A procurement compliance intelligent auditing system based on multi-source data fusion

Abstract

The application provides a procurement compliance intelligent auditing system based on multi-source data fusion, and relates to the technical field of data processing, comprising: an acquisition module for acquiring logistics waybill data, warehouse delivery data and financial reconciliation data of a target procurement business to obtain multi-source business data; a fusion module for standardizing and correlating the multi-source business data to obtain fusion auditing records; according to weight difference and freight difference in the fusion auditing records, weight difference characteristic values and freight difference characteristic values are obtained; a positioning module for obtaining a target auditing point according to the positions of the weight difference characteristic values and the freight difference characteristic values in an auditing characteristic coordinate system, and obtaining a target reference area according to the coordinate range where the target auditing point is located. The application realizes automatic hierarchical auditing of procurement business compliance, and improves the consistency and auditing efficiency of the auditing results.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an intelligent audit system for procurement compliance based on multi-source data fusion. Background Technology

[0002] In procurement management, compliance auditing is a crucial step in ensuring the standardized operation of the procurement process. Typically, procurement involves multiple business processes, such as logistics, warehousing, and financial reconciliation. Each process generates corresponding business data. Currently, auditors often need to manually compare and verify the data from different processes to determine if there are any anomalies in the procurement process.

[0003] However, due to differences in data sources and recording methods across different business processes, data formats and field standards may not be entirely consistent across systems. This leads to auditors needing to spend considerable time organizing and verifying data when comparing data across different processes. For example, in a procurement transaction involving bulk materials, there may be a difference between the weight of goods recorded in the logistics waybill and the weight recorded in the warehouse delivery note. Simultaneously, there may be discrepancies between the freight amount in the logistics waybill and the settlement amount in the financial reconciliation data. Faced with such situations, auditors may need to rely on their experience to judge the weight and freight differences separately to determine whether the procurement transaction needs further manual review. However, this approach relies to some extent on the auditor's personal experience, and different auditors may have different understandings and judgment standards regarding the degree of difference, potentially leading to less than ideal consistency in the audit results. Summary of the Invention

[0004] This invention provides an intelligent audit system for procurement compliance based on multi-source data fusion, which enables automated hierarchical audit of procurement compliance, improving the consistency and efficiency of audit results.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a procurement compliance intelligent auditing system based on multi-source data fusion includes: The acquisition module is used to acquire logistics waybill data, warehouse outbound data, and financial reconciliation data of the target procurement business to obtain multi-source business data; The fusion module is used to standardize and correlate multi-source business data to obtain fusion audit records; based on the weight difference and freight difference in the fusion audit records, the weight difference feature value and freight difference feature value are obtained. The positioning module is used to obtain the target audit point based on the position of the weight difference feature value and the freight difference feature value in the audit feature coordinate system, and to obtain the target reference area based on the coordinate range of the target audit point. The analysis module is used to obtain the compliance benchmark line based on the first and second compliance benchmark points corresponding to the target benchmark area; and to obtain the vertical distance and positioning value along the line based on the positional relationship between the target audit point and the compliance benchmark line. The adjustment module is used to determine the vertical distance as the baseline deviation value and the positioning value along the line as the column positioning value. Based on the value range of the two values, the row index and column index are obtained. The two-dimensional mapping table is queried based on the row index and column index to obtain the candidate review level. The execution module is used to adjust the candidate review level based on the continuous comparison results between the benchmark deviation value and the fixed distance value to obtain the target review level; generate review processing instructions based on the target review level, control the approval process nodes to execute the corresponding review processing, and obtain the intelligent review result of the procurement compliance of the target procurement business.

[0006] The above-described solution of the present invention has at least the following beneficial effects: This system achieves deep integration and standardized processing of multi-source procurement business data, improving the integrity and usability of data association. Simultaneously, it constructs an audit feature coordinate system based on weight and freight differences, and calculates vertical deviation and along-line positioning values ​​using a compliance benchmark line. This transforms the difference features into quantifiable two-dimensional spatial location information, avoiding the limitations of single threshold judgments. Furthermore, it dynamically adjusts the audit level through two-dimensional mapping table queries and continuous comparison of fixed distance values. This not only improves the flexibility and rationality of audit grading results but also achieves automated linkage between audit decisions and approval process nodes, enhancing the intelligence level, response speed, and processing accuracy of procurement compliance audits. Attached Figure Description

[0007] Figure 1 This is a flowchart illustrating an intelligent audit system for procurement compliance based on multi-source data fusion, provided by an embodiment of the present invention.

[0008] Figure 2 This is a schematic diagram of the target reference area provided in an embodiment of the present invention.

[0009] Figure 3 This is a schematic diagram of the vertical distance and line positioning values ​​provided in an embodiment of the present invention. Detailed Implementation

[0010] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0011] To make the purpose, technical solution, and technical effects of this application clearer, the following description, in conjunction with specific implementation methods, further explains the intelligent audit system for procurement compliance based on multi-source data fusion. The described implementation methods are used to explain the data sources, processing order, calculation relationships, state changes, and abnormal branches of each module in the system. The technical features in each implementation method can be combined with each other when there is no logical conflict.

[0012] In this implementation, "first" and "second" are used to distinguish objects with similar names and do not necessarily limit the order of processing or importance; "multiple" means two or more; "and / or" means any one of the connected objects or a combination of at least two objects; "in response to" means that the processing corresponding to the condition is executed after the corresponding condition is met; "audit period" means the time range within which the same set of regional boundaries, benchmarks, value ranges, two-dimensional mapping tables, fixed distance values, and audit level order in the audit rule base remain valid. The same version of rule data is used within the same audit period to avoid continuous audit records losing comparability due to rule version changes.

[0013] The target procurement transaction is the one that requires compliance review. The logistics waybill data, warehouse outbound data, and financial reconciliation data represent the data records generated during the logistics transportation, warehouse outbound, and financial reconciliation stages of this procurement transaction, respectively. The multi-source business data is not a simple collection of concatenated files, but a collection of records to be processed that retains data source identifiers, business identifiers, and field semantics. The integrated review record is the review data carrier formed by standardizing the fields, unifying the units of measurement, unifying the currency, and associating the business data with the multi-source business data. It is used to provide data that can be compared with each other for subsequent difference calculations.

[0014] The weight difference feature value is used to characterize the degree of deviation between the recorded weight and the billing weight and the weight difference benchmark value. The freight difference feature value is used to characterize the degree of deviation between the benchmark invoice freight and the benchmark actual freight and the freight difference benchmark value. Both feature values ​​are obtained by subtracting data of the same dimension and normalizing them with a non-zero benchmark value of the same dimension. Therefore, the two feature values ​​obtained are non-negative and dimensionless data, which can be used as the horizontal axis and vertical axis coordinates of the audit feature coordinate system, respectively.

[0015] The target audit point represents the joint position of the weight difference feature value and the freight difference feature value in the audit feature coordinate system; the target reference area represents the candidate reference area containing the target audit point in the area boundary data; the compliance reference line represents the reference line determined by the first compliance reference point and the second compliance reference point corresponding to the target reference area; the vertical distance represents the unsigned distance from the target audit point to the compliance reference line; the positioning value along the line represents the direction and position of the projection of the target audit point in the reference direction relative to the first compliance reference point; since both coordinate axes of the audit feature coordinate system use dimensionless feature values, the vertical distance calculated based on the two coordinate axes is also a dimensionless quantity, and this vertical distance can be directly determined as the reference deviation value.

[0016] The two-dimensional mapping table stores the correspondence between deviation value ranges, positioning value ranges, and audit levels; the candidate audit level is obtained by querying the two-dimensional mapping table according to the row and column indexes formed by a single audit; the target audit level is the final level determined based on the candidate audit levels, combined with the comparison results of the benchmark deviation value of a consecutive preset number of times and the distance to a fixed value; the audit processing instruction is the process control data corresponding to the target audit level or abnormal result, used to control the execution of compliance release, compliance warning, or pending verification processing at the approval process nodes; the intelligent audit result for procurement compliance is the output of this audit formed by integrating audit records, target audit level, audit processing instructions, and audit execution status.

[0017] The intelligent auditing described in this system is characterized by its ability to automatically complete multi-source record association, differential feature extraction, two-dimensional location determination, geometric deviation analysis, mapping grading, continuous state correction, and approval process control within a unified data processing chain. This processing does not rely on auditors temporarily selecting calculation methods. The audit rule base provides each module with coordinate system configuration data, regional boundary data, compliance benchmark points, interval data, two-dimensional mapping tables, fixed distance values, audit level order, and the correspondence between levels and processing types, all determined within the same audit cycle. This ensures that the same input yields verifiable processing results under the same rule version.

[0018] like Figure 1As shown, in one implementation, the procurement compliance intelligent audit system includes an acquisition module, a fusion module, a positioning module, an analysis module, an adjustment module, and an execution module. Each module can be deployed in the same computing device or implemented separately by multiple computing nodes capable of communicating with each other. The acquisition module is connected to the logistics business data interface, the warehousing business data interface, and the financial business data interface. The fusion module receives multi-source business data output by the acquisition module. The positioning module receives weight difference characteristic values ​​and freight difference characteristic values ​​output by the fusion module. The analysis module receives the target audit point, target benchmark area, and area identifier output by the positioning module. The adjustment module receives the vertical distance and along-line positioning value output by the analysis module. The execution module receives the candidate audit level and benchmark deviation value for continuous comparison output by the adjustment module.

[0019] The audit rule base can be set as an internal data storage structure of the system or as an independent data storage structure accessed by the system through a rule interface. Each set of rule data in the audit rule base is associated with a rule version identifier, an audit cycle identifier, and a region identifier. Under the same region identifier, the meaning of the coordinate axes, the region's horizontal axis boundary, the region's vertical axis boundary, the boundary attribution mark, the coordinates of the first compliance benchmark point, the coordinates of the second compliance benchmark point, the deviation value range, the positioning value range, the two-dimensional mapping table identifier, the fixed distance value, the audit level order, and the audit processing type mapping are recorded. When the system reads the rule data, it also records the rule version identifier so that subsequent continuous comparisons only use data belonging to the same audit cycle and using the same rule version.

