An intelligent short message resending system and method based on a receipt state recognition

Abstract

The application relates to the technical field of short message transmission optimization, and discloses an intelligent short message resending system and method based on a return receipt state identification. A communication data acquisition module of the system acquires historical short message transmission records in a target communication area in a preset time period and generates a communication state data set; a feature extraction storage module extracts return receipt features from the data set, obtains a historical return receipt feature set, and establishes a return receipt state change matrix; a topology modeling module constructs a terminal state atlas according to the matrix, calculates the topology aggregation degree of historical short message failure events to determine an abnormal feature index set; a real-time analysis module extracts real-time return receipt features of current short message transmission data to generate a real-time state vector, and calculates a topology correlation degree with the abnormal feature index set in the terminal state atlas to generate a real-time risk factor; and a decision generation module generates a short message resending strategy instruction according to the real-time risk factor, the historical return receipt feature set and the return receipt state change matrix.

Description

Technical Field

[0001] This invention relates to the field of SMS transmission optimization technology, specifically to an intelligent SMS resending system and method based on receipt status recognition. Background Technology

[0002] In today's era of rapid development in digital communication technology, SMS, as a fundamental communication method, is widely used in scenarios such as verification code issuance, business notification push, and emergency information transmission. The stability and timeliness of its transmission directly affect enterprise service quality, user experience, and even the normal operation of critical businesses. However, in actual SMS transmission, due to various factors, SMS delivery failures are difficult to completely avoid, posing a significant challenge to communication service quality.

[0003] From the current practical application scenarios of SMS transmission, the factors affecting the success rate of SMS delivery are diverse and complex. On the one hand, there are many uncertainties at the terminal device level, such as the user's mobile phone being turned off, in airplane mode, in a signal dead zone, or insufficient terminal memory or abnormal SMS application, resulting in the inability to receive SMS receipts normally. On the other hand, fluctuations in the communication network environment can also interfere with SMS transmission. Problems such as weak base station signal coverage, network congestion, and operator transmission link failures can all lead to SMS loss or delay during transmission, thus preventing the sender from obtaining effective receipt feedback. In addition, with the continuous increase in the types of mobile terminals and the frequent updates of operating system versions, the compatibility differences of different terminals with SMS protocols are gradually becoming apparent. Some older terminals or devices from niche brands may not be able to correctly return the receipt status due to protocol compatibility issues, further increasing the difficulty of judging the SMS transmission status.

[0004] Traditional SMS transmission management solutions often lack a systematic and intelligent approach to handling SMS delivery failures. Most solutions rely solely on simple receipt information to determine successful delivery. If a failure is detected, resending the SMS at fixed time intervals is typically used as a remedy, but this approach has significant limitations. First, traditional solutions cannot deeply analyze the specific reasons for SMS delivery failures, making it difficult to distinguish between terminal device issues, network environment problems, and protocol compatibility issues. This results in a lack of targeted resending strategies, wasting communication resources and potentially interfering with users and degrading their experience due to frequent resending. Second, traditional solutions do not fully utilize the value of historical SMS transmission data. They cannot predict potential SMS delivery risks through historical data analysis and can only passively handle failures after they occur, failing to achieve proactive prevention and optimization. Furthermore, with the continuous growth of SMS traffic, traditional manual intervention and simple automated processing methods can no longer meet the needs of large-scale SMS transmission management. An automated system capable of real-time analysis and intelligent decision-making is urgently needed to improve the efficiency and accuracy of SMS resending.

[0005] Although some telecommunications companies have attempted to optimize SMS resending strategies by introducing statistical data analysis, such as collecting historical SMS success rates and failure types, to formulate resending rules based on statistical patterns, these solutions still have significant technical shortcomings. The data analysis dimensions are relatively singular, mostly focusing only on the SMS sending result itself, while ignoring key factors such as the changing patterns of receipt status over time, the dynamic evolution of terminal device status, and real-time fluctuations in the network environment. This results in an incomplete and inaccurate assessment of SMS transmission status. Existing solutions lack effective topology modeling capabilities, failing to organically combine terminal devices, network nodes, and SMS transmission links to construct a comprehensive terminal status map. This makes it difficult to accurately calculate the clustering degree of SMS failure events in complex communication networks, thus hindering the timely identification of potential anomalies. Consequently, the formulation of resending strategies remains at an empirical level, failing to achieve truly intelligent decision-making. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent SMS resending system based on receipt status recognition, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides an intelligent SMS resending system based on receipt status recognition, the system comprising:

[0008] The communication data acquisition module is used to acquire historical SMS transmission records of the target communication area within a preset time period, and generate a communication status dataset based on the historical SMS transmission records.

[0009] The feature extraction and storage module performs receipt feature extraction processing on the communication status dataset to obtain a historical receipt feature set, and establishes a receipt status change matrix based on the historical receipt features changing over time.

[0010] The topology modeling module constructs a terminal state map based on the receipt state change matrix, calculates the topological clustering degree of historical SMS failure events in the terminal state map, and determines the abnormal feature index set based on the topological clustering degree.

[0011] The real-time analysis module extracts real-time receipt features based on the current SMS transmission data, generates a real-time state vector, and performs topological correlation calculation with the abnormal feature index set within the terminal state map to generate a real-time risk factor.

