Audit report generation method and system based on dynamic allocation engine

Abstract

The application provides an audit report generation method and system based on a dynamic allocation engine, relates to the technical field of audit data processing, and comprises the following steps: obtaining to-be-allocated data by acquiring original data and matching audit rules; performing space-time dimension decomposition and multi-level topological allocation on the to-be-allocated data to calculate node allocation values; identifying abnormalities based on time sequence evolution characteristics, backtracking and correcting the allocation values, correlating compliance states, and aggregating to generate a structured audit report. The application realizes automatic and refined allocation and abnormal traceability of audit data, and improves the accuracy and efficiency of report generation.

Description

Technical Field

[0001] This invention relates to the field of audit data processing technology, and in particular to an audit report generation method and system based on a dynamic allocation engine. Background Technology

[0002] In the field of audit report generation, existing technologies typically rely on static, rule-based data processing procedures that directly extract raw data from business systems, conduct compliance checks according to preset audit rules, and summarize and present the check results according to the static hierarchy of the organizational structure.

[0003] However, existing technologies have significant limitations. The generation and impact of audit data often exhibit complex spatiotemporal correlations and hierarchical transmission. Treating audit events as isolated points ignores the dynamic responsibility relationships between business entities and the coupling and transmission effects of data across organizational levels. Static aggregation methods cannot characterize the dynamic allocation and cascading impact processes, resulting in coarse audit attribution. Furthermore, reports generated by existing technologies are mostly flat lists of items or simple hierarchical summaries, lacking the ability to analyze the evolution of abnormal patterns over time and trace their origins within the organizational structure. Report users struggle to quickly identify trending abnormal deviations from massive compliance items, let alone accurately pinpoint the root causes or links that trigger a series of compliance issues. The foresight and initiative in process intervention and decision support are also weak. Summary of the Invention

[0004] This invention provides an audit report generation method and system based on a dynamic allocation engine, which can at least solve some of the problems existing in the prior art.

[0005] A first aspect of this invention provides a method for generating audit reports based on a dynamic cost-sharing engine, comprising:

[0006] Obtain the mapping relationship between raw data and organizational structure, extract the business entity identifier and business event identifier corresponding to the raw data, determine the hierarchical and responsibility attribution relationships corresponding to the business entity identifier based on the organizational structure mapping relationship, match the business event identifier with the preset audit rule base and mark the compliance status to obtain the data set to be allocated;

[0007] The data set to be allocated is decomposed into time dimension features and spatial dimension features. Based on the time dimension features, the data set to be allocated is sliced ​​into segmented data units. Based on the spatial dimension features and the hierarchical relationship, a multi-level allocation topology is constructed. The segmented data units are coupled and solved according to the multi-level allocation topology and decomposed layer by layer to obtain the node allocation value.

[0008] Based on the node allocation value and the responsibility attribution relationship, extract all node allocation values ​​of the target audit subject and perform frequency domain transformation to extract time-series evolution features. Calculate the abnormal deviation metric value based on the time-series evolution features and backtrack to determine the influencing source node in conjunction with the multi-level allocation topology. Iteratively correct and decompose the transmission attenuation factor corresponding to the influencing source node to obtain the corrected allocation value. Associate the corrected allocation value and the compliance status to generate an audit item set and aggregate it according to the hierarchical attribution relationship to obtain a structured audit report.

[0009] In one alternative implementation,

[0010] Obtaining the mapping relationship between raw data and organizational structure, extracting the business entity identifiers and business event identifiers corresponding to the raw data, and determining the hierarchical and responsibility relationships corresponding to the business entity identifiers based on the organizational structure mapping relationship include:

[0011] The system obtains the original data from the data source of the business object to be audited and the organizational structure mapping relationship from the organizational structure management system. It parses the field structure of the data records in the original data and identifies the field type identifiers. Based on the field type identifiers, the data records are divided into entity attribute field sets and event attribute field sets. The system extracts the field values ​​corresponding to the entity attribute field sets to generate business entity identifiers. It performs timestamp parsing and event type matching on the event attribute field sets to generate business event identifiers. It performs multi-level matching on the business entity identifiers and the organizational structure mapping relationship and calculates the matching score. Based on the matching score, it determines the affiliated organizational node of the business entity identifier and extracts the corresponding hierarchical path. Based on the parent-child relationship between different nodes in the hierarchical path, it determines the hierarchical affiliation relationship.

[0012] Determine the set of superior responsibility nodes and the set of peer responsibility nodes of the affiliated organization node. Calculate the superior responsibility weight based on the hierarchical span between each node in the superior responsibility node set and the affiliated organization node. Calculate the peer responsibility weight based on the organizational affiliation path similarity between each node in the peer responsibility node set and the affiliated organization node. Construct a responsibility weight vector based on the superior responsibility weight and the peer responsibility weight, and generate responsibility affiliation relationships.

[0013] In one alternative implementation,

[0014] The business event identifier is matched with a preset audit rule base and its compliance status is marked to obtain the data set to be allocated, including:

[0015] The system extracts rule condition fields and rule constraint logic from a preset audit rule base, extracts the field values ​​of the event attribute fields in the business event identifier and maps them to the rule condition fields, determines the applicable rule set based on the field mapping results, parses the logical expressions of each rule in the applicable rule set, obtains the logical expressions, extracts the constraint conditions and constraint threshold ranges, substitutes the field values ​​corresponding to the business event identifier into the logical expressions to determine the constraint conditions and calculates the rule compliance degree, and generates a compliance status identifier based on the rule compliance degree.

[0016] The process involves querying the attributing organization node corresponding to the business event identifier and extracting the corresponding responsibility weight vector. The compliance status identifier is then associated and bound with the responsibility weight vector, and an event allocation record is constructed by combining the numerical field values ​​in the business event identifier. The event allocation records are grouped and aggregated according to the attributing organization node to obtain grouping results. Compliance status identifiers are extracted from the event allocation records within each group in the grouping results, and compliance filtering is performed. Finally, the event allocation records that pass the compliance filtering are summarized to obtain the set of data to be allocated.

[0017] In one alternative implementation,

[0018] The data set to be allocated is decomposed into temporal and spatial features. Based on the temporal features, the data set to be allocated is time-series sliced ​​to obtain segmented data units. A multi-level allocation topology is constructed based on the spatial features and the hierarchical relationship, including:

[0019] The timestamp field of the data records in the data set to be allocated is extracted and parsed into a standard time format. The time interval is calculated based on the standard time format and the time period pattern is identified to obtain the time dimension feature. The organization node identifier corresponding to the business entity identifier of the data record in the data set to be allocated is extracted. The organization hierarchy depth and organization affiliation path are extracted based on the organization node identifier to obtain the spatial dimension feature. The time slice boundary is determined based on the time dimension feature and the data set to be allocated is divided into multiple time periods according to the time slice boundary. The data records in each time period are aggregated and calculated, and the numerical fields are extracted to generate segmented data units.

[0020] Extract the hierarchical node identifiers corresponding to each business entity identifier from the hierarchical affiliation relationship and construct a hierarchical node set based on the hierarchical node identifiers. Based on the spatial dimension features, classify the hierarchical node identifiers in the hierarchical node set into layers and mark the hierarchical positions. Based on the hierarchical positions and the organizational affiliation path, determine the hierarchical node identifiers with the same superior node and establish a peer-level association relationship. Based on the hierarchical positions and the hierarchical affiliation relationship, establish a hierarchical transmission relationship between different levels and construct a multi-level distribution topology structure by combining the peer-level association relationship.

[0021] In one alternative implementation,

[0022] Based on the multi-level amortization topology, the segmented data units are coupled and solved, and the node amortization values ​​are obtained by decomposing them layer by layer, including:

[0023] Node identifiers and layer depths are extracted from the multi-level allocation topology to determine the initial distribution position of the segmented data units in the multi-level allocation topology. Responsibility weight vectors are extracted from the responsibility attribution relationship and a layer weight tensor is constructed according to the pre-obtained layer position. Tensor decomposition is performed on the layer weight tensor to extract the core tensor and factor matrix. Based on the factor matrix, the intra-layer coupling coefficient and inter-layer coupling coefficient are calculated and combined with the core tensor to reconstruct the global coupling weight tensor.

[0024] The values ​​of the segmented data units are loaded at the bottom-level nodes according to the initial distribution positions to construct an initial state vector. The initial state vector is then subjected to tensor contraction operation with the global coupling weight tensor to obtain a coupled state field. The coupled state field is then subjected to Laplace diffusion, and diffusion boundary conditions are set based on the sibling relationships in the multi-level amortization topology. The steady-state field distribution after diffusion equilibrium is solved to obtain the coupled amortization value. The coupled amortization value is then integrated and aggregated layer by layer according to the hierarchical structure of the multi-level amortization topology to obtain an intermediate amortization value. The inter-layer gradient operator is extracted from the global coupling weight tensor, and the intermediate amortization value is then subjected to variational descent decomposition starting from the top-level node based on the inter-layer gradient operator to obtain the node amortization value.

[0025] In one alternative implementation,

[0026] Based on the node allocation value and the responsibility attribution relationship, extract all node allocation values ​​of the target audit subject and perform frequency domain transformation to extract time-series evolution features. Based on the time-series evolution features, calculate the abnormal deviation metric value and combine it with the multi-level allocation topology to backtrack and determine the influencing source nodes, including:

[0027] Extract the allocated value sequence corresponding to the target audit subject from the allocated value of the node and construct the audit subject time series matrix. Perform wavelet transform decomposition on the audit subject time series matrix to obtain multi-scale wavelet coefficients and extract instantaneous frequency and instantaneous amplitude through Hilbert transform. Calculate the frequency drift rate based on the instantaneous frequency and the amplitude envelope based on the instantaneous amplitude. Combine the frequency drift rate with the time-frequency plane to obtain the time series evolution characteristics.

