Operator billing period data quality assessment and automatic switching methods and systems

Abstract

This application provides a method and system for assessing and automatically switching operator billing period data quality, including: generating a task execution queue based on operator configuration information; detecting the operator's execution status and generating an operator detection task list; filtering target detection numbers from historical customer call logs to obtain billing period data and billing period update status; collecting a number sample set when the billing period is updated to compare the billing period data before and after the update and obtain a data consistency assessment result; and executing billing period switching or triggering an alarm based on the assessment result. This application realizes automated monitoring and intelligent switching of operator billing period data, significantly improving the timeliness and accuracy of billing period switching, reducing operation and maintenance costs, and effectively solving the problems of response delay, incomplete data anomaly detection, and high operation and maintenance costs existing in traditional semi-automated management methods.

Description

Technical Field

[0001] This application relates to the field of communication service data processing technology, and in particular to a method and system for assessing the quality of operator billing period data and for automatic switching. Background Technology

[0002] Data quality management in the telecommunications service sector has always been a critical aspect of ensuring business continuity and data accuracy. This is especially true in telecommunications service systems involving multiple operators and large data volumes, where the accuracy and timeliness of billing data have a significant impact on downstream business operations.

[0003] Traditional billing cycle data management typically employs a combination of basic monitoring and scheduled tasks. For example, it involves setting up scheduled inspection tasks to check data update status or using simple data comparison logic to determine data availability. While these methods are simple and easy to implement, they lack in-depth analytical capabilities and intelligent decision-making abilities, making it difficult to cope with the complex and ever-changing data environment of telecom operators.

[0004] Current advanced solutions typically combine manual monitoring with automated tools to monitor and switch operator billing cycle data. These solutions involve manual sampling to check data quality, and then manually triggering the billing cycle switchover process once confirmed to be error-free. Despite incorporating some automation elements, this approach still has significant technical limitations when dealing with terabyte-scale data volumes and complex billing models.

[0005] This semi-automated management approach suffers from three core problems: severe response delays, averaging 4-6 hours to complete billing cycle switching; limited manual sampling, resulting in approximately 15% of data anomalies going undetected in a timely manner; and the need for dedicated auditors for manual verification, leading to high maintenance costs. The limitations of this solution are particularly pronounced outside of working hours or when data formats are constantly changing, failing to meet the high demands of modern communication services for real-time data accuracy. Summary of the Invention

[0006] In view of this, this application provides a method and system for assessing the quality of operator billing data and for automatic switching, which solves the problems of severe response delay, incomplete data anomaly detection, and high operation and maintenance costs in the prior art.

[0007] This application provides a method for assessing the quality of operator billing period data and automatically switching over, including:

[0008] Based on the preset operator configuration information, a task execution queue is generated;

[0009] The execution status of operators in the task execution queue is detected, operators with normal execution status are identified, and an operator detection task list is generated.

[0010] Based on the operator's detection task list, target detection numbers are filtered from historical customer call logs, and billing period data and billing period update status are obtained based on the target detection numbers;

[0011] When the billing period update status is updated, a number sample set is collected from the historical customer call log according to a preset distribution strategy. The billing period data before the update and the billing period data after the update are compared using the number sample set to obtain a data consistency assessment result.

[0012] When the data consistency assessment result meets the preset comparison pass conditions, the billing period switching operation is executed; otherwise, an alarm is triggered, and the billing period is automatically switched.

[0013] Based on preset operator configuration information, a task execution queue is generated, including:

[0014] A timed task scheduler is built based on a distributed architecture. The timed task scheduler iterates through the operator list in the configuration database at a preset frequency to generate independent detection tasks for each operator. The independent detection tasks are then aggregated to generate the task execution queue.

[0015] The execution status of operators in the task execution queue is detected, operators with normal execution status are identified, and an operator detection task list is generated, including:

[0016] Perform online and offline status checks on the operators in the task execution queue, verify the reachability of the interface address and the validity of the authorization credential, and obtain the status verification result;

[0017] Based on the status verification results, operators that are offline or whose interface addresses are unreachable are removed, resulting in the operator detection task list.

[0018] Based on the operator's detection task list, target detection numbers are filtered from historical customer call logs, and billing period data and billing period update status are obtained based on the target detection numbers, including:

[0019] Perform a multi-condition joint query on the historical customer call logs to filter out records of the target operator whose call billing period data is the current billing period data, and sort them in descending order by the absolute value of the interface return value to obtain a set of candidate numbers;

[0020] Based on the candidate number set, the first N valid numbers are selected as the detection targets using a sliding window mechanism to obtain the target detection numbers;

[0021] Based on the target detection number, the results are parsed through a pre-configured status code mapping table to obtain each billing period data and the corresponding billing period update status.

[0022] Filter target detection numbers from historical customer call logs, including:

[0023] For each number in the historical log data, a multi-dimensional state vector is constructed, including historical query frequency, return value volatility, return value size, frequency of special values, and query success rate, to obtain the number feature state space;

[0024] Based on the number feature state space, the value boundary, stability boundary, and responsiveness boundary are determined. A flux function is set for each boundary region, and a jump condition is set at each boundary intersection. Neumann boundary conditions are applied to the boundary of the number feature state space. By calculating the Laplace equation containing the jump condition, the flux distribution of the number feature state space is obtained.

[0025] Based on the flux distribution, random sample points conforming to the distribution are generated by constructing a cumulative distribution function and an inverse function sampling method, and based on the random sample points, number groups with an approximation higher than a preset threshold in the number feature state space are determined;

[0026] Based on the number group, a target detection number is generated.

[0027] When the billing cycle update status is "updated," a sample set of phone numbers is collected from the historical customer call logs according to a preset distribution strategy. The billing cycle data before and after the update are compared using this sample set to obtain a data consistency assessment result, including:

[0028] Based on a preset distribution strategy, the number sample set is collected from the historical customer call logs to obtain distributed sample data;

[0029] For the distributed sample data, the return values ​​of the billing period data before and after the update are obtained through the operator interface, and compared one by one to obtain the comparison data structure.

[0030] Based on the comparison data structure, the overall complete inconsistency rate and the overall fuzzy inconsistency rate are calculated. When either exceeds a preset threshold, an abnormal termination mechanism is triggered to obtain the data consistency evaluation result.

[0031] Based on the aforementioned comparison data structure, the overall complete inconsistency rate and the overall fuzzy inconsistency rate are calculated, including:

[0032] Based on the sample data of the payment period data before the update and the sample data of the updated payment period data in the comparison data structure, feature extraction is performed to obtain a multidimensional feature vector, which includes the sample mean, median, variance, skewness and kurtosis of the multidimensional feature vector.

[0033] The multidimensional feature vector is standardized to construct the sample data of the payment period data before the update and the empirical distribution function of the payment period data after the update. Based on the size of the sample data, the maximum deviation statistic and significance probability are calculated.

[0034] Multiple sub-distributions are constructed for the multidimensional feature vectors according to their dimension types, and Kolmogorov tests are performed independently on each sub-distribution to obtain the Kolmogorov analysis results;

[0035] Based on the maximum deviation statistic, the significance probability, and the Kolmogorov analysis results, the overall complete inconsistency rate and the overall fuzzy inconsistency rate are calculated.

[0036] When the data consistency assessment result meets the preset comparison pass conditions, the billing period switching operation is executed; otherwise, an alarm is triggered, and the automatic billing period switching result is obtained, including:

[0037] Determine the decision variable vector containing switching decisions, alarm levels, and detection time adjustment coefficients; construct the constraint matrix and basic condition vector; establish a mathematical model of the linear complementarity problem; construct the interval-extended constraint matrix and vector based on the total complete inconsistency rate and the total fuzzy inconsistency rate; apply interval arithmetic rules to calculate the constraint conditions; and obtain the interval-based constraint expression.

[0038] A parameter space is constructed based on the interval constraint expression, and the stable switching region, stable non-switching region, sensitive boundary region and uncertain region in the parameter space are determined, and the boundaries of each region and the corresponding confidence index are determined.

[0039] Determine the corresponding region of the data consistency assessment result in the parameter space, and determine the corresponding decision rule based on the confidence index corresponding to the region;

[0040] The decision results are output according to the decision rules, and a historical database is built based on the decision results. The optimal decision boundary is learned through a vector machine, and the constraint parameters of the linear complementarity problem are updated periodically to perform adaptive optimization of the decision rules.

[0041] Determining the stable switching region, stable non-switching region, sensitive boundary region, and uncertain region in the parameter space includes:

[0042] For the decision variable vector, construct the feasible region of the linear complementarity problem, apply principal component transformation to simplify the parameter space dimension, determine the boundary equation of each region in the parameter space based on critical point theory, calculate the symmetry characteristics of the solution set, including the center point, symmetry axis and invariant subspace, analyze the topological structure and geometric characteristics of the solution set under parameter changes, and obtain the shape description of the symmetric solution set.

[0043] Based on the symmetric solution set shape description, the stability region and sensitivity boundary of the solution set are identified, the stability radius and perturbation tolerance of the solution are calculated, the uniqueness and existence conditions of the solution are calculated for each region, the mapping relationship between solution set shape change and parameter change is established, and the solution set sensitivity analysis results are obtained.

[0044] Based on the symmetric solution set shape description and the solution set sensitivity analysis results, the parameter space is adaptively partitioned to determine the stable switching region, the stable non-switching region, the sensitive boundary region, and the uncertain region.

[0045] This application also provides a system for assessing and automatically switching data quality during operator billing periods, including:

[0046] The queue generation module is used to generate task execution queues based on preset operator configuration information.

[0047] The task scheduling module is used to detect the execution status of operators in the task execution queue, determine the operators whose execution status is normal, and generate a list of operator detection tasks.

[0048] The status detection module is used to filter target detection numbers from historical customer call logs based on the operator's detection task list, and obtain billing period data and billing period update status based on the target detection numbers;

[0049] The consistency assessment module is used to collect a number sample set from the historical customer call log according to a preset distribution strategy when the billing period update status is updated, and to compare the billing period data before and after the update with the number sample set to obtain the data consistency assessment result.