[0020] All configuration data in the audit rule base, including coordinate system configuration data, area boundary data, compliance benchmark points, deviation value ranges, positioning value ranges, two-dimensional mapping tables, fixed distance values, and audit level order, are pre-established and written by the rule configuration device; the rule configuration device performs the following processing: The regional boundary data configuration uses past audit records of the industry to which the target procurement business belongs as samples. It statistically analyzes the distribution range of each sample point in the audit feature coordinate system. According to the audit granularity required by the audit rules, the coordinate plane is divided into multiple non-overlapping candidate benchmark regions. The horizontal and vertical boundaries of each candidate benchmark region correspond to the quantile values ​​of the statistical distribution. The selection of quantile values ​​is set by the audit rule setter based on the business risk tolerance. After setting, the region identifier, the lower horizontal boundary, the upper horizontal boundary, the lower vertical boundary, the upper vertical boundary, and the boundary attribution mark are written into the region configuration table.

[0021] For each candidate benchmark region, the audit rule makers select two reference coordinate points within the region as the first and second compliance benchmarks based on the risk control objectives corresponding to that region. The horizontal axis coordinates or vertical axis coordinates of the two compliance benchmarks are different to ensure that a compliance benchmark straight line in a non-zero direction can be determined. The coordinate values ​​of the two compliance benchmarks are written into the benchmark configuration table.

[0022] The two-dimensional mapping table configuration involves setting multiple deviation value intervals and multiple location value intervals for each candidate baseline region. Each interval is assigned a unique row index or column index. Then, the corresponding review level is specified for the intersection of each set of row and column indices, forming a two-dimensional mapping table for that region. The review levels are arranged from low to high according to the review processing intensity, forming a review level order.

[0023] For each candidate baseline region, the review rule setter sets a non-negative fixed distance value. This value is used to measure whether the baseline deviation values ​​of multiple consecutive reviews within the same review cycle continuously exceed the tolerance range.

[0024] Rule version management: Each time any of the above configuration data is modified, the system generates a new rule version identifier and records the old version identifier and effective time period in the rule version log. When each module executes an audit task, it reads the configuration data using the rule version identifier bound when the audit task is created, and uses the same version of rule data within the same audit cycle. After a new rule version is released, existing but incomplete audit tasks continue to use the old version of the rules until completion, while newly created audit tasks use the new version of the rules. When switching versions, comparisons of continuous deviations in data are only performed between audit records of the same rule version, and cross-version merging is not performed.

[0025] The modules can exchange processing results through memory objects, message queues, service interfaces, or database records. However, regardless of the transmission method, each subsequent processing step takes the data object output by the previous step as input. For example, the analysis module does not read compliance benchmarks from arbitrary areas, but uses the area identifier output by the positioning module to query two compliance benchmarks corresponding to the target benchmark area. Similarly, the adjustment module does not directly read the original weight and freight, but performs interval positioning based on the vertical distance and positioning values ​​along the line output by the analysis module. Thus, the target audit point, target benchmark area, compliance benchmark line, benchmark deviation value, two-dimensional mapping unit, and target audit level form a continuous data dependency chain.

[0026] The system can create an audit task identifier for each audit and sequentially write the following statuses into the task status record: data acquisition completion status, fusion completion status, feature calculation completion status, regional matching completion status, baseline line verification completion status, geometric analysis completion status, mapping query completion status, level adjustment completion status, and instruction execution completion status. When an abnormal result occurs at any stage, the system stops relying on the subsequent normal calculation based on the abnormal result and passes the already formed fusion audit record, abnormal type, abnormal location, and rule version identifier to the execution module, which then generates the instruction to be verified. This status record allows the audit task to continue processing from the already confirmed valid stage when it is re-executed or recovers from a fault, while avoiding the misinterpretation of incomplete intermediate results as intelligent audit results for procurement compliance.

[0027] Referring to the main process diagram, the intelligent procurement compliance audit system acquires multi-source business data for the target procurement business, standardizes and correlates the multi-source business data, calculates weight difference characteristic values ​​and freight difference characteristic values, establishes target audit points and determines target benchmark areas, obtains two compliance benchmark points corresponding to the area and forms a compliance benchmark line, calculates the vertical distance and positioning value along the line, determines the benchmark deviation value and row index and column index, queries the two-dimensional mapping table to obtain candidate audit levels, determines the target audit level based on continuous comparison results, generates audit processing instructions and controls the approval process nodes; each step is explained below.

[0028] The acquisition module receives the business identifier of the target procurement business and sends data read requests to the logistics business data interface, warehousing business data interface, and financial business data interface respectively based on the business identifier. The logistics business data interface returns logistics waybill data containing waybill identifier, outbound box number, batch identifier, recorded weight, chargeable weight, invoiced freight, and data source identifier. The warehousing business data interface returns warehousing outbound data containing outbound box number, batch identifier, outbound weight, and data source identifier. The financial business data interface returns financial reconciliation data containing waybill identifier, reconciliation detail identifier, actual freight, currency identifier, exchange rate conversion details, and data source identifier. The fields returned by each interface are based on the actual records. The aforementioned fields are used to describe the data categories required for subsequent association and calculation, and different source systems are not required to use the same field names.

[0029] The acquisition module performs integrity checks on the returned records, confirming that the records can be associated with the target procurement business, the data source is identifiable, and the necessary fields are present. When the logistics waybill data, warehouse outbound data, and financial reconciliation data all pass the checks, the acquisition module encapsulates the records according to their data sources and combines the record sets from the three sources with the business identifier of the target procurement business into multi-source business data. When the record weight, billing weight, invoice freight, actual freight, or exchange rate conversion details used to determine the target audit point are missing, the acquisition module retains the acquired records and writes a data missing marker, which triggers calculation anomaly handling. Since the acquisition module retains both the source and association identifiers, it can locate the business link that generated the data when discrepancies occur later.

[0030] The fusion module reads multi-source business data and standard field mapping configurations, and performs standardization based on the standard field name, data type, data format, weight unit, and currency of each source field. For field names, the fusion module uses the source system identifier and the source field name as a mapping key to query the field mapping relationship and obtain the standard field name. For date, number, and numerical formats, the fusion module parses the source field according to the target format and generates a standard format value. For weight data, the fusion module converts the data according to the conversion relationship between the unit of measurement in the record and the target weight unit of measurement. For monetary data, the fusion module retains the original currency amount, currency identifier, and exchange rate conversion details for conversion to the same currency.

[0031] During field standardization, if the source field lacks a corresponding standard field mapping, the numerical field cannot be parsed, or the unit of measurement lacks conversion relationships, the fusion module does not replace the original content with null values. Instead, it retains the source field value, data source identifier, and failure reason, forming a standardization anomaly marker. When no such anomaly exists, the fusion module obtains standardized logistics waybill records, standardized warehouse outbound records, and standardized financial reconciliation records, and combines these three types of standardized records into standardized business data. Since the standardization process first unifies the field semantics and measurement basis, the subsequent difference calculation uses directly comparable data, avoiding misidentification of differences in field names or units of measurement as business differences.

[0032] The fusion module locates the target logistics waybill record based on the waybill identifier in the standardized business data. When the same waybill identifier corresponds to multiple records, the fusion module uses the set of records associated with the target procurement business as the processing scope, and retains the business detail identifier of each record, without unconditionally overwriting different details. It constructs a warehousing association key based on the outbound box number and batch identifier in the target logistics waybill record, and uses this warehousing association key to query standardized warehousing outbound records, obtaining the target warehousing outbound record that simultaneously satisfies both the outbound box number and batch identifier. Finally, it constructs a financial association key based on the waybill identifier and reconciliation detail identifier in the target logistics waybill record, and uses this financial association key to query standardized financial reconciliation records, obtaining the target financial reconciliation record that simultaneously satisfies both the waybill identifier and reconciliation detail identifier.

[0033] For each target logistics waybill record, the fusion module merges the corresponding target warehouse outbound record and target financial reconciliation record into the same audit data object. For standard fields with the same meaning and consistent values ​​from the three sources, the fusion module stores a unified field value and retains the source set. For standard fields with the same meaning but inconsistent values, the fusion module retains the values ​​from each source and the corresponding source identifiers, without overwriting other source values ​​with any source value, and writes a field difference marker. For fields provided by only one source, the fusion module retains the field and its source identifier. After completing the association of all target logistics waybill records, the fusion module outputs a fusion audit record, which provides the record weight, chargeable weight, billed freight, actual freight, exchange rate conversion details, and association identifier for subsequent use.

[0034] When a warehousing association key does not match a target warehouse outbound record or a financial association key does not match a target financial reconciliation record, the fusion module writes the missing association key, the matched target logistics waybill record, and the type of the missing record into the fusion audit record and sets a data missing flag. When a single association key matches multiple records that cannot be distinguished based on existing identifiers, the fusion module retains all candidate records and sets a non-unique association flag, without speculatively selecting candidate records. The data missing flag or non-unique association flag causes the system to enter pending verification processing when information is insufficient, avoiding the generation of compliance release or compliance warning results based on uncertain data. When a single association key matches multiple records that cannot be distinguished based on existing identifiers, the fusion module further... The system first performs a time priority judgment, reads the record generation timestamps of each candidate record, and selects the candidate record whose generation time is closest to that of the target logistics waybill record as the target record. At the same time, it writes the non-unique association marker and the timestamp of the selected record into the fusion audit record. If multiple candidate records have the same time difference as the target logistics waybill record, the target record cannot be uniquely determined based on time priority. In this case, the fusion module retains all candidate records and sets a non-unique association marker, and does not make speculative selections of candidate records. The data missing marker or non-unique association marker causes the system to enter the pending verification process when there is insufficient information, avoiding the generation of compliance release or compliance warning results based on uncertain data.

