Network quality sensing and intelligent switching method of three-in-one network Internet of Things card

Abstract

The invention discloses a network quality sensing and intelligent switching method for a three-in-one network Internet of Things card, and relates to the technical field of Internet of Things communication, and the method comprises the steps: continuously collecting real-time network quality index data of a current wireless network of the Internet of Things card; and performing multi-dimensional aggregation and cleaning on the data to form a standardized network state snapshot. And comparing the snapshot with a pre-established historical network quality mode library, and identifying a deviation trend and a potential degradation mode of the current network connection state. And detecting other two wireless networks which can be accessed by the Internet of Things card in parallel, and obtaining network quality evaluation parameters of the candidate network. And predicting a quality change track of the current network in a specified time period in the future based on the association rule in the historical mode library, and evaluating the quality stability of the candidate network. And generating a decision set comprising a plurality of switching strategy options by integrating the deviation trend, the potential degradation mode, the candidate network evaluation parameters and the prediction trajectory.

Description

Technical Field

[0001] This invention belongs to the field of Internet of Things (IoT) communication technology, specifically a method for network quality perception and intelligent switching of a triple-play IoT card. Background Technology

[0002] In IoT applications, devices using triple-play IoT SIM cards can access different operator networks. To maintain service continuity, it is necessary to switch to a better network when network quality deteriorates. Existing common technical solutions mainly rely on real-time monitoring of a single or a few indicators of the currently connected network, such as signal strength, and setting fixed thresholds to trigger switching. Another approach is to periodically probe alternative networks and select the network with the stronger signal based on instantaneous probe results. These existing technical solutions mainly react based on isolated state parameters at the current moment. The drawback of existing technical solutions is that their decision-making mechanism is passive and lagging, only able to respond to quality degradation that has already occurred, and unable to predict it in advance; at the same time, instantaneous or short-term quality sampling cannot reflect the fluctuation patterns and long-term trends of network status, which may lead to frequent and unnecessary switching or switching to networks whose quality deteriorates in the short term, affecting stable communication and energy efficiency of devices. This invention aims to solve the problem of how to shift from passive response to proactive predictive switching, and how to go beyond instantaneous quality comparison to select the best network for stability based on historical behavior patterns. Summary of the Invention

[0003] This invention aims to solve at least one of the technical problems existing in the prior art; To this end, the present invention proposes a network quality perception and intelligent switching method for a triple-play IoT card, comprising: The system continuously collects real-time network quality index data of the IoT card in the currently connected wireless network. The network quality index data includes signal strength data, network latency data, packet loss rate data, and available bandwidth data. The collected network quality index data is aggregated and cleaned in multiple dimensions to form a standardized network status snapshot, which contains a summary of the network connection status of the IoT card at a specific moment. The network status snapshot is compared and analyzed with a pre-established historical network quality pattern library to identify deviation trends and potential degradation patterns in the current network connectivity status. The network quality of the other two wireless networks that the IoT card can access is detected in parallel to obtain the network quality evaluation parameters of the candidate networks; Based on the association rules in the historical network quality pattern library, predict the quality change trajectory of the current network connection state within a specified future time period, and evaluate the quality stability of the candidate network. Based on the deviation trend, potential degradation mode, network quality evaluation parameters of candidate networks, and predicted quality change trajectory, a decision set containing multiple switching strategy options is generated.

[0004] Furthermore, the process of multi-dimensional aggregation and cleaning of the collected network quality index data to form a standardized network status snapshot includes: The collected signal strength data, network latency data, packet loss rate data, and available bandwidth data are input into the corresponding data calibration pipelines, which correct the errors in the original data according to preset physical layer characteristics. The error-corrected data is averaged over a preset time window to obtain the mean signal strength, mean network latency, mean packet loss rate, and mean available bandwidth. The fluctuation variance of the signal strength data within the preset time window is calculated to obtain the signal stability parameter; the jitter value of the network delay data within the preset time window is calculated to obtain the delay jitter parameter. The average signal strength, average network latency, average packet loss rate, average available bandwidth, signal stability parameters, and latency jitter parameters are encapsulated according to a predefined format to generate a standardized network state snapshot containing a timestamp.

[0005] Furthermore, the step of comparing and analyzing the network state snapshot with a pre-established historical network quality pattern library to identify deviation trends and potential degradation patterns in the current network connectivity status includes: Retrieve a set of historical network status snapshots that match the current geographical location, current time period, and current service type from the historical network quality pattern library; Calculate the Euclidean distance between the standardized network state snapshot and each snapshot in the historical network state snapshot set to obtain a set of similarity distance values; If all the similarity distance values ​​are greater than the preset similarity threshold, it is determined that the current network state deviates from the historical normal mode, and the deviation direction and magnitude of each indicator of the current snapshot relative to the historical average are calculated to form the deviation trend. Analyze the slope of change of network latency data and packet loss rate data in the standardized network status snapshot in the short term. If the slope of change exceeds the warning threshold, it is marked as a potential degradation mode.

[0006] Furthermore, the parallel detection of network quality for two other wireless networks accessible to the IoT card, and the acquisition of network quality evaluation parameters for candidate networks, include: While maintaining the current network connection, the baseband chip of the IoT card sends probe request frames to the base stations or access points of the other two wireless networks through pre-configured frequency points and signaling. Receive probe response frames from base stations or access points of two other wireless networks, and parse the broadcast information of the candidate network from the probe response frames; Based on the broadcast information, the reference signal received power and the signal-to-interference-plus-noise ratio of the candidate network are measured as signal quality parameters of the candidate network. Send a set of probe packets to a specific server in the candidate network, record the round-trip time and reception success rate of the packets, and calculate the estimated network latency and packet reception rate of the candidate network. By combining the signal quality parameters, estimated network latency, and data packet reception rate of the candidate networks, network quality evaluation parameters for the candidate networks are generated.

[0007] Furthermore, the step of predicting the quality change trajectory of the current network connection state over a specified future time period based on the association rules in the historical network quality pattern library, and evaluating the quality stability of the candidate network, includes: Extract several historical scene fragments that are similar to the current deviation trend from the historical network quality pattern library. The historical scene fragments record the process of network state change from a certain initial state. Using a time series prediction algorithm, starting from the current standardized network state snapshot, and combining the evolution patterns of the historical scene fragments, the predicted network state at multiple future time points is extrapolated to form the quality change trajectory. By analyzing long-term records of the candidate network in the historical network quality pattern library within the same geographical location and time period, the historical variance and mutation frequency of each network quality evaluation parameter are calculated to evaluate the quality stability of the candidate network.

