A small and medium-sized enterprise business management whole-process digitalization supervision service system

Abstract

The application relates to the technical field of e-government data processing, and discloses a small and medium-sized enterprise industrial and commercial management whole-process digitalized supervision service system, which comprises a data mapping module, a benchmark calculation module, a feedforward modulation module, a reverse drive module and a closed-loop output module; multi-source data is separated into discrete state flow and continuous characteristic flow, an aggregated finite state machine is constructed for an enterprise, and topological conflict detection is performed; the benchmark calculation module extracts group characteristic flow to calculate a mean vector and a covariance matrix; when a legal state transition occurs, the feedforward modulation module truncates and resets a sliding window, and performs Hadamard product operation on the covariance matrix by using a compensation matrix to generate a modulation covariance matrix; the reverse drive module calculates Mahalanobis distance and deviation degree gradient in this way, and triggers an interruption to force a state transition and overwrite a mapping function when the Mahalanobis distance and the deviation degree gradient exceed a limit. The application absorbs legal business data fluctuation through feedforward matrix compensation, suppresses system data false alarm, and avoids normal approval processes from being wrongly blocked.

Description

Technical Field

[0001] This invention relates to the field of e-government data processing technology, specifically to a digital supervision service system for the entire process of business management of small and medium-sized enterprises. Background Technology

[0002] In the current e-government system architecture, the regulatory approach for the business administration of SMEs has gradually integrated data nodes from multiple departments. Current e-government regulatory systems typically acquire asynchronous data from multiple sources, including business registration, taxation, and social security. This data is obtained by extracting periodically submitted figures and registration information from enterprises, and by setting fixed statistical baselines and error tolerance ranges within the system, compliance comparisons are performed on the received data. According to conventional processing logic, when the calculated continuous indicator values ​​submitted by an enterprise exceed the preset static threshold range, the calculation module directly outputs an exception command, triggering the upper-level business blocking mechanism and suspending the relevant approval process.

[0003] In the course of their actual operations, SMEs may undergo legal administrative registration status migrations, such as relocation across administrative regions, changes in their main business category, or increases or decreases in registered capital. Such discrete status changes will inevitably cause short-term numerical fluctuations in related continuous characteristic data (such as monthly tax declaration amount, social security contribution base and number of people, etc.) during the transition period of business implementation. The existing data processing architecture exposes underlying technical defects when dealing with the above scenarios.

[0004] In the data parsing and detection stages, the existing system has severed the temporal coupling between discrete state flow and continuous feature flow. When a legitimate state transition occurs in an enterprise, it still uses a fixed historical time statistical window and a static distribution benchmark to calculate the data deviation at the current moment. Due to the lack of feedforward perception and underlying calculation matrix compensation mechanism for legitimate state changes, the system cannot adaptively converge to modulate reasonable data fluctuations during the transition period.

[0005] This single and static data verification logic directly leads the system to equate legitimate business data fluctuations with abnormal violations during the distance measurement and gradient calculation stages. The deviation calculation value is mechanically amplified, frequently exceeding the blocking threshold, and thus outputting a large number of false conflict alerts and generating invalid verification work orders. This causes normal enterprise status change requests to be forcibly suspended by the transaction lock at the system's underlying level, resulting in unnecessary stagnation of the government approval process. It also increases the system communication overhead for external nodes to intervene in manual due diligence and status reset. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a digital supervision service system for the entire process of business management for small and medium-sized enterprises. This system solves the problem that existing government supervision systems, when enterprises undergo a legal status migration, cannot adaptively absorb normal continuous business data fluctuations due to their reliance on static calculation benchmarks, resulting in frequent false anomaly alerts and erroneous blocking of normal government approval processes.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solution: This invention provides a digital supervision service system for the entire process of business management for small and medium-sized enterprises, comprising: The data mapping module is used to align multi-source data to generate discrete state streams and continuous feature streams, construct an aggregated finite state machine for the target enterprise, and use a joint state topology mapping function to perform topology conflict detection on the discrete state streams to obtain conflict detection results. The benchmark calculation module divides the target enterprise into peer groups and calculates the mean vector and covariance matrix based on the continuous feature flow of the peer groups within the sliding window. The feedforward modulation module truncates the sliding window when the aggregated finite state machine undergoes a legal state transition, extracts the compensation coefficient matrix, and multiplies it with the covariance matrix to obtain the modulation covariance matrix. The reverse driving module calculates the Mahalanobis distance of the feature vector using the mean vector and the modulation covariance matrix, extracts the deviation direction gradient, and when the deviation direction gradient is greater than the critical value, controls the joint state vector in the aggregated finite state machine to transition to the hidden inference state, and overwrites the joint state topology mapping function as a strict mode function to obtain the overwriting result. The closed-loop output module receives the conflict detection result and the overwrite result to generate a regulatory closed-loop work order.

