A Telecom Invoicing Content Matching System Based on Multi-Dimensional Business Proportion Analysis

Abstract

This application relates to the field of telecommunications billing and financial and tax data processing, and discloses a telecommunications invoicing content matching system and method based on multi-dimensional business proportion analysis. The system is located between the underlying billing gateway and the external tax invoicing subsystem. The system extracts business consumption to generate a standard usage matrix; evaluates the overall activity scalar to determine the performance status, and extracts historical data to generate a baseline entropy weight vector when the performance is dormant; calculates the temporal information entropy and combines it with the performance status to generate a final dynamic entropy weight vector; performs tensor scalar dot product projection along the direction of the final dynamic entropy weight vector to split the total invoicing amount scalar into various tax item nodes; performs tax red line interval boundary checks on the split amounts, and when a boundary violation occurs, calls multi-dimensional tax base Lagrange constraints to perform quadratic programming optimization calculations to correct the amount data set; finally, it encapsulates the data into an invoicing instruction to generate an electronic invoice. This invention solves the problem of dormant option distortion and achieves accurate splitting of amounts and automated tax compliance correction.

Description

Technical Field

[0001] This invention relates to the field of telecommunications billing and financial and tax data processing technology, specifically a telecommunications invoicing content matching system based on multi-dimensional business proportion analysis. Background Technology

[0002] With the evolution of telecommunications services, converged packages typically include heterogeneous network resources such as data traffic, voice calls, and various value-added services. The consumption of these services has completely different physical dimensions, making it difficult to directly use the underlying call detail record (CDR) data for unified weight calculation. At the same time, users exhibit significant temporal fluctuations when using various services. In traditional billing and invoicing systems, the calculation of invoicing weights relies excessively on transient data of the current billing cycle. When a user's activity level is extremely low or they are in a dormant state during a billing cycle, the weight calculated directly based on the minimum consumption value of the current period will experience extreme changes or a precipitous drop, resulting in a serious distortion of service weight allocation and failing to objectively reflect the user's long-term historical service structure.

[0003] In the process of completing the underlying business calculation and converting it into tax invoices, it involves the spatial mapping of business attributes to tax attributes. Most of the existing splitting logic adopts static and coarse calculation, which fails to build an accurate business-tax mapping topology. There is a significant boundary calculation loophole in this conversion process: when a specific business has zero consumption for a continuous period of time during the billing cycle, the traditional algorithm will force the invoicing ratio of the corresponding tax item to be directly reduced to zero. This processing logic ignores the resource pre-occupancy and maintenance costs that the underlying communication network continuously generates to maintain the availability of the business, resulting in a disconnect between the business-tax amount conversion and the actual resource consumption.

[0004] Furthermore, tax regulations impose strict compliance requirements on invoice issuance. The distribution of invoice amounts across different tax categories must be controlled within specific tax red line ranges. When faced with abnormal split amounts exceeding the limits, the existing telecommunications invoicing subsystem can typically only execute simple process blocking and error reporting. Subsequent processes heavily rely on manual intervention for verification and manual accounting adjustments. The system lacks a closed-loop automated correction and recalculation mechanism, making it unable to automatically calculate the optimal weight solution that satisfies multidimensional constraints under the strict constraints of tax compliance red lines and maintaining the conservation of total financial invoice amounts. This results in low processing efficiency for abnormal invoice orders and significantly increases the operational costs of business, finance, and tax collaboration. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a telecommunications invoicing content matching system based on multi-dimensional business proportion analysis, which solves the problems of heterogeneous business volume differences and distortion caused by dormant option redistribution.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a telecommunications invoicing content matching system based on multi-dimensional business proportion analysis, comprising: The heterogeneous service call detail record (CDR) collection and time series matrix normalization processing module is used to extract service consumption from the billing gateway and generate a standard usage matrix through range normalization. The compliance obligation dormancy detection and inter-period time window status assessment module is used to evaluate the overall activity scalar of the standard usage matrix to determine the compliance status, and outputs the historical baseline entropy weight vector when the compliance dormancy status is determined. The temporal information entropy calculation and baseline calibration evolution module is used to calculate the temporal information entropy of each business dimension and generate the final dynamic entropy weight vector in combination with the performance status. The amount tensor projection module is used to obtain the total invoice amount scalar, and perform dot product projection on the total invoice amount scalar along the direction of the final dynamic entropy weight vector to split it into each tax item node, generating a data set containing the final split invoice amount of each tax item node. The tax base red line verification and compliance blocking module is used to perform out-of-bounds verification on the final split invoice amount and output compliance release data or out-of-bounds blocking signal; The multidimensional tax base Lagrange recalculation module is used to perform quadratic programming optimization calculations when an out-of-bounds blocking signal is received, and to generate a corrected set of split invoice amount data. The business, finance and tax instruction encapsulation and electronic invoice generation module is used to encapsulate the compliant release data or the modified split invoice amount data set into a standard business, finance and tax invoicing instruction with a digital signature, and send it to the tax invoicing subsystem to generate an electronic invoice.

[0007] Preferably, the heterogeneous service call detail record (CDR) collection and time series matrix normalization processing module performs range normalization to generate a standard usage matrix, specifically including: The total time span of the target billing cycle is divided into multiple time slices at equal intervals in the time domain, and multiple business dimensions are extracted based on the configuration data of the telecom converged package ordered by the user. The original consumption values ​​of each item under the same user identifier are accumulated and aggregated to the corresponding spatiotemporal cross node according to timestamp and business type to construct the original usage matrix; By using the range transformation method, and combining the extreme values ​​of consumption of each service within the billing cycle, the matrix elements in the original usage matrix are transformed into dimensionless standardized values ​​to generate the standard usage matrix.

[0008] Preferably, the compliance obligation dormancy detection and inter-period time window status assessment module determines the compliance status and outputs the historical baseline entropy weight vector, specifically including: The total activity scalar is obtained by summing all elements in the standard usage matrix, and the dynamic dormancy threshold is generated by multiplying the preset basic activity tolerance coefficient with the number of multiple time slices and the number of multiple business dimensions. When the overall activity scalar is greater than or equal to the dynamic dormancy threshold, the performance status is determined to be a normal performance active period; When the overall activity scalar is less than the dynamic dormancy threshold, the performance status is determined to be a performance dormancy state. In the performance dormancy state, a cross-period sliding time window is constructed by backtracking along the time series towards the historical record, extracting the historical dynamic entropy weight values ​​corresponding to the effective historical billing period, and calculating the average value to generate the historical baseline entropy weight vector.

[0009] Preferably, the temporal information entropy calculation and baseline calibration evolution module generates the final dynamic entropy weight vector, specifically including: Calculate the feature weight of each business dimension under each time slice, and calculate the temporal information entropy in combination with the limit convergence rule to generate the original transient entropy weight; When the performance status is in the normal performance active period, the weight direct execution logic is triggered, and the original transient entropy weight is directly assigned to the final dynamic entropy weight vector. When the performance status is in the performance dormant state, the exponential smoothing baseline calibration logic is triggered. Based on the overall activity scalar and the dynamic dormant threshold, a smoothing factor is calculated. The smoothing factor is then used to perform exponential smoothing fitting calculation on the original transient entropy weight and the historical baseline entropy weight vector to generate the final dynamic entropy weight vector.

[0010] Preferably, before performing dot product projection, the monetary tensor projection module is further configured to activate the orthogonal penalty function with zero cost dimension, specifically including: The number of time slices in which the consumption of each business dimension is continuously zero within the target billing period is counted. When the number is greater than or equal to the zero consumption tolerance threshold preset by the system, the corresponding business dimension is determined to have triggered the zero consumption condition. Read the preset basic maintenance coefficients and construct the orthogonal minimum guarantee basis vector corresponding to the business dimension in the business weight space; The orthogonal minimum guaranteed basis vector is superimposed onto the final dynamic entropy weight vector to generate the penalized corrected business weights. The superimposed corrected full weight vector is then normalized to generate a projection reference direction vector containing orthogonal penalty correction, which replaces the original final dynamic entropy weight vector.