[0050] The payment period switching decision module is used to execute the payment period switching operation when the data consistency assessment result meets the preset comparison pass conditions; otherwise, it triggers an alarm and obtains the automatic payment period switching result.

[0051] This application embodiment also provides a computer device, the computer device comprising:

[0052] At least one processor; and,

[0053] A memory communicatively connected to the at least one processor; wherein,

[0054] The memory stores instructions that can be executed by the at least one processor, which enables the at least one processor to perform the above-described method for assessing and automatically switching operator billing data quality.

[0055] This application also provides a computer-readable storage medium storing computer instructions for causing a computer to execute the above-described method for assessing and automatically switching operator billing data quality.

[0056] This application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the above-described operator billing data quality assessment and automatic switching method.

[0057] This application has the following technical effects:

[0058] Through automated detection and intelligent decision-making processes, the automatic update detection and quality assessment of operator billing period data has been achieved, effectively solving the response delay problem in traditional solutions and reducing the billing period switching time from 4-6 hours to minutes.

[0059] By employing a multi-dimensional data consistency comparison mechanism and advanced statistical analysis methods, the accuracy and comprehensiveness of data anomaly detection have been improved, increasing the anomaly detection rate from 85% to over 98%.

[0060] Based on the linear complementarity problem and symmetric solution set shape analysis of interval data, robust decision-making under uncertain conditions is achieved, significantly reducing the misjudgment rate and the need for manual intervention, and reducing operation and maintenance costs by about 70%. Attached Figure Description

[0061] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 This is a flowchart of a method for assessing and automatically switching operator billing period data, provided in an embodiment of this application.

[0063] Figure 2 This is a flowchart of detecting the operator's execution status and generating an operator detection task list provided in an embodiment of this application;

[0064] Figure 3 This is a flowchart provided in an embodiment of the present application for filtering target detection numbers from historical customer call logs and obtaining billing period data;

[0065] Figure 4 This is a schematic diagram of the structure of a carrier billing period data quality assessment and automatic switching system provided in an embodiment of this application. Detailed Implementation

[0066] 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.

[0067] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0068] like Figure 1 As shown in the figure, this application provides a method for assessing the quality of operator billing period data and automatically switching over, including:

[0069] S1: Generate a task execution queue based on preset operator configuration information;

[0070] The system generates a task execution queue based on preset operator configuration information. Specifically, the system maintains a configuration database that stores core information about all operators to be detected, including operator identifiers, interface addresses, request parameter formats, response structure parsing rules, and other basic information. Simultaneously, the system configures core parameters for each operator individually, such as detection time window parameters, number of data samples, billing cycle switching thresholds, and special value handling rules. Based on these configurations, the system constructs a distributed architecture scheduled task system. The scheduler iterates through the operator list in the configuration database at a preset frequency (default once per minute), generating independent detection tasks for each operator and aggregating these tasks into the task execution queue. The system tracks the execution status and results of each detection task through task IDs. When an abnormal status occurs, failure retries and timeout controls are implemented to ensure the reliability and efficiency of the entire detection process.

[0071] S2: Detect the execution status of operators in the task execution queue, determine the operators with normal execution status, and generate an operator detection task list;

[0072] The system monitors the execution status of operators in the task execution queue, identifies operators with normal status, and generates an operator detection task list. Before each detection begins, the system first verifies the operator's current status and configuration integrity, including checking the operator's online / offline status identifier, interface address reachability, authorization credential validity, and the reasonableness of the detection time window setting. If an operator is found to be in a manually marked offline state, with an unreachable interface address, or missing critical configurations (such as a missing authorization key), the system automatically skips the subsequent detection process for that operator, records detailed reasons for the skip, and generates corresponding system event logs. This mechanism ensures that system resources are not wasted on operators that cannot respond normally, improves overall detection efficiency, and also avoids system anomalies that may be caused by incomplete configurations.

[0073] S3: Based on the operator's detection task list, filter the target detection number from the historical customer call log, and obtain the billing period data and billing period update status based on the target detection number;

[0074] The system, based on the operator's detection task list, filters target detection numbers from historical customer call logs and obtains billing period data and update status based on these numbers. The system includes a historical call log database that records key information from all customer query requests. First, a multi-condition joint query is executed to filter records of the target operator whose requested billing period is the currently in use, and these records are sorted in descending order of the absolute value of the interface return value to form a candidate number set. Then, based on a sliding window mechanism, the top N valid numbers are selected as detection targets to ensure that numbers recently used for detection are not reused, thereby improving sample diversity. For the selected target numbers, standardized interface request parameters are constructed, the billing period parameter is set to the new billing period to be detected, and the interface call is initiated. Through a pre-configured status code mapping table, the system parses the returned results to identify whether the billing period data exists, thereby determining whether the operator's latest billing period has been updated. When any number returns a "found" status code and valid service data, the system determines that the operator's billing period has been updated and records detailed detection process data.

[0075] S4: When the billing period update status is updated, collect a number sample set from the historical customer call log according to a preset distribution strategy, and compare the billing period data before the update and the billing period data after the update using the number sample set to obtain a data consistency assessment result;

[0076] When the billing period is confirmed to have been updated, the system collects a sample set of numbers from historical customer call logs according to a preset distribution strategy. These samples are then used to compare the billing period data before and after the update to obtain a data consistency assessment result. The system defines the updated billing period as the new billing period and the currently used online billing period as the old billing period. Based on a preset strategy (e.g., 20% for special values, 40% for positive values, 20% for zero values, and 20% for negative values), the system collects samples from historical logs. For each collected number, the system calls the operator's interface with the new billing period as a parameter to obtain new data and compares it one-to-one with the old billing period data to form a standard comparison data structure. Based on the comparison results, the system calculates key indicators such as the Overall Complete Inconsistency Rate (OMR) and the Overall Fuzzy Inconsistency Rate (OFMR). Simultaneously, the system monitors abnormal indicators in real time, such as the inconsistency rate of specific categories, the number of consecutive inconsistent records, and changes in data type distribution. If necessary, it triggers a circuit breaker mechanism to terminate the comparison, preventing erroneous judgments due to sample or interface anomalies.

[0077] S5: When the data consistency assessment result meets the preset comparison pass condition, the billing period switching operation is executed; otherwise, an alarm is triggered, and the billing period automatic switching result is obtained.

[0078] Based on the data consistency assessment results, the system decides whether to perform a billing period switch or trigger an alarm. When the assessment results meet preset conditions (e.g., no circuit breaker mechanism is triggered, and OMR is below 5%, OFMR is below 15%), the system automatically performs the billing period switch, updates the "Current In-Service Billing Period" field in the operator's configuration database to the latest detected billing period value, and updates the distributed cache to ensure real-time effectiveness. The system records detailed switch information, including the billing period values ​​before and after the switch, the judgment criteria, and key indicator values, and stores the results in a notification queue, sending a summary notification at a preset time. If the assessment results do not meet the conditions, the system does not perform the switch, but keeps the current billing period unchanged and triggers a multi-level alarm mechanism to notify the responsible party. The system generates a detailed anomaly analysis report, including sample data, anomaly indicators, and possible cause analysis, to assist manual intervention decisions. In addition, the system implements continuous monitoring after the switch, verifying the new billing period data at a higher frequency in the first hour after the switch. If an anomaly is detected, a rollback operation is automatically triggered to restore the state before the switch, ensuring business continuity.

[0079] The operator billing period data quality assessment and automatic switching method provided in this application, through automated detection and intelligent decision-making processes, achieves automatic update detection and quality assessment of operator billing period data, solving the problems of severe response delays, incomplete data anomaly detection, and high operation and maintenance costs in traditional solutions. The technical solution of this application reduces billing period switching time from 4-6 hours to minutes, increases the anomaly detection rate from 85% to over 98%, and significantly reduces the need for manual intervention, lowering operation and maintenance costs by approximately 70%.

[0080] S1 generates a task execution queue based on preset operator configuration information, including:

[0081] A timed task scheduler is built based on a distributed architecture. The timed task scheduler iterates through the operator list in the configuration database at a preset frequency to generate independent detection tasks for each operator. The independent detection tasks are then aggregated to generate the task execution queue.

[0082] In one embodiment of this application, the scheduled task operator employs a primary-backup dual-node hot standby mode to ensure high availability. The scheduler iterates through all preset operator lists in the configuration database at a preset frequency (default once per minute, adjustable via configuration). For each operator, the system generates an independent detection task and submits it to the task execution queue, using an asynchronous parallel processing mechanism to ensure that multi-operator detection does not interfere with each other. The system tracks the execution status and results of each detection task through the task ID, and can perform retry and timeout control when an abnormal status occurs.

[0083] Specifically, the system pre-defines and centrally manages the core configuration information of all operators to be tested, including basic information such as operator identifier, interface address, request parameter format, and response structure parsing rules. Simultaneously, it individually configures core parameters for each operator, including the detection time window (e.g., 0:00-23:59), the number of data samples per run (default 200, configurable), the billing period switching threshold (e.g., default difference rate threshold of 10%), special value handling rules, and the distribution logic for billing period comparison values ​​(20% for special values, 40% for positive values, 20% for zero values, and 20% for negative values). All configuration parameters are stored in the configuration database and can be dynamically modified without a system restart.

[0084] like Figure 2 As shown, S2 detects the execution status of operators in the task execution queue, determines operators with normal execution status, and generates an operator detection task list, including:

[0085] S2.1: Perform online and offline status checks on the operators in the task execution queue, verify the reachability of the interface address and the validity of the authorization certificate, and obtain the status verification result;

[0086] S2.2: Based on the status verification results, remove operators that are offline or whose interface addresses are unreachable to obtain the operator detection task list.