[0008] Preferably, the data mapping module includes: The data stream parsing unit receives the enumerated state label vector and numerical business indicators returned by the nodes to which the multi-source data belongs, and converts them into the discrete state stream and the continuous feature stream according to the timestamp alignment; The state machine instantiation unit extracts the set of state nodes from the discrete state flow to construct the aggregated finite state machine; The topology conflict detection unit extracts the joint state vector of the discrete state flow at the current time and inputs it into the joint state topology mapping function to perform a validity comparison and generate the conflict detection result.

[0009] Preferably, the topology conflict detection unit extracts the target joint state vector when the discrete state flow experiences a state transition event, inputs the target joint state vector into the joint state topology mapping function to calculate the topology mapping value, and determines that a hard state concurrent conflict has occurred when the topology mapping value is zero and records the blocking flag in the conflict detection result.

[0010] Preferably, the benchmark calculation module includes: The group clustering unit extracts objective attribute vectors and inputs them into the clustering algorithm to classify the target enterprise into the peer group; The mean calculation unit accumulates the time mean vector of the continuous feature flow of all enterprises in the peer group within the sliding window and performs a division operation with the total number of enterprises to obtain the mean vector; The covariance calculation unit calculates the transpose product of the time mean vector and the difference vector of the mean vector, and accumulates the transpose product to obtain the covariance matrix.

[0011] Preferably, the group clustering unit extracts the industry code and registered capital value to generate the objective attribute vector, and the distance vector between the objective attribute vector and the preset cluster center is calculated by the distance clustering algorithm to obtain the peer group.

[0012] Preferably, the feedforward modulation module includes: The window truncation unit receives the sliding window initialized by the legal state transition truncation output by the aggregated finite state machine, and resets the start time of the sliding window to the occurrence time of the legal state transition; The matrix extraction unit retrieves a preset state compensation lookup table to extract the compensation coefficient matrix corresponding to the legal state transition; The product modulation unit calculates the Hadamard product of the compensation coefficient matrix and the covariance matrix to obtain the modulation covariance matrix.

[0013] Preferably, the product modulation unit extracts elements of the covariance matrix and the compensation coefficient matrix, performs element-wise multiplication of corresponding elements, and generates the modulation covariance matrix containing the multiplication results to replace the initialized covariance matrix input to the reverse drive module.

[0014] Preferably, the reverse drive module includes: The distance calculation unit extracts the mean value of continuous features in the current calculation cycle to generate the feature vector, and calculates the square root of the product of the feature vector, the mean vector and the inverse matrix of the modulation covariance matrix to obtain the Mahalanobis distance; The gradient calculation unit extracts the feature difference between adjacent calculation cycles, projects the magnitude of the change, and calculates the rate of change to obtain the deviation direction gradient. The state transition control unit determines that when the deviation direction gradient is greater than the critical value, it triggers an internal interrupt event to force the joint state vector to transition to the hidden deduced state. The function overwrite unit receives the trigger signal of the hidden derivation state and uses the strict mode function to replace the joint state topology mapping function to obtain the overwrite result.

[0015] Preferably, the function overwrite unit resets the output value of the preset high-risk change path in the strict mode function to an invalid value, extracts the currently received discrete state change request and inputs it into the strict mode function to calculate the function output result. When the function output result is the invalid value, the discrete state change request is blocked and written into the overwrite result.

[0016] Preferably, the closed-loop output module includes: The blocking output unit generates a level-one blocking verification work order to suspend the business approval flow when the conflict detection result contains a hard state concurrent conflict identifier. The due diligence output unit generates a secondary penetration due diligence work order to restrict the operation permissions of business terminals when the overwrite result contains an invalid value blocking flag. The state reset unit receives feedback signals and sends recovery instructions to the aggregated finite state machine to reload the initialized joint state topology mapping function.

[0017] This invention provides a digital supervision service system for the entire process of business management for small and medium-sized enterprises. It has the following beneficial effects: 1. This invention divides target enterprises into peer groups and uses continuous feature flow to calculate mean vectors and covariance matrices. Based on this, it calculates the Mahalanobis distance of the target enterprise feature vectors. In this way, it uses the relative data distribution characteristics of the same industry to replace the preset static absolute value threshold, eliminates the interference of industry cyclical fluctuations and differences in enterprise size on the indicator calculation results, and reduces the false alarm rate of data caused by uniform fixed thresholds.

[0018] 2. This invention uses a feedforward modulation module to truncate the sliding window when the aggregated finite state machine determines that a legitimate state transition has occurred, and uses the extracted compensation coefficient matrix to perform product modulation operation on the covariance matrix. When the system identifies that an enterprise has entered a legitimate business transition period, the tolerance for fluctuations in continuous business data during this period is relaxed by updating the matrix parameters, thus avoiding the system misjudging the normal decay phase of legitimate business data as data anomalies.