[0035] The fusion module reads the recorded weight and billing weight, which use the same unit of measurement, from the fusion audit records. The recorded weight can be taken from the warehouse outbound weight associated with the target logistics waybill record and with a unified unit. The billing weight is taken from the billing weight in the target logistics waybill record. The specific field correspondence is determined by the standard field mapping configuration. The fusion module subtracts the billing weight from the recorded weight to obtain the weight difference value. Then, it obtains the absolute value of the weight difference value and divides this absolute value by the non-zero weight difference benchmark value in the audit rule base that corresponds to the current audit rule to obtain the weight difference feature value. The weight difference benchmark value uses the same unit of measurement as the recorded weight and the billing weight, so the ratio is dimensionless.

[0036] The weight difference baseline value is used to provide a normalized reference under the same rule version. Before performing division, the fusion module checks whether the value exists and is not equal to 0. If the weight difference baseline value is equal to 0 or cannot be read, the system does not perform the weight difference feature value calculation, generates an abnormal calculation result, and records the abnormal field. If the recorded weight or billing weight is negative, cannot be parsed into a numerical value, or lacks a unit of measurement, the system generates a data missing or calculation abnormality mark according to the data validity check result. By confirming that the units are consistent and the denominator is valid before calculation, the weight difference feature value can be guaranteed to have a definite meaning.

[0037] The fusion module reads the currency identifiers and target currency identifiers of the invoice shipping fee and the actual shipping fee, along with their corresponding conversion relationships, from the exchange rate conversion details. It converts the invoice shipping fee into the base invoice shipping fee and the actual shipping fee into the base actual shipping fee. If both shipping fees already use the same target currency, their amounts remain unchanged. If conversion is required, multiplication or division is performed according to the amount direction recorded in the exchange rate conversion details. The conversion direction is jointly determined by the definitions of the original currency, target currency, and exchange rate fields. The system does not automatically select the calculation method when the conversion direction is missing. After the conversion is completed, both base shipping fees use the same currency and the same amount precision.

[0038] The fusion module calculates the freight difference by subtracting the actual freight from the benchmark invoice freight. It then obtains the absolute value of the freight difference and divides it by the non-zero freight difference benchmark value in the audit rule base that corresponds to the current audit rule to obtain the freight difference characteristic value. The freight difference benchmark value and the two benchmark freight costs use the same currency and monetary unit, so the freight difference characteristic value is dimensionless. When the freight difference benchmark value is equal to 0, the exchange rate conversion details are missing, the conversion direction cannot be determined, or the two freight costs cannot be converted to the same currency, the system stops calculating the freight difference characteristic value and generates an abnormal calculation result.

[0039] The positioning module establishes an audit feature coordinate system based on the coordinate system configuration data in the audit rule base. The horizontal axis represents the weight difference feature value, and the vertical axis represents the freight difference feature value. The positioning module uses the weight difference feature value as the horizontal axis coordinate value of the target audit point and the freight difference feature value as the vertical axis coordinate value of the target audit point to obtain the target audit point. The target audit point also retains the business identifier, audit task identifier, and rule version identifier of the target procurement business, so that the coordinate position can be correlated with the integrated audit record and subsequent audit results.

[0040] The positioning module reads multiple candidate reference regions based on the rule version identifier. Each candidate reference region is defined by the lower boundary of the horizontal axis, the upper boundary of the horizontal axis, the lower boundary of the vertical axis, the upper boundary of the vertical axis, and the attribution markers of each boundary. The region boundary data adopts boundary attribution rules that ensure that the candidate reference regions do not overlap. When adjacent regions share numerical boundaries, only the attribution marker of one region contains that boundary. The positioning module sequentially compares the horizontal axis coordinate value of the target audit point with the horizontal axis boundary of each candidate reference region, and compares the vertical axis coordinate value with the corresponding vertical axis boundary. When both the horizontal axis and the vertical axis meet the boundary conditions of the same candidate reference region, the candidate reference region is determined as the target reference region, and its region identifier is output.

[0041] The localization module stops region matching after traversing all candidate reference regions under the current rule version. If only one candidate reference region contains the target audit point, the target reference region and its identifier are output. If no candidate reference region contains the target audit point, a region matching error result is output. If the region boundary data is misconfigured, causing two or more candidate reference regions to simultaneously contain the target audit point, the system does not arbitrarily select one of the regions, but instead writes the overlapping region identifier and the target audit point into the region matching error result. Using boundary attribution markers can handle the case where the point is located on a common boundary and ensures that the region selection is deterministic.

[0042] The analysis module uses the region identifier output by the positioning module as the query key to access the audit rule base and reads the first and second compliance benchmark points corresponding to the target benchmark region. Both compliance benchmark points include horizontal and vertical coordinate values ​​and use the same coordinate definition as the audit feature coordinate system. The analysis module compares the horizontal and vertical coordinate values ​​of the two compliance benchmark points. When the horizontal coordinate values ​​are not equal or the vertical coordinate values ​​are not equal, a valid benchmark point result is formed. When the horizontal coordinate values ​​are equal and the vertical coordinate values ​​are also equal, the two benchmark points coincide, and the non-zero benchmark direction cannot be determined, resulting in an abnormal calculation result.

[0043] After obtaining valid benchmark results, the analysis module subtracts the horizontal axis coordinate of the first compliance benchmark from the horizontal axis coordinate of the second compliance benchmark to obtain the change in horizontal axis position, and subtracts the vertical axis coordinate of the first compliance benchmark from the vertical axis coordinate of the second compliance benchmark to obtain the change in vertical axis position. The changes in horizontal and vertical axis positions together constitute the benchmark direction from the first compliance benchmark to the second compliance benchmark. The analysis module uses the first compliance benchmark as a known point on the straight line and the benchmark direction as the extension direction of the straight line to obtain the compliance benchmark straight line corresponding to the target benchmark area. Since the two benchmarks are obtained by querying the area identifier of the target benchmark area, the compliance benchmark straight line maintains a clear association with the area where the determined target audit point is located.

[0044] The analysis module again uses the coordinate position difference between the first compliance benchmark point and the second compliance benchmark point to construct a benchmark direction vector. The horizontal axis component of the benchmark direction vector is the change in horizontal position, and the vertical axis component is the change in vertical position. The analysis module subtracts the horizontal axis coordinate value of the first compliance benchmark point from the horizontal axis coordinate value of the target audit point to obtain the horizontal axis component of the target displacement vector, and subtracts the vertical axis coordinate value of the first compliance benchmark point from the vertical axis coordinate value of the target audit point to obtain the vertical axis component of the target displacement vector. The two direction components together constitute the target displacement vector pointing from the first compliance benchmark point to the target audit point.

[0045] The analysis module calculates the squares of the two directional components of the reference direction vector, and adds the two squares to obtain the squared value of the reference direction length. If the squared value of the length is greater than 0, the projection calculation continues; if it is equal to 0, it indicates that the two compliant reference points coincide, the system generates an abnormal calculation result, and stops the geometric calculation of the current audit task. Subsequently, the analysis module multiplies the horizontal component of the target displacement vector with the horizontal component of the reference direction vector, and multiplies the vertical component of the target displacement vector with the vertical component of the reference direction vector, and then adds the two products to obtain the directional correlation value. The directional correlation value represents the degree of correlation of the target displacement vector in the reference direction.

[0046] The analysis module divides the direction correlation value by the square of the non-zero reference direction length to obtain the projection scale value; the projection scale value is multiplied by the horizontal and vertical components of the reference direction vector to obtain the two direction components of the target projection vector; the horizontal coordinate value of the first compliance reference point is added to the horizontal component of the target projection vector, and the vertical coordinate value of the first compliance reference point is added to the vertical component of the target projection vector to obtain the horizontal and vertical coordinate values ​​of the perpendicular point; thus, the perpendicular point is located on the compliance reference line and corresponds to the position of the target audit point projected along the direction perpendicular to the line.

[0047] The analysis module calculates the first coordinate difference by subtracting the x-axis coordinate of the perpendicular point from the x-axis coordinate of the target audit point, and the second coordinate difference by subtracting the y-axis coordinate of the perpendicular point from the y-axis coordinate of the target audit point. The squares of the two coordinate differences are calculated, and the sum of the squares is taken as the non-negative square root to obtain the vertical distance between the target audit point and the perpendicular point. Since the coordinates of the audit feature coordinate system are dimensionless feature values, both coordinate differences and the vertical distance are dimensionless quantities. The vertical distance does not retain the sign of which side of the straight line the target audit point is on; it only represents the degree to which the target audit point deviates from the compliance baseline straight line.

[0048] In one direct implementation, the analysis module determines the projection scale value as the positioning value along the line. When the positioning value along the line is less than 0, the perpendicular point is located in the direction away from the second compliance benchmark point from the first compliance benchmark point; when the positioning value along the line is equal to 0, the perpendicular point coincides with the first compliance benchmark point; when the positioning value along the line is greater than 0, the perpendicular point is located in the direction from the first compliance benchmark point toward the second compliance benchmark point; when the positioning value along the line is equal to 1, the perpendicular point coincides with the second compliance benchmark point; when it is greater than 1, the perpendicular point is located at the position extending along the benchmark direction after crossing the second compliance benchmark point. By simultaneously outputting the vertical distance and the positioning value along the line, the system can distinguish target audit points with the same degree of deviation but different positions along the compliance benchmark line. The vertical distance characterizes the unsigned degree of deviation of the target audit point from the compliance benchmark line; the larger the value, the higher the joint deviation of the weight difference characteristics and freight difference characteristics of the target procurement business. The greater the audit risk, the higher the risk level. The positioning value along the line represents the relative position of the target audit point on the compliance benchmark line. When the value is less than 0, it means that the perpendicular point is located on the side of the first compliance benchmark point away from the second compliance benchmark point, indicating that the difference feature shows a deviation trend opposite to the benchmark direction. When the value is equal to 0, it means that the perpendicular point coincides with the first compliance benchmark point. When the value is between 0 and 1, it means that the perpendicular point is located between the two compliance benchmark points, indicating that the difference feature is within the reference range determined by the benchmark line. When the value is greater than 1, it means that the perpendicular point extends beyond the second compliance benchmark point and continues to extend in the benchmark direction, indicating that the deviation of the difference feature in the benchmark direction has exceeded the reference range limited by the second compliance benchmark point. By simultaneously outputting the vertical distance and the positioning value along the line, the system can distinguish target audit points with the same degree of deviation but different positions along the compliance benchmark line, thereby achieving a more refined risk classification than a single deviation threshold judgment.