[0028] Based on the attribution of responsibility and the audit subject's time series matrix, a baseline evolution trajectory is calculated. The time series evolution features are decomposed into singular spectra and the dominant singular vector is extracted. The subspace angle between the dominant singular vector and the feature vector corresponding to the baseline evolution trajectory is calculated, and the abnormal deviation metric is obtained by combining the Hausdorff distance between the time series evolution features and the baseline evolution trajectory.

[0029] From the multi-level allocation topology, the associated nodes corresponding to the target audit entity are extracted, the node allocation value corresponding to the associated node is extracted and the mutual information value with the time series matrix of the audit entity is calculated, the associated nodes with mutual information values ​​exceeding a preset threshold are selected to construct a candidate impact source set, the responsibility weights corresponding to each edge on the transmission path from each associated node in the candidate impact source set to the node corresponding to the target audit entity are extracted and the cumulative weight value is calculated, and the impact contribution is calculated in combination with the abnormal deviation metric value to determine the impact source node.

[0030] In one alternative implementation,

[0031] The propagation attenuation factor corresponding to the source node is iteratively corrected and decomposed to obtain a corrected allocation value. The corrected allocation value and the compliance status are associated to generate an audit item set, which is then aggregated according to the hierarchical relationship to obtain a structured audit report, including:

[0032] Extract the transmission path from the source node to the target audit entity node from the multi-level allocation topology, extract the initial transmission attenuation factor corresponding to each edge of the transmission path from the responsibility attribution relationship, calculate the correction gradient based on the abnormal deviation metric and the pre-acquired impact contribution, iteratively update the initial transmission attenuation factor according to the correction gradient, and perform exponential attenuation correction based on the hierarchical span of the transmission path to obtain the hierarchical correction transmission attenuation factor.

[0033] The node allocation value corresponding to the source node of the influence is passed layer by layer according to the transmission path, and the hierarchical correction transmission attenuation factor is applied at each layer for attenuation calculation. Based on the same-level association relationship in the multi-level allocation topology, horizontal diffusion correction is performed, and the corrected allocation value is obtained by aggregating at the target audit subject node.

[0034] The modified allocation value is associated with the compliance determination result in the compliance status, and the influence source node identifier is added to construct the audit entry. The audit entries are grouped and aggregated according to the hierarchical relationship and sorted in descending order according to the abnormal deviation metric to obtain the structured audit report.

[0035] A second aspect of this invention provides an audit report generation system based on a dynamic allocation engine, comprising:

[0036] The data extraction unit is used to obtain the mapping relationship between raw data and organizational structure, extract the business entity identifier and business event identifier corresponding to the raw data, determine the hierarchical and responsibility attribution relationships corresponding to the business entity identifier based on the organizational structure mapping relationship, and match the business event identifier with the preset audit rule base and mark the compliance status to obtain the data set to be allocated.

[0037] The apportionment calculation unit is used to perform dimensional decomposition on the data set to be apportioned to obtain time dimension features and spatial dimension features, perform time-series slicing on the data set to be apportioned based on the time dimension features to obtain segmented data units, construct a multi-level apportionment topology based on the spatial dimension features and the hierarchical affiliation relationship, and perform coupled solution on the segmented data units according to the multi-level apportionment topology and decompose layer by layer to obtain the node apportionment value.

[0038] The audit generation unit is used to extract all node allocation values ​​of the target audit subject based on the node allocation value and the responsibility attribution relationship, and perform frequency domain transformation to extract time-series evolution features. Based on the time-series evolution features, it calculates the abnormal deviation metric value and combines it with the multi-level allocation topology to backtrack and determine the influencing source node. It iteratively corrects and decomposes the transmission attenuation factor corresponding to the influencing source node to obtain the corrected allocation value. It associates the corrected allocation value and the compliance status to generate an audit item set and aggregates it according to the hierarchical attribution relationship to obtain a structured audit report.

[0039] A third aspect of the present invention provides an electronic device, comprising:

[0040] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0041] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0042] In this invention, the combination of raw data and organizational structure mapping enables precise matching of business entity identifiers and event identifiers with audit rules, thereby automatically marking compliance status and forming a data set to be allocated. This avoids data omissions or misjudgments caused by reliance on manual verification, ensuring the integrity and reliability of the audit foundation. The data set to be allocated is decomposed in terms of time and space dimensions, and a multi-level allocation topology is constructed, realizing the structured processing of complex audit data. The application of time-series slicing and coupled solution techniques allows segmented data units to be decomposed into node allocation values ​​layer by layer according to hierarchical relationships. This not only reflects the real flow path of data within the organization but also quantifies the responsibility contribution of each level of node, providing refined data support for subsequent audit analysis. Based on the node allocation values, time-series evolution characteristics are extracted, and through the calculation and backtracking analysis of abnormal deviation metrics, the source nodes of influence can be automatically identified, the correlation between allocation values ​​and compliance status can be corrected, and a set of items with clear audit conclusions can be formed. Finally, these items are aggregated into a structured audit report, achieving full automation from data collection to report generation, significantly reducing the need for manual intervention, and enhancing the objectivity and traceability of audit conclusions. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating the audit report generation method based on a dynamic allocation engine according to an embodiment of the present invention.

[0044] Figure 2 This is a flowchart illustrating the analysis of the relationship between node allocation values ​​and responsibility attribution in the audit report generation method based on a dynamic allocation engine, as described in an embodiment of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0047] Figure 1 This is a flowchart illustrating the audit report generation method based on a dynamic cost-sharing engine according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0048] Obtain the mapping relationship between raw data and organizational structure, extract the business entity identifier and business event identifier corresponding to the raw data, determine the hierarchical and responsibility attribution relationships corresponding to the business entity identifier based on the organizational structure mapping relationship, match the business event identifier with the preset audit rule base and mark the compliance status to obtain the data set to be allocated;

[0049] The data set to be allocated is decomposed into time dimension features and spatial dimension features. Based on the time dimension features, the data set to be allocated is sliced ​​into segmented data units. Based on the spatial dimension features and the hierarchical relationship, a multi-level allocation topology is constructed. The segmented data units are coupled and solved according to the multi-level allocation topology and decomposed layer by layer to obtain the node allocation value.

[0050] Based on the node allocation value and the responsibility attribution relationship, extract all node allocation values ​​of the target audit subject and perform frequency domain transformation to extract time-series evolution features. Calculate the abnormal deviation metric value based on the time-series evolution features and backtrack to determine the influencing source node in conjunction with the multi-level allocation topology. Iteratively correct and decompose the transmission attenuation factor corresponding to the influencing source node to obtain the corrected allocation value. Associate the corrected allocation value and the compliance status to generate an audit item set and aggregate it according to the hierarchical attribution relationship to obtain a structured audit report.

[0051] In one alternative implementation,

[0052] Obtaining the mapping relationship between raw data and organizational structure, extracting the business entity identifiers and business event identifiers corresponding to the raw data, and determining the hierarchical and responsibility relationships corresponding to the business entity identifiers based on the organizational structure mapping relationship include:

[0053] The system obtains the original data from the data source of the business object to be audited and the organizational structure mapping relationship from the organizational structure management system. It parses the field structure of the data records in the original data and identifies the field type identifiers. Based on the field type identifiers, the data records are divided into entity attribute field sets and event attribute field sets. The system extracts the field values ​​corresponding to the entity attribute field sets to generate business entity identifiers. It performs timestamp parsing and event type matching on the event attribute field sets to generate business event identifiers. It performs multi-level matching on the business entity identifiers and the organizational structure mapping relationship and calculates the matching score. Based on the matching score, it determines the affiliated organizational node of the business entity identifier and extracts the corresponding hierarchical path. Based on the parent-child relationship between different nodes in the hierarchical path, it determines the hierarchical affiliation relationship.

[0054] Determine the set of superior responsibility nodes and the set of peer responsibility nodes of the affiliated organization node. Calculate the superior responsibility weight based on the hierarchical span between each node in the superior responsibility node set and the affiliated organization node. Calculate the peer responsibility weight based on the organizational affiliation path similarity between each node in the peer responsibility node set and the affiliated organization node. Construct a responsibility weight vector based on the superior responsibility weight and the peer responsibility weight, and generate responsibility affiliation relationships.

[0055] Raw data is obtained from the data sources of the configured auditable business objects. These data sources include various types such as financial system databases, business operation system logs, and data returned from third-party interfaces. Simultaneously, organizational structure mapping relationships are obtained from the organizational structure management system via interface calls. These mapping relationships include key fields such as department code, department name, superior department identifier, department level depth, and department head identifier, forming a complete organizational tree structure data.

[0056] The acquired raw data undergoes field structure parsing to read the metadata information of each field in the data record, including field name, field data type, field length, and field constraints. Field type identifiers are identified using preset field type mapping rules. For example, fields such as "department code," "employee number," and "supplier code" are identified as entity-type fields, while fields such as "transaction time," "operation type," and "approval process" are identified as event-type fields. Based on the identification results, all fields in the data record are divided into an entity attribute field set and an event attribute field set. The entity attribute field set describes the static characteristics of business entities, while the event attribute field set describes the dynamic characteristics of business activities.

[0057] The system extracts corresponding field values ​​from the entity attribute field set and generates business entity identifiers using field value concatenation or encoding conversion. When the entity attribute field set contains multiple fields, the field values ​​are concatenated according to preset combination rules. For example, "department code" and "employee number" are combined using a separator to form a unique business entity identifier, ensuring that the business entity identifier can uniquely locate a specific business entity within the entire audited business object scope. For the event attribute field set, the timestamp fields are parsed, and different time representations are uniformly converted to a standard time format, extracting time dimension information such as year, month, day, hour, minute, and second. Predefined event type matching is performed on event type-related fields, semantically matching the field values ​​with event definitions in the event type library to identify specific event types such as "purchase event," "payment event," and "reimbursement event." The time information and event type information are combined to generate a business event identifier, which not only identifies the time point of the event but also the business semantic category of the event.