[0019] 3. This invention extracts the deviation direction gradient through the reverse driving module. When the gradient is greater than the critical value, it forces the underlying aggregated finite state machine to transition to the hidden inference state and overwrites the topology mapping function with a strict mode function. Then, it converts the calculation results of continuous values ​​into the interruption intervention instruction of the state machine. Through the strict mode function, the output value of the preset high-risk state change path is directly set to an invalid value, thereby realizing the closed-loop control of blocking the business front-end state change request by using the data deviation calculation result. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the system architecture of the present invention; Figure 2 This is a comparison curve of continuous characteristic deviation data during enterprise state transitions according to the present invention. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Reference Figure 1 This invention provides a digital supervision service system for the entire process of business management of small and medium-sized enterprises, which may include: a data mapping module, a benchmark calculation module, a feedforward modulation module, a reverse driving module, and a closed-loop output module.

[0023] The regulatory service system is deployed in a computing device that includes a processor, memory, and communication interface. The memory stores computer programs, and the processor is connected to the memory. The data processing logic of each module in the regulatory service system is realized by calling and executing the computer programs.

[0024] The input of the data mapping module is connected to the communication interface to acquire multi-source data and align it according to the time base. The multi-source data is separated and converted into discrete state streams and continuous feature streams. The data mapping module instantiates an aggregated finite state machine for the target enterprise in memory, performs topological conflict detection on the discrete state stream using the joint state topology mapping function, and transmits the generated conflict detection results to the closed-loop output module.

[0025] The benchmark calculation module and the data mapping module are connected via their data output ports. The benchmark calculation module extracts the objective attribute parameters of the target enterprise and categorizes it into peer groups. It then extracts the continuous feature streams of all enterprises within each peer group within a specified sliding window. Based on the extracted feature stream data, the benchmark calculation module performs matrix operations and outputs the mean vector and covariance matrix.

[0026] The feedforward modulation module is connected to the state event output port of the data mapping module and the matrix output port of the benchmark calculation module, respectively. When the aggregated finite state machine undergoes a set legal state transition, the feedforward modulation module receives a trigger signal and performs a sliding window truncation and reset operation. The feedforward modulation module extracts the corresponding compensation coefficient matrix and performs a Hadamard product operation with the received covariance matrix to output the modulation covariance matrix.

[0027] The inputs of the reverse-drive module receive the mean vector, the modulation covariance matrix, and the continuous feature stream of the current period. The reverse-drive module uses the aforementioned input data to calculate the Mahalanobis distance of the feature vectors and extracts and numerically judges the directional gradient of the deviation on the time series.

[0028] When the deviation gradient is determined to be greater than a preset critical value, the reverse drive module sends an interrupt intervention signal to the data mapping module via the control bus. This signal forces the aggregated finite state machine to undergo a state transition into a hidden derivation state, and uses a strict pattern function to overwrite the existing joint state topology mapping function, generating the corresponding overwrite result.

[0029] The receiving end of the closed-loop output module is connected to both the data mapping module and the reverse drive module to receive conflict detection results and overwrite results. Based on a preset rule set, the closed-loop output module parses the received results, generates a corresponding level of regulatory closed-loop work order, and outputs it to the business terminal via a communication interface.

[0030] Reference Figure 1 The data mapping module includes a data flow parsing unit, a state machine instantiation unit, and a topology conflict detection unit.

[0031] The data stream parsing unit receives data packets returned by the nodes to which the multi-source data belongs through the communication interface, and extracts the timestamp and unique identification code from the header of the data packet. Using the unified social credit code as the unique identification code and the timestamp as the time reference, the data stream parsing unit performs time-series alignment operations on the received asynchronous multi-source data.

[0032] After alignment, the data flow parsing unit categorizes and parses the data content, targeting the specific enterprise. At the time point The data stream parsing unit generates a discrete state stream from the received enumerated state data. Its vector form is defined as: ; In the formula, Represents discrete state flow; Representing the data source of industry and commerce The discrete enumerated states returned at each time step; Represents the source of tax data. The discrete enumerated states returned at each time step; Represents the data source of social security. The discrete enumerated state returned at each time step. Transpose symbol. This indicates that the status parameters of each department are organized into a column vector.

[0033] For this target company, the data stream parsing unit extracts periodically reported numerical business indicators and generates a continuous feature stream. Its vector form is defined as: ; In the formula, Represents a continuous characteristic flow; The total number of dimensions of the extracted continuous features; to Representing the first Wei Zhi Di Continuous feature values ​​of dimension; This indicates that the continuous feature flow vector belongs to 3D real space; transpose symbol This indicates that the status parameters of each department are organized into a column vector. The symbol indicates that the feature flow vector generated on the left belongs to the set of spaces indicated on the right.

[0034] The state machine instantiation unit receives the parsed discrete state stream. It allocates independent computational space for the target enterprise in system memory to construct an aggregate finite state machine. The single-system state space of the business system is defined as follows: The single-system state space of the tax system is The single-system state space of the social security system is The state machine instantiation unit extracts the set of state nodes from the discrete state flow, and integrates the state spaces of the single system through Cartesian product operations to generate a joint state space containing three-dimensional state variables.