[0008] Furthermore, based on the deviation trend, potential degradation patterns, network quality evaluation parameters of candidate networks, and predicted quality change trajectories, a decision set containing multiple switching strategy options is generated, including: Develop an immediate switching strategy option: The immediate switching strategy option is defined as selecting a candidate network for switching immediately based on the best current network quality assessment parameters once the potential degradation mode is identified; Develop a delayed handover strategy option: The delayed handover strategy option is defined as follows: if the predicted quality change trajectory shows that the current network will recover in the near future, then the evaluation and handover will be carried out after a preset time based on the network quality evaluation parameters at that time. Conditional switching strategy options: The conditional switching strategy options are defined as setting a set of network quality threshold conditions, and switching is triggered if and only if the current network quality index is lower than the threshold and the candidate network quality evaluation parameter is better than another higher threshold. The immediate switching strategy option, the delayed switching strategy option, and the conditional switching strategy option, together with their respective triggering conditions and expected target networks, constitute the decision set.

[0009] Furthermore, the method also includes: invoking a preset switching strategy evaluation model to comprehensively score each switching strategy option in the decision set, including: The comprehensive score is based on network tariff costs, service continuity maintenance, and the resource consumption of the handover operation itself; The feature vector of each handover strategy option in the decision set is input into the handover strategy evaluation model. The feature vector includes the estimated handover time, the unit price of the target network, the predicted service interruption duration, and the quality stability score of the target network. The first fully connected layer inside the switching strategy evaluation model calculates the basic score of the switching strategy option in the network quality dimension based on the pre-trained weight matrix. The second fully connected layer inside the switching strategy evaluation model calculates the basic score of the switching strategy option in the cost control dimension based on another set of pre-trained weight matrices. The third fully connected layer inside the switching strategy evaluation model calculates the basic score of the switching strategy option in the business assurance dimension based on another set of pre-trained weight matrices. The basic scores of the network quality dimension, the cost control dimension, and the service assurance dimension are weighted and summed according to preset importance coefficients to obtain the comprehensive score of the switching strategy option.

[0010] Furthermore, the method also includes: selecting the switching strategy option with the highest overall score as the target switching strategy, including: Sort the comprehensive scores of all switching strategy options in the decision set in descending order; Select the top-ranked switching strategy option as the initial target switching strategy; Check whether the initial target switching strategy and the currently executing background task or business protocol have any mandatory network locking constraints; If the mandatory network locking constraint exists, the initial target switching strategy is excluded, and the next ranked switching strategy option is selected as the new initial target switching strategy. The constraint check is repeated until a switching strategy option without constraints is found, and this option is determined as the final target switching strategy.

[0011] Furthermore, the method further includes: generating a specific network handover instruction sequence according to the target handover strategy, including: The network handover instruction sequence includes the handover timing, the target network identifier, and parameter reconfiguration information during the handover process; Analyze the target switching strategy to determine its type as immediate switching, delayed switching, or conditional switching; If the switch is immediate, a sequence of instructions is generated, including "Disconnect current connection", "Configure target network access parameters" and "Initiate access request". If a delayed handover is required, the instruction "Start Delay Timer" will be inserted before the sequence of instructions containing "Disconnect Current Connection", "Configure Target Network Access Parameters" and "Initiate Access Request", and the subsequent instructions will be triggered after the timer expires. If it is a conditional switch, a "start conditional monitoring thread" instruction is generated. The start conditional monitoring thread continuously checks the preset network quality threshold conditions. Once the conditions are met, a switch execution instruction is triggered. Each instruction in the instruction sequence is appended with context parameters, which include the target network's operating frequency band, access point name, authentication key, and desired IP address acquisition method.

[0012] Furthermore, the method also includes: executing the network switching instruction sequence to control the IoT card to disconnect from the current wireless network and access the target wireless network; performing network connection verification after access is completed; and updating the verification result and the network status information after the switch to the historical network quality mode library, including: Control commands are sent to the baseband chip and protocol stack of the IoT card one by one in the order of the network switching command sequence. After receiving confirmation that the "Disconnect current connection" command has been executed, send the "Configure target network access parameters" command. After receiving confirmation of successful parameter configuration, send the "Initiate Access Request" command and wait for the access permission response from the target network; After confirming successful access, network connection verification is performed, which includes sending a test data packet to a predefined verification server and successfully receiving a response. Collect network quality indicator data after successful verification, generate a new network status snapshot, and store the complete log of this switching event, the description of the target switching strategy, and the new network status snapshot as a new record in the historical network quality pattern library.

[0013] Compared with the prior art, the beneficial effects of the present invention are: By comparing real-time network status snapshots with a historical network quality pattern database, not only is the current absolute quality value determined, but also deviation trends and potential patterns relative to historical normal or deteriorated patterns are identified. This pattern-matching-based deep analysis can identify early signs of systemic degradation from micro-fluctuations, thus issuing warnings and initiating handover assessment processes before significant declines or interruptions in connection quality occur. This transforms network handover behavior from traditional passive remediation to proactive prevention, reducing the risk of service interruption and data loss due to sudden network deterioration.

[0014] Based on association rules mined from a historical network quality pattern database, predictive evaluations are performed on the performance of the current network and candidate networks over a specified future time period. This includes predicting the current network's degradation trajectory and assessing the likelihood of candidate networks maintaining stable connectivity during the same period. The decision-making basis thus shifts from "which is better now" to "which is more stable and reliable in the future." This method selects target networks that better match the continuity requirements of services, avoids unnecessary switching triggered by only momentary signal fluctuations, and reduces the probability of needing to switch again due to rapid changes in the target network's quality after a previous switch, thereby improving the long-term stability and decision-making intelligence of the overall connectivity solution. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the network quality perception and intelligent switching method of the triple-play IoT card described in this invention. Figure 2 A flowchart for obtaining parameters for candidate network quality detection and evaluation; Figure 3 A bar chart showing the network quality stability rating of a triple-play IoT SIM card; Figure 4 A radar chart for multi-dimensional evaluation of IoT three-network switching strategies; Figure 5 A grouped bar chart showing the execution time of network switching commands for IoT cards. Detailed Implementation

[0016] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] See Figure 1The network quality perception and intelligent switching method for a three-network integrated IoT card of the present invention is implemented as follows: The IoT card continuously collects real-time network quality indicator data in the currently connected wireless network. This data includes signal strength data, network latency data, packet loss rate data, and available bandwidth data. The collected network quality indicator data undergoes multi-dimensional aggregation and cleaning to form a standardized network status snapshot, which contains a summary of the IoT card's network connection status at a specific moment. The network status snapshot is compared and analyzed with a pre-established historical network quality pattern library to identify deviation trends and potential degradation patterns in the current network connection status. Simultaneously, the network quality of the other two wireless networks that the IoT card can access is probed in parallel to obtain network quality evaluation parameters for candidate networks. Based on the association rules in the historical network quality pattern library, the quality change trajectory of the current network connection status within a specified future time period is predicted, and the quality stability of the candidate networks is evaluated. Based on the deviation trend, potential degradation patterns, network quality evaluation parameters of the candidate networks, and the predicted quality change trajectory, a decision set containing multiple switching strategy options is generated.