[0012] The decision generation module generates SMS resending strategy instructions based on the real-time risk factors, historical receipt feature set, and receipt status change matrix.

[0013] Preferably, the communication data acquisition module is specifically used for:

[0014] Collect historical transmission records of SMS gateways within the target communication area;

[0015] Based on the type of receipt code, response delay, and network signal-to-noise ratio in historical transmission records, perform data validity verification and remove abnormal records that exceed the preset reasonable range;

[0016] The verified historical transmission records are categorized and integrated according to terminal type, base station location, and timestamp to generate a communication status dataset containing terminal response parameters, base station load parameters, and channel quality parameters.

[0017] Preferably, the feature extraction and storage module is specifically used for:

[0018] Multidimensional feature extraction is performed on the communication status dataset to obtain the historical receipt feature set;

[0019] Establish a node attribute table for the terminal state graph based on different feature dimensions;

[0020] Calculate the rate of change of each feature in the historical receipt feature set within a specified time window;

[0021] Arrange the rate of change values ​​in a time series to form a receipt status change matrix, where the rows of the matrix correspond to different receipt characteristics and the columns correspond to time sampling points.

[0022] Preferably, the topology modeling module is specifically used for:

[0023] The dimension of the terminal state map is defined based on the number of features in the receipt state change matrix, and the node coordinates of the map are composed of the change rate values ​​of the historical receipt features.

[0024] Calculate the spatial density value of the node corresponding to the historical SMS failure event in the terminal state map, and generate the topological clustering value;

[0025] A dynamic threshold is set based on the topological aggregation degree value, and the acknowledgment features corresponding to the coordinates of nodes that exceed the dynamic threshold are selected to form an abnormal feature index set.

[0026] Preferably, the real-time analysis module is specifically used for:

[0027] Analyze the terminal response delay, base station signaling load, and receipt code distribution characteristics from the current SMS transmission data;

[0028] The parsing results are mapped to a real-time state vector;

[0029] The Euclidean distance between the real-time state vector and the set of abnormal feature indicators is calculated in the terminal state graph to generate a real-time risk factor.

[0030] Preferably, the system further includes:

[0031] The strategy conflict detection module is used to perform a combined effect analysis on the real-time risk factors and the preset reissue rule priority, and calculate the strategy conflict probability value.

[0032] When the probability value of the policy conflict exceeds the preset risk threshold, a reissue rule optimization instruction is triggered.

[0033] Preferably, the system further includes:

[0034] The model correction module is used to extract the phase offset features of the real-time state vector when the deviation between the real-time state vector and the historical feedback feature set continues to exceed a set number of times.

[0035] Adjust the node weight coefficients in the terminal state map according to the phase offset characteristics;

[0036] The corrected node weight coefficients are stored in the historical feature database.

[0037] Preferably, the decision generation module is specifically used for:

[0038] Construct a decision input vector for an intelligent SMS resending system based on receipt status recognition, which includes real-time risk factors, policy conflict probability values, and historical topological clustering.

[0039] The decision input vector is input into the SMS resending strategy generation model, and the SMS resending strategy instruction is output.

[0040] Preferably, the system further includes:

[0041] The feedback learning module is used to receive terminal response data after the retransmission operation and generate feedback feature vectors.

[0042] The feedback feature vector is correlated and compared with the decision input vector to update the parameter weights of the re-sending strategy generation model.

[0043] Preferably, it includes all modules and method processes of the above-mentioned intelligent SMS resending system based on receipt status recognition.

[0044] Compared with the prior art, the beneficial effects of the present invention are:

[0045] This intelligent SMS resending system based on receipt status recognition, through its communication data acquisition module, can comprehensively acquire historical SMS transmission records of the target communication area within a preset time period and generate a complete communication status dataset. Compared to traditional methods that can only obtain partial and fragmented SMS transmission information, this module can systematically collect historical data, providing a comprehensive data foundation for subsequent data analysis and strategy formulation. By integrating a large amount of historical data, it can cover SMS transmission status of different terminal models, different network operators, and different time periods, avoiding analytical biases caused by incomplete data.

[0046] The feature extraction and storage module performs receipt feature extraction on the communication status dataset and establishes a receipt status change matrix. This enables the extraction of key features related to SMS receipt status from complex historical data, clearly presenting the changing patterns of these features over time in matrix form. This processing method transforms previously chaotic data into ordered and analyzable data, intuitively reflecting the changing trends of SMS receipt status under the influence of various factors at different time points. It overcomes the limitations of traditional methods in effectively extracting and summarizing receipt features, providing precise feature support for subsequent topology modeling and real-time analysis.

[0047] The topology modeling module constructs a terminal state map based on the receipt status change matrix and calculates the topological clustering degree of historical SMS failure events to determine the set of abnormal feature indicators. It visualizes abstract terminal states and SMS failure events in a graphical form, clearly showing the relationships between different terminals, between terminals and the network environment, and the distribution characteristics of SMS failure events within the terminal group. By calculating the topological clustering degree, it can accurately locate terminal clusters prone to SMS transmission failures and key abnormal features. Compared to traditional methods that cannot identify terminal state relationships and key anomalies, this module provides clear anomaly judgment criteria for the real-time analysis module, making subsequent risk assessments more targeted.