[0049] The adjustment module determines the vertical distance output by the analysis module as the baseline deviation value and the line positioning value output by the analysis module as the column positioning value. The baseline deviation value is non-negative and dimensionless data, and the column positioning value can be negative, 0, or positive. The adjustment module first determines whether the column positioning value belongs to a negative value range, a zero value range, or a positive value range according to the sign of the column positioning value, and then determines a more specific range position in the positioning value range corresponding to the target baseline area. The zero value range can be represented by a single range containing 0, or the positioning value range to which 0 belongs can be clearly defined by the boundary assignment mark in the area rules.

[0050] The adjustment module queries the audit rule base using the region identifier of the target baseline region, and reads multiple deviation value intervals and multiple positioning value intervals arranged in numerical direction. Each interval includes a lower boundary, an upper boundary, and a marker indicating whether the boundary is included. The interval number corresponds to the row position and column position of the two-dimensional mapping table, respectively. The adjustment module compares the baseline deviation value with the boundary of the deviation value interval in turn, and determines the row position of the deviation value interval that meets the boundary condition as the row index. The adjustment module compares the column positioning value with the boundary of the positioning value interval in turn, and determines the column position of the positioning value interval that meets the boundary condition as the column index. Both the row index and the column index use non-negative integers and are consistent with the storage location of the two-dimensional mapping table.

[0051] When the baseline deviation value or column positioning value is exactly equal to the common boundary of adjacent intervals, the adjustment module selects a unique target interval according to the boundary attribution mark pre-recorded in the interval data; when any value does not belong to all the configured intervals, the adjustment module generates an interval matching anomaly and includes it as part of the calculation anomaly result; when the same value hits two intervals at the same time, it indicates that the interval configurations overlap, the adjustment module records the identifier of the duplicated intervals and generates a calculation anomaly result; the above processing ensures that the row index and column index used by the two-dimensional mapping table query have a unique source.

[0052] The adjustment module obtains a two-dimensional mapping table corresponding to the target baseline area based on the area identifier of that area. Multiple row positions in the two-dimensional mapping table correspond one-to-one with multiple deviation value intervals, and multiple column positions correspond one-to-one with multiple positioning value intervals. The intersection of each row position and column position forms a mapping unit, and each valid mapping unit stores an audit level identifier. The two-dimensional mapping table also records the rule version identifier to which it applies. Before querying, the adjustment module checks whether the two-dimensional mapping table, deviation value intervals, and positioning value intervals belong to the same rule version. If the versions are inconsistent, the query is not executed.

[0053] The adjustment module locates the target row using the row index and the target column using the column index, and determines the intersection of the target row and the target column as the target mapping unit. When the target mapping unit exists and stores the audit level, the audit level is read and determined as the candidate audit level. At the same time, the region identifier, row index, column index, mapping table identifier, target mapping unit position, candidate audit level, and rule version identifier are written to the level determination record. When the target mapping unit does not exist, the row index or column index exceeds the mapping table range, or the target mapping unit does not store the audit level, the system generates an abnormal calculation result. The level determination record retains the lookup path from the baseline deviation value and column positioning value to the candidate audit level, which facilitates the review of the audit results.

[0054] The execution module or adjustment module queries the audit rule base based on the regional identifier of the target benchmark area to obtain a fixed distance value, a consecutive preset number of times, and a preset audit level order. The consecutive preset number of times ranges from 3 to 10 times, and the specific value is set by the audit rule setter according to the business audit cycle and risk control requirements. In this embodiment, it is preferably set to 5 times. The preset audit level order is arranged from low to high audit processing intensity, and is exemplarily set as low risk level, medium risk level, and high risk level. The adjacent previous audit level refers to the adjacent level that is one audit processing intensity level higher than the candidate audit level in the audit level order. For example, when the candidate audit level is low risk level, the adjacent previous audit level is medium risk level. The end of the audit level order represents the level with the highest audit processing intensity, i.e., high risk level. When the candidate audit level is already at the end, no further level increase adjustment is made. This definition makes the direction of level adjustment independent of the literal meaning of the level name.

[0055] The system uses the business identifier of the target procurement business, the regional identifier of the target benchmark area, the audit cycle identifier, and the rule version identifier as search criteria to read the benchmark deviation values ​​formed by the most recent consecutive preset number of audits for the target procurement business within the current audit cycle. The system arranges these benchmark deviation values ​​according to the audit time or audit sequence number and confirms that there are no rule version changes or invalid audit records between the records to obtain continuous deviation data. Each benchmark deviation value in the continuous deviation data is compared with a fixed distance value. When each benchmark deviation value is greater than the fixed distance value, the result is that the grade adjustment condition is met. If any benchmark deviation value is less than or equal to the fixed distance value, the grade adjustment condition is not met.

[0056] When the level adjustment conditions are met and the candidate review level is not at the end of the preset review level order, the system obtains the previous review level adjacent to the candidate review level along the direction of increasing review processing intensity and determines that level as the target review level; when the candidate review level is already at the end of the order, the system keeps the candidate review level unchanged and determines it as the target review level; when there is a baseline deviation value less than or equal to the fixed distance in the continuous deviation data, the system also keeps the candidate review level unchanged; if the number of valid records in the current review cycle has not reached the preset number of consecutive times, the system saves the current baseline deviation value for subsequent reviews and determines the candidate review level as the target review level in this process, without presuming that the continuous condition is met based on an insufficient number of data.

[0057] Continuous comparison is only performed on valid benchmark deviations under the same target procurement business, the same target benchmark area, the same audit cycle, and the same rule version. If the audit cycle changes, the rule version changes, or the target audit point enters a different target benchmark area, the system will regenerate continuous deviation data under the corresponding new combination of conditions, without incomparable historical records being included in the current continuous sequence. Thus, the level adjustment reflects the continuous deviation status under the same audit conditions, avoiding the increase in audit processing intensity due to rule switching or regional changes.

[0058] The execution module obtains the target audit processing type based on the correspondence between the target audit level and the audit processing type. When the target audit processing type is a compliance release type, the execution module generates a compliance release instruction. When the target audit processing type is a compliance warning type, the execution module generates a compliance warning instruction. The audit processing instruction includes the audit task identifier, the business identifier of the target procurement business, the target audit level, the target audit processing type, the rule version identifier, and the instruction status. The approval process node locates the process instance to be processed based on the business identifier in the instruction and executes the node processing corresponding to the instruction type.

[0059] When an abnormal result is obtained in the region matching, the execution module generates a verification instruction; when there are missing data in the fusion audit record for determining the target audit point, the obtained baseline direction length square value is equal to 0, no unique row index or column index is obtained, or the target mapping unit does not store the audit level, the execution module generates a verification instruction based on the corresponding abnormal information; the verification instruction includes the abnormality type, the stage of the abnormality, the available fusion audit record and rule version identifier, but does not include the audit level inferred based on incomplete calculation, so that the approval process node can transfer the business to the verification process.

[0060] After receiving a compliance release instruction, the approval process node executes the node corresponding to the compliance release type, processes the data, and returns an audit execution status of release completed, release failed, or pending retry. After receiving a compliance warning instruction, it executes the warning processing node corresponding to the compliance warning type and returns an audit execution status of warning generated, warning processing failed, or pending retry. After receiving a pending verification instruction, it transfers the process instance to the verification node and returns a pending verification status. The execution module updates the instruction status based on the returned status, and retains the same audit task identifier when the node returns pending retry to avoid generating duplicate independent audit results.

[0061] The execution module integrates audit records, weight difference characteristic values, freight difference characteristic values, target audit points, regional identifiers, benchmark deviation values, column positioning values, grade determination records, target audit grades, audit processing instructions, and audit execution status, and establishes a correlation based on the audit task identifier to obtain the intelligent audit results for the procurement compliance of the target procurement business. For abnormal processes, the intelligent audit results for procurement compliance include available integrated audit records, abnormal regional matching results or calculation results, instructions to be verified, and status to be verified. This result not only shows the status of the approval process that has been executed, but also retains the intermediate basis from multi-source business data to processing instructions, thereby supporting the review of results under the same rule version.

[0062] In one implementation, the audit rule base stores a region configuration table, a benchmark configuration table, an interval configuration table, a level mapping table, and a process mapping table according to rule versions. The region configuration table uses the region identifier as the primary key and records the horizontal axis boundary, vertical axis boundary, and boundary attribution marker. The benchmark configuration table uses the region identifier as the query key and records the coordinates of the first compliance benchmark point and the second compliance benchmark point. The interval configuration table records the deviation value interval, the positioning value interval, and their corresponding row and column positions. The level mapping table records the correspondence between the region identifier, row position, column position, and audit level. The process mapping table records the correspondence between the audit level and the audit processing type. Each table also stores the rule version identifier so that the system can perform version consistency verification when reading.

[0063] Multiple deviation value intervals are arranged in ascending order of the baseline deviation value, and multiple location value intervals are arranged in ascending order of the location value along the line. Interval data are not required to be divided with equal width, but adjacent intervals must not overlap without explanation, and the common boundary is uniquely assigned by the boundary attribution mark. When establishing a two-dimensional mapping table, the system establishes a row index set based on the row position of the deviation value interval, establishes a column index set based on the column position of the location value interval, and then establishes a mapping unit for each set of valid row and column positions, and writes the audit level identifier determined by the audit rules into the mapping unit.

[0064] Two-dimensional mapping tables can be implemented using two-dimensional arrays, key-value structures with composite keys composed of row and column indices, or relational data tables. When using two-dimensional arrays, the row and column indices directly correspond to the first and second dimensions of the array, respectively. When using key-value structures, the system combines the region identifier, row index, and column index into a mapping key. When using relational data tables, the system uses the region identifier, row position, and column position as the conditions for joint queries. Regardless of the data structure used, when a match is found, the audit level field in the target mapping unit is read; when a match is not found, an abnormal calculation result is generated, and the missing level is not replaced by the level of the adjacent mapping unit.