[0035] The topology conflict detection unit is connected to the state machine instantiation unit. The topology conflict detection unit extracts the joint state vector of the discrete state flow at the current time step and inputs it into the joint state topology mapping function to perform a validity comparison. The joint state topology mapping function is configured as follows: ; In the formula, Represents the joint state topology mapping function; This represents one of the single-system state spaces that constitute the input parameters of the joint state space (as can be seen from the previous text, it is the single-system state space of the industrial and commercial system). This represents one of the single-system state spaces that constitute the input parameters of the joint state space (as can be seen from the previous text, it is the single-system state space of the tax system). This represents one of the single-system state spaces constituting the input parameters of the joint state space (as mentioned earlier, it is the single-system state space of the social security system). This state topology mapping function receives input parameters from... , , The input parameters constitute the joint state space, and the output set is... Output the binary numerical result. This indicates that the currently received cross-departmental joint state combination is valid; output the numerical value. This indicates that there is a temporal topology conflict in the currently received cross-departmental joint state combination.

[0036] When the data stream parsing unit detects a state transition event in the discrete state stream, it sends a trigger detection signal to the topology collision detection unit. The topology collision detection unit then acquires the target joint state vector at the time of the state transition event. ,in This is the timestamp of the state transition event.

[0037] The topology collision detection unit will combine the target joint state vector Substitute the values ​​into the joint state topology mapping function for calculation. When the function output satisfies... At that time, the topology conflict detection unit determines that a hard state concurrent conflict has occurred.

[0038] After a conflict is determined to be established, the topology conflict detection unit generates a conflict detection result data packet, writes a blocking flag in the field of the data packet, and then transmits the conflict detection result to the system bus, waiting for the closed-loop output module to receive it.

[0039] Reference Figure 1 The benchmark calculation module includes a group clustering unit, a mean calculation unit, and a covariance calculation unit.

[0040] The clustering unit connects to the system database and reads the static registration information of the target enterprise. The clustering unit extracts the industry code and registered capital value from the information, converts the industry code into a numerical classification label, and combines it with the registered capital value to generate an objective attribute vector.

[0041] The system memory contains pre-set coordinates of multiple cluster centers generated through training with historical full data. The group clustering unit calls a clustering algorithm to calculate the Euclidean distance between the objective attribute vector and each pre-set cluster center coordinate, generating a distance vector. The group clustering unit extracts the minimum value from the distance vectors and assigns the target enterprise to the peer group represented by the cluster center corresponding to that minimum value.

[0042] After receiving the group partitioning results, the mean calculation unit sets a sliding window. Sliding window Defined as ,in For the current calculation time, The set time span for tracing historical data.

[0043] The mean calculation unit extracts all companies within the peer group from the data bus in a sliding window. The system uses continuous feature stream data. For each enterprise, the mean calculation unit calculates the average value over time for each feature dimension within the sliding window, generating a time mean vector. Subsequently, the mean calculation unit accumulates the time mean vectors of all enterprises in the peer group and performs a division operation with the total number of enterprises to obtain the final mean vector. The mean calculation unit then executes the mean vector calculation. The calculation formula is as follows: ; In the formula, This indicates the result calculated for a peer group. 3D mean vector; This indicates the peer group into which the target company has been classified; This indicates the total number of companies included in the peer group; Representative set Individual businesses within; Indicates enterprise In the sliding window The time-mean vector of the continuous feature stream generated internally; The symbol indicates that the feature flow vector generated on the left belongs to the set of spaces indicated on the right.

[0044] The covariance calculation unit is connected to the mean calculation unit and receives the generated mean vector. The covariance calculation unit extracts the time mean vector for each enterprise and performs matrix subtraction operations with the mean vector to obtain the corresponding difference vector.

[0045] The covariance calculation unit multiplies each difference vector with its own transpose to generate the covariance bias matrix for the corresponding firm. Then, the covariance calculation unit sums the covariance bias matrices of all firms within the peer group and divides them by the degree of freedom parameter to calculate the final covariance matrix. The covariance calculation unit performs the covariance matrix calculation... The calculation formula is as follows: ; In the formula, The covariance matrix representing the distribution of continuous characteristic flows within a peer group; The sample degrees of freedom constant under unbiased estimation; For enterprises The difference vector between the time mean vector and the peer group mean vector; transpose sign This indicates that the status parameters of each department are organized into a column vector. Representative set Individual businesses within; The symbol indicates that the feature flow vector generated on the left belongs to the set of spaces indicated on the right.

[0046] The calculated mean vector and covariance matrix are cached in the system memory address range, serving as a baseline for the distribution characteristics of the peer group within the current sliding window period, and providing numerical inputs to the feedforward modulation module and the reverse drive module.

[0047] Reference Figure 1 The feedforward modulation module includes a window truncation unit, a matrix extraction unit, and a product modulation unit.