[0058] The generated business entity identifier is matched against the organizational structure mapping relationship obtained from the organizational structure management system at multiple levels. At the lowest level of the organizational structure mapping relationship, a precise match is performed to determine if the business entity identifier is completely identical to the code of a specific lowest-level organizational node. If a precise match cannot be achieved, a fuzzy match is performed at each level of the organizational structure, calculating a matching score by evaluating multiple dimensions such as string similarity, encoding rule consistency, and business domain affiliation consistency between the business entity identifier and the identifiers of each organizational node. The matching score calculation considers the consistency weight of the encoding prefix, the correspondence weight of the business type, and the confidence weight of historical association records, and the weighted sum of these dimensions yields a comprehensive matching score.

[0059] Based on the calculated matching scores of each organizational node, the organizational node with the highest matching score is selected as the belonging organizational node corresponding to the business entity identifier. When multiple organizational nodes have the same or similar matching scores, the node with a deeper organizational hierarchy is preferentially selected as the belonging organizational node. After determining the belonging organizational node, the complete hierarchical path from that node to the root node is extracted from the organizational structure mapping relationship. The hierarchical path is an ordered sequence of nodes arranged in order from the root node to the belonging organizational node, recording the complete belonging chain of the business entity in the organizational structure.

[0060] Based on the extracted hierarchical path, the parent-child relationships between adjacent nodes in the path are analyzed, and the management relationship and data inheritance relationship of each parent node to its child nodes are recorded. The parent-child relationships of each node in the hierarchical path are organized into a tree-structured local view, forming a data representation of the hierarchical belonging relationship. The hierarchical belonging relationship records not only the direct parent-child node relationship, but also the indirect ancestor node relationship, which facilitates subsequent cross-level data aggregation and responsibility tracing.

[0061] The direct superior node of the belonging organization node in the organizational structure is determined as the superior responsibility node. The hierarchical path is recursively traversed upwards to extract all ancestor nodes of the belonging organization node, forming a set of superior responsibility nodes. For each node in the set of superior responsibility nodes, the hierarchical span between it and the belonging organization node is calculated. The hierarchical span represents the number of levels separating the two nodes in the hierarchical path. The superior responsibility weight is calculated based on the hierarchical span; a smaller hierarchical span indicates a more direct management relationship, and the corresponding superior responsibility weight is greater. A decreasing function is used to map the hierarchical span. For example, the direct superior node with a hierarchical span of 1 is assigned the highest weight. For each increase of 1 in the hierarchical span, the weight decreases according to an exponential or linear decay rule, ensuring that superior nodes closer to the belonging organization node bear a greater responsibility weight.

[0062] Identify the sibling organizational nodes of the parent organizational node. These sibling nodes are other child nodes in the organizational structure that share the same parent node, and are at the same level in the organizational hierarchy. Form a set of sibling responsibility nodes. For each node in this set, extract its complete hierarchical path from the root node and compare it with the hierarchical path of the parent organizational node to calculate the similarity between the two paths. The path similarity is calculated by considering multiple factors, including the number of common ancestor nodes, the consistency of path length, and the similarity of business attributes of the path nodes, reflecting the close relationship between the sibling node and the parent organizational node in the organizational structure. Based on the calculated organizational affiliation path similarity, calculate the sibling responsibility weight. A higher path similarity indicates a stronger business connection between the sibling node and the parent organizational node, and a greater sibling responsibility weight to be borne in collaborative or joint liability scenarios.

[0063] The calculated superior responsibility weights of each node in the set of superior responsibility nodes and the peer responsibility weights of each node in the set of peer responsibility nodes are normalized to ensure that the sum of the weights of all responsibility nodes is a preset value, for example, normalized to a sum equal to 1. The normalized node responsibility weights are then organized into a responsibility weight vector according to the node identifier order. This responsibility weight vector is a multi-dimensional vector, where each component corresponds to a responsibility node and its weight value. The dimension of the vector is equal to the total number of nodes in the sets of superior and peer responsibility nodes. Based on the constructed responsibility weight vector, a responsibility attribution relationship is generated. This relationship records the responsibility allocation ratio between the attributing organization node and each responsibility node in a structured form, providing a quantitative basis for subsequent audit responsibility allocation and traceability analysis. Through the responsibility attribution relationship, the proportion of audit responsibility that each level and peer organization node should bear when an abnormal or non-compliant event occurs in a business entity can be clearly identified, achieving precise decomposition and traceable management of audit responsibility.

[0064] In one alternative implementation,

[0065] The business event identifier is matched with a preset audit rule base and its compliance status is marked to obtain the data set to be allocated, including:

[0066] The system extracts rule condition fields and rule constraint logic from a preset audit rule base, extracts the field values ​​of the event attribute fields in the business event identifier and maps them to the rule condition fields, determines the applicable rule set based on the field mapping results, parses the logical expressions of each rule in the applicable rule set, obtains the logical expressions, extracts the constraint conditions and constraint threshold ranges, substitutes the field values ​​corresponding to the business event identifier into the logical expressions to determine the constraint conditions and calculates the rule compliance degree, and generates a compliance status identifier based on the rule compliance degree.

[0067] The process involves querying the attributing organization node corresponding to the business event identifier and extracting the corresponding responsibility weight vector. The compliance status identifier is then associated and bound with the responsibility weight vector, and an event allocation record is constructed by combining the numerical field values ​​in the business event identifier. The event allocation records are grouped and aggregated according to the attributing organization node to obtain grouping results. Compliance status identifiers are extracted from the event allocation records within each group in the grouping results, and compliance filtering is performed. Finally, the event allocation records that pass the compliance filtering are summarized to obtain the set of data to be allocated.

[0068] The system retrieves a pre-defined audit rule base, stored in a structured table format. This base includes core information such as rule number, rule type, applicable business scope, rule condition fields, rule constraint logic, and trigger thresholds. The system then parses the audit rule base, extracting the rule condition fields. These fields define the data fields to be checked when the rule takes effect, such as transaction amount, transaction time, counterparty, and business type code. Simultaneously, the system extracts the rule constraint logic, described using Boolean expressions or conditional statements. This logic includes logical operators such as greater than, less than, equal to, belong to, and not belong, as well as combinations of logical AND, logical OR, and logical NOT relationships.

[0069] For each pending business event identifier, the associated event attribute fields are read. These fields contain multi-dimensional information such as the business occurrence timestamp, business occurrence location identifier, business type code, business amount, business object identifier, and business description text. Field values ​​are extracted from these event attribute fields using a field name matching mechanism to obtain the actual numerical value or text content corresponding to each field. A field mapping relationship is established between the event attribute fields and rule condition fields, employing either exact name matching or semantic similarity matching. When field names are identical, a mapping relationship is established directly; when field names differ but semantics are the same, a connection is established using a predefined field mapping table.

[0070] Based on the field mapping results, an applicable rule set is selected, including all rules whose condition fields can be mapped to corresponding fields in the business event identifier. Each rule in the applicable rule set is processed individually, parsing the rule constraint logic into an executable logical expression structure. During parsing, constraints are identified, including single and compound conditions. Single conditions represent a comparison between a single field and a fixed value or threshold range, while compound conditions represent a combination of multiple single conditions connected by logical operators. The constraint threshold range is extracted from the logical expression: for numerical constraints, the upper and lower bounds are extracted; for enumerated constraints, the allowed value set is extracted; and for time-related constraints, the time window boundary is extracted.

[0071] Substitute the field values ​​corresponding to the business event identifier into the logical expression and perform constraint condition judgment. For each constraint, a truth value is evaluated; if the field value satisfies the constraint, the condition is judged as true; otherwise, it is false. For compound logical expressions containing multiple constraints, the truth value of the overall expression is calculated according to the rules of logical operators. During the constraint condition judgment process, judgment details are recorded, including the judgment results of each individual condition, the judgment result of the overall expression, and the specific reasons for non-compliance.

[0072] The rule compliance score is calculated based on the constraint determination results. A 100% compliance score is achieved when the business event identifier fully satisfies the rule constraints. If only some constraints are satisfied, the compliance score is calculated as the proportion of satisfied constraints to the total number of constraints. For constraints with weighted configurations, a weighting factor is introduced during compliance score calculation, and a weighted summation method is used to obtain the overall compliance score. A compliance threshold is set; if the rule compliance score reaches or exceeds this threshold, the business event identifier is determined to comply with the rule; otherwise, it is determined to be non-compliant.

[0073] A compliance status identifier is generated based on rule compliance. The compliance status identifier uses standardized coding and includes status types such as fully compliant, partially compliant, non-compliant, and undetermined. For a fully compliant status, a list of compliant rule numbers is appended to the compliance status identifier; for a partially compliant status, a list of compliant rule numbers and a list of non-compliant rule numbers are appended; for a non-compliant status, a list of non-compliant rule numbers and a description of the specific reasons for non-compliance are appended; for an undetermined status, a list of missing fields or a description of data anomalies is appended.

[0074] The system queries the business event identifier to locate the organizational node to which the business event belongs through organizational structure mapping. It then extracts the responsibility weight vector corresponding to that organizational node from the organizational structure data. This vector records the node's weight allocation within a multi-dimensional responsibility system, including financial responsibility weight, operational responsibility weight, and compliance responsibility weight. The sum of the values ​​of each component of the responsibility weight vector is typically normalized to 1, reflecting the relative importance of different responsibility dimensions.

