical data volume of a terminal device exceeds 100GB, its weight coefficient is increased by 1.5 times. Alternatively, a logarithmic function is used to map data volume to a weight coefficient, specifically weight percentage = log(1 + terminal device data volume) / Σlog(1 + data volume of each device).

[0069] Specifically, when the cloud center receives encrypted weight factor updates uploaded by devices, it first classifies and aggregates them based on the scenario tags reported by the devices. For example, in high-speed rail mobile scenarios, updates from devices traveling at speeds above 200 km / h during the same period are grouped together, while in indoor fixed scenarios, updates from devices connected to the same base station for 12 consecutive hours are grouped together. For each scenario set, when calculating the weighted average, the aggregation weight is dynamically adjusted based on the data volume share. For example, if a device contributes 80% of the data volume in the high-speed rail scenario, its weight factor updates will account for 80% of the weight when aggregated within the scenario. After completing the scenario classification and aggregation, the weighted results for each scenario are combined into a global weight factor according to a preset ratio, such as a 3:7 weight ratio for mobile and fixed scenarios. This process not only preserves the ability of scenario characteristics to optimize the weight factor in a targeted manner, but also strengthens the dominant role of high-reliability data in the global model by introducing data volume weighting. The resulting global weight factor exhibits greater scenario adaptability and decision fairness in dynamic network switching decisions.

[0070] Preferably, after the cloud center receives the weight factor updates uploaded by multiple terminal devices, it first classifies these updates according to the network usage scenario. For example, the scenarios can be divided into high-speed rail commuting, indoor fixed use, outdoor mobile and other categories. For each scenario category, the cloud center calculates the weighted average of the weight factor updates of all terminal devices in the scenario. When calculating the weighted average, the data volume of the terminal device is used as the weight coefficient. Specifically, assuming that there are n terminal devices in a certain scenario, the data volume of the i-th device is D i , the weight factor update amount is ΔW i , then the weighted average value ΔW of the weight factor update amount in this scenario can be expressed as: ΔW = (D1× ΔW1+ D2× ΔW2+ ... + D n × ΔW n ) / (D1+ D2+ ... + D n ) Among them, the data volume D i It can be the number of valid samples collected by the terminal device within a certain period of time. In this way, the terminal device with a larger amount of data has a greater impact on the final weight factor update amount.

[0071] Therefore, if Figure 6As shown in Figure 1, the cloud center can aggregate the weight factor updates for multiple users calculated in various scenarios, take a weighted average, generate a new global weight factor, and send it to all terminal devices (i.e., the new global weight factor is sent to all terminal devices), completing the weight factor update.

[0072] This process leverages the advantages of federated learning to achieve global optimization of the model without exposing personal privacy data. Finally, the updated global weight factors are sent to all terminal devices, completing one iteration of federated learning.

[0073] In order to achieve the above-mentioned purpose, the present invention also provides a decision system for dynamic network switching of an eSIM card, characterized in that it uses the decision method described in any of the aforementioned embodiments, including a server, a terminal device and a cloud center that are mutually communicatively connected.

[0074] The server with the model deployed locally is configured as a core data processing node, establishing a real-time data channel with the terminal device via a two-way communication link. This channel supports the transmission of signal quality parameters, package information, and service type data, and uses a low-latency transmission protocol to ensure the timeliness of data acquisition and preprocessing. The terminal device is configured as a network switching execution node, establishing an encrypted transmission channel with the cloud center via an independent communication link. This channel supports uploading local weight factor updates and receiving global model parameters. The cloud center is configured as a global optimization node, generating a global model by aggregating the weight factors of multiple terminal devices and sending the optimized parameters to all terminal devices via broadcast or multicast. The two-to-two interconnected architecture allows for a direct connection between the server and the cloud center to synchronize the metadata and system-level policy parameters required for federated learning, avoiding processing delays caused by relay forwarding.

[0075] Specifically, during operation, the server continuously receives real-time network status data reported by terminal devices and calculates a comprehensive score locally. The calculation results are pushed directly to the terminal device through a dedicated interface. The terminal device triggers a network switching operation based on the received priority list, and at the same time, performs local weight factor fine-tuning and uploads the adjusted parameters to the cloud center via an encrypted link. The cloud center aggregates the multi-source weight update data, generates a global optimization model, and sends it to the server and terminal device through independent channels. Through a communication topology with three nodes directly connected in pairs, the three processes of signal quality assessment, local policy adjustment, and global model update can be executed in parallel, reducing the generation cycle of network switching decisions to milliseconds while ensuring that the synchronization error of federated learning parameters is controlled within the preset threshold range.