[0048] The real-time analysis module extracts real-time receipt features from current SMS transmission data and calculates topological correlation to generate real-time risk factors, enabling dynamic monitoring and rapid assessment of the current SMS transmission status. This module can correlate real-time data with historically established terminal status maps and abnormal feature indicator sets to promptly identify potential risks during the current SMS transmission process. Compared to traditional methods that cannot perceive the current transmission status in real time and suffer from delayed risk assessment, this module can identify SMS transmission failure risks at their initial stage, buying time for timely adjustments to resend strategies and effectively reducing the probability of SMS transmission failure.

[0049] The decision generation module combines real-time risk factors, historical receipt feature sets, and receipt status change matrices to generate SMS resending strategy instructions, enabling intelligent and personalized adjustments to the resending strategy. This module no longer relies on traditional fixed retry mechanisms but comprehensively considers the current risk situation, patterns summarized from historical data, and trends in receipt characteristics to formulate differentiated resending strategies for SMS transmission failures under different terminals and network environments. For example, for terminals with high topological correlation and large real-time risk factors, shorter monitoring intervals and more flexible link switching strategies can be selected; for terminals with stable historical receipt characteristics and small real-time risk factors, the resending interval can be appropriately adjusted to save resources. This dynamically adjusted resending strategy effectively improves the success rate of SMS resending while avoiding the waste of communication resources caused by invalid retries, enhancing the reliability and efficiency of SMS transmission, improving the user experience during SMS reception, and ensuring the stable operation of various services that rely on SMS for information transmission. Attached Figure Description

[0050] Figure 1 This is a timing diagram of the intelligent SMS resending system based on receipt status recognition described in this invention;

[0051] Figure 2 This is a flowchart of the feature extraction and storage module.

[0052] Figure 3 Workflow diagram for the topology modeling module;

[0053] Figure 4 This is a flowchart of the model correction module's workflow. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0055] Please see Figure 1 This invention provides an intelligent SMS resending system and method based on receipt status recognition, the system comprising:

[0056] This system collects historical transmission records from SMS gateways within the target communication area to generate a communication status dataset containing terminal response parameters, base station load parameters, and channel quality parameters. Based on this dataset, the system extracts multidimensional features from receipts and constructs a state change matrix, thereby establishing a terminal state map that reflects the clustering patterns of historical SMS failure events and identifying a set of abnormal feature indicators. During real-time processing, the system extracts real-time receipt features from current SMS transmission data and generates a state vector. By calculating its topological correlation with the abnormal feature indicator set in the terminal state map, accurate risk factors are generated. Combining real-time risk factors, historical feature sets, and the state change matrix, the system outputs intelligent resend strategy instructions through a resend strategy generation model. This system achieves continuous monitoring and intelligent decision-making of SMS transmission status, continuously optimizing the decision model through a closed-loop feedback mechanism, effectively improving the reliability of SMS delivery.

[0057] Example 1: See Figure 2 In the specific implementation process, the communication data acquisition module obtains historical transmission records from multiple SMS gateway devices within the target communication area. These records are stored in the form of structured logs, including millisecond-level timestamps, terminal device identifiers, base station cell numbers, transmission status receipt codes, and network signal-to-noise ratio measurements during transmission. The data acquisition process is completed through a distributed log collection system, which polls each gateway node hourly to collect all communication transaction records within a specified time period. The raw data fields include more than twenty dimensions such as transmission sequence number, terminal IMSI code, base station CI code, receipt code, latency in milliseconds, and signal-to-noise ratio in dB.

[0058] The data validity verification phase employs a multi-level filtering mechanism. The first level of filtering targets receipt codes, retaining records with complete success or failure status identifiers and removing entries with intermediate or undefined statuses. The second level of filtering is based on the reasonableness of response latency, setting upper and lower bounds between 50 milliseconds and 800 milliseconds; records exceeding this range are considered abnormal measurements and excluded. The third level of filtering checks the network signal-to-noise ratio, removing extreme values ​​consistently below 10dB or above 40dB, which typically indicate sensor malfunctions or environmental interference. After three levels of filtering, the system generates a validity verification report, recording the rejection rate and the main reasons for rejection. Verified data enters the classification and integration process. Terminal type classification is based on device identifier prefixes: mobile terminals include smartphones, IoT devices, and other mobile devices; fixed terminals include server alarm systems, monitoring equipment, and other fixed-location devices. Base station location classification uses a geographic grid coding system, dividing the target area into 1km × 1km grid units, with each base station mapped to a specific grid based on its latitude and longitude coordinates. Timestamp aggregation uses 15-minute intervals as the basic time unit, grouping records from consecutive time periods into the same time bucket. During the integration process, statistical indicators are calculated for each classification unit: terminal response parameters include average latency, latency standard deviation, and histogram of receipt code distribution; base station load parameters include peak concurrent connections and signaling message throughput; and channel quality parameters include median signal-to-noise ratio and bit error rate percentage. The final output communication status dataset is stored in a columnar format, containing fields such as time unit identifier, grid code, terminal type, and statistical indicator array.