[0087] In one embodiment of this application, before each polling detection begins, the system first verifies the operator's current status and configuration integrity. Specifically, the system checks the operator's online / offline status identifier, interface address reachability, authorization credential validity, and the reasonableness of the detection time window setting. If an operator is detected to be in a manually marked offline state, its interface address is unreachable, or critical configurations are missing (such as a missing authorization key), the system automatically skips the subsequent detection process for that operator and records a detailed reason for skipping, while generating a corresponding system event log. This mechanism ensures that system resources are not wasted on operators that cannot respond normally, improving overall detection efficiency.

[0088] like Figure 3 As shown, S3, based on the operator's detection task list, filters target detection numbers from historical customer call logs, and obtains billing period data and billing period update status based on the target detection numbers, including:

[0089] S3.1: Perform a multi-condition joint query on the historical customer call logs, filter out records of the target operator whose call billing period data is the currently used billing period data, sort them in descending order by the absolute value of the interface return value, and obtain a set of candidate numbers;

[0090] In one embodiment of this application, the system establishes a dedicated historical call log database to record key information of all customer query requests in real time, including the query number, operator identifier, call billing period, interface return status code, and service return value. Based on this database, the system performs multi-condition joint queries: first, it filters out records of the target operator, and then further filters out records whose call billing period is the currently used online billing period. In the set of records that meet the above conditions, the system sorts them in descending order according to the absolute value of the interface return value, and prioritizes the numbers with larger return values ​​as detection targets. The system adopts a sliding window mechanism to ensure that the numbers used for detection in the most recent N times (default N=10) are not reused, thereby improving sample diversity.

[0091] S3.2: Based on the candidate number set, the first N valid numbers are selected as detection targets using a sliding window mechanism to obtain the target detection numbers;

[0092] In addition, the system uses the selected target numbers (the first 3 valid numbers are selected by default) to construct standardized interface request parameters, where the billing period parameter is set to the new billing period to be tested (usually the current date or the current date minus 1 day). The system initiates interface calls in parallel and performs unified status code parsing on the returned results. The system uses a pre-configured status code mapping table to identify states such as "No such number" and "Not found," indicating that the data does not exist. If all test numbers return a "Data not found" status, it is determined that the latest billing period of the operator has not yet been updated; if any number returns a "Found" status code and valid service data, it is determined that the latest billing period of the operator has been updated. The system writes the judgment results into the detection result table and records detailed detection process data.

[0093] S3.3: Based on the target detection number, the results are parsed through the pre-configured status code mapping table to obtain each of the billing period data and the corresponding billing period update status.

[0094] A billing cycle update monitoring mechanism is established. When a billing cycle of a particular operator is continuously monitored and exceeds a preset time (e.g., 48 hours without update), the system automatically generates a high-priority alarm event and notifies a pre-defined group of responsible personnel through multiple channels (SMS, email, enterprise instant messaging tools, etc.). Secondly, the system employs a dual-verification mechanism to ensure handover security. When it detects that the latest billing cycle of an operator has been updated, it does not directly trigger a billing cycle switch. Instead, it marks the detection status as "Pending Verification" and automatically initiates the billing cycle data consistency comparison process. The system maintains detailed status transition logs, recording the entire process from "Detecting" to "Pending Verification" and then to the final status.

[0095] S3.1 filters target detection numbers from historical client call logs, further including:

[0096] S3.1.1: For each number in the historical log data, construct a multi-dimensional state vector including historical query frequency, return value volatility, return value size, frequency of special values, and query success rate to obtain the number feature state space;

[0097] In one embodiment of this application, the system first constructs a multi-dimensional state space describing the characteristics of the number and defines state vectors. Each dimension represents: This refers to the historical query frequency (the number of times a query is called within a unit of time). For the volatility of the return value (standard deviation of the most recent N queries). The size of the return value (the normalized value). This represents the frequency of special values ​​(the percentage of special values ​​returned). This represents the query success rate (the percentage of queries that return a successful response). Identify business attributes (such as the number's carrier and region code). Analyze historical log data, calculate the state vector for each number, and apply Principal Component Analysis (PCA) to reduce dimensionality, retaining the principal components needed to explain 90% of the variance, thus reducing computational complexity.

[0098] S3.1.2: Based on the number feature state space, determine the value boundary, stability boundary and responsiveness boundary, set a flux function for each boundary region and set a jump condition at each boundary intersection, apply von Neumann boundary conditions to the boundary of the number feature state space, and obtain the flux distribution of the number feature state space by calculating the Laplace equation containing the jump condition;

[0099] To address the specific needs of billing cycle update detection, the system defines state space boundaries with non-uniform throughput characteristics and identifies key boundaries in the state space, including: high-value boundaries (number regions with large and stable return values), high-stability boundaries (number regions with low fluctuations), and high-responsiveness boundaries (number regions with high success rates). A throughput function is defined for each boundary region. This indicates the tendency to select samples from this region, and a jump condition is set at the boundary: ,in For jump functions, and These represent the flux potential function values ​​taken by the same point S when approached from both sides at the boundary intersection, which can also be understood as the "left and right limit values / positive and negative side values" on both sides of the boundary. Based on business rules, the sigmoid function is used to smooth the jump function and avoid overly drastic changes.

[0100] S3.1.3: Based on the flux distribution, random sample points conforming to the distribution are generated by constructing a cumulative distribution function and an inverse function sampling method, and based on the random sample points, number groups with an approximation higher than a preset threshold in the number feature state space are determined;

[0101] The system introduces von Neumann boundary conditions to handle the sampling problem on the state space boundary. Application conditions: ,in For the boundary normal vector, Represents the state space. For the preset boundary gradient function, Representing the state space The boundary. Design the boundary gradient function. This allows the gradient to: point inwards from the high-value boundary, increasing the selection probability; point outwards from the low-stability boundary, decreasing the selection probability; and dynamically adjust the gradient direction based on historical response times at the responsiveness boundary. Solve the Laplace equation with von Neumann boundary conditions: This yields the flux distribution across the entire state space.

[0102] S3.1.4: Generate target detection numbers based on the number group.

[0103] Based on the aforementioned flux distribution, the system implements an adaptive importance sampling strategy and defines an importance function. Normalize the flux values ​​to the [0,1] interval and construct the cumulative distribution function (CDF):

[0104]

[0105] in Let Φ(S) be the maximum value of the flux potential function Φ(S) over the entire state space Ω, and S' be the integration variable, corresponding to any point in the state space. The integration range is the entire state space. A conformal distribution is generated using the inverse function sampling method. For each random sample point, the closest actual number in the state space is found. The Markov Chain Monte Carlo (MCMC) method is introduced, and the sampling efficiency is optimized by the Metropolis-Hastings algorithm.

[0106] First, the system constructs a multi-dimensional state vector for each number in the historical log data. For example, for number 186xxxx5678, the system analyzes its call records over the past 30 days and finds that this number is queried an average of 15 times per day (high historical query frequency), the standard deviation of the return values ​​of the most recent 10 queries is 2.3 (low return value volatility), the average return value is 87.6 (relatively high return value), the occurrence rate of special values ​​is 5% (low frequency of special values), and the query success rate is 99% (high query success rate). The system combines these features into the state vector of this number, and similarly processes all historical numbers, ultimately forming a feature state space containing tens of thousands of numbers.

[0107] Secondly, the system defines three key boundary regions based on this state space. Taking mobile operators as an example, the system identifies: high-value boundary regions include numbers with large and stable return values, such as corporate switchboard numbers and VIP customer numbers; high-stability boundary regions include ordinary user numbers that have been stable for a long time; and high-responsiveness boundary regions include numbers with fast query responses and high success rates. The system sets different sampling weights for each region; for example, the weight for high-value regions is 0.5, the weight for high-stability regions is 0.3, and the weight for high-responsiveness regions is 0.2. At the boundaries between regions, the system sets smooth transition weight change rules to avoid sampling bias towards specific regions. For boundary condition processing, the system ensures continuous change in sampling probability at the edge of the region; for example, for numbers that are about to be cancelled (state space boundary), the system will correspondingly reduce their probability of being selected.

[0108] Through this boundary setting and processing, the system generates a flux distribution map of the entire number space. In practical applications, such as billing update detection for telecom operators, the system may find that number ranges with return values ​​between 80 and 120, query success rates above 95%, and volatility below 5% have the highest flux values. This means that these numbers are most suitable for detecting billing updates.

[0109] Based on flux distribution, taking a carrier billing cycle update detection as an example, the system randomly generates 100 sampling points that conform to the distribution characteristics, and then finds the real numbers in the actual number database that are closest in characteristics to these sampling points. For example, the system might select numbers such as 186xxxx5678 (enterprise user, high-value area), 139xxxx4321 (ordinary stable user, high-stability area), and 156xxxx5678 (newly registered but highly active user, high-responsiveness area). In this way, the system ensures the diversity and representativeness of the sampled number groups, avoiding sampling bias that may occur in traditional methods.

[0110] In a single test, the system might select 30 of the most representative numbers from 1 million historical records using the aforementioned mechanism. These numbers come from different user groups and usage scenarios, comprehensively reflecting the billing cycle update status. By testing these 30 numbers, the system can accurately determine whether the billing cycle has been updated and the quality of the update, greatly improving the accuracy and efficiency of the detection. It's worth noting that the system also records the results of each test, continuously optimizing the sampling strategy. For example, if it finds that the detection success rate for a certain type of number (such as newly registered users) is low, the system will adjust the sampling weight of that type of number accordingly in subsequent tests, thereby achieving self-learning and optimization.

[0111] This number selection strategy based on multidimensional state vectors and boundary conditions, compared with the traditional simple sorting and selection method, can more comprehensively consider various characteristics and applicable scenarios of numbers, significantly improving the accuracy and efficiency of billing cycle update detection. Its advantages are even more obvious when dealing with large-scale, complex and ever-changing operator data environments.