[0075] A data association record is established by associating compliance status identifiers with responsibility weight vectors. This record includes a business event identifier, a compliance status identifier, a responsibility weight vector, and the identifier of the organization node to which the event belongs. Numerical field values ​​are extracted from the business event identifiers. These numerical fields include quantifiable business indicators such as business amount, number of transactions, and resource consumption. An event allocation record is constructed by combining the compliance status identifier, the responsibility weight vector, and the numerical field values. This event allocation record clearly records the amount to be allocated for each business event, the allocation weight of that amount in each responsibility dimension, the compliance determination result of the event, and the organization node to which the event belongs.

[0076] All event allocation records are grouped and aggregated according to their affiliated organizational nodes, with records belonging to the same organizational node grouped into the same group. During the grouping process, the integrity of the event allocation records is maintained, ensuring that the compliance status identifier, responsibility weight vector, and numerical field values ​​of each record are not lost or tampered with. For business events with multi-level organizational affiliations, the corresponding organizational node is selected for grouping based on the current processing level, ensuring that the grouping results are consistent with the organizational structure hierarchy.

[0077] Each group in the grouping results is processed, and the compliance status identifier is extracted for all event allocation records within the group. A compliance screening operation is performed, setting screening rules. Typically, only event allocation records with a compliance status identifier of fully compliant or partially compliant and a compliance degree reaching a preset threshold are retained. Records with a compliance status identifier of non-compliant or undetermined are individually marked or excluded. Compliance screening ensures that all data entering subsequent allocation calculations has been verified by audit rules, preventing non-compliant data from interfering with the allocation results.

[0078] All event allocation records that have passed the compliance screening across all groups are aggregated to form a dataset to be allocated. This dataset is organized in a structured data table format, with each record containing a business event identifier, the responsible organization node identifier, a compliance status identifier, a responsibility weight vector, and the numeric field value to be allocated. Each record in the dataset has undergone audit rule matching and compliance verification, providing compliance-approved foundational data for subsequent dimensional decomposition and allocation calculations, ensuring that the audit report generation process is based on a trusted data source.

[0079] In one alternative implementation,

[0080] The data set to be allocated is decomposed into temporal and spatial features. Based on the temporal features, the data set to be allocated is time-series sliced ​​to obtain segmented data units. A multi-level allocation topology is constructed based on the spatial features and the hierarchical relationship, including:

[0081] The timestamp field of the data records in the data set to be allocated is extracted and parsed into a standard time format. The time interval is calculated based on the standard time format and the time period pattern is identified to obtain the time dimension feature. The organization node identifier corresponding to the business entity identifier of the data record in the data set to be allocated is extracted. The organization hierarchy depth and organization affiliation path are extracted based on the organization node identifier to obtain the spatial dimension feature. The time slice boundary is determined based on the time dimension feature and the data set to be allocated is divided into multiple time periods according to the time slice boundary. The data records in each time period are aggregated and calculated, and the numerical fields are extracted to generate segmented data units.

[0082] Extract the hierarchical node identifiers corresponding to each business entity identifier from the hierarchical affiliation relationship and construct a hierarchical node set based on the hierarchical node identifiers. Based on the spatial dimension features, classify the hierarchical node identifiers in the hierarchical node set into layers and mark the hierarchical positions. Based on the hierarchical positions and the organizational affiliation path, determine the hierarchical node identifiers with the same superior node and establish a peer-level association relationship. Based on the hierarchical positions and the hierarchical affiliation relationship, establish a hierarchical transmission relationship between different levels and construct a multi-level distribution topology structure by combining the peer-level association relationship.

[0083] When extracting timestamp fields from data records, the system identifies the time formats that different data sources may use, including Unix timestamps, ISO 8601 format, or custom date and time strings. During parsing, regular expressions are used to match time format patterns, and timestamps of different formats are uniformly converted to a standard time format, such as the ISO 8601 extended format in the UTC time zone. The standardized timestamps can support cross-system time calculation and comparison.

[0084] The time interval is calculated based on the timestamp difference between adjacent data records. The dataset to be allocated is sorted in ascending order by timestamp, and the time difference between the (i+1)th record and the ith record is calculated, forming a time interval sequence. Statistical analysis of this time interval sequence identifies the most frequent interval as the baseline period. The identification of time periodic patterns uses autocorrelation function analysis to calculate the correlation coefficient of the data sequence at different time offsets. The time offset corresponding to a periodic peak in the correlation coefficient is the periodic pattern. Common periodic patterns include daily, weekly, and monthly cycles; these periodic characteristics collectively constitute the time dimension features.

[0085] Spatial dimension features are extracted starting from business entity identifiers. Each data record's associated business entity identifier is mapped to a specific organizational node identifier through an organizational structure mapping. Organizational node identifiers typically use a hierarchical encoding method, such as dotted decimal notation, where the first-level department is coded as 01, and its subordinate second-level departments are coded as 01.01 and 01.02. The organizational hierarchy depth can be determined by parsing the number of separators in the node identifier; the number of separators plus one indicates the node's level. The organizational affiliation path is obtained by extracting node identifiers level by level. For example, the affiliation path of node 01.02.03 is root node, 01, 01.02, 01.02.03, describing the complete path from the top level to the current node.

[0086] The time-series slice boundaries are determined based on the periodic patterns identified in the time dimension features. When a daily periodic pattern is identified, midnight of each day is used as the slice boundary; when a weekly periodic pattern is identified, the start day of each week is used as the slice boundary. For irregular business cycles, the time points that ensure a balanced distribution of data across time intervals are selected as slice boundaries based on the distribution characteristics of the time interval sequence. After the slice boundaries are determined, each record in the dataset to be allocated is traversed, and its time interval is determined based on its timestamp. Data records within the same time interval are grouped according to the business entity identifier, and summation, averaging, or weighted calculations are performed on the numerical fields of each group to generate segmented data units containing the time interval identifier, business entity identifier, and aggregated value.

[0087] The hierarchical node set is constructed by extracting all business entity identifiers from the hierarchical affiliation relationship. The data set to be allocated is traversed, collecting all unique business entity identifiers and retrieving the corresponding hierarchical node identifiers through the organizational structure mapping relationship. These hierarchical node identifiers are then deduplicated to form the hierarchical node set. The hierarchical depth is calculated for each node identifier in the set, and the depth value is determined based on the hierarchical separator in the encoding structure. Node identifiers with a depth value of 1 are classified as the first level, nodes with a depth value of 2 as the second level, and so on, completing the hierarchical operation. The hierarchical results are marked by a hierarchical position attribute, which records the hierarchical value of the node.

[0088] Establish sibling relationships by identifying nodes with the same parent node. For nodes at the same level, compare the parent node identifier along their organizational affiliation path to see if they are identical. For example, if nodes 01.02.01 and 01.02.02 both have 01.02 as their parent node, then these two nodes are established as siblings. Sibling relationships are stored in an adjacency list; each node maintains a list of sibling nodes, storing the identifiers of all nodes at its level.

[0089] A hierarchical relationship is established based on hierarchical location and attribution. For example, for a node at hierarchical location L, its parent node is located at hierarchical location L-1, and the parent node's identifier can be directly located through the organizational attribution path. When establishing a transitive relationship from a parent node to a child node, a list of child nodes is added to the parent node record, and the parent node is marked as a reference in the child node record. This bidirectional association structure supports top-down distribution and bottom-up aggregation and tracing.

[0090] A multi-level allocation topology is constructed by integrating peer-to-peer relationships and hierarchical transmission relationships. A graph structure is used to represent the topology, with each level node identifier as a vertex. Peer-to-peer relationships form horizontal edges, and hierarchical transmission relationships form vertical edges. Vertex attributes include node identifier, level position, organizational affiliation path, and associated segmented data units. Vertical edges are appended with a transmission weight attribute, which reflects the proportional coefficient when a higher-level node allocates data to a lower-level node. Horizontal edges are appended with a collaboration coefficient attribute, used to describe the strength of business associations between peer-to-peer nodes.

[0091] After the topology is constructed, its validity is verified. Check for isolated nodes (vertices without any edges connecting them); the presence of isolated nodes usually indicates an incomplete hierarchical structure. Verify that the vertical edges form a directed acyclic graph to avoid circular dependencies that could cause amortization calculations to fail to converge. Check that each non-leaf node has at least one child node to ensure that amortization propagation occurs layer by layer. For the root node, verify that its level is the top level and that it is not referenced by any parent node.

[0092] In practical applications, multi-level amortization topologies may contain hundreds of nodes and thousands of edges. To improve the efficiency of subsequent amortization calculations, the topology is preprocessed and optimized. Nodes are topologically sorted according to their hierarchical position, ensuring that higher-level nodes precede lower-level nodes in the calculation sequence. Nodes within the same level are arranged lexicographically according to their organizational affiliation path, guaranteeing the deterministic nature of the calculation order. The topology is serialized into a matrix form, where rows represent higher-level nodes, columns represent lower-level nodes, and matrix elements are transitive weights. This matrix representation supports efficient matrix operations for amortization value calculation.

[0093] For dynamically changing organizational structures, the topology construction process supports an incremental update mechanism. When organizational mapping relationships change, the affected node identifiers and their associated edges are identified. For new nodes, the hierarchical position and organizational affiliation path are extracted according to standard procedures, and the association with existing nodes is established before insertion into the topology. For deleted nodes, it is first checked whether there are subordinate nodes. If so, the subordinate nodes need to be reassigned to other superior nodes or the hierarchical affiliation needs to be adjusted, and then the node and its associated edges are removed. For changes in node attributes, the corresponding hierarchical position and organizational affiliation path are updated, and sibling relationships and hierarchical transitive relationships are recalculated.