[0048] The control port of the window truncation unit is connected to the event output bus of the aggregated finite state machine. When the aggregated finite state machine determines that a valid state transition has occurred in the discrete state flow that satisfies the joint state topology mapping function, the window truncation unit receives a trigger signal containing the timestamp of the event.

[0049] Let the timestamp of the legal state transition be . The window truncation unit performs a time-axis truncation operation on the initial sliding window set by the baseline calculation module. The window truncation unit resets the starting time parameter of the sliding window from its original time starting point to... Generate the updated sliding window range This reset operation changes the statistical interval of the continuous characteristic indicator to the time period after the event occurs and outputs it to the system time control bus.

[0050] The addressing port of the matrix extraction unit is connected to the system memory, which is pre-written with a state compensation lookup table. The state compensation lookup table stores the mapping relationship between various legal state transition identifiers and matrix index data.

[0051] The matrix extraction unit parses the received identification codes of legitimate state transitions and uses them as search keywords to perform a matching query on the state compensation lookup table. Based on the addressing results returned by the query, the matrix extraction unit retrieves the corresponding compensation coefficient matrix from the system memory. The data dimension of this compensation coefficient matrix is ​​consistent with the total number of feature dimensions of the continuous feature stream. 3D numerical matrix.

[0052] The data input port of the product modulation unit is connected to the data output port of the matrix extraction unit and the matrix output port of the benchmark calculation module, respectively. The product modulation unit synchronously receives the generated compensation coefficient matrix and the initialized covariance matrix.

[0053] The product modulation unit extracts the coordinate element data within the covariance matrix and the element data at the same row and column coordinate positions in the compensation coefficient matrix. The product modulation unit performs an item-by-item multiplication operation on the element data at the corresponding coordinate positions to generate a modulation covariance matrix containing the multiplication results. The calculation formula for this modulation operation performed by the product modulation unit is as follows: ; In the formula, This indicates that the value at position i in the calculated modulation covariance matrix is... row and number The value of the element at the intersection of column coordinates; This indicates that the initial covariance matrix input from the benchmark calculation module contains elements at position i. row and number The value of the element at the intersection of column coordinates; This indicates that the extracted compensation coefficient matrix is ​​located at the th position. row and number The element value at the intersection of column coordinates; operators This indicates that the Hadamard product operation is performed on the elements of two matrices of the same dimension.

[0054] The product modulation unit traverses all row and column coordinates of the matrix according to the above formula to obtain the complete modulation covariance matrix. Subsequently, the product modulation unit writes the modulation covariance matrix into the system operation cache address, replacing the original covariance matrix data of the system, and providing the feedforward modulated matrix reference input parameters for the subsequent reverse drive module.

[0055] Reference Figure 1 The reverse drive module includes a distance calculation unit, a gradient calculation unit, a state transition control unit, and a function overwrite unit.

[0056] The input of the distance calculation unit is connected to the output of the feedforward modulation module and the baseline calculation module. The distance calculation unit receives the continuous feature stream data of the target enterprise in the current calculation period, calculates the time mean within that period, and generates a feature vector.

[0057] The distance calculation unit retrieves the mean vector and modulation covariance matrix from the system memory. It performs a matrix inversion operation on the modulation covariance matrix to obtain the inverse modulation covariance matrix. The distance calculation unit then calculates the difference vector between the eigenvector and the mean vector. It performs a product operation using the difference vector, the inverse modulation covariance matrix, and the transpose of the difference vector, and calculates the square root of the product to obtain the Mahalanobis distance. The formula for calculating the Mahalanobis distance is as follows: ; In the formula, This represents the calculated Mahalanobis distance value; Indicates the number of times data is constructed within the current computation cycle. 3D feature vector; This indicates the result calculated for a peer group. 3D mean vector; This represents the inverse matrix of the modulation covariance matrix input from the feedforward modulation module; transpose symbol. This indicates that the status parameters of each department are organized into a column vector.

[0058] The gradient calculation unit is connected to the distance calculation unit. The gradient calculation unit extracts the difference between the feature vector and the mean vector of the current calculation cycle, as well as the difference between the feature vector and the mean vector of the previous calculation cycle. The gradient calculation unit calls the system's preset weight vector and calculates the rate of change of the magnitude of the above difference data along the projection direction defined by the weight vector, generating the deviation direction gradient. The gradient calculation unit executes the following approximate calculation formula for the deviation direction gradient: ; In the formula, This represents the calculated deviation direction gradient; and These represent the feature vector and mean vector at the current time, respectively; and These represent the feature vector and mean vector at the previous calculation time, respectively; This represents a pre-defined weight vector that reflects the sensitivity of the indicators. This indicates that the Hadamard product operation is performed; This represents the magnitude of the calculated vector. This represents the time increment between two adjacent calculation cycles.