[0112] Furthermore, for the above screening mechanism, sampling optimization is performed using the following methods: record the numbers used in each detection and their detection results (success / failure, response time, return value, etc.), construct a Bayesian network model, analyze the conditional probability relationship between number characteristics and detection effect, calculate the posterior probability P(detection success|S), update the importance function: w'(S) = w(S)·P(detection success|S), periodically (default every 24 hours) recalculate the throughput distribution to adapt to the dynamic changes in operator data characteristics, set the exploration-utilization balance parameter ε (default 0.1), select purely random samples with probability ε, and avoid getting trapped in local optima.

[0113] Compared to the simple descending selection strategy in the original scheme, the Monte Carlo method based on non-uniform throughput can select statistically superior detection numbers, improving the detection accuracy from 92% to 98.5%. By introducing von Neumann boundary conditions, the system can better handle boundary situations, such as special number segments and newly launched numbers, reducing false positives caused by boundary effects by approximately 35%. The dynamic feedback mechanism allows the system to continuously learn and optimize the sampling strategy, adapting to changes in the data characteristics of different operators, resulting in an average improvement in detection efficiency of 40%. Importance sampling technology significantly reduces the number of samples required, reducing API calls by more than 50% while maintaining the same detection accuracy, thus reducing system load and operator interface pressure. The non-uniform throughput model has a stronger adaptability to sudden changes in data distribution, enabling rapid adjustment of the sampling strategy to maintain detection accuracy when the operator's data structure changes.

[0114] S4 When the billing period update status is "updated", a number sample set is collected from the historical customer call log according to a preset distribution strategy. The billing period data before and after the update are compared using the number sample set to obtain a data consistency assessment result, including:

[0115] S4.1: Based on a preset distribution strategy, collect the number sample set from the historical customer call logs to obtain distributed sample data;

[0116] In one embodiment of this application, the function is activated immediately upon detecting a billing period update, defining the updated billing period as the new billing period (denoted as Value_new) and the currently used online billing period as the old billing period (denoted as Value_old). Call records for the currently used online billing period are extracted from historical customer call logs. Intelligent data collection is performed according to a preset strategy (usually 200 records by default, configurable adjustment supported) and the following proportion rules: 20% for special values, 40% for values ​​greater than zero, 20% for values ​​equal to zero, and 20% for values ​​less than zero. The collection results include the query number corresponding to the old billing period and its return value (Value_old).

[0117] S4.2: For the distributed sample data, obtain the return values ​​of the billing period data before and after the update through the operator interface, and compare them one by one to obtain the comparison data structure;

[0118] The system uses the collected number set and the latest billing period as the query parameter to re-call the operator's interface to obtain the return value (Value_new) for the new billing period. To avoid excessive interface pressure, the system adopts an adaptive concurrency control strategy, initially set to 10 concurrent requests, dynamically adjusting the number of concurrent requests based on the interface response time to ensure efficient data collection without impacting the operator's system. The system standardizes the query results for each number, uniformly extracting the core business return value (Value_new) and matching it one-to-one with the corresponding value (Value_old) in the historical records, forming a standard comparison data structure of <number, Value_old, Value_new, query status>. All comparison data is written to a temporary comparison result table in real time, and complete request and response raw data is recorded to facilitate subsequent problem localization and analysis.

[0119] S4.3: Based on the comparison data structure, calculate the overall complete inconsistency rate and the overall fuzzy inconsistency rate. When either exceeds a preset threshold, trigger the abnormal termination mechanism to obtain the data consistency evaluation result.

[0120] The system performs multi-dimensional discrepancy analysis based on the compared data structure. First, according to predefined consistency classification criteria, the system labels each record as "completely consistent," "fuzzy consistent," or "completely inconsistent." Completely consistent means that Value_old and Value_new have exactly the same value; fuzzy consistent means that the difference rate does not exceed a preset threshold R (default 10%), calculated using the formula: |Value_new - Value_old| / max(|Value_old|, baseline value), where the baseline value defaults to 1.0 and is configurable; completely inconsistent includes changes in status (e.g., from "found" to "not found") or a difference rate exceeding the threshold R. Second, the system calculates the proportion of completely inconsistent samples according to data type (special values / positive values / zero values / negative values), and calculates the classification inconsistency rate (CMR_category) for each category. Finally, the system calculates the overall complete inconsistency rate (OMR) and the overall fuzzy inconsistency rate (OFMR) as key decision indicators. The system also implements a dynamic threshold adjustment mechanism, which automatically adjusts the judgment threshold based on the actual number of days between the old and new payment periods. When the interval exceeds 1 day, the tolerance threshold is linearly increased by 5% for each additional day.

[0121] During the comparison process, the system monitors three types of abnormal indicators in real time: classification inconsistency rate, consecutive inconsistency count, and data type distribution change. When the inconsistency rate (CMR_category) of any category exceeds the preset threshold T_c (default 5%), the system immediately triggers a classification inconsistency rate circuit breaker. When N (default N=10, configurable) completely inconsistent records are detected consecutively, the system triggers a consecutive inconsistency circuit breaker. When the system detects a significant change in data type distribution between the old and new accounting periods (e.g., more than 30% of positive data becomes negative or special values ​​in the new accounting period), the system triggers a data type distribution drastic change circuit breaker. Upon triggering any circuit breaker condition, the system immediately terminates the current comparison task, marks the status as "abnormal termination," and generates a detailed anomaly report, including the triggering conditions, anomaly data examples, and suggested handling solutions. Simultaneously, the system sends an emergency alarm notification to the preset responsible personnel group.

[0122] S4.3 Based on the aforementioned comparison data structure, calculate the overall complete inconsistency rate and the overall fuzzy inconsistency rate, including:

[0123] S4.3.1: Based on the sample data of the payment period data before the update and the sample data of the updated payment period data in the comparison data structure, feature extraction is performed to obtain a multidimensional feature vector, wherein the multidimensional feature vector sample means, median, variance, skewness and kurtosis are included.

[0124] In one embodiment of this application, the system first extracts features from data samples of new and old payment periods, constructs a multi-dimensional feature vector, and classifies all data values ​​according to preset intervals (such as negative intervals, zero intervals, low intervals, median intervals, and high intervals) to obtain a discrete frequency distribution. Key statistical indicators, including higher-order moments such as sample mean, median, variance, skewness, and kurtosis, are calculated to form a distribution feature fingerprint. The feature vector is then standardized to eliminate the influence of dimensions, making data of different scales comparable.

[0125] S4.3.2: Standardize the multidimensional feature vector to construct the sample data of the payment period data before the update and the empirical distribution function of the payment period data after the update, and calculate the maximum deviation statistic and significance probability based on the size of the sample data;

[0126] Based on the standardized feature vectors, the system performs an enhanced version of the Kolmogorov-Smirnov test to construct an empirical distribution function for the new and old payment period data: and Calculate the maximum deviation statistic: That is, the maximum vertical distance between the two distribution functions, where, and These are the empirical distribution functions for the new and old payment period data, respectively, based on the sample size. Calculate the critical value : ,in significance level The corresponding constants, n1 and n2, represent the sample sizes for the old and new payment terms, respectively. The decision rule is as follows: If the null hypothesis that "the two distributions are the same" is rejected, then a significant difference is considered to exist.

[0127] S4.3.3: Construct multiple sub-distributions for the multidimensional feature vectors according to their dimension types, and independently perform the Kolmogorov test on each sub-distribution to obtain the Kolmogorov analysis results;

[0128] To improve sensitivity to local anomalies, the system implements hierarchical Kolmogorov analysis, constructing sub-distributions according to data types (special values, positive values, zero values, and negative values). Kolmogorov tests are performed independently on each sub-distribution to obtain the bias statistics of the sub-distributions. and value A weighted fusion strategy is used to calculate the comprehensive deviation index: ,in The weights for each sub-distribution are determined based on business importance and sample size. Multi-level thresholds are set based on the Comprehensive Deviation Index (DI): Normal (…) ),warn( ),abnormal( T1 and T2 are the grading thresholds of DI, respectively.

[0129] S4.3.4: Based on the maximum deviation statistic, the significance probability, and the Kolmogorov analysis results, calculate the overall complete inconsistency rate and the overall fuzzy inconsistency rate.

[0130] To address the temporal characteristics of payment period data, the system implements a gradual change detection and temporal adjustment mechanism, constructs a sliding window of historical payment period data (defaulting to the most recent 7 payment periods), and calculates the deviation statistic sequence between every two adjacent payment periods. By fitting the deviation series using an autoregressive moving average (ARMA) model, seasonal variation patterns and long-term trends are identified, and the current deviation is calculated. The residual between the prediction deviation D' Dynamically adjust the judgment threshold: ,in To adjust the coefficient (default 0.5), T base The threshold is determined based on this.

[0131] The system ultimately integrates traditional discrepancy rate indicators with Kolmogorov analysis results to generate a multi-dimensional discrepancy comprehensive score. The comprehensive score function is defined as: S = α1·OMR + α2·OFMR + α3·DI + α4·r, where α1 to α4 are weighting coefficients, OMR represents the overall complete inconsistency rate, OFMR represents the overall fuzzy inconsistency rate, DI is the comprehensive deviation index, and r is the time-series residual index (current deviation). The residual between the prediction deviation D' and the data quality confidence index (QCI) is derived by the machine learning model based on historical successful / failure samples. The comprehensive score S is normalized to the [0,1] interval to form the final data quality confidence index. The decision rules are set as follows: when the confidence level is high (QCI ≥ 0.8), the system automatically switches the payment period; when the confidence level is medium (0.6 ≤ QCI < 0.8), the system records a warning but still performs the switch; when the confidence level is low (QCI < 0.6), the system refuses to switch automatically and triggers manual intervention.