[0059] The feature extraction and storage module performs in-depth processing on the communication state dataset. The feature extraction process focuses on statistical indicators within each time unit, calculating derived features with representative meaning. It extracts the coefficient of variation of response delay from terminal response parameters, reflecting the degree of delay fluctuation; it extracts the proportion of failure codes from the receipt code distribution, paying particular attention to the ratio of permanent to temporary failure codes; it extracts the intensity of load fluctuation from base station load parameters, obtained by calculating the absolute value of load changes in adjacent time units; and it extracts quality trend indicators from channel quality parameters, using linear fitting to calculate the slope of signal-to-noise ratio changes. These features collectively constitute the historical receipt feature set, with each feature accompanied by a source identifier and timestamp information.

[0060] A node attribute table for constructing a terminal state graph based on feature dimensions is used for storage in a graph database. Each node represents a feature dimension, and node attributes include metadata such as feature name, calculation method, numerical range, and unit of measurement. Edges are established between nodes to represent the statistical correlation between features; the correlation coefficient is calculated using historical data. A sliding time window mechanism is used to calculate the rate of change, with the window size set to five consecutive time units. For each feature, the percentage change between the current time unit value and the historical values ​​within the window is calculated, and the average rate of change within the window is taken as the feature's rate of change value at that time point. These rate of change values ​​are organized in a time series to form a status change matrix. The matrix structure uses dense matrix representation, with row indices corresponding to feature identifiers, column indices corresponding to time unit numbers, and matrix elements storing floating-point rate of change values. The matrix data is persisted to a time-series database, along with complete metadata description information.

[0061] The implementation process involves large-scale data processing, employing a distributed computing framework for parallel execution. The data acquisition phase utilizes a log collection agent deployed on each gateway node; the verification phase employs an in-memory computing engine for streaming filtering; the integration phase uses online analytical processing (OLAP) tools for multi-dimensional aggregation; the feature extraction phase calls machine learning libraries for feature computation; and the storage phase combines relational and time-series databases for hybrid storage. Data quality checkpoints are set up at all processing stages to record statistical indicators and anomaly detection results during data processing.

[0062] Example 2: See Figure 3 In constructing the terminal state graph, the topology modeling module first processes the receipt state change matrix from the feature extraction and storage module. The rows of this matrix represent different receipt feature dimensions, such as response latency variation coefficient, failure code percentage, load fluctuation intensity, and quality trend indicators. The columns represent sampling points arranged in chronological order, and the matrix elements store the rate of change value of each feature at the corresponding time point. The graph construction is based on the concept that each feature dimension constitutes an independent coordinate axis in the rate of change space, and the system state at each time point can be represented by a multi-dimensional vector composed of the rate of change values ​​of all features. The number of dimensions of the graph is equal to the number of features in the receipt state change matrix, meaning that the system state will be mapped to a high-dimensional feature space. Each node in the terminal state graph corresponds to a historical moment, and its coordinates are directly composed of the rate of change values ​​of all features at that moment. Specifically, for each time point… Its node coordinates It can be represented as:

[0063]

[0064] in: This represents the total number of receipt features. Indicates the first Individual receipt characteristics at a given time point The rate of change is numerical. In this way, time series data is transformed into a set of points in a high-dimensional space, with each point carrying dynamic information about the system's behavior at that moment. The identification of historical SMS failure events is based on the receipt code status in the original communication data. The system filters out all records where the receipt code indicates "failure" or "not delivered" and extracts the precise timestamps of these events. Each failure event is mapped to a corresponding node in the terminal status graph based on its timestamp. If multiple failure events occur at the same time point, they will share the same graph node.

[0065] The core of calculating topological clustering lies in analyzing the distribution characteristics of failed event nodes in the feature space. The system employs a spatial density calculation method based on kernel density estimation to quantify the concentration of failed events within a specific region. For each node in the graph, the density of failed event nodes in its surrounding neighborhood is calculated. The neighborhood is defined by a preset radius parameter, which defines the boundary of a hypersphere in high-dimensional space. The density value is calculated by counting the number of failed event nodes falling within the hypersphere, taking distance weights into account. Nodes that are closer contribute more to the density. The final density value is normalized and converted into a topological clustering value, ranging from 0 to 1. A higher value indicates a greater concentration of failed events in the region.

[0066] Setting the dynamic threshold is an adaptive process. The system collects the topological clustering values ​​of all nodes and analyzes their statistical distribution characteristics. The threshold is set to a high quantile of historical clustering values, specifically the 90th percentile. This means that only the top 10% of nodes with the highest clustering are considered anomalous. This dynamic adjustment mechanism allows the system to adapt to changes in the distribution of failure events under different time periods and network conditions, avoiding oversensitivity or insensitivity that might result from using a fixed threshold. The generation of the anomaly feature index set is based on nodes exceeding the dynamic threshold. The system extracts the coordinate vectors of these nodes, i.e., the combination of feature change rates that leads to high clustering. By analyzing these vectors, frequently occurring feature patterns are identified. For example, it may be found that high response delay coefficient of variation and high failure code proportion often occur simultaneously, and their change rate values ​​are within a specific range. These recurring feature combinations and their numerical ranges are recorded, constituting the anomaly feature index set. This index set essentially defines an anomalous region in a high-dimensional space; any state vector falling into or approaching this region will be considered a potentially risky state.