[0059] The state transition control unit receives the deviation direction gradient and compares it with a system-defined threshold. When the deviation direction gradient is greater than the threshold, it indicates that the change in the continuous data distribution along the preset weight direction exceeds the threshold range. The state transition control unit generates an internal interrupt event and sends it to the data mapping module via the system control bus. This instruction forces the underlying aggregated finite state machine to modify the current state parameters of the target enterprise, jumping the address pointer of the joint state vector to the preset hidden derivation state.

[0060] The function overriding unit receives a trigger signal indicating the completion of a hidden inference state transition. The function overriding unit retrieves a strict mode function from memory, which pre-sets the output values ​​of high-risk change paths to invalid values. The function overriding unit then replaces the existing joint state topology mapping function in system memory with this strict mode function.

[0061] When the communication interface receives a discrete state change request for the target enterprise, the function overriding unit extracts the target discrete state parameters from the request and inputs them into the strict mode function to calculate the function output. When the function output is determined to be invalid, the function overriding unit blocks the discrete state change request from entering the state change queue of the aggregated finite state machine at the underlying level, and compiles the request content, triggering conditions, and timestamp information of this blocking operation into an overriding result data packet, which is then sent to the closed-loop output module.

[0062] Reference Figure 1 The closed-loop output module includes a blocking output unit, a due diligence output unit, and a state reset unit.

[0063] The blocking output unit connects its data receiving port to the data mapping module to obtain the conflict detection result data packets generated by the module. The blocking output unit extracts the status bit field from the data packets and performs bitmap mask matching operations. When a hard-state concurrent conflict identifier is parsed, the blocking output unit extracts the unique identification code of the target enterprise that triggered the conflict and the timestamp when the conflict was triggered.

[0064] The blocking output unit encapsulates the extracted code and timestamp information to generate a Level 1 blocking verification work order. This Level 1 blocking verification work order contains a transaction suspension instruction message. The blocking output unit pushes the Level 1 blocking verification work order to the corresponding government business approval node through the system communication interface, triggering the database transaction lock mechanism of that node, so that the process data of the current business approval flow involved remains in a read-only state, stopping subsequent write operations.

[0065] The due diligence output unit is connected to the data output bus of the reverse drive module and receives the overwrite result data packet generated by the reverse drive module. The due diligence output unit decodes the overwrite result data packet and extracts the invalid value blocking flag and the type code of the associated discrete state change request.

[0066] The due diligence output unit retrieves the extracted request type code from a pre-defined permission mapping table, generating a second-level penetration due diligence work order containing a permission mask matrix. After this second-level penetration due diligence work order is sent to the front-end business terminal, the system restricts the business terminal's operation permissions based on the mask data embedded in the work order. The formula for calculating the restriction of business terminal operation permissions is as follows: ; In the formula, This represents the calculated and restricted operation permission vector for the business terminal. This represents the vector of original business operation permissions possessed by the target enterprise before receiving the work order; This represents the binary permission mask matrix consisting of 0s and 1s extracted from the second-level penetration due diligence work order; operator symbols This indicates that the Hadamard product operation is performed on the elements of two matrices of the same dimension. The specific business operation entry point corresponding to the dimension in the permission vector that results in 0 is then blocked by the system.

[0067] The external signal receiving port of the status reset unit is connected to the system communication interface. After the external verification system completes the entity verification operation, it returns a feedback signal containing a digital signature and a transaction release identifier to the system communication interface.

[0068] The state reset unit receives and verifies the digital signature of the feedback signal. Upon successful verification, the state reset unit generates a recovery instruction and sends it to the control layer of the aggregated finite state machine. This recovery instruction triggers the aggregated finite state machine to release the memory space pointer lock operation for the hidden deduced state.

[0069] After releasing the lock, the state reset unit sends an overwrite revocation command to the function scheduling bus, clearing the existing strict mode function configuration parameters. The system function addressing pointer is redirected to the memory address of the initialized joint state topology mapping function, and the multi-source discrete state detection logic in standard mode is reloaded.

[0070] Specific application examples: Reference Figure 2 , Figure 2 This is a comparison curve of the conventional static threshold method according to an embodiment of the present invention and the continuous characteristic deviation data of the present invention during enterprise state transition. Figure 2 In the figure, the horizontal axis represents the calculation time period in months, with the statistical interval from the 1st to the 12th month; the vertical axis represents the dimensionless deviation detection value of the continuous feature flow in the calculation output.

[0071] In a specific application example, the target enterprise is a specialized equipment manufacturing company in a period of expansion. The regulatory service system captures the company's application for relocation of its production base across administrative regions during the sixth calculation cycle (month). In the multi-source data mapping stage, the data mapping module receives three discrete enumerated states through the data flow parsing unit: change of registered address from the industrial and commercial system, tax registration relocation / relocation from the tax system, and change of social security pooling entity from the social security system. The aggregated finite state machine constructed by the state machine instantiation unit receives the above inputs, and the topology conflict detection unit calls the joint state topology mapping function to detect and determine that the joint state combination is a legitimate state transition and does not trigger a hard state concurrent conflict.