[0011] Preferably, the amount tensor projection module performs dot product projection to generate a data set containing the final split invoice amount for each tax item node, specifically including: Establish the mapping relationship between each business dimension and each tax item node to construct the business and tax topology mapping matrix, extract the final dynamic entropy weight vector and perform tensor dot product operation with the business and tax topology mapping matrix to generate tax item weight vector; The total invoice amount scalar is multiplied by each projection transformation weight in the tax item weight vector to obtain the initial split invoice amount for all tax item nodes. The initial split invoicing amount of each tax item node except the last tax item node is truncated to two decimal places according to the rules. The total invoicing amount is then subtracted from the truncated amount, and the remaining difference is taken as the final split invoicing amount of the last tax item node. The tail difference is then balanced.

[0012] Preferably, the tax base red line verification and compliance blocking module performs red line overrun verification, specifically including: For each tax item node, the corresponding historical benchmark invoicing ratio and preset fluctuation tolerance are read; a boundary extreme value function is introduced, and numerical calculations are performed in combination with the total invoicing amount scalar to construct a tax red line range that includes a minimum amount threshold and a maximum amount threshold, so as to prevent the threshold from becoming negative or exceeding the total invoicing amount scalar. The final split invoice amount of the corresponding tax item node is introduced into the tax red line range for numerical comparison. If none of them exceed the limit, the compliant release data is output. If it is determined that the final split invoice amount is less than the minimum amount threshold or greater than the maximum amount threshold, then the verification is confirmed to have failed, and the corresponding over-limit blocking signal and the amount split details are generated and written into the abnormal review queue.

[0013] Preferably, the multidimensional tax base Lagrange recalculation module performs quadratic programming optimization calculations, specifically including: Extract the total invoice amount scalar that triggers the boundary blocking signal, obtain the initial projection transformation weights corresponding to each tax item node, and the minimum amount threshold and the maximum amount threshold; A quadratic objective function is constructed with the goal of minimizing the weight adjustment range, and a set of constraints is constructed by combining the equality constraint of the total weight and the inequality constraint of the tax red line interval. By introducing Lagrange multiplier vectors to transform into Lagrange functions, and based on the Caro-Kun-Tucker conditions, calling the quadratic programming algorithm to perform a finite-step iterative calculation, the optimal set of target weight values ​​is obtained. Perform a scalar multiplication operation between the total invoice amount scalar and the optimal target weight value set, and then perform financial precision truncation and tail difference compensation operations again to generate a corrected split invoice amount data set.

[0014] Preferably, the encapsulation of business and tax instructions and the encapsulation of standard business and tax invoicing instructions in the electronic invoice generation module specifically include: Obtain basic tax information, using the tax item node code in the compliant release data or the corrected split invoice amount data set as an index, match and retrieve the statutory tax rate in the basic tax information, and map and bind the split amount, tax rate and the basic tax information containing the main accounting parameters. Based on the pre-defined specifications, the bound data is converted into a standard business and tax invoicing instruction that includes an invoice header and a commodity transaction details section; The data digest of the invoice instruction is calculated using a hash algorithm and a digital signature with an encrypted private key is attached. The digest is then sent to an external electronic invoice service platform via network communication to trigger the coding and generation process. When the receipt indicates that the invoice issuance failed, the error code is extracted and written to the exception retry queue.

[0015] A method for matching telecommunications invoice content based on multi-dimensional business proportion analysis includes the following steps: Extract heterogeneous service consumption data from the underlying billing gateway and perform range normalization to generate a standard usage matrix; The overall activity scalar of the standard usage matrix is ​​evaluated to determine the performance status of the current billing cycle, and historical data is extracted and the historical baseline entropy weight vector is output when the performance status is determined to be dormant. Calculate the temporal information entropy of each business dimension, and generate the final dynamic entropy weight vector by combining it with the determined performance status; Obtain the total invoice amount scalar, and perform tensor scalar dot product projection along the direction of the final dynamic entropy weight vector to split it into each tax item node, generating a data set containing the final split invoice amount of each tax item node. An out-of-bounds check is performed on the final split invoice amount. When the check passes, compliant release data is output. When an out-of-bounds error occurs, an out-of-bounds blocking signal is generated. When the boundary blocking signal is received, the multidimensional tax base Lagrange constraint is invoked to perform a quadratic programming optimization calculation to generate a corrected set of split invoice amount data; The compliant release data or the corrected split invoice amount data set is encapsulated into a standard business and tax invoicing instruction with a digital signature, and sent to an external tax invoicing subsystem to generate an electronic invoice.

[0016] This invention provides a telecommunications invoicing content matching system based on multi-dimensional service proportion analysis. It has the following beneficial effects: 1. This invention effectively eliminates the interference of inconsistent physical dimensions of massive heterogeneous telecommunications services on invoice weight calculation by combining range normalization processing with time-series information entropy calculation. At the same time, the system introduces a performance obligation dormancy detection and inter-period sliding time window mechanism. When the overall activity scalar falls below the dynamic dormancy threshold, the exponential smoothing baseline calibration logic is automatically triggered, and the historical baseline entropy weight is adaptively fitted with the original transient entropy weight. When the user's service consumption is extremely low or in a dormant state, the system can still output an objective dynamic entropy weight vector based on the historical stable baseline, avoiding the extreme change phenomenon of invoice weight caused by single-cycle data sparsity.

[0017] 2. This invention employs tensor-scalar dot product projection operations combined with an orthogonal penalty function of zero consumption dimension to enhance the dimensionality reduction and decomposition capability from the business system to the financial and tax system. By constructing a topological mapping matrix between the underlying business and the upper-level tax system, the system can accurately decompose the total invoiced amount to each tax item node along the direction of the dynamic entropy weight vector. For extreme cases where some businesses have continuous zero consumption during the billing cycle, the system activates the orthogonal penalty function and corrects the business weights by superimposing the orthogonal minimum guaranteed basis vector, thereby compensating for the resource pre-occupancy cost on the underlying network side. This effectively avoids the loophole that the invoicing ratio of the corresponding tax item will be forcibly reduced to zero due to zero consumption of a single business, ensuring the integrity of the amount decomposition logic.

[0018] 3. This invention constructs a closed-loop compliance mechanism based on red line boundary verification and multi-dimensional tax base Lagrange recalculation, which significantly improves the compliance security of the system's invoice issuance. The system constructs a strict tax red line range based on historical benchmark invoice ratios and fluctuation tolerance, and performs hard boundary interception on the split amount. When the boundary blocking signal is triggered, the system uses minimizing the weight adjustment range as the calculation objective, constructs a Lagrange quadratic objective function with equality and inequality constraints, and calls a quadratic programming algorithm. Under the premise of strictly satisfying the tax red line boundary and the conservation of total amount, it can calculate the optimal target weight value set, realize the system's automatic optimization and correction in the case of boundary anomalies, and reduce the intervention cost of manual verification. Attached Figure Description

[0019] Figure 1 This is a system architecture diagram of the telecommunications invoicing content matching system based on multi-dimensional business proportion analysis according to the present invention; Detailed Implementation Example: This system sits between the underlying billing gateway and the external tax invoicing subsystem, and includes: The heterogeneous service call detail record collection and time series matrix normalization processing module 100 is used to extract service consumption from the billing gateway and generate a standard usage matrix through range normalization.

[0020] The compliance obligation dormancy detection and inter-period time window status assessment module 200 is used to assess the overall activity of the matrix to determine the compliance status, and output the historical baseline entropy weight vector when in dormancy.

[0021] The temporal information entropy calculation and baseline calibration evolution module 300 is used to calculate the temporal information entropy and generate the final dynamic entropy weight vector in combination with the performance status.

[0022] The amount tensor projection module 400 is used to perform dot product projection of the total invoice amount along the direction of the final dynamic entropy weight vector and split it into each tax item node.

[0023] The Tax Base Red Line Verification and Compliance Blocking Module 500 is used to perform red line boundary verification on the split amount and output compliant release data or boundary blocking signal.

[0024] The Multidimensional Tax Base Lagrange Recalculation Module 600 is used to perform quadratic programming optimization calculations when exceeding limits, generating compliant corrected amount data.

[0025] The Business Finance and Tax Instruction Encapsulation and Electronic Invoice Generation Module 700 is used to encapsulate the final amount into an invoice instruction with a digital signature and send it to the tax invoice issuance subsystem to generate an invoice.

[0026] In this embodiment of the invention, the specific implementation method of the heterogeneous service call detail record collection and timing matrix normalization processing module 100 performing physical signaling data collection and raw usage construction includes the following steps: S201, the underlying communication network nodes generate heterogeneous original call detail record data based on the service-triggered signaling of the user terminal.