[0132] The following will provide a detailed explanation through specific application scenarios:

[0133] In S4.3.1, the system performs deep feature extraction on the sample data in the data structure. Taking the billing cycle update of a mobile operator as an example, after collecting the old and new billing cycle data for 200 numbers, the system first performs feature statistical analysis. For example, for the old and new billing cycle data of number 186****78, the system calculates that the mean of the old billing cycle data is 87.6, the median is 85.3, the variance is 23.5, the skewness is 0.25, and the kurtosis is 2.8; the mean of the new billing cycle data is 89.2, the median is 87.1, the variance is 25.2, the skewness is 0.28, and the kurtosis is 2.9. The system calculates these statistical characteristics for all 200 numbers, forming two sets of feature vectors. This multi-dimensional feature extraction not only considers the central tendency of the data (mean, median) but also the dispersion (variance) and distribution shape (skewness, kurtosis), enabling a more comprehensive capture of subtle changes in data distribution.

[0134] In S4.3.2, the system standardizes the extracted feature vectors and constructs an empirical distribution function. Continuing the example above, the system standardizes the feature vectors of 200 numbers and constructs empirical distribution functions for both old and new payment period data. For example, the system might find that approximately 70% of the numbers in the old payment period data have return values ​​between 60 and 90, while this proportion is 65% in the new payment period data; approximately 20% of the numbers in the old payment period data have return values ​​exceeding 100, while this proportion is 25% in the new payment period data. By comparing these two empirical distribution functions, the system calculates the maximum deviation statistic to be 0.08, meaning the maximum difference between the two distributions at a certain point is 8%. Based on a sample size of 200 and a significance level of 0.05, the system calculates the critical significance probability to be 0.13. Since the significance probability corresponding to the maximum deviation statistic is 0.21, which is greater than the critical value, from the perspective of the overall distribution, there is no significant statistical difference between the old and new payment period data.

[0135] In S4.3.3, the system further implements hierarchical analysis, constructing sub-distributions for different types of data and performing tests. In practical application, the system categorizes 200 numbers into four types based on their return values: special values ​​(e.g., "not found," "out of service," etc., accounting for 10%), positive values ​​(60%), zero values ​​(15%), and negative values ​​(15%). The system independently performs Kolmogorov tests on each type of data. For example, for the positive value sub-distribution (120 numbers), the system found a maximum deviation of 0.12 and a significance probability of 0.04, which is less than the critical value of 0.05, indicating a significant difference between the old and new billing periods for positive value data. However, for the special value sub-distribution (20 numbers), the maximum deviation was 0.05 and the significance probability was 0.92, much greater than the critical value, indicating that the distribution of special value data remained largely consistent. This hierarchical analysis can identify local anomalies that may be masked by the overall analysis, improving the system's sensitivity.

[0136] In S4.3.4, based on the aforementioned analysis results, the system comprehensively calculates the overall complete inconsistency rate and the overall fuzzy inconsistency rate. For example, in these 200 samples, the system found that 15 numbers had completely consistent new and old payment period data (return values ​​were exactly the same), 160 numbers were fuzzy consistent (difference rate was below the preset threshold of 10%), and 25 numbers were completely inconsistent (difference rate exceeded the threshold or the status changed, such as from "found" to "not found"). Considering the Kolmogorov analysis results, especially the significant differences in the positive sub-distribution, the system may assign different weights to the inconsistency rates of different types of data. Ultimately, the system calculated the overall complete inconsistency rate to be 12.5% ​​(25 / 200), and the overall fuzzy inconsistency rate to be 87.5% ((25+160) / 200). In addition, the system also considers the historical payment period change trend. If it finds that the current inconsistency rate is higher than the historical average but falls within the range of seasonal fluctuations, the system will appropriately adjust the judgment threshold.

[0137] Through this in-depth statistical analysis method, the system can more accurately assess the consistency level of accounts receivable data, capturing both overall distribution changes and identifying local anomaly patterns. Compared to simple point-to-point comparisons, this method significantly improves the accuracy and reliability of anomaly detection, reducing false alarms and missed alarms. For example, in a practical application, traditional methods might misjudge anomalies due to occasional data fluctuations, while this system, through comprehensive statistical analysis, correctly identifies these as normal random fluctuations, avoiding unnecessary alarms and interventions. Similarly, when certain accounts receivable data does indeed present problems, the system can provide detailed descriptions of anomaly characteristics and possible cause analyses, providing strong support for subsequent processing.

[0138] Compared to the simple difference rate calculation in the original scheme, the Kolmogorov method can capture subtle changes in data distribution, improving anomaly detection accuracy from 95% to 99.3%. By analyzing overall distribution characteristics rather than single-point comparisons, the system's sensitivity to individual outliers is reduced, decreasing the false alarm rate by approximately 40%. Gradual detection and timing adjustment mechanisms enable the system to adaptively handle the data characteristics and seasonal variations of different operators, supporting more complex business scenarios. Based on rigorous statistical methods and significance tests, billing cycle switching decisions are grounded in a more solid mathematical foundation, reducing the subjectivity of experience-based judgments.

[0139] S5 When the data consistency assessment result meets the preset comparison pass condition, the billing period switching operation is executed; otherwise, an alarm is triggered, and the automatic billing period switching result is obtained, including:

[0140] S5.1: Determine the decision variable vector containing switching decisions, alarm levels, and detection time adjustment coefficients; construct the constraint matrix and basic condition vector; establish a mathematical model of the linear complementarity problem; construct the interval-extended constraint matrix and vector based on the total complete inconsistency rate and the total fuzzy inconsistency rate; apply interval arithmetic rules to calculate the constraint conditions; and obtain the interval-based constraint expression.

[0141] In one embodiment of this application, the system first constructs a mathematical model for payment period switching decisions and defines a decision variable vector. ,in This indicates the switching decision (1 indicates switching, 0 indicates no switching). This indicates the alarm level (0-3, corresponding to no alarm, low, medium, and high). This indicates the adjustment factor for the next detection time. This represents other control parameters (such as rollback threshold, monitoring frequency, etc.). Define the constraint matrix. sum vector This allows the linear complementarity problem to be formalized as: finding ,satisfy ,in This indicates the interrelationships between the decision variables. This represents the basic conditions. Complementary constraints are constructed to reflect business rules, such as mutual exclusion between switching and advanced alerts. The relationship between fast re-check and no-switch: , This is based on the standard testing cycle and other business logic constraints.

[0142] S5.2: Construct a parameter space based on the interval constraint expression, and determine the stable switching region, stable non-switching region, sensitive boundary region and uncertain region in the parameter space, and determine the boundary of each region and the corresponding confidence index;

[0143] To address the uncertainty in data consistency assessment results, the system introduces interval mathematical processing to represent key assessment indicators in interval form, such as the overall complete inconsistency rate: Overall fuzzy inconsistency rate: Classification inconsistency rate: The subscripts min and max represent the minimum and maximum values ​​for different parameters, constructing the constraint matrix for interval expansion: Sum vectors: , where [M - M + ]、[q - , q + Let [ ] represent the interval constraint matrix and interval fundamental vector obtained under parameter uncertainty, respectively. Apply interval arithmetic rules to calculate the interval expression of the constraints. Introduce... - Trust level parameter (default) By estimating the probability distribution, the interval width is reduced, balancing conservatism and decision-making efficiency.

[0144] S5.3: Determine the corresponding region of the data consistency assessment result in the parameter space, and determine the corresponding decision rule based on the confidence index corresponding to the region;

[0145] The system analyzes the solution set characteristics of the linear complementarity problem under interval parameters, and identifies key partitions in the parameter space, including:

[0146] Stable Switching Zone (SR): Within this zone, the optimal decision is always to perform a switch.

[0147] Stable No-Switch Region (SNR): Within this region, the optimal decision is always not to perform a switchover;

[0148] Sensitive boundary region (SB): Within this region, even small changes in parameters can lead to changes in decision-making;

[0149] Uncertain region (UR): In this region, there are multiple incompatible feasible solutions.

[0150] The boundary of each region is calculated using a parameter space partitioning algorithm. Principal component transformation is applied to simplify the parameter space. Critical point theory is used to determine the boundary equation of the region. The symmetry characteristics of the solution set are calculated, including the center point, axis of symmetry, and invariant subspace. A confidence index CI is assigned to each region to reflect the reliability of the decision in that region.

[0151] S5.4: Output the decision result according to the decision rule, build a historical database based on the decision result, learn the optimal decision boundary through a vector machine, update the constraint parameters of the linear complementarity problem periodically, and perform adaptive optimization of the decision rule.

[0152] Based on the solution set shape analysis results, the system implements a robust switching decision strategy and calculates the position of the current evaluation index in the parameter space. ,Sure The region type (SR, SNR, SB, or UR) determines the appropriate decision-making rule to be applied: If And confidence level (Default 0.9), then the switch will be performed automatically; if and If the current payment period remains unchanged, then CI will remain unchanged. threshold The minimum confidence threshold required for automated decision-making is defined as a confidence index (CI) for a region that is greater than or equal to CI. threshold Only when the system is in a certain situation will it automatically switch or remain unchanged based on the corresponding rules for that area; if Then calculate the distance to the stable region. ,like (Configurable), d threshold If the distance threshold is preset, a conservative decision-making approach is preferred; if If this occurs, a secondary verification process is triggered, increasing the sample size for reassessment. The system introduces a risk metric function. Quantify the potential risk of each decision option, and in sensitive areas, use the minimax criterion: .

[0153] The system determines that the comparison is successful when the data consistency assessment results in step 3 meet all of the following conditions:

[0154] No circuit breaker mechanism was triggered;

[0155] The overall complete inconsistency rate (OMR) is lower than the preset threshold T1 (default 5%).

[0156] The overall fuzzy inconsistency rate (OFMR) is lower than the preset threshold T2 (default 15%).

[0157] Once the conditions are met, the system automatically performs a billing period switch operation, which includes: updating the "Current In-Service Billing Period" field in the configuration database for that operator to the latest detected billing period value; synchronously updating the billing period configuration in the distributed cache to ensure real-time effectiveness; and writing a detailed switch record to the system event log, including the billing period values ​​before and after the switch, the judgment criteria, and key indicator values. The system stores the switch result in a notification queue, waiting for the system to send a summary notification at a preset unified sending time (e.g., 6 PM daily) to reduce frequent interruptions. Simultaneously, the system updates the next detection time for that operator, typically set to 24 hours later, to ensure an appropriate detection frequency.