[0094] In one alternative implementation,

[0095] Based on the multi-level amortization topology, the segmented data units are coupled and solved, and the node amortization values ​​are obtained by decomposing them layer by layer, including:

[0096] Node identifiers and layer depths are extracted from the multi-level allocation topology to determine the initial distribution position of the segmented data units in the multi-level allocation topology. Responsibility weight vectors are extracted from the responsibility attribution relationship and a layer weight tensor is constructed according to the pre-obtained layer position. Tensor decomposition is performed on the layer weight tensor to extract the core tensor and factor matrix. Based on the factor matrix, the intra-layer coupling coefficient and inter-layer coupling coefficient are calculated and combined with the core tensor to reconstruct the global coupling weight tensor.

[0097] The values ​​of the segmented data units are loaded at the bottom-level nodes according to the initial distribution positions to construct an initial state vector. The initial state vector is then subjected to tensor contraction operation with the global coupling weight tensor to obtain a coupled state field. The coupled state field is then subjected to Laplace diffusion, and diffusion boundary conditions are set based on the sibling relationships in the multi-level amortization topology. The steady-state field distribution after diffusion equilibrium is solved to obtain the coupled amortization value. The coupled amortization value is then integrated and aggregated layer by layer according to the hierarchical structure of the multi-level amortization topology to obtain an intermediate amortization value. The inter-layer gradient operator is extracted from the global coupling weight tensor, and the intermediate amortization value is then subjected to variational descent decomposition starting from the top-level node based on the inter-layer gradient operator to obtain the node amortization value.

[0098] After obtaining the segmented data units and the multi-level amortization topology, a mapping relationship needs to be established between them to achieve accurate amortization calculations. The multi-level amortization topology contains multiple levels, each containing several nodes, with parent-child and sibling relationships between nodes. Node identifiers and level depths are extracted from this topology. Node identifiers use a hierarchical encoding method; for example, first-level nodes are encoded as 1.0, 2.0, second-level nodes as 1.1, 1.2, 2.1, and so on. Level depth represents the vertical position of a node in the topology, with the root node having a level depth of 0, increasing progressively downwards. Based on the business entity identifiers contained in the segmented data units, they are matched with the node identifiers to determine the lowest-level node to which the data unit should belong, forming an initial distribution location mapping table.

[0099] The responsibility attribution relationship includes the weight ratio of each business entity in different responsibility dimensions. For example, a business entity may have 30% responsibility for costs and 45% responsibility for revenue. These weight ratios are extracted and organized into a responsibility weight vector. The dimensions of this vector correspond to the number of responsibility types, and each element in the vector corresponds to a weight value for a specific responsibility type. Based on the hierarchical position of each node in the multi-level allocation topology, the responsibility weight vector is arranged and combined at different levels to construct a three-level hierarchical weight tensor. The three dimensions of this tensor correspond to the node number, hierarchical depth, and responsibility type, respectively, and the element values ​​in the tensor are the weight values ​​of the corresponding node at a specific level and responsibility type.

[0100] Tensor decomposition of the hierarchical weight tensor is performed using the Tucker decomposition method, decomposing the tensor into a product of a core tensor and three factor matrices. The core tensor reflects the interaction relationships between the three dimensions, while the factor matrices describe the inherent patterns of nodes, hierarchies, and responsibility types, respectively. The decomposition process is implemented using an iterative optimization algorithm, with a convergence threshold of 0.001. Iteration stops when the tensor reconstruction error of two consecutive iterations is less than this threshold. From the decomposed factor matrices, node factor matrices and hierarchical factor matrices are extracted, and intra-layer coupling coefficients are calculated. Intra-layer coupling coefficients reflect the mutual influence strength between different nodes within the same hierarchical level and are obtained by calculating the cosine similarity of the row vectors of the node factor matrices. Inter-layer coupling coefficients reflect the transitive relationships between different hierarchies and are obtained by performing inner product operations on the column vectors of the hierarchical factor matrices. The core tensor, intra-layer coupling coefficients, and inter-layer coupling coefficients are then reconstructed using tensor multiplication to generate a global coupling weight tensor, which fully describes the coupling relationships between all nodes in the multi-level amortized topology.

[0101] The values ​​in the segmented data units are loaded onto the corresponding bottom-level nodes according to the initial distribution location mapping table. Each bottom-level node corresponds to a value. The values ​​of all bottom-level nodes are arranged into a vector according to their node indices to form the initial state vector. The length of this vector is equal to the total number of bottom-level nodes, and the elements in the vector are the values ​​of the data units carried by the corresponding nodes. For nodes that have not been assigned data, their corresponding vector elements are set to zero. The initial state vector is then subjected to tensor shrinkage with the global coupling weight tensor. Specifically, the initial state vector is expanded by multiplication along the first dimension of the tensor to obtain the coupled state field. The coupled state field is a second-order tensor, whose two dimensions correspond to the hierarchy depth and responsibility type, respectively. The element values ​​in the tensor represent the coupling state strength at a specific hierarchy and responsibility type.

[0102] Laplace diffusion is applied to the coupled state field to simulate the propagation of numerical data within the topology. Laplace diffusion follows the diffusion equation, with the diffusion coefficient adjusted according to the hierarchical depth. The diffusion coefficient for deeper nodes is set smaller to ensure the stability of the allocation results. Diffusion boundary conditions are set based on the sibling relationships in the multi-level allocation topology. For nodes with sibling relationships, numerical data is allowed to diffuse between sibling nodes; for independent nodes, a zero-flux boundary condition is set to prevent outward diffusion. The diffusion equation is solved using a numerical iterative method. When the rate of change of the diffusion field falls below a set threshold of 0.0001, a diffusion equilibrium state is considered reached, and the field distribution at this point is the steady-state field distribution. The numerical values ​​at each location in the steady-state field distribution are the coupled allocation values, which comprehensively consider the coupling relationships between nodes and the diffusion propagation effect.

[0103] The coupled allocation values ​​are aggregated layer by layer according to the hierarchical structure of the multi-level allocation topology. Starting from the bottom layer, the coupled allocation values ​​of all child nodes under the same parent node are summed to obtain the intermediate allocation value of that parent node. Aggregation is performed layer by layer upwards until the top-level node is reached. During the aggregation process, the responsibility type dimension needs to be considered, and the allocation values ​​under different responsibility types are aggregated separately to obtain the intermediate allocation value of each node under each responsibility type. The intermediate allocation value reflects the cumulative effect from the bottom up, but does not yet reflect the decomposition constraints of the top layer on the bottom layer.

[0104] Inter-layer gradient operators are extracted from the global coupled weight tensor. The inter-layer gradient operator describes the rate of change of numerical values ​​between adjacent layers and is obtained by differencing the global coupled weight tensor along the layer dimension. This operator is a matrix, with rows corresponding to upper-layer nodes and columns corresponding to lower-layer nodes, and matrix elements representing gradient coefficients from upper to lower layers. Based on the inter-layer gradient operator, variational descent decomposition is performed on the intermediate amortized values ​​starting from the top-layer nodes. Variational descent is an optimization process that aims to distribute the total amortized value of the top layer to the lower-layer nodes according to a reasonable gradient, while satisfying bottom-up aggregation constraints. Variational descent uses the gradient descent algorithm, iteratively updating the amortized values ​​of each node with a learning rate of 0.01 and 500 iterations. In each iteration, the residual between the current amortized value and the target constraint is calculated, the gradient direction is calculated based on the inter-layer gradient operator, and the amortized value is updated along the gradient direction. After iterative optimization, the final node allocation value is obtained. This allocation value satisfies both bottom-up aggregation consistency and top-down decomposition rationality, achieving balanced allocation under bidirectional constraints. The node allocation value provides an accurate data foundation for subsequent anomaly detection and audit analysis.

[0105] Figure 2 This is a flowchart illustrating the analysis of the relationship between node allocation values ​​and responsibility attribution in the audit report generation method based on a dynamic allocation engine, as described in an embodiment of the present invention.

[0106] In one alternative implementation,

[0107] Based on the node allocation value and the responsibility attribution relationship, extract all node allocation values ​​of the target audit subject and perform frequency domain transformation to extract time-series evolution features. Based on the time-series evolution features, calculate the abnormal deviation metric value and combine it with the multi-level allocation topology to backtrack and determine the influencing source nodes, including:

[0108] Extract the allocated value sequence corresponding to the target audit subject from the allocated value of the node and construct the audit subject time series matrix. Perform wavelet transform decomposition on the audit subject time series matrix to obtain multi-scale wavelet coefficients and extract instantaneous frequency and instantaneous amplitude through Hilbert transform. Calculate the frequency drift rate based on the instantaneous frequency and the amplitude envelope based on the instantaneous amplitude. Combine the frequency drift rate with the time-frequency plane to obtain the time series evolution characteristics.

[0109] Based on the attribution of responsibility and the audit subject's time series matrix, a baseline evolution trajectory is calculated. The time series evolution features are decomposed into singular spectra and the dominant singular vector is extracted. The subspace angle between the dominant singular vector and the feature vector corresponding to the baseline evolution trajectory is calculated, and the abnormal deviation metric is obtained by combining the Hausdorff distance between the time series evolution features and the baseline evolution trajectory.

[0110] From the multi-level allocation topology, the associated nodes corresponding to the target audit entity are extracted, the node allocation value corresponding to the associated node is extracted and the mutual information value with the time series matrix of the audit entity is calculated, the associated nodes with mutual information values ​​exceeding a preset threshold are selected to construct a candidate impact source set, the responsibility weights corresponding to each edge on the transmission path from each associated node in the candidate impact source set to the node corresponding to the target audit entity are extracted and the cumulative weight value is calculated, and the impact contribution is calculated in combination with the abnormal deviation metric value to determine the impact source node.