[0027] In practice, different network element entities, such as the core network, IP multimedia subsystem, or broadband access server, are responsible for recording service data with different physical dimensions, such as traffic bytes, call minutes, or online duration, thus forming heterogeneous dimensional record data.

[0028] S202, the billing gateway aggregates the raw call detail record data generated by each underlying communication network node.

[0029] The billing gateway periodically pulls or receives in real time through the communication interface to achieve unified aggregation and protocol conversion of call detail records pushed by all network element nodes across the network.

[0030] S203, the heterogeneous service call detail record collection and time series matrix normalization processing module 100 parses and extracts fields from the aggregated raw record data.

[0031] The heterogeneous service call detail record collection and time series matrix normalization processing module 100 extracts data fields including user identifier, service type code, service occurrence timestamp, and original consumption value, and maps the recorded data to N predefined service dimensions corresponding to the sub-items of the telecom converged package service.

[0032] S204, Heterogeneous service call detail record collection and time series matrix normalization processing module 100 constructs the original usage matrix and performs range normalization to generate the standard usage matrix.

[0033] The heterogeneous service call detail record (CDR) collection and time-series matrix normalization processing module 100 performs equidistant time-domain slices on the target billing cycle, and aggregates the original consumption values ​​of each item under the same user identifier according to timestamp and service type to the corresponding spatiotemporal cross node, thereby constructing an original usage matrix reflecting the underlying physical signaling triggering state. Subsequently, in order to eliminate the difference in physical dimensions of consumption of different services, the module uses the range transformation method to combine the extreme values ​​of consumption of each service to transform the matrix elements into dimensionless standardized values, and finally generates a standard usage matrix.

[0034] The technical principles of the present invention will be explained in detail below in conjunction with the above steps, specifically including: The heterogeneous service call detail record (CDR) collection and time series matrix normalization processing module 100 obtains the total time span of the target billing period and divides this total time span into T time slices according to a preset sampling step size, where T is a positive integer greater than or equal to 1. The sampling step size is set according to the data statistics period of the billing gateway, for example, 24 hours or 1 hour. These T time slices constitute the time series dimension of the original usage matrix.

[0035] Simultaneously, the heterogeneous service call detail record (CDR) collection and time-series matrix normalization processing module 100 reads the configuration data of the user's subscribed telecom converged package, parses and extracts the various service sub-items included in the package. Each service sub-item is mapped to a service dimension, resulting in a total of N service dimensions, where N is a positive integer greater than or equal to 1. These N service dimensions constitute the service attribute dimensions of the original usage matrix. The aforementioned service dimensions cover the classification of heterogeneous network resources such as data traffic, voice calls, and various value-added services.

[0036] After determining the time-series dimension and the service attribute dimension, the heterogeneous service call detail record (CDR) collection and time-series matrix normalization processing module 100 allocates a two-dimensional array with T rows and N columns in the system memory space, using T time slices as row indices and N service dimensions as column indices to construct the original usage matrix X. To avoid matrix operation anomalies caused by no service occurring in certain time periods, the module initializes all elements in the original usage matrix X to zero.

[0037] The matrix elements in the original usage matrix are defined as follows: ,in This is the time slice number, with a value ranging from 1 to... , This is the business dimension index, with a value range from 1 to... The heterogeneous service call detail record (CDR) collection and time-series matrix normalization processing module 100 extracts consumption data, service occurrence timestamps, and service type codes from the aggregated CDR data. The module calculates the difference between the service occurrence timestamp and the start time of the target billing cycle, divides it by the sampling step size, and rounds down to obtain the corresponding time-series row index. Simultaneously, it matches the corresponding service column index i based on the service type code. The calculated row index... With column index i, the heterogeneous service call detail record collection and time series matrix normalization processing module 100 accumulates the extracted consumption data and fills it into the corresponding element of the original consumption matrix. In the middle. After traversing and processing all call detail records (CDRs) within the target billing period, complete all matrix elements. The numerical loading. The original usage matrix after loading. The output is sent to the system buffer as the base data source for performing range transformation processing to generate the standard usage matrix.

[0038] The heterogeneous service call detail record collection and time series matrix normalization processing module 100 reads the original usage matrix from the system cache. Obtain the original usage matrix. Total number of time slices , and the total number of business dimensions Simultaneously, allocate a dimension of [dimension value] in system memory. OK The standard usage matrix is ​​pre-constructed and initialized in the two-dimensional array space of the column. .

[0039] Subsequently, the heterogeneous service call detail record collection and time series matrix normalization processing module 100 processes the data according to the column index. Extract the original usage matrix one by one. The Middle A collection of usage time series data for each business dimension This set Including the Each business dimension All consumption values ​​under each time slice.

[0040] For the extracted set The heterogeneous service call detail record (CDR) collection and time series matrix normalization processing module 100 performs a sorting operation on the values ​​to obtain the minimum value in the set. and maximum value And calculate the difference between the two.

[0041] The heterogeneous service call detail record collection and time series matrix normalization processing module 100 determines whether the difference is zero: when the maximum value is zero. and minimum value When the difference is zero, it is determined that the usage of this service is stable or has not been consumed within the billing cycle. To avoid calculation overflow errors caused by dividing by zero, the heterogeneous service call detail record collection and time series matrix normalization processing module 100 will process the first... Standardized values ​​of each business dimension across all time slices Assign it to zero directly and write it into the standard usage matrix. In the corresponding element position.

[0042] When the maximum value and minimum value When the difference is not zero, the heterogeneous service call detail record collection and time series matrix normalization processing module 100 performs range transformation calculation for the original usage matrix. any element in Calculate the corresponding standardized value according to the following formula. : In the formula, Indicates the first Consumption of each business dimension in the first time slice; Indicates the first The minimum consumption of this type of service within the current billing cycle; Indicates the first The maximum consumption of this type of service within the current billing cycle.

[0043] Standardized values ​​were calculated. Subsequently, the heterogeneous service call detail record (CDR) collection and time series matrix normalization processing module 100 indexes them according to the corresponding time slice rows. Indexes of business dimension columns Fill in the standard usage matrix In the corresponding element position.

[0044] By iterating through all In terms of business operations, the heterogeneous business call detail record collection and time series matrix normalization processing module 100 completes the range transformation calculation and filling of all matrix elements. This process converts consumption quantities with different dimensions into dimensionless values ​​in the range of 0 to 1, so as to eliminate the differences in physical units of data traffic, voice duration, service call count, and other dimensions.

[0045] Finally, the heterogeneous service call detail record collection and time series matrix normalization processing module 100 outputs the filled standard usage matrix to the performance obligation dormancy detection and inter-period time window status assessment module 200, as the data source for subsequent calculation of the overall activity scalar and spatiotemporal information entropy.

[0046] This invention provides a specific implementation method for global activity scalar calculation and dynamic threshold determination. This method is executed by the aforementioned performance obligation dormancy detection and inter-period time window status assessment module 200, and specifically includes the following processing steps: The performance obligation dormancy detection and inter-period time window status assessment module 200 receives the standard usage matrix output by the heterogeneous service call detail record collection and time series matrix normalization processing module 100. And directly to the standard usage matrix All elements are summed to obtain the overall activity scalar for the current billing cycle. The specific calculation formula is as follows: In the formula, This represents the overall activity level scalar for the current billing cycle; Representing the standard usage matrix The Middle The business dimension in the first Standardized values ​​under a time slice; This indicates the total number of time slices in the current billing cycle; Indicates the total number of business dimensions; This represents the time slice number, with values ​​ranging from 1 to... Positive integers; This represents the business dimension index, with values ​​ranging from 1. A positive integer; represents the summation operator.

[0047] After obtaining the overall activity scalar Subsequently, the fulfillment obligation dormancy detection and inter-period time window status assessment module 200 reads the system's preset basic activity tolerance coefficient. To enable the threshold to adaptively adjust with changes in the number of days in a calendar month and the complexity of the service package, the fulfillment obligation dormancy detection and inter-period time window status assessment module 200 will adjust the basic activity tolerance coefficient. Total number of time slices Total number of business dimensions Perform a product operation to generate a dynamic dormancy threshold that is adapted to the current billing cycle. The specific calculation formula is as follows: Among them, the basic activity tolerance coefficient This is an empirical constant preset by the system. In specific implementation, this coefficient is obtained by the system based on usage sample data of historical offline users or long-term dormant users, and its value ranges from greater than 0 to less than 1; as a preferred implementation method, the basic activity tolerance coefficient... The value range is set to 0.01 to 0.05. In the formula: Indicates the dynamic sleep threshold; This represents the tolerance coefficient for basic activity. This indicates the total number of time slices in the current billing cycle; This indicates the total number of business dimensions.