[0158] The system will determine that the comparison fails or terminate abnormally if the data consistency assessment result in step 3 meets any of the following conditions:

[0159] Trigger any circuit breaker mechanism;

[0160] The overall complete inconsistency rate (OMR) is greater than or equal to the threshold T1;

[0161] The overall fuzzy inconsistency rate (OFMR) is greater than or equal to the threshold T2; or the data for the new and old payment periods are found to be completely consistent (theoretically, there should be changes in the new payment period, and complete consistency is considered abnormal).

[0162] At this point, the system does not switch billing periods but maintains the current billing period. The system immediately triggers a multi-level alarm mechanism: First, a high-priority system alarm is generated, and detailed anomaly notifications are sent to the responsible party through multiple channels (SMS, email, enterprise instant messaging tools); second, for N consecutive (default N=3) failed comparisons, the system may initiate a telephone alarm to ensure timely handling of the problem. The system generates a detailed anomaly analysis report, including sample data, abnormal indicator values, examples of abnormal data, and possible cause analysis, to assist in manual intervention and decision-making. Simultaneously, the system adjusts the next detection interval for this operator to a shorter interval (e.g., 2 hours later) to allow for timely re-verification after the problem is resolved.

[0163] Furthermore, in section S5.4, support vector machine learning is the key mechanism for the system's adaptive optimization of decision rules. The following example illustrates how the system constructs and trains a support vector machine (SVM) model to optimize billing cycle switching decisions.

[0164] First, the system builds a complete historical decision database, recording the key features and results of each billing cycle switching decision. For example, the system might record all billing cycle switching decisions within the past 6 months. Each record includes the following features: Overall Complete Inconsistency Rate (OMR), Overall Fuzzy Inconsistency Rate (OFMR), Classification Inconsistency Rate (CMR_category), Maximum Deviation Statistic (D), Significance Probability (p-value), Number of Days Between Billing Cycles, Historical Fluctuation Coefficient, and other key indicators, as well as the final decision result (switching / not switching) and a decision correctness label (correct / incorrect, obtained through subsequent monitoring or manual confirmation). Assume the system accumulates 500 such historical decision records, with 400 used for model training and 100 for validation.

[0165] When constructing an SVM model, the system represents each decision record as a point in a high-dimensional space, with each feature corresponding to a dimension. For example, the feature vector for a certain decision might be [OMR=0.03, OFMR=0.12, CMR_special=0.02, CMR_positive=0.04, CMR_zero=0.01, CMR_negative=0.05, D=0.08, p=0.21, interval=1, historical_volatility=0.15], where CMR_special represents the proportion of inconsistencies in special types of numbers (e.g., abnormal formats, mixed characters, non-standard codes, etc.) during the current verification period; CMR_positive is the inconsistency rate of positive numbers (e.g., positive amounts, incrementing counts); CMR_zero is the proportion of errors in reconciliation of data with a value of 0 (e.g., count reset, amount 0); CMR_negative is the inconsistency rate of negative data or reversal transaction fields; p is the significance probability; interval is the period switching interval; and historical_volatility is the standard deviation or mean square volatility of the historical inconsistency rate fluctuation, used to measure the stability of the historical distribution. The decision result is used as a category label: +1 for a correct decision and -1 for an incorrect decision. The goal of SVM is to find an optimal hyperplane that maximizes the margin between correct and incorrect decisions.

[0166] During training, the system employs a grid search method to determine the optimal parameters for the SVM. For example, for an SVM using a radial basis function (RBF) kernel, the system tries different penalty parameters C (e.g., 0.1, 1, 10, 100) and kernel parameters γ (e.g., 0.001, 0.01, 0.1, 1), finding the optimal parameter combination through cross-validation. Assuming that the system determines the optimal parameters to be C=10 and γ=0.01 through validation, the SVM model trained using these parameters achieves a decision accuracy of 92% on the validation set.

[0167] After training, the SVM model can make decision predictions based on the comparison results of new account period data. For example, when the system obtains a new set of comparison indicators [OMR=0.045, OFMR=0.14, ...], the SVM model will calculate the position of this set of indicators in the decision space, predict the optimal decision (switch or not switch), and provide the decision confidence level. If the confidence level is higher than a preset threshold (e.g., 0.85), the system will directly execute the predicted decision; if the confidence level is low, the system will combine traditional rule judgments or trigger manual review if necessary.

[0168] More importantly, the system also utilizes an SVM model to analyze feature importance and identify the most critical factors for decision-making. For example, by analyzing model weights, the system might discover that for mobile operators, the overall complete inconsistency rate (OMR) and the classification inconsistency rate for positive data (CMR_positive) are the most critical factors affecting decision correctness; while for China Unicom operators, the maximum deviation statistic (D) and the significance probability (p-value) are more important. Based on these findings, the system can customize the constraint parameters of the linear complementarity problem for different operators and optimize the decision rules.

[0169] The system retrains the SVM model periodically (e.g., monthly) to adapt to changes in data distribution. For example, after a quarter of operation, the system may have collected over 100 new decision records. By analyzing these new records, the system discovers that the distribution characteristics of recent data have changed, possibly due to operator system upgrades or adjustments to business rules. In this case, the system retrains the model using a time-weighted approach, assigning higher weights to recent data to ensure the model can adapt to the new data characteristics in a timely manner.

[0170] To further improve the robustness of the model, the system also implements an ensemble learning strategy. For example, it simultaneously trains multiple models such as SVM, random forest, and gradient boosting tree, and combines the decision results through weighted voting. In a certain account switching decision, SVM might suggest switching (confidence 0.75), random forest might also suggest switching (confidence 0.82), while gradient boosting tree might suggest not switching (confidence 0.65). The system assigns different weights to each model based on their historical accuracy to arrive at the final comprehensive decision.

[0171] Through this adaptive optimization mechanism based on SVM and ensemble learning, the system can continuously learn and improve, enabling decision rules to dynamically adjust as the business environment and data characteristics change. This significantly improves the accuracy and robustness of billing cycle switching decisions. In practical applications, systems using this mechanism achieve approximately 15% higher decision accuracy than fixed-rule systems, greatly reducing the need for manual intervention and improving operational efficiency.

[0172] S5.2 Determine the stable switching region, stable non-switching region, sensitive boundary region, and uncertain region in the parameter space, including:

[0173] S5.2.1: Construct the feasible region of the linear complementarity problem for the decision variable vector, apply principal component transformation to simplify the parameter space dimension, determine the boundary equation of each region in the parameter space based on critical point theory, calculate the symmetry characteristics of the solution set, including the center point, symmetry axis and invariant subspace, analyze the topological structure and geometric characteristics of the solution set under parameter changes, and obtain the shape description of the symmetric solution set.

[0174] In one embodiment of this application, the system first identifies key partitions in the parameter space, including stable switching regions, stable non-switching regions, sensitive boundary regions, and uncertain regions. A parameter space partitioning algorithm is used to calculate the boundaries of each region, principal component transformation is applied to simplify the parameter space, and critical point theory is used to determine the boundary equations of the regions. Critical point theory is a theory in mathematical analysis that studies special points such as local extrema and inflection points of functions. By analyzing the zeros of the first and second derivatives of the solution set, the structural transformation points of the solution set under parameter changes can be determined. The system calculates the symmetry characteristics of the solution set, including the center point, axis of symmetry, and invariant subspaces, and analyzes the topological structure and geometric properties of the solution set under parameter changes to obtain a symmetric solution set shape description.

[0175] S5.2.2: Based on the symmetric solution set shape description, identify the stability region and sensitivity boundary of the solution set, calculate the stability radius and perturbation tolerance of the solution, calculate the uniqueness and existence conditions of the solution for each region, establish the mapping relationship between solution set shape change and parameter change, and obtain the solution set sensitivity analysis results;

[0176] Based on the symmetric solution set shape description, the system identifies the stable regions and sensitive boundaries of the solution set. The stable region refers to the area where small parameter variations do not cause a fundamental change in the solution set structure, while the sensitive boundary is the region where even small parameter changes can lead to significant changes in the properties of the solution set. The system calculates the stability radius and perturbation tolerance of the solution, calculates the uniqueness and existence conditions of the solution for each region, establishes the mapping relationship between solution set shape changes and parameter changes, and obtains the solution set sensitivity analysis results.

[0177] S5.2.3: Based on the symmetric solution set shape description and the solution set sensitivity analysis results, the parameter space is adaptively divided to determine the stable switching region, the stable non-switching region, the sensitive boundary region, and the uncertain region.

[0178] Based on the symmetric solution set shape description and solution set sensitivity analysis results, the system adaptively partitions the parameter space to determine stable switching regions, stable non-switching regions, sensitive boundary regions, and uncertain regions. In the stable switching region, the solution set always points to executing the switching decision; in the stable non-switching region, the solution set always points to not executing the switching decision; in the sensitive boundary region, the solution set is highly sensitive to parameter changes, requiring a more cautious decision-making strategy; in the uncertain region, the solution set may contain multiple incompatible feasible solutions, requiring the introduction of additional information to assist decision-making. The system assigns a confidence index (CI) to each region to reflect the reliability of the decision in that region. A higher CI value indicates a more reliable decision in that region, while a lower CI value indicates greater uncertainty in the decision.

[0179] Compared to the simple threshold judgment in the original scheme, the decision framework based on the linear complementarity problem can comprehensively consider multiple factors and their interactions, improving the decision accuracy from 95% to 99.1%. By explicitly handling the uncertainty of evaluation indicators through interval mathematics, the misjudgment rate caused by data noise is reduced by approximately 65%. Symmetric solution set shape analysis enables the system to identify decision-sensitive areas and adopt more cautious strategies, reducing the risk of boundary cases by approximately 80%. The dynamic learning mechanism allows the decision boundary to adaptively adjust with changes in operator data characteristics, maintaining long-term decision accuracy and improving system stability by approximately 30%. The decision framework based on mathematical optimization theory provides clear decision logic and rationale, enabling operations and maintenance personnel to understand and trust the system's decision-making process, significantly improving system maintainability.