[0018] In one embodiment of the present invention, the IoT card performs multi-dimensional aggregation and cleaning processing on the collected network quality indicator data to form a standardized network status snapshot. Specifically, the collected signal strength data, network latency data, packet loss rate data, and available bandwidth data are input into corresponding data calibration pipelines. The data calibration pipelines correct the original data for errors based on preset physical layer characteristics. The error-corrected data is then averaged over a preset time window to obtain the mean signal strength, mean network latency, mean packet loss rate, and mean available bandwidth. The variance of the signal strength data within the preset time window is calculated to obtain a signal stability parameter; the jitter value of the network latency data within the preset time window is calculated to obtain a latency jitter parameter. The mean signal strength, mean network latency, mean packet loss rate, mean available bandwidth, signal stability parameter, and latency jitter parameter are encapsulated according to a predefined format to generate a standardized network status snapshot containing a timestamp.

[0019] In practical implementation, the IoT SIM card performs multi-dimensional aggregation and cleaning of collected network quality indicator data to form a standardized network status snapshot. This process is achieved through a data processing thread running on the IoT SIM card's embedded processor. The data processing thread reads the raw network quality indicator data from the baseband chip and protocol stack interface at fixed intervals. The raw network quality indicator data includes signal strength data in dBm, network latency data in milliseconds, packet loss rate data as a percentage, and available bandwidth data in kbps. This raw data is input into four independent data calibration pipelines. Each data calibration pipeline corrects the raw data for errors based on the physical layer characteristics of its corresponding indicator and a known system error model. The signal strength data calibration pipeline compensates for the raw received signal strength indication value based on the antenna gain compensation table and the current temperature sensor reading; the network latency data calibration pipeline subtracts a fixed processing latency offset within the protocol stack; the packet loss rate data calibration pipeline normalizes the statistical results based on the size of the currently transmitted data packets; and the available bandwidth data calibration pipeline recalculates the measured value based on the current physical layer coding scheme. The error-corrected data is fed into a sliding window buffer, the size of which corresponds to a preset time window, with a length of 60 seconds. The data processing thread performs a moving average calculation on the corrected data in the buffer according to the indicator type, updating the average value each time a new data point is received, thereby continuously generating the average signal strength, average network latency, average packet loss rate, and average available bandwidth. For a signal strength data sequence contained within the preset time window, its signal stability parameter is obtained by calculating the average of the sum of squared deviations of all signal strength samples within the preset time window from their average value.

[0020] In some embodiments, the average signal strength, average network latency, average packet loss rate, and average available bandwidth are calculated using an exponentially weighted moving average algorithm, which assigns higher weights to recent data. A data processing thread maintains a first-in-first-out queue to store raw sampling points within a preset time window. Whenever new data is added to the queue and old data is removed, the signal stability parameter and latency jitter parameter are recalculated based on the updated queue content. In a specific implementation, the calculated average signal strength, average network latency, average packet loss rate, average available bandwidth, signal stability parameter, and latency jitter parameter are organized into a structured data object. This data object is encapsulated according to a predefined JSON format, which includes fixed field names and data type specifications. During the encapsulation process, a high-precision timestamp obtained from the IoT card's security chip is written into the data object, thereby generating a standardized network state snapshot containing the timestamp. Optionally, the standardized network state snapshot is temporarily stored in the IoT card's non-volatile memory after generation, awaiting subsequent access by the analysis module. It is understood that the error correction coefficients used in the data calibration pipeline are obtained through laboratory calibration before shipment and are stored in the read-only memory of the IoT card. In some embodiments, the length of the preset time window can be dynamically adjusted according to the type of service currently running on the IoT card; for services with high real-time requirements, the length of the preset time window is set to be shorter. Optionally, before calculating the moving average, the data calibration pipeline also performs an outlier filtering step, removing sampled values ​​that significantly deviate from the normal range according to the Laida criterion. It is understood that the generation frequency of standardized network state snapshots is consistent with the data acquisition frequency, ensuring that each acquisition cycle corresponds to one snapshot.

[0021] See Figure 2 In one embodiment of the present invention, a network status snapshot is compared and analyzed with a pre-established historical network quality pattern library to identify deviation trends and potential degradation patterns in the current network connection status. Specifically, this includes: retrieving a set of historical network status snapshots from the historical network quality pattern library that match the current geographical location, current time period, and current service type; calculating the Euclidean distance between the standardized network status snapshot and each snapshot in the historical network status snapshot set to obtain a set of similarity distance values; if all similarity distance values ​​are greater than a preset similarity threshold, it is determined that the current network status deviates from the historical normal pattern, and the deviation direction and magnitude of each indicator of the current snapshot relative to the historical average are calculated to form a deviation trend; analyzing the slope of change of network latency data and packet loss rate data in the standardized network status snapshot in the short term, if the slope exceeds a warning threshold, it is marked as a potential degradation pattern.

[0022] The IoT SIM card simultaneously probes the network quality of two other accessible wireless networks to obtain network quality evaluation parameters for candidate networks. Specifically, this involves: controlling the IoT SIM card's baseband chip to send probe request frames to the base stations or access points of the other two wireless networks using pre-configured frequency points and signaling while maintaining the current network connection; receiving probe response frames from the base stations or access points of the other two wireless networks and parsing the broadcast information of the candidate networks from these frames; measuring the reference signal received power and signal-to-interference-plus-noise ratio (SNR) of the candidate networks based on the broadcast information, using these as signal quality parameters; sending a set of probe data packets to a specific server of the candidate networks, recording the round-trip time and reception success rate of the data packets, and calculating the estimated network latency and data packet reception rate of the candidate networks; and combining the signal quality parameters, estimated network latency, and data packet reception rate of the candidate networks to generate network quality evaluation parameters for the candidate networks.