[0048] Subsequently, the compliance obligation dormancy detection and inter-period time window status assessment module 200 calculates the overall activity scalar. With dynamic sleep threshold Compare the results. When the overall activity scalar... Greater than or equal to the dynamic sleep threshold At that time, the fulfillment obligation dormancy detection and cross-period time window status assessment module 200 determines that the service consumption status within the current billing cycle is a normal fulfillment active period. When the overall activity scalar Less than the dynamic hibernation threshold When the performance obligation dormancy detection and cross-period time window status assessment module 200 determines the service consumption status within the current billing cycle, it triggers the performance dormancy state.

[0049] After the determination is completed, the performance obligation dormancy detection and cross-period time window status assessment module 200 records the determination result of the performance status in the system memory or database, as the trigger condition for the subsequent control system to execute the timing information entropy direct pass mechanism or the cross-period historical time window tracing mechanism.

[0050] Additional notes: Indicates the first The business dimension in the first Standardized values ​​under a time slice; Indicates the first The business dimension in the first Consumption under each time slice; Indicates the first A collection of usage time series for each business dimension within the current billing cycle; Represents a set of time series data on usage. The minimum consumption in; Represents a set of time series data on usage. The maximum consumption amount.

[0051] The performance obligation dormancy detection and cross-period time window status assessment module 200 reads the performance status determination result of the current billing cycle recorded in the system memory or database. When the determination result is in the performance dormancy state, the performance obligation dormancy detection and cross-period time window status assessment module 200 triggers the cross-period data tracing logic and generates a historical data query instruction.

[0052] Subsequently, the performance obligation dormancy detection and inter-period time window status assessment module 200 obtains the system's preset time window length parameter. This parameter represents the total number of billing cycles that need to be traced back to history, denoted as... , It is a positive integer greater than or equal to; in specific implementation, the time window length parameter The preferred value is 3. The performance obligation dormancy detection and inter-period time window status assessment module 200 uses the start time of the current billing cycle as the reference node, and traces back along the time series to the historical record to construct a continuous period covering the current billing cycle. A sliding time window spanning a historical billing cycle.

[0053] The fulfillment obligation dormancy detection and inter-period time window status assessment module 200 extracts the user identifier of the current service, sends the user identifier and the time span of the inter-period sliding time window as search conditions to the historical billing database, and extracts the historical service records of the user identifier within the inter-period sliding time window from the database. The extracted historical service records include the user's continuous Historical standard usage matrix and historical consumption data for each service within a historical billing cycle.

[0054] To accommodate situations where historical data for newly registered users is insufficient, the performance obligation dormancy detection and cross-period time window status assessment module 200 performs periodic data integrity checks on the extracted business history records. If the actual number of historical billing cycles retrieved reaches the time window length parameter If the historical data extraction is complete, then the number of historical billing periods actually retrieved is less than the time window length parameter. The compliance obligation dormancy detection and inter-period time window status assessment module 200 uses the actual number of historical billing cycles acquired as an adjustment value to numerically correct the time window length parameter and reset the range boundary of the inter-period sliding time window based on the corrected value. After completing the extraction and verification of historical data, the compliance obligation dormancy detection and inter-period time window status assessment module 200 writes all business history records within the final inter-period sliding time window into the system cache space. The business history records stored in the system cache space serve as the data analysis source for subsequent calculation of the historical baseline entropy weight vector.

[0055] This invention provides a specific implementation method for extracting historical stable baseline features. This method is executed by the aforementioned performance obligation dormancy detection and inter-period time window status assessment module 200, and specifically includes the following processing steps: The performance obligation dormancy detection and inter-period time window status assessment module 200 reads the business history records in the system cache space, parses and obtains the historical performance status identifiers for H consecutive historical billing cycles within the inter-period sliding time window, as well as the historical dynamic entropy weight vectors that have been fixed and stored in these historical cycles. Based on the obtained historical performance status identifiers, the performance obligation dormancy detection and inter-period time window status assessment module 200 performs status filtering processing on the H consecutive historical billing cycles, removing billing cycles identified as performance dormancy states, and retaining billing cycles identified as normal performance active periods, thereby generating a set of valid historical billing cycles. The number of billing cycles contained in this set of valid historical billing cycles is denoted as Q, where Q is an integer greater than or equal to 0. When the value of Q is greater than 0, it indicates that there is valid active data within the historical window. The compliance obligation dormancy detection and cross-period time window status assessment module 200 extracts the historical dynamic entropy weight values ​​corresponding to the Q valid historical billing periods for the predefined i-th business dimension, and calculates the historical baseline entropy weight value according to the following formula: In the formula: Indicates the first Historical baseline entropy weights for each business dimension; This represents the total number of billing cycles in the set of valid historical billing cycles; This indicates the serial number of the valid historical billing period, with a value ranging from 1 to... Positive integers; Indicates the first In the first valid historical billing cycle, the Historical dynamic entropy weight values ​​corresponding to each business dimension; This represents the business dimension index, with a value range from 1 to... A positive integer. When When the value is 0, it indicates that all historical billing periods within the inter-period sliding time window are in a dormant state or have no valid historical records. At this time, the performance obligation dormancy detection and inter-period time window status assessment module 200 triggers the default baseline allocation mechanism, assigning the first... Historical baseline entropy weights for each business dimension direct assignment ,in This represents the total number of business dimensions. The fulfillment obligation dormancy detection and inter-period time window status assessment module traverses all 200... For each business dimension, complete the calculation or assignment of all historical baseline entropy weights. Then, based on the resulting... Each historical baseline entropy weight is used to construct a historical baseline entropy weight vector in system memory. Finally, the compliance obligation dormancy detection and inter-period time window status assessment module 200 will generate a historical baseline entropy weight vector. The output is sent to the temporal information entropy calculation and baseline calibration evolution module 300 as the basic data source for subsequent baseline smoothing calibration operations.

[0056] The invention provides a specific implementation method for transient temporal information entropy calculation, which is executed by the aforementioned temporal information entropy calculation and baseline calibration evolution module 300, and specifically includes the following processing steps: The temporal information entropy calculation and baseline calibration evolution module 300 receives the standard usage matrix output by the heterogeneous service call detail record acquisition and temporal matrix normalization processing module 100. It then iterates through the columns by index to extract the standardized numerical sequence of each business throughout the entire cycle.

[0057] To prevent division-by-zero overflow exceptions in subsequent calculations, the time-series information entropy calculation and baseline calibration evolution module 300 first calculates the... The sum of the standardized values ​​of each business dimension across all time slices And determine whether the summation value is zero: when At that time, it indicates that the first Each business dimension is in a zero-consumption state during the current billing cycle. At this time, the time-series information entropy calculation and baseline calibration evolution module 300 directly calculates the information difference degree of this business dimension. Assign a value of 0 and skip the information entropy calculation for that dimension.

[0058] when At that time, the time-series information entropy calculation and baseline calibration evolution module 300 calculates the first... The business dimension in the first Feature proportions in each time slice The specific calculation formula is as follows: In the formula: Indicates the first The business dimension in the first Feature weighting under each time slice; Representing the standard usage matrix The Middle The business dimension in the first Standardized values ​​under a time slice; This indicates the total number of time slices in the current billing cycle; Indicates the time slice number, incrementing from 1 to... ; Indicates the business dimension number, incrementing from 1 to... , This represents the total number of business dimensions.

[0059] After obtaining the feature weights, the time-series information entropy calculation and baseline calibration evolution module 300 performs time-series information entropy calculation. To avoid calculation errors caused by the logarithmic function having zero independent variables, the system has an embedded limit convergence rule: when the feature weights... At that time, the calculation result of this item is specified. Based on this rule, module 300 calculates the temporal information entropy of the th business dimension, using the following formula: In the formula: Indicates the first Temporal information entropy of each business dimension; It represents the natural logarithm with the natural constant e as the base; This represents the total number of time slices in the current billing cycle. To ensure the existence of the time-series fluctuation sequence and the validity of the denominator, The value of is limited to positive integers greater than or equal to 2; Indicates the first The business dimension in the first Feature weighting under each time slice; Indicates the time slice number, incrementing from 1 to... .