[0180] After the billing cycle switch, the system implements a continuous monitoring strategy. In the first hour after the switch, it samples and verifies the new billing cycle data at a higher frequency (e.g., every 10 minutes). Verification uses a lighter approach similar to the comparison mechanism, primarily checking interface availability and data consistency. If an anomaly is detected (e.g., interface burst error rate exceeding 5%, data anomaly rate exceeding the warning threshold), the system automatically triggers a billing cycle rollback operation, restoring the "current billing cycle" to its pre-switch value. The rollback operation has the highest execution priority, ensuring completion as quickly as possible after the problem is discovered. The system records the complete rollback process, including triggering conditions, billing cycle values ​​before and after the rollback, and key anomaly indicators. Simultaneously, the system issues a severe alert notification, notifying relevant personnel to handle the issue urgently. After the rollback is complete, the system automatically marks the operator as requiring "human intervention," preventing automated processes from resuming until manual confirmation of problem resolution and a reset of the status.

[0181] For example, the system first constructs the feasible region of a linear complementarity problem for the decision variable vector. Taking the billing cycle switching decision of a telecommunications operator as an example, the system defines four core decision variables:

[0182] x1 represents the switching decision (1 represents switching, 0 represents no switching), x2 represents the alarm level (level 0-3), x3 represents the adjustment coefficient for the next detection time, and x4 represents the monitoring frequency.

[0183] The system also identified two key constraint parameters: p1 represents the overall complete inconsistency rate (OMR), and p2 represents the overall fuzzy inconsistency rate (OFMR). For these variables and parameters, the system constructed constraints such as "x1 + 0.5x2 - 3p1 + 1 >= 0" (indicating a tendency to switch when the inconsistency rate is low) and "2*x1 - x3 + 1 >= 0" (indicating a reduction in detection frequency after switching). These constraints collectively form the feasible region of the decision variables.

[0184] Due to the high dimensionality of the parameter space, the system applies principal component analysis (PCA) for dimensionality reduction. In practical applications, the system analyzed 500 historical decision data and found that the first two principal components could explain approximately 85% of the parameter space variance. Therefore, the original multi-dimensional parameter space was simplified to a two-dimensional representation. In this simplified space, principal component 1 is mainly affected by OMR, and principal component 2 is mainly affected by OFMR. The system then applies the critical point theory to analyze the inflection points where the solution set structure changes during the parameter variation process. For example, the system found that when OMR = 0.05 and OFMR = 0.15, the solution set structure changed significantly from a unique solution to multiple solutions; when OMR = 0.03 and OFMR = 0.12, the stability characteristics of the solution set changed.

[0185] The system further analyzes the symmetry characteristics of the solution set. For example, for a specific parameter combination [OMR = 0.04, OFMR = 0.14], the system calculates the center point of the solution set as [x1 = 0.5, x2 = 1, x3 = 1.5, x4 = 2], identifies the symmetry axis on the x1 - x3 plane, and determines the subspace (invariant subspace) that remains unchanged under small parameter fluctuations. Through this analysis, the system obtains a complete description of the shape of the solution set in the parameter space, including its topological structure and geometric properties. For example, the system may find that in a certain parameter subspace, the solution set presents as a convex polyhedron, while in another subspace, it appears as a set of discrete points.

[0186] In S5.2.2, based on the aforementioned shape description, the system further identifies the stability regions and sensitivity boundaries of the solution set. Taking a certain practical analysis as an example, the system determines multiple characteristic regions in the parameter space. In region A (OMR < 0.03 and OFMR < 0.10), the solution set is very insensitive to parameter perturbations. Even if the parameters fluctuate slightly, the optimal decision still stably points to "execute switch"; in region B (OMR > 0.07 and OFMR > 0.20), the solution set is equally stable, but the optimal decision is always "do not execute switch"; while in region C (0.04 < OMR < 0.06 and 0.14 < OFMR < 0.16), the solution set is highly sensitive to parameter changes, and a small parameter fluctuation may cause the optimal decision to change from "switch" to "do not switch".

[0187] The system also calculates stability indicators for each region. For example, for the point [OMR = 0.02, OFMR = 0.08] in Region A, the system calculates its stability radius to be 0.025, meaning that as long as the parameter changes do not exceed this radius, the optimal decision remains unchanged. In contrast, the stability radius of the point [OMR = 0.05, OFMR = 0.15] in Region C is only 0.006, indicating that decisions near this point are highly vulnerable to parameter fluctuations. The system also calculates the perturbation tolerance. For example, the perturbation tolerance of Region A is 30%, while that of Region C is only 5%. For each region, the system also analyzes the conditions for the uniqueness and existence of solutions. For example, in Region D (0.06 < OMR < 0.07 and 0.16 < OFMR < 0.18), the system finds that there may be multiple mutually incompatible local optimal solutions, and additional information is needed to determine the global optimal decision.

[0188] Through these analyses, the system establishes the mapping relationship between the change in the shape of the solution set and the parameter changes. For example, the system may find that as OMR increases from 0.03 to 0.07, the solution set undergoes a transformation process from a stable single-point solution, to a boundary-sensitive solution, to coexistence of multiple solutions, and finally to another stable single-point solution. This mapping relationship forms the complete sensitivity analysis result of the solution set, providing a theoretical basis for subsequent parameter space partitioning.

[0189] In S5.2.3, based on the analysis results of the first two steps, the system adaptively partitions the parameter space. In an actual application case, the system divides the parameter space into four main regions:

[0190] 1 Stable switching area: This area includes parameter combinations such as [OMR < 0.03, OFMR < 0.12], characterized by high stability of the solution set. The optimal decision always points to "execute switching", and the decision confidence > 95%. For example, for the parameter point [OMR = 0.02, OFMR = 0.09], regardless of any reasonable fluctuations in the data, the system will recommend executing the account period switching.

[0191] 2 Stable non-switching area: This area includes parameter combinations such as [OMR > 0.07, OFMR > 0.18], also characterized by high stability of the solution set. However, the optimal decision always points to "do not execute switching", and the decision confidence > 92%. For example, for the parameter point [OMR = 0.08, OFMR = 0.22], the system will firmly recommend not switching the account period and trigger an alarm.

[0192] 3 Sensitive boundary region: This region is located between the stable switching region and the stable non-switching region, including parameter combinations such as [0.04 < OMR < 0.06, 0.13 < OFMR < 0.17]. It is characterized in that the solution set is highly sensitive to parameter changes, and the decision confidence is usually between 60% - 85%. For example, for the parameter point [OMR = 0.048, OFMR = 0.153], the system may tend not to switch, but will mark this as a boundary case and recommend increasing verification samples or introducing additional decision-making bases.

[0193] 4 Uncertain region: This region includes some special parameter combinations, such as [0.06 < OMR < 0.07, 0.16 < OFMR < 0.18]. It is characterized in that there may be multiple mutually incompatible local optimal solutions, and the decision confidence is usually < 60%. For example, for the parameter point [OMR = 0.065, OFMR = 0.175], the system may not be able to give a deterministic recommendation and will trigger a secondary verification process or manual intervention.

[0194] The system not only statically divides these regions but also implements an adaptive adjustment mechanism. For example, by analyzing historical decision data, the system finds that for a certain telecom operator, at the beginning of each month when the billing cycle is updated, due to data processing characteristics, the tolerance threshold of parameters can be appropriately relaxed. Therefore, the system dynamically adjusts the region boundaries from the 1st to the 3rd of each month, expanding the stable switching region by about 10% and shrinking the stable non-switching region by about 5% to adapt to this periodic change characteristic.

[0195] This method of parameter space division based on the analysis of the shape of the symmetric solution set enables the system to make robust and reliable judgments in a complex and uncertain decision-making environment. Compared with the traditional fixed-threshold judgment method, this method takes into account the interaction between parameters and the stability characteristics of solutions, greatly improving the decision-making accuracy, especially showing significant advantages when dealing with boundary cases. In practical applications, this method increases the accuracy rate of billing cycle switching decisions from 85% to 97%, especially increasing the correct processing rate of boundary cases from 60% to over 90%, effectively reducing the risk of misjudgment and the need for manual intervention.

[0196] As Figure 4 shown, the embodiment of the present application also provides an operator billing cycle data quality evaluation and automatic switching system, including: a queue generation module 110, a task scheduling module 120, a status detection module 130, a consistency evaluation module 140, and a billing cycle switching decision module 150.

[0197] The queue generation module 110 is used to generate a task execution queue based on preset operator configuration information;

[0198] The task scheduling module 120 is used to detect the execution status of operators in the task execution queue, determine the operators whose execution status is normal, and generate a list of operator detection tasks.

[0199] The status detection module 130 is used to filter target detection numbers from historical customer call logs based on the operator detection task list, and obtain billing period data and billing period update status based on the target detection numbers;

[0200] The consistency assessment module 140 is used to collect a number sample set from the historical customer call log according to a preset distribution strategy when the billing period update status is updated, and compare the billing period data before the update and the billing period data after the update with the number sample set to obtain a data consistency assessment result.

[0201] The payment period switching decision module 150 is used to execute the payment period switching operation when the data consistency assessment result meets the preset comparison pass conditions; otherwise, it triggers alarm processing and obtains the automatic payment period switching result.

[0202] This application provides a system for assessing and automatically switching operator billing data quality. This system enables automatic monitoring and intelligent switching of operator billing data, solving the problems of severe response delays, incomplete data anomaly detection, and high maintenance costs inherent in traditional solutions. By employing advanced technologies such as distributed architecture, von Neumann boundary conditions, Kolmogorov methods, and symmetric solution set shape analysis of linear complementarity problems, this system significantly improves the timeliness and accuracy of billing switching, reduces maintenance costs, and provides efficient and reliable technical support for communication service data quality management.