[0067] The entire graph construction and anomaly detection process involves a large amount of linear algebra and statistical computation. The system employs a distributed graph computing framework to handle distance calculations and density estimations in high-dimensional space, and uses dimensionality reduction techniques to visualize and verify the effectiveness of the graph. Anomaly feature indicator sets are stored in a queryable data structure for rapid matching and comparison by the real-time analysis module. This implementation establishes a correlation model between system state and failure risk through deep spatial analysis of historical data, providing a foundation for subsequent real-time risk identification.

[0068] Example 3: The real-time analysis module operates based on continuous monitoring of current SMS transmission data. The system captures signaling messages and data packets flowing through the network in real time using data probes deployed at the SMS gateway. This raw data contains rich underlying information and requires multi-step parsing processing to be transformed into useful features. The parsing process first focuses on terminal response latency. The system extracts timestamp information from each SMS transaction and calculates the time interval between sending the SMS and receiving the acknowledgment. To eliminate the impact of instantaneous fluctuations, a sliding time window is used to calculate the average latency value. The window size is set to 60 seconds, sliding every 5 seconds to ensure the real-time performance and smoothness of the data.

[0069] Base station signaling load analysis requires integrating information from multiple data sources. The system monitors performance statistics counters generated by the base station controller, including indicators such as the number of signaling messages processed per second, the number of currently connected terminals, and channel occupancy. Simultaneously, traffic data is collected from the packet core network to analyze the type distribution and throughput changes of control plane signaling messages. This data is normalized and converted into a load index between 0 and 1, reflecting the base station's processing pressure in real time. The analysis of receipt code distribution characteristics focuses on statistically analyzing the frequency of occurrence of various receipt codes. The system maintains a circular buffer to store receipt code records for all transactions within the last 300 seconds. The proportion of each type of receipt code is calculated every 10 seconds, with particular attention paid to error codes indicating transmission failure, such as "network congestion," "user unreachable," and "insufficient resources." By calculating the proportion of these error codes in the total transactions, a real-time failure rate indicator is obtained.

[0070] The process of mapping the parsed results to a real-time state vector involves multi-dimensional data fusion. The system constructs a fixed-dimensional vector space, with each dimension corresponding to a key performance indicator. The first dimension of the vector carries terminal response latency information, storing a standardized latency value calculated as the ratio of the current latency to the historical baseline latency. The second dimension represents the base station signaling load, directly using a normalized load index. The third and subsequent dimensions record the frequency of occurrence of various important acknowledgment codes, especially error code types closely related to transmission failures. Each dimension value undergoes appropriate scaling to ensure all features are within a similar numerical range, facilitating distance calculation.

[0071] When calculating topological correlation in the terminal state graph, the system first projects the real-time state vector into the same feature space as the graph. The projection process needs to consider differences in feature scale and importance. The system configures appropriate weight coefficients for each dimension; these coefficients are derived from historical data analysis and reflect the contribution of each feature to the failure risk. The Euclidean distance between the real-time state vector and each point in the set of abnormal feature indicators is calculated using a weighted Euclidean distance formula, fully considering the differences between different feature dimensions.

[0072] The real-time risk factor is generated based on the minimum distance principle. The system searches for the point closest to the current state vector in the set of abnormal feature indicators and converts this distance value into a risk score. The conversion process uses a non-linear mapping function; when the distance is small, the risk score increases rapidly; when the distance is large, the risk score increases slowly. The final real-time risk factor is a continuous value between 0 and 1. The higher the value, the closer the current system state is to the characteristic pattern of historical SMS failures, indicating a higher probability of transmission failure.

[0073] The real-time analysis process employs a pipelined architecture, with data parsing, feature extraction, vector construction, and risk calculation forming a continuous processing chain. The system incorporates a caching mechanism for each processing stage to ensure stable processing latency even during data traffic fluctuations. All intermediate results and final risk factors are timestamped and linked to the original transmitted data, creating a complete analysis log. These logs are used not only for current re-transmission decisions but also fed back into the historical database for model optimization and system improvement.

[0074] During real-time monitoring data processing, the system places particular emphasis on data timeliness and consistency. Each processing step is equipped with a timeout mechanism to ensure that even if a data source is temporarily unavailable, the system can still generate meaningful analysis results based on the latest available data. Simultaneously, data quality checks are implemented to filter and correct outliers, preventing erroneous data from affecting the accuracy of the analysis. The entire real-time analysis module is designed to provide accurate risk assessments with minimal latency, offering timely and reliable input for remedial decisions.

[0075] Example 4: See Figure 4The policy conflict detection module continuously monitors the risk factor data stream from the real-time analysis module and simultaneously accesses a pre-defined re-transmission rule base. This rule base defines multi-level re-transmission policies and their triggering conditions, including three basic types: immediate re-transmission, delayed re-transmission, and no re-transmission. Each type is associated with a risk factor threshold range and a business priority weight. The system maintains a policy execution history log, recording the actual effect of each re-transmission decision, paying particular attention to decision events that lead to abnormal increases in base station load or trigger other chain reactions. The conflict detection algorithm analyzes the matching between newly generated real-time risk factors and various re-transmission rules, assessing the potential cumulative effect when different rules are triggered simultaneously.