[0060] After calculating the temporal information entropy, the temporal information entropy calculation and baseline calibration evolution module 300 calculates the entropy according to the formula. Calculate the first Information differences across business dimensions .

[0061] Traverse all the branches described above. After each business dimension, the time-series information entropy calculation and baseline calibration evolution module 300 obtains all... The information difference in each dimension is used to generate the original transient entropy weight for the current billing cycle. Original transient entropy weights for each business dimension The calculation formula is: In the formula: Indicates the first Original transient entropy weights for each business dimension; Indicates the first Information discrepancies across business dimensions; Indicates all The information differences across each business dimension are summed.

[0062] Finally, the time-series information entropy calculation and baseline calibration evolution module 300 will calculate the... Each original transient entropy weight is constructed in memory as the original transient entropy weight vector for the current billing cycle. This original transient entropy weight vector serves as the basis data for subsequent execution of the weight pass-through mechanism during the normal performance period or the exponential smoothing baseline calibration mechanism during the performance dormancy period.

[0063] The time-series information entropy calculation and baseline calibration evolution module 300 reads the performance status determination result of the current billing cycle recorded in the system memory or database. This determination result is generated by the performance obligation dormancy detection and inter-period time window status assessment module 200 in the pre-state assessment process.

[0064] When the determination result indicates a normal active performance period, it means that the user's communication network consumption in the current billing cycle can effectively reflect the current service preferences, and there is no need for historical benchmark smoothing. At this time, the time-series information entropy calculation and baseline calibration evolution module 300 triggers the weight pass-through execution logic.

[0065] Under the weighted pass-through execution logic, the temporal information entropy calculation and baseline calibration evolution module 300 obtains the original transient entropy weight vector calculated in the aforementioned steps, and directly assigns this original transient entropy weight vector as the final dynamic entropy weight vector. For any of the... The final business weight assignment for each business dimension. The equation is: In the formula: Indicates the first The final business weight for each business dimension; Indicates the first Original transient entropy weights for each business dimension; Indicates the business dimension number, incrementing from 1 to... ; This indicates the total number of business dimensions.

[0066] The temporal information entropy calculation and baseline calibration evolution module traverses all 300... Each business dimension completes the pass-through assignment operation, generating a value in system memory containing... The final dynamic entropy weight vector of each final business weight .

[0067] To support the data iteration of the system's inter-period baseline calibration mechanism, the time-series information entropy calculation and baseline calibration evolution module 300 will generate the final dynamic entropy weight vector. The status identifier of the current billing cycle's normal active performance period is also synchronously and persistently written to the historical billing database. This written data will serve as historical sample features when triggering cross-period sliding time windows for subsequent billing cycles.

[0068] Finally, the time-series information entropy calculation and baseline calibration evolution module 300 generates the final dynamic entropy weight vector. The output is sent to the topological constraint-based amount tensor projection module 400 as the spatial direction reference parameter for subsequent tensor-scalar dot product operations and tax item invoice amount splitting.

[0069] This invention provides a specific implementation of an exponentially smoothed baseline calibration algorithm for a dormant performance status. This method is executed by the aforementioned time-series information entropy calculation and baseline calibration evolution module 300, specifically including: the time-series information entropy calculation and baseline calibration evolution module 300 reads the performance status determination result of the current billing cycle. When the determination result is a dormant performance status, it indicates that the user activity level has fallen below a threshold. To avoid sudden changes in weight data, the time-series information entropy calculation and baseline calibration evolution module 300 triggers the exponentially smoothed baseline calibration logic.

[0070] When executing the index smoothing baseline calibration logic, the time-series information entropy calculation and baseline calibration evolution module 300 obtains the original transient entropy weight vector of the current billing cycle generated by the previous steps; and simultaneously receives the historical baseline entropy weight vector, overall activity scalar, and dynamic dormancy threshold output by the performance obligation dormancy detection and inter-period time window status assessment module 200.

[0071] Based on the data obtained above, the time-series information entropy calculation and baseline calibration evolution module 300 performs a smoothing factor calculation operation to determine the fitting ratio between the original transient entropy weight of the current period and the historical baseline entropy weight, and the smoothing factor. The calculation formula is: In the formula: Indicates the smoothing factor; This represents the overall activity level scalar for the current billing cycle; This represents the dynamic sleep threshold.

[0072] Based on the judgment logic, the overall activity scalar under the performance dormancy state... Strictly less than the dynamic hibernation threshold And the overall activity scalar The smoothing factor is greater than or equal to zero, therefore the calculated smoothing factor is... The range of values ​​is in This satisfies the mathematical constraints of the data smoothing weighting coefficients. The essential technical control significance of this formula lies in: when the overall user activity scalar... When the smoothing factor approaches 0 (i.e., extremely dormant), Approaching 0, the system's final business weight will depend heavily on the historical stable baseline; when the overall activity scalar... Approaching the dormancy threshold (i.e., during mild dormancy) smoothing factor As the value approaches 1, the system retains more of the original transient characteristics of the current cycle. This linear decay fitting method enables a seamless adaptive transition from normal performance to deep sleep.

[0073] The smoothing factor was calculated. Subsequently, the time-series information entropy calculation and baseline calibration evolution module 300 targets the first... For each business dimension, extract the corresponding original transient entropy weights and historical baseline entropy weights, and perform exponential smoothing fitting calculations. Final business weights for each business dimension The calculation formula is: In the formula: Indicates the first The final business weight for each business dimension; Indicates the smoothing factor; Indicates the first Original transient entropy weights for each business dimension; Indicates the first Historical baseline entropy weights for each business dimension; Indicates the business dimension number, incrementing from 1 to... ; This indicates the total number of business dimensions.

[0074] The temporal information entropy calculation and baseline calibration evolution module traverses all 300... Each business dimension completes smooth fitting calculations for all business dimensions. Through the above calculations, the time-series information entropy calculation and baseline calibration evolution module 300 generates a data set in system memory containing... The final dynamic entropy weight vector of each final business weight .

[0075] Similar to the persistence mechanism during normal active performance periods, the time-series information entropy calculation and baseline calibration evolution module 300 will generate the final dynamic entropy weight vector. The current billing cycle's performance dormancy status identifier is persistently written to the historical billing database. The final dynamic entropy weight vector written to the historical billing database will serve as historical sample data when triggering cross-period sliding time window retrospective in subsequent billing cycles.

[0076] Finally, the temporal information entropy calculation and baseline calibration evolution module 300 will generate the final dynamic entropy weight vector. The output is sent to the topologically constrained amount tensor projection module 400. This is the final dynamic entropy weight vector. This serves as the spatial directional reference parameter for subsequent tensor-scalar dot product operations and tax item invoice amount splitting.

[0077] The Amount Tensor Projection Module 400 reads the system's predefined... Each business dimension, and the Each business dimension is strictly aligned with the business dimension in the final dynamic entropy weight vector output by the preceding temporal information entropy calculation and baseline calibration evolution module 300.

[0078] Simultaneously, the amount tensor projection module 400 connects to the basic configuration database of the tax invoicing subsystem and extracts the currently applicable set of tax invoice items from this database. This set of tax invoice items contains... A tax item node with an independent tax code and corresponding tax rate, It is a positive integer greater than or equal to 1. In practice, the tax item node covers tax categories such as basic telecommunications services, value-added telecommunications services, and hardware sales.

[0079] After acquiring the business dimensions and tax item nodes, the amount tensor projection module 400 allocates the following dimensions in system memory: OK A two-dimensional array space of columns is used to construct the business tax topology mapping matrix. The amount tensor projection module 400 maps the tax industry topology matrix. The matrix elements are defined as follows: .

[0080] Amount Tensor Projection Module 400 Traversals Each business dimension and For each tax item node, a mapping relationship is established between business attributes and tax attributes. In practice, the system memory or database pre-loads an accounting mapping table containing the correspondence between underlying business type codes and upper-level tax codes. For the th business dimension and the th tax item node, the amount tensor projection module 400 extracts the ... The business type code for each business dimension is used to retrieve whether it is associated with the first business dimension in the accounting mapping table. Tax code for each tax item node: When the match is successful, the first step is determined. The business dimension belongs to the first When dealing with individual tax item nodes, the amount tensor projection module 400 will map the corresponding matrix elements in the business tax topology mapping matrix. The value is assigned to 1; When the matching retrieval fails, that is, when the first step is determined... The business dimension does not belong to the first When dealing with individual tax item nodes, the amount tensor projection module 400 will map the corresponding matrix elements in the business tax topology mapping matrix. The value is assigned to 0.