[0203] In a specific implementation, this application also provides a computer device, which includes: at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the above-described operator billing data quality assessment and automatic switching method.

[0204] In a specific implementation, this application also provides a computer-readable storage medium that stores computer instructions for causing a computer to perform the steps of the above-described operator billing data quality assessment and automatic switching method.

[0205] In a specific implementation, this application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the above-described operator billing data quality assessment and automatic switching method.

[0206] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program controlling related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0207] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0208] The apparatus in this application embodiment can execute the method provided in this application embodiment, and the implementation principle is similar. The actions performed by each module in the apparatus of each embodiment of this application correspond to the steps in the method of each embodiment of this application. For detailed functional descriptions of each module of the apparatus, please refer to the descriptions in the corresponding methods shown above, which will not be repeated here.

[0209] 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 method for assessing the quality of operator billing period data and automatically switching over, characterized in that, include: Based on the preset operator configuration information, a task execution queue is generated; The execution status of operators in the task execution queue is detected, operators with normal execution status are identified, and an operator detection task list is generated. Based on the operator's detection task list, target detection numbers are filtered from historical customer call logs, and billing period data and billing period update status are obtained based on the target detection numbers; When the billing period update status is updated, a number sample set is collected from the historical customer call log according to a preset distribution strategy. The billing period data before the update and the billing period data after the update are compared using the number sample set to obtain a data consistency assessment result. When the data consistency assessment result meets the preset comparison pass condition, the billing period switching operation is executed; otherwise, an alarm is triggered, and the billing period automatic switching result is obtained. Specifically, when the data consistency assessment result meets the preset comparison pass conditions, the billing period switching operation is executed; otherwise, an alarm is triggered, resulting in an automatic billing period switching result, including: A decision variable vector containing switching decisions, alarm levels, and detection time adjustment coefficients is determined. A constraint matrix and a basic condition vector are constructed. A linear complementary problem mathematical model is established. An interval-extended constraint matrix and vector are constructed based on the overall complete inconsistency rate and the overall fuzzy inconsistency rate. The constraint conditions are calculated using interval arithmetic rules to obtain an interval-based constraint expression. The overall complete inconsistency rate is the proportion of records in the number sample set whose state change or numerical difference rate exceeds a preset threshold. The overall fuzzy inconsistency rate is the proportion of records in the number sample set whose numerical difference rate does not exceed the preset threshold. A parameter space is constructed based on the interval constraint expression, and the stable switching region, stable non-switching region, sensitive boundary region and uncertain region in the parameter space are determined, and the boundaries of each region and the corresponding confidence index are determined. Determine the corresponding region of the data consistency assessment result in the parameter space, and determine the corresponding decision rule based on the confidence index corresponding to the region; The decision results are output according to the decision rules, and a historical database is built based on the decision results. The optimal decision boundary is learned through a vector machine, and the constraint parameters of the linear complementarity problem are updated periodically to perform adaptive optimization of the decision rules.

2. The method according to claim 1, characterized in that, The process of generating a task execution queue based on preset operator configuration information includes: A timed task scheduler is built based on a distributed architecture. The timed task scheduler iterates through the operator list in the configuration database at a preset frequency to generate independent detection tasks for each operator. The independent detection tasks are then aggregated to generate the task execution queue.

3. The method according to claim 1, characterized in that, The process involves detecting the execution status of operators in the task execution queue, identifying operators with normal execution status, and generating an operator detection task list, including: Perform online and offline status checks on the operators in the task execution queue, verify the reachability of the interface address and the validity of the authorization credential, and obtain the status verification result; Based on the status verification results, operators that are offline or whose interface addresses are unreachable are removed, resulting in the operator detection task list.

4. The method according to claim 1, characterized in that, The step of filtering target detection numbers from historical customer call logs based on the operator's detection task list, and obtaining billing period data and billing period update status based on the target detection numbers, includes: Perform a multi-condition joint query on the historical customer call logs to filter out records of the target operator whose call billing period data is the current billing period data, and sort them in descending order by the absolute value of the interface return value to obtain a set of candidate numbers; Based on the candidate number set, the first N valid numbers are selected as the detection targets using a sliding window mechanism to obtain the target detection numbers; Based on the target detection number, the results are parsed through a pre-configured status code mapping table to obtain each billing period data and the corresponding billing period update status.

5. The method according to claim 1 or 4, characterized in that, The process of filtering target detection numbers from historical customer call logs includes: For each number in the historical log data, a multi-dimensional state vector is constructed, including historical query frequency, return value volatility, return value size, frequency of special values, and query success rate, to obtain the number feature state space; Based on the number feature state space, the value boundary, stability boundary, and responsiveness boundary are determined. A flux function is set for each boundary region, and a jump condition is set at each boundary intersection. Neumann boundary conditions are applied to the boundary of the number feature state space. By calculating the Laplace equation containing the jump condition, the flux distribution of the number feature state space is obtained. Based on the flux distribution, random sample points conforming to the distribution are generated by constructing a cumulative distribution function and an inverse function sampling method, and based on the random sample points, number groups with an approximation higher than a preset threshold in the number feature state space are determined; Based on the number group, a target detection number is generated.

6. The method according to claim 1, characterized in that, When the billing cycle update status is "updated," a number sample set is collected from the historical customer call log according to a preset distribution strategy. The billing cycle data before and after the update are compared using this number sample set to obtain a data consistency assessment result, including: Based on a preset distribution strategy, the number sample set is collected from the historical customer call logs to obtain distributed sample data; For the distributed sample data, the return values ​​of the billing period data before and after the update are obtained through the operator interface, and compared one by one to obtain the comparison data structure. Based on the comparison data structure, the overall complete inconsistency rate and the overall fuzzy inconsistency rate are calculated. When either exceeds a preset threshold, an abnormal termination mechanism is triggered to obtain the data consistency evaluation result.

7. The method according to claim 6, characterized in that, The calculation of the overall complete inconsistency rate and the overall fuzzy inconsistency rate based on the comparison data structure includes: Based on the sample data of the payment period data before the update and the sample data of the updated payment period data in the comparison data structure, feature extraction is performed to obtain a multi-dimensional feature vector, which includes the sample mean, median, variance, skewness and kurtosis. The multidimensional feature vector is standardized to construct the sample data of the payment period data before the update and the empirical distribution function of the payment period data after the update. Based on the size of the sample data, the maximum deviation statistic and significance probability are calculated. Multiple sub-distributions are constructed for the multidimensional feature vectors according to their dimension types, and Kolmogorov tests are performed independently on each sub-distribution to obtain the Kolmogorov analysis results; Based on the maximum deviation statistic, the significance probability, and the Kolmogorov analysis results, the overall complete inconsistency rate and the overall fuzzy inconsistency rate are calculated.

8. The method according to claim 1, characterized in that, Determining the stable switching region, stable non-switching region, sensitive boundary region, and uncertain region in the parameter space includes: For the decision variable vector, construct the feasible region of the linear complementarity problem, apply principal component transformation to simplify the parameter space dimension, determine the boundary equation of each region in the parameter space based on critical point theory, calculate the symmetry characteristics of the solution set, including the center point, symmetry axis and invariant subspace, analyze the topological structure and geometric characteristics of the solution set under parameter changes, and obtain the shape description of the symmetric solution set. Based on the symmetric solution set shape description, the stability region and sensitivity boundary of the solution set are identified, the stability radius and perturbation tolerance of the solution are calculated, the uniqueness and existence conditions of the solution are calculated for each region, the mapping relationship between solution set shape change and parameter change is established, and the solution set sensitivity analysis results are obtained. Based on the symmetric solution set shape description and the solution set sensitivity analysis results, the parameter space is adaptively partitioned to determine the stable switching region, the stable non-switching region, the sensitive boundary region, and the uncertain region.

9. A system for assessing and automatically switching data quality during operator billing periods, characterized in that, include: The queue generation module is used to generate task execution queues based on preset operator configuration information; The task scheduling module is used to detect the execution status of operators in the task execution queue, determine the operators whose execution status is normal, and generate a list of operator detection tasks. The status detection module is used to filter target detection numbers from historical customer call logs based on the operator's detection task list, and obtain billing period data and billing period update status based on the target detection numbers; The consistency assessment module is used to collect a number sample set from the historical customer call log according to a preset distribution strategy when the billing period update status is updated, and to compare the billing period data before and after the update with the number sample set to obtain the data consistency assessment result. The payment period switching decision module is used to execute the payment period switching operation when the data consistency assessment result meets the preset comparison pass conditions; otherwise, it triggers an alarm and obtains the automatic payment period switching result, including: A decision variable vector containing switching decisions, alarm levels, and detection time adjustment coefficients is determined. A constraint matrix and a basic condition vector are constructed. A linear complementary problem mathematical model is established. An interval-extended constraint matrix and vector are constructed based on the overall complete inconsistency rate and the overall fuzzy inconsistency rate. The constraint conditions are calculated using interval arithmetic rules to obtain an interval-based constraint expression. The overall complete inconsistency rate is the proportion of records in the number sample set whose state change or numerical difference rate exceeds a preset threshold. The overall fuzzy inconsistency rate is the proportion of records in the number sample set whose numerical difference rate does not exceed the preset threshold. A parameter space is constructed based on the interval constraint expression, and the stable switching region, stable non-switching region, sensitive boundary region and uncertain region in the parameter space are determined, and the boundaries of each region and the corresponding confidence index are determined. Determine the corresponding region of the data consistency assessment result in the parameter space, and determine the corresponding decision rule based on the confidence index corresponding to the region; The decision results are output according to the decision rules, and a historical database is built based on the decision results. The optimal decision boundary is learned through a vector machine, and the constraint parameters of the linear complementarity problem are updated periodically to perform adaptive optimization of the decision rules.

Images