[0076] In this embodiment, the entire terminal device can be understood as a distributed local server, which is deployed with a lightweight AI model (a model of sustainable evolution and federated learning). By collecting the signal quality parameters (RSRP / RSRQ / SINR) of the terminal device, user package tariff data, service type requirements and network environment status in real time, a dynamic weight distribution model is constructed, and networks that meet high signal quality, low tariff costs and service QoS requirements are given priority. At the same time, reinforcement learning and federated learning mechanisms are introduced to achieve continuous self-optimization of model parameters based on user switching feedback and network environment changes while protecting privacy.

[0077] Optionally, a long-term communication channel can be established between the server and the terminal device via the MQTT protocol for real-time transmission of signal quality parameters and priority lists; the weight factor update data generated by local federated learning can be transmitted between the terminal device and the cloud center using the TLS-encrypted HTTP / 2 protocol; and global model parameter synchronization can be performed between the server and the cloud center via a dedicated VPN channel. On the terminal device side, the embedded system deploys a lightweight federated learning client that periodically collects local network switching records and calculates the weight factor gradient difference. The gradient data is encrypted with AES-256 and then uploaded to the cloud center in batches; the cloud center deploys a distributed parameter server cluster that aggregates the encrypted weights uploaded by multiple terminals based on the differential privacy algorithm, generates a global weight update package, and distributes it to all connected terminal devices via CDN nodes; the server side integrates a real-time stream processing engine that performs sliding window statistics on the original signal quality data reported by the terminal device and dynamically adjusts the input feature dimensions of the comprehensive scoring calculation model.

[0078] In order to achieve the above-mentioned purpose, the present invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor runs the program, the steps of the decision-making method described in any of the foregoing embodiments can be implemented.

[0079] The processor and memory may be provided separately or integrated together, for example, integrated into a system-on-chip (SOC) of a terminal device.

[0080] In order to achieve the above objectives, the present invention also provides a computer-readable storage medium, which stores computer-executable instructions or computer programs. When the computer-executable instructions or computer programs are processed and executed, the decision-making method as described above is implemented.

[0081] The computer-readable storage medium is, for example, a memory. The memory may be a volatile memory or a non-volatile memory, or the memory may include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as 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 link dynamic random access memory (SLDRAM), and direct memory bus random access memory (DRRAM).

[0082] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above-mentioned computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the existing technology, or all or part of the technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes a number of instructions for causing one or more computer devices (such as personal computers, servers, or network devices) to execute all or part of the steps of the methods described in various embodiments of the present invention.

[0083] The preferred embodiments of the present invention have been described in detail above. It should be understood that numerous modifications and variations based on the concepts of the present invention are possible without inventive effort by those skilled in the art. Therefore, any technical solution that can be derived by one skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A decision method for dynamic network switching of an eSIM card, characterized in that: The steps include: Step S1: Real-time collection and preprocessing of multimodal data; the server continuously collects signal quality parameters and package information of user terminal devices and identifies service types and their QoS requirements, and performs comprehensive scoring based on the collected data information to obtain a comprehensive signal quality score S signal , tariff cost score S cost And business matching score S qos ; Step S2: Dynamic weight allocation: define the weight factors, determine the comprehensive utility value U of each available network and determine the network priority; Among them: comprehensive utility value U=αS signal +βS cost +γS qos ; α, β, and γ are the weight factors of signal quality, tariff cost, and service matching, respectively; Step S3: Network switching: triggering a network interface switching instruction to switch to the optimal network; Step S4: Federated learning self-optimization; based on user feedback and changes in the network environment, the weight factor is fine-tuned locally, and only the updated weight factor is encrypted and uploaded to the cloud center. The cloud center generates a new global weight factor and sends it to all terminal devices, that is, the weight distribution is dynamically optimized for all user devices at the same time through the federated learning mechanism.