[0081] The logical relationship between the above-mentioned business tax topology mapping matrix and its elements is expressed as follows: In the formula: Represents the business tax topology mapping matrix The middle is located in the first Line number Column matrix elements; This represents the business dimension index, with values ​​increasing from 1 to... ; This represents the tax item node number, with values ​​increasing from 1 to... ; Indicates the total number of business dimensions; This indicates the total number of tax item nodes.

[0082] Through the above retrieval and assignment operations, the amount tensor projection module 400 completes the assignment operation for all matrix elements. The completed business tax topology mapping matrix... It records the mapping structure from the underlying communication services to the upper-level tax invoice commodity codes. Finally, the amount tensor projection module 400 generates the business and tax topology mapping matrix. It is stored in the system cache and serves as the basis for subsequent tensor-scalar dot product operations.

[0083] This invention provides a specific implementation of an orthogonal penalty function activation mechanism with zero consumption dimension. This method is executed by the aforementioned amount tensor projection module 400, specifically including: the amount tensor projection module 400 obtains the original usage matrix output by the heterogeneous service call detail record collection and time series matrix normalization processing module 100, and extracts the time series data of each service within the current billing cycle.

[0084] For the extracted time series data, the amount tensor projection module 400 statistically analyzes the first... The system calculates the number of time slices within a billing cycle where the consumption for each business dimension is consecutively zero, and compares this number with a preset zero-consumption tolerance threshold. In practice, this zero-consumption tolerance threshold can be set to the total number of time slices in the current billing cycle. The preset ratio, for example 80% or 100%.

[0085] When the number of time slices with consecutive zero consumption is greater than or equal to the zero consumption tolerance threshold, the first time slice is determined to be zero. A zero-consumption condition is triggered in one business dimension. At this time, the amount tensor projection module 400 applies this condition to the first business dimension. Each business dimension activates orthogonal penalty function operations to compensate for the resource pre-occupancy cost on the underlying communication network side.

[0086] Under the orthogonal penalty function operation logic, the amount tensor projection module 400 accesses the basic configuration database, reads the preset basic maintenance coefficient configuration table, and retrieves and extracts the values ​​related to the first... Basic maintenance coefficients corresponding to each business dimension .

[0087] Based on the extracted basic maintenance coefficient The amount tensor projection module 400 constructs a target for the first dimension business weight space within the first dimension business weight space. Orthogonal minimum guaranteed basis vectors for each business dimension The orthogonal minimum guaranteed basis vector In the The component values ​​in each dimension form the basic maintenance coefficient. ,the remaining The component values ​​in each dimension are all 0.

[0088] The amount tensor projection module 400 obtains the final dynamic entropy weight vector output by the pre-processing. Due to the first Each business dimension triggered the zero-consumption condition, which in The original weight values ​​in Approaching or equal to 0. The amount tensor projection module 400 uses vector addition to transform the orthogonal minimum guaranteed basis vector. Superimposed on the final dynamic entropy weight vector In the process, generate the business weights after penalty correction. Its mathematical expression is: The amount tensor projection module iterates through all 400 times. For each business dimension that triggers the zero-consumption condition, the above penalty compensation component calculation and summation operation is performed sequentially. If the zero-consumption condition is not triggered, then... .

[0089] To ensure the consistency of spatial orientation in subsequent tensor projections and the conservation of total financial amount, after completing all... After traversing each dimension, the amount tensor projection module 400 performs normalization processing on the superimposed and corrected full weight vector to generate a projection reference direction vector including orthogonal penalty correction. Among them, the first The formula for calculating the corrected weights for each business dimension is as follows: In the formula: Represents the normalized i-th Adjusted weights for each business dimension; Indicates the first Business weights adjusted for penalties across each business dimension; Indicates all The sum of the weighted values ​​of penalties from each business dimension; and These are all business dimension numbers, with values ​​increasing from 1 to... .

[0090] Through this orthogonal penalty function activation mechanism, the amount tensor projection module 400 avoids the calculation loophole where zero business consumption leads to a forced zeroing of the corresponding tax item's invoicing ratio. The generated projection reference direction vector includes orthogonal penalty correction. Stored in system memory, it is used to replace the original final dynamic entropy weight vector as the spatial direction parameter for performing subsequent total amount dot product splitting.

[0091] The amount tensor projection module 400 receives the total invoice amount scalar for the current billing period from the service billing system. Simultaneously, the projection reference direction vector stored in the system memory is read. Business tax topology mapping matrix Among them, the projection reference direction vector Include Weight values ​​for each business dimension Business tax topology mapping matrix Include OK Column mapping elements .

[0092] Subsequently, the amount tensor projection module 400 extracts the projection reference direction vector. Business tax topology mapping matrix Perform tensor dot product operations to calculate the projection transformation weights for each tax item node. For the Each tax item node has its corresponding projection transformation weight. The formula for the tensor dot product operation is as follows: In the formula: Indicates the first Projection transformation weights for each tax item node; Represents the first in the projection reference direction vector Weight values ​​for each business dimension; Represents the business tax topology mapping matrix The middle is located in the first Line number Column matrix elements; This represents the business dimension index, with values ​​increasing from 1 to... ; This represents the tax item node number, with values ​​increasing from 1 to... ; This indicates the total number of business dimensions.

[0093] By traversing all After each tax item node completes the above calculations, the amount tensor projection module 400 reduces the dimensionality of the business weight space to map to... V's tax invoicing space generates invoices containing Tax item weight vector of each projection transformation weight .

[0094] After deriving the tax item weight vector Then, the amount tensor projection module 400 performs a scalar multiplication projection operation, transforming the total invoice amount into a scalar... Respectively with the tax item weight vector The projection transformation weights in Perform the product calculation. For the first... Individual tax item node, its initial split invoicing amount The calculation formula is: In the formula: Indicates the first Initial split invoicing amount for each tax item node; This represents the total invoice amount for the current billing cycle. Indicates the first Projection transformation weights for each tax item node.

[0095] To meet the tax system's requirements for the accuracy of electronic invoice amounts and ensure the balance of the general ledger, after calculating and obtaining the initial split invoice amounts for all tax item nodes, the amount tensor projection module 400 performs financial accuracy truncation and tail-difference adjustment operations. Specifically, the amount tensor projection module 400 will... The initial split invoice amount for individual income tax items is rounded to two decimal places (i.e., accurate to the cent), and then the total invoice amount is quantified. Subtract before The sum of the amounts after processing each tax item, the difference obtained by subtracting them, will be directly used as the first tax item. The final breakdown of individual income tax items into invoice amounts ensures that the sum of all tax item amounts is strictly equal to the total invoice amount.

[0096] Finally, the amount tensor projection module 400 encapsulates the data set containing the final split invoice amounts of all tax item nodes and outputs it to the tax invoicing subsystem as the basis for triggering the amount split execution of the electronic invoice issuance action.

[0097] The tax base red line verification and compliance blocking module 500 receives the data set containing the final split invoice amount of all tax item nodes output by the amount tensor projection module 400, and obtains the total invoice amount scalar for the current billing period. .

[0098] Regarding the first For each tax item node, the tax base red line verification and compliance blocking module 500 accesses the system rules database and reads the historical benchmark invoicing ratio and preset fluctuation tolerance corresponding to that tax item node. This fluctuation tolerance is a pre-configured numerical fluctuation threshold based on tax compliance requirements.

[0099] Based on the historical benchmark invoicing ratio and fluctuation tolerance, the tax base red line verification and compliance blocking module 500 performs numerical calculations to construct a tax red line range for the given tax item node. This tax red line range is defined by a minimum amount threshold. With the maximum amount threshold Definition. To prevent abnormal parameter configurations from causing the calculated amount to exceed the legal boundary, the tax base red line verification and compliance blocking module 500 introduces a boundary extreme value function, the specific calculation formula of which is: In the formula: Indicates the first Minimum amount threshold for the tax red line range of individual tax items; Indicates the first The maximum amount threshold within the tax red line range for each tax item; This function represents the maximum value and is used to prevent the minimum amount threshold from being negative; the function represents the minimum value and is used to prevent the maximum amount threshold from exceeding the total invoice amount. This represents the total invoice amount for the current billing cycle. Indicates the first Historical benchmark invoicing ratio corresponding to each tax item node; Indicates the first Preset fluctuation tolerance for each tax item node; This represents the tax item node number, with values ​​increasing from 1 to... .