[0065] Before rule data is published, structural verification can be performed. Structural verification includes checking whether two compliance benchmarks overlap, whether the deviation value range covers the benchmark deviation value range to be processed, whether the positioning value range covers the positioning value range along the line to be processed, whether there is duplicate attribution between intervals, whether the row and column positions are non-negative integers, whether each expected mapping unit has an audit level, whether the distance fixed value is a valid non-negative number, and whether all levels in the audit level order have processing type mappings. Structural verification does not change the specific content of the business rules, but confirms that the configured rules can be read and executed by the above processing chain. Rule versions that fail structural verification will not be used for new audit tasks.

[0066] Within the same review cycle, review tasks that have already started processing use the rule version bound to them when they were created; when a new rule needs to be released, the new rule uses a different rule version identifier and is used by the new review task after the rule takes effect boundary; continuous deviation data is grouped by rule version, and the baseline deviation value formed by the old rule does not directly participate in the continuous comparison of the new rule; by adopting this version isolation method, it can be ensured that the adjustment basis of the target review level comes from the same coordinate boundary, the same baseline line, the same interval mapping, and the same fixed distance value.

[0067] The standard field mapping configuration is used to convert source fields from different data sources into standard fields with unified semantics. Each mapping record includes the source system identifier, source field name, standard field name, source data type, target data type, source data format, target data format, source unit of measurement, target unit of measurement, and enabled status. When this configuration is established, the configuration receiving unit receives the above fields, combines the source system identifier and source field name into a mapping key, first queries existing mapping records, and creates a new record if no matching mapping key is found. If the matching mapping key is found and the mapping content is consistent, the existing record is reused. If the matching mapping key is found but the standard field name, data type, format, or unit is different, a duplicate key conflict is formed, and the configuration with duplicate key conflict is not published.

[0068] The field mapping configuration is not inferred by the system based on the similarity of field names, but is input into the system as audit rule configuration data. The configuration receiving unit checks the existence of required fields item by item, whether the mapping key composed of the source system identifier and the source field name is empty, whether the target data type has the corresponding parsing processing, whether the weight field has the source unit of measurement and the target weight unit of measurement, and whether the amount field has the currency field or the exchange rate conversion details associated field. After all checks pass, the rule version identifier and effective status are written to the mapping record. When standardizing fields, only mapping records that are consistent with the current rule version and are in an effective state are queried.

[0069] When performing format conversion on a numeric field, the normalization unit first reads the text content and source data format of the source field, identifies the positive and negative markers, integer part, decimal part, and allowed separator characters according to the source data format, removes separator characters that are explicitly defined by the source data format for display purposes, and then parses the remaining characters into numeric values. After parsing, the normalization unit checks whether the numeric value is within the representable range according to the target data type and writes it into the normalized field according to the target data format. If the source field is empty, contains characters not defined in the source data format, has only positive and negative markers without a numeric part, or exceeds the representable range of the target data type, a format parsing exception is generated and the content of the source field is preserved.

[0070] When performing format conversion on the identifier field, the standardization unit determines whether to retain the leading character, remove blank characters used only for display, and unify the case of characters according to the mapping record. The processed identifier is still saved in string form to avoid incorrectly converting waybill identifiers, box numbers, batch identifiers, or reconciliation detail identifiers with leading characters into numerical values. When performing format conversion on the date and time field, the standardization unit reads the year, month, day, and time components in sequence according to the source data format, checks whether each component constitutes a valid time, and then generates a standard time field according to the target data format. If the date and time components are missing or cannot constitute a valid time, a format parsing error will occur.

[0071] Weight unit conversion is performed through unit conversion configuration. Each unit conversion record includes the source weight unit, target weight unit, conversion direction, and non-zero conversion factor. The standardization unit uses the source weight unit and target weight unit as the query key. When the source unit and target unit are the same, the source weight value is directly determined as the standard weight value. When the source unit and target unit are different, the corresponding conversion record is read. When the conversion direction indicates that the source unit value is multiplied by the conversion factor to obtain the target unit value, the source weight value is multiplied by the conversion factor to obtain the standard weight value. When the conversion direction indicates that the source unit value is divided by the conversion factor to obtain the target unit value, the source weight value is divided by the non-zero conversion factor to obtain the standard weight value. A unit conversion anomaly occurs when the conversion record is not matched, the conversion direction is missing, or the conversion factor used in the division is equal to 0.

[0072] Currency conversion uses the conversion record corresponding to the target audit record in the exchange rate conversion details. The conversion record includes the original currency identifier, target currency identifier, exchange rate value, exchange rate expression direction, and applicable record identifier. When the original currency and target currency are the same, the original amount is determined as the base amount. When they are different, the conversion record is queried using the original currency identifier and the target currency identifier. When the exchange rate expression direction represents the target currency amount corresponding to one unit of the original currency, the original amount is multiplied by the exchange rate value to obtain the base amount. When the exchange rate expression direction represents the original currency amount corresponding to one unit of the target currency, the original amount is divided by the non-zero exchange rate value to obtain the base amount. If the conversion record is not matched, the applicable record identifier is inconsistent, the exchange rate value is equal to 0, or the expression direction cannot be identified, the amount conversion is stopped and an exchange rate conversion anomaly is generated.

[0073] The merged audit records can adopt a record structure with a fixed set of fields. Each record includes a business identifier for the target procurement business, a target logistics waybill record identifier, a target warehouse outbound record identifier, a target financial reconciliation record identifier, a standard field set, a source field set, a field difference marker set, a data missing marker set, and a rule version identifier. When merging records with the same field, the record association unit uses the standard field name as the merge key. When the standard field is read for the first time, a target field item is created and the field value and source identifier are written. When the same standard field name is read subsequently, the subsequent field value is compared with the existing field value in the target field item according to the same data type.

[0074] When a subsequent field value is equal to an existing field value, the record association unit does not store the same value repeatedly, but instead adds the subsequent data source identifier to the source set of the target field item. When a subsequent field value is not equal to an existing field value, the record association unit writes each field value and its corresponding source identifier into the source field set, and writes the standard field name into the field difference marker set, without selecting any source value to overwrite other values. When subsequent steps require specifying a source field, the corresponding source value is read according to the field purpose already determined in the standard field mapping configuration. If the field purpose is not determined in the configuration or the required source value does not exist, a data missing result is generated.

[0075] The weight difference benchmark value, freight difference benchmark value, and distance fixed value are all used as input configuration values ​​in the audit rule base and are not derived temporarily during a single online audit. When establishing the corresponding configuration record, the configuration receiving unit receives the configuration value, parameter name, applicable area identifier, applicable audit cycle identifier, rule version identifier, and parameter unit, respectively. The parameter unit of the weight difference benchmark value is consistent with the standard weight unit, the parameter unit of the freight difference benchmark value is consistent with the benchmark amount unit, and the distance fixed value adopts the same dimensionless representation as the benchmark deviation value. The configuration verification unit checks that the two difference benchmark values ​​are greater than 0, checks that the distance fixed value is a valid non-negative number, and checks that there is a correspondence between the applicable identifier of each parameter and the current rule version. Only after the checks are passed will the configuration record be set to the effective state.

[0076] The coordinate system configuration data includes the horizontal axis field identifier, the vertical axis field identifier, the coordinate data type, the coordinate validity conditions, and the rule version identifier. When establishing the audit feature coordinate system, the coordinate construction unit reads the horizontal axis field identifier to confirm that it points to the weight difference feature value, reads the vertical axis field identifier to confirm that it points to the freight difference feature value, and then checks that both feature values ​​are finite non-negative numbers that can participate in comparison and multiplication / division operations. After the check is passed, a target audit point with two coordinate components is formed according to the order specified by the horizontal axis field identifier and the vertical axis field identifier. If any feature value is missing, cannot be parsed, or does not meet the coordinate validity conditions, the target audit point is not established.

[0077] Each candidate reference region can be represented by a rectangular region record. The rectangular region record includes a region identifier, a lower boundary of the horizontal axis, an upper boundary of the horizontal axis, a lower boundary of the vertical axis, an upper boundary of the vertical axis, and inclusion markers for each of the four boundaries. When writing a region record, the configuration verification unit compares the lower boundary of the horizontal axis with the upper boundary of the horizontal axis. If the former is less than the latter, the horizontal axis range is valid. It also compares the lower boundary of the vertical axis with the upper boundary of the vertical axis. If the former is less than the latter, the vertical axis range is valid. If the lower boundary of any coordinate direction is greater than or equal to the upper boundary, a region boundary anomaly is formed, and the region record is rejected.

[0078] When checking whether two candidate reference regions overlap, the configuration verification unit compares the projection intervals of the two regions in the horizontal and vertical directions respectively. If the horizontal projection intervals of the two regions have common internal values ​​and the vertical projection intervals also have common internal values, the two regions have area overlap, forming a region overlap anomaly. If the two regions only connect at the horizontal or vertical boundary, the inclusion marker of the connecting boundary is further read. If both regions contain the same common boundary, a boundary duplication anomaly is formed. If only one region contains the common boundary, the two regions are allowed to connect. The verification unit performs the above comparison on each candidate reference region in the same rule version. The region configuration is released when all regions have been checked and there are no anomalies.

[0079] The configuration records for the first and second compliance benchmarks include the region identifier, point type, horizontal axis coordinate value, vertical axis coordinate value, and rule version identifier. When writing the benchmark configuration, the configuration verification unit checks whether there is a first compliance benchmark and a second compliance benchmark under the same region identifier, checks whether each coordinate value can be parsed according to the coordinate data type, and then compares the horizontal axis coordinate value and vertical axis coordinate value of the two points. If the coordinate values ​​in both directions are equal, a benchmark overlap anomaly is formed, and the benchmark configuration for that region is not published. If the coordinate values ​​in at least one direction are not equal, a valid benchmark configuration result is formed.