[0111] When conducting time-series characteristic analysis on a target audit entity, data relevant to that entity is selected from all node allocation values. Based on the attribution of responsibility, all node identifiers corresponding to the target audit entity are located. The allocation values ​​of these nodes in different time slices are extracted from each segmented data unit and arranged chronologically to form an allocation value sequence. This allocation value sequence reflects the numerical changes of the audit entity over time. Organizing the allocation value sequences of different nodes by rows or columns constructs an audit entity time-series matrix. One dimension of the matrix corresponds to the time scale, and the other dimension corresponds to different nodes or business categories, enabling structured analysis of subsequent time evolution patterns.

[0112] When performing multi-scale analysis on the audit subject's time series matrix, wavelet transform is used for decomposition. By selecting appropriate wavelet basis functions, discrete or continuous wavelet transforms are applied to each row of data in the time series matrix, decomposing the original signal into different scale levels. Each scale corresponds to a specific time-frequency range, and wavelet coefficients at each scale are obtained through decomposition. These wavelet coefficients characterize the local properties of the signal within different time-frequency windows. Multi-scale wavelet coefficients can reveal the variation patterns of data at different time granularities, preserving low-frequency trend components while capturing high-frequency abrupt changes.

[0113] To further extract the instantaneous characteristics of the time-series signal, the coefficients obtained from the wavelet transform are processed by the Hilbert transform. The Hilbert transform converts the real-valued signal into an analytic signal. By calculating the amplitude and phase of the analytic signal, the instantaneous amplitude and instantaneous phase are obtained. The instantaneous frequency is obtained by taking the derivative of the instantaneous phase with respect to time, reflecting the rate of change of the signal frequency with time. The instantaneous amplitude characterizes the intensity of the signal energy distribution along the time axis. Instantaneous frequency and instantaneous amplitude are extracted for the wavelet coefficients at each scale, forming a three-dimensional representation of time, frequency, and amplitude.

[0114] Based on instantaneous frequency sequences, the frequency drift rate is calculated to quantify the rate of frequency change. By performing a difference operation or differentiation on the instantaneous frequency sequence, the rate of change of frequency over time is obtained, i.e., the frequency drift rate. This indicator reflects the accelerating or decelerating trend of numerical changes in the audited entity; a positive value indicates an increase in frequency, and a negative value indicates a decrease in frequency. Abnormal fluctuations in the frequency drift rate may indicate sudden changes or instability in business activities.

[0115] For instantaneous amplitude sequences, an amplitude envelope is constructed to capture the overall trend of energy changes. The upper envelope curve of the amplitude is obtained by connecting the local maxima of the instantaneous amplitude or by using a sliding window smoothing method. The amplitude envelope eliminates high-frequency oscillations, preserves the main trend of energy changes, and facilitates the identification of significant energy enhancement or decay phases.

[0116] The frequency drift rate and amplitude envelope are jointly represented in the time-frequency plane to construct a two-dimensional feature space. The horizontal axis represents time, and the vertical axis represents frequency. The intensity of the amplitude envelope is represented by color or contour lines, while the direction and magnitude of the frequency drift rate are indicated by arrows or gradient directions. This joint representation method forms a time-series evolution feature, comprehensively depicting the dynamic characteristics of the numerical changes of the audit subject, including the rate of change, acceleration of change, and energy distribution pattern.

[0117] To determine whether the temporal evolution of the target audit entity is abnormal, a baseline evolutionary trajectory needs to be constructed as a reference standard. Based on the attribution of responsibility, other audit entities at the same level or comparable to the target audit entity are identified, and their temporal matrix data is extracted. Statistical analysis is performed on the temporal data of multiple comparable entities, calculating the mean, median, or weighted average at each time point to form the baseline evolutionary trajectory. This trajectory represents the temporal change pattern that an audit entity should exhibit under normal circumstances, reflecting the regular evolutionary patterns of business activities.

[0118] Singular spectral decomposition (SSD) is performed on the temporal evolution characteristics of the target audit subject, decomposing the time-series signal into trend, periodic, and noise components. A trajectory matrix of the time-series matrix is ​​constructed, and singular value decomposition (SSD) is performed on the trajectory matrix to obtain a series of singular values ​​and corresponding singular vectors. The magnitude of the singular values ​​reflects the contribution of each component to the original signal; the singular vectors corresponding to the largest singular values ​​are selected as dominant singular vectors. The dominant singular vectors represent the main change patterns of the time-series signal and can extract the core features of the signal. SSD is then performed on the baseline evolution trajectory to extract its dominant singular vector as the baseline feature vector. The subspace angle between the dominant singular vector of the target audit subject and the baseline feature vector is calculated to quantify the degree of deviation between the two in the feature space. The subspace angle is obtained by calculating the principal angle between the subspaces spanned by the two sets of vectors; a larger angle indicates a more significant difference between the evolution pattern of the target subject and the baseline pattern.

[0119] To more comprehensively measure the degree of anomalies, the Hausdorff distance between the temporal evolution feature and the baseline evolution trajectory is calculated. The Hausdorff distance measures the maximum mismatch between two sets of points. It is calculated by taking the maximum value of the minimum distances from each point on the temporal evolution feature curve to the baseline trajectory. This distance reflects the magnitude of the deviation of the temporal curve at its most extreme position, and can capture significant local anomalies.

[0120] The subspace angle and Hausdorff distance are combined to calculate the anomaly deviation metric. The two metrics can be normalized and then weighted and summed, with the weights adjusted according to actual needs. A larger anomaly deviation metric indicates a more severe deviation between the target audit subject's temporal evolution characteristics and the normal pattern, and a higher probability of anomalies.

[0121] When an abnormal deviation metric is detected exceeding a preset warning threshold, it is necessary to backtrack through the multi-level allocation topology to determine the source node. The node corresponding to the target audit subject is located within the topology, and all associated nodes of that node are extracted, including direct upstream nodes, downstream nodes, and related nodes at the same level. Associated nodes are connected by edges in the topology; the existence of edges indicates the transfer of business data or responsibility relationships.

[0122] For each associated node, its corresponding node allocation value sequence is extracted, and a mutual information value is calculated with the audit subject's time series matrix. Mutual information measures the statistical dependence between two random variables. By calculating the difference between the joint probability distribution and the marginal probability distribution, it quantifies the degree to which one variable reduces the uncertainty of the other. The larger the mutual information value, the stronger the correlation between the associated node and the target audit subject, and the more likely the node is a source of transmission or influence of anomalies.

[0123] A mutual information threshold is set, and associated nodes with mutual information values ​​exceeding this threshold are selected to construct a candidate source of influence. Nodes in the candidate set are potential sources of anomalous influence, and their influence intensity needs to be further quantified.

[0124] In a multi-level attribution topology, the transmission path is traced from each associated node in the candidate impact source set to the target audit subject node. The transmission path consists of a series of nodes and edges, with each edge labeled with a responsibility weight, indicating the strength of data or impact transmission along that edge. The responsibility weight of each edge on the path is extracted, and the responsibility weights of all edges on the path are accumulated to obtain the cumulative weight value. The accumulation method can be a product, representing a progressively decaying transmission effect, or a weighted sum, representing the cumulative contribution.

[0125] The influence contribution of each candidate node is calculated by combining the anomaly deviation metric and the weighted cumulative value. The influence contribution comprehensively considers the correlation strength between the node and the target subject, the weight of the transmission path, and the severity of the anomaly. The calculation formula can multiply or weight the mutual information value, the weighted cumulative value, and the anomaly deviation metric to obtain the contribution score of each candidate node to the anomaly of the target subject.

[0126] The nodes in the candidate source of influence set are sorted according to their contribution, and the nodes with the highest contribution are selected as the source of influence nodes. These source nodes are the main causes of abnormal deviations from the target audit entity, and they require focused attention and corrective action. This backtracking process traces anomalies back to specific business nodes or responsible entities, providing a clear problem identification for audit decisions.

[0127] In one alternative implementation,

[0128] The propagation attenuation factor corresponding to the source node is iteratively corrected and decomposed to obtain a corrected allocation value. The corrected allocation value and the compliance status are associated to generate an audit item set, which is then aggregated according to the hierarchical relationship to obtain a structured audit report, including:

[0129] Extract the transmission path from the source node to the target audit entity node from the multi-level allocation topology, extract the initial transmission attenuation factor corresponding to each edge of the transmission path from the responsibility attribution relationship, calculate the correction gradient based on the abnormal deviation metric and the pre-acquired impact contribution, iteratively update the initial transmission attenuation factor according to the correction gradient, and perform exponential attenuation correction based on the hierarchical span of the transmission path to obtain the hierarchical correction transmission attenuation factor.

[0130] The node allocation value corresponding to the source node of the influence is passed layer by layer according to the transmission path, and the hierarchical correction transmission attenuation factor is applied at each layer for attenuation calculation. Based on the same-level association relationship in the multi-level allocation topology, horizontal diffusion correction is performed, and the corrected allocation value is obtained by aggregating at the target audit subject node.

[0131] The modified allocation value is associated with the compliance determination result in the compliance status, and the influence source node identifier is added to construct the audit entry. The audit entries are grouped and aggregated according to the hierarchical relationship and sorted in descending order according to the abnormal deviation metric to obtain the structured audit report.

[0132] After calculating the abnormal deviation metric and locating the source nodes, the detected anomalies need to be traced and corrected, and a structured report for the auditing entity needs to be generated. This process involves extracting the transmission path, dynamically correcting the attenuation factor, and aggregating and presenting multi-level data.