[0023] In practice, the process of comparing network status snapshots with a pre-established historical network quality pattern library to identify deviation trends and potential degradation patterns is executed by the network status analysis engine running within the IoT card. Upon receiving a standardized network status snapshot, the analysis engine first extracts the geolocation code, timestamp, and service type label embedded in the standardized snapshot. In practice, the geolocation code originates from the IoT card's built-in GPS module or base station-based positioning results, the timestamp is in Coordinated Universal Time (UTC), and the service type label is specified by the upper-layer application when establishing a network connection. The network status analysis engine uses the geolocation code, the time period of the timestamp, and the service type label as a composite query key to initiate a retrieval of the historical network quality pattern library stored in the secure storage area of ​​the IoT card. The historical network quality pattern library is a database with multi-dimensional indexes based on geolocation, time period, and service type. The retrieval operation returns a set of historical network status snapshots, containing all standardized network status snapshots recorded under similar past conditions. For each historical snapshot in the historical network state snapshot set, the network state analysis engine calculates its Euclidean distance to the current standardized network state snapshot. The Euclidean distance calculation encompasses six dimensions: mean signal strength, mean network latency, mean packet loss rate, mean available bandwidth, signal stability parameters, and latency jitter parameters. The Euclidean distance between two network state snapshots is obtained by calculating the square root of the weighted sum of squared differences among the six network quality metrics. Each metric's difference is multiplied by a preset weighting coefficient.

[0024] In some embodiments, the network state analysis engine compares this set of similarity distance values ​​with a preset similarity threshold. The preset similarity threshold is obtained by offline analysis of the data distribution under normal operating conditions in a historical network quality pattern library. If all similarity distance values ​​are greater than the preset similarity threshold, the network state analysis engine determines that the current network state deviates from the historical normal pattern. Subsequently, the network state analysis engine calculates the difference between each indicator value in the current standardized network state snapshot and the average value of the corresponding indicator in the historical network state snapshot set. The direction and magnitude of the difference are recorded, and the direction and magnitude of the difference together constitute a deviation trend report on the current network connectivity state. For example, a deviation trend report may indicate that the average signal strength has decreased by 5 dBm compared to the historical average, while the average network latency has increased by 20 milliseconds. In specific implementations, the network state analysis engine simultaneously initiates a short-term trend analysis routine, which focuses on the network latency data sequence and packet loss rate data sequence in the standardized network state snapshot. The short-term trend analysis routine takes network latency data from the most recent 10 sampling periods and calculates its slope, i.e., the rate of change of network latency over time; it also calculates the slope of change of packet loss rate over the same time period. If the absolute value of the slope of change of network latency or the slope of change of packet loss rate exceeds its respective independent warning threshold, the network state analysis engine marks this phenomenon as a potential degradation mode. The warning thresholds are preset based on the sensitivity of the service type to latency and packet loss.

[0025] In practical implementation, the IoT SIM card simultaneously probes the network quality of two other accessible wireless networks to obtain network quality assessment parameters. This function is performed by the parallel probing unit of the baseband chip. The IoT SIM card is currently connected to the operator's LTE network. Based on a pre-configured list of operator networks and frequency information, the parallel probing unit controls the baseband chip's radio frequency front-end to periodically switch to the frequency band of the target candidate network while maintaining a connection with the current LTE base station. Optionally, the target candidate networks include 5G New Radio networks and narrowband IoT networks. The baseband chip sends probe request frames to the base station or access point of the target candidate network via pre-configured frequency points and signaling conforming to the candidate network's air interface standard. The probe request frame is a specially designed low-overhead signaling used to request network broadcast information without performing a complete access procedure. In practical implementation, the baseband chip then listens for probe response frames from the target candidate network's base station or access point. Upon receiving a probe response frame, the baseband chip's protocol stack parsing module extracts the candidate network's broadcast information from the probe response frame. This broadcast information includes network identity, cell load indication, supported quality of service (QoS) level, and other information. The parsing module transmits the broadcast information to the signal quality measurement module.

[0026] It is understood that the signal quality measurement module performs physical layer measurements for each candidate network based on the parsed broadcast information. The measurements include the candidate network's reference received power and the signal-to-interference-plus-noise ratio (SNR). The reference received power and SNR together serve as signal quality parameters characterizing the candidate network's signal quality. In some embodiments, after completing the signal quality measurement, the IoT card's application processor generates a set of small probe data packets. Optionally, these probe data packets are Internet Control Message Protocol (ICP-IP) echo request packets. These probe data packets are sent to a specific server in the candidate network via a temporarily established, uncertified, simplified signaling link by the baseband chip. The ICP-IP address of the specific server is pre-installed in the IoT card. The application processor records the sending and receiving timestamps of each probe data packet and counts the number of successfully received response data packets. Based on these records, the average round-trip time of the probe data packets is calculated as the estimated network latency of the candidate network, and the ratio of the number of received response data packets to the number of sent response data packets is calculated as the candidate network's packet reception rate. Finally, the network state analysis engine merges data from three dimensions—signal quality parameters of the candidate network, estimated network latency of the candidate network, and packet reception rate of the candidate network—to generate a structured network quality assessment parameter report for the candidate network.

[0027] In one embodiment of the present invention, based on association rules in a historical network quality pattern library, the quality change trajectory of the current network connection state over a specified future time period is predicted, and the quality stability of candidate networks is evaluated. Specifically, this includes: extracting several historical scene fragments from the historical network quality pattern library that are similar to the current deviation trend; these historical scene fragments record the change process of the network state from a certain initial state; using a time series prediction algorithm, starting with the current standardized network state snapshot, and combining the evolution patterns of the historical scene fragments, extrapolating to generate predicted network states for multiple future time points, thus forming a quality change trajectory; and analyzing long-term records of candidate networks in the historical network quality pattern library within the same geographical location and time period, calculating the historical variance and mutation frequency of various network quality evaluation parameters to evaluate the quality stability of the candidate networks.

[0028] Based on deviation trends, potential degradation patterns, network quality assessment parameters of candidate networks, and predicted quality change trajectories, a decision set containing multiple switching strategy options is generated. Specifically, these include: an immediate switching strategy option, defined as immediately selecting a candidate network for switching based on the best current network quality assessment parameters once a potential degradation pattern is identified; a delayed switching strategy option, defined as delaying evaluation and switching based on network quality assessment parameters at a predetermined time if the predicted quality change trajectory indicates that the current network will recover in the near future; and a conditional switching strategy option, defined as setting a set of network quality threshold conditions, triggering switching only if the current network quality index is below the threshold and the candidate network quality assessment parameters are better than another higher threshold. The immediate, delayed, and conditional switching strategies, along with their respective trigger conditions and the expected target network, together constitute the decision set.