[0100] After completing the construction of the tax red line range, the tax base red line verification and compliance blocking module 500 extracts the first... Final split invoicing amount for individual tax item nodes and the amount Introduce a tax red line range for numerical comparison and perform out-of-bounds verification: When determining the final split invoice amount Less than the minimum amount threshold or greater than the maximum amount threshold When the tax base red line verification and compliance blocking module 500 confirms that the invoice amount for that tax item node has failed verification, it generates a corresponding over-limit blocking signal; conversely, when it is determined that the final split invoice amount is greater than or equal to the minimum amount threshold, the module will block the invoice amount. And less than or equal to the maximum amount threshold At that time, the tax base red line verification and compliance blocking module 500 confirms that the invoice amount of the tax item node has passed the verification.

[0101] The tax base red line verification and compliance blocking module traversed all 500 times. For each tax item node, perform the above interval construction and out-of-bounds verification operations sequentially, and then summarize the verification results of all tax item nodes: If all The final split invoice amount for each tax item node is within the corresponding tax red line range. The tax base red line verification and compliance blocking module 500 outputs a compliance release instruction to the tax invoice issuance subsystem, triggering the formal issuance process of electronic invoices. If any tax item node in the summary result fails the verification, the tax base red line verification and compliance blocking module 500 will block the invoice issuance action of that transaction and write the generated over-boundary blocking signal and the corresponding amount breakdown details into the system's abnormal review queue as the data basis for triggering manual intervention or abnormal parameter recalculation mechanism.

[0102] The multidimensional tax base Lagrange recalculation module 600 reads the out-of-bounds blocking signals and corresponding amount breakdown details recorded in the anomaly review queue.

[0103] When an out-of-bounds blocking signal is detected, the multidimensional tax base Lagrange recalculation module 600 triggers the abnormal parameter recalculation mechanism and extracts the total invoice amount scalar of the current billing cycle that triggered the out-of-bounds blocking signal. The original set of projection transformation weights and the minimum amount threshold corresponding to each tax item node. With the maximum amount threshold .

[0104] To preserve the original business consumption characteristics as much as possible while meeting tax compliance requirements, the multidimensional tax base Lagrange recalculation module 600 constructs a quadratic objective function with the goal of minimizing the weight adjustment magnitude. For... For each tax item node, let the adjusted target weight be... The original projection transformation weights are The constructed quadratic objective function is : In the formula: This represents a quadratic objective function with the adjusted target weights as independent variables; Indicates the first Target weights after adjustments to individual tax item nodes; Indicates the first Original projection transformation weights for each tax item node; This represents the tax item node number, with values ​​increasing from 1 to... ; This indicates the total number of tax item nodes.

[0105] Simultaneously, the multidimensional tax base Lagrange recalculation module 600 constructs a set of constraints based on tax compliance boundaries, including equality constraints for the sum of weights and inequality constraints for tax red line intervals. The specific mathematical expressions are as follows: After constructing the objective function and the set of constraints, the multidimensional tax base Lagrange recalculation module 600 introduces Lagrange multiplier vectors to transform the objective function and the set of constraints into an unconstrained Lagrange function. Its expression is: In the formula: This represents the Lagrange function that combines equality and inequality constraints. This represents the Lagrange multiplier corresponding to the weighted sum equality constraint; Let represent the Caro-Kuhn-Tucker multiplier corresponding to the lower bound inequality constraint of the tax red line interval, and satisfy . ; Let represent the Caro-Kuhn-Tucker multiplier corresponding to the upper bound inequality constraint of the tax red line interval, and satisfy . ; Indicates the first Minimum amount threshold for each tax item; Indicates the first Maximum amount threshold for each tax item node; This represents the total invoice amount for the current billing cycle. Indicates the first Target weights after adjustments to individual tax item nodes; Indicates the first The original projection transformation weights of each tax item node.

[0106] After constructing the Lagrange function, the multidimensional tax base Lagrange recalculation module 600 calculates the partial derivatives of the Lagrange function according to the Caro-Kun-Tucker (KKT) conditions and sets the partial derivatives to zero. In specific implementation, the multidimensional tax base Lagrange recalculation module 600 calls the system's built-in Quadratic Programming (QP) solution algorithm (such as the effective set method or interior point method) to solve the KKT equation system through finite-step iterative calculation, thereby obtaining the optimal target weight numerical set that satisfies all constraints.

[0107] After obtaining the optimal solution, the multidimensional tax base Lagrange recalculation module 600 uses the target weight value set obtained from the solution to overwrite and replace the original projection transformation weight set in the system memory; then it extracts the total invoice amount scalar and performs scalar multiplication operation on it with each value in the target weight value set.

[0108] After the multiplication operation is completed, the multidimensional tax base Lagrange recalculation module 600 performs financial precision truncation and tail difference compensation operations on the calculation result again to generate a corrected set of split invoice amount data with a total amount strictly equal to the original.

[0109] Finally, the multidimensional tax base Lagrange recalculation module 600 encapsulates the corrected split invoice amount data set and outputs it to the tax invoicing subsystem as the data basis for re-triggering the electronic invoice issuance operation after compliance blocking.

[0110] After acquiring the above data, the business, finance and tax instruction encapsulation and electronic invoice generation module 700 uses the tax item node code in the data set as an index to match the statutory tax rate corresponding to each business, and maps and binds the split amount, tax rate and basic tax information; then, according to the preset specifications, the bound data is converted into a standard business, finance and tax invoicing instruction containing a message header and commodity transaction details.

[0111] To ensure secure cross-system transmission, the business, finance and tax instruction encapsulation and electronic invoice generation module 700 uses a hash algorithm to calculate the data digest of the instruction and attaches a private key digital signature. Then, it sends the invoicing instruction to an external electronic invoice service platform via network communication to trigger the coding and generation process.

[0112] Finally, the business, finance, and tax instruction encapsulation and electronic invoice generation module 700 receives the invoice status receipt and performs branch parsing: when the receipt indicates successful invoicing, it extracts the invoice file and number, stores them in the invoiced database, and returns an invoicing completion signal to end the current operation; when the receipt indicates invoicing failure, it extracts the error code and writes the invoicing instruction along with the error code into the abnormal retry queue, triggers an alarm, and waits for resending or manual intervention to ensure the eventual consistency of billing and tax invoicing data.

[0113] 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 telecommunications invoicing content matching system based on multi-dimensional business proportion analysis, characterized in that, include: The heterogeneous service call detail record (CDR) collection and time series matrix normalization processing module is used to extract service consumption from the billing gateway and generate a standard usage matrix through range normalization. The compliance obligation dormancy detection and inter-period time window status assessment module is used to evaluate the overall activity scalar of the standard usage matrix to determine the compliance status, and outputs the historical baseline entropy weight vector when the compliance dormancy status is determined. The temporal information entropy calculation and baseline calibration evolution module is used to calculate the temporal information entropy of each business dimension and generate the final dynamic entropy weight vector in combination with the performance status. The amount tensor projection module is used to obtain the total invoice amount scalar, and perform dot product projection on the total invoice amount scalar along the direction of the final dynamic entropy weight vector to split it into each tax item node, generating a data set containing the final split invoice amount of each tax item node; the tax base red line verification and compliance blocking module is used to perform out-of-bounds verification on the final split invoice amount and output compliance release data or out-of-bounds blocking signal. The multidimensional tax base Lagrange recalculation module is used to perform quadratic programming optimization calculations when a boundary blocking signal is received, and generate a corrected split invoice amount data set; the business, finance and tax instruction encapsulation and electronic invoice generation module is used to encapsulate the compliant release data or the corrected split invoice amount data set into a standard business, finance and tax invoice instruction with a digital signature, and send it to the tax invoicing subsystem to generate an electronic invoice.