[0133] When extracting the propagation path from the source node to the target audit entity node in a multi-level allocation topology, a breadth-first traversal strategy is adopted. The source node is added to the queue as the starting point, and child nodes are visited layer by layer until the level of the target audit entity node is reached. During the traversal, the node sequence and connecting edges of all reachable paths are recorded, forming a path set. When multiple propagation paths exist, the shortest path chain is selected first. If the path lengths are the same, the path with the larger cumulative edge weight is selected as the primary propagation path. The initial propagation attenuation factor corresponding to each edge on the propagation path is extracted from the responsibility attribution relationship. The initial propagation attenuation factor reflects the propagation strength of business responsibility between adjacent levels. The initial propagation attenuation factor is usually preset according to business rules. For example, when a first-level department propagates to a second-level department, the attenuation factor may be 0.85, indicating that 85% of the responsibility intensity is retained during the propagation.

[0134] When calculating the correction gradient based on the anomaly deviation metric and the impact contribution, it is necessary to quantify the contribution of the source node to the anomaly of the target node. The impact contribution is obtained by analyzing the correlation between the numerical fluctuation of the source node and the anomaly deviation metric of the target node, and the delay correlation coefficient between the two is calculated using time-series cross-correlation analysis. The correction gradient is directly proportional to the anomaly deviation metric and inversely proportional to the deviation of the current propagation attenuation factor. When the anomaly deviation metric is large and the impact contribution is high, it indicates that the current propagation attenuation factor setting is unreasonable and requires a significant correction. The calculation of the correction gradient also needs to consider the hierarchical span of the propagation path; the larger the span, the smaller the single-step correction magnitude should be to avoid over-correction.

[0135] When iteratively updating the initial transmission attenuation factor, a gradient descent strategy is employed. The corrected gradient is multiplied by a preset learning rate and then weighted and fused with the initial transmission attenuation factor to obtain the first-round corrected transmission attenuation factor. In each iteration, the node amortization value based on the new attenuation factor is recalculated, and it is checked whether the abnormal deviation metric of the target audit subject node converges below a threshold. When the abnormal deviation metric continuously decreases and the change is less than one percent for three consecutive iterations, the iteration is considered to have converged. An exponential decay correction mechanism is introduced during the iteration process to adjust the corrected transmission attenuation factor a second time based on the hierarchical span of the transmission path. The hierarchical span is obtained by calculating the shortest path length between the source node and the target node in the topology. When the span is three layers, the attenuation coefficient is 0.9, and when the span is five layers, the attenuation coefficient drops to 0.75. Multiplying the corrected transmission attenuation factor by the corresponding attenuation coefficient yields the hierarchical corrected transmission attenuation factor, which more accurately reflects the cumulative attenuation effect in long-distance transmission.

[0136] The initial transmission value is the node allocation value corresponding to the source node, which is then passed down layer by layer along the extracted transmission path. During transmission at each layer, the value transmitted from the upper layer is multiplied by the layer-level correction transmission attenuation factor corresponding to the current edge to obtain the attenuated value transmitted to the lower layer. When a node receives transmission values ​​from multiple upper-level nodes, all transmission values ​​are weighted and summed, with the weights determined based on the business association strength between each upper-level node and the current node. Simultaneously with vertical transmission, horizontal diffusion correction between sibling nodes must also be considered. Sibling relationships are extracted from the multi-level allocation topology, identifying node pairs at the same level with business collaboration relationships. For these node pairs, the values ​​are horizontally adjusted based on the collaboration strength coefficient. When a node's allocation value is too high due to anomalies, some values ​​will diffuse to adjacent collaborative nodes, with the diffusion ratio typically controlled between 5% and 15%. Horizontal diffusion correction avoids numerical abrupt changes caused by anomalies concentrated in a single node, making the corrected allocation result more consistent with the actual business distribution. The transmitted values ​​are finally aggregated at the target audit entity node. The values ​​that reach this node through all transmission paths, as well as the values ​​of the horizontal diffusion adjustment, are added together to obtain the corrected allocation value for this node.

[0137] When associating the revised allocation value with the compliance judgment result in the compliance status, the correspondence between the two needs to be clearly defined. The compliance judgment result originates from the matching process between the original data and the preset audit rule base, and includes status indicators such as compliant, non-compliant, and partially compliant. For each node allocation value, its corresponding original business event identifier is extracted, and the compliance status of that business event in the data set to be allocated is queried. When there is a significant difference between the revised allocation value and the original node allocation value of a node, and the compliance judgment result of the corresponding business event is non-compliant or partially compliant, the node will be marked as an audit object requiring key attention. When constructing audit entries, the revised allocation value is used as a numerical field, the compliance judgment result is used as a status field, and the impact source node identifier is added as a tracing field. The impact source node identifier can help auditors quickly locate the root cause of the current node's anomaly, improving audit efficiency. Audit entries also include metadata such as timestamps, business entity identifiers, and business event types, forming a complete audit record unit.

[0138] When grouping and aggregating audit items according to hierarchical affiliation, a multi-level grouping strategy is adopted. The first level of grouping is based on the top-level organization of the business entity, grouping all audit items belonging to the same group or headquarters together. The second level of grouping is based on second-level departments or branches, and the third level of grouping is refined to specific business units or responsibility centers. Within each group, audit items are sorted in descending order based on their abnormal deviation metrics, placing the audit item with the highest degree of abnormality at the top of the group to facilitate auditors prioritizing high-risk items. During the descending sorting process, when the abnormal deviation metrics of multiple audit items are close, the deviation ratio between the corrected allocation value and the original node allocation value is further compared, with items having a larger deviation ratio ranked higher. After sorting, summary statistics are calculated at each group level, including the total number of audit items in the group, the number of non-compliance items, the average and maximum values ​​of the abnormal deviation metrics, and the sum of the corrected allocation values. These statistical indicators are appended as group-level summary information to the group title, facilitating auditors' rapid assessment of the overall compliance status at each organizational level.

[0139] The aggregated and sorted audit entries are formatted and formatted according to a predefined report template to generate a structured audit report. The report employs a hierarchical tree structure. The top layer displays a global summary, including the total number of audit entries, overall compliance rate, average outlier metric, and a list of major impacting nodes. The middle layer expands according to the organizational structure hierarchy, with each organizational node displaying its group-level summary and links to its subordinate groups. The bottom layer provides detailed information on each audit entry, including corrected allocation values, compliance determination results, impacting node identifiers, timestamps, and business event descriptions. The report supports hierarchical collapse and expansion, allowing auditors to select the level of detail they need. The report also offers multi-dimensional filtering capabilities, allowing filtering by compliance status, outlier metric range, time range, and impacting node to quickly pinpoint specific types of audit issues. The generated structured audit report is stored in a standardized data format and can be exported to various document formats for subsequent analysis.

[0140] A second aspect of this invention provides an audit report generation system based on a dynamic allocation engine, comprising:

[0141] The data extraction unit is used to obtain the mapping relationship between raw data and organizational structure, extract the business entity identifier and business event identifier corresponding to the raw data, determine the hierarchical and responsibility attribution relationships corresponding to the business entity identifier based on the organizational structure mapping relationship, and match the business event identifier with the preset audit rule base and mark the compliance status to obtain the data set to be allocated.

[0142] The apportionment calculation unit is used to perform dimensional decomposition on the data set to be apportioned to obtain time dimension features and spatial dimension features, perform time-series slicing on the data set to be apportioned based on the time dimension features to obtain segmented data units, construct a multi-level apportionment topology based on the spatial dimension features and the hierarchical affiliation relationship, and perform coupled solution on the segmented data units according to the multi-level apportionment topology and decompose layer by layer to obtain the node apportionment value.

[0143] The audit generation unit is used to extract all node allocation values ​​of the target audit subject based on the node allocation value and the responsibility attribution relationship, and perform frequency domain transformation to extract time-series evolution features. Based on the time-series evolution features, it calculates the abnormal deviation metric value and combines it with the multi-level allocation topology to backtrack and determine the influencing source node. It iteratively corrects and decomposes the transmission attenuation factor corresponding to the influencing source node to obtain the corrected allocation value. It associates the corrected allocation value and the compliance status to generate an audit item set and aggregates it according to the hierarchical attribution relationship to obtain a structured audit report.

[0144] A third aspect of the present invention provides an electronic device, comprising:

[0145] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0146] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0147] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0148] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An audit report generation method based on a dynamic allocation engine, characterized in that, include: Obtain the mapping relationship between raw data and organizational structure, extract the business entity identifier and business event identifier corresponding to the raw data, determine the hierarchical and responsibility attribution relationships corresponding to the business entity identifier based on the organizational structure mapping relationship, match the business event identifier with the preset audit rule base and mark the compliance status to obtain the data set to be allocated; The data set to be allocated is decomposed into time dimension features and spatial dimension features. Based on the time dimension features, the data set to be allocated is sliced ​​into segmented data units. Based on the spatial dimension features and the hierarchical relationship, a multi-level allocation topology is constructed. The segmented data units are coupled and solved according to the multi-level allocation topology and decomposed layer by layer to obtain the node allocation value. Based on the node allocation value and the responsibility attribution relationship, extract all node allocation values ​​of the target audit subject and perform frequency domain transformation to extract time-series evolution features. Calculate the abnormal deviation metric value based on the time-series evolution features and backtrack to determine the influencing source node in conjunction with the multi-level allocation topology. Iteratively correct and decompose the transmission attenuation factor corresponding to the influencing source node to obtain the corrected allocation value. Associate the corrected allocation value and the compliance status to generate an audit item set and aggregate it according to the hierarchical attribution relationship to obtain a structured audit report.