[0029] In practice, the prediction and evaluation module predicts the quality change trajectory of the current network connection state over a specified future time period based on association rules in a historical network quality pattern library, and evaluates the quality stability of candidate networks. This process is completed by the prediction and evaluation module. The prediction and evaluation module receives deviation trend reports from the network state analysis engine regarding the current network connection state. These reports may describe a decreasing trend in the average signal strength and an increasing trend in the average network latency. The prediction and evaluation module uses this deviation trend as a feature vector to perform pattern matching in the historical network quality pattern library. The matching process involves retrieving all historical data fragments containing similar initial states and directions of change from the historical network quality pattern library. These historical data fragments are extracted as historical scene fragments, each recording the complete change process of the network state from a certain initial state over several subsequent sampling periods. After collecting a set of similar historical scene fragments, the prediction and evaluation module employs a time series prediction algorithm, using the current standardized network state snapshot as the starting point for prediction. The time-series prediction algorithm analyzes the evolution patterns of each historical scene segment, such as the growth function of network latency over time or the attenuation rate of signal strength. The algorithm extrapolates and generates predicted network state values ​​for multiple future time points by weighted fusion of these historical evolution patterns. These predicted network state values, arranged chronologically, are then connected to form a trajectory of quality changes in the current network connectivity over a specified future time period. For the quality stability assessment of candidate networks, the prediction and evaluation module queries the historical network quality pattern library for all historical network quality assessment parameter records of the candidate network within the same geographical location and current time period. The evaluation module calculates various network quality assessment parameters in these historical records, such as the historical variance of the reference signal received power and the estimated network latency. The historical variance measures the degree of parameter fluctuation. Simultaneously, the evaluation module counts the number of times parameter values ​​exceed a preset mutation threshold between adjacent sampling points, i.e., the mutation frequency. The quality stability of the candidate network is quantified by a stability score, which is inversely proportional to the historical variance and the mutation frequency. Finally, the prediction and evaluation module outputs an evaluation report containing the quality change trajectory curve and the quality stability score.

[0030] In some embodiments, the prediction and evaluation module generates a decision set containing multiple switching strategy options based on deviation trends, potential degradation patterns, network quality assessment parameters of candidate networks, and predicted quality change trajectories. The generation process is executed by a strategy generator, which first checks for the existence of marked potential degradation patterns. If a potential degradation pattern exists, the strategy generator immediately creates an immediate switching strategy option. The immediate switching strategy option is defined as follows: once a potential degradation pattern is identified, the candidate network with the highest overall score is selected for switching based on the network quality assessment parameters of the candidate networks obtained at the current time. The trigger condition for the immediate switching strategy option is that the flag of the potential degradation pattern is true. Simultaneously, the strategy generator analyzes the predicted quality change trajectory. If the quality change trajectory shows a clear trend of a key indicator of the current network recovering to a normal level within a preset short-term time window, the strategy generator creates a delayed switching strategy option. The delayed switching strategy option is defined as follows: after a preset time delay, all available networks are re-evaluated and switched based on the network quality assessment parameters at that time. The trigger condition for the delayed switching strategy option is that the predicted recovery time point of the quality change trajectory is earlier than a preset tolerance time threshold. The policy generator also sets a set of network quality threshold conditions, including a lower limit for signal strength, an upper limit for network latency, an upper limit for packet loss rate, and a threshold for the magnitude at which the network quality evaluation parameters of candidate networks are better than those of the current network. A conditional switching policy option will only be triggered if and only if the current network quality index is lower than its corresponding threshold and the network quality evaluation parameter of at least one candidate network is better than its corresponding higher threshold. The definition of a conditional switching policy option involves setting this set of network quality threshold conditions and continuously monitoring them; switching is automatically triggered when the conditions are met. In specific implementation, the policy generator encapsulates the immediate switching policy option, the delayed switching policy option, and the conditional switching policy option, along with their respective specific triggering condition logic and the target network identifier for the expected switching, into three independent structured objects. These three structured objects together constitute a decision set that can be selected by subsequent modules. It can be understood that each switching policy option object in the decision set contains a complete description of the execution logic and applicable scenarios. Optionally, the specific values ​​of the network quality threshold conditions are remotely configured by the IoT card management platform according to business policies or preset by the device at the factory. In some embodiments, the preset delay time in the delay handover policy option is not a fixed value, but is dynamically calculated based on the recovery time predicted by the quality change trajectory. It can be understood that the generation of the decision set is periodic; whenever a new network state snapshot and evaluation report are generated, the policy generator re-runs to update the contents of the decision set.

[0031] See Figure 3This is a bar chart showing the network quality stability score of a triple-play IoT SIM card, used to quantitatively compare the quality stability of different networks. The stability score directly reflects the probability of sudden network failures. Choosing a high-scoring China Unicom network can significantly reduce the risk of IoT service interruptions due to network fluctuations. Operators or IoT platforms can dynamically adjust network resource allocation based on this score, such as reserving more bandwidth for highly stable networks, thereby improving overall service quality. The long-term accumulated stability score data can be used to optimize the historical network quality pattern library, improving the accuracy of subsequent predictions and forming a closed-loop iteration of "data-evaluation-optimization". By comparing the stability scores of the three networks, clear priorities are provided for different switching strategies, prioritizing the most stable China Unicom network to ensure service continuity.

[0032] In one embodiment of the present invention, a preset handover strategy evaluation model is invoked to comprehensively score each handover strategy option in the decision set. The comprehensive score is based on network tariff cost, service continuity maintenance, and the resource consumption of the handover operation itself. The feature vector of each handover strategy option in the decision set is input into the handover strategy evaluation model. The feature vector includes the estimated handover time, the unit price of the target network, the predicted service interruption duration, and the quality stability score of the target network. The first fully connected layer within the handover strategy evaluation model calculates the basic score of the handover strategy option in the network quality dimension based on a pre-trained weight matrix; the second fully connected layer calculates the basic score of the handover strategy option in the cost control dimension based on another set of pre-trained weight matrices; and the third fully connected layer calculates the basic score of the handover strategy option in the service assurance dimension based on yet another set of pre-trained weight matrices. The basic scores in the network quality dimension, cost control dimension, and service assurance dimension are weighted and summed according to preset importance coefficients to obtain the comprehensive score of the handover strategy option.

[0033] The switching strategy option with the highest overall score is selected as the target switching strategy. Specifically, this involves: sorting all switching strategy options in the decision set in descending order of their overall scores; selecting the top-ranked switching strategy option as the initial target switching strategy; checking whether the initial target switching strategy has any mandatory network locking constraints with the currently executing background task or business protocol; if mandatory network locking constraints exist, excluding the initial target switching strategy and selecting the next-ranked switching strategy option as the new initial target switching strategy, repeating the constraint check until a switching strategy option without constraints is found and determined as the target switching strategy.