[0076] The combined effect analysis employs a probabilistic statistical method based on historical data. The system constructs a policy interaction matrix, recording the frequency and severity of conflicts arising from various rule combinations over a past period. For the current risk state, the algorithm identifies all potentially triggered re-issuance rules, queries historical execution records of similar rule combinations, and calculates the probability of negative impact. This probability value comprehensively considers multiple factors such as base station processing capacity, network congestion, and time period characteristics, ultimately outputting a policy conflict probability value between 0 and 1. When this probability value exceeds a preset risk threshold of 0.7, the system generates a re-issuance rule optimization instruction, which includes specific rule adjustment suggestions, such as raising the trigger threshold for certain high-risk rules or introducing execution time delays.

[0077] The model correction module operates based on continuous monitoring of the system state. This module maintains a comparison record between a real-time state vector and a historical feedback feature set, recording the degree and duration of deviation between the two. Deviation calculation employs a multi-dimensional spatial distance metric, considering not only numerical differences but also the consistency of trends. When the system detects a deviation value exceeding a set threshold five times consecutively, the phase shift analysis process is triggered. Phase shift features are extracted by comparing the alignment of real-time data with historical patterns in the time dimension. The analysis focuses on identifying shifts, compressions, or expansions in system behavior patterns over time.

[0078] Based on phase shift characteristics, the system initiates a node weight adjustment process for the terminal state graph. This weight adjustment is not a simple numerical correction, but rather considers the changes in the relative importance of different feature dimensions under the current environment. Feature dimensions exhibiting significant phase shifts are assigned new weight coefficients, calculated using machine learning algorithms. These coefficients reflect both the current degree of shift and the stability requirements of historical weight configurations. The corrected node weight coefficients are encapsulated in a weight configuration file, which contains a complete version identifier, effective timestamp, and modification description. Important parameters and decision records generated during system implementation are compiled, as shown in Table 1.

[0079] Table 1: Key Parameters for Strategy Conflict Detection and Model Correction

[0080] Parameter categories Parameter name Parameter meaning Typical values Adjustment mechanism Conflict detection Risk threshold Conflict probability threshold for triggering rule optimization 0.7 Dynamically adjust based on network conditions. Conflict detection Historical Retrospective Window Historical data range used for probability calculations 24 hours Fixed time range Conflict detection Rule priority weight Priority of different reissue rules 0.1-1.0 Based on business importance Model Correction Number of deviations Minimum number of consecutive deviations to trigger phase analysis 5 Adjusted according to system stability requirements Model Correction Phase offset tolerance Permissible phase deviation range ±15% Based on historical fluctuation range settings Model Correction Weight update range Maximum change in a single weight adjustment 0.2 Constraints to prevent over-adjustment

[0081] The conflict detection and model correction process employs a layered processing architecture. The bottom data acquisition layer is responsible for real-time monitoring and raw data collection; the middle analysis layer performs complex algorithmic calculations and decision reasoning; and the top control layer is responsible for final instruction generation and system parameter adjustment. Data exchange between layers is achieved through well-defined interfaces, ensuring the system's modularity and maintainability. Detailed audit logs are maintained for all decision-making processes, recording input data, processing steps, and output results, providing comprehensive data support for system behavior analysis. During system operation, the policy conflict detection module and the model correction module work collaboratively. When the conflict detection module identifies a potential risk, it not only triggers rule optimization instructions but also sends an environmental change alert to the model correction module. Similarly, when the model correction module detects significant changes in system behavior, it notifies the conflict detection module to reassess the applicability of the current rules. This two-way information exchange mechanism ensures that the system can adapt to constantly changing network environments and business needs, maintaining the accuracy and effectiveness of decision-making.

[0082] During implementation, special attention was paid to handling boundary conditions and abnormal states. The system is equipped with multiple protection mechanisms to prevent erroneous decisions due to data anomalies or calculation errors. These include validating input data, capturing anomalies in the calculation process, and checking the reasonableness of output results. All critical operations have rollback mechanisms, automatically restoring to the most recent safe state when an anomaly is detected. Simultaneously, the system provides a manual intervention interface, allowing operations personnel to take over system control in special circumstances, ensuring the reliability of critical business scenarios.

[0083] Example 5: The decision generation module operates on the basis of deep fusion of multi-source heterogeneous data. This module receives real-time risk factors from the real-time analysis module. These factors are quantitative indicators generated by comprehensively analyzing the terminal response delay, base station signaling load, and receipt code distribution characteristics of the current SMS transmission link, accurately reflecting the degree of abnormality in the current system state. Simultaneously, it acquires the conflict probability value output by the policy conflict detection module. This value, based on historical policy execution records and the current network state, predicts potential systemic problems that may result from the proposed retransmission strategy, including potential risks such as base station overload and signaling storms. Furthermore, the module extracts topology clustering data from the historical feature database. This data originates from statistical analysis of the spatiotemporal distribution patterns of historical failure events in the terminal state map, revealing the clustering characteristics of failure events under specific network environments. These input data undergo a rigorous standardization preprocessing procedure. The real-time risk factors are converted to relative values ​​within the 0-1 range using the min-max normalization method; the conflict probability value retains its original probability form; and the historical topology clustering data is standardized using Z-score to eliminate the influence of dimensions. The processed data is organized into a three-dimensional decision input vector, with each dimension representing a decision factor and the values ​​uniformly ranging to the [0,1] interval to ensure the numerical stability of subsequent neural network processing.