[0072] In the benchmark calculation stage, the benchmark calculation module classifies the enterprise and other equipment manufacturing enterprises with the same registered capital in the same region into the same peer group. The mean calculation unit and covariance calculation unit extract the continuous feature streams of the group within a set sliding window, such as business revenue, tax declaration amount, and social security contribution base, and calculate and output the mean vector and covariance matrix.

[0073] In response to this cross-regional relocation event, the company's continuous feature flows (such as tax payments and the number of employees contributing to social security) will experience short-term fluctuations consistent with business logic during the transition period (months 6 to 8). At this time, the feedforward modulation module receives a trigger signal from the data mapping module, and the window truncation unit resets the starting time of the sliding window to month 6. The matrix extraction unit extracts the compensation coefficient matrix corresponding to the cross-regional relocation from the state compensation lookup table, and the product modulation unit performs a Hadamard product operation on this compensation coefficient matrix and the initialized covariance matrix to generate a modulation covariance matrix. This operation relaxes the system's sensitivity restrictions to feature dimension numerical fluctuations during the transition period at the matrix level.

[0074] The reverse drive module receives the aforementioned modulation covariance matrix, and the distance calculation unit calculates the Mahalanobis distance of the enterprise's feature vector based on this. Since the covariance matrix has been feedforward compensated, the calculated Mahalanobis distance value remains within a reasonable range, and the deviation direction gradient does not exceed the preset critical value. The state transition control unit does not trigger the hidden inference state transition, and the closed-loop output module does not generate a first-level or second-level blocking work order. The enterprise's normal cross-regional relocation business process can be executed smoothly, avoiding system false alarms caused by drastic numerical fluctuations.

[0075] To verify the technical effectiveness of the regulatory service system in controlling false alarms and identifying anomalies, objective continuous feature stream data of the target enterprise over 12 consecutive months from the above specific application examples were extracted for comparative testing. The tests employed both the traditional static threshold determination method relying on fixed statistical benchmarks and the dynamic collaborative detection method of this invention to calculate the deviation (Mahavira distance metric) values ​​in parallel, setting the deviation threshold for determining data anomalies at 8.0.

[0076] Reference Figure 2 The experimental comparison data curves shown have the horizontal axis representing the time period (months) and the vertical axis representing the dimensionless deviation detection value calculated by the system.

[0077] During the regular operating period from month 1 to month 5, the deviation values ​​calculated by the two methods were basically consistent, fluctuating between 2.0 and 3.5, indicating that when the enterprise is in a stable state, the basic monitoring capabilities of the two methods are consistent.

[0078] In the sixth month, the target company initiated a legitimate status migration (inter-regional relocation), causing objective fluctuations in its continuous characteristic stream data. The deviation calculation results using the traditional static threshold method rapidly increased from the sixth month onwards, reaching 12.1, 13.5, and 10.8 in the sixth, seventh, and eighth months respectively, breaching the threshold of 8.0. In actual business operations, this would lead to the approval system being incorrectly locked, generating invalid false alarms.

[0079] The dynamic collaborative detection method of this invention, due to the synchronous triggering of matrix compensation and sliding window reset of the feedforward modulation module in the 6th month, resulted in the deviation detection values ​​being converged and modulated to 5.8, 6.4, and 5.5 in the 6th, 7th, and 8th months, respectively. Although these values ​​objectively reflect data fluctuations, the overall deviation was controlled within the preset threshold of 8.0 due to the introduction of a compensation mechanism. Entering the 9th month, with the completion of the target company's relocation, the two sets of curves converged again and returned to the normal range.

[0080] The comparative experimental data verified that the closed-loop computing architecture of this invention can adaptively absorb the fluctuations of legitimate business data through feedforward modulation while ensuring the legitimacy of the underlying topology logic. This eliminates the problem of false alarms caused by the solidification of static rules without reducing the overall regulatory sensitivity of the system.

Claims

1. A digital supervision and service system for the entire process of business management of small and medium-sized enterprises, characterized in that, include: The data mapping module is used to align multi-source data to generate discrete state streams and continuous feature streams, construct an aggregated finite state machine for the target enterprise, and use a joint state topology mapping function to perform topology conflict detection on the discrete state streams to obtain conflict detection results. The benchmark calculation module divides the target enterprise into peer groups and calculates the mean vector and covariance matrix based on the continuous feature flow of the peer groups within the sliding window. The feedforward modulation module truncates the sliding window when the aggregated finite state machine undergoes a legal state transition, extracts the compensation coefficient matrix, and multiplies it with the covariance matrix to obtain the modulation covariance matrix. The reverse driving module calculates the Mahalanobis distance of the feature vector using the mean vector and the modulation covariance matrix, extracts the deviation direction gradient, and when the deviation direction gradient is greater than the critical value, controls the joint state vector in the aggregated finite state machine to transition to the hidden inference state, and overwrites the joint state topology mapping function as a strict mode function to obtain the overwriting result. The closed-loop output module receives the conflict detection result and the overwrite result to generate a regulatory closed-loop work order.

2. The digital supervision and service system for the entire process of business administration of small and medium-sized enterprises as described in claim 1, characterized in that, The data mapping module includes: The data stream parsing unit receives the enumerated state label vector and numerical business indicators returned by the nodes to which the multi-source data belongs, and converts them into the discrete state stream and the continuous feature stream according to the timestamp alignment; The state machine instantiation unit extracts the set of state nodes from the discrete state flow to construct the aggregated finite state machine; The topology conflict detection unit extracts the joint state vector of the discrete state flow at the current time and inputs it into the joint state topology mapping function to perform a validity comparison and generate the conflict detection result.

3. The digital supervision and service system for the entire process of business administration of small and medium-sized enterprises as described in claim 2, characterized in that, The topology conflict detection unit extracts the target joint state vector when the discrete state flow experiences a state transition event, inputs the target joint state vector into the joint state topology mapping function to calculate the topology mapping value, and determines that a hard state concurrent conflict has occurred when the topology mapping value is zero and records the blocking flag in the conflict detection result.

4. The digital supervision and service system for the entire process of business administration of small and medium-sized enterprises as described in claim 1, characterized in that, The benchmark calculation module includes: The group clustering unit extracts objective attribute vectors and inputs them into the clustering algorithm to classify the target enterprise into the peer group; The mean calculation unit accumulates the time mean vector of the continuous feature flow of all enterprises in the peer group within the sliding window and performs a division operation with the total number of enterprises to obtain the mean vector; The covariance calculation unit calculates the transpose product of the time mean vector and the difference vector of the mean vector, and accumulates the transpose product to obtain the covariance matrix.

5. The digital supervision and service system for the entire process of business administration of small and medium-sized enterprises as described in claim 4, characterized in that, The group clustering unit extracts the industry code and registered capital value to generate the objective attribute vector, and the distance vector between the objective attribute vector and the preset cluster center is calculated by the distance clustering algorithm to obtain the peer group.

6. The digital supervision and service system for the entire process of business administration of small and medium-sized enterprises as described in claim 1, characterized in that, The feedforward modulation module includes: The window truncation unit receives the sliding window initialized by the legal state transition truncation output by the aggregated finite state machine, and resets the start time of the sliding window to the occurrence time of the legal state transition; The matrix extraction unit retrieves a preset state compensation lookup table to extract the compensation coefficient matrix corresponding to the legal state transition; The product modulation unit calculates the Hadamard product of the compensation coefficient matrix and the covariance matrix to obtain the modulation covariance matrix.

7. The digital supervision and service system for the entire process of business administration of small and medium-sized enterprises as described in claim 6, characterized in that, The product modulation unit extracts the elements of the covariance matrix and the elements of the compensation coefficient matrix, performs element-wise multiplication on the corresponding elements, and generates the modulation covariance matrix containing the multiplication results to replace the initialized covariance matrix input to the reverse drive module.

8. The digital supervision and service system for the entire process of business administration of small and medium-sized enterprises as described in claim 1, characterized in that, The reverse drive module includes: The distance calculation unit extracts the mean value of continuous features in the current calculation cycle to generate the feature vector, and calculates the square root of the product of the feature vector, the mean vector and the inverse matrix of the modulation covariance matrix to obtain the Mahalanobis distance; The gradient calculation unit extracts the feature difference between adjacent calculation cycles, projects the magnitude of the change, and calculates the rate of change to obtain the deviation direction gradient. The state transition control unit determines that when the deviation direction gradient is greater than the critical value, it triggers an internal interrupt event to force the joint state vector to transition to the hidden deduced state. The function overwrite unit receives the trigger signal of the hidden derivation state and uses the strict mode function to replace the joint state topology mapping function to obtain the overwrite result.

9. The digital supervision and service system for the entire process of business administration of small and medium-sized enterprises as described in claim 8, characterized in that, The function overwrite unit resets the output value of the preset high-risk change path in the strict mode function to an invalid value, extracts the currently received discrete state change request and inputs it into the strict mode function to calculate the function output result. When the function output result is the invalid value, the discrete state change request is blocked and written into the overwrite result.

10. The digital supervision and service system for the entire process of business administration of small and medium-sized enterprises according to claim 1, characterized in that, The closed-loop output module includes: The blocking output unit generates a level-one blocking verification work order to suspend the business approval flow when the conflict detection result contains a hard state concurrent conflict identifier. The due diligence output unit generates a secondary penetration due diligence work order to restrict the operation permissions of business terminals when the overwrite result contains an invalid value blocking flag. The state reset unit receives feedback signals and sends recovery instructions to the aggregated finite state machine to reload the initialized joint state topology mapping function.

Images