[0080] Both deviation value intervals and location value intervals can be represented by interval records. Each interval record includes interval type, lower boundary, upper boundary, lower boundary containment marker, upper boundary containment marker, and location index. When establishing an interval set, the configuration verification unit first checks that the lower boundary of each interval record is less than the upper boundary, and then arranges the interval records in ascending order of the lower boundary value. For two adjacent records after arrangement, if the upper boundary of the first interval is greater than the lower boundary of the second interval, an interval overlap anomaly is formed. If the two are equal and both intervals contain a common boundary, a boundary duplication anomaly is formed. If the two are equal and only one interval contains a common boundary, the boundary is valid. If the upper boundary of the first interval is less than the lower boundary of the second interval, the interval gap is recorded.

[0081] Gaps in the interval set are not automatically filled by the system; these gaps are retained during configuration release. When a value falls into a gap during online review, an interval matching anomaly is generated. The configuration verification unit also checks whether the position index within the same interval type is a non-negative integer and whether there are duplicate position indices. After the checks pass, the position index of the deviation value interval is used as the row index of the two-dimensional mapping table, and the position index of the location value interval is used as the column index of the two-dimensional mapping table. This process clarifies the establishment process between interval configuration, index value, and mapping position.

[0082] When determining the row index online, the interval positioning unit reads the interval records in the order of the deviation value intervals and compares the baseline deviation value with the lower and upper boundaries of the current interval respectively. If the baseline deviation value is greater than the lower boundary and less than the upper boundary, the current interval is hit. If it is equal to either boundary, the corresponding boundary inclusion flag is used to determine whether it is hit. After a hit, the position index of the current interval is temporarily stored and the remaining intervals are checked to confirm that there is no second hit interval. When there is only one hit result after traversal, the row index is output. If there is no hit result or there are more than two hit results, the interval matching error is output.

[0083] The online column index determination uses the same traversal process as the row index, but the data used for comparison are column positioning values, and the data used for traversal are positioning value ranges. The range positioning unit first determines the sign state based on whether the column positioning value is less than 0, equal to 0, or greater than 0, and then checks whether the matched positioning value range is compatible with the sign state. For example, negative values ​​cannot match ranges where all boundaries are non-negative, and positive values ​​cannot match ranges where all boundaries are non-positive. After comparing all positioning value ranges, the column index is output if there is only one match result; otherwise, the range matching error is output.

[0084] When establishing a two-dimensional mapping table, the mapping configuration unit reads the set of deviation value intervals and the set of location value intervals that have passed the verification, and obtains all row indices and all column indices respectively. For each row index, the mapping configuration unit establishes the corresponding mapping key according to the order of the column indices. The mapping key includes the region identifier, row index, column index and rule version identifier. Then, the audit level corresponding to the mapping key in the input configuration is written into the mapping unit. After completing all row and column combinations, the mapping configuration unit counts the expected mapping key set and the actual mapping key set. When the actual set lacks the expected mapping key, a mapping unit missing anomaly is formed. When the actual set contains duplicate mapping keys, a mapping unit duplicate anomaly is formed.

[0085] When querying the two-dimensional mapping table, the mapping query unit combines the region identifier, row index, column index, and current rule version identifier of the target base area into a target mapping key in the same field order as when the mapping table was created. The mapping unit is then searched using this target mapping key. If a record is found, the review level field is read and a candidate review level is obtained. If no record is found, a mapping unit missing exception is generated. If more than two records are found, a mapping key duplicate exception is generated. If the read review level does not belong to the current review level order, a level configuration exception is generated. None of the above exceptions are filled with levels from adjacent rows or columns.

[0086] The preset review level order is represented by an ordered set of level records. Each level record includes the review level, order position, and review processing type. When establishing the level order, the configuration verification unit checks that the order position is a non-negative integer, checks that different review levels do not use the same order position, and arranges the level records according to the direction specified by the review processing intensity. When determining the adjacent previous review level, the continuous adjustment unit first queries the current order position of the candidate review level, and then queries the level corresponding to the adjacent order position in the direction of increasing processing intensity. When the level is found, the first-level adjustment is completed. If there is no adjacent order position, it indicates that the candidate review level is at the end of the order, and the candidate review level remains unchanged.

[0087] When generating continuous deviation data, the continuous adjustment unit uses the business identifier of the target procurement business, the regional identifier of the target benchmark area, the audit cycle identifier, and the rule version identifier as search keys to query audit records that have completed geometric calculations and whose benchmark deviation values ​​are valid. The continuous adjustment unit removes records marked with missing data, abnormal regional matching, or abnormal calculation, and sorts the remaining records in ascending order according to the audit completion sequence number. When the audit completion sequence number is repeated or missing, it forms a continuous data order abnormality. After sorting, it retrieves a preset number of consecutive records, including the current record, from the end of the sequence. When the number of records retrieved reaches the preset number of consecutive records, it stops reading and obtains continuous deviation data. When the number of records is insufficient, it forms a continuous number unmet state.

[0088] The continuous adjustment unit sequentially reads each baseline deviation value in the continuous deviation data, compares the current baseline deviation value with the fixed distance value, and writes the comparison result to the comparison status corresponding to the audit record. When the current baseline deviation value is greater than the fixed distance value, it is written as "exceeded"; when it is less than or equal to the fixed distance value, it is written as "not exceeded" and the level adjustment condition is set to "not true". During the traversal, once the "not exceeded" status is obtained, the level improvement condition can be stopped. The remaining records are still retained in the continuous deviation data for result verification. The level adjustment condition is considered true only when all records are in the "exceeded" status.

[0089] When transmitting intermediate results between multiple computing nodes, the sending node performs serialization according to a predefined data record structure, sequentially writing the record type, audit task identifier, business identifier, rule version identifier, number of fields, and the field name, data type, field length, field value, and exception flag for each field. The receiving node reads the record header in the same field order, first verifying the record type and number of fields, and then parsing the field values ​​according to the field name and data type. If the parsed rule version identifier is inconsistent with the local task context, the number of fields is inconsistent, the field length exceeds the receiving range, or a required field is missing, deserialization is stopped and a data transmission exception is generated, and the unparsed object is not handed over to the subsequent module.

[0090] The review and processing instructions are represented by an instruction record structure. The instruction record includes an instruction identifier, an review task identifier, a business identifier for the target procurement business, a target review level, a target review processing type, an exception type, a rule version identifier, and an instruction status. When a normal instruction is generated, the instruction generation unit queries the process mapping table using the target review level as the mapping key, reads the unique target review processing type, and writes a compliance release instruction type when the target review processing type is a compliance warning type, and writes a compliance warning instruction type when it is a compliance warning type. When the process mapping is not matched or is matched repeatedly, a processing type mapping exception is formed, and the instruction is changed to be generated as a pending verification instruction.

[0091] When generating intelligent audit results for procurement compliance, the result generation unit uses the audit task identifier as the association key to sequentially read the integrated audit record, target audit point record, geometric calculation record, level determination record, audit processing instruction, and audit execution status. For each type of record read, the business identifier and rule version identifier are verified to be consistent with the audit task context. If they are consistent, they are written into the result object; if they are inconsistent, a result association anomaly is written and a pending verification status is generated. After all necessary records are associated, the result generation unit sets the result status to "formed." If there is an abnormal branch, the result status is set to "pending verification." If the approval process node return status has not yet been obtained, the result status is kept as "processing," and the "processing" status is not recorded as a completed audit result.

[0092] Below is an example of an end-to-end operation without specific numerical values: After receiving the audit task for the target procurement business, the acquisition module obtains the logistics waybill data, warehouse outbound data, and financial reconciliation data corresponding to the business, and confirms that the records from the three sources all have waybill identifiers, outbound box numbers, batch identifiers, or reconciliation detail identifiers that can establish a connection; The fusion module converts the weight fields from different sources to the same weight measurement unit, retains the amount field and its currency and exchange rate conversion details in the standardized record, and then locates the target logistics waybill record by waybill identifier, locates the target warehouse outbound record by outbound box number and batch identifier, and locates the target financial reconciliation record by waybill identifier and reconciliation detail identifier, forming a fused audit record.

[0093] The fusion module obtains the recorded weight and chargeable weight from the fusion audit records, calculates the difference between the two, the absolute value of the difference, and the ratio of the absolute value to the non-zero weight difference benchmark value to obtain the weight difference characteristic value; the fusion module converts the invoice freight and actual freight to the same currency based on the exchange rate conversion details, calculates the difference between the benchmark invoice freight and the benchmark actual freight, the absolute value of the difference, and the ratio of the absolute value to the non-zero freight difference benchmark value to obtain the freight difference characteristic value; the two characteristic values ​​become the horizontal axis coordinate value and the vertical axis coordinate value of the target audit point, respectively.

[0094] like Figure 2 As shown, the positioning module compares the target audit point with the horizontal and vertical boundaries of each candidate benchmark region under the current rule version to determine the unique target benchmark region containing the target audit point and obtain the region identifier. The analysis module uses the region identifier to obtain the first compliance benchmark point and the second compliance benchmark point. After confirming that the two points do not coincide, it obtains the benchmark direction vector based on the coordinate difference between the two points, and obtains the target displacement vector based on the coordinate difference between the first compliance benchmark point and the target audit point. Then, it obtains the squared value of the benchmark direction length, the direction correlation value, and the projection scale value.

[0095] like Figure 3As shown, the analysis module obtains the target projection vector based on the projection scale value and the reference direction vector. It adds the target projection vector to the first compliance reference point to obtain the perpendicular foot point. Then, it obtains the vertical distance based on the two coordinate differences between the target audit point and the perpendicular foot point, and determines the projection scale value as the positioning value along the line. The adjustment module determines the vertical distance as the reference deviation value and the positioning value along the line as the column positioning value. It compares the two values ​​with the deviation value interval and positioning value interval corresponding to the target reference area to obtain a unique row index and column index.

[0096] The adjustment module obtains a two-dimensional mapping table based on the region identifier, locates the target mapping unit using row and column indices, reads the candidate review level, and generates a level determination record. The system then reads the baseline deviation values ​​for a preset number of consecutive times within the same review cycle and compares them one by one with the fixed distance value. If all baseline deviation values ​​are greater than the fixed distance value and the candidate review level is not at the end of the review level order, the candidate review level is adjusted to the adjacent previous review level. If a baseline deviation value is less than or equal to the fixed distance value, or the candidate review level is at the end of the order, the candidate review level remains unchanged.