2. The method according to claim 1, characterized in that, Obtaining the mapping relationship between raw data and organizational structure, extracting the business entity identifiers and business event identifiers corresponding to the raw data, and determining the hierarchical and responsibility relationships corresponding to the business entity identifiers based on the organizational structure mapping relationship include: The system obtains the original data from the data source of the business object to be audited and the organizational structure mapping relationship from the organizational structure management system. It parses the field structure of the data records in the original data and identifies the field type identifiers. Based on the field type identifiers, the data records are divided into entity attribute field sets and event attribute field sets. The system extracts the field values ​​corresponding to the entity attribute field sets to generate business entity identifiers. It performs timestamp parsing and event type matching on the event attribute field sets to generate business event identifiers. It performs multi-level matching on the business entity identifiers and the organizational structure mapping relationship and calculates the matching score. Based on the matching score, it determines the affiliated organizational node of the business entity identifier and extracts the corresponding hierarchical path. Based on the parent-child relationship between different nodes in the hierarchical path, it determines the hierarchical affiliation relationship. Determine the set of superior responsibility nodes and the set of peer responsibility nodes of the affiliated organization node. Calculate the superior responsibility weight based on the hierarchical span between each node in the superior responsibility node set and the affiliated organization node. Calculate the peer responsibility weight based on the organizational affiliation path similarity between each node in the peer responsibility node set and the affiliated organization node. Construct a responsibility weight vector based on the superior responsibility weight and the peer responsibility weight, and generate responsibility affiliation relationships.

3. The method according to claim 1, characterized in that, The business event identifier is matched with a preset audit rule base and its compliance status is marked to obtain the data set to be allocated, including: The system extracts rule condition fields and rule constraint logic from a preset audit rule base, extracts the field values ​​of the event attribute fields in the business event identifier and maps them to the rule condition fields, determines the applicable rule set based on the field mapping results, parses the logical expressions of each rule in the applicable rule set, obtains the logical expressions, extracts the constraint conditions and constraint threshold ranges, substitutes the field values ​​corresponding to the business event identifier into the logical expressions to determine the constraint conditions and calculates the rule compliance degree, and generates a compliance status identifier based on the rule compliance degree. The process involves querying the attributing organization node corresponding to the business event identifier and extracting the corresponding responsibility weight vector. The compliance status identifier is then associated and bound with the responsibility weight vector, and an event allocation record is constructed by combining the numerical field values ​​in the business event identifier. The event allocation records are grouped and aggregated according to the attributing organization node to obtain grouping results. Compliance status identifiers are extracted from the event allocation records within each group in the grouping results, and compliance filtering is performed. Finally, the event allocation records that pass the compliance filtering are summarized to obtain the set of data to be allocated.

4. The method according to claim 1, characterized in that, The data set to be allocated is decomposed into temporal and spatial features. Based on the temporal features, the data set to be allocated is time-series sliced ​​to obtain segmented data units. A multi-level allocation topology is constructed based on the spatial features and the hierarchical relationship, including: The timestamp field of the data records in the data set to be allocated is extracted and parsed into a standard time format. The time interval is calculated based on the standard time format and the time period pattern is identified to obtain the time dimension feature. The organization node identifier corresponding to the business entity identifier of the data record in the data set to be allocated is extracted. The organization hierarchy depth and organization affiliation path are extracted based on the organization node identifier to obtain the spatial dimension feature. The time slice boundary is determined based on the time dimension feature and the data set to be allocated is divided into multiple time periods according to the time slice boundary. The data records in each time period are aggregated and calculated, and the numerical fields are extracted to generate segmented data units. Extract the hierarchical node identifiers corresponding to each business entity identifier from the hierarchical affiliation relationship and construct a hierarchical node set based on the hierarchical node identifiers. Based on the spatial dimension features, classify the hierarchical node identifiers in the hierarchical node set into layers and mark the hierarchical positions. Based on the hierarchical positions and the organizational affiliation path, determine the hierarchical node identifiers with the same superior node and establish a peer-level association relationship. Based on the hierarchical positions and the hierarchical affiliation relationship, establish a hierarchical transmission relationship between different levels and construct a multi-level distribution topology structure by combining the peer-level association relationship.

5. The method according to claim 1, characterized in that, Based on the multi-level amortization topology, the segmented data units are coupled and solved, and the node amortization values ​​are obtained by decomposing them layer by layer, including: Node identifiers and layer depths are extracted from the multi-level allocation topology to determine the initial distribution position of the segmented data units in the multi-level allocation topology. Responsibility weight vectors are extracted from the responsibility attribution relationship and a layer weight tensor is constructed according to the pre-obtained layer position. Tensor decomposition is performed on the layer weight tensor to extract the core tensor and factor matrix. Based on the factor matrix, the intra-layer coupling coefficient and inter-layer coupling coefficient are calculated and combined with the core tensor to reconstruct the global coupling weight tensor. The values ​​of the segmented data units are loaded at the bottom-level nodes according to the initial distribution positions to construct an initial state vector. The initial state vector is then subjected to tensor contraction operation with the global coupling weight tensor to obtain a coupled state field. The coupled state field is then subjected to Laplace diffusion, and diffusion boundary conditions are set based on the sibling relationships in the multi-level amortization topology. The steady-state field distribution after diffusion equilibrium is solved to obtain the coupled amortization value. The coupled amortization value is then integrated and aggregated layer by layer according to the hierarchical structure of the multi-level amortization topology to obtain an intermediate amortization value. The inter-layer gradient operator is extracted from the global coupling weight tensor, and the intermediate amortization value is then subjected to variational descent decomposition starting from the top-level node based on the inter-layer gradient operator to obtain the node amortization value.

6. The method according to claim 1, characterized in that, Based on the node allocation value and the responsibility attribution relationship, extract all node allocation values ​​of the target audit subject and perform frequency domain transformation to extract time-series evolution features. Based on the time-series evolution features, calculate the abnormal deviation metric value and combine it with the multi-level allocation topology to backtrack and determine the influencing source nodes, including: Extract the allocated value sequence corresponding to the target audit subject from the allocated value of the node and construct the audit subject time series matrix. Perform wavelet transform decomposition on the audit subject time series matrix to obtain multi-scale wavelet coefficients and extract instantaneous frequency and instantaneous amplitude through Hilbert transform. Calculate the frequency drift rate based on the instantaneous frequency and the amplitude envelope based on the instantaneous amplitude. Combine the frequency drift rate with the time-frequency plane to obtain the time series evolution characteristics. Based on the attribution of responsibility and the audit subject's time series matrix, a baseline evolution trajectory is calculated. The time series evolution features are decomposed into singular spectra and the dominant singular vector is extracted. The subspace angle between the dominant singular vector and the feature vector corresponding to the baseline evolution trajectory is calculated, and the abnormal deviation metric is obtained by combining the Hausdorff distance between the time series evolution features and the baseline evolution trajectory. From the multi-level allocation topology, the associated nodes corresponding to the target audit entity are extracted, the node allocation value corresponding to the associated node is extracted and the mutual information value with the time series matrix of the audit entity is calculated, the associated nodes with mutual information values ​​exceeding a preset threshold are selected to construct a candidate impact source set, the responsibility weights corresponding to each edge on the transmission path from each associated node in the candidate impact source set to the node corresponding to the target audit entity are extracted and the cumulative weight value is calculated, and the impact contribution is calculated in combination with the abnormal deviation metric value to determine the impact source node.

7. The method according to claim 1, characterized in that, The propagation attenuation factor corresponding to the source node is iteratively corrected and decomposed to obtain a corrected allocation value. The corrected allocation value and the compliance status are associated to generate an audit item set, which is then aggregated according to the hierarchical relationship to obtain a structured audit report, including: Extract the transmission path from the source node to the target audit entity node from the multi-level allocation topology, extract the initial transmission attenuation factor corresponding to each edge of the transmission path from the responsibility attribution relationship, calculate the correction gradient based on the abnormal deviation metric and the pre-acquired impact contribution, iteratively update the initial transmission attenuation factor according to the correction gradient, and perform exponential attenuation correction based on the hierarchical span of the transmission path to obtain the hierarchical correction transmission attenuation factor. The node allocation value corresponding to the source node of the influence is passed layer by layer according to the transmission path, and the hierarchical correction transmission attenuation factor is applied at each layer for attenuation calculation. Based on the same-level association relationship in the multi-level allocation topology, horizontal diffusion correction is performed, and the corrected allocation value is obtained by aggregating at the target audit subject node. The modified allocation value is associated with the compliance determination result in the compliance status, and the influence source node identifier is added to construct the audit entry. The audit entries are grouped and aggregated according to the hierarchical relationship and sorted in descending order according to the abnormal deviation metric to obtain the structured audit report.

8. An audit report generation system based on a dynamic allocation engine, used to implement the method of any one of claims 1-7, characterized in that, include: The data extraction unit is used to obtain the mapping relationship between raw data and organizational structure, extract the business entity identifier and business event identifier corresponding to the raw data, determine the hierarchical and responsibility attribution relationships corresponding to the business entity identifier based on the organizational structure mapping relationship, and match the business event identifier with the preset audit rule base and mark the compliance status to obtain the data set to be allocated. The apportionment calculation unit is used to perform dimensional decomposition on the data set to be apportioned to obtain time dimension features and spatial dimension features, perform time-series slicing on the data set to be apportioned based on the time dimension features to obtain segmented data units, construct a multi-level apportionment topology based on the spatial dimension features and the hierarchical affiliation relationship, and perform coupled solution on the segmented data units according to the multi-level apportionment topology and decompose layer by layer to obtain the node apportionment value. The audit generation unit is used to extract all node allocation values ​​of the target audit subject based on the node allocation value and the responsibility attribution relationship, and perform frequency domain transformation to extract time-series evolution features. Based on the time-series evolution features, it calculates the abnormal deviation metric value and combines it with the multi-level allocation topology to backtrack and determine the influencing source node. It iteratively corrects and decomposes the transmission attenuation factor corresponding to the influencing source node to obtain the corrected allocation value. It associates the corrected allocation value and the compliance status to generate an audit item set and aggregates it according to the hierarchical attribution relationship to obtain a structured audit report.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images