[0084] The re-sending strategy generation model employs a three-layer feedforward neural network architecture. The input layer has three neurons, corresponding to the three core dimensions of the decision input vector, with each node receiving a standardized input feature value. The hidden layer contains five neurons, using the ReLU activation function to capture the non-linear interactions between input factors. These nodes are fully connected to the input layer, learning the influence patterns of different feature combinations on the decision outcome. The output layer has one neuron, using the Sigmoid activation function to compress the output value to the 0-1 range, and then transforming it into specific re-sending strategy instructions through piecewise function mapping. The neural network parameters are trained using supervised learning, employing a training dataset containing historical decision records and actual effect annotations. The training process uses mean squared error as the loss function, calculating gradients and updating weight parameters through backpropagation. An adaptive adjustment strategy is used for the learning rate, initially set to 0.01, decreasing to 0.95 times the original value every 100 iterations. To prevent overfitting, early stopping and L2 regularization are used during training.

[0085] The interpretation of the model's output values ​​follows a strict segmented decision-making rule. When the output value is between 0 and 0.3, the system generates a no-retransmission instruction, indicating that the current risk state is within the system's acceptable range and no remedial measures are needed. When the output value is between 0.3 and 0.7, a delayed retransmission instruction is generated. The system automatically calculates the optimal delay time based on the current network load, typically adjusting dynamically between 30 seconds and 5 minutes. When the output value exceeds 0.7, an immediate retransmission instruction is triggered, and the system prioritizes allocating transmission resources to ensure the retransmission operation is executed immediately. This risk-level-based segmented decision-making mechanism ensures both the system's responsiveness and avoids resource waste caused by overreaction.

[0086] The feedback learning module constructs a complete closed-loop optimization system. This module continuously monitors the execution effect of resend commands, collecting key performance indicators such as the final delivery status of resend SMS messages, response delay data for secondary transmission, and base station load change trends. This data undergoes feature engineering and is transformed into a feedback feature vector consistent with the dimensions of the decision input vector, ensuring data comparability. The correlation comparison analysis employs a multi-dimensional correlation calculation method. The system calculates the Pearson correlation coefficient between the feedback feature vector and the original decision input vector across various feature dimensions, analyzing the changes in system state before and after the resend operation. Particular attention is paid to cases where the decision effect deviates from expectations; by comparing the feature distribution differences between successful and failed cases, blind spots and weaknesses in the decision model are identified. These analytical results are transformed into specific directions for model optimization, including suggestions for adjusting feature weights and optimization schemes for decision thresholds.

[0087] The parameter optimization process employs an incremental learning algorithm based on gradient descent. The system calculates the error loss between the actual and expected outputs of the model, propagates the error gradient layer by layer via backpropagation, and adjusts the connection weights based on the contribution of each layer's parameters. An adaptive adjustment strategy is used for the learning rate: a relatively large step size of 0.01 is used in the early stages of training to quickly approximate the optimal solution, while the step size is reduced to 0.001 in the later stages to improve convergence accuracy. After each parameter update, the system uses an independent validation dataset to evaluate model performance metrics, including decision accuracy, recall, and F1 score, to ensure the optimization direction is correct.

[0088] The updated model parameters are stored in a versioned strategy database. This database employs a multi-version concurrency control mechanism, and each parameter version contains complete metadata information, including training timestamps, data sample ranges, validation set performance metrics, and model complexity parameters. The system supports a canary release mechanism, where new parameter sets are first piloted on select nodes and then fully rolled out after thorough validation. If a new version experiences performance degradation, it can be quickly rolled back to a previous stable version, ensuring the continuity of system services.

[0089] The decision optimization process forms a virtuous cycle of continuous improvement. Each resend decision instance generates corresponding feedback data, which, after quality assessment, is added to the training sample library for incremental model learning. The system performs full-data training weekly, retraining the model using all historical data to eliminate potential bias accumulation from incremental learning. Simultaneously, a multi-level monitoring indicator system is established to track key performance indicators such as decision accuracy, response latency, and resource utilization in real time, ensuring the overall system performance remains optimal. The system places particular emphasis on the transparency and interpretability of the decision-making process. In addition to generating resend commands, it outputs detailed decision analysis reports, including contribution analysis of each input feature, decision confidence assessment, and risk factor explanations. This information helps operations personnel understand the system's decision logic and provides a basis for manual intervention. All decision-making processes are logged in complete audit logs, including input data, processing procedures, output results, and feedback information, supporting the traceability and analysis of historical decisions. The system also provides decision simulation functionality, allowing operations personnel to test decision results under different scenarios and better understand system behavior characteristics. Through the organic combination of intelligent decision-making and continuous optimization, the system achieves accurate generation and dynamic adjustment of SMS resend strategies. The segmented decision-making mechanism based on a neural network model ensures differentiated treatment for different risk levels, while the closed-loop feedback learning system guarantees the continuous evolution of the decision-making model. Multi-version management and canary release mechanisms provide reliable guarantees for stable system operation, and the transparent decision-making process enhances the system's credibility and operability.