[0034] In practical implementation, a pre-defined handover strategy evaluation model is invoked to comprehensively score each handover strategy option in the decision set. This function is performed by the strategy evaluation engine integrated within the IoT card. The strategy evaluation engine loads a pre-trained neural network model as the handover strategy evaluation model. The handover strategy evaluation model is designed to quantitatively evaluate each handover strategy option across three dimensions: network cost, service continuity maintenance, and resource consumption of the handover operation itself. The strategy evaluation engine reads a handover strategy option from the decision set and constructs a feature vector based on the definition of the handover strategy option and the current network environment information. The feature vector contains the following fields: the estimated handover time estimated based on historical handover logs, the unit price of the target network under the latest tariff policy, the service interruption duration predicted based on the current service traffic model, and the quality stability score of the target network obtained from the prediction and evaluation module. Refer to Table 1, which shows a feature vector containing three handover strategy options.

[0035] Table 1: Feature Vector Data Table of Switching Strategy Options In implementation, the strategy evaluation engine inputs feature vectors into the handover strategy evaluation model. The first fully connected layer within the model, based on its pre-trained weight matrix, performs linear transformations and non-linear activations on the network quality-related dimensions of the feature vectors, outputting a base score representing the network quality dimension. These network quality-related dimensions include the target network's quality stability score and a portion of the estimated handover time. The second fully connected layer, based on another set of pre-trained weight matrices, specifically handles cost-related features, i.e., tariff unit price, and calculates a base score for the cost control dimension. The third fully connected layer, based on yet another set of pre-trained weight matrices, focuses on analyzing service assurance-related features, i.e., the predicted service interruption duration, and outputs a base score for the service assurance dimension. These three base scores are passed to the model's output layer, which performs a weighted sum based on preset importance coefficients, ultimately generating a comprehensive score for the handover strategy option. The comprehensive score for the handover strategy option is obtained by multiplying the three base scores for the network quality dimension, cost control dimension, and service assurance dimension by preset importance coefficients and then summing them; the sum of the three importance coefficients is one. The strategy evaluation engine repeats the above process for each switching strategy option in the decision set, resulting in a list with serial numbers and comprehensive scores.

[0036] In some embodiments, the operation of selecting the switching strategy option with the highest overall score as the target switching strategy is performed by a decision selector. The decision selector receives a list of scores from the strategy evaluation engine and first sorts it in descending order of overall score. The decision selector selects the top-ranked switching strategy option as the initial target switching strategy. The decision selector then checks whether the initial target switching strategy has any mandatory network locking constraints with the currently executing background task or business protocol. Mandatory network locking constraints refer to certain critical tasks or protocols that explicitly prohibit network switching at a specific stage. Optionally, such constraint information is set by the upper-layer application through an application programming interface and stored in the task status register of the IoT card. If mandatory network locking constraints exist, the decision selector excludes the initial target switching strategy and selects the next-ranked switching strategy option from the sorted list as the new initial target switching strategy. The decision selector repeats the network locking constraint check on the new initial target switching strategy. This checking and exclusion process continues until the decision selector finds a switching strategy option without any mandatory network locking constraints. The decision selector identifies the found handover strategy options as the final target handover strategy and outputs a complete description of the target handover strategy, including its type, triggering conditions, and target network identifier. In some embodiments, if all handover strategy options in the sorted list are excluded due to network locking constraints, the decision selector will output a specific decision to maintain the current network connection. It is understood that the check for network locking constraints ensures that the handover decision does not violate the critical requirements of upper-layer services.

[0037] See Figure 4 This is a radar chart for multi-dimensional evaluation of IoT three-network switching strategies. It quantitatively compares the overall performance of different networks across five dimensions: network quality, stability, switching time, service assurance, and cost control, providing an intuitive basis for intelligent switching decisions. By transforming abstract network performance indicators into a visual radar chart, it intuitively presents the strengths and weaknesses of each network across different dimensions, solving the problem of the one-sidedness of single-indicator evaluation. If prioritizing service continuity, Network B is the optimal choice, excelling in service assurance, cost control, and stability. Evaluation based on the "switching time" dimension can avoid networks with low switching efficiency, reducing service interruption time, which is particularly suitable for latency-sensitive IoT services. Based on the evaluations of "stability" and "service assurance," networks with low fluctuations and strong support capabilities can be prioritized, improving the overall reliability of IoT services and user experience.

[0038] In one embodiment of the present invention, a specific network handover instruction sequence is generated according to the target handover strategy. The network handover instruction sequence includes the handover timing, the target network identifier, and parameter reconfiguration information during the handover process. The target handover strategy is parsed to determine its type as immediate handover, delayed handover, or conditional handover. If it is an immediate handover, an in-order instruction sequence containing "disconnect current connection," "configure target network access parameters," and "initiate access request" is generated. If it is a delayed handover, a "start delay timer" instruction is inserted before the in-order instruction sequence containing "disconnect current connection," "configure target network access parameters," and "initiate access request," and subsequent instructions are triggered after the timer expires. If it is a conditional handover, a "start condition monitoring thread" instruction is generated, which continuously checks preset network quality threshold conditions. Once the conditions are met, a handover execution instruction is triggered. Context parameters are added to each instruction in the instruction sequence. The context parameters include the target network's operating frequency band, access point name, authentication key, and desired IP address acquisition method.

[0039] The system executes a network handover command sequence to control the IoT SIM card to disconnect from the current wireless network and connect to the target wireless network. After connection is established, network connection verification is performed, and the verification results and post-handover network status information are updated to the historical network quality pattern library. Specifically, this includes: sending control commands to the IoT SIM card's baseband chip and protocol stack sequentially according to the network handover command sequence; sending a "Configure target network access parameters" command after receiving confirmation of the "Disconnect current connection" command completion; sending an "Initiate access request" command after receiving confirmation of successful parameter configuration and waiting for access permission from the target network; performing network connection verification after successful access confirmation, including sending test data packets to a predefined verification server and successfully receiving a response; collecting network quality indicator data after successful verification; generating a new network status snapshot; and storing the complete log of this handover event, a description of the target handover strategy, and the new network status snapshot as a new record in the historical network quality pattern library.