2. The telecommunications invoicing content matching system based on multi-dimensional business proportion analysis according to claim 1, characterized in that, The heterogeneous service call detail record (CDR) collection and time series matrix normalization processing module performs range normalization to generate a standard usage matrix, specifically including: The total time span of the target billing cycle is divided into multiple time slices at equal intervals in the time domain, and multiple business dimensions are extracted based on the configuration data of the telecom converged package ordered by the user. The original consumption values ​​of each item under the same user identifier are accumulated and aggregated to the corresponding spatiotemporal cross node according to timestamp and business type to construct the original usage matrix; By using the range transformation method, and combining the extreme values ​​of consumption of each service within the billing cycle, the matrix elements in the original usage matrix are transformed into dimensionless standardized values ​​to generate the standard usage matrix.

3. The telecommunications invoicing content matching system based on multi-dimensional business proportion analysis according to claim 2, characterized in that, The compliance obligation dormancy detection and inter-period time window status assessment module determines the compliance status and outputs the historical baseline entropy weight vector, specifically including: The total activity scalar is obtained by summing all elements in the standard usage matrix, and the dynamic dormancy threshold is generated by multiplying the preset basic activity tolerance coefficient with the number of multiple time slices and the number of multiple business dimensions. When the overall activity scalar is greater than or equal to the dynamic dormancy threshold, the performance status is determined to be a normal performance active period; When the overall activity scalar is less than the dynamic dormancy threshold, the performance status is determined to be a performance dormancy state. In the performance dormancy state, a cross-period sliding time window is constructed by backtracking along the time series towards the historical record, extracting the historical dynamic entropy weight values ​​corresponding to the effective historical billing period, and calculating the average value to generate the historical baseline entropy weight vector.

4. The telecommunications invoicing content matching system based on multi-dimensional business proportion analysis according to claim 3, characterized in that, The temporal information entropy calculation and baseline calibration evolution module generates the final dynamic entropy weight vector, specifically including: Calculate the feature weight of each business dimension in each time slice, and calculate the temporal information entropy in combination with the limit convergence rule to generate the original transient entropy weight; When the performance status is in the normal performance active period, the weight direct execution logic is triggered, and the original transient entropy weight is directly assigned to the final dynamic entropy weight vector. When the performance status is in the performance dormant state, the exponential smoothing baseline calibration logic is triggered. Based on the overall activity scalar and the dynamic dormant threshold, a smoothing factor is calculated. The smoothing factor is then used to perform exponential smoothing fitting calculation on the original transient entropy weight and the historical baseline entropy weight vector to generate the final dynamic entropy weight vector.

5. A telecommunications invoicing content matching system based on multi-dimensional business proportion analysis according to claim 2, characterized in that, Before performing dot product projection, the monetary tensor projection module is also used to activate the orthogonal penalty function for zero-cost dimensions, specifically including: The number of time slices in which the consumption of each business dimension is continuously zero within the target billing period is counted. When the number is greater than or equal to the zero consumption tolerance threshold preset by the system, the corresponding business dimension is determined to have triggered the zero consumption condition. Read the preset basic maintenance coefficients and construct the orthogonal minimum guarantee basis vector corresponding to the business dimension in the business weight space; The orthogonal minimum guaranteed basis vector is superimposed onto the final dynamic entropy weight vector to generate the penalized corrected business weights. The superimposed corrected full weight vector is then normalized to generate a projection reference direction vector containing orthogonal penalty correction, which replaces the original final dynamic entropy weight vector.

6. A telecommunications invoicing content matching system based on multi-dimensional business proportion analysis according to claim 2, characterized in that, The amount tensor projection module performs dot product projection to generate a data set containing the final split invoice amount for each tax item node, specifically including: Establish the mapping relationship between each business dimension and each tax item node to construct the business and tax topology mapping matrix, extract the final dynamic entropy weight vector and perform tensor dot product operation with the business and tax topology mapping matrix to generate tax item weight vector; The total invoice amount scalar is multiplied by each projection transformation weight in the tax item weight vector to obtain the initial split invoice amount for all tax item nodes. The initial split invoicing amount of each tax item node except the last tax item node is truncated to two decimal places according to the rules. The total invoicing amount is then subtracted from the truncated amount, and the remaining difference is taken as the final split invoicing amount of the last tax item node. The tail difference is then balanced.

7. A telecommunications invoicing content matching system based on multi-dimensional business proportion analysis according to claim 1, characterized in that, The tax base red line verification and compliance blocking module performs red line overrun verification, specifically including: For each tax item node, the corresponding historical benchmark invoicing ratio and preset fluctuation tolerance are read; a boundary extreme value function is introduced, and numerical calculations are performed in combination with the total invoicing amount scalar to construct a tax red line range that includes a minimum amount threshold and a maximum amount threshold, so as to prevent the threshold from becoming negative or exceeding the total invoicing amount scalar. The final split invoice amount of the corresponding tax item node is introduced into the tax red line range for numerical comparison. If none of them exceed the limit, the compliant release data is output. If it is determined that the final split invoice amount is less than the minimum amount threshold or greater than the maximum amount threshold, then the verification is confirmed to have failed, and the corresponding over-limit blocking signal and the amount split details are generated and written into the abnormal review queue.

8. A telecommunications invoicing content matching system based on multi-dimensional business proportion analysis according to claim 7, characterized in that, The multidimensional tax base Lagrange recalculation module performs quadratic programming optimization calculations, specifically including: Extract the total invoice amount scalar that triggers the boundary blocking signal, obtain the initial projection transformation weights corresponding to each tax item node, and the minimum amount threshold and the maximum amount threshold; A quadratic objective function is constructed with the goal of minimizing the weight adjustment range, and a set of constraints is constructed by combining the equality constraint of the total weight and the inequality constraint of the tax red line interval. By introducing Lagrange multiplier vectors to transform into Lagrange functions, and based on the Caro-Kun-Tucker conditions, calling the quadratic programming algorithm to perform a finite-step iterative calculation, the optimal set of target weight values ​​is obtained. Perform a scalar multiplication operation between the total invoice amount scalar and the optimal target weight value set, and then perform financial precision truncation and tail difference compensation operations again to generate a corrected split invoice amount data set.

9. A telecommunications invoicing content matching system based on multi-dimensional business proportion analysis according to claim 1, characterized in that, The encapsulation of business and tax instructions and the encapsulation of electronic invoice generation modules encapsulate standard business and tax invoicing instructions, specifically including: Obtain basic tax information, using the tax item node code in the compliant release data or the corrected split invoice amount data set as an index, match and retrieve the statutory tax rate in the basic tax information, and map and bind the split amount, tax rate and the basic tax information containing the main accounting parameters. Based on the pre-defined specifications, the bound data is converted into a standard business and tax invoicing instruction that includes an invoice header and a commodity transaction details section; The data digest of the invoice instruction is calculated using a hash algorithm and a digital signature with an encrypted private key is attached. The digest is then sent to an external electronic invoice service platform via network communication to trigger the coding and generation process. When the receipt indicates that the invoice issuance failed, the error code is extracted and written to the exception retry queue.

10. A method for matching telecommunications invoice content based on multi-dimensional service proportion analysis, implemented based on the telecommunications invoice content matching system based on multi-dimensional service proportion analysis as described in any one of claims 1-9, characterized in that, Includes the following steps: Extract heterogeneous service consumption data from the underlying billing gateway and perform range normalization to generate a standard usage matrix; The overall activity scalar of the standard usage matrix is ​​evaluated to determine the performance status of the current billing cycle, and historical data is extracted and the historical baseline entropy weight vector is output when the performance status is determined to be dormant. Calculate the temporal information entropy of each business dimension, and generate the final dynamic entropy weight vector by combining it with the determined performance status; Obtain the total invoice amount scalar, and perform tensor scalar dot product projection along the direction of the final dynamic entropy weight vector to split it into each tax item node, generating a data set containing the final split invoice amount of each tax item node. An out-of-bounds check is performed on the final split invoice amount. When the check passes, compliant release data is output. When an out-of-bounds error occurs, an out-of-bounds blocking signal is generated. When the boundary blocking signal is received, the multidimensional tax base Lagrange constraint is invoked to perform a quadratic programming optimization calculation to generate a corrected set of split invoice amount data; The compliant release data or the corrected split invoice amount data set is encapsulated into a standard business and tax invoicing instruction with a digital signature, and sent to an external tax invoicing subsystem to generate an electronic invoice.

Images