[0090] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0091] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart SMS resending system based on receipt status recognition, characterized in that, include: The communication data acquisition module is used to acquire historical SMS transmission records of the target communication area within a preset time period, and generate a communication status dataset based on the historical SMS transmission records. The feature extraction and storage module performs receipt feature extraction processing on the communication status dataset to obtain a historical receipt feature set, and establishes a receipt status change matrix based on the historical receipt features changing over time. The topology modeling module constructs a terminal state map based on the receipt state change matrix, calculates the topological clustering degree of historical SMS failure events in the terminal state map, and determines the abnormal feature index set based on the topological clustering degree. The real-time analysis module extracts real-time receipt features based on the current SMS transmission data, generates a real-time state vector, and performs topological correlation calculation with the abnormal feature index set within the terminal state map to generate a real-time risk factor. The decision generation module generates SMS resending strategy instructions based on the real-time risk factors, historical receipt feature set, and receipt status change matrix.

2. The intelligent SMS resending system based on receipt status recognition according to claim 1, characterized in that, The communication data acquisition module is specifically used for: Collect historical transmission records of SMS gateways within the target communication area; Based on the type of receipt code, response delay, and network signal-to-noise ratio in historical transmission records, perform data validity verification and remove abnormal records that exceed the preset reasonable range; The verified historical transmission records are categorized and integrated according to terminal type, base station location, and timestamp to generate a communication status dataset containing terminal response parameters, base station load parameters, and channel quality parameters.

3. The intelligent SMS resending system based on receipt status recognition according to claim 2, characterized in that, The feature extraction and storage module is specifically used for: Multidimensional feature extraction is performed on the communication status dataset to obtain the historical receipt feature set; Establish a node attribute table for the terminal state graph based on different feature dimensions; Calculate the rate of change of each feature in the historical receipt feature set within a specified time window; Arrange the rate of change values ​​in a time series to form a receipt status change matrix, where the rows of the matrix correspond to different receipt characteristics and the columns correspond to time sampling points.

4. The intelligent SMS resending system based on receipt status recognition according to claim 3, characterized in that, The topology modeling module is specifically used for: The dimension of the terminal state map is defined based on the number of features in the receipt state change matrix, and the node coordinates of the map are composed of the change rate values ​​of the historical receipt features. Calculate the spatial density value of the node corresponding to the historical SMS failure event in the terminal state map, and generate the topological clustering value; A dynamic threshold is set based on the topological aggregation degree value, and the acknowledgment features corresponding to the coordinates of nodes that exceed the dynamic threshold are selected to form an abnormal feature index set.

5. The intelligent SMS resending system based on receipt status recognition according to claim 4, characterized in that, The real-time analysis module is specifically used for: Analyze the terminal response delay, base station signaling load, and receipt code distribution characteristics from the current SMS transmission data; The parsing results are mapped to a real-time state vector; The Euclidean distance between the real-time state vector and the set of abnormal feature indicators is calculated in the terminal state graph to generate a real-time risk factor.

6. The intelligent SMS resending system based on receipt status recognition according to claim 5, characterized in that, The system also includes: The strategy conflict detection module is used to perform a combined effect analysis on the real-time risk factors and the preset reissue rule priority, and calculate the strategy conflict probability value. When the probability value of the policy conflict exceeds the preset risk threshold, a reissue rule optimization instruction is triggered.

7. The intelligent SMS resending system based on receipt status recognition according to claim 6, characterized in that, The system also includes: The model correction module is used to extract the phase offset features of the real-time state vector when the deviation between the real-time state vector and the historical feedback feature set continues to exceed a set number of times. Adjust the node weight coefficients in the terminal state map according to the phase offset characteristics; The corrected node weight coefficients are stored in the historical feature database.

8. The intelligent SMS resending system based on receipt status recognition according to claim 7, characterized in that, The decision generation module is specifically used for: Construct a decision input vector that includes real-time risk factors, policy conflict probability values, and historical topological clustering. The decision input vector is input into the SMS resending strategy generation model, and the SMS resending strategy instruction is output.

9. The intelligent SMS resending system based on receipt status recognition according to claim 8, characterized in that, The system also includes: The feedback learning module is used to receive terminal response data after the retransmission operation and generate feedback feature vectors. The feedback feature vector is correlated and compared with the decision input vector to update the parameter weights of the re-sending strategy generation model.

10. A method for intelligent SMS resending based on receipt status recognition, characterized in that, The communication data acquisition module obtains historical SMS transmission records of the target communication area within a preset time period and generates a communication status dataset based on the historical SMS transmission records. The feature extraction and storage module performs receipt feature extraction processing on the communication status dataset to obtain a historical receipt feature set, and establishes a receipt status change matrix based on the historical receipt features changing over time. The topology modeling module constructs a terminal state map based on the receipt state change matrix, calculates the topology clustering degree of historical SMS failure events in the terminal state map, and determines the abnormal feature index set based on the topology clustering degree. The real-time analysis module extracts real-time receipt features based on the current SMS transmission data, generates a real-time status vector, and performs topological correlation calculation with the abnormal feature index set within the terminal status map to generate a real-time risk factor. The decision generation module generates SMS resending strategy instructions based on the real-time risk factors, historical receipt feature set, and receipt status change matrix.

Images