[0097] The execution module obtains the target audit processing type based on the target audit level, generates a compliance release instruction or a compliance warning instruction, and sends it to the approval process node. After the approval process node performs the corresponding processing, it returns the audit execution status. The execution module integrates the audit record, target audit level, audit processing instruction, and audit execution status to form a procurement compliance intelligent audit result. Each intermediate result in this running example is obtained from the output of the previous step. The target benchmark area determines the compliance benchmark point and compliance benchmark line, the vertical distance and the positioning value along the line determine the two-dimensional mapping position, and the candidate audit level and continuous deviation data jointly determine the target audit level.

[0098] In the case of data anomalies, when the acquisition module or fusion module discovers that the data used to determine the target audit point is missing, it still saves the logistics waybill data, warehouse outbound data, and financial reconciliation data that have already been obtained, and marks the missing fields, data sources, and related identifiers in the fusion audit record; the system does not replace the missing record weight, chargeable weight, billed freight, actual freight, or exchange rate conversion information with 0, because substituting 0 may create difference feature values ​​that appear valid but do not have real business meaning; the execution module generates a verification instruction based on this, and the approval process node is transferred to the verification process.

[0099] In the implementation of normalization anomalies, when the weight difference benchmark value or freight difference benchmark value is equal to 0, the denominator of the corresponding ratio is invalid, and the system will not continue to obtain the weight difference feature value or freight difference feature value; in the implementation of coordinate anomalies, when two compliant benchmark points coincide, the square value of the benchmark direction length is equal to 0, and the system will not perform the division between the direction association value and the square value of the benchmark direction length; both of the above situations will generate calculation anomaly results, and the calculation anomaly results will explain the anomaly parameters, the steps in which the anomaly occurred, and the current rule version.

[0100] In the implementation of regional anomalies, if the target audit point does not fall into any candidate benchmark area, or falls into multiple candidate benchmark areas formed due to configuration overlap, the positioning module outputs a regional matching anomaly result. In the implementation of interval anomalies, if the benchmark deviation value or column positioning value does not hit the interval, or hits more than two intervals, the adjustment module outputs a calculation anomaly result. In the implementation of mapping anomalies, if the target mapping unit does not exist, the index is out of range, or the mapping unit has no audit level, the adjustment module outputs a calculation anomaly result. The execution module uniformly generates verification instructions for these anomalies, and retains the specific anomaly type in the intelligent audit results of procurement compliance.

[0101] When the target audit point is located on the boundary of the candidate benchmark area, the positioning module reads the attribution mark of the boundary. Only the attribution mark indicates that the candidate benchmark area containing the boundary can match the point. When the benchmark deviation value or column positioning value is located on the common boundary of adjacent value intervals, the adjustment module reads the interval boundary attribution mark in the same way to obtain a unique row index or column index. When the positioning value along the line is equal to 0, the column index is determined according to the zero value interval or the positioning value interval containing 0. Through explicit boundary attribution, the system does not need to rely on the numerical comparison order to determine the category of the boundary point.

[0102] When an approval process node is temporarily unable to complete instruction processing, the execution module records the review execution status as pending retry, while keeping the review task identifier, instruction identifier, and instruction content unchanged. During subsequent retries, the execution module queries existing execution records based on the instruction identifier. Completed instructions are not executed repeatedly, while incomplete instructions continue to be sent to the same type of process node. This process ensures that the process execution status corresponds to the determined target review level, preventing communication or node processing anomalies from causing conflicting execution statuses for the same review task.

[0103] The acquisition module may include a task receiving unit, an interface access unit, and a data verification unit. The task receiving unit is used to receive the business identifier and audit task identifier of the target procurement business. The interface access unit is used to read logistics waybill data, warehouse outbound data, and financial reconciliation data according to the business identifier. The data verification unit is used to check the data source, association identifier, and required fields, and encapsulates the check results together with the three types of data into multi-source business data. The output of the acquisition module is directly used as the input of the fusion module.

[0104] The fusion module can include a standardization unit, a record association unit, and a difference feature unit. The standardization unit is used to unify field names, data formats, weight measurement units, and currency based on field mapping relationships. The record association unit is used to generate target logistics waybill records, target warehouse outbound records, and target financial reconciliation records according to waybill identifiers, outbound box numbers, batch identifiers, and reconciliation detail identifiers, and merges the three types of records into a fusion audit record. The difference feature unit is used to obtain weight difference feature values ​​and freight difference feature values ​​based on the fusion audit record.

[0105] The positioning module may include a coordinate construction unit and a region matching unit. The coordinate construction unit is used to read coordinate system configuration data and construct the target audit point with weight difference feature value and freight difference feature value. The region matching unit is used to read multiple non-overlapping candidate reference regions, compare the horizontal axis coordinate value and vertical axis coordinate value of the target audit point with the region boundary, and output the target reference region and region identifier or region matching anomaly result.

[0106] The analysis module may include a baseline determination unit and a geometric analysis unit. The baseline determination unit is used to obtain the first compliance benchmark point and the second compliance benchmark point according to the area identification, and to verify that the two points do not overlap to obtain the compliance benchmark straight line. The geometric analysis unit is used to form the benchmark direction vector and the target displacement vector, and to obtain the benchmark direction length square value, direction association value, projection scale value, target projection vector, perpendicular foot point, vertical distance and positioning value along the line in sequence.

[0107] The adjustment module may include an interval positioning unit, a mapping query unit, and a continuous adjustment unit. The interval positioning unit is used to determine the vertical distance as the baseline deviation value, the positioning value along the line as the column positioning value, and obtain the row index and column index according to the corresponding value interval. The mapping query unit is used to query the two-dimensional mapping table with the row index and column index to obtain the candidate review level. The continuous adjustment unit is used to obtain continuous deviation data and compare it with the fixed distance value one by one to determine the target review level.

[0108] The execution module may include an instruction generation unit, a process control unit, and a result generation unit. The instruction generation unit is used to generate compliance release instructions or compliance warning instructions based on the target review level, and to generate verification instructions based on regional matching abnormal results or calculation of abnormal results. The process control unit is used to send instructions to approval process nodes and receive the review execution status. The result generation unit is used to associate the integrated review records, target review level, review processing instructions, and review execution status into intelligent review results for procurement compliance.

[0109] The aforementioned units can be implemented by software programs, dedicated logic circuits, or a combination of software programs and logic circuits. The division of modules or units is used to describe functional relationships and does not restrict physical deployment boundaries. While keeping the input-output relationships unchanged, multiple units in a module can be executed by the same processor, and different modules can be deployed on multiple computing nodes. Regardless of the deployment method, the system transmits intermediate results according to the aforementioned data dependencies.

[0110] In one device implementation, the electronic device includes a processor, a memory, and a communication interface. The processor is connected to both the memory and the communication interface. The communication interface is used to receive logistics waybill data, warehouse outbound data, and financial reconciliation data, and to send audit processing instructions to approval process nodes and receive audit execution status. The memory is used to store audit rule base data, fused audit records, intermediate calculation results, level determination records, audit processing instructions, audit execution status, and computer programs that can be executed by the processor. When the processor executes the computer program, it performs the above-mentioned acquisition, fusion, location, analysis, adjustment, and execution processing.

[0111] The processor can assign an independent processing context to each target procurement business based on the audit task identifier. The processing context stores the current step status, rule version identifier, and intermediate results that have been generated. When multiple audit tasks are processed in parallel, different processing contexts are isolated by the audit task identifier. Continuous deviation data of the same target procurement business are associated in a controlled manner through the business identifier, region identifier, audit cycle identifier, and rule version identifier. This allows for parallel processing of different procurement businesses without mixing target audit points, indexes, or level records of different tasks.

[0112] In one chip implementation, the chip includes a processing circuit and an interface circuit. The interface circuit is used to receive multi-source business data and audit rule data, and output audit processing instructions or intelligent audit results for procurement compliance. The processing circuit is used to complete field standardization, record association, difference calculation, coordinate positioning, region matching, line determination, projection analysis, interval positioning, two-dimensional mapping query, and continuous level adjustment according to the above steps. The chip can be installed in procurement business management equipment, audit server, or computing device capable of executing approval process control.

[0113] In one implementation of a computer-readable storage medium, the computer-readable storage medium stores a computer program or computer instructions. When the processor reads and executes the computer program or computer instructions, the program reads the output of the previous step and generates the input of the next step according to the step status record. When encountering data missing, zero denominator, region mismatch, reference point coincidence, interval mismatch or empty mapping unit, it enters the corresponding abnormal branch and generates a verification instruction.

[0114] In one implementation of a computer program product, the computer program product includes a computer program. When the computer program is executed by a processor, it controls an electronic device to obtain multi-source business data of the target procurement business, form a fused audit record and two difference feature values, determine the target audit point and its target benchmark area, calculate the benchmark deviation value and column positioning value of the target audit point relative to the corresponding compliance benchmark line of the area, query a two-dimensional mapping table and perform continuous comparisons, and finally control the approval process nodes to obtain the intelligent audit result of procurement compliance.

[0115] In one optional implementation, the acquisition module, fusion module, and positioning module are deployed on the data processing node, the analysis module and adjustment module are deployed on the audit calculation node, and the execution module is deployed on the process control node. The data processing node sends the fused audit record, weight difference feature value, freight difference feature value, and target audit point to the audit calculation node. The audit calculation node sends the target audit level, level determination record, or abnormal result to the process control node. The data transmitted between each node carries the audit task identifier and rule version identifier. The receiving node verifies that the identifiers are consistent before continuing processing.

[0116] In another optional implementation, six modules are deployed within the same electronic device, with each module passing intermediate results through task objects in memory. The task objects are written sequentially with multi-source business data, integrated audit records, target audit points, regional identifiers, compliance baselines, baseline deviations, column positioning values, row indexes, column indexes, candidate audit levels, target audit levels, and audit execution status. Fields in the task objects that have not yet been generated remain in an incomplete state and are not replaced with default values, thus distinguishing between the actual value of 0 and the case where no data has been obtained.