2. The decision method for dynamic network switching of an eSIM card according to claim 1, characterized in that: The step S1 specifically includes the following steps: Step S1.1: Acquire a signal quality parameter according to a preset frequency, wherein the signal quality parameter includes at least one of the following: reference signal received power RSRP, reference signal received quality RSRQ, and signal to interference plus noise ratio SINR; The reference signal received power RSRP, reference signal received quality RSRQ, and signal to interference plus noise ratio SINR are scored, and the comprehensive signal quality score S is obtained by combining the three scores and adding weights. signal ; Step S1.2: Parse the user's eSIM card package information, which includes at least one of the following: data limit, roaming charges, and priority rules, and score the package cost using the following formula: Tariff cost score S cost =(total package traffic / remaining traffic×100)×w1+(1 / unit price×100)×w2; Among them, w1 is the weight of the remaining traffic; w2 is the weight of the tariff unit price; the tariff unit price is the tariff within the package; the higher the proportion of remaining traffic, the higher the score, and the lower the tariff unit price, the higher the score; Step S1.3: Determine the service type of the user's current eSIM card based on the port number and data packet size distribution, and output the QoS indicator requirement level; The business matching score S qos The calculation of the key QoS indicators of the business type is combined to set the weight w i , using the following formula: The ratio of negative indicators = demand threshold / actual value i ; Positive indicator ratio = actual value i / the ratio of the required threshold. If the actual value of the positive indicator exceeds the required threshold, the ratio result will be directly calculated as 1.0 to avoid excessive rewards; Then the weighted sum of the ratio of negative indicators to positive indicators is calculated. Business matching score S qos =(ratio 带宽 ×w 带宽 +Ratio 时延 ×w 时延 +Ratio 抖动 ×w 抖动 )×100, the maximum score is 100 points; Among them, w represents the weight coefficient of each QoS indicator. At the same time, these weight coefficients satisfy w 带宽 +w 时延 +w 抖动 =1.

3. The decision method for dynamic network switching of an eSIM card according to claim 2, characterized in that: In step S1.1, the scoring is based on the following logic and formula: When the RSRP value is less than or equal to -118 dBm, the signal is determined to be unavailable and the RSRP score is set to 0. When the SINR value is less than or equal to -5dB, the SINR score is 0; When the reference signal reception quality RSRQ is less than or equal to -19dB, the reference signal reception quality RSRQ score is 0; The specific formula is as follows: RSRP score = max(0, min(100, (RSRP + 118) × 2.083)) Signal-to-interference-and-noise ratio (SINR) score = max(0, min(100, (SINR+5)×4)) Reference signal reception quality RSRQ score = max(0, min(100, (RSRQ+19)×7.14)).

4. The decision method for dynamic network switching of an eSIM card according to claim 2, characterized in that: In step S1.3, the service type includes at least one of the following: video, voice, and instant messaging; and the QoS indicator includes at least one of the following: bandwidth, delay, and jitter.

5. The decision method for dynamic network switching of an eSIM card according to claim 1, characterized in that: In step S4, the three weight factors α, β, and γ are fine-tuned based on each user's historical usage records. in, If user A often experiences video freezes on the high-speed train, the value of α will be automatically increased; If user B's package traffic is frequently overspent, the value of β will be automatically increased; If user C experiences video freezes due to insufficient network bandwidth, the value of γ is automatically increased.

6. The decision method for dynamic network switching of an eSIM card according to claim 5, characterized in that: The user feedback includes at least one of the following: manual network switching and complaints about lag; privacy protection is performed during step S4, and the user's privacy data is not uploaded to the cloud center. The privacy data includes at least one of the following: location and package usage, and only the updated amount of the weight factor modification is uploaded.

7. The decision method for dynamic network switching of an eSIM card according to claim 6, characterized in that: The cloud center classifies the updated weight factors by scenario and then performs weighted average. The terminal devices with larger data volumes have higher weight proportions.

8. A decision system for dynamic network switching of an eSIM card, characterized in that: The decision-making method according to any one of claims 1 to 7 includes a server, a terminal device and a cloud center that are communicatively connected to each other.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor runs the program, the decision-making method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions or a computer program, and when the computer-executable instructions or the computer program are processed and executed, the decision-making method according to any one of claims 1 to 7 is implemented.

Images