[0040] In practice, the operation of generating a specific network handover instruction sequence based on the target handover strategy is executed by the instruction sequence compiler, which receives the target handover strategy description from the decision selector. The network handover instruction sequence is a list of atomic operations, containing the handover timing, the target network identifier, and parameter reconfiguration information during the handover process. The instruction sequence compiler first parses the target handover strategy to determine whether it is an immediate handover, a delayed handover, or a conditional handover. If the target handover strategy is an immediate handover, the instruction sequence compiler generates a list of three sequential instructions: the first instruction is "disconnect current connection," the second is "configure target network access parameters," and the third is "initiate access request." If the target handover strategy is a delayed handover, before generating the above three sequential instruction lists, the instruction sequence compiler inserts a "start delay timer" instruction as the first instruction in the list. The "start delay timer" instruction carries the delay duration parameter parsed from the target handover strategy. This instruction sets a hardware timer, which, upon expiration, will automatically trigger the execution of subsequent instructions such as "disconnect current connection." If the target handover strategy is conditional handover, the instruction sequence compiler generates a "Start Conditional Monitoring Thread" instruction. This instruction creates an independent monitoring thread that continuously checks preset network quality thresholds. Once the thresholds are met, the monitoring thread triggers an internal event to activate subsequent handover execution instructions. In practice, the instruction sequence compiler adds context parameters to each instruction in the network handover instruction sequence. The specific content of these context parameters is retrieved from the IoT card's network configuration file. These parameters include the target network's operating frequency band, the target network's access point name, the target network's authentication key, and the desired Internet Protocol address acquisition method. After adding the context parameters, a complete network handover instruction sequence is generated and passed to the instruction executor. In some embodiments, the network handover instruction sequence generates a checksum. This checksum is calculated by performing a cyclic redundancy check on the concatenated string of the instruction opcode sequence and all context parameters. The checksum is used to verify the integrity of the network handover instruction sequence during transmission.

[0041] In practice, the instruction executor module is responsible for executing the network switching instruction sequence and controlling the IoT card to complete the network switching and verification update. The instruction executor sends control commands to the IoT card's baseband chip and network protocol stack one by one, according to the order of the instructions in the network switching instruction sequence. The instruction executor first sends a "Disconnect Current Connection" instruction. Upon receiving the instruction, the baseband chip executes the signaling release procedure with the current serving base station. After completing the procedure, it returns a confirmation message indicating that the instruction execution is complete to the instruction executor. After receiving the confirmation of the "Disconnect Current Connection" instruction completion, the instruction executor immediately sends the next instruction, "Configure Target Network Access Parameters." This instruction carries accompanying context parameters. The network protocol stack updates the frequency point, access point name, and authentication key information in its internal registers based on the context parameters. Upon successful configuration, it sends a confirmation message indicating successful parameter configuration. After receiving the confirmation of successful parameter configuration, the instruction executor sends an "Initiate Access Request" instruction. The baseband chip uses the newly configured parameters to initiate a random access or association request to the target network and waits for an access permission response from the target network. After successful access confirmation, the instruction executor initiates the network connection verification process. This process includes sending a set of Internet Control Message Protocol (ICP) echo request test packets to a predefined verification server and successfully receiving response packets from the server. The address of the verification server is stored in the non-volatile memory of the IoT card. Following successful network connection verification, the instruction executor triggers the data acquisition module to collect initial network quality indicator data. This network quality indicator data is then sent to the data processing thread to generate a new network state snapshot.

[0042] In some embodiments, the instruction executor binds the complete log record of this handover event, the descriptive text of the target handover strategy, and the new network state snapshot into a single data packet. This data packet is appended as a new record to the corresponding category in the historical network quality pattern library. Optionally, the complete log record includes the sending timestamp of each instruction, the confirmation receiving timestamp, any error codes generated during the handover process, and the response latency of the verification server. It is understood that the addition of new records enriches the historical network quality pattern library, providing an updated data foundation for subsequent pattern matching and prediction. After completing the data update, the instruction executor sends a final status notification indicating that the handover process has been completed to the main controller of the IoT card.

[0043] See Figure 5 This is a grouped bar chart showing the execution time of IoT SIM card network switching commands. It is the core visualization content of the "command execution stage" and is used to compare the command execution time and command confirmation time of the four key steps in the switching process. Figure 5The study clearly revealed that "network verification" is the main time-consuming step in the entire handover process. This conclusion can be used to optimize the verification logic, thereby reducing the overall handover time. The time consumption data of each step directly reflects the potential duration of service interruption, providing a quantitative basis for the handover strategy of highly sensitive services. The accumulated time consumption data can be used to optimize the scheduling logic of the instruction executor, such as by processing some instructions in parallel to compress the overall time consumption, or by adjusting the timeout threshold to improve the handover success rate.

[0044] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for network quality perception and intelligent switching of a triple-play IoT SIM card, characterized in that, The method includes: The system continuously collects real-time network quality index data of the IoT card in the currently connected wireless network. The network quality index data includes signal strength data, network latency data, packet loss rate data, and available bandwidth data. The collected network quality index data is aggregated and cleaned in multiple dimensions to form a standardized network status snapshot, which contains a summary of the network connection status of the IoT card at a specific moment. The network status snapshot is compared and analyzed with a pre-established historical network quality pattern library to identify deviation trends and potential degradation patterns in the current network connectivity status. The network quality of the other two wireless networks that the IoT card can access is detected in parallel to obtain the network quality evaluation parameters of the candidate networks; Based on the association rules in the historical network quality pattern library, predict the quality change trajectory of the current network connection state within a specified future time period, and evaluate the quality stability of the candidate network. Based on the deviation trend, potential degradation mode, network quality evaluation parameters of candidate networks, and predicted quality change trajectory, a decision set containing multiple switching strategy options is generated.

2. The network quality perception and intelligent switching method for a triple-play IoT card according to claim 1, characterized in that, The process of multi-dimensional aggregation and cleaning of the collected network quality index data to form a standardized network status snapshot includes: The collected signal strength data, network latency data, packet loss rate data, and available bandwidth data are input into the corresponding data calibration pipelines, which correct the errors in the original data according to preset physical layer characteristics. The error-corrected data is averaged over a preset time window to obtain the mean signal strength, mean network latency, mean packet loss rate, and mean available bandwidth. The fluctuation variance of the signal strength data within the preset time window is calculated to obtain the signal stability parameter; the jitter value of the network delay data within the preset time window is calculated to obtain the delay jitter parameter. The average signal strength, average network latency, average packet loss rate, average available bandwidth, signal stability parameters, and latency jitter parameters are encapsulated according to a predefined format to generate a standardized network state snapshot containing a timestamp.

3. The network quality perception and intelligent switching method for a triple-play IoT card according to claim 1, characterized in that, The step of comparing and analyzing the network status snapshot with a pre-established historical network quality pattern library to identify deviation trends and potential degradation patterns in the current network connectivity status includes: Retrieve a set of historical network status snapshots that match the current geographical location, current time period, and current service type from the historical network quality pattern library; Calculate the Euclidean distance between the standardized network state snapshot and each snapshot in the historical network state snapshot set to obtain a set of similarity distance values; If all the similarity distance values ​​are greater than the preset similarity threshold, it is determined that the current network state deviates from the historical normal mode, and the deviation direction and magnitude of each indicator of the current snapshot relative to the historical average are calculated to form the deviation trend. Analyze the slope of change of network latency data and packet loss rate data in the standardized network status snapshot in the short term. If the slope of change exceeds the warning threshold, it is marked as a potential degradation mode.

4. The network quality perception and intelligent switching method for a triple-play IoT card according to claim 3, characterized in that, The method involves simultaneously probing the network quality of two other wireless networks accessible to the IoT card to obtain network quality evaluation parameters for the candidate networks, including: While maintaining the current network connection, the baseband chip of the IoT card sends probe request frames to the base stations or access points of the other two wireless networks through pre-configured frequency points and signaling. Receive probe response frames from base stations or access points of two other wireless networks, and parse the broadcast information of the candidate network from the probe response frames; Based on the broadcast information, the reference signal received power and the signal-to-interference-plus-noise ratio of the candidate network are measured as signal quality parameters of the candidate network. Send a set of probe packets to a specific server in the candidate network, record the round-trip time and reception success rate of the packets, and calculate the estimated network latency and packet reception rate of the candidate network. By combining the signal quality parameters, estimated network latency, and data packet reception rate of the candidate networks, network quality evaluation parameters for the candidate networks are generated.

5. The network quality perception and intelligent switching method for a triple-play IoT card according to claim 4, characterized in that, The step of predicting the quality change trajectory of the current network connection state over a specified future time period based on association rules in the historical network quality pattern library, and evaluating the quality stability of the candidate network, includes: Extract several historical scene fragments that are similar to the current deviation trend from the historical network quality pattern library. The historical scene fragments record the process of network state change from a certain initial state. Using a time series prediction algorithm, starting from the current standardized network state snapshot, and combining the evolution patterns of the historical scene fragments, the predicted network state at multiple future time points is extrapolated to form the quality change trajectory. By analyzing long-term records of the candidate network in the historical network quality pattern library within the same geographical location and time period, the historical variance and mutation frequency of each network quality evaluation parameter are calculated to evaluate the quality stability of the candidate network.

6. The network quality perception and intelligent switching method for a triple-play IoT card according to claim 5, characterized in that, The step involves generating a decision set containing multiple switching strategy options based on the deviation trend, potential degradation patterns, network quality evaluation parameters of candidate networks, and predicted quality change trajectories, including: Develop an immediate switching strategy option: The immediate switching strategy option is defined as selecting a candidate network for switching immediately based on the best current network quality assessment parameters once the potential degradation mode is identified; Develop a delayed handover strategy option: The delayed handover strategy option is defined as follows: if the predicted quality change trajectory shows that the current network will recover in the near future, then the evaluation and handover will be carried out after a preset time based on the network quality evaluation parameters at that time. Conditional switching strategy options: The conditional switching strategy options are defined as setting a set of network quality threshold conditions, and switching is triggered if and only if the current network quality index is lower than the threshold and the candidate network quality evaluation parameter is better than another higher threshold. The immediate switching strategy option, the delayed switching strategy option, and the conditional switching strategy option, together with their respective triggering conditions and expected target networks, constitute the decision set.

7. The network quality perception and intelligent switching method for a triple-play IoT card according to claim 6, characterized in that, The method further includes: invoking a preset switching strategy evaluation model to comprehensively score each switching strategy option in the decision set, including: The comprehensive score is based on network tariff costs, service continuity maintenance, and the resource consumption of the handover operation itself; The feature vector of each handover strategy option in the decision set is input into the handover strategy evaluation model. The feature vector includes the estimated handover time, the unit price of the target network, the predicted service interruption duration, and the quality stability score of the target network. The first fully connected layer inside the switching strategy evaluation model calculates the basic score of the switching strategy option in the network quality dimension based on the pre-trained weight matrix. The second fully connected layer inside the switching strategy evaluation model calculates the basic score of the switching strategy option in the cost control dimension based on another set of pre-trained weight matrices. The third fully connected layer inside the switching strategy evaluation model calculates the basic score of the switching strategy option in the business assurance dimension based on another set of pre-trained weight matrices. The basic scores of the network quality dimension, the cost control dimension, and the service assurance dimension are weighted and summed according to preset importance coefficients to obtain the comprehensive score of the switching strategy option.

8. The network quality perception and intelligent switching method for a triple-play IoT card according to claim 7, characterized in that, The method further includes: selecting the switching strategy option with the highest overall score as the target switching strategy, including: Sort the comprehensive scores of all switching strategy options in the decision set in descending order; Select the top-ranked switching strategy option as the initial target switching strategy; Check whether the initial target switching strategy and the currently executing background task or business protocol have any mandatory network locking constraints; If the mandatory network locking constraint exists, the initial target switching strategy is excluded, and the next ranked switching strategy option is selected as the new initial target switching strategy. The constraint check is repeated until a switching strategy option without constraints is found, and this option is determined as the final target switching strategy.

9. The network quality perception and intelligent switching method for a triple-play IoT card according to claim 8, characterized in that, The method further includes: generating a specific network handover instruction sequence according to the target handover strategy, including: The network handover instruction sequence includes the handover timing, the target network identifier, and parameter reconfiguration information during the handover process; Analyze the target switching strategy to determine its type as immediate switching, delayed switching, or conditional switching; If the switch is immediate, a sequence of instructions is generated, including "Disconnect current connection", "Configure target network access parameters" and "Initiate access request". If it is a delayed handover, insert the "Start Delay Timer" instruction before the sequential instruction sequence containing "Disconnect Current Connection", "Configure Target Network Access Parameters" and "Initiate Access Request", and trigger the subsequent instructions after the timer expires; If it is a conditional switch, a "start conditional monitoring thread" instruction is generated. The start conditional monitoring thread continuously checks the preset network quality threshold conditions. Once the conditions are met, a switch execution instruction is triggered. Context parameters are added to each instruction in the instruction sequence. These context parameters include the target network's operating frequency band, access point name, authentication key, and desired IP address acquisition method.

10. The network quality perception and intelligent switching method for a triple-play IoT card according to claim 9, characterized in that, The method further includes: executing the network switching instruction sequence to control the IoT card to disconnect from the current wireless network and access the target wireless network; performing network connection verification after access is completed; and updating the verification result and the network status information after the switch to the historical network quality mode library, including: Control commands are sent to the baseband chip and protocol stack of the IoT card one by one in the order of the network switching instruction sequence. After receiving confirmation that the "Disconnect current connection" command has been executed, send the "Configure target network access parameters" command; After receiving confirmation of successful parameter configuration, send the "Initiate Access Request" command and wait for the access permission response from the target network; After confirming successful access, network connection verification is performed, which includes sending a test data packet to a predefined verification server and successfully receiving a response. Collect network quality indicator data after successful verification, generate a new network status snapshot, and store the complete log of this switching event, the description of the target switching strategy, and the new network status snapshot as a new record in the historical network quality pattern library.

Images