[0117] In the absence of logical conflicts, the technical features in the above implementation methods can be combined with each other; the name, deployment location or communication method of the processing module can be adjusted according to the equipment architecture, but the processing relationships from multi-source business data to fused audit records, from fused audit records to differential feature values, from differential feature values ​​to target audit points, from target audit points to target benchmark areas, from target benchmark areas to compliance benchmark lines, from compliance benchmark lines to benchmark deviation values ​​and column positioning values, from two positioning values ​​to two-dimensional mapping levels, from candidate audit levels to target audit levels, and from target audit levels to approval process control remain unchanged.

[0118] The above content describes a complete implementation path for a procurement compliance intelligent audit system. Those skilled in the art can implement the system based on the disclosed field mapping, association key, difference normalization, regional boundary comparison, vector projection, interval positioning, two-dimensional mapping query, continuous comparison, and abnormal branching. Without changing the above technical processing chain and its input-output relationship, modules can be merged, split, or implemented using equivalent software, hardware, or a combination of software and hardware to form corresponding specific implementation methods.

[0119] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A procurement compliance intelligent audit system based on multi-source data fusion, characterized in that, include: The acquisition module is used to acquire logistics waybill data, warehouse outbound data, and financial reconciliation data of the target procurement business to obtain multi-source business data; The fusion module is used to standardize and correlate multi-source business data to obtain fusion audit records; based on the weight difference and freight difference in the fusion audit records, the weight difference feature value and freight difference feature value are obtained. The positioning module is used to obtain the target audit point based on the position of the weight difference feature value and the freight difference feature value in the audit feature coordinate system, and to obtain the target reference area based on the coordinate range of the target audit point. The analysis module is used to obtain the compliance benchmark line based on the first and second compliance benchmark points corresponding to the target benchmark area; Based on the positional relationship between the target audit point and the compliance baseline line, the vertical distance and the positioning value along the line are obtained; The adjustment module is used to determine the vertical distance as the baseline deviation value and the positioning value along the line as the column positioning value. Based on the value range of the two values, the row index and column index are obtained. The two-dimensional mapping table is queried based on the row index and column index to obtain the candidate review level. The execution module is used to adjust the candidate review level based on the continuous comparison results between the benchmark deviation value and the fixed distance value, so as to obtain the target review level; Based on the target review level, a review processing instruction is generated, and the corresponding review processing is executed at each node of the approval process to obtain the intelligent review result of the procurement compliance of the target procurement business.

2. The intelligent procurement compliance audit system based on multi-source data fusion as described in claim 1, characterized in that, The fusion audit record includes: Based on the mapping relationship between each data field and the standard field, the field names, data formats, weight units, and currency of the multi-source business data are unified to obtain standardized business data; The target logistics waybill record is obtained based on the waybill identifier in the standardized business data; Based on the outbound box number and batch identifier in the target logistics waybill record, obtain the target warehouse outbound record; based on the waybill identifier and reconciliation detail identifier in the target logistics waybill record, obtain the target financial reconciliation record. The target logistics waybill record, target warehouse outbound record, and target financial reconciliation record are merged in the same field to obtain the integrated audit record.

3. The intelligent procurement compliance audit system based on multi-source data fusion as described in claim 2, characterized in that, The weight difference characteristic value and the freight difference characteristic value include: Based on the integrated audit records, the recorded weight and billing weight using the same weight measurement unit are obtained; The weight difference value is obtained by comparing the recorded weight with the billed weight. The weight difference characteristic value is obtained by comparing the absolute value of the weight difference with the non-zero weight difference baseline value. Based on the exchange rate conversion details in the integration audit record, the bill shipping fee and actual shipping fee are converted to the same currency to obtain the base bill shipping fee and base actual shipping fee; The freight difference value is obtained by comparing the base invoice freight cost with the base actual freight cost. The characteristic value of freight difference is obtained by comparing the absolute value of the freight difference with the non-zero freight difference benchmark value.

4. The intelligent procurement compliance audit system based on multi-source data fusion according to claim 3, characterized in that, The target reference region includes: Based on the coordinate system configuration data in the audit rule base, an audit feature coordinate system is obtained. The horizontal axis of the audit feature coordinate system represents the weight difference feature value, and the vertical axis represents the freight difference feature value. The target audit point is obtained based on the position of the weight difference characteristic value on the horizontal axis and the position of the freight difference characteristic value on the vertical axis; Based on the region boundary data in the audit rule base, multiple non-overlapping candidate benchmark regions are obtained; By comparing the horizontal and vertical coordinates of the target audit point with the horizontal and vertical boundaries of each candidate benchmark region, a candidate benchmark region including the target audit point is obtained. Based on the candidate benchmark area including the target audit point, the target benchmark area and its corresponding area identifier are obtained.

5. The intelligent procurement compliance audit system based on multi-source data fusion according to claim 4, characterized in that, The compliance benchmark line includes: Based on the regional identifier of the target benchmark region, query the audit rule base to obtain the first compliance benchmark point and the second compliance benchmark point corresponding to the target benchmark region; Position verification is performed based on the coordinates of the first and second compliance benchmarks. When the horizontal or vertical coordinate values ​​of the two compliance benchmarks are not equal, a valid benchmark result is obtained. Based on the valid results of the benchmark points and the changes in the horizontal and vertical positions between the two compliant benchmark points, the benchmark direction is obtained; Based on the first compliance benchmark point, the second compliance benchmark point, and the benchmark direction, the compliance benchmark line corresponding to the target benchmark area is obtained.

6. The intelligent audit system for procurement compliance based on multi-source data fusion as described in claim 5, characterized in that, The vertical distance and the positioning value along the line include: Based on the difference in coordinate positions between the first compliance benchmark point and the second compliance benchmark point, the benchmark direction vector is obtained; Based on the difference in coordinate positions between the first compliance benchmark point and the target audit point, the target displacement vector is obtained; The squared value of the reference direction length is obtained by summing the squares of the components of each direction in the reference direction vector. The direction correlation value is obtained by summing the products of the target displacement vector and the corresponding direction components in the reference direction vector; The projection scale value is obtained by the ratio between the direction correlation value and the square of the reference direction length. The target projection vector is obtained based on the projection scale value and the reference direction vector; The perpendicular point is obtained based on the first compliance benchmark point and the target projection vector; The vertical distance is obtained based on the distance between the target audit point and the perpendicular point, and the positioning value along the line is determined based on the projection scale value.

7. The intelligent audit system for procurement compliance based on multi-source data fusion as described in claim 6, characterized in that, The row index and column index include: Based on the vertical distance, a non-negative and dimensionless benchmark deviation value is obtained, which represents the degree to which the target audit point deviates from the compliance benchmark line corresponding to the target benchmark area; Based on the positioning values ​​along the line, the value range is determined. The value range includes a negative value range, a zero value range, and a positive value range. The column index is obtained according to the interval mapping. The column index is a non-negative integer and is used to distinguish the different directions and near and far positions of the perpendicular point relative to the first compliance reference point in the two-dimensional mapping table. Based on the regional identifier of the target benchmark area, query the audit rule base to obtain multiple deviation value ranges and multiple positioning value ranges corresponding to the target benchmark area; The row index is obtained based on the deviation range in which the baseline deviation value falls; The column index is obtained based on the range of column positions where the column position value is located.

8. The intelligent audit system for procurement compliance based on multi-source data fusion according to claim 7, characterized in that, The candidate review levels include: Based on the region identifier of the target reference region, a two-dimensional mapping table corresponding to the target reference region is obtained; Based on multiple deviation value intervals, the positions of multiple rows in the two-dimensional mapping table are obtained; Based on multiple location value ranges, the positions of multiple columns in the two-dimensional mapping table are obtained; Based on the intersection of each row position and each column position, multiple mapping units are obtained, and each mapping unit stores the corresponding review level. Locate the target mapping unit in the two-dimensional mapping table based on the row index and column index; Based on the review levels stored in the target mapping unit, candidate review levels are obtained; Based on the mapping unit position corresponding to the candidate review level, the level determination record is obtained.

9. The intelligent audit system for procurement compliance based on multi-source data fusion as described in claim 8, characterized in that, The target review level includes: Based on the regional identifier of the target benchmark area, query the audit rule base to obtain a fixed distance value that remains unchanged within the same audit cycle; Based on the benchmark deviation value obtained from the target procurement business in a series of predetermined audits, continuous deviation data is obtained. Based on the comparison of continuous deviation data with the fixed distance value, the result is that the level adjustment condition is met when each benchmark deviation value is greater than the fixed distance value; Based on the results of the grade adjustment conditions and the preset review grade order, the candidate review grade is adjusted to the adjacent previous review grade to obtain the target review grade; When the candidate review level is already at the end of the review level sequence, the target review level is obtained based on the candidate review level; When there is a baseline deviation value in the continuous deviation data that is less than or equal to the distance from the fixed value, the target audit level is obtained based on the candidate audit level.

10. The intelligent audit system for procurement compliance based on multi-source data fusion according to claim 9, characterized in that, Based on the correspondence between the target review level and the review processing type, the target review processing type is obtained; Depending on the target review and processing type, a compliance release instruction or a compliance warning instruction will be obtained; Based on the positional relationship between the target audit point and each candidate benchmark region, when there is no candidate benchmark region containing the target audit point, an abnormal region matching result is obtained; Based on the fusion audit records, the squared value of the baseline direction length, and the target mapping unit, anomaly results are obtained when the data used to determine the target audit point is missing, the squared value of the baseline direction length is equal to 0, or the target mapping unit does not store the audit level. Based on the abnormal results of regional matching or calculation, the verification instruction is obtained; Based on compliance release instructions, compliance warning instructions, or pending verification instructions, control the execution of corresponding review processes at each node of the approval process to obtain the review execution status; Based on the integrated audit records, target audit level, corresponding audit processing instructions, and audit execution status, the intelligent audit results for the procurement compliance of the target procurement business are